Characterizing.Pdf

The International Rice Research Institute (IRRI) was established in 1960 by the Ford and Rockefeller Foundations with the help and approval of the Government of the Philippines. Today IRRI is one of the 16 nonprofit international research centers supported by the Consultative Group on International Agricultural Re- search (CGIAR). The CGIAR is sponsored by the Food and Agriculture Organi- zation of the United Nations, the International Bank for Reconstruction and De- velopment (World Bank), the United Nations Development Programme (UNDP), and the United Nations Environment Programme (UNEP). Its membership comprises donor countries, international and regional organizations, and private foundations. As listed in its most recent Corporate Report, IRRI receives support, through the CGIAR, from donors such as UNDP, World Bank, European Union, Asian Development Bank, International Fund for Agricultural Development (IFAD), Rockefeller Foundation, and the international aid agencies of the following governments: Australia, Bangladesh, Belgium, Brazil, Canada, People’s Republic of China, Denmark, France, Germany, India, Islamic Republic of Iran, Japan, Republic of Korea, Mexico, The Netherlands, Norway, Philippines, Spain, Sweden, Switzerland, Thailand, United Kingdom, and United States. The responsibility for this publication rests with the International Rice Research Institute.

Mailing address: DAPO Box 7777, Metro Manila, Philippines Phone: (63-2) 845-0563, 844-3351 to 53 Fax: (63-2) 891-1292, 845-0606 Email: [email protected] URL: http://www.cgiar.org.irri Riceweb: http://www.riceweb.org Riceworld: http://www.riceworld.org Courier address: Suite 1099, Pacific Bank Building 6776 Ayala Avenue, Makati Metro Manila, Philippines Tel. (63-2) 891-1236, 891-1174, 891-1258, 891-1303

Suggested citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Characterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Layout and design: Ariel Paelmo Figures and illustrations: Ariel Paelmo Cover design: Juan Lazaro IV

ISBN 971-22-0152-X Contents

FOREWORD vii

ACKNOWLEDGMENTS ix

Overview Characterizing rainfed rice environments: an overview of the biophysical aspects 3 V.P. Singh, T.P. Tuong, and S.P. Kam

Characterizing environments for sustainable rice production 33 Van Nguu Nguyen

Tools and methodologies for biophysical characterization Effect of climate, agrohydrology, and management on rainfed rice production in Central Java, Indonesia: a modeling approach 57 A. Boling, T.P. Tuong, B.A.M. Bouman, M.V.R. Murty, and S.Y. Jatmiko

Perception, understanding, and mapping of soil variability in rainfed lowlands of northeast Thailand 75 T. Oberthür and S.P. Kam

Identifying soil suitability for subsoil compaction to improve water- and nutrient-use efficiency in rainfed lowland rice 97 D. Harnpichitvitaya, G. Trébuil, T. Oberthür, G. Pantuwan, I. Craig, T.P. Tuong, L.J. Wade, and D. Suriya-Arunroj

Modeling water availability, crop growth, and yield of rainfed lowland rice genotypes in northeast Thailand 111 S. Fukai, J. Basnayake, and M. Cooper

Using reference lines to classify multienvironment trials to the target population of environments, and their potential role in environmental characterization 131 G.C. McLaren and L.J. Wade

Biophysical characterization and mapping Biophysical characterization of rainfed systems in Java and South Sulawesi and implications for research 145 I. Amien and I. Las

iii Monitoring rainfed and irrigated rice in Southeast Asia using radar remote sensing 157 R. Verhoeven, H. van Leeuwen, and E. van Valkengoed

Characterizing soil phosphorus and potassium status in lowland and upland rice-cropping regions of Indonesia 169 A. Clough, I.P.G. Widjaja-Adhi, J. Sri Adiningsih, A. Kasno, and S. Fukai

Planning and managing rice farming through environmental analysis 191 K. Borkakati, V.P. Singh, A.N. Singh, R.K. Singh, A.S.R.A.S. Sastri, and S.K. Mohanty

Agroclimatic inventory for environmental characterization of rainfed rice-based cropping systems of eastern India 215 A.S.R.A.S. Sastri and V.P. Singh

Agrohydrologic and drought risk analyses of rainfed cultivation in northwest Bangladesh 233 A.F.M. Saleh, M.A. Mazid, and S.I. Bhuiyan

Characterizing biotic stresses Characterizing biotic constraints to production of Cambodian rainfed lowland rice: limitations to statistical techniques 247 G.C. Jahn, Pheng Sophea, Pol Chanthy, and Khiev Bunnarith

Weed communities of gogarancah rice and reflections on management 269 H. Pane, E. Sutisna Noor, M. Dizon, and A.M. Mortimer

Socioeconomic characterization The role of characterization in ex ante assessment of research programs: a study in the rainfed rice production system 291 P.K. Joshi and Suresh Pal

Constraints to the adoption of modern varieties of rice in Bihar, eastern India 305 A. Kumar and A.K. Jha

Rainfed rice, risk, and technology adoption: microeconomic evidence from eastern India 323 H.N. Singh, S. Pandey, and R.A. Villano

Using gender analysis in characterizing and understanding farm-household systems in rainfed lowland rice environments 339 T. Paris, A. Singh, M. Hossain, and J. Luis

Agricultural commercialization and land-use intensification: a microeconomic analysis of uplands of northern Vietnam 371 Nguyen Tri Khiem, S. Pandey, and Nguyen Huu Hong

Economics of intensive rainfed lowland rice-based cropping systems in northwest Luzon, Philippines 391 M.P. Lucas, S. Pandey, R.A. Villano, D.R. Culannay, and T.F. Marcos

iv Integrating biophysical and socioeconomic characterization Socioeconomic and biophysical characterization of rainfed versus irrigated rice production in Myanmar 407 Y.T. Garcia, M. Hossain, and A.G. Garcia

Integration of biophysical and socioeconomic constraints in rainfed lowland rice farm characterization: techniques, issues, and ongoing IRRI research 441 C.M. Edmonds and S.P. Kam

Regional land-use analysis to support agricultural and environmental policy formulation 471 B.A.M. Bouman, R. Roetter, R.A. Schipper, and A.G. Laborte

v Foreword

Rainfed rice areas are associated with a high incidence of poverty, mainly because of low and unstable yields in a difficult environment where rainfall and water availability are both seasonal and highly variable, and soil conditions are highly heterogeneous. These areas have not benefited much from the technological advances of the Green Revolution, which have targeted the more favorable irrigated environments. Yet even small improvements made in increasing and stabilizing rice yields could significantly alleviate poverty and improve food security among the poorest of the farming communities in these areas. The challenges to attaining even these small improvements are great, however, because of the variable biophysical conditions and socioeconomic circumstances in these difficult ecosystems. Over the years, IRRI, in partnership with national agricultural research systems in several Asian countries, especially the member countries of the Rainfed Lowland and Upland Rice Research Consortia, has been carrying out studies to understand rainfed rice environments and cropping practices so as to translate research into appropriate technological interventions. This research ranges from broad regional-scale characterization to detailed farm-level studies. Studies have been and are being carried out in different geographical areas by different institutions. Different methodologies have been tested and several cross-cutting issues have emerged. IRRI, together with the Central Research Institute for Food Crops, Indonesia, organized a thematic workshop on Characterizing and Understanding Rainfed Envi- ronments held 5-9 December 1999 in Bali, Indonesia. The objectives were to review progress on research related to characterizing and understanding rainfed lowland rice environments, with emphasis on work done at RLRRC sites, and consider future research issues and opportunities for collaboration. Seventy scientists from 15 coun-

Foreword vii tries gathered to discuss subjects covering these objectives. This book, which was partially supported by the Australian Centre for International Agricultural Research, contains the papers presented at the workshop. We hope that this will be a useful source of information for translating our understanding into appropriate technological interventions in the highly heterogeneous and variable rainfed environments.

Ronald P. Cantrell Director General

viii Foreword Acknowledgments

IRRI is most grateful to the Australian Centre for International Agricultural Research for its financial contribution to the publication of this book. Many people contributed to the success of the international workshop upon which it is based. The workshop was co-hosted by the Central Research Institute for Food Crops, Indonesia. The local organizing committee included Andi Hasanuddin, Sunendar Kartaatmadja, Suprapto, Hasil Sembiring, and Mahyuddin Syam. The overall organizing and technical review committee included the following IRRI researchers: T.P. Tuong (chair), B.A.M. Bouman, S.P. Kam, S. Pandey, and L. Wade. L.L. Garcia ably handled international logistics for the workshop and Hekki provided technical and logistical support to the local committee. We thank D. Dawe, C. Edmonds, S. Fukai (Queensland University, Australia), T. George, K.L. Heong, S. Morin, and T. Paris for reviewing papers in this book.

Foreword ix Overview

Characterizing rainfed rice environments: an overview . . . 1 2 Singh et al Characterizing rainfed rice environments: an overview of the biophysical aspects

V.P. Singh, T.P. Tuong, and S.P. Kam

The literature reveals that rainfed rice environments have been characterized for various purposes at differing scales using a range of techniques. This chapter reviews the biophysical aspects of characterization undertaken in rainfed rice environments, more specifically at sites of the Rainfed Lowland Rice Research Consortium. It presents the status of the work done; provides an inventory of techniques applied; discusses scale, variability, and accuracy; pinpoints gaps in knowledge; and provides future directions for work from which these sites can benefit. The chapter brings out commonalities of characteristics across rainfed sites, provides insights into which issues new technological developments should focus on when addressing major limitations to production enhancement, and identifies opportunities for making future work more efficient and relevant to needs. Examples for each of these aspects are drawn from case studies in the rainfed rice regions, and specific cases where environmental characterization has made a significant impact on national systems are cited.

Rainfed rice environments are characterized for various purposes. Some of the predominant uses are prioritizing research on a broad scale, extrapolating technology, interpreting multilocation network research, and identifying recommendation zones. Each of these uses requires different degrees of classification subdivisions, from only a few at broad levels to many at specific levels, using a combination of parameters and available tools and techniques. Whatever the purpose, characterization in general is supposed to help increase the productivity of rainfed environments via a better understanding of their properties, particularly in rainfed areas that face a high degree of temporal variability and spatial heterogeneity. It is within this context that rainfed rice research sites, including the Rainfed Lowland Rice Research Consortium (RLRRC) sites, are characterized at different scales with corresponding degrees of detail. However, characterization alone cannot lead to the impact of technology without addressing aspects of

Characterizing rainfed rice environments: an overview . . . 3 development, for example, provisions for technology application, such as the availability of inputs, etc. This chapter focuses on the biophysical aspects of rainfed rice environments. It reviews some of the broad environmental classifications and systems of rice ecosystem analysis, and reviews work done at different sites and the tools and techniques applied. The chapter highlights the commonalities across rainfed sites, pinpoints gaps and methodological limitations, and identifies opportunities for making future work more efficient and relevant to needs.

Broad environmental classifications Garrity et al (1996) extensively reviewed the systems of environmental classification and posed a fundamental question: To what extent can rice research rely upon the several efforts at broad agroecological classification at the global level, specifically the climatic classification systems and the FAO agroecological zone studies? There are natural advantages in assessing (if this is feasible) the impact of these broad-scale classifications, since these efforts are generalized (relatively fewer classes) and widely known. Among the global climatic classifications, those of Koppen (1936), Thornthwaite (1948), Holdridge et al (1971), and Papadakis (1975) have had a wide currency. The Koppen system recognizes 13 tropical and subtropical climatic zones. The Thornthwaite system identifies 21 classes. These relate locations to general climates, but are broad. For example, Koppen’s class labeled AW is subhumid tropics with one season. Papadakis’s work led to a more complex classification system (more than 500 classes). Though moving toward comprehensiveness, this sacrificed simplicity. The FAO agroecological zone system (Kassam et al 1982) includes a broadly defined climate component based on temperature in the growing period and length of the growing period based on soil-water balance. The major objective of the FAO system was to assess the suitability of land for different crops (Higgins et al 1987). The major climates can be used independently or combined with growing-period zones. When the climate classification is combined with soils data, it yields a more comprehensive and complex global agroecological zone (AEZ). In addition to the global and continental level studies, the agroecological zone studies have been conducted in several countries, for example, Bangladesh (FAO 1988). A world data bank of the AEZ system is maintained at FAO headquarters in Rome. The Technical Advisory Committee (TAC 1990) of the Consultative Group on International Agricultural Re- search (CGIAR) adopted the FAO system of seven basic agroecological units for its analysis of CGIAR priorities.

Systems of rice ecosystem analysis The methods discussed above provide a range of flexibility in data requirement and aggregation and are widely used elsewhere. They have not been widely applied in characterizing and classifying rice environments, mainly because of the uniqueness

4 Singh et al of rice’s environmental situation compared with other major crops. Rice environments are dominated by surface flooding patterns. Therefore, all the rice classification systems recognize surface hydrology as the dominant delineating variable (Garrity 1984). In addition, a meaningful classification of rice environments must proceed independently of the commonly known global systems. The agroclimatic classification for rice and rice-based cropping systems that has been widely adopted (Oldeman 1980, Oldeman and Frere 1982) is based on the length of the rice-growing season—months in which surface flooding can be maintained (monthly rainfall of above 200 mm). National agroclimatic maps based on this system were derived for several countries, such as the Philippines, Bangladesh, and Indonesia (Oldeman 1980). Huke (1982) compiled maps that uniformly classified all the countries of South and Southeast Asia in this system. Agro-hydrological factors have been studied in only some cases: as land qualities in land evaluation for rice- cropping patterns by Tuong et al (1991) at the macro level and by Minh (1995) at the meso-micro level in the Mekong Delta. However, the information on the dynamics of surface hydrology is sparse in most of the studies. They also generally do not contain information on soil type and topography, which are essential land features and are strongly related to field hydrology as well as to surface flooding patterns to some extent. The International Terminology of Rice Growing Environments (IRRI 1984) established a standardized scheme of rice ecosystems, which subdivided the commonly accepted rice environments into a varying number of subecosystems, based on hydrological, temperature, and, in some cases, soil factors (Table 1). Five

Table 1. Terminology for rice-growing environments. Environment Characteristics Irrigated Irrigated, with favorable temperature Irrigated, low-temperature, tropical zone Irrigated, low-temperature, temperate zone Rainfed lowland Rainfed shallow, favorable Rainfed shallow, drought-prone Rainfed shallow, drought- and submergence-prone Rainfed shallow, submergence-prone Rainfed medium deep, waterlogged Deepwater Deep water Very deep water Upland Favorable upland with long growing season (LF) Favorable upland with short growing season (SF) Unfavorable upland with long growing season (LU) Unfavorable upland with short growing season (SU) Tidal wetlands Tidal wetlands with perennially fresh water Tidal wetlands with seasonally or perennially saline water Tidal wetlands with acid sulfate soils Tidal wetlands with peat soils

Source: IRRI (1984).

Characterizing rainfed rice environments: an overview . . . 5 subecosystems were recognized within the rainfed lowland and four within the upland. Subsequent efforts have attempted to sharpen the classes and provide better estimates of their overall extent. Four distinct levels of ecosystems analysis, in terms of geographic scope and mapping scale, were recognized later (IRRI 1992): mega, macro, meso, and micro. Table 2 gives the indicative objectives and activities associated with work at each of these levels. Many technical problems still arise in ecosystems analysis and they are due to the difficulties of working at different mapping scales and transferring information across scales. Table 3 presents some of the methods that have been commonly used in evaluating the parameters of rice ecosystems and Table 4 lists the parameters that have been used for characterization at different scales. These methods have a wide range and flexibility in their application and accordingly provide various outputs: from simple descriptions to semi- or fully quantitative measurements, and computer-simulated estimates. The techniques employed range from informal and formal interviews and surveys including reconnaissance, rapid rural appraisals, field visits, and farm/household surveys to simple accounting and bookkeeping, application of procedures for specific field and laboratory measurements, remotely sensed image and aerial photograph interpretation, applications of detailed simulation models of crop growth, and the use of geographic information systems (GIS). Examples of the tools used are structured and unstructured questionnaires, field books, specific instruments, and data recorders. However, the tools and techniques in characterization cannot be separated. They are very closely linked, as a tool is basically an instrument and the application of it is a technique. The parameters or factors in the context of characterization are those properties, obtained from primary and/or secondary sources, that can help describe sufficiently the unit (site/system) in terms of its properties, thereby enhancing the understanding of the system and serving as a diagnostic criterion to differentiate it into its subclasses. While the field surveys provide ground truth as well as other needed ancillary information, the advent of remote sensing and GIS has greatly enhanced characterization and mapping capabilities. However, these methods are not fixed for any given level of analysis and can also be used at other levels. For example, methods 1 and 2 in Table 3 can also be used at the micro and other levels.

Mega-level analysis When IRRI restructured its research programs (IRRI 1989a) to explicitly address the rice ecosystems, decision criteria for allocating funds on an ecosystem basis became more explicit. The rice area and production in each ecosystem were fundamental information in applying a resource allocation model. Aggregate data and maps on the rice area by cultural type have been standardized for about two decades (Huke 1982, Huke and Huke 1997) on the basis of judicious estimates. The accuracy of mega-level data on the amount of rice land in the four major ecosystems is still uncertain, except in a few countries, because the na-

6 Singh et al colation);

included

Example of

parameters/factors

geopolitical and policy

flooding and drought

zones; land forms, flooding, forms, land zones; and drought characteristics; drought and land use; cross-cultural and cross-cultural use; land community resource base, needs, and policy

farming systems; community resource base,

land type, hydrology (rain-

seepage, and per

varieties, and pest pressure; household resource base and needs

Selected; soil groups, land Selected; soil groups,

Common (many); soil units,

eas type, drought and flooding

Activities

units

and mapping key ecological parameters as mapping patterns; land use;

with homologous conditions, characteristics, rainfall; using selected factors as mapping units; identification of potential research sites needs, and actions

factors, transects, agroeco- surface water depth, system maps, and systems diagrams; selection of field cropping patterns, yields, experimental sites

Broad international-/national- Key (few); land for

esource as mapping units

oad spatial National- and regional-level Dominant; soil and climatic

Objectives

systems, establishing their relative importance for research prioritization

regional research planning using dominant factors and prioritization; r alloca-ion and broad extrapolation; indicative variability within ecosystems

environments, extrapolation and recommendation domains

ments, selection of represen- considering all prevalent fall, water supply sources, tative experimental sites for crop adaptation, extension recommendations

Understanding br

Selection of representative Regional- and zonal-level

analysis

systems areas for systems. Target ecosystems, differentiated

Scale of Unit of

1:5 million ecosystems distribution of different eco- level analysis; characterizing including hydrology,

1:25,000 systems design of on-farm experi- analysis; detailed study

<1,000,000 1:500,000 subeco- differentiation of subeco- analysis; detailed study of

<500,000 1:50,000 environments research sites in different analysis; delineating ar

These are indicative figures based on current mapping studies; adopted from IRRI (1993).

Table 2. Indicative basic objective and activities in ecosystems analysis at four different levels of analysis. Table Level of analysis Area (ha) mapping

Mega >1,000,000 1:500,000– Rice Identification and geographic

Macro >500,000 to 1:50,000– Rice

Meso >50,000 to 1:10,000– Rice Micro <5,000 1:5,000– Rice agroeco- Determining agroecosystems, Municipal- and village-level a

Characterizing rainfed rice environments: an overview . . . 7 Reference

et al (1993, 1999)

WARDA (1993) WARDA

Garrity and Agustin (1984)

e IRRI (1993)

e (1993) WARDA

Southeast Asia

districts now)

in eastern India

in Thailand

One district in India (40 mor

Mega Countries of South and

Mega Mekong Delta, Vietnam et al (1991) Tuong

Level of Geographical area

analysis

mation

owing

vations

etation and Micro Some research sites IRRI (1989b)

etation and Mega Entire Cambodia Garrity and Bruce (1992)

etation and Meso Mekong Delta Kam et al (2000)

etation and Meso Some sites in Côte d’Ivoir

etation and Meso Entire Côte d’Ivoire (1993) WARDA

Description

Index for different parameters of classification of upland rice-gr environment

Agro-hydrological parameters Satellite image interpr Satellite image interpr ground truth

Satellite image interpr ground truth; also use of available ancilliary information

Satellite image interpr ground truth Aerial photograph interpr ground truth for collecting infor on field experimental sites Household and village surveys, Micro More than 100 villages Lightfoot et al (1989), Singh

farmers, field observations and measurements, use of secondary data

Interview with extension agents and Macro Entire Côte d’Ivoire farmers, some field obser

●

● ● ●

●

method

based method ground truth

appraisal (RRA) interview with extension agents and

Adopted from Singh and Singh (1995).

Table 3. Methods commonly used in evaluating parameters of rice ecosystems at different levels. Table

Method

1. Parametric

2. Remote sensing- 3. Rapid rural a

8 Singh et al Table 4. Parameters used in some published rainfed rice environmental characterizations and analysis at different scales.

Scale of analysis/ characterization Geographical Environmental characteristics/ and rice area area parameters studied Reference covered

Mega South and ● Rainfall in 5-mo growing season Garrity et al (1986) (30 million ha) Southeast ● No. of months R>PE + 20% Asia ● Soil textural class ● Slope class ● Soil units (physical and chemical characteristics and constraints such as low CEC, high P fixation, alkalinity and salinity, Zn deficiency; organic, acid sulfate, and shallow soils Mega Eastern India ● Irrigation IRRI (1993) (26.8 million ● Water depth ha) ● Land form Macro (–) Bangladesh ● Various soil physical and chemical Ahmed et al (1992) properties ● Land use ● Rainfall FAO (1988) ● Length of growing season ● Crop establishment techniques ● Water quality Macro Eastern India ● Rice yield reductions due to major Widawsky and O’Toole (10.6 million ha) biophysical stresses (1990), IRRI (1993) ● Drought ● Flooding pattern ● Water balance ● Length of growing season ● Rice yield Macro Côte d’Ivoire ● Rainfall WARDA (1992) (329,000 ha) ● Toposequence position ● Tillage method ● Rice variety, sowing, and intercropping technique ● Land tenure ● Decision-making by gender ● Production objective Meso Bahraich ● Land form Singh and Pathak (200,000 ha) District, India ● Physiography and slope (1990), IRRI ● Soil texture (1992, 1996) ● Soil fertility ● Water sources ● Irrigation

continued on next page

Characterizing rainfed rice environments: an overview . . . 9 Table 4 continued.

Scale of analysis/ characterization Geographical Environmental characteristics/ and rice area area parameters studied Reference covered

● Groundwater table ● Spatial extent under different environments ● Water regime ● Floods ● Droughts ● Rice cultivation practices ● Rice yields ● Fertilizer and other inputs ● Problems and suggestions Meso Solana, Cagayan, ● Rainfall IRRI (1987, 1990), (176,000 ha) Philippines ● Land form Garrity et al (1992) ● Hydrology ● Drainage ● Soil properties ● Cropping pattern ● Rice-farming technology Meso Faizabad ● Rice area IRRI (1993) (181,000 ha) District, India ● Rainfall pattern ● Irrigation ● Land use ● Soil constraints ● Submergence ● Drought ● Soil texture and fertility ● Groundwater table ● Rice area under different subecosystems Meso/micro Masodha ● Long-term trend in rice area IRRI (1993) (21,000 ha) Block, Faizabad ● Rice subecosystems, extent and District, India spatial distribution ● Physiography ● Rainfall pattern, flood and drought occurrence ● Soil texture and area coverage ● Soil constraints and their respective area coverage ● Drought-affected area ● Irrigation source, supply, and irrigated area

continued on next page

10 Singh et al Table 4 continued.

Scale of analysis/ characterization Geographical Environmental characteristics/ and rice area area parameters studied Reference covered

● Available groundwater potential or development ● Major environmental constraints to production Micro (85 ha) Chandpur ● Land types and their area IRRI (1993) Village, Masodha ● Soil texture and fertility Block, Faizabad in each land type District, India ● Groundwater table in each land type ● Flooding ● Crops grown in each land type ● Farm-related and other enterprises ● Farmers’ problem identification and its influence on rice productivity ● Household number and social group composition Micro (385 ha) Four villages ● Land type and use pattern IRRI (1991), Singh in Hazaribagh ● Varieties grown et al (1994) District, India ● Rice yield by land-type varieties ● Social class ● Land sharing and division ● Credit availability ● Water resources and irrigation ● Soil characteristics ● Pest, disease, and weed pressure ● Farming system ● System problems Micro (–) Claveria research ● Land form Magbanua and site, Northern ● Soils Garrity (1988) Mindanao, ● Slope Philippines ● Land use ● Farm size and land tenure ● Infrastructure ● Climate and long-term rainfall pattern ● Grouping of landholding size by social class ● Rice crop area by social class and landholding size ● Rice yields by social class and landholding size ● Fertilizer and other input use by farm household

continued on next page

Characterizing rainfed rice environments: an overview . . . 11 Table 4 continued.

Scale of analysis/ characterization Geographical Environmental characteristics/ and rice area area parameters studied Reference covered

Micro (–) Rainfed lowland ● Land form IRRI (1989b) rice research ● Physiography site, Khukhar ● Slope Village, Thailand ● Land use ● Soil fertility and other characteristics ● Detailed field hydrology ● Groundwater table ● Rice yields in different soil and land combinations Mega South and ● Climate-rainfall, temperature, and Garrity (1984) (11.6 million Southeast Asia growing season length ha) ● Slope, soil texture, soil groups, and inherent fertility status Macro Eastern India ● Irrigation extent IRRI (1993) (26.8 million ● Land form ha) ● Fertility-related constraints Macro Côte d’Ivoire ● Rainfall WARDA (1992) (329,000 ha) ● Toposequence position ● Tillage method ● Rice variety, sowing and intercropping technique ● Land tenure ● Decision-making by gender ● Production objective

tional-level statistics on various types of nonirrigated rice lands are generally unavailable. The work of Huke yielded standard maps of the regional allocations of rice land by ecosystem. The maps provided the basis for more comprehensive mega-level geographic databases, for focusing characterization on selected micro-regions, and for classifying them into subecosystems. The upland ecosystem geographic database was the initial product (Garrity 1984, Jones and Garrity 1986). This database contained data on several agroclimatic and soil parameters for each of the approximately 4,000 upland locations on the Huke maps for South and Southeast Asia. The sites were classified consistently according to a two-factor upland rice environmental classification based on the length of the growing season and inherent soil fertility constraints, and also a three-factor system that included an estimate of seasonal moisture sufficiency. The two-factor classification conforms with the four broad upland subecosystems specified in the International Terminology of Rice Environments (IRRI 1984). The three-factor classification recognizes 12 major classes and, at a more de-

12 Singh et al tailed level, that is, by using three categories of growing-season length, six categories of soil fertility constraints, and four categories of water balances, a total of 72 classes. The rainfed lowland rice ecosystem database (Garrity et al 1986) was compiled using a similar methodology. Approximately 6,300 rainfed lowland sites were classified in a three-factor environmental classification that included growing-season length, water balance, and soil constraints as delimiters (Garrity 1984). The other mega-level characterization study is of eastern India (IRRI 1993), with a similar basis of assessment. The Asian rice-land soil constraints database covers all rice land in South and Southeast Asia, including irrigated and deepwater area (IRRI 1987). This database includes data from the FAO Soil Maps of the World (FAO 1977, 1979), with soil constraints interpreted according to the fertility capability classification (Buol and Cuoto 1981). These databases were intended to be useful in research prioritization. Their impact has been significant in terms of a major shift in upland breeding and agronomic research in the early 1980s, from recent volcanic soil to acid upland soils, and from flat land to sloping land. The rainfed lowland database had a lesser impact initially most likely because it did not include data on surface water depth regime, particularly the frequency and duration of crop submergence. Although the database included length of the growing season and a crude water balance classification, sufficient data did not exist to classify the surface water accumulation dynamics at the micro-region level. In its absence, there was no way to definitely classify and map the rainfed lowlands into the five subecosystems. However, with the development of remote-sensing and GIS-based methodology (Singh 1987, 1988), the assessment of flood as well as drought became easily possible. Further development of this methodology (Singh and Singh 1996), which uses the crop vegetation index as a comprehensive reflection of prevailing conditions, including hydrological ones, in the crop, allowed taking full account of seasonally variable environmental conditions and, thus, the reliable classification and mapping of rainfed rice lowlands into various subecosystems. This methodology, coupled with the rainfed lowland database, is now being used extensively throughout eastern India to analyze and map about 21 million ha of rainfed rice into subecosystems, and is presented in the next sections.

Macro-level analysis The generalized nature and small scale of mega-level databases strongly limit their application beyond regional issues. Many nations are interested in research prioritization and extrapolation that typically require much more detailed information, at least at the macro level (national/state level for large states). In several countries, some remarkably comprehensive and useful data sets have been developed (FAO 1988, Widawsky and O’Toole 1990, IRRI 1993, WARDA 1992). The challenge is to make better use of these data. An excellent example is the case of Bangladesh (Ahmed et al 1992), where extensive soil and land-use data sets and maps were developed by the Soil Resources and Development Institute in collaboration with FAO (FAO 1988).

Characterizing rainfed rice environments: an overview . . . 13 This included standard countrywide data on surface flooding regime, data which are rare in most countries. The entire rainfed rice area in eastern India, covering the states of Assam, Bihar, and West Bengal and eastern parts of Uttar Pradesh and Madhya Pradesh, has been delineated in different rice ecosystems using the information on rice cultural types coupled with growing-season length, a field water balance, and soil quality (Singh and Sastri 1998, IRRI 1993). An agroclimatic analysis of this region includes databases on major climatic variables, a full account of the climatic water balance, and a detailed agronomic interpretation of planting schedules, varieties to be used, and other crop management practices according to moisture availability. Analysis of rice areas in different eco-classes has provided more reliable estimates of their extent and assisted the government in setting priorities. An agroclimatic atlas of eastern India depicting this analysis and detailing the climatic water balance for 45 stations in the region has been prepared (Singh et al 1999) as a reference for research and development agencies. Other such efforts at the regional level, primarily looking into hydrological aspects, are those of Tuong et al (1991) and Kam et al (2000).

Meso-level analysis A major positive trend in many countries is the regionalization of research across subnational border linkages, for example, across districts/states in India. National and subnational governmental efforts have enabled the development of institutions that can identify the unique problems and research priorities of the specific areas where they are located. These institutions seek methods to establish priorities that are suited to smaller geographic areas and larger mapping scales (Garrity et al 1996). The process is more suited to direct feedback from extension personnel and on-farm adop- tive research. Meso-level analysis is typically associated with a cultivated area of about 100,000 hectares, using a mapping scale of 1:25,000 to 1:100,000. Analysis at this scale can efficiently identify and delineate rice ecosystems and subecosystems in terms of surface hydrology, land form, and soil classes. Associated with each rice ecosystem are the flood and drought frequencies and duration, prevalent cropping patterns, and crop management practices. An example of a useful meso-level analysis was that conducted for Baharaich District, Uttar Pradesh, India, to identify the problems causing low rice yields and to prepare the priority research agenda at a district level in eastern India (Singh and Pathak 1990). The analysis determined the ecological variability of the district in terms of hydrology (rainfall pattern, water table depth, irrigation sources, and drainage), land form and slope, length of the growing season, frequency and duration of flood and drought, and major insect, disease, and weed pressures. After characterization, the factors were combined to identify and delineate the area into homologous zones. Cropping pattern, varieties used, crop management practices and input use, and socioeconomic conditions were superimposed separately on

14 Singh et al each of the rice ecosystem maps. The rice ecosystems were then prioritized for research on the basis of extent of area, number of affected households, and potential possibilities of research success. Similar characterization was done in representative districts in each of the eastern India states (IRRI 1992, 1993, 1996) and was used to develop a comprehensive rice research plan for eastern India. Faced with the hydrological complexities in the rainfed situations, Singh and Singh (1996) developed a remote-sensing and GIS-based methodology that could reliably delineate different rice ecosystems and their subecosystems, particularly by their hydrology—drought-prone, submergence-prone, etc. Using this methodology, 40 of the 93 rainfed rice-growing districts in eastern India have been characterized and mapped into the principal rainfed lowland subecosystems (unpublished) by teams of state agricultural universities, Indian Council for Agricultural Research centers, state departments of agriculture, remote-sensing application centers, nongovernment organizations, and groups of local farmers. This analysis is already being used extensively in regional research planning and to develop and delineate application domains of the promising technologies in eastern India (Singh and Sastri 1998). Ex- amples of these efforts are Sastri and Singh (this volume) and Borkakati et al (this volume). IRRI and the Department of Agriculture Regional Office for the Cagayan Val- ley, Philippines (Region II), developed a meso-level classification of the valley’s complex mosaic of rainfed rice lands (IRRI 1987, 1990, Garrity et al 1992). They explored the utility of a computerized geographic database correlated with village-level maps of rainfed rice land types. The information was packaged as a field manual for extension personnel. Six rainfed rice subecosystems were recognized on a hydrological basis. They were explicitly correlated with a range of associated information to specify their identification and the technology associated with them. The data on rice area and the yield constraints associated with each rainfed rice land type have facilitated regional rice research efforts, particularly the relative emphasis given to applied and adaptive research among land types (Garrity et al 1992, IRRI 1991).

Micro-level analysis Agroecosystem analysis has become very popular in micro-level prioritization (Conway 1986, KEPAS 1985). Micro-level analysis studies have been done in several cases (IRRI 1989b, 1991, 1993, Singh et al 1994, Magbanua and Garrity 1988). This level of analysis has been used extensively in the rainfed regions of eastern India, covering the states of Uttar Pradesh, Madhya Pradesh, Bihar, Assam, West Bengal, and Orissa, to set research priorities within and among dominant rice-farming systems. Various techniques of agroecosystem analysis such as site descriptions, problem diagnosis, farming systems analysis, and rapid rural appraisal methodology were extensively employed (Lightfoot et al 1990). The methodology involved a two- tier training program for researchers on the methodology for setting research priorities by agroecosystem analysis with farmer participation. The analysis was carried out by 15 research centers in the region covering upland, rainfed lowland, and

Characterizing rainfed rice environments: an overview . . . 15 deepwater rice ecosystems. The research diagnosis and prioritization at this level were conducted by multidisciplinary teams in the respective centers, with regular involvement and interaction from groups of farmers (IRRI 1989b). The micro-level agroecosystem analyses (100 locations in eastern India) included detailed characterization and classification of the static and variable factors that differentiate agroecosystems by soils, hydrology, farming practices, and socioeconomic conditions (IRRI 1989b, 1993, Singh et al 1993). The sites were mapped on the scale of 1:2,000–1:5,000. At all sites, the static factors studied were land types, landholding size, source of water supply, and soil properties. The dynamic factors were land use; field water depth; rainfall and cropping patterns; crop yields; varieties and management practices; insects, diseases, and weeds; production costs and returns; labor supply pattern; prices, assets, and income distribution; and demography by social class. The geographic area was zoned into agroecosystems and the problems and opportunities elucidated in each major agroecosystem. Among the different agroecosystems, the highest priority was given to the one with the largest extent. The research problems were then prioritized on the basis of their physical extent within agroecosystems (coverage), number of affected households, complexity of the problem, severity of the problem (crop loss estimates), frequency of problem occurrence, and the importance of the affected enterprise in the farming system (IRRI 1989b). This type of analysis has greatly facilitated on-farm research for the development of problem-solving technologies, the selection of representative research sites, and the actual adoption/impact of technologies in the region (Singh and Singh 2000). These methodologies are also increasingly being used in Thailand, Bangladesh, Indonesia, and Bhutan, as is reflected in the requests for training on them by the respective countries and their research programs developed through these methods. The characterization studies carried out at the consortium (RLRRC) sites are mostly diagnostic in nature and in detail (micro, field level, Table 5). Consortium sites serve as the “hot spots” for research on certain thematic issues, such as low-P soils, and their reports show that they lack the characterization of other existing conditions. This is not site characterization but the characterization of a certain theme, such as submergence or drought, at the consortium site. The characterization of only certain themes without consideration of other existing conditions at the research site limits the utility of research results because the technological output cannot be extrapolated without a complete description of the research as well as the target site. This underutility points out the need for a complete characterization of a site that will include certain major themes along with other prevailing conditions. Also, a higher level (e.g., regional) analysis is carried out only at some consortium sites, such as Faizabad and Cuttack (Table 5). However, higher level analysis has been scheduled to be carried out at other sites, such as Jorhat, Assam, India; the Mekong Delta and Red River Basin, Vietnam; and northeastern Thailand, which are affiliated with the RLRRC key sites. It is essential to foster functional linkages between key and affiliated sites to ensure effective interchange and to enhance the utility of research.

16 Singh et al needs/gaps

continued on next page

by farmers

adopted

developed technologies

and adopted distribution

technology

developed and

Cropping systems

applied

and deep-

P-efficient

techniques developed and

techniques and adopted

rooted genotypes

applied

sequence develop

toposequence assessment technology

basin

block drain- monitoring/ technologies promising

topo- alternatives, developed of spatial

Main Methodology

needs

availability

and floodwater drainage techniques quality monitoring

soil physical and hydraulic properties

duration

drought availability period block screening

submergence; availability period cropping and floodwater age basin screening system; quality monitoring response of rice genotypes

P-deficient, chemical and acidic properties, soil; drought drought, and

Rajshahi Drought Determining water Field/crop Moisture availability Cropping systems

Narsindhi, Seasonal Flooding patterns, Drainage basin Kishore Ganj flooding depth, and (flood-prone systems)

Cuttack Submergence/ Determining moisture Watershed/ Flood quality and

Faizabad Drought and Determining moisture Watershed/ balance Water On-farm Promotion of

Hazaribagh Shallow, Analysis of soil Field/cross P-management Rice varieties Regional analysis

●

Table 5. Characterization at some consortia sites. Table

Country Site Main issues characterization Scale presently developed/ Impact Unfilled

Bangladesh

India

Characterizing rainfed rice environments: an overview . . . 17 P buildup/decline

effects

needs/gaps

continued on next page

Impact Unfilled

P management

and adopted

strategies for to be continued to

adopted

Economical P Monitoring of P

Economical General soil

applied

Methodology

of soil-P status

surface-applied subsoil acidity physical and

of cumulative uplands better understand

measurement

crop planning developed technologies

lime and liming amelioration chemical effect on subsoil for uplands measurements root profile

experiment; study

(PDSS); GIS tions and policy

and crop responses

phosphorus management dynamics needs

applications

characterization using PDSS

effects of P

Long-term

Soil sample and

Study of leaching of

applied

basin

watershed monitoring and technologies promising

land slope decision

block and screening technique promising

drainage techniques developed and technologies

Field

District/ On-farm testing of Site-specific Newly proposed

Main

needs

dynamics; soil

depth, and duration

availability period

cations and anions; water movement, soil acidity

physical and chemical characterization

limitation and regional/ phosphorus adaptation of range of potential crop responses to class support system recommenda- P application in acid uplands of Indonesia

hydraulic properties

Extent of P

Soil physical and Field

in uplands leaching of

in acid upland term soil-P soils

Low soil-P supply Characterizing long-

Site Main issues characterization Scale presently developed/

Jorhat Submergence Flooding pattern, Watershed/ Flood quality Cropping systems Promotion of

Raipur Drought Determining moisture Field/farmer/ balance Water On-farm Promotion of

Sitiung Subsoil acidity Characterizing Field

●

Table 5 continued. Table

Country

India

Indonesia

18 Singh et al distributions

water balance:

applicability

and spatial

needs/gaps

continued on next page

Impact Unfilled

by farmers regional

released

and adapted characteristics,

nutrient management improved

dynamics and nutrient

Dry-seeded rice Crop Need for short- Regional data on

Knowledge on Regional data on

New varieties

water enhanced; availability and

applied

E and

Methodology

water balance variety monitoring and stressed their dynamics

on-farm reservoirs developed micro-catchment

G × nutrient experiment farmers’ dynamics

table depths; duration water table and

simulation

economics of technologies hydraulic data,

experiments; nutrient toposequence

Water balance andWater On-farm Soil physical and

Monitoring water

Long-term

applied

Field

Main

needs

different toposequence

landscape and soil properties for the adoption of on-farm reservoirs

and its spatial distribution

Characterizing Field

Nutrient dynamics

stable yield balance, taking of the second into account water rainfed rice table depth and

and low input and adoption lead to nutrient mining, especially K and P

Low and un- Multiscale water

High intensity screening Variety Field

Site Main issues characterization Scale presently developed/

Jakenan High population

●

Table 5 continued. Table

Country

Indonesia

Characterizing rainfed rice environments: an overview . . . 19 different

population in

gical dynamics

agrohydrolo-

toposequence,

needs/gaps

Weed dynamics, Weed

continued on next page

problem; of areas

management adaptation

recommendations

reduce amount prone to of fertilizer contamination applied

dynamics enhanced and screening techniques improved

Better nutrient Wide-scale

applied

monitoring and water quality identification

techniques nutrient to other areas

study

cropping sequence

modeling

and productivity

Nutrient balance

Weed surveyWeed

Groundwater Awareness of Regional

applied

Field

Field Screening Knowledge of Extrapolation

Main Methodology

needs

patterns

Nutrient dynamics

infestation; in different high pest toposequence infestation and cropping

sustainability in highly intensified and diversified cropping systems; groundwater Groundwater quality sustainability and quantity

sustainability market structure

Profuse weed dynamics Weed Field

Economic Farmer income and Farm Farm survey of

Batac Soil nutrient Nutrient dynamics Field

Claveria/ Acidic, P-deficient Cavinti soil

●

Table 5 continued. Table

Country Site Main issues characterization Scale presently developed/ Impact Unfilled

Philippines

20 Singh et al of drought

and spatial distributions

and dynamics

availability

distribution

their dynamics

needs/gaps

continued on next page

development

genotypes areas

developed

germplasm of temporal

and adapted

by farmers

Knowledge Regional data on

Potential Regional analysis

Agrohydrology of Regional data

water

applied

mapping

simulation; G experiments

experiment generated nutrient

secondary

balance study;

climatic data; identified and spatial in situ variety testing

Water balanceWater Suitable Analysis of target

Nutrient by water

Analysis of

Field

Field and Monitoring water

Main Scale presently Methodology

needs

and its spatial distribution

water table depth sequence monitoring and identified balance: and different toposequence

moisture availability

availability

periods

Multiscale water

season, quality analysis province provincial-level cropping under drought, poor of drought soil quality, low cropping intensity

yield due to balance, taking across table depths; dry-seeded on water table drought into account topo- water balance rice system and water

Low nutrient Nutrient dynamics

Iloilo Short rainy Characterize soil Field/farm/ On-farm and Double rice

Tarlac Low and unstable

Ubon Drought balance, Water Field

Upland Red Drought Determining water Field River Basin (Vietnam)

●

Thailand

Table 5 continued. Table

Country Site Main issues characterization applied developed/ Impact Unfilled

Philippines

Vietnam

Characterizing rainfed rice environments: an overview . . . 21 useful from

agronomic point of view

needs/gaps

Models for long-

recommenda- term policy

recommenda- term policy tions made analysis

tions made analysis

Better nutrient Regional soil map

applied

management management that can be experiments

Nutrient

Farm, Survey; econometric Policy

Main Scale presently Methodology

needs

food supply and household modeling

chemical properties

demand

Factors affecting Farm, Survey; econometric Policy

pressure tenure policies household modeling

upland

Nutrition (low P) Quantifying soil Field

High population Land-use and land-

Food security for

Table 5 continued. Table

Country Site Main issues characterization applied developed/ Impact Unfilled

22 Singh et al General discussion The complexities in biophysical characterization and classification of rainfed environments stem from the existence of different ecosystems and/or their classes in a continuum, unclearly defined boundaries between them, overlapping or transitional zones between the subsystems, and the inability of the methodologies and parameters used to discriminate among them. In many places, rice ecosystems and/or their classes co-exist, even at short physical distances, sometimes even within a farm. Overlaps or transitional zones also occur between the subsystems, which shift from one class to another from season to season. Recognizing these facts can be helpful in dealing with the characterization methodology and technologies for such areas. This also brings out the limitation of zoning with fixed boundaries, particularly at more detailed resolution and for technology targeting. The characterization done either by using diverse parameters, land type in some cases and other parameters in other cases, such as upland and lowland ecosystems by land type and deepwater ecosystems by water depth, or by using related parameters, but exclusive of each other, has posed difficulties in the synthesis of interpretation in an integrated manner. This is particularly relevant where the effect of prevailing environmental conditions is to be related to crop performance and ways and means are to be developed for improving it. Otherwise, the results of characterization would reflect an “ex post facto”) scenario wherein no mid-term corrective measures could be devised or applied. Therefore, we suggest that different systems be characterized by using similar and related parameters. This should include a reflection of crop conditions and synthesis of the interpretation in a comprehensive manner. In which eco-category any given geographic location would be under when environmental conditions are variable often can’t be predicted reliably by using conventional analytical approaches. Because of this weakness, the remote-sensing and GIS-based methodology (Singh and Singh 1996), which uses the crop vegetation index (a comprehensive reflection of prevailing conditions and their effect on the crop) as a major criterion, shows good promise and may be pursued further for the reliable delineation of an area in specific ecologies under variable conditions from season to season. The availability of optical remote-sensing data may, however, be a limitation in tropical areas during the monsoon season. Radar-based sensory data are helpful in overcoming this limitation and they are increasingly becoming available for such areas. The above cases show how the methodologies evolved over time and how current work, which is site-specific, has benefited from past work. But, in doing so, issues emerge related to new perspectives, new demands, and new tools; scale, variability, and accuracy; parameters that are often overlooked at different levels; and outsourcing skills, data sharing, and partnership. These issues are related to the fundamentals of characterization: 1. It is a sequential and in some cases multientity (institution/agency) activity wherein the participating agencies may have different mandates or different

Characterizing rainfed rice environments: an overview . . . 23 levels of involvement. For example, United Nations agencies such as FAO may have a global mandate, whereas the CGIAR centers may have a commodity-specific mandate or an ecoregional mandate, and the country/state/ provincial organizations may have a national or subnational mandate. 2. It can be an activity that is done to fulfill an objective, such as when technology development is an objective and characterization is a means for achieving that objective. Impact, of course, is an outcome of the application of technology for which the agroecological requirements of the technology should be known and conditions suitable for its application identified. The data requirement, techniques and tools used, and responsibilities for doing activities will mainly depend on the above two considerations. Other issues are discussed below.

Shifts in emphasis As has been noted in Tables 4 and 5, the perspectives of characterization have shifted. One such shift is from broad or general uses of characterization to detailed and specific uses through sharper tools and techniques, such as from developing broadly adapted plants to developing plants for target environments. The other shift has been in the ways characterizations are used. Characterization can be used for technology extrapolation and recommendation domains and as a tool for diagnostics in technology development. It is in this context when scaling up or scaling down the characterization becomes important, that is, interpreting and interrelating the information across scales. There is a need to bridge these two perspectives. The operative support scale, particularly for technology targeting and development, would be the scale at which (1) the technology will have an impact and (2) common interest groups (users, local administrative units) can effectively take actions.

Data sourcing and sharing This is a most sensitive and important issue, particularly if the demand is high and urgent. It is also voluminous and involves sensitive information, such as topo sheets, national borders involving national/geopolitical sensitivity, etc. Many national agricultural research systems (NARS) are reluctant to give their data to outside agencies; however, they are willing to share the processed information if data processing is done in their own places. The issue of data sharing is not confined to NARS versus international agencies. In many countries, institutional barriers constrain data sharing even within the country, but it is easier to resolve this among agencies within a country by involving higher authorities. In both cases, a discussion among the partners from the beginning, clearly outlining the benefits to all involved, and devising an agreed-upon procedure to follow may be quite useful in overcoming this problem.

Scale, accuracy, and integration ● Inadequacies in techniques and structure of the classification systems. In the literature, there appears to be no specified classification system or criteria for different levels of analysis, except that of IRRI (1984, 1993). There-

24 Singh et al Factors: Biophysical Biophysical Biophysical Biophysical Biophysical Biophysical and (soil, pedon, (soil units, (soil units, (soil groups, (soil and socioeconomic water and water and land, land, hydrology, climate zones, (same as for micro- micro- hydrology, climate, land, hydrology, agroclimatic climate) climate climate) vegetation) vegetation) zone level and Economic Socioeconomic Socioeconomic Socioeconomic geopolitical) (household (household (community (cross-cultural, resource resource base resource base, community base) and needs) needs, and resource base, actions) needs, and policy)

Levels: Soil level Field level Farm level Watershed level Agroclimate Agroecological (pedon, pot, zone level zone/ experimental ecoregional plot) level

Systems: Soil-plant Field-crop Agro- Agroecosystems Agroclimatic- Agroeco- system system systems sociocultural logical- (agro, agrosystems sociopolitical (farm- forest, agro-aqua, systems household nonfarm and (multiwater- and other off-farm sheds, agro, (geo, agro, related enterprises) forest, socio- nonagro, enterprises) cultural, etc.) policy, etc.)

Scale: Micro Micro Micro Micro Meso Macro Mega units: <1–20 m2 0.1– 0.5– <5,000 ha >50,000 ha >50,000 ha >1,000,000 ha 0.5 ha 5.0 ha (Micro Meso) (Meso Macro) (Macro Mega) (5,000–<50,000 ha) (>50,000–<500,000 ha) (>500,000– <1,000,000 ha)

Focus: Temporal Temporal Spatial and Spatial and Spatial and Spatial and temporal temporal temporal temporal

Fig. 1. Relative features of various levels of agrosystems analysis.

fore, any site analyses, even at a particular level, may be carried out using different sets of parameters on account of their relative importance and researchers’ access to the databases. Although this provides much flexibility to researchers, it runs the risk of resulting in differential classification, which could contribute to inappropriate research and development planning for the ecology represented by the site. This inadequacy points out the need to follow a well-defined set of classification criteria that operate at well-defined spatial scales. We suggest that such criteria be developed through a “workshop-mode” agreement. A framework developed, from among others, by Garrity et al (1986, 1996), IRRI (1984, 1993), and Jones and Garrity (1986) is provided for consideration in Figure 1 and Table 2. Such a classification, which uses multiple criteria in a hierarchical manner, reflecting the system’s properties, has been found quite useful and is suggested for use in the future (Fig. 1). ● Lacking classification criteria and parameters for each level. In spite of the uniqueness of rice’s environmental situation (surface hydrology), which depends on land characteristics, the land properties (parameters) used in rice

Characterizing rainfed rice environments: an overview . . . 25 ecosystem classifications are diverse. Surface hydrology and land characteristics have often been used exclusive of each other. The lack of specific criteria available in the classification systems has also hindered the reliable differentiation of the systems. For example, by using only drought occurrence as a criterion, Ubon (Thailand), Tarlac (Phil- ippines), and Raipur (Madhya Pradesh, India) are classified only as drought- prone rainfed lowlands, despite the vast difference in their biophysical characteristics (soil type, rainfall, amount and pattern of water retention, etc.). Such a classification reflects only the hydrological effect and assumes that its generic causes are the same across sites, or does not take into account the causes at all. Because of this, all the areas classified in a particular ecology appear to be homogeneous in all respects across locations, which is far from reality. This also indicates (proposes) the application of the same remedial measures to overcome the stress across locations. This issue mostly relates to aspects of scale and the spatial heterogeneity that exists at every level. The issue here really is of the level of understanding that is possible and needed. A better understanding is always more helpful and can be achieved only through detailed analysis. Therefore, the analysis of the spatial pattern of both the nature of occurrence of stress (e.g., drought in early, mid, and late season or at different crop growth stages) and severity of stress (e.g., degree of drought and its temporal feature and effect on the crop) should be included in the characterization, which can provide more meaningful interpretations, especially on management aspects, than only classification of the site in an ecological class. ● Unrelated criteria/parameters used in characterization at different levels. Even within a location/site, the parameters used at different levels are not only different but also generically diverse. In such cases, the analysis at different levels cannot be integrated. Table 4 shows an example of parameter diversity used at different levels. The information in this table also shows that, in certain types of analysis, there is no direct relationship of any parameter used in one level to the parameters used in the next level. Therefore, the use of related parameters becomes necessary if the analysis at different levels is to be integrated. ● Exclusiveness of the levels of analysis. Unless the analysis at different levels is done in a continuum, that is, within a given area, ecosystem, or agroecological zone, and by using related (common) parameters, its integration and synthesis are extremely difficult. ● Unspecified definitions of rice environmental classes below the subecosystem level. There is no classification subdivision below the rice subecosystem level in any of the classification systems. Hence, different workers use different terminologies, such as land management units, production units, rice environments, land-use units, etc. This is not a problem, however, if there are equivalents and these can be specified. Also, from the review in this chapter, it has been noted that, at the higher levels of analysis, such as mega

26 Singh et al and macro scales, rice environments are generally classified as ecosystems and subecosystems, which are then characterized in terms of the prevailing biophysical conditions using the key parameters and at the lower scales. For example, at the micro level, the analysis is simply a more detailed description of the same, or similar, biophysical parameters. Such an analysis con- veys the same output (more detailed interpretation at the micro level) when, in fact, at different levels, it is expected to fulfill different objectives.

Some important parameters that are often overlooked The following parameters are often overlooked in characterization. ● Groundwater table information. The groundwater information needed to determine the nature of water fluctuation, to interpret the duration and severity of drought, and to understand the groundwater contribution by capillary rise and its simulation is missing in most of the characterization studies reviewed in this chapter. This is probably because of the difficulty in gathering the information, which, by its own nature, is highly variable over time and heterogeneous over space. However, rice yields are very sensitive to groundwater table depth, especially when it fluctuates within 1 m from the soil surface (Wopereis 1993). Exclusion of this term may result in a 30% underestimation of yields at 4 t ha–1 and a 90% underestimation at 1 t ha–1 (Bolton and Zandstra 1981). ● Surface hydrology. The information on sustained surface water depth, which is highly dynamic, is also sparse in most of the studies reviewed in this chapter, probably for the same reason: the difficulty in gathering this information. Rice yields are also sensitive to surface flooding if it occurs at the active tillering stage and the surface water depth patterns determine, to a greater extent, both the nature and severity of effects of hydrological stress on the crop and crop management practices. ● Seepage and percolation (S&P) rates. These are the main components of water balance and they strongly influence the presence or disappearance of surface water. In drought-prone areas, one of the reasons for frequent water stress is high S&P rates, and this affects yields significantly. Using crop simulation, Fukai et al (1995) estimated that a 2 mm d–1 reduction in S&P, from 6 to 4 mm d–1, would increase rice yields by more than 60% in Thai- land. ● Soil fertility mapping. Mapping units delineated in soil maps show polygons of soil type/series and are based on soil genesis and classification. But existing soil maps explain only 0–27% of the variances for the soil fertility parameters that directly influence yield (Oberthur et al 1995). To be more relevant to increasing rice yields, technology development should focus on parameters that may limit yield (macro- and micronutrients, cation exchange capacity, texture). ● Biotic stress profile. Biotic stresses (weeds, insect pests, diseases) are as important in rainfed systems as in the irrigated one. Savary et al (1997) showed

Characterizing rainfed rice environments: an overview . . . 27 clearly that they are different in the two systems. Therefore, without understanding biotic stresses, technology development may be addressing less relevant problems or may require modifications.

Conclusions Characterization of rainfed rice environments for specific objectives is very useful because it enhances research prioritization and technology development, delivery, and impact. Characterization for the sake of characterizing has limited value. Charac- terization without linkages to its higher or lower hierarchies also has limited value. Agronomic interpretations and developing relationships tremendously increase the value of characterization and mapping. Characterization can be efficiently accomplished by first determining an objective based on requirements, inventorying what is available and making use of it, and then embarking on fresh data collection on new items when necessary. Adding information on the critical parameters that have been missing, such as hydrology, will add value to the already accomplished characterizations. Similarly, there is a need to establish linkages over scales, disciplines, and institutions. Functional collaboration among agencies seems to be an essential prerequisite. However, different institutional interests and sensitivities may be involved in data sharing, which one needs to be aware of in the collaborative efforts. A discussion on this among the partners from the beginning may be quite useful and a clear outline of the responsibilities of and benefits to all involved is expected to enhance collaboration to serve the respective interests. Several mechanisms exist, such as the Rainfed Lowland Rice Research Consor- tium and other consortia, and network projects, for exploring these opportunities to further enhance characterization and use it for productivity gains.

References Ahmed NU, Magor NP, Islam S, Khan AH, Naseem SB. 1992. Extrapolation and testing of technologies in Bangladesh. In: Proceedings of the 1990 Planning Workshop on Ecosys- tems Analysis for Extrapolation of Agricultural Technologies. Los Baños (Philippines): International Rice Research Institute. p 26-59. Bolton F, Zandstra HG. 1981. A soil moisture based yield model of wetland rainfed rice. IRRI Research Paper Series 62. Los Baños (Philippines): International Rice Research Insti- tute. Buol SW, Cuoto W. 1981. Soil-fertility-capability assessment for use in the humid tropics. In: Greenland DJ, editor. Characterization of soils. Oxford (UK): Clarendon Press. p 254- 261. Conway G. 1986. Agroecosystem analysis for research development. Bangkok (Thailand): Winrock International. FAO (Food and Agriculture Organization of the United Nations). 1977. FAO-UNESCO soil map of the world. Vol. VII. South Asia. Paris (France): UNESCO. FAO (Food and Agriculture Organization of the United Nations). 1979. FAO-UNESCO soil map of the world. Vol IX. Southeast Asia. Paris (France): UNESCO.

28 Singh et al FAO (Food and Agriculture Organization of the United Nations). 1988. Land resources appraisal of Bangladesh for agricultural development. BGD/81/035. Technical Reports 1-7. Fukai S, Rajatsasereekul S, Boonjung H, Skulkhu E. 1995. Simulation modeling to quantify the effect of drought for rainfed lowland rice in Northeast Thailand. In: Fragile lives in fragile ecosystems. Los Baños (Philippines): International Rice Research Institute. p 657-674. Garrity DP. 1984. Asian upland rice environments. In: An overview of upland rice research. Proceedings of the Upland Rice Workshop, Bouaké, Côte d’Ivoire, 1982. Los Baños (Philippines): International Rice Research Institute. p 161-163. Garrity DP, Agustin PC. 1984. A classification of Asian upland rice growing environments. Paper presented at the Workshop on Characterization and Classification of Upland Rice Environments, August 1984. Goiânia, Goiás (Brazil): CNPAF, EMBRAPA. Garrity DP, Oldeman LR, Morris RA, Lenka D. 1986. Rainfed lowland rice ecosystems: characterization and distribution. In: Progress in rainfed lowland rice. Los Baños (Philip- pines): International Rice Research Institute. p 3-23. Garrity DP, Agustin PC, Dacumos RQ, Pernito RN. 1992. A method for extrapolating rainfed cropping systems by land type. In: Proceedings of the 1990 Planning Workshop on Eco- system Analysis for Extrapolation of Agricultural Technologies, 22-25 May 1990, Tuguegarao, Cagayan, Philippines. Los Baños (Philippines): International Rice Research Institute. Garrity DP, Bruce RC. 1992. Rice ecosystems in Cambodia. Paper presented at the seminar on Remote Sensing and GIS in Agricultural Research, 20-22 July 1992. Los Baños (Philip- pines): International Rice Research Institute. 11 p. (In mimeo.) Garrity DP, Singh VP, Hossain M. 1996. Rice ecosystems analysis for research prioritization. In: Evenson RE, Herdt RW, Hossain M, editors. Rice research in Asia: progress and priorities. Los Baños (Philippines): CAB International and International Rice Research Institute. p 35-58. Higgins GM, Kassam AH, van Velthuizen HT, Prnell MF. 1987. Agricultural environments: characterization, classification and mapping. In: Bunting AH, editor. Proceedings of the Rome workshop on agroecological characterization, classification, and mapping, 14-18 April 1986. Wallingford (UK): CAB International. p 171-183. Holdridge LR, Grenke WC, Hatheway WH, Liang T, Tosi JA. 1971. Forest environments in tropical life zones: a pilot study. Oxford (UK): Pergamon. Huke RE. 1982. Rice area by type of culture: South, Southeast, and East Asia. Los Baños (Philippines): International Rice Research Institute. 32 p. Huke RE, Huke EU. 1997. Rice area by type of culture: South, Southeast and East Asia—a revised and updated database. Los Baños (Philippines): International Rice Research In- stitute. 59 p. IRRI (International Rice Research Institute). 1984. Terminology for rice growing environments. Los Baños (Philippines): IRRI. 35 p. IRRI (International Rice Research Institute). 1987. Annual report for 1986. Los Baños (Philip- pines): IRRI. p 432-436. IRRI (International Rice Research Institute). 1989a. IRRI toward 2000 and beyond. Los Baños (Philippines): IRRI. IRRI (International Rice Research Institute). 1989b. Program report for 1988. Los Baños (Phil- ippines): IRRI. p 430-436.

Characterizing rainfed rice environments: an overview . . . 29 IRRI (International Rice Research Institute). 1990. Program report for 1989. Los Baños (Phil- ippines): IRRI. p 68-89. IRRI (International Rice Research Institute). 1991. Program report for 1990. Los Baños (Phil- ippines): IRRI. p 100-102. IRRI (International Rice Research Institute). 1992. Program report for 1991. Los Baños (Phil- ippines): IRRI. p 146-157. IRRI (International Rice Research Institute). 1993. Program report for 1992. Los Baños (Phil- ippines): IRRI. p 62-66, 199-230. IRRI (International Rice Research Institute). 1996. Program report for 1995. Los Baños (Phil- ippines): IRRI. p 54-56. Jones PG, Garrity DP. 1986. Toward a classification system for upland rice growing environments. In: Progress in upland rice research. Proceedings of the 1985 Jakarta Conference. Los Baños (Philippines): International Rice Research Institute. p 107-116. Kam SP, Tuong TP, Hoanh CT, Ngoc NV, Minh VQ. 2000. Integrated analysis of changes in rice cropping systems in the Mekong River Delta, Vietnam, by using remote sensing, GIS and hydraulic modeling. Poster presented at the XIXth ISPRS (International Soci- ety for Photometry and Remote Sensing) Congress, Amsterdam, The Netherlands, 16- 23 July 2000. Kassam AH, van Helthuizen HT, Higgins GM, Christoforides A, Voortman RL, Spiers B. 1982. Assessment of land resources for rainfed crop production in Mozambique. FAO:AGOA:MOZ/75/011, Field Documents 32-37. Rome (Italy): Food and Agricul- ture Organization of the United Nations. KEPAS. 1985. The critical uplands of eastern Java: an agro-ecosystems analysis. Kelompok Penelitian Agro-Ekosistem, Agency for Agricultural Research and Development, Re- public of Indonesia. Koppen W. 1936. Gas geographische System der Klimate. In: Handbuch der Climatologie. Berlin (Germany): Borntrager. Lightfoot C, Axinn N, Singh VP, Bottral A, Conway G. 1989. Training resource book for agroecosystems mapping. Los Baños (Philippines): International Rice Research Insti- tute and the Ford Foundation, India. Lightfoot C, Singh VP, Paris T, Mishra P, Salman A. 1990. Training resource book for farming systems diagnosis. Manila (Philippines): International Rice Research Institute and In- ternational Center for Living Aquatic Resources Management. Magbanua RD, Garrity DP. 1988. Acid upland agroecosystems: a micro-level analysis of the Claveria research site. Proceedings of the Acid Upland Research Design Workshop. International Rice Research Institute and Department of Agriculture. Region 10, Cagayan de Oro City, Philippines. p 1-20. Minh VQ. 1995. Use of soil and agro-hydrological characteristics in developing technology extrapolation methdology: a case study of the Mekong Delta, Vietnam. M.S. thesis. University of the Philippines Los Baños, Los Baños, Philippines. 164 p. Oberthur T, Doberman A, Neue HU. 1995. Spatial modeling of soil fertility: a case study in Nueva Ecija Province, Philippines. In: Fragile lives in fragile ecosystems. Los Baños (Philippines): International Rice Research Institute. p 689-705. Oldeman LR. 1980. The agroclimatic classification of rice-growing environments in Indone- sia. In: Cowell RL, editor. Proceedings of a symposium on the agrometeorology of the rice crop. Los Baños (Philippines): International Rice Research Institute. p 47-66. Oldeman LR, Frere M. 1982. A study of the agroclimatology of humid tropics of Southeast Asia. Rome (Italy): FAO.

30 Singh et al Papadakis J. 1975. Climates of the world and their agricultural potentialities. Buenos Aires (Argentina): Published by the author. Savary S, Srivastava RK, Singh HM, Elazegui FA. 1997. A characterization of rice pests and quantification of yield losses in rice-wheat system in India. Crop Prot. 16(4):387-398. Singh AN. 1987. Assessing extent of damage caused by flooding and drought in predominantly rice cropland area using Landsat data. In: Proceedings of the Eighth Asian Conference on Remote Sensing, Jakarta, Indonesia. Tokyo (Japan): Asian Association of Remote Sensing. p 14:1-10. Singh AN. 1988. Assessing impact of drought in a rainfed cropland area using Landsat data. Paper presented at the Ninth Asian Conference on Remote Sensing, 23-27 Nov. 1988, Bangkok, Thailand. Singh RK, Singh VP, Variar M. 1994. Environmental characterization for on-farm research in rainfed uplands of Bihar, India: a case study. Rice Farm. Syst. Tech. Exch. 4:7-9. Singh VP, Pathak MD. 1990. Rice growing environments in Bahraich district of Uttar Pradesh. Lucknow (India): Uttar Pradesh Council of Agricultural Research. 100 p. Singh VP, Singh RK, Singh RK, Chauhan VS. 1993. Developing integrated crop-livestock-fish farming system for rainfed uplands in Eastern India. J. Asian Farm. Syst. Assoc. 1(4):523- 536. Singh VP, Singh AN. 1995. Integration of different levels of rice ecosystems analysis. In: Frag- ile lives in fragile ecosystems. Los Baños (Philippines): International Rice Research Institute. p 481-506. Singh VP, Singh AN. 1996. A remote sensing and GIS-based methodology for the characterization and classification of rainfed environments. Int. J. Remote Sensing 17(7):1377-1390. Singh VP, Sastri ASRAS. 1998. Sustaining rice farming in Eastern India: a conceptual and factual analysis. Paper presented at the National Seminar on Farming Rice in Eastern India, 19-22 March 1998. Lucknow (India): Uttar Pradesh Council of Agricultural Re- search. Singh VP, Singh RK, Sastri ASRAS, Baghel SS, Chaudhary JL. 1999. Rice growing environments of Eastern India: an agroclimatic analysis. Raipur (India): Indira Gandhi Agricul- tural University and Los Baños (Philippines): International Rice Research Institute. 76 p. Singh VP, Singh RK. 2000. Rainfed rice: best practices and strategies in eastern India. Manila (Philippines): International Rice Research Institute, Indian Council of Agricultural Re- search (India), International Fund for Agricultural Development (Italy), and Interna- tional Institute for Rural Reconstruction (Philippines). 292 p. TAC (Technical Advisory Committee). 1990. Towards a review of CGIAR priorities and strategies. Progress report by the Technical Advisory Committee (TAC/CGIAR). Washing- ton, D.C. (USA): Consultative Group on International Agricultural Research. Thornthwaite CW. 1948. An approach towards a rational classification of climate. Geog. Rev. 38:55-94. Tuong TP, Hoanh CT, Khiem NT. 1991. Agro-hydrological factors as land qualities in land evaluation for rice cropping patterns in the Mekong Delta of Vietnam. In: Deturck P, Ponnamperuma FN, editors. Rice production on acid soils of the tropics. Kandy (Sri Lanka): Institute of Fundamental Studies. p 23-30. WARDA (West Africa Rice Development Association). 1992. Annual report for 1991. Bouaké (Côte d’Ivoire): WARDA. WARDA (West Africa Rice Development Association). 1993. Annual Report for 1992. Bouaké (Côte d’Ivoire): WARDA.

Characterizing rainfed rice environments: an overview . . . 31 Widawsky DA, O’Toole JC. 1990. Prioritizing the rice technology research agenda for Eastern India. New York, N.Y. (USA): Rockefeller Foundation. Wopereis MCS. 1993. Quantifying the impact of soil and climatic variability on rainfed rice production. Ph.D. thesis. Wageningen Agricultural University, Wageningen, The Neth- erlands.

Notes Authors’ address: International Rice Research Institute, DAPO Box 7777, Metro Manila, Philippines. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

32 Singh et al Characterizing environments for sustainable rice production

Van Nguu Nguyen

Growth in the world’s rice production has slowed down. Since 1990, the growth rate of rice production has been lower than that of the population. This indicates the need to increase efforts to improve rice productivity and bring more land area under rice cultivation. Substantial efforts have been made to characterize and classify the environments for agricultural production in general and for rice production in particular during the past three decades. Major efforts in the characterization and classification of these environments are reviewed and selected examples of the contribution of these exercises to rice production via the development of rice technologies, expansion of rice area, transfer of rice technologies, and others, such as the assessment of investments in agricultural research, are provided and discussed. Rice production factors such as varieties, water and land/soil resources, insects and diseases, socioeconomic issues, and global climate have also evolved substantially during the past 30 years. A new generation of rice varieties with higher yield potentials and better resistance to abiotic and biotic stresses has been developed using hybrid rice and biotechnology. The increasing deficiency of nutrient elements and pressure from insects and diseases have been observed in intensive rice production systems. Socioeco- nomic factors such as labor availability and wages and availability of inputs and credits have changed with improvements in the economies of rice-producing countries. The global climate has been warming and discussions cover the implications of these changes for the suitability of rice production under different environments. The globalization of the world economy and the decline in public investments for agricultural research activities require that future efforts in characterization and classification be evaluated in terms of returns to agricultural production. Discussions refer to the issues affecting the efficiency and applicability of characterizing and classifying the environments of rice production such as universality, completeness, objectivity, and scales and costs of the exercises.

Characterizing environments for sustainable rice production 33 Rice is the staple food crop for more than half the world’s population. Its popularity has increased steadily, not only in rice-eating countries but also in countries where traditionally it is not an important food crop. The slowdown in the growth of rice production is serious because of the continuing growth of population. Sustainable rice production in the near future therefore requires more efforts to improve rice productivity and to bring more land area into rice cultivation. During the past three decades, substantial efforts have been made to characterize and classify the environments of agricultural production, especially the production of important food crops such as rice. The results of these exercises have been applied in developing crop production technologies, expanding new crop areas, transferring technologies, and others. Recently, results of characterization and classification of environments have also been applied in assessing investments in agricultural research. Rice yields are affected by variety, ecological conditions during the growing season, and socioeconomic factors that affect farmers’ crop management. Rice crops yield highest when planted in the best-suited environments. Factors affecting rice production, however, have evolved much during the past three decades and need to be included in future efforts to characterize and classify the environments of rice production. In the competitive markets created by the globalization of the world economy, improving rice productivity and expanding rice areas must be done in the most efficient manner. Public investments in rice research have also been declining. In the future, therefore, characterizing and classifying the environments of rice production must take into consideration the following areas: universality, completeness, objectiveness, and scales and costs.

The challenge to the world’s rice production Rice is the world’s most important food crop. In 1997, about 2.9 billion people depended mainly on rice for food calories and protein. The popularity of rice has also increased in Africa and Latin America, where traditionally rice has not been an important food crop. The worldwide annual growth rates of population and rice production, harvested area, and yield since 1970 (Table 1) show that world rice production has increased continuously, but at varying growth rates. The annual growth rate was 2.7% in the 1970s, 3.1% in the 1980s, and 1.3% in the first half of the 1990s. A comparison between the growth rates of rice production and population since 1970 shows that, for the first time since 1990, rice production has grown more slowly than population. During the 1970s, the high annual growth rate of rice production was caused by both a high increase in yield and a moderate increase in rice area, whereas the rapid growth in rice production during the 1980s came principally from improvements in rice productivity. Rice yield grew 1.8% annually in the 1970s, 2.8% in the 1980s, and only 1.1% in the first half of the 1990s, whereas the annual growth rate of harvested rice area decreased from 0.8% during the 1970s to 0.2% in the 1980s (Table 1). The trend of reduced growth of rice harvested area indicates that future rice production increases will come mainly from improvements in productivity unless

34 Van Nguu Nguyen Table 1. Annual growth rates (%) of the world’s population and rice production, harvested area, and yield.

Rice

Period Population Production Harvested Rice yield area

1970-79 2.03 2.71 0.80 1.76 1980-89 1.86 3.14 0.23 2.80 1990-96 1.55 1.31 0.23 1.10

major development activities are undertaken to bring more land under rice cultivation. The very low annual growth rate of rice yield observed since 1990 is therefore a cause for concern and it has been the topic of numerous reviews (Pingali and Rosegrant 1994, Cassman and Pingali 1995, Pingali et al 1997). Regardless of food consumption trends, the slowdown in growth of rice production is particularly serious considering the continuing growth of population. This declining growth trend in rice production needs to be reversed if the world’s rice production is to meet popular demand. This requires efforts to increase rice yield, expand rice area, or a combination of both. Historical evidence indicates that the world’s efforts in these directions could be effectively assisted through vigorous characterization of the environments of rice production.

Characterizing environments and their impacts on rice production Chandler (1979) considered the inventory/analysis of natural resources as an essential element for a successful national rice program. Rice is grown from about 50°N to about 35°S, from below sea level to about 2,700 m above, and in a wide range of ecological conditions, from dry land where soils are freely drained to flooded land with the depth of flooded water reaching several meters. The development of irrigation further modified the ecological environments of rice production. Human intervention in rice production increases the complexity of the already diverse rice production environments. Several systems have characterized and classified the environments of rice production. These activities in characterization and classification have provided the basis for increasing rice production by developing improved technologies, developing new rice areas, transferring improved technologies, and making other investments aimed at supporting rice production.

Major activities in characterizing and classifying environments Early attempts to characterize and classify the environments of rice production. Rice scientists and researchers from different national and international institutions have made considerable efforts to classify and characterize the environments of rice production. In late 1973, the International Rice Research Institute (IRRI) organized its

Characterizing environments for sustainable rice production 35 first working group to identify the macro soil/climate zones of rice production in Southeast Asia (IRRI 1974). A comprehensive agroclimatic classification for evaluating cropping systems in Southeast Asia based on the amount of monthly rainfall and length of consecutive wet months was reported by Oldeman in 1974. A wet month is defined as a month with at least 200 mm rainfall. The agroclimatic environments of Southeast Asia were classified into four zones (Oldeman 1974): ● Zone I: more than 9 consecutive wet months ● Zone II: 5–9 consecutive wet months; zone II has four subzones ● Zone III: 2–5 consecutive wet months; zone III has three subzones ● Zone IV: less than 2 consecutive wet months Since then, several attempts to characterize and classify rice production environments have been carried out by different scientists working in different locations around the world. The bases for characterizing rice production environments used up to the mid-1980s, however, are different (Table 2). Recognizing the importance of having a common agreement on environmental terminology so that plant type can be better related to environments, IRRI established in 1982 an international committee to develop an agreed-upon terminology and classification for rice (IRRI 1984). The recent characterization and classification of the environments of rice production in inland valleys in West Africa. Rice is an important food crop in West Af- rica, but local production has not been able to meet popular demand, resulting in the spending of large amounts of foreign exchange by governments in the region to import rice. The region has large wetland areas in inland valleys, however, that are still not fully exploited. Considerable efforts have therefore been made recently to characterize and classify the environments in these inland valleys, where rice has been a traditional crop. The efforts have been carried out since 1982 by various national and international institutions, such as the International Institute of Tropical Agriculture (IITA) (Windmeijer and Andriesse 1993), Conférence des Responsables de Recher- che Agricole Africains (CORAF) (Albergel et al 1993), the Centre de Coopération Internationale en Recherche Agronomique pour le Développement (CIRAD) (Legoupil and Bidon 1995), and the West Africa Rice Development Association (Becker and

Table 2. The classification of rice production environments during the 1970s to early 1980s.

Basis of classification/characterization Exercises (no.)

General surface hydrology 15 Physiographic source of water 3 Landform and soil units 3 Matrix of ecological factors 9 Soil suitability 3 Crop season, intensity, and management 3 Comprehensive (combination of above 6 items) 2 Total 38

Source: Bowles and Garrity (1988).

36 Van Nguu Nguyen Diallo 1992). In 1990, the FAO Regional Office in Accra, Ghana, established a Tech- nical Co-operation Network in Wetland Development and Management (WEDEM) aimed at promoting the environmentally sound development and management of wetland resources for sustainable food production and exchange of information. The characterization and classification of the agrosystems in inland valleys in West Africa by the earlier-mentioned institutions were based on physical (e.g., climate, lithology, landform, soil, and hydrology), biotic (e.g., vegetation, insects, diseases, and weeds), and management (land use, distance from farm to market) factors. Land use was described by socioeconomic parameters such as labor, capital input, and management. The FAO’s agroecological zoning for agricultural production. Recognizing the importance of classifying and characterizing natural resource bases, the Food and Agriculture Organization of the United Nations (FAO) began in 1976 to develop the agroecological zones (AEZ) methodology (FAO 1976) and supporting databases and software packages to provide solutions for land resources analysis in member countries, linking land-use outputs with other developmental goals in areas such as food production, food self-sufficiency, cash crop requirements, and population-supporting capacity. The key elements of AEZ are based on the FAO’s Framework for Land Evaluation (FAO 1976), which emphasizes the need to characterize land-use types (LUTs) as a necessary precursor to land evaluation and land-use planning. Climatic, soil, and plant parameters were used to calculate the length of the growing period (LPG) for various crops and to determine crop suitability. Land productivity is estimated at two levels of input application: the minimum and the optimum. The first use of the AEZ methodology was to assess the production potential of land resources in the developing world based on climatic data and the 1:5,000,000 scale FAO/UNESCO Soil Map of the World. The AEZ methodology was applied entensively to evaluate the suitability and land productivity under rainfed conditions of five major food crops—wheat, rice, maize, barley, pearl millet, and sorghum; three root crops—white potato, sweet potato, and cassava; two leguminous crops—soybean and Phaseolus bean; and other cash crops in 117 countries in Africa, Central and South America, and Asia. One of the major products of this exercise was the mapping of zones suitable for producing various food and cash crops under rainfed conditions in the developing countries of Africa, Central and South America, and Asia, an example of which appears in Figure 1. Although maps were produced for other crops, however, none was made for rice (FAO 1978, 1980, 1981), perhaps because of the uniqueness of the rice-growing ecological environments. Outside of upland ecologies, flooding and its patterns mostly dominate rice environments. The methodology and findings of the AEZ methodoloy were presented at a FAO conference in 1983, which, recognizing the importance of such work for development, recommended that similar work be undertaken at the national level (Antoine 1994). Since then, FAO has assisted several countries in learning the methodology and applying and adapting it to tackle issues of land, food, and people at the national and subnational level. FAO’s agroecological zone models were successfully linked

Characterizing environments for sustainable rice production 37 Tropics Subtropics Temperate

Summer/winter rainfall Normal isoline Intermediate isoline High altitudes/cold temperatures High altitudes/cool temperatures Very suitable Suitable Marginally suitable Not suitable

Fig. 1. Agroclimatic suitability assessment for rainfed pearl millet production in South America (after FAO 1981). with geographic information systems (GIS) to appraise natural resources to support the population in Kenya, Nigeria, and the Sahel countries in West Africa (Antoine 1994).

Selected examples of the contribution of the characterization and classification of environments to sustainable rice production Example 1. Developing cropping systems to intensify rice production in rainfed lowlands. Rice-cropping intensification—or the growing of two or more rice crops on the

38 Van Nguu Nguyen same piece of land in a year—increases not only rice production but also farmers’ incomes and employment opportunities. After the development of short-growth-duration and photoperiod-insensitive rice varieties, cropping intensification has been widely practiced in irrigated areas under tropical climate. On the other hand, the traditional method of growing rice in rainfed lowlands allows farmers to grow only one rice crop per year during the wet season. The results of the agroclimatic classification mentioned earlier (IRRI 1974, Oldeman 1974) led to a study at IRRI aimed at developing new rice-based cropping systems for the rainfed lowland rice areas in the early 1970s. With the encouraging results of IRRI’s study, the Philippine Council for Agriculture and Resources Re- search and Development agreed in 1975 to join with IRRI to implement a cooperative applied research project on rainfed cropping systems. The project aimed at developing, evaluating, and disseminating improved rice-based cropping systems for the rainfed lowland rice areas in the Philippines. The most significant result of this cooperative project was the development and transfer of a new cropping system that allowed the growing of two rice crops per year in rainfed lowland areas where the wet season is at least 6 mo with rainfall of 200 mm or more (Fig. 2). The Philippines’ national average rice yield before implementation of the project was about 1.2 t ha–1. Adopting this new cropping system enabled farmers to grow two rice crops and an upland crop per year in many rainfed lowland areas, with a possible annual yield of 9–10 t ha–1 for rice (Cardenas et al 1980). The higher rice production multiplied the incomes of farmers several times. With the impressive results obtained in the Philippines, many rice production programs in Asia have adopted and then modified this new cropping system and trans-

Rainfall (mm)

Direct seeding Harvesting Transplanting 700 Seedbed preparation

600 Introduced First 500 system crop Second 400 crop Traditional 300 system Normal transplanting 200

100

0 J FMAMJ J A SOND Month Fig. 2. Cropping calendar of introduced and traditional cropping systems in some rainfed lowland rice areas in the Philippines (after Cardenas et al 1980).

Characterizing environments for sustainable rice production 39 ferred it for wide adoption by farmers for increasing both rice production and the production of other food crops and farmers’ income. The environments of rice production in many Asian countries have zones with agroecological conditions similar to those in the Philippines where this new cropping system was tested (Oldeman and Frere 1982, Huke 1982a,b). Example 2. Expanding rainfed lowland and irrigated rice areas in West Africa. West Africa has abundant wetland areas, but traditional rice production was carried out mainly in upland ecologies. As mentioned earlier, activities to characterize and classify the environments in inland valleys in West Africa were strengthened starting in 1982. The rapid increase in harvested rainfed lowland and irrigated rice area in West Africa after 1985 (Table 3) could be partially attributable to using the results of the work on characterization and classification of the environments in inland valleys in the region. In 1975, the harvested area from rainfed lowland rice was only 441,000 ha. It increased to 516,000 in 1985, an increase of about 17% after 10 years. In 1995, it was 979,000 ha, an increase of about 90% after 15 years. A similar trend in area expansion was also observed in irrigated rice (Table 3). The expansion in rice area has greatly contributed to the rapid growth of rice production in the region. Example 3. Guidelines for fertilizer recommendations in rice production in Bangladesh. Rice is the dominant food crop in Bangladesh’s agriculture and is the staple food for more than 120 million Bangladeshis. Intensification of rice production has resulted in higher demand for fertilizer because of higher crop removal. This, coupled with the imbalance in fertilizer application, mainly nitrogen, has led to increasing deficiencies of nutrient elements such as phosphorus, potassium, zinc, and sulfur in many rice soils, thus endangering the sustainability of national rice production. The Bangladesh Agricultural Research Council began and implemented a project aimed at applying agroecological zoning to support rice production during the early 1980s, with financial and technical support from the United Nations Development Programme (UNDP) and FAO. The project’s activities resulted in “a fertilizer recommendation guide” for extension workers to use in planning their activities and assist- ing rice farmers. This has greatly contributed to the sustained growth of rice production in the country (Karim 1994). Table 4 shows part of the fertilizer recommendation guide for rainfed rice-cropping systems. Other examples of impact of the classification and characterization of environments. Advances in geographic information systems (GIS), crop simulation model-

Table 3. West Africa’s harvested rice area, 1975 to 1995 (000 ha).

Year Total Irrigated Rainfed Upland Others lowland

1975 2,292 82 441 1,379 401 1980 2,561 171 507 1,574 310 1985 2,706 231 516 1,599 370 1990 3,380 337 753 1,916 376 1995 3,886 492 979 1,983 432

40 Van Nguu Nguyen Table 4. Selected fertilizer recommendation guide for rainfed rice-based cropping systems in agroecological region 11 in Bangladesh: High Ganges River floodplain.

Fertilizer recommendation Cropping pattern (kg ha–1) Land and soil characteristicsa

Season Crop NP2O5 K2OS Zn

Land type: highland; organic Rabi Wheat 60 30 30 10 – matter: low; pH: 6.1–7.9; Kharif 1 B. aus (L)b 30–––– texture: silt-loam, K-bearing Kharif 2 Fallow ––––– minerals: medium Rabi Mustard 80 60 40 30 3 Kharif 1 Groundnut/jute 40 – 10 – – Kharif 2 B. aus (L) 30 ––––

Rabi Chickpea/lentil 20 40 30 10 – Kharif 1 B. aus 30 –––– Kharif 2 Fallow –––––

Land type: medium highland; Rabi Khesari 10 20 – – – OM: low; pH: 6.1–7.9; Kharif 1 Aus + aman (L) 30 20 20 – – texture: loamy; K-bearing Kharif 2 – –––– minerals: medium Rabi Chickpea 20 50 30 10 5 Kharif 1 B. aus (L) 30 20 20 – – Kharif 2 T. aman (L) or 50 20 20 – – T. aman (HYV) 70 20 20 10 –

Rabi Wheat 60 30 30 10 3 Kharif 1 Jute 45 – 10 – – Kharif 2 T. aman (L) or 50 20 20 – – T. aman (HYV) 70 20 20 10 –

Land type: medium lowland; Rabi Khesari 10 30 – – – OM: medium; pH: 6.1–7.9; Kharif 1 Aus + aman (L) 30 –––– texture: clayey; K-bearing Kharif 2 – ––––– minerals: high Rabi Khesari 10 30 – – – Kharif 1 B. aman (L) 30 –––– Kharif 2 – –––––

Rabi Boro (L) 60 40 20 10 3 Kharif 1 Kharif 2 aOM = organic matter. bB = broadcast, T = transplanted, L = local variety, HYV = high-yielding variety. Source: Karim (1994).

Characterizing environments for sustainable rice production 41 ing, remote sensing, and computing technologies have further strengthened the applicability of AEZ in agricultural planning and development. Many countries have used the AEZ methodology (FAO 1994) in assessing land productivity and population- supporting capacity; in land evaluation and land-use planning; in assessing environmental degradation due to agricultural production; and in research planning, technology transfer, farming systems analysis, and recommendations for input application and supply. Detailed information on these applications was published in the FAO’s World Soil Resources Report Number 75 (FAO 1994). Application of the characterization and classification of the environments of agricultural production has recently been extended to quite a new area: the evaluation of investments in agricultural research. The methodology for evaluating alternative research investments calls for assessing the impacts of past research on production (ex post analysis) and the possible or potential impact of the currently proposed research alternatives (ex ante analysis). Characterizing and classifying the environments, especially those that employ the AEZ methodology, can provide quantified inputs to such assessment (Pardey and Wood 1991, Wood and Pardey 1993). Dividing geographic space into AEZs provides an estimate of the area that could benefit from the results of the proposed research investments. The homogeneous conditions under each AEZ, on the other hand, facilitate quantifying the response to (or outputs of) the application of new technologies resulting from the proposed research investments. Research investments could then be disaggregated to commodities, subtypes, environments, and problem and discipline domains. For example, investments in rice research can be grouped into irrigated, rainfed lowland, upland, and deepwater and tidal wetland, and then into genetic improvement, crop management, and crop protection. The products of the research investments could be further classified into non-site-specific (applied equally to all AEZs), site-specific (applied to only one AEZ), or multivariable site-specific (variable AEZs).

Rice production factors as guidelines for environmental characterization Variety, ecological conditions during the growing season, and the socioeconomic factors that affect farmers’ crop management determine the yield of a rice crop. The factors affecting rice production have undergone substantial evolution during the past 30 years. This requires more attention to the classification and characterization of the environments of rice production in the future.

Rice varietal development The rapid growth in world rice production during the 1980s (Table 1) came mainly from the gain in productivity. The results of the maximum yield studies carried out by IRRI during the 1970s showed that, in tropical climate areas, the potential yield of rice varieties was about 3.7–6.8 t ha–1 in irrigated ecologies, 2.5–4 t ha–1 in rainfed lowlands, and 2 t ha–1 in upland and other ecologies (Table 5). Cassman et al (1997) reported that yields of 4 to 5 t ha–1 are normally obtained in irrigated areas in several tropical countries. The average yields in 1997 in major rice-producing countries such

42 Van Nguu Nguyen Table 5. Estimated maximum farm yields (t ha–1) for different types of rice land in 11 Asian countries.

Country Irrigated Supplemental Rainfed Upland and dry-season irrigated wet-season lowland deepwater

Philippines 5.94.6 3.5 2.0 India 6.8 5.4 4.0 2.0 Indonesia 5.94.8 3.6 2.0 Thailand 4.4 3.7 2.5 2.0 Bangladesh 6.6 4.93.7 2.0 Vietnam 5.8 4.1 3.1 2.0 Sri Lanka 5.7 5.3 4.0 2.0 Myanmar 6.0 4.8 3.6 2.0 Pakistan 6.0 – – – Nepal – 4.8 3.6 2.0 Malaysia 6.0 4.8 3.6 2.0

Source: Chandler (1979).

as Bangladesh, Brazil, India, Myanmar, Nigeria, the Philippines, Thailand, and Viet- nam were still below 4 t ha–1, indicating the limited improvement in rice productivity in other ecologies. In 1997, about 54% of the world’s rice harvested area came from irrigated ecologies, 30% from rainfed lowland ecologies, 11% from upland ecologies, and 5% from other ecologies such as deepwater and tidal wetlands or mangroves. Irrigated rice was responsible for about three-quarters of the world’s total rice production, indicating the limited contribution of rice production in other ecologies. This further confirms the observation on the limited success of activities aimed at improving rice productivity in rainfed environments during the past 30 years. The yield potentials of high-yielding rice varieties (HYVs) in tropical areas, however, have not improved further after the development of IR8 in the late 1960s, although yield efficiency of rice varieties has been improved with the development of early maturing HYVs. Increasing efforts have therefore been made to develop new rice varieties with higher yield potentials. Since 1982, the Japanese government has been promoting a project to develop super-high-yielding varieties with the target of increasing rice yield by 50% in 15 years based on wide crosses between indica and japonica varieties. Breeding for the new plant type that could increase the yield potential of tropical rice by 25% to 50% began at IRRI in 1989. Tropical japonica varieties have been used as sources for desirable traits in this project. After learning of the successful application of this technology for increasing rice production in China, FAO, IRRI, and several other national institutions have been promoting the development and use of hybrid rice (Tran and Nguyen 1998). The West Africa Rice Development Association has made considerable efforts to develop rice varieties for low-input management areas in West Africa from crosses between Oryza sativa and O. glaberrima. Therefore, new and more vigorous characterization and classification of

Characterizing environments for sustainable rice production 43 rice environments will be needed to develop management techniques to allow new rice varieties to fully express their potential. Biotechnological tools have been increasingly used to develop rice varieties with better resistance to insects, diseases, and abiotic stresses. Transgenic rice plants with Bt genes have been created for stem borer resistance. Herbicide-resistant transgenic rice varieties have also been developed. Substantial progress has been made in the use of biotechnological tools for creating drought- and salinity-tolerant rice varieties (Khush et al 1999). These developments may influence the classification of the suitability of rice varieties for production under different environments.

Water supplies Rice depends on water for normal growth and development. An insufficient water supply leads to drought stress, whereas an oversupply of water results in complete submergence. Rice yields under both situations are usually low. The water supply in rice production is controlled best under irrigation systems. Irrigation water is increasingly becoming less and less available for rice production in many countries because of the depletion of aquifers, salinity, and competition for water from urbanization and industrialization. In several other countries, the high costs of irrigation infrastructure have limited the building of new irrigated rice schemes. In the near future, therefore, large rice areas will remain under rainfed conditions. Considerable deepwater rice areas in Bangladesh, Cambodia, and Vietnam have been converted for growing boro and/or dry-season irrigated rice. Sizeable rice areas, however, still suffer from frequent deep floods. Drought. Rice plants generally have shallow root systems; therefore, water deficiency or drought stress has been considered to be the most important yield-limiting factor in rainfed lowland and upland rice. It is also a major constraint to rice production in most deepwater rice areas and in some irrigated and tidal wetland areas. Under rainfed conditions, water deficiency can occur at any time during the cropping season, but especially during the early and late stages of the crop. The degree of damage to rice crops depends on the time in relation to the development stage of the rice plants when water deficiency occurs and its intensity and duration. The damage is usually heavy and irreparable when intensive water deficiency occurs during the reproductive and flowering stages. Yield losses of 1 t ha–1 or more may occur after 10 d of continuous water deficiency during these stages. Some rice varieties exhibit the capacity to fully recover and resume normal growth after being exposed to drought stress during the vegetative stages, but their growth duration is generally prolonged. This could make rice varieties unsuitable in areas or zones that were originally classified as suitable based on plant growth duration. The main sources of the water supply to rainfed lowland rice in undulating terrain are rainfall, interflows, and streams activated by local rain. In basins, deltas, estuaries, and lake fringe, the groundwater table may rise during the rainy season and come within easy reach of the rice roots. Under rainfed conditions, therefore, the water supply to wetland rice fields depends on both rainfall and the hydrology of the rice field. Farmers often carry out bunding and leveling of rice fields to conserve

44 Van Nguu Nguyen water for rice production. The feasibility of field bunding and leveling depends greatly on the topography of the area. Furthermore, the soil’s capacity to retain water has an important influence on the water supply for rainfed rice during its growing season. Therefore, parameters such as rainfall and its distribution, hydrology of rice fields, topography, and soil water retention capacity should be used when estimating the period of water deficiency and its duration. Complete submergence. Although rice plants are well known for their ability to transport oxygen from the air into their root systems, flooding, with consequent crop submergence, may severely damage the rice crop. Most wetland rice varieties, including deepwater ones, can withstand complete submergence for at least 6 d before 50% of the crop dies, whereas 100% mortality occurs in all varieties within 14 d of complete submergence (Setter et al 1995). The growth and development of rice plants under complete submergence are also affected by the quality of floodwater. In addition to direct damage to rice crops from submergence, stagnant water causes excessive soil reduction, which alters the chemistry of wetland soils and usually causes nutrient deficiency or toxicity or both to the rice crop. Rice yields from fields where prolonged stagnant flood occurs are usually low. Susceptibility to complete submergence therefore needs to be considered when we characterize rice production environments.

Adverse soil conditions A decline in rice yield in long-term intensive rice production systems has been observed, at least at the experimental level (Cassman et al 1997). At the farm level, a decline in total factor productivity has been observed. Farmers have to apply more and more production inputs to obtain the same rice yield (FAO 1997). Undoubtedly, soil physicochemical and biological environments change under continuous submergence for a long period. Deficiencies in phosphorus, potassium, zinc, and sulfur have expanded in many lowland rice areas (Cassman et al 1997). Increased salinity in rice soils under intensive and continuous irrigated rice cultivation has also been increasingly observed (Pingali and Rosegrant 1994). These parameters need to be well quantified and characterized if solutions to the reversing of declining trends are to be found. In Asia, land areas that are suitable to rice production have been used. Substan- tial wetland areas in coastal plains in some countries, however, are still available and could be developed for rice production. Many of these land areas, however, are influenced by tidal water and have unfavorable soil conditions such as salinity, acidity, and peat. A high level of iron concentration is a major constraint to yield in many lowland rice areas, especially the inland valleys in sub-Saharan Africa, where about 140 million ha of this type of land are still available (Ton That 1982). Iron toxicity has been cited as a yield constraint to wetland rice in Brazil (Pulver, personal communication). The characterization and classification of rice production environments that take into account these parameters would be useful, especially for selecting areas for expanding rice production.

Characterizing environments for sustainable rice production 45 Insects and diseases Insects and diseases reduce rice yield. Intensification in rice production has changed the types and pressures of many insects and diseases. Although insects and diseases can be controlled with appropriate management techniques, their pressures need to be characterized and classified, at least for varietal improvement activities. The use of resistant varieties is still the cheapest and most effective measure for controlling insects and diseases in rice.

Socioeconomic issues Although an area may be suitable for rice production based on ecological environments, rice production may not necessarily be the best suited based on socioeconomics. Many areas suitable to rice production in southern Brazil, Argentina, and Venezuela are either planted to other crops or under fallow due to the high costs of rice production and irrigation infrastructure. With the globalization of the world economy, rice production needs to be evaluated for its competitiveness not only with rice production in other areas but also with the production of other crops in the same area. The costs of providing favorable conditions for rice production such as irrigation infrastructure, an adequate supply of inputs and credit, and better marketing of rice need to be included in characterization. It is well known that improved rice technologies, such as high-yielding varieties, could not make a significant contribution without the adequate availability of production inputs. The yield of IR8 does not differ significantly from that of its parent, Peta, unless fertilizers are applied. The labor force for rice production in many countries has been shrinking rapidly due to a combination of population control, improvement in income, and employment opportunities created to improve national economies. Also, in many countries, the population of rice farmers has become older and older because of the migration of young people to urban centers and job preferences. The scarcity of labor and high rural wages have led to a shift in the application of rice technologies in these countries. Land preparation, harvesting, and threshing operations have become more and more mechanized, direct seeding has increasingly replaced transplanting as the main crop establishment method, and chemical weeding has become more popular than hand weeding. In countries such as India, the Philippines, and Thailand, women have been participating more and more actively in rice production operations. Socioeconomic factors vary with changes in the national economy. Understand- ing socioeconomic factors could have an important bearing on activities aimed at developing new rice land and at creating the appropriate conditions for adopting improved technologies. Socioeconomic factors were considered as major constraints to the agricultural development of inland valleys in West Africa (Table 6).

Global climate change Global climate change has been increasingly observed. The global climate has warmed because of the emission of CO2 and other gases such as methane and nitrite oxide and their accumulation in the atmosphere. Temperature and CO2 concentration are important parameters of the photosynthetic pathway, whereas temperature influences respi-

46 Van Nguu Nguyen Table 6. Major constraints to agricultural development of inland valleys in West Africa.

Constraint Rankinga

Weeds 4.7 Lack of water control 4.7 Lack of inputs 4.6 Labor shortage 4.4 Crop diseases 4.3 Lack of credit 4.3 Nutrient status 4.2 Land tenure 4.1 Poorly adapted varieties 3.9 Soil erosion 3.7 Soil structure 3.6 Insects/pests 3.6 Marketing 3.6 Iron toxicity 3.4 Acidification 3.2 Land clearing 3.2 Human health 3.1

aOn a scale of 1–5, where 1 = not important and 5 = very important. Source: Jasmin and Andriesse (1993). ration. A change in temperature regime may also lead to changes in rainfall and its distribution and cloudiness and its distribution, and thus solar radiation, wind speed, and biological activities in the soil-air-plant system. These factors have large effects on rice growth, development, and yield.

Major issues in characterizing rice environments Experiences gained in characterizing and classifying rice production environments in the past could provide insights for future exercises in this domain. The following are some of the issues that future activities in characterizing rice production environments may have to consider.

The universality and compatibility of the systems used Table 2 shows that the bases used in many systems for classifying and characterizing rice production environments before the mid-1980s were not similar. This lack of universality has limited the usefulness of environmental classification and characterization, at least in terms of collaboration among rice research institutions and scientists to develop and transfer rice technologies. This was perhaps the main reason for IRRI to establish the International Committee on Terminology of Rice Growing En- vironments in 1982.

Characterizing environments for sustainable rice production 47 The completeness of characterization and classification Incompleteness in characterization was another weakness of many early attempts to characterize and classify rice-growing environments. Several environmental factors may be static, but several other parameters are very dynamic in nature. For example, although environmental parameters such as altitude, longitude, latitude, soil texture, landform and slope, and even soil pH are more or less static, parameters such as groundwater table, depth of flooded water layer, temperature, solar radiation, and others are very dynamic in nature. An incomplete characterization may lead to imperfect classification. For example, theoretically, rainfed rice production is not feasible in areas where rainfall is less than 800 mm. Rainfed rice production has been successful, however, in many inland valleys in the Sudan Savanna Zone in West Africa, where annual precipitation ranges from 550 to 900 mm (Windmeijer and Andriesse 1993). The water supply in rice production in this area comes not only from rainfall but also from the groundwater table, runoff, and seepage from surrounding areas. The concept of minimum data set (MDS) has been proposed for characterizing the ecological conditions of crop production. This concept has been used in many projects, such as the project on the International Benchmark Sites Network for Agrotechnology Transfer (Uehara and Tsuji 1991). The MDS is supposed to provide basic environmental factors that have potential effects on plant growth, health, and development. The MDS, however, may be too simple for some studies. For example, results of experiments to evaluate rice germplasm for salinity tolerance may be misleading if soil alkalinity was not included in the environmental characterization of the experimental sites. Soil salinity is usually associated with soil alkalinity and both have negative effects on the growth and development of rice plants. Similarly, Pen- ning de Vries et al (1989) considered that the minimum data sets used in many experiments to evaluate drought tolerance of rice varieties are not adequate for simulation modeling. They opined that, to evaluate the level of drought stress in a given environment, parameters such as rainfall, humidity, light, temperature, and soil characteristics in the root zone are needed. Andriesse and Fresco (1991) noted the weakness of the broad agroecological classification of rice environments when they reviewed the results characterizing rice-growing environments in West Africa.

The objectiveness of characterization and classification The sources of information used to characterize and classify rice production environments are another factor. Most activities to characterize and classify rice production environments in the past were carried out mainly by researchers and scientists or people responsible for rice development. The characterization and classification may therefore be subjective, which consequently renders them less useful. Several technologies and infrastructures supporting rice production developed based on the characterization and classification of rice production environments in the past have proven to be unsuitable or unsustainable when they are viewed environmentally and socio- economically. Many farmers in several countries still plant many traditional rice varieties regardless of the availability of improved rice varieties and extension efforts. Similarly, many large irrigated rice schemes built in Africa during the 1970s and

48 Van Nguu Nguyen 1980s have not proven to be economically profitable. Insect control techniques based solely on chemical application were unsustainable. Therefore, the participation of all stakeholders in environmental characterization, in the selection of production sites, and in the development and selection of technologies needs to be encouraged. Many FAO field projects, including projects on rice research and development, have recently employed the participatory approach (with stakeholders and farmers) during their implementation (Nguyen 1998).

The costs and scale of characterization and classification Environmental characterization and classification also need to be evaluated in terms of returns to agricultural production. More detailed characterization would lead to more precise classification and thus better inputs for assessing investments, but it is also more costly. As a rule of thumb, the more details of the characterization, the larger the scale or ratio between the area under characterization and the area in mapping of AEZ, and the higher the cost involved. Detailed characterization and classification are relative, depending on the objectives of the exercise. Some general characterization at a smaller mapping scale is adequate for certain objectives, whereas detailed characterization at a larger mapping scale is needed for other objectives. Andriesse et al (1994) proposed several levels and scales for the agroecological characterization of inland valleys in West Africa (Table 7). For agricultural production at different administrative levels, Koohafkan et al (1998) proposed four scales and levels of analysis in the characterization of lands and water-use planning and management (Table 8).

Conclusions The slowdown in the growth of the world’s rice production after 1990 has led to an increasing call for the renewal of efforts to sustain rice production. Historical evidence indicates that efforts to increase the world’s rice production could be strengthened by the characterization of rice production environments. As in the past, future environmental characterization should provide the basis for the efficient development of improved rice technologies and new rice area, the effective transfer of technologies, and other profitable investments aimed at supporting sustainable rice production. Rice production environments have evolved substantially during the past 30 years. Factors affecting rice production and changes in these factors therefore need to be considered in any new characterization and classification. Future activities in the characterization and classification of rice production environments also need to consider issues such as universality and compatibility, completeness, objectiveness, and the costs and scales of the exercises.

Characterizing environments for sustainable rice production 49 ucture.

, soil

, period of flooding

, conservation of natural resources and biodiversity, land degradation, land conservationbiodiversity, , resourcesand natural of

op pests and animal diseases.

oecological potential and use of irrigation water resources, drought and flood

farming systems and cropping inputs-outputs, intensity, crop varieties, and cropping calendar. Quantification of constraints.

Selection of key areas.

level), crops and crop rotation, socioeconomic factors (market, credit, etc.), and infrastr Selection of inland valleys.

fertility soil physics (infiltration, permeability), flooding and groundwater dynamics, and toxicity,

and shallow groundwater, size of watersheds, land-use ratio (per land settlement and at valley and shallow groundwater,

landform, lithology, drainage density, major upland soils, major land use, and population density. population and use, land major soils, upland major density, drainage lithology, landform,

water pollutants, low level of cr

public awareness.

regional and global institutions.

iable crop production systems, food requirements, economic and social needs satisfied, awareness by

Scale Unit of analysis Objective

resolution

5,000,000 water pollution, population growthclimate change and and food agricultural security, potential, awareness of

10,000

to 5,000,000 zone regions (landform and lithology).

50,000

250,000

Scale/spatial Issue

1:25,000 to Key area Characterization of key valley systems based on soils and valley morphology

level

(site-specific)

(local level) 50,000 farmers.

subnational) 2,500,000security food and production food risks,

Table 7. Levels and scales for agroecological characterization of inland valleys. Table

Characterization level

Macro level 1:1,000,000land of Agroecologicalbasis the on units agroecological into subdivided zones agroecological of Characterization

Reconnaissance 1:100,000 to Country Characterization of agroecological subunits on the basis of precipitation, length of humid period,

Semidetailed level

Detailed level 1:5,000 to Inland valley Characterization of inland valleys on the basis of variation of soils and valley morpholog

Source: Andriesse et al (1994).

Table 8. Scales of land-use and water-use planning and management. 8. Scales of land-use and water-use Table

Level of analysis

Field/production unit < 1:5,000 Productive crops and animals, conservation of soil and water, high levels of soil fertility, low level of soil and

Farm/village 1:1,000 to V

Country (national/ 1:25,000 to Judicious development of agr

Continent/world 1:1,000,000 tosharing, water and management basin water biodiversity, dessertification,conservationof and degradation Land

Source: Koohafkan et al (1998).

50 Van Nguu Nguyen References Albergel J, Lamachere JM, Lidon B, Mokadem A, Vandriel W. 1993. Mise en valeur agricole des bas-fonds au Sahel – Typologie, fonctionnement hydrologique, potentialités agricoles. Rapport final d’un projet CORAF-R3S. Ouagadougou, Burkina Faso. 335 p. Andriesse W, Fresco LO. 1991. Characterization of rice growing environments in West Africa. Agric. Ecosyst. Environ. 33:377-395. Andriesse W, Fresco LO, van Duivenbooden N, Windmeijer PN. 1994. Multi-scale characterization of inland valley agro-ecosystems in West Africa. Neth. J. Agric. Sci. 42(2):159- 179. Antoine J. 1994. Linking geographic information systems and FAO’s agro-ecological zone models for land resource appraisal. In: World Soil Resources Report 75–AEZ in Asia. Rome (Italy): Food and Agriculture Organization of the United Nations. p 35-52. Becker L, Diallo R. 1992. Characterization and classification of rice agro-ecosystems in Côte d’Ivoire. Bouaké (Côte d’Ivoire): West Africa Rice Development Association. 252 p. Bowles EJ, Garrity DP. 1988. Development of a comprehensive classification of rice ecosystems. Manila (Philippines): International Rice Research Institute. Cardenas AC, Dilag RT, Pantastico Jr. EB, Haws LD. 1980. An approach to rainfed farming. Laguna (Philippines): Philippine Council for Agriculture and Resources Research and Development. 27 p. Cassman KG, Pingali PL. 1995. Intensification of irrigated rice systems: learning from the past to meet future challenges. GeoJournal 35:299-305. Cassman KG, Olk DC, Dobermann A. 1997. Scientific evidence of yield and productivity declines in irrigated rice systems of tropical Asia. IRC Newsl. 46:7-18. Chandler Jr. RF. 1979. Rice in the tropics: a guide to the development of the national programme. Boulder, Colorado (USA): Westview Press. 256 p. FAO (Food and Agriculture Organization of the United Nations). 1976. Framework for land evaluation. Rome (Italy): FAO. FAO (Food and Agriculture Organization of the United Nations). 1978. World Soil Resources Report 48/1 – Africa. Rome (Italy): FAO. FAO (Food and Agriculture Organization of the United Nations). 1980. World Soil Resources Report 48/4 – Asia. Rome (Italy): FAO. FAO (Food and Agriculture Organization of the United Nations). 1981. World Soil Resources Report 48/3 – Central and South America. Rome (Italy): FAO. FAO (Food and Agriculture Organization of the United Nations). 1994. World Soil Resources Report 75 – AEZ in Asia. Rome (Italy): FAO. FAO (Food and Agriculture Organization of the United Nations). 1997. Trends of yield and productivity of modern rice in irrigated rice systems in Asia. IRC Newsl. 46:19-28. Huke RE. 1982a. Map of South and Southeast Asia. Manila (Philippines): International Rice Research Institute. Huke RE. 1982b. Rice area by type of culture: South, Southeast, and East Asia. Manila (Philip- pines): International Rice Research Institute. IRRI (International Rice Research Institute). 1974. An agro-climatic classification for evaluating cropping systems potentials in Southeast Asian rice-growing regions. Los Baños (Philippines): IRRI. 10 p. IRRI (International Rice Research Institute). 1984. International terminology of rice growing environments. Manila (Philippines): IRRI.

Characterizing environments for sustainable rice production 51 Jamin JY, Andriesse W. 1993. Discussion synthesis. In: Proc. 1st Annual Workshop of Inland Valley Consortium, 8-10 Oct 1993. Bouaké (Côte d’Ivoire): West Africa Rice Develop- ment Association. p 8-16. Karim Z. 1994. Cropping systems based fertilizer recommendations by agro-ecological zones in Bangladesh. In: World Soil Resources Report 75 – AEZ in Asia. Rome (Italy): Food and Agriculture Organization of the United Nations. p 53-77. Khush GS, Bennett J, Datta SK, Brar DS, Li Z. 1999. Advances in rice genetics and biotechnology. In: Proceedings of the 19th Session of the IRC, 7-9 Sep 1998, Cairo, Egypt. Rome (Italy): Food and Agriculture Organization of the United Nations. p 64-76. Koohafhan P, Nachtengale F, Antoine J. 1998. Use of agro-ecological zones and resources management domains for sustainable management of African wetlands. In: Proceedings of a sub-regional consultation workshop on Wetland Characterization and Classification for Sustainable Agricultural Development, 3-6 Dec 1997, Harare, Zimbabwe. p 107- 132. Legoupil JC, Bidon B. 1993. What level of water control for inland valley intensification in West Africa? In: Proceedings of the first Workshop of the IVC held from 8-10 Jun 1993. Bouaké (Côte d’Ivoire): West Africa Rice Development Association. p 45-60. Nguyen VN. 1998. Factors affecting wetland rice production and the classification of wetland for agricultural development. In: Proceedings of a sub-regional consultation workshop on Wetland Characterization and Classification for Sustainable Agricultural Develop- ment, 3-6 Dec 1997, Harare, Zimbabwe. p 175-90. Oldeman LR. 1974. An agro-climatic classification for evaluation of cropping systems in South- east Asia. Paper presented at FAO/UNDP international expert consultation on the use of improved technology for food production in rainfed areas of tropical Asia held at Hyderabad, India, 24-30 Nov; Khon Kaen, Thailand, 1-7 Dec; and Kuala Lumpur, Malaysia, 8-13 Dec 1974. Oldeman LR, Frere M. 1982. A study of the agro-climatology of the humid tropics of Southeast Asia. Rome (Italy): FAO. Pardey PG, Wood SR. 1991. Targetting research by agricultural environments. Chapter 31 in: Anderson JR, editor. Agricultural technology policy: issues for the international community. Wallingford (UK): CAB International. Penning de Vries EWT, Jansen DM, ten Berge HFM, Bakema A. 1989. Simulation of agroecological processes of growth in several annual crops. Manila (Philippines): Inter- national Rice Research Institute. Pingali PL, Rosegrant MW. 1994. Confronting the environmental consequences of the green revolution. In: Proceedings of the Eighteenth Session of the International Rice Commis- sion, 5-9 Sep 1994, Rome, Italy. p 59-69. Pingali PL, Hossain M, Gerpacio RV. 1997. Asian rice bowls: the returning crisis? Wallingford (UK): CAB International. Setter TL, Ingram KT, Tuong TP. 1995. Environmental characterization requirement for strategic research in rice grown under adverse conditions of drought, flooding, or salinity. In: Ingram KT, editor. Rainfed lowland rice: agricultural research for high-risk environments. Manila (Philippines): International Rice Research Institute. p 3-18. Ton That T. 1982. Potentialities and constraints of rainfed lowland rice development in tropical Africa. IRC Newsl. 31(2):1-6. Tran DV, Nguyen VN. 1998. Global hybrid rice: progress, issues and challenges. IRC Newsl. 47:16-28.

52 Van Nguu Nguyen Uehara G, Tsuji GY. 1991. Progress in crop modelling in IBSNAT project. In: Muchow RC, Bellamy JA, editors. Climate risk in crop production: models and management for the semi-arid tropics and subtropics. Wallingford (UK): CAB International. p 143-156. Windmeijjer PN, Andriesse W. 1993. Inland valleys in West Africa: an agro-ecological characterization of rice-growing environments. The Netherlands: International Institute for Land Reclamation and Improvement. Wood SR, Pardey PG. 1993. Agro-ecological dimensions of evaluating and prioritizing research from a regional perspective: Latin America and the Caribbean. The Hague (The Netherlands): International Service of National Agricultural Research.

Notes Author’s address: Agricultural Officer, Crop and Grassland Service, Plant Production and Pro- tection Division, Agriculture Department, Food and Agriculture Organization of the United Nations, Vialle delle Terme di Caracalla, 00100 Rome, Italy. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Characterizing environments for sustainable rice production 53 Tools and methodologies for biophysical characterization

Effect of climate, agrohydrology, and management . . . 55 56 Boling et al Effect of climate, agrohydrology, and management on rainfed rice production in Central Java, Indonesia: a modeling approach

A. Boling, T.P. Tuong, B.A.M. Bouman, M.V.R. Murty, and S.Y. Jatmiko

The typical rainfed cropping system in Central Java includes a dry-seeded rice crop grown from November to February (gogorancah), followed by a transplanted rice crop from March to June (walik jerami). Earlier studies showed that the yield of the walik jerami crop was lower and less stable than that of the gogorancah crop. This study assessed the climatic and agrohydrologic (groundwater depth) constraints to rice production and explored management strategies to increase the yield and yield stability of the double-rice cropping system using the crop growth simulation model ORYZA. The model was validated with data of field experiments in 1995-96 in Jakenan. Long- term simulation of potential and rainfed rice yield of cultivar IR64 was carried out on a 15-d planting interval for the period 1977-98. Three water table depth scenarios (medium, shallow, and deep), which were derived from 1995 to 1999 measurements, were used. The average simulated potential yield of walik jerami rice (about 7 t ha–1) was higher than that of gogorancah rice (about 6 t ha–1), indicating that radiation and temperature are not the determinants of the observed relatively low yields of walik jerami rice. Simulated yields of rainfed rice sown with a shallow water table depth in mid-November- February equaled the potential yield. Rainfed rice yield was reduced by 45% in the medium water table scenario and by 70% in the deep water table scenario. With medium water table depths, simulated rainfed yields of walik jerami crops declined sharply if planted later than early March. Supplemental irrigation increased the yields of rainfed walik jerami crops. The combined yields of gogorancah and walik jerami rice could be increased by using shorter- duration varieties. The results highlight the critical planting dates of the walik jerami crop, and indicate the potentials of using on-farm reservoirs and of growing shorter-duration varieties to increase the combined yields and yield stability in the area.

Effect of climate, agrohydrology, and management . . . 57 Rainfed lowland rice in Central Java covers about 293,600 ha, which is equivalent to 30% of the rice area (Amien and Las, this volume). In this area, farmers practice a high degree of crop intensification. The common cropping pattern includes two rainfed rice crops (see Fig. 1A). At the onset of the rainy season, a dry-seeded rice crop is grown, called gogorancah. Immediately after the harvest of gogorancah, a second, transplanted rice crop is grown under minimum tillage in submerged conditions. This particular cultivation of the second rice crop is called walik jerami. If residual soil moisture is still adequate, or where irrigation from shallow groundwater or from on- farm water reservoirs is possible, farmers may afterwards grow an upland crop in the dry season (called palawija). Earlier studies showed that the average yield of gogorancah rice is 4–6 t ha–1, whereas walik jerami yields are only 1.5–3.0 t ha–1 (Fagi 1995, Mamaril et al 1995, Wihardjaka et al 1998). According to Fagi (1995), Mamaril et al (1995), and Wihardjaka et al (1998), the relatively low and unstable yield of the walik jerami crop is attributed

A Cropping system Dry-seeded rice Transplanted rice Upland (gogorancah) (walik jerami) crop (palawija)

Solar radiation (MJ m–2 d–1) 22 B 20 18 16 14 12 10 8 Rainfall (mm 10 d–1) 160 C 140 Exceedance probability (P) P = 0.20 120 P = 0.50 100 P = 0.80 80 60 40 20 0 ONDJ FMAMJ JASO Month Fig. 1. Cropping system (A), daily solar radiation (B), and 10-d rainfall (C) in Jakenan, Central Java, Indonesia.

58 Boling et al to water deficit because the crop’s reproductive stage coincides with the recession of rainfall in the area. Wihardjaka et al (1998) further attributed the higher yield of gogorancah crops to more favorable temperature and radiation regimes. To date, the quantitative effect of temperature, radiation, and rainfall on yield of both the gogorancah and the walik jerami crop in Central Java has yet to be determined. The rainfed lowland areas in Central Java are characterized by an undulating landscape. In these areas, the water table depth probably varies across the toposequence. Shallow water tables at the bottom of the toposequence may contribute significantly to the water requirements of the rice crop via capillary rise. On the other hand, deep water tables at the top of the toposequence may have a negligible contribution to the crop’s water requirements. Therefore, the position of a rice field in the landscape may have a pronounced effect on rice yields. Pests and diseases may also depress rice yields. Several management strategies that alleviate the effect of drought in the area can be explored. Planting dates of both the gogorancah and the walik jerami crops can be adapted so that the combined yields are maximized by avoiding the planting dates that expose the crops to adverse conditions. The recent introduction of on-farm water reservoirs (Syamsiah et al 1994) offers scope for supplemental irrigation, especially to alleviate drought during the reproductive stage of the walik jerami crop at the recession of the rainy season. The use of shorter-duration varieties could minimize the exposure to drought of the walik jerami crop during its reproductive stage since shorter-duration varieties advance the flowering date compared with the commonly used varieties. The relative increase in rice yield because of drought escape, however, may be offset by the lower yield potentials of shorter-duration crops. The impact of the above drought alleviation strategies remains to be quantified in Central Java. Simulation models have been used to evaluate rainfed rice ecosystems and to explore management strategies for constraint alleviation (e.g., Wopereis et al 1995, Jongdee et al 1997). Our study uses the ecophysiological crop growth model ORYZA to (1) assess the climatic and agrohydrologic (groundwater depth) constraints to rainfed rice production in Central Java, and (2) explore the efficacy of management strategies aimed at increasing yield and yield stability of the double-rice cropping system.

Materials and methods Site description The study was carried out at the Jakenan Experiment Station (6°45′S, 111°10′E, 7 m above sea level). The landscape is undulating. The soil is alluvial, with 20-cm light- textured surface soil (44% sand, 46% silt, 11% clay) and a clayey 21–40-cm subsurface layer (35% sand, 34% silt, 31% clay). The top 20-cm soil layer has a dry bulk density of 1.46 g cm–3, low organic carbon (0.41%), relatively low CEC (3.48 meq 100 g–1), and low exchangeable bases (0.03 K, 2.80 Ca, 0.22 meq Mg 100 g–1). The experimental site represents a rainfed lowland area that covers about 150,000 ha in Central Java (Mamaril et al 1995).

Effect of climate, agrohydrology, and management . . . 59 According to Oldeman (1975), the climate of the site can be classified as D3. Maximum temperature is 31.7 ± 0.1 °C and minimum temperature is 23.5 ± 0.1 °C throughout the year. Solar radiation is low from December to February and high from August to October (Fig. 1B). The rainy season usually starts in October, peaks in January, and ends in May or June (Fig. 1C). The average annual rainfall (1953-98) is 1,540 mm, with 1,050 mm falling in the 5-mo period from November to March. The area has a rather long (≥5 mo) growing season (monthly rainfall exceeds evapotranspiration, Garrity et al 1986).

Experiments Field experiments using rice variety IR64 were conducted in April-July 1995 and March-July 1996 at the Jakenan Experiment Station for model calibration and evaluation. In 1995, rice was transplanted on 8 April and treatments were laid out in a split- plot design with four replications using two water treatments in the main plot and three tillage treatments in the subplots. The water treatments were full irrigation and no irrigation (rainfed). The main plots were lined with polyethylene sheets up to 40- cm depth to minimize subsurface lateral water flow. The tillage treatments were normal tillage (hoed once to 10-cm depth), deep tillage (hoed twice to 30-cm depth), and deep tillage with puddling (puddled after hoeing). In 1996, the experiment included two water treatments (fully irrigated and rainfed) in the main plot, two tillage (normal and deep) treatments in the subplots, and two transplanting dates (10 March and 8 April) in the sub-subplots. The staggered transplanting dates were used to expose the rice to different levels of drought stress. A basal fertilizer equivalent to 30 N, 22.5 P, 30 K, and 20 kg S ha–1 was applied a day before planting. Additional fertilizer equivalent to 60 N and 30 kg K ha–1 was applied at maximum tillering, 30 kg N ha–1 at panicle initiation, and 30 kg K ha–1 at flowering. Hand weeding and pesticide spray- ing were used to minimize pest damage. Daily rainfall, solar radiation, maximum and minimum temperature, relative humidity, and wind speed were measured at the Jakenan weather station. We measured biomass and its partitioning (leaves, culm and sheath, and panicle) at 30, 45, and 66 d after transplanting (DAT), at flowering, and at physiological maturity. Rice yield was determined from an 8-m2 sampling area at harvest. During the periods when there was no water standing in the fields, soil water potentials in the rainfed plots were measured daily using tensiometers installed at 5-, 10-, 20-, and 30-cm depth. The soil water content at the same depths was measured when soil water potentials were beyond the air entry values (approximately 60– 80 kPa) of the tensiometers. The groundwater table for the experiments was measured daily in four of the rainfed plots using 2.5-cm-diameter, 150-cm-long PVC tubes perforated with 3-mm- diameter holes along a 100-cm length from the bottom. In addition, we used the groundwater table measurements during the period April-June 1997, December 1997-June 1998, and November 1998-February 1999 to characterize the water table fluctuation. Since we measured the groundwater table depths only during the crop growth period, the measuring period was not continuous. The number of measurements thus differed

60 Boling et al for different periods of the year: four sets of data for April-May, three sets for June, two sets for December-January, and one set for October, November, February, and March. No data were measured during the July-October peak of the dry season. Dur- ing this period, the water table was assumed to be 150 cm below the soil surface. To quantify the fluctuation of the groundwater table depths, we constructed the shallow, medium, and deep water table scenarios as follows: 1. For periods with more than one measurement, the means of the measured values were used to represent the medium water table depths, the mean + standard error of measured values to represent the shallow depths, and mean – standard error for the deep water table. The medium water table scenario was presented by the line connecting the calculated medium depths. The enveloping line connecting the mean + standard error represented the shallow water table scenario, and the line connecting the mean – standard error the deep water scenario. 2. For periods with only one measurement, the measured values were used for the medium water table scenario. The line that connects the shallow water table depths of the adjacent periods (with more than one measurement) was used to represent the shallow water table scenario and the line that connects the deep water table represents the deep water table scenarios. 3. For the period at the peak of the dry season (i.e., July-October), the water table was assumed to be at 150 cm below the soil surface in the three water table scenarios. The shallow, medium, and deep water table scenarios were used as inputs to the model simulating rainfed rice production.

The ORYZA model ORYZA is an updated and integrated version of the ecophysiological models ORYZA1 (Kropff et al 1994) and ORYZA_W (Wopereis et al 1996). ORYZA simulates crop growth and development of lowland rice in potential and water-limited production situations. Under potential situations, water and nutrients are in ample supply and growth rates are determined by weather conditions only (radiation and temperature). Under water-limited production, growth is limited by water shortage in at least part of the growing period, but nutrients are still considered to be in ample supply. In both production situations, the crop is supposed to be well protected against pests, diseases, and weeds. In our study, water-limited production is synonymous with rainfed production. ORYZA consists of separate modules to calculate growth and development of the crop, evapotranspiration, effects of drought on growth and development, and the water balance of puddled and nonpuddled soils. The crop module is a photosynthesis- driven model. On each day, the solar radiation profile in the canopy is calculated on the basis of incident radiation, leaf area index, and the vertical distribution of leaves in the canopy. The daily canopy assimilation rate is calculated by integrating the calculated photosynthesis of single leaves over the height of the canopy and over the day. After subtracting respiration requirements and accounting for losses from the

Effect of climate, agrohydrology, and management . . . 61 conversion of carbohydrates into structural dry matter, the net daily growth rate is obtained. The dry matter produced is partitioned among the various plant organs according to the stage of development of the crop, which is tracked as a function of ambient mean daily air temperature. Daily evapotranspiration rates are calculated using modified Penman equations (Van Kraalingen and Stol 1997). Effects of drought (defined as the condition when soil water contents are lower than saturation) on crop growth and development include leaf rolling, reduced leaf growth rate, accelerated leaf senescence, reduced evapotranspiration and photosynthesis, reduced development rate, reduced sink size, and reduced spikelet fertility. The drought stress response functions were derived from pot experiments (Wopereis 1993, Wopereis et al 1996; unpublished experiments by the authors) and the literature (Turner et al 1986). The soil water balance, called PADDY, is a universal multiple-layer model that can be used for both puddled and nonpuddled conditions, for freely draining soils, and for soils with impeded drainage (Wopereis et al 1996). Capillary rise from groundwater into the rooting zone of the soil profile is taken into account. ORYZA requires input data on crop characteristics, soil properties, management, daily water table depth, and daily weather. ORYZA was extensively calibrated and evaluated for potential production situations (Kropff et al 1994, Matthews et al 1994, 1995). The model was validated for rainfed production at the IRRI farm in Los Baños, Laguna, Philippines (Wopereis 1993, Wopereis et al 1996).

Model simulations Four series of simulation runs were made. The first series consists of model calibration and performance evaluation. ORYZA was parameterized and evaluated using the 1995-96 experimental data and the observed daily weather data. For the crop, standard physiological characteristics for IR64 were used (IRRI, unpublished data set). The irrigated treatments of both 1995 and 1996—supposedly representative for potential production situations—were used to derive the required empirical parameters of development rate, assimilate partitioning, and leaf death rate. The rainfed treatments were used for evaluation (independent data set). For all simulations, measured soil physical properties and water table depths were used as inputs in ORYZA’s soil water balance. The second series of simulations concerned the exploration of long-term rice yield under potential and rainfed conditions. Potential and rainfed yields were calculated for different planting dates at 15-d intervals using historical weather data of 1977-98. To keep the simulation as much as possible in line with farmers’ practices, simulation from January to May was performed for transplanted rice, and for the rest of the year for direct-seeded rice. The evaluation of potential rice yield sheds light on the (interactive) contribution of solar radiation and temperature to yield formation in the gogorancah and walik jerami crops. The simulation of rainfed rice yields quantifies the effect of drought on yield reduction and allows us to quantify critical dates for planting of the walik jerami crop in relation to the recession of the rains at the end of the rainy season. The rainfed rice yields were calculated for three groundwater table

62 Boling et al depth scenarios (shallow, medium, and deep) to quantify the effect of groundwater contribution on yield formation. These scenarios were derived from the groundwater depth measurements in 1995-99 (see “Experiments”). The third simulation series focused on the sensitivity of the gogorancah crop to initial soil moisture at sowing. Transplanting of walik jerami rice is commonly done in saturated soil at the peak of the rainy season. Therefore, model simulations with ORYZA can safely begin with saturated soil conditions. On the other hand, the seeding of the gogorancah crop starts at the beginning of the rainy season when the initial soil moisture varies across years depending on rainfall distribution. The initial soil moisture status during the establishment of the gogorancah crop may affect the vegetative growth and yield of rice crops grown in the area, and therefore warrants a sensitivity analysis with ORYZA. Four scenarios, that is, initial soil moisture status at saturation, field capacity, –100 kPa soil water potential, and at permanent wilting point, were simulated. The fourth simulation set explored the effect of supplementary irrigation and of the use of relatively short-duration varieties on rice yield. Three supplementary irrigation scenarios were constructed: (1) irrigation water was applied whenever the top- 3 –3 soil water content fell to field capacity (= 0.34 cm cm , scenario I1), (2) a daily amount of 7.5 mm water was applied from panicle initiation to crop maturity (scenario I2), and (3) a daily amount of 3.3 mm water was applied from panicle initiation to crop maturity (scenario I3). To study the effect of crop duration on rice yield, simulations were performed, besides for IR64 (V1), for two hypothetical varieties with 5- d (V2) and 10-d (V3) reduction in growth duration compared with IR64. The reduction in growth duration was accomplished by diminishing the parameterized value for the development rate of IR64. In the supplementary irrigation scenarios, irrigated yield was simulated (i.e., rainfed plus supplementary irrigation) and, in the crop duration scenarios, both potential and rainfed situations were simulated, with 15-d-interval planting dates and using historical weather data of 1977-98.

Results and discussion Model evaluation Figure 2 gives simulated and measured total dry canopy biomass in 1996. The simulated potential biomass falls mostly within the standard error range of measured biomass in the highest yielding irrigated plot (deep tillage treatment, supposedly the best treatment approaching potential production situations). It is to be expected that simulated potential biomass values are higher than measured biomass values in the irrigated treatments since it is extremely difficult to realize potential growth conditions in field experiments (e.g., Kropff et al 1994). Some yield reduction might have occurred because of nutrient limitation, pests, and diseases. Even so, simulated rainfed biomass values were generally somewhat higher than measured biomass values in the two rainfed treatments with different sowing dates. ORYZA mostly performed about the same in the simulation of yield as in the simulation of total biomass (data not shown). Only under severe drought stress occur-

Effect of climate, agrohydrology, and management . . . 63 Dry matter (kg ha–1) 16,000 Irrigated, measured Early rainfed, measured 12,000 Late rainfed, measured Potential, simulated Early rainfed, simulated Late rainfed, simulated 8,000

4,000

0 FMA MJ J A Month Fig. 2. Simulated and measured aboveground dry matter (mean ± SE) during the 1996 walik jerami season in Jakenan. The solid line is the simulated potential production, which may be compared with the measurements of the best irrigated treatment. The broken and dotted lines are the simulated rainfed productions using early and late sowing. The measurements of the rainfed treatments include two tillage treatments. ring during the reproductive phase did ORYZA overestimate grain yield by a factor of 4 to 10 (measured values were 200–500 kg ha–1 while simulated values were 2,000 kg ha–1). Whereas ORYZA partitions all assimilates produced in the reproductive phase to the storage organs, it was observed that, in reality, when the stress was relieved, the crop produced new tillers, indicating that not all assimilates went to the grains as modeled. Though a statistical analysis of model performance still needs to be made, the preliminary results indicate that ORYZA performed sufficiently well for the purposes of our study.

Water table fluctuation Figure 3 presents the shallow, medium, and deep water table scenarios constructed from measurements in six cropping seasons of 1995-99. Large standard errors (in periods with more than one measured value) indicated that the water table depth varied widely in different years. The large standard errors, however, may be attributed to the small number of available data. The mean water table depth fluctuated from 10 cm below the soil surface in December to around 30–60 cm in January and in April. The mean water table started to drop off from the middle of May to 70–100 cm in June. This conforms to the declining rainfall in these periods and suggests that, starting from June, the contribution of the capillary water from the groundwater to the root zone decreased substantially. The water table depth in February-March appeared very deep and might not adequately reflect the “medium” value because we had only one set of data that were measured in 1998, a year with particularly low rainfall.

64 Boling et al Water table depth (cm) 50 Shallow water table Medium water table Deep water table 0

–50

–100

–150 OND J FMAMJ J ASO Month Fig. 3. Shallow, medium, and deep water table scenarios used in ORYZA simulations to explore rainfed rice yields in Jakenan. The method of deriving the scenarios from 1995-98 measured data is described in the text. Solid triangles indicate the means of the measured data (or the actual data when there was only one water table value measured). Vertical bars indicate standard errors of the means (when there was more than one water table value).

Yield exploration Potential yield. The simulated, long-term average potential rice yield is shown in Figure 4 as a function of date of sowing. The potential yields ranged from 6 to 8 t ha–1, and were within the yield range of well-managed irrigated rice as observed in Bogor, West Java, and in Genteng, East Java, by Makarim and Las (1993). The standard errors of the average values were low, indicating little variation in potential yield across years. Rice sown in the typical gogorancah period (November-December) had a lower potential yield (on average 6 t ha–1) than rice planted in the typical walik jerami period (February-March, average yield = 7 t ha–1). This lower potential yield in the gogorancah period was caused mainly by the low radiation during the crop’s reproductive stage (Fig. 1B). The variation in potential yield for crops sown in No- vember-December, however, was relatively larger than for crops sown in February- March. Rainfed yield. The simulated, long-term average rainfed rice yield is shown in Figure 4 as a function of date of sowing for situations with a shallow, medium, and deep groundwater table. The yield of rainfed rice sown from mid-November to the end of March differed significantly among the three water table depth scenarios. Dur- ing this period, rainfed yields reached the potential yield level with the shallow water table, but decreased with deeper water table depths. For crops sown at the end of the rainy season and in the dry season (roughly April-November), the yield was very low and yield differences among the water table scenarios disappeared. The low yields in this season are attributed to inadequate water supply from rainfall and from the ground-

Effect of climate, agrohydrology, and management . . . 65 Simulated yield (kg ha–1) 12,000 Rainfed, shallow water table 10,000 Rainfed, medium water table Rainfed, deepwater table Potential 8,000

6,000

4,000

2,000

Simulation for Simulation for Simulation for dry-seeded rice transplanted rice dry-seeded rice

OND J FMAM J J A SO Day of seeding by month Fig. 4. Simulated potential and rainfed rice yield (mean ± SE) in Jakenan, Central Java, Indonesia.

water. Considering the overestimation of simulated yield under severe drought conditions by ORYZA (see “Model evaluation”), we can in fact assume that simulated rainfed yields in the dry season were virtually zero. In the shallow water table scenario, the yield of crops sown in the typical gogorancah period (November-December) was significantly lower than that of crops sown in the typical walik jerami period (February-March) (similar to potential yield differences; see above). However, in the deep water table scenario, the opposite is true: rainfed rice yields are higher in the typical gogorancah period than in the typical walik jerami period. This latter is consistent with field studies reported by Fagi (1995), Mamaril et al (1995), and Wihardjaka et al (1998).

Sensitivity analysis Figure 5 shows the simulated rainfed yield of gogorancah rice sown in October- December with medium water table depths at four levels of initial soil moisture content. In general, rice sown in soil with a high initial moisture content had a high yield. The observed yield differences among initial soil moisture statuses decreased as the planting dates moved from October to December. Simulated yield of rainfed rice crops sown at a given soil moisture differed among seeding dates. Yields of rainfed rice sown in a low soil moisture status in October-early November were significantly lower than those of crops sown from mid-November onward. The yield difference of rainfed rice for crops sown before and after mid-November, however, was considerably lower in soils with high initial moisture contents. For crops sown from mid-

66 Boling et al Simulated yield (kg ha–1) 6,000

5,000

4,000

3,000

Initial water status: 2,000 Saturation Field capacity Soil water potential = –100 kPa 1,000 Permanent wilting point

0 SOND JF Day of seeding by month

Fig. 5. Simulated yield (mean ± SE) of dry-seeded rice with a medium water table depth as a function of sowing date at four different initial soil moisture contents in Jakenan.

November onward, the relatively high yields obtained regardless of initial soil moisture status indicate that rice yield is stable if rice is sown after early November.

Management options to increase yield Establishment date. Figure 4 shows that the three water table scenarios caused the yield of rainfed crops to start to decline at different seeding dates. Yields in the shallow water table scenario started to decline in mid-March, whereas yields in the deep water table scenario started to decline in mid-January (Fig. 4). The yield decline was opposite to the yield-increasing effect of increasing radiation levels toward the recession of the rainy season (Fig. 1B) and was attributed to water stress because of the decreasing rainfall during the reproductive stage of the crops (Fig. 1C). The observed decline in simulated yield of rainfed crops planted toward the recession of the rainy season (Fig. 4) highlighted the importance of planting dates of the walik jerami crop. The planting date of the walik jerami crop, however, depends on the harvest date of the preceding gogorancah crop. A late onset of the rains may delay the seeding of the gogorancah crop, which consequently delays the establishment of the walik jerami crop. If the delay is beyond a critical transplanting date, the farmer may choose whether to forgo the walik jerami crop or risk having a low walik jerami yield. From the simulation results, the critical sowing date of the walik jerami crop with a shallow water table was mid-March, with a medium water table mid- February, and with a deep water table mid-January. Supplementary irrigation. The simulated rice yield in the three supplementary irrigation scenarios with medium water table depth and initial soil water potential of –100 kPa is presented in Figure 6A. For crops sown in the typical walik jerami period

Effect of climate, agrohydrology, and management . . . 67 Fig. 6. Simulated yield of rainfed and irrigated rice (A) and amount of irrigation water in three irrigation scenarios with medium water table depth (B) as a function of sowing date. I1 = irrigation when moisture of the topsoil falls below field 3 –3 capacity (0.34 cm cm ), I2 = daily irrigation with 7.5 mm from panicle initiation (PI) to maturity (M), I3 = daily irrigation with 3.3 mm from PI to M.

(February-March), simulated yields in irrigation scenarios I1, I2, and I3 were on average about 2.1–3.2 t ha–1, 1.5–3.1 t ha–1, and 0.5–1.2 t ha–1 higher, respectively, than the simulated yields under purely rainfed conditions. Figure 6B shows the average total amount of irrigation water applied in the three irrigation scenarios. The irrigation water for crops sown in the walik jerami period in scenario I1 varied from 347 mm with sowing in mid-February to 555 mm with sowing in mid-April. On the other hand, the irrigation water for crops sown from mid-February to mid-April in scenarios I2 and I3 is fixed at 345 mm and 153 mm, respectively. The increase in rice yields for every m3 of irrigation water applied in the three irrigation scenarios is presented in Figure 7. The maximum yield increases in the three scenarios occurred for crops sown in early March. For crops sown from mid-

68 Boling et al Perception, understanding, and mapping of soil variability in the rainfed lowlands of northeast Thailand

T. Oberthür and S.P. Kam

Previous studies that assessed the soil resources in northeast Thailand create the impression that soils are universally infertile because of their light texture and low inherent nutrient contents. In reality, variations within micro- catchments are sufficient to influence land productivity factors over short distances. This chapter describes a study carried out to examine soil variability in the rainfed lowlands of northeast Thailand and to develop quantitative methods of spatial prediction that provide useful soil resource information for agronomic management. Our methodology is based on geostatistical mapping, using soil data collected in soil surveys supplemented with less costly auxiliary information. The auxiliary information included knowledge of local farmers and soil experts about soil-landscape relations (providing a 5- category classification referred to in this study as the updated farmers’ field classification, or UFFC). A soil sampling scheme was devised equivalent to that employed for producing soil maps at a mapping scale of 1:50,000 to 1:100,000. Hence, application of the proposed method in areas that have been mapped at these scales will only require a reanalysis of existing data and collection of complementary data at lower costs. Soils of loamy sand or loam texture were found in many parts of the study region, indicating that the soil texture of these soils is not altogether unsuitable for rice cultivation. Conventional statistical analysis of the soil survey data reveals very high spatial variability that cannot be ignored. Many topsoil and subsoil properties related to nutrient availability (Bray-II P, cation exchange capacity, exchangeable bases) had large coefficients of variation (CVs), including those properties that are considered relatively stable, such as organic matter and clay content. Further statistical analysis shows that this soil map accounts for less than only 8% of the variance in measured soil properties, not enough to provide agronomically important information. There was no discernible distinction in soil fertility characteristics among the mapped soil types. The UFFC accounted for higher proportions of the total variance (ranging from 0% to 43% for both soil depths), and indicated that soil productivity declines in the order alluvial fields > fields in low topographic position > fields in medium and high topographic position > fields with upper, deep sandy horizons.

Perception, understanding, and mapping of soil variability . . . 75 Geostatistical indicator approaches that can either use or ignore auxiliary information were adopted for modeling of soil heterogeneity to enable estimation and mapping of mean, median, conditional variance, and conditional CV, and also of the 0.2- and 0.8-quantiles of the conditional cumulative distribution function. Quantile maps are very useful for mapping soil properties tailored to specific land management questions. Maps produced without auxiliary information show distinct spatial distribution patterns of average clay, silt, and sand contents in relation to the regional physiography. Maps generated with auxiliary information reveal more spatial detail; texture changes gradually and follows the local topography and drainage patterns. These maps assign much land (25%) to a soil texture class that is suitable for rice production (loam and heavier). In conclusion, our results depart from the general belief that soils of the Korat Plateau are universally coarse-textured and infertile. The reliance on soil maps has contributed to the long-held views about soil texture in northeast Thailand. Our maps strongly support an alternative hypothesis of soil genesis in the Korat Plateau that combines colluvio-alluvial erosion processes over short distances and in situ soil development with long-range Quaternary alluvial sedimentation.

Understanding the heterogeneity of the environment is a cornerstone to increasing and sustaining rice productivity in the rainfed lowlands (Zeigler and Puckridge 1995). Modern techniques for generating and analyzing geographic information (geographic information systems, remote sensing, and global positioning systems) used in conjunction with systems approaches can help improve environmental characterization. So far, however, data-intensive techniques have been attempted only in irrigated systems at the field scale in the rice-growing areas of Asia (Dobermann and White 1999). Reliable quantitative data are often lacking beyond specific field locations in rainfed lowland rice areas of Southeast Asia, and this shortcoming impedes a broad-based assessment of land resources as a prerequisite for agronomic management decisions. This is particularly true for soil information. Over the past 20 years, land evaluation has benefited from the introduction of geographic information systems (GIS), which enable spatial modeling of information. GIS explicitly provide the spatial dimension for land resource assessment and the means for integrating data of different subject matters and from disparate sources. Conventional (GIS-based) approaches for deriving soil information at regional and subregional scales (1:20,000 to 1:100,000), however, suffer from the shortcoming of traditional sources of soil information, principally soil taxonomic maps. Because of their low spatial resolution, focus on pedological characterization, lack of quantitative data, large between-farm variability in crop management, and dynamic changes in many soil nutrients, existing soil maps in the developing countries of Asia often do not provide sufficient information for agronomic purposes (Oberthür et al 1996). Spe- cific spatial assessment and monitoring of soil-related constraints to nutrient uptake

76 Oberthür and Kam and rice yield at regional or more detailed scales are needed to identify problem areas as a basis for making strategic agronomic decisions. This chapter discusses this problem and suggests solutions to improve the availability of soil data and characterize the associated uncertainty whereby GIS tools are used in conjunction with related techniques such as geostatistics.

Perception of soil variability Previous studies that assessed the soil resources in northeast Thailand created the impression that soils are universally infertile because of their light texture and low inherent nutrient contents (Ragland and Boonpuckdee 1988). These generalizations might be grounded in the fact that the finite elements of the assessment were soil series or soil groups; see, for instance, Suddhiprakarn and Kheoruenromne (1987) or Patcharapreecha (1988). In reality, however, variations within micro-catchments are sufficient to influence land productivity factors over short distances. This short-range variation of soil fertility indicators within the landscape has long been recognized as important in studies on land resource assessment (Craig and Pisone 1988, Grandstaff 1988) but is often ignored or smoothed out in conventional soil maps. Acceptance of Moormann’s soil development theory (Moormann et al 1964), which implied gradual changes in soil properties over long distances, has contributed to this practice. Land formation and soil genesis in northeast Thailand have been the subject of controversial discussions. Pendleton and Montrakun (1960) favored an in situ soil development of particular soil types at certain topographic positions. Moormann et al (1964) refuted the in situ soil genesis hypothesis and linked soil formation to vast Pleistocene sediments deposited by the Mekong River and its tributaries. According to Moormann et al (1964), deposition of sediments and soil formation occurred in four distinct phases, resulting in the high, middle, and lower terraces and the present floodplain. Recently, however, several workers have independently questioned this theory. Paiboon et al (1985), Tamura (1986), Mitsuchi et al (1989), and Miura (1990) all support a combination of in situ soil development, mainly through lessivage, and soil development by colluvial processes over short distances. The topography that resembles bedrock relief and soil texture distribution within the soil matrix support this hypothesis. Research by Michael (1982), Löffler et al (1984), and Kubiniok (1990) revealed strong relationships among regional tectonics, sedimentation, climatic conditions, and soil genesis. Most soils of the Korat Plateau more likely developed on an erosional relief over different rock formations. Two relief generations, which have formed under different climatic conditions, can be distinguished. Initially, Early Ter- tiary Red Soil (Oxisols) covered most of the plateau. These soils developed under humid tropical conditions and are currently found as isolated remnants, mainly on hills. Subsequently, increasing Miocene tectonic activity destroyed the old relief. Furthermore, the climate became increasingly drier with the glacial maximum of the Late Pleistocene. Consequently, yellowish, brownish soils (Oxisols, Ultisols, Alfisols) developed on the now gently sloping landscape. Quaternary weakly developed alluvial soils (Entisols and Inceptisols) are found in the alluvial plains of the Mun River

Perception, understanding, and mapping of soil variability . . . 77 and its tributaries (Kubiniok 1990). As these geomorphologic processes led to the formation of soils with distinctively different characteristics, knowledge of soil-landscape relationships helps in a better appreciation of variation in soil properties for agronomic decision making. This is illustrated by an example using a soil texture classification for a subarea of the Ubon Ratchathani Land Reform Area in Ubon Ratchathani Province of northeast Thailand (the rectangular area in Figure 1). A soil map (Fig. 2A) of northeast Thailand at 1:50,000 scale (Changprai et al 1971) was used to derive a thematic map of soil texture by recoding the soil classes with their associated texture classes (Fig. 2B). Using this method, 93% of the mapped area would be assigned to a medium- textured class. A very different picture (Fig. 2C) emerges when soil texture was mapped using improved interpolation techniques (see detailed explanation below). This map clearly shows short-range variation in soil texture, with 40%, 36%, and 24% of the area under light, medium, and heavy texture, respectively. Insufficient spatial resolution, resulting in incorrect allocation of land to soil texture classes, however, renders the recoded soil map unsuitable to support agronomic decisions at the subregional scale. Can we then use existing soil classification systems as a basic source of soil information in the rainfed rice lands in northeast Thailand? Use would require improving the existing soil maps to obtain spatial estimates of those land characteristics that are not yet well predicted. The actual decision as to whether revision or upgrading is the preferred choice depends on the quality of the existing database, the type of problems to be solved, and available funds (Brus et al 1992). Collection of additional quantitative data is tedious and costly. Collecting qualitative, auxiliary information is less costly, but this information has been infrequently used in the past, perhaps because of the lack of appropriate data management techniques and a research approach that emphasizes quantitative data. The knowledge of local experts is particularly beneficial because of their strong bonds with the local environment. The identification of methods that integrate sparse quantitative and available auxiliary data is a prerequisite for producing meaningful soil information at a regional scale in the heterogeneous rainfed lands of northeast Thailand.

Understanding soil variability Data needs A case study was carried out within the 61,000-ha Ubon Ratchathani Land Reform Area (URLRA), located about 45 km southeast of Ubon Ratchathani City (Fig. 1) in the southwestern corner of the Korat Plateau, which is the largest plateau in Southeast Asia extensively used to grow rice (Mackill et al 1996). The URLRA, which is ad- ministered under the Thailand Agricultural Land Reform Act, straddles two districts, Amphoe Det Udom and Amphoe Nachaluai, of Ubon Ratchathani Province, and is bordered by the Lam Dom Yai River. The starting point would be a set of point-based soil data and field observations that would normally be collected in a soil survey. Previous studies have mapped soil series as a result of semidetailed surveys (Changprai

78 Oberthür and Kam Subarea

Sampling sites: Grid Transect Random

Soil types (Changprai et al 1971) Alluvial Complex Korat Korat/Phon Phisay Nam Phong Phen Phon Phisay Roi Et Roi Et/Phen Ubon

0 2 4 6 8 km

Fig. 1. The project region in the Ubon Ratchathani Land Reform Area (URLRA) in northeast Thailand.

Perception, understanding, and mapping of soil variability . . . 79 A. Soil map 1:50,0000 B. Soil texture class map C. Improved soil texture class (Changprai et al 1971) (reclassified soil map) map (soft indicator kriging of field texture)

Alluvial Complex soils Light texture Light texture Korat series Medium texture Medium texture Korat/Phon Phisay association Heavy texture Heavy texture Nam Phong series No data Phen series Phon Phisay series Roi-Et series 0 2 4 6 8 10 km Roi-Et/Phen association Ubon series

Fig. 2. Mapping soil texture classes. Map A shows the soil map of Changprai et al (1971) for an area in the Ubon Ratchathani Land Reform Area (URLRA). Map B is the soil texture class map derived by recoding map A. Map C was generated using advanced interpolation techniques (soft indicator kriging and classification). et al 1971) but the accompanying point information that is important for agronomic land assessment and management was not accessible for us. Therefore, we designed a survey scheme to describe the spatial distribution of agronomically important soil properties at greater resolution. The development of methodologies for regional characterization was the main objective in this study, and within-field variability was not considered in the survey. It is important to note that the sampling density of this survey was sufficient to produce soil maps at scales of 1:50,000 to 1:100,000. Appli- cation of the proposed method in areas that have been mapped at these scales will hence require only a reanalysis of existing data and collecting some complementary data.

Methodology for data collection and processing The sampling layout (Fig. 1) comprised three different but complementary sets. First, a sampling grid was drawn on a map and the fields nearest to the grid nodes were identified using a global positioning system (Set 1). Spacing between the grid nodes

80 Oberthür and Kam was approximately 2,000 m with a total of 91 sampling locations in the area. Posi- tional deviations from the original grid node locations were due to site inaccessibility. Second, to depict local variations, 117 sampling sites were located within 29 transects (Set 2). Each transect comprised three to five sampling locations with 50 m to 150 m spacing. Transects were positioned along the slopes of micro-catchments. Third, a validation set (Set 3) with 70 locations was obtained using a stratified random sampling design. The area was split into 14 strata of equal size and five locations were randomly selected for each stratum. For all three sampling sets, each sampling site represented a rice field. Soil samples were collected from 0–0.15-m depth (topsoil; soil monolith 0.2 × 0.2 × 0.15 m by spade) and from 0.15–0.4-m depth (subsoil; Dutch auger with 0.1-m diameter). Five soil samples were bulked, one from the center of the field and four samples within a 6-m radius around the center of the field. The samples were air-dried, ground to pass through a 2-mm sieve, and analyzed (Table 1). Qualitative soil profile descriptions were conducted on auger borings at each location. Where possible, interviews were conducted with the farmers of the sampled fields to record their perceptions on soil fertility of their fields (ranked on an ordinal scale: very low, low, medium, high, very high), the probability that water is sufficient for rice production (two, four, six, eight, or ten out of ten years), and the likelihood of iron toxicity (high, medium, low, very low, none). Interviews also revealed information about the field’s topographical position in the toposequence (low, medium, high). Farmers make use of this information and combine it with knowledge about the performance of different rice varieties to build strategies to minimize risks of crop failure. Survey observations suggested that fields with an upper sandy horizon deeper than 1.5 m and fields frequently rejuvenated by alluvial sediments of the Lam Dom Yai River should be added to the farmers’ field classification (FFC) of the three topographical positions. This updated classification is referred to as the UFFC. Exhaus- tive coverage with the UFFC was achieved in an 18,000-ha subarea of the URLRA, delineated by a rectangle in Figure 1. Scanned, pan-chromatic aerial photographs (1:4,000) for this subarea were provided by the Thailand Agricultural Land Reform Office. Within a GIS, a 250 × 250-m grid was generated for the subarea and one of the UFFC classes was assigned to each of the 2,864 grid nodes, visually aided by the scanned aerial photographs and an elevation map. The procedure was validated in 1997 when locations of 100 randomly selected grid nodes were visited in the field. Classification accuracy of 96% was obtained.

Summary statistics About 50% of the samples had more than 70% sand and less than 7% clay in the topsoil. Clay contents increased slightly in the subsoil. However, soil texture was loamy sand or loam in many parts of the study region because silt contents reached almost 20% in about 50% of the samples (topsoil and subsoil). About 25% of all topsoil samples had organic carbon contents of more than 8.2 g kg–1. Subsoil organic carbon values were very low, however, with three-quarters of the samples having less than 4.4 g kg–1. Bray-II P content was more than 1.1 mg kg–1 and more than 4 mg

Perception, understanding, and mapping of soil variability . . . 81 )

–1

)kg

–1

)kg

–1

Walkley and Black wet oxidation (Black oxidation wet Black and Walkley

)kg

–1

g )kg

–1

Saturation extract. Saturation

CEC

In %. i

) (cmol (cmol (cmol (cmol (cmol

–1

1 N ammonium acetate at pH 7 using a steam distillation technique distillation steam a using 7 pH at acetate ammonium N 1 f g

) (mg kg

e –1

) (g kg

–1

(dS m

Sand

At 1:1 soil to solution (water), measured with a glass electrode. glass a with measured (water), solution to soil 1:1 At

F at 1:10 soil to solution for 40 s (Bray and Kurtzand (Bray s 40 1945). for solution to soil 1:10 at F

Silt

5.8 12.3 56.2 4.9 0.13 2.1 0.97 0.96 0.10 0.07 0.011 0.03

0.7 0 22.7 4.2 0.10 0 0.50 0.04 0.01 0.02 0.001 0

Clay

16.1 26.1 80.5 5.3 0.20 4.4 2 3.68 0.51 0.15 0.057 0.12

13.1 19.2 67.7 5.2 0.21 3.4 1.81 2.84 0.53 0.13 0.043 0.19

10.5 17.9 70.3 5.1 0.16 3.0 1.09 1.92 0.22 0.10 0.028 0.06

56.5 46.3 96.7 8.1 0.97 14.3 14 15.36 7.36 1.48 0.412 4.80

78 48 24 9 77 6 84 94 164 88 121 268

Hydrometer method. Hydrometer

1 N ammonium acetate at pH 7 using a 1:5 soil to solution (Page 1982, p 159-165).

h a

6 1.0 0.1 23.3 3.9 0.11 0.5 0.6 0.24 0.01 0 0.001 0

13 9.4 20.5 70.1 4.8 0.24 7.0 5.2 2.18 0.31 0.11 0.039 0.07

13 7.2 18.2 73.1 4.7 0.21 6.6 4 1.68 0.19 0.09 0.028 0.05

18 77 51 21 8 42 47 92 89 136 65 94 124

0.1 N HCl + 0.03 N NH N 0.03 + HCl N 0.1

Ap f

(cm) (%) (%) (%)

i i

Estimated on auger borings. auger on Estimated

Table 1. Statistics of topsoil (0–15 cm) properties in the area (n = 278). Means, variances, and coefficients of variation (CVs) were estimated by the by estimated were (CVs) variation of coefficients and variances, Means, 278). = (n area the propertiesin cm) (0–15 topsoil of Statistics 1. Table method of moments.

Mean Minimum Lower quartile 12 5.3 12.1 60.9 4.6 0.17 5.1 3 1.04 0.11 0.07 0.015 0.03 Median Upper quartile 15 10.4 27.9 82.2 5.0 0.27 8.2 6 2.40 0.35 0.13 0.050 0.07 Maximum 21 48.4 54.1 92.5 6.4 1 28.1 56 13.12 3.37 0.59 0.217 0.63 CV

Mean Minimum Lower quartile Median Upper quartile

1965, p 1372-1376). p 1965, (Page 1982, p 893-895).

Maximum Topsoil Subsoil CV a

82 Oberthür and Kam kg–1 in 50% of the topsoil and subsoil samples, respectively. Contents of CEC and exchangeable bases were low in a large proportion of the study area. Nevertheless, –1 1N NH4-acetate extractable K, for example, was more than 0.06 cmol kg in at least 25% of all samples (Table 1). Statistical analysis of the soil samples underlined that accounting for variability is important in heterogeneous rainfed rice lands. Only a few soil properties had little variability within the study region. Many topsoil and subsoil properties related to nutrient availability (Bray-II P, CEC, exchangeable bases) had large CVs (Tables 1 and 2). Even properties that are considered relatively stable, such as organic carbon or clay content (Oberthür et al 1996), had high CVs. Compared with results published in the seminal review of Beckett and Webster (1971), CVs found for soil attributes such as texture, organic carbon content, or the base complex were generally much higher than the proposed yardstick values. Even coefficients of variation determined by Davis et al (1995) for soils similar to the soils in northeast Thailand were on average lower than the CVs presented here. However, we ask the reader to interpret the CV values with caution as the values depend on the mean of the sample. A set of observations with a mean close to zero has a larger CV than a set of observations of the same property with a higher mean. The high CVs of the bases are partly caused by this effect. The UFFC incorporates local changes in topography, yet CVs of many soil properties were only slightly lower than those based on soil series (Table 2), suggesting that steep gradients of change exist over very short distances. But unlike the mapped series of the 1971 soil map, the units of the UFFC have clear differences in their soil fertility properties. Grouping the individual samples using the UFFC shows characteristics of extremely unfavorable to favorable soil fertility. Differences between the class medians were large for most soil properties, except for soil pH, EC, and Bray-II P in the topsoil and pH and EC in the subsoil (Table 3). Class medians of soil fertility characteristics, including organic carbon, texture, CEC, and exchangeable bases, suggest that soil productivity in the region declines in the order alluvial fields > fields in low topographic position > fields in medium and high topographic position > fields with upper, deep sandy horizons. Table 4 shows that the proportion of variance in the data explained by the soil map, as measured by the complement of the relative variance RVc (Webster and Oliver 1990), was low for all the soil variables determined. The RVc ranged from 0% to 8% for topsoil and subsoil, even for relatively stable soil properties such as texture, organic carbon, and CEC. This illustrates the inability of the soil map to account for much of the variance of agronomically important soil properties (Table 4). In other words, the mapped soil units are highly variable in their indicators of soil fertility and are of limited use for agronomic interpretation. On the other hand, the higher RVc values obtained with the UFFC classification (ranging from 0% to 43% for both soil depths) show that this classification accounted better for the variability of the agronomically important soil properties, with the exception of EC and K content, than the conventional soil mapping units (Table 4). Relatively high RVc values were achieved for soil properties such as Ca and Mg.

Perception, understanding, and mapping of soil variability . . . 83 s of the of s

ABS = alluvial sites on the

MTP HTP DUSH

Categories of the UFFC

LTP

ABS

RE/PH

KO/PP RE

Mapping units of 1971 soil taxonomic map

KO = Korat (n = 98); KO/PP = Korat/Phon Phisay (n = 122); RE = Roi Et (n = 12); RE/PH = = topsoil, Roi Sub Et/Phen = (n subsoil. = 15). Top

Soil layer Top Sub Top Sub Top Sub Top Sub Top Sub Top Sub Top Sub Top Sub Top Sub

Clay 70 78 80 76 91 86 68 70 56 48 70 64 55 68 73 70 30 78 Silt 51 44 54 50 52 58 35 39 25 46 33 36 57 49 49 48 35 49 Sand 22 24 20 22 32 37 21 23 25 28 24 28 20 20 16 16 5 10 H985216 8108109877 8951251664 pH99 EC 31 79 45 73 55 33 59 89 28 16 50 80 32 72 36 69 28 109 OM 44 57 35 59 57 40 31 53 50 59 45 48 36 60 41 54 46 73 P 114 81 79 93 29 77 73 83 60 84 87 64 55 70 78 96 161 292 CEC 75 93 96 101 80 89 81 51 69 58 75 74 59 85 78 118 49 87 Ca 108 144 127 164 92 120 100 177 97 115 88 101 73 193 182 230 70 172 Mg 64 103 40 49 69 86 44 23 53 48 54 60 47 57 71 135 35 41 K 95 119 89 137 114 133 76 71 41 68 78 79 104 138 96 113 107 122 Na 117 273 100 269 127 195 130 233 133 72 101 214 135 243 141 322 73 89

Table 2. The coefficient of variation (%) was calculated for four mapping units of the soil map of Changprai et al (1971) and for five categorie five for and (1971) al et Changprai of map soil the of units mapping four for calculated was (%) variation coefficientof The 2. Table updated farmers’ field classification (UFFC).

backslopes of the Lam Dom Yai River (n = 9); LTP = low topographic position (n = 69); MTP = medium topographic position (n = 107); HTP = high topographic position (n = 65); = (n position topographic high = HTP 107); = (n position topographic medium = MTP 69); = (n position topographic low = LTP 9); = (n River Yai Dom Lam the of backslopes a DUSH = upper sandy horizon deeper than 1.5 m (n = 23).

84 Oberthür and Kam 0

3.80

4.86 0.13 0.30 2.65 0.72 0.06 0.07

0.05

12.10 86.20

3.90

4.59 0.20 0.57 4.00 1.04 0.07 0.07 0.02 0.04

11.00 86.10

ABS = alluvial sites on the

7.63

5.08 0.16 0.28 1.84 1.28 0.16 0.10 0.02 0.03 b

14.29 76.70

6.66

4.72 0.20 0.66 4.00 1.52 0.16 0.09 0.03 0.04

15.37 77.49

9.85

5.09 0.16 0.30 1.53 2.08 0.20 0.10 0.03 0.07

16.50 71.30

MTP HTP DUSH

6.80

4.74 0.20 0.59 4.00 1.44 0.16 0.08 0.03 0.04

17.93 75.01

Categories of the UFFC

5.24 0.17 0.30 1.00 3.04 0.51 0.13 0.04 0.12

15.60 24.10 56.73

LTP

4.84 0.25 0.69 4.00 2.32 0.34 0.12 0.04 0.07

10.70 24.10 63.90

5.06 0.16 0.53 2.04 5.04 0.52 0.21 0.08 0.06

14.80 18.30 67.40

ABS

4.88 0.21 1.17 6.86 4.16 0.55 0.22 0.09 0.05

13.60 21.90 67.60

5.22 0.17 0.34 1.00 2.96 0.32 0.10 0.04 0.09

12.50 20.07 65.52

RE/PH

7.47

4.70 0.27 0.60 4.00 1.76 0.16 0.08 0.03 0.06

21.93 70.74

5.05 0.17 0.30 1.00 1.84 0.23 0.09 0.03 0.06

14.00 19.20 65.80

8.80

4.71 0.21 0.54 4.00 1.76 0.18 0.09 0.03 0.04

18.90 68.10

9.20

5.13 0.16 0.26 1.00 1.76 0.21 0.09 0.02 0.06

16.29 71.20

KO/PP RE

6.20

4.79 0.20 0.60 4.00 1.36 0.16 0.08 0.03 0.05

18.20 74.90

Mapping units of 1971 soil taxonomic map

5.04 0.16 0.33 1.53 1.92 0.22 0.13 0.03 0.05

10.17 17.93 71.09

8.00

4.76 0.22 0.72 4.00 1.88 0.24 0.11 0.03 0.05

18.09 73.55

KO = Korat (n = 98); KO/PP = Korat/Phon Phisay (n = 122); RE = Roi Et (n = 12); RE/PH = = topsoil, Roi Sub Et/Phen = (n subsoil. = 15). Top

Soil layer Top Sub Top Sub Top Sub Top Sub Top Sub Top Sub Top Sub Top Sub Top Sub

Clay Silt Sand pH EC OM P CEC Ca Mg K Na

Table 3. Class medians were calculated for four mapping units of the soil map of Changprai et al (1971) and for five categories of the updated farmers’ updated the of categories five for and (1971) al et Changprai of map soil the of units mapping four for calculated were medians Class 3. Table field classification (UFFC).

Perception, understanding, and mapping of soil variability . . . 85 Table 4. The complement of the relative variance (%) calculated for four mapping unitsa of the soil map of Changprai et al (1971), for three categoriesb of the farmers’ field classification (FCC) and for five categoriesc of the updated farmers’ field classification (UFFC).

Soil mapa FCCb UFFCc

Topsoil Subsoil Topsoil Subsoil Topsoil Subsoil

Clay 3 1 10 17 21 24 Silt 0 0 16 15 25 17 Sand 0 1 18 24 30 29 pH02020 4 EC40818 0 OM40002515 P 01032 3 CEC 0 1 17 15 29 20 Ca 0 1 23 11 34 13 Mg85044320 K30323011 Na21427 2

aKorat, Korat/Phon Phisay, Roi Et, Roi Et/Phen. bLow, medium, high position. cLow, medium, high position, deep sands, alluvial backslope sites.

Mapping soil variability Managing soil variability requires knowledge about the spatial distribution of soil properties. Geostatistical interpolation approaches permit the user to first elucidate the spatial structure of soil properties using variography (i.e., examining plots of the semivariance, which is a measure of the dissimilarity of pairs of data points, against distance between them), and second to use this information in interpolation algorithms to generate thematic soil maps (Oliver and Webster 1991). The indicator approach (Goovaerts 1997) was chosen because it accounts for class-specific patterns of spatial continuity through the different indicator variogram models at each threshold, allows for the incorporation of additional (secondary soft) information, and does not depend on normality of the data. Qualitative and quantitative data can be processed using the indicator approach. In the case of quantitative data, one is able to determine the conditional variance σ2 of the data around their mean. Unlike the variance obtained for nonindicator approaches, the conditional variance depends on the actual data and not on the covariance model. Nonindicator-based kriging variances would be the same for a given covariance model and data configuration even when different data were used to derive estimates of a variable Z. Soft indicator kriging (SIK) and indicator simulations (IS) were used for the interpolation. The flow of operations for this study is outlined in Figure 3. Kriging algorithms are best locally, whereas simulation models the joint uncertainty about attribute values at several locations. Connected strings of large and small values are

86 Oberthür and Kam Hard point data Hard point data Soft point data Area information Clay, silt, sand Soil texture classes UFFC classes Aerial photography

Geographic information systems

Soft grid data UFFC classes

Sequential IS SIK Maps of clay, silt, sand Maps of texture classes

Geostatistical indicator interpolation

Fig. 3. The outline of operations that are needed to generate maps from point data using sequential indicator simulation (IS) and soft indicator kriging (SIK).

better reproduced by simulation than kriging, and this is important for an assessment of soil property distribution in very heterogeneous lands. Suppose that {z*(u), uεA)} is a set of kriging estimates of the soil property Z for locations u in a region A. The local error variance Var{Z*(u) – Z(u)} is minimum and the best estimate in the “least-square sense” if each estimate z*(u’) is considered independently of its neighboring estimates. These best local estimates may not be best as a map and, instead of generating such a kriged map, simulation produces one or a set of possible realizations of the spatial distribution of the z* values. For example, a map of z* values {z*(l)(u), uεA}, where l denotes the lth realization, can be generated that reproduces the statistics of an area that are most important for the problem at hand. Simulated maps honor data values at their locations and reproduce closely the declustered sample histogram and the set of covariance models for various indicator thresholds (Goovaerts 1997). In this study, indicator variograms were modeled for the nine deciles of quantitative soil properties and for the classes of qualitative properties. Estimates of mean (Figs. 4A–6A), median (Figs. 4B–6B), conditional variance (Figs. 4E–6E), and conditional CV (Figs. 4F–6F) and the 0.2- and 0.8-quantiles (Figs. 4C,D–6C,D) of the conditional cumulative distribution function were mapped. Quantile maps are often more useful than maps showing mean values; for instance, overestimation of soil nutrients in regions with low soil fertility is more harmful (yield loss or crop failure) than underestimation (undue application of treatments). Conversely, underestimation of nutrient contents has negative impacts in very fertile regions. In this situation, excess fertilizer nutrients are discharged in the groundwater and have detrimental effects on ecosystems. Using quantile maps, one can over- or underestimate the true values and minimize risks of environmental pollution or crop failure (see Figs. 4–6).

Perception, understanding, and mapping of soil variability . . . 87 A. Clay (%) B. 0.5 ccdf

C. 0.2 ccdf C. 0.8 ccdf

E. Con. Var. F. Con. CV

Fig. 4. Applying estimators to probabilities of occurrence (obtained using sequential indicator simulation) for created maps of clay (topsoil). Map A shows the mean value and was generated using the e-type estimator. Quantile estimators were applied to produce maps B, C, and D that show the values at the 0.5-, 0.2-, and 0.8-quantile of the conditional cumulative distribution function (ccdf), respectively. Estimates can be interpreted as the value that has 50% (B), 80% (C), or 20% (D) probability of being exceeded by the true value in the field. Maps E and F depict the spread about the mean at each location as the conditional variance (map E; Con. Var.) and in relative terms as the conditional coefficient of variation (map F; Con. CV).

88 Oberthür and Kam A. Silt (%) B. 0.5 ccdf

C. 0.2 ccdf D. 0.8 ccdf

E. Con. Var. F. Con. CV

Fig. 5. Applying estimators to probabilities of occurrence (obtained using sequential indicator simulation) for created maps of silt (topsoil). Map A shows the mean value and was generated using the e-type estimator. Quantile estimators were applied to produce maps B, C, and D that show the values at the 0.5-, 0.2-, and 0.8-quantile of the conditional cumulative distribution function (ccdf), respectively. Estimates can be interpreted as the value that has 50% (B), 80% (C), or 20% (D) probability of being exceeded by the true value in the field. Maps E and F depict the spread about the mean at each location as the conditional variance (map E; Con. Var.) and in relative terms as the conditional coefficient of variation (map F; Con. CV).

Perception, understanding, and mapping of soil variability . . . 89 A. Sand (%) B. 0.5 ccdf

B. 0.2 ccdf D. 0.8 ccdf

E. Con. Var. F. Con. CV

Fig. 6. Applying estimators to probabilities of occurrence (obtained using sequential indicator simulation) for created maps of sand (topsoil). Map A shows the mean value and was generated using the e-type estimator. Quantile estimators were applied to produce maps B, C, and D that show the values at the 0.5-, 0.2-, and 0.8-quantile of the conditional cumulative distribution function (ccdf), respectively. Estimates can be interpreted as the value that has 50% (B), 80% (C), or 20% (D) probability of being exceeded by the true value in the field. Maps E and F depict the spread about the mean at each location as the conditional variance (map E; Con. Var.) and in relative terms as the conditional coefficient of variation (map F; Con. CV).

90 Oberthür and Kam In summary, the chosen geostatistical indicator algorithms have various comparative advantages over other mapping approaches, including the ability to handle quantitative and/or qualitative variables, to account for uncertainty about the estimated data and to offer various types of estimates (e.g., mean, median, quantiles, probability of exceeding a given threshold). Maps were generated for all analyzed soil properties and can be obtained from the IRRI-GIS laboratory. In this chapter, we discuss only the geostatistical analysis of topsoil texture classes and the mapping of topsoil particle size distribution. Soil particle size distribution was mapped for the URLRA using sequential indicator simulation with hard data from laboratory analysis of all soil survey samples, without soft information such as the UFFC classification. The resulting maps (Figs. 4–6) show that average clay, silt, and sand contents have a similar spatial distribution. Clay is low in most parts of the region and moderately high only along the Lam Dom Yai River in the southwestern part of the study area. Silt follows similar patterns but, unlike clay, has another inclusion with higher values in the northeastern part of the study area. High sand contents form a wide belt that stretches diagonally from the northwestern to the southeastern part of the study area. The map outputs (Figs. 4–6) differ from those generated using soft information, such as Figure 2C. The soil texture class map shown in Figure 2C, based on field texturing, was generated for the rectangular subarea of the URLRA using a modified indicator kriging approach (soft indicator kriging or SIK) that incorporates descriptive soft information into the mapping procedure (Oberthür et al 1999). A probability vector of occurrence of each soil texture class at unsampled locations was estimated from hard information (field estimations of soil texture classes) and soft information about field location (formalized knowledge of farmers and experts in the UFFC). Our data sets would not have permitted meaningful kriging of soil texture classes if we hadn’t reduced the number of possible classes by merging some of those previously recognized. Original topsoil texture classes as estimated in the field were aggregated into three major classes to obtain sufficient samples for geostatistical modeling in each class. Relative to the average soil texture in the study area, these classes represent soils with light (class one), medium (class two), and heavy (class three) texture. Consequently, class one represents sandy and loamy sand soils and class two sandy loam. All samples classified as loam and heavier are assigned to class three. The SIK map (Fig. 2C) reveals a dendritic pattern (similar to tree branches) and much spatial detail of soil texture class distribution. Texture classes change gradually from heavy to medium to light and follow the main topographic and drainage patterns. In the undulating rainfed lands of northeast Thailand, knowledge of farmers and soil experts implicitly reflected the spatial variation of soil texture classes and this soft information was incorporated in the geostatistical approach to improve mapping accuracy. The dendritic pattern reflects partly the local topography and drainage system. However, considering only topography and drainage pattern as soft information would not be sufficient because correspondence between these features and soil texture class distribution is only partial: the size of micro-watersheds in the area is smaller than the extent of the contiguous areas of soil texture classes.

Perception, understanding, and mapping of soil variability . . . 91 Northeast Thailand field texture—Class 1 Class 2 Class 3 Semivariance 0.25 0.25 0.07 0.06 0.20 0.20 0.05 0.15 0.15 0.04 0.03 0.00 0.10 0.00 0.00 012345678 012345678 012345678 Distance (km) Fig. 7. The indicator variograms for three finger-estimated soil texture classes (light, medium, heavy).

Table 5. Effective rangea (given in m) and the relative nugget effectb of the exponential models and pure nugget variation that were fitted to the indicator variogramsc: depth of clay, silt, and sand of topsoil and subsoil.

T 0.1 T 0.2 T 0.3 T 0.4 T 0.5 T 0.6 T 0.7 T 0.8 T 0.9

Clay Effective range 613 613 759 846 847 1,372 2,015 2,803 1,518 Relative nugget 0.25 0.49 0.39 0.55 0.55 0.34 0.40 0.44 0.38 effect

Silt Effective range 2,131 2,453 2,803 993 788 934 1,693 1,168 – Relative nugget 0.20 0.27 0.37 0.31 0.28 0.31 0.31 0.25 1 effect

Sand Effective range – 993 701 1,226 701 1,401 4,088 3,737 4,380 Relative nugget 1 0.40 0.26 0.26 0.27 0.49 0.47 0.24 0.20 effect

aFor practical purposes, the effective range, i.e., three times the distance parameter of the exponential model, is given to indicate the limit of spatial dependence. bThe relative nugget effect was calculated as the ratio of nugget variance (discontinuity at the beginning of the variogram caused by sampling error and short-scale variability not resolved by the sampling scheme) to sill (value where the variogram reaches a plateau). cIndicator variograms were calculated with values for thresholds (T) corresponding to the nine deciles (0.1–0.9) of the conditional cumulative distribution function.

The SIK map confirms the short-range variation in soil texture that is indicated by variograms of the texture classes (Fig. 7). The variograms display mainly short- range structures, indicating a change of texture classes over very short distances. Variography results listed in Table 5 reveal that low clay values change rapidly in the topsoil over short distances. No spatial dependence was detected beyond 1 km for values corresponding to indicator thresholds < 0.5 of the conditional cumulative distribution function (corresponding to 7.2% clay content). Large contents of topsoil clay are spatially dependent over longer distances of up to 2.8 km. Low silt values are

92 Oberthür and Kam related over slightly longer distances than high silt contents. High sand values are spatially dependent over long distances of up to 4.4 km, whereas low sand contents change rapidly over short distances (Table 5). Our results are thus not consistent with the general belief that soils of the Korat Plateau are universally coarse-textured and infertile. The SIK map assigns much land (25%) to soil texture class three (loam and heavier). A loamy soil is expected to be suitable for rice production. The reliance on soil maps may have contributed to the long-held views about soil texture in northeast Thailand. In reality, soil texture in this region varies over short distances and the soil maps do not depict these changes. Colluvio-alluvial processes over short distances and in situ soil development are as important factors for the distribution of soils in northeast Thailand as is the gradually changing Quaternary alluvial sedimentation.

Conclusions 1. Soil maps at scales of 1:100,000 to 1:125,000 highlight soil trends in the region but don’t depict much of the agronomically important spatial variability and cannot be used for land resource assessment. Agronomic interpretations should be based on maps generated by detailed soil surveys or on combining field data with existing information to produce refined thematic maps with relevant spatial resolution. 2. The UFFC that is based on local knowledge captures much of the variation in soil properties and does provide a basis for land resource assessment. Its applicability, however, excludes some soil properties (see RVc values) and little variation of pH, EC, Bray-II P, K, and Na is accounted for. Variation of more stable soil properties characterizing the inherent soil potential (texture, organic carbon, CEC) and Ca and Mg of the base complex can be assessed sufficiently. 3. Although the average values of soil fertility indicators are low, their range and spatial distribution, in conjunction with soil development processes, confirm information by farmers that soil fertility levels are sufficient for rainfed rice production in a large proportion of the land. 4. The soft information used here is readily available in many regions or does not cost much to obtain. Therefore, instead of concentrating on additional sampling with greater density, SIK provides an interesting cheap alternative for updating or upgrading soil maps and describing spatial soil variability. 5. Integration of hard and soft information revealed visually that colluvio-alluvial processes act over short distances and in situ soil development is likely. Large-scale variation, as suggested by the variograms and maps, is not consistent with the widely held belief that Quaternary alluvial sedimentation was the sole process responsible for the distribution of soils in the region. 6. Geostatistical indicator approaches offer a suite of tools for characterizing and mapping soil variability in the rainfed rice lands, including variography

Perception, understanding, and mapping of soil variability . . . 93 and the possibility to generate quantile maps or maps that exceed specified thresholds. 7. The results presented indicate the need for software that implements the SIK and other relevant algorithms in a standardized and user-friendly mode.

References Beckett PHT, Webster R. 1971. Soil variability: a review. Soils Fert. 34:1-15. Black CA. 1965. Methods of soil analysis. Madison, Wis. (USA): American Society of Agronomy. Bray RH, Kurtz LT. 1945. Determination of total, organic, and available forms of phosphorus in soils. Soil Sci. 59:39-45. Brus DJ, De Gruijter JJ, Breeuwsma A. 1992. Strategies for updating soil survey information: a case study to estimate phosphate sorption characteristics. J. Soil Sci. 43:567-581. Changprai C, Chotimon A, Thunduan V, Thipsuwan C, Lepananontha J, Kittiyarak S. 1971. Detailed reconnaissance soil map of Ubon Ratchathani Province. Bangkok (Thailand): Department of Land Development. Craig IA. 1988. Agronomic, economic and socially sustainable strategies for soil fertility management in northeast Thailand. In: Chareonwatana T, Rambo AT, editors. Sustainable rural development in Asia. Fourth SUAN Regional Symposium on Agroecosystem Re- search. Khon Kaen (Thailand): Farming Systems Research Project and Southeast Asian Universities Agroecosystem Network. p 153-165. Davis JG, Hossner LR, Wilding LP, Manu A. 1995. Variability of soil chemical properties in two sandy, dunal soils of Niger. Soil Sci. 159:321-330. Dobermann A, White P. 1999. Strategies for nutrient management in irrigated and rainfed lowland rice systems. Nutr. Cycl. Agroecosyst. 53:1-18. Goovaerts P. 1997. Geostatistics for natural resources evaluation. New York (USA): Oxford University Press. Grandstaff TB. 1988. Environment and economic diversity in Northeast Thailand. In: Chareonwatana T, Rambo AT, editors. Sustainable rural development in Asia. Fourth SUAN Regional Symposium on Agroecosystem Research. Khon Kaen (Thailand): Farm- ing Systems Research Project and Southeast Asian Universities Agroecosystem Net- work. p 11-22. Kubiniok J. 1990. Relief- und Bodengeneration auf dem Khorat-Plateau (NE-Thailand). Zeitschrift für Geomorphol. 34:149-164. Löffler E. 1984. Quarternary geomorphological development of the Lower Mun River Basin, North East Thailand. Catena 11:321-330. Mackill DJ, Coffman WR, Garrity DP. 1996. Rainfed lowland rice improvement. Manila (Phil- ippines): International Rice Research Institute. 242 p. Michael P. 1982. The landforms of Thailand: ideas about their genesis and influence on soil property distribution. In: Nutalaya P, Karasudhi P, Tanasuthipitak T, Kheoruenromme I, Sudhiprakarn A, editors. First International Symposium on Soil, Geology and Land- forms: Impact on Land Use Planning in Developing Countries. Bangkok (Thailand): Association of Geoscientists for International Development. p C10.1-C10.14. Mitsuchi M, Wichaidit P, Jeungnijnirund S. 1989. Soils of the Northeast Plateau, Thailand. Tsukuba (Japan): Tropical Agriculture Research Center.

94 Oberthür and Kam Miura K. 1990. Genetic features of the major soils in northeast Thailand. Khon Kaen (Thai- land): Agricultural Development Research Project in Northeast Thailand. Moormann FR, Montrakun S, Panichapong S. 1964. Soils of northeastern Thailand: a key to their identification and survey. Bangkok (Thailand): Land Development Department. Oberthür T, Dobermann A, Goovaerts P. 1999. Mapping soil texture classes using field texturing, particle size distribution, and local knowledge by both conventional and geostatistical methods. Eur. J. Soil Sci. 50:459-479. Oberthür T, Dobermann A, Neue HU. 1996. How good is a reconnaissance soil map for agronomic purposes? Soil Use Manage. 12:33-43. Oliver MA, Webster R. 1991. How geostatistics can help you. Soil Use Manage. 7:206-218. Page AL. 1982. Methods of soil analysis. Madison, Wis. (USA): American Society of Agronomy. Paiboon P, Liengsakul M, Engkagul V. 1985. Grain size analysis of some sand rises and stream sediments in the northeast of Thailand in order to indicate depositional environment. In: Proceedings of the Conference on Geology and Mineral Resource Development in North- east Thailand. Khon Kaen University, Khon Kaen, Thailand. p 235-253. Patcharapreecha P. 1988. Physico-chemical properties of the Northeast paddy soils in relation to fertility. In: Panichapong S, editor. Proceedings of the First International Symposium on Paddy Soil Fertility. Paddy soil fertility working group, Chiang Mai, Thailand. p 405-414. Pendleton RL, Montrakun S. 1960. The soils of Thailand. In: Proceedings of the 9th Pacific Science Congress. Bangkok (Thailand): Department of Rice, Ministry of Agriculture. Ragland J, Boonpuckdee L. 1988. Soil fertility management in northeast Thailand. Khon Kaen (Thailand): Northeast Rainfed Agricultural Development Project (NERAD). Suddhiprakarn A, Kheoruenromne I. 1987. A study on some alfisols and ultisols in ustic soil moisture regime, northeast Thailand. Kasetsart J. (Nat. Sci.) 21:214-229. Tamura T. 1986. Geomorphological development in northeast Thailand with reference to problem soil formation. Agricultural Development Research Project in Northeast Thailand. Khon Kaen (Thailand): JICA. Webster R, Oliver MA. 1990. Statistical methods in soil and land resource survey. Oxford (UK): Oxford University Press. Zeigler RS, Puckridge DW. 1995. Improving sustainable productivity in rice-based rainfed lowland systems of South and Southeast Asia. GeoJournal 35:307-324.

Notes Authors’ addresses: Thomas Oberthür, Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia; Suan Pheng Kam, International Rice Research Institute, Los Baños, Philippines. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Perception, understanding, and mapping of soil variability . . . 95 Identifying soil suitability for subsoil compaction to improve water- and nutrient-use efficiency in rainfed lowland rice

D. Harnpichitvitaya, G. Trébuil, T. Oberthür, G. Pantuwan, I. Craig, T.P. Tuong, L.J. Wade, and D. Suriya-Arunroj

For coarse-textured soils of high subsoil permeability, research has demonstrated the benefits of subsoil compaction for improved water- and nutrient- use efficiency in rainfed lowland rice (Oryza sativa L.). To better define soil conditions suited to this subsoil compaction, on-farm experiments were carried out on soils varying in subsoil clay content. Rice (cv. KDML105 in 1993 and cv. RD6 in 1994) was grown in main plots comparing shallow dry tillage without compaction, shallow dry tillage with subsoil compaction, and deep dry tillage with subsoil compaction. Soil was compacted in seven farmers’ fields in the south of Ubon Ratchathani Province, with 10 passes of a vibrating road roller on 19 and 23 May 1993. The effects of subsoil compaction on changes in soil physical and hydrological properties differed according to subsoil clay content, which ranged from 1.4% to 12.0%. Subsoil compaction decreased soil hydraulic conductivity sufficiently for fields with subsoil clay content greater than 2%. When subsoil clay content was higher than 10%, the justification for using relatively costly subsoil compaction was questionable, as the hydraulic conductivity was already low and the gains in water- storing capacity seemed limited. Based on these results, proportions of soils with subsoils of <2%, 2–5%, 5–10%, or >10% clay were mapped using geographic information systems for an area of about 40,000 ha within the Ubon Ratchathani Land Reform Area. About 40% of the mapped area had subsoils with clay % between 2 and 10, suggesting that substantial areas could be suitable for subsoil compaction. More investigations are needed to assess its economic and social acceptability, to better understand variability and probability of response, and to further refine soil suitability in relation to clay type, clay content, and groundwater changes at the toposequence level.

Identifying soil suitability for subsoil compaction . . . 97 Rainfed lowland rice, growing with an average yield of only 1.5 t ha–1 on some 4.5 million ha of coarse-textured and low-fertility soils in northeastern Thailand, may suffer from water stress as soon as rains are interrupted for about 1 wk (Sharma 1992). Percolation rate can be as high as a few centimeters d–1. Puddling is not fully effective due to high sand content. Efficient management of rainwater and nutrients is a key factor in improving the productivity and stability of rice. On-station experiments, carried out at the Ubon Ratchathani Rice Research Center on a soil in the Ubon series with a clay content of 12% in the subsoil (30–60 cm), showed that water- and nutrient-use efficiency may be improved by reducing percolation through the creation of a subsurface barrier (Garrity et al 1992). Subsequent field-oriented research to identify practical and cost-effective techniques for decreasing subsoil permeability showed that multiple passes of a 12-t conventional road roller with vibration action (Dynapac CA25) that compacts the subsoil to 75-cm depth was most effective (Sharma et al 1995a,b, Trébuil et al 1998). As subsoil compaction was observed to concentrate rice roots in the plow layer above the compacted zone, a deeper (15–20 cm) tillage depth than the one currently achieved in farmers’ fields (7–10 cm) was found to be desirable. All previous studies were carried out at the experiment station of the Ubon Rice Research Center. The effects of subsoil compaction across a range of subsoil clay contents have not been investigated, and the combined effects of soil compaction and green manuring have not been investigated either. The objectives of the on-farm study were 1. To quantify the effects of subsoil compaction and tillage depth on changes in sandy soil physical and hydrological properties across a range of subsoil clay contents. 2. To quantify the interactions between subsoil compaction and green manure fertilization and their effects on the productivity of rainfed lowland rice. 3. To define soil suitability for subsoil compaction by using geographic information systems (GIS) techniques applied to the Ubon Ratchathani Land Re- form Area.

Materials and methods Experimental site Two-factor on-farm field experiments were conducted in the 1993 and 1994 wet seasons in Ban Klang and Ban Mak Mai villages of Klang subdistrict of Det Udom District, Ubon Ratchathani Province, lower northeast Thailand. The experimental sandy soils (Table 1) belonged to the Nam Phong series and were acidic, low in organic matter content and available N, P, and K, and with a high concentration of ferrous iron in some fields.

Experimental treatments We selected seven farmers’ fields for this study. The elementary experimental plot size varied from 100 to 200 m2. Each plot was surrounded by a single levee, approximately 20 cm high and 40 cm wide.

98 Harnpichitvitaya et al Table 1. Properties of the experimental soils belonging to the Nam Phong series in Det Udom District, Ubon Ratchathani Province, 1993 wet season.

Soil parameter Range Mean

Particle size distribution: 0–30 cm (%) Sand 88.2–94.6 90.9 Silt 3.6– 9.4 6.5 Clay 1.7– 3.6 2.6 Particle size distribution: 30–60/70 cm (%) Sand 88.0–95.4 91.2 Silt 3.1–10.3 6.7 Clay 1.4– 4.5 2.1 Chemical parameters: 0–30 cm pH (1:1) 3.9– 5.2 4.2 Organic matter content (%) 0.39–1.79 0.85 Total N (%) 0.02–0.08 0.04 Extractable P (Bray II, ppm) 6.1–19.0 9.8 Extractable K (ammonium acetate, pH 7, ppm) 5.0–12.6 8.6 CEC (meq 100 g–1) 0.32–1.28 0.83

In the 1993 wet season (WS), the following three subsoil compaction and tillage treatments were imposed in each of the selected fields (but without replications in each field): ● C0 = uncompacted dry tillage; soil was disc-plowed dry to 7–10-cm depth. ● C1 = subsoil compacted by 10 passes of a 12-t Dynapac vibrating road roller on 19 or 23 May 1993 (with soil moisture brought close to field capacity by using a watering truck), followed by dry tillage to 7–10-cm depth. ● C2 = subsoil compacted as above but with dry tillage to 15–20-cm depth. In the 1994 WS, we selected the C0 and C1 plots of six farmers’ fields, all having been included in the previous year’s on-farm experiment. C2 plots were not included because the tillage treatment did not show any effects on the monitored parameters in 1993. Each plot was further divided into two to accommodate a split- plot design, with the main and subplot treatments as follows: ● Main plots: C0 = uncompacted with shallow (7–10 cm) dry tillage. C1 = compacted with shallow dry tillage (plots were not compacted again in 1994). ● Subplots: F0 = no green manure before rice (control treatment). F1 = with green manure before rice.

Cultural practices The experimental cultivars were the similar and widely grown photoperiod-sensitive, medium-tall KDML105 and RD6 during the 1993 and 1994 wet seasons, respec-

Identifying soil suitability for subsoil compaction . . . 99 tively. In 1993, following dry tillage and harrowing, rice was direct-seeded at a rate of 62.5 kg ha–1 on 24-25 May 1993. Rice seeds were covered by a second harrowing and, as in the surrounding farmers’ fields, no weeding was implemented because only limited weed infestations occurred. Later, two fertilizer applications were carried out at sowing (25-11-21 kg NPK ha–1) and panicle initiation (19 kg N ha–1). In 1994, because green manure before rice (a combination of Sesbania rostrata, Aeschynomene afraspera, cowpea, and sword bean seeds was sown in mid-May 1994) was grown in subplots, rice was transplanted at 20 × 20-cm intervals on 29 July 1994 for three fields and 8 August 1994 for another three fields. A basal application of 11 kg P ha–1 was made in F1 subplots before the sowing of green manure seeds. A similar P basal application was applied to subplots without green manure before transplanting of rice. All subplots also received 25 kg N and 21 kg K ha–1 at 32 days after transplanting (DAT) and 25 kg N ha–1 again at 48 DAT.

Field-level soil and water observations A piezometer was installed to 150-cm depth in each of the experimental fields for weekly monitoring of fluctuations in the groundwater table. We did not monitor the groundwater table depth in each individual plot because the groundwater table depth was governed by the regional hydrology rather than the soil management at the plot level. A shorter piezometer (40-cm PVC pipes with perforations in the bottom 10 cm) was installed to 30-cm depth to monitor the perched water table in each plot during the 1994 wet season only. Floodwater depth was regularly recorded using sloping gauges permanently installed in each subplot, for seven fields in the 1993 WS and six fields in the 1994 WS. Volumetric soil moisture content in the 0–15-cm layer was measured twice a week from the last rain until rice harvest. The Det Udom District meteorological station and two rain gauges installed at the experimental sites in each of the two villages of Klang subdistrict provided daily rainfall data. Soil penetration resistance was measured just before flowering on 12–13 Octo- ber 1993 using a recording-type cone penetrometer (Eijkelkamp, cone base area = 1 cm2) for each 5-cm-thick soil layer from 0 to 75-cm depth, for seven fields in the 1993 WS only. Saturated hydraulic conductivity was also determined close to maturity in mid-November 1993, using the constant head method for seven fields and the on-station experiment at the Ubon Rice Research Center. Fresh and dry weights of the green manure biomass incorporated at land preparation for transplanted rice were measured 1 wk before rice transplanting in 1-m2 sampling squares replicated three times in each subplot, for five fields in the 1994 WS only. In both years, grain yield of rice from harvest areas of 6 m2 was recorded and adjusted to 14% grain moisture content. Yield data were obtained from seven fields in 1993 and six fields in 1994.

GIS-based mapping of subsoil texture Soil texture data were collected in a 2,000 × 2,000-m grid and at sites in transects along the slopes of micro-watersheds to reveal soil heterogeneity over different spatial scales in an area of 40,000 ha within the Ubon Ratchatani Land Reform Area. The

100 Harnpichitvitaya et al locations for sampling were drawn a priori on a map, and the coordinates of the 278 sampling locations were identified in the field using a global positioning system and the nearest rice field sampled. Soil samples were collected from 0 to 15-cm (soil monolith 0.2 m × 0.2 m × 0.15 m by spade) and 15 to 40-cm (dutch auger) depths. Five samples were bulked, one from the center of the field and four samples within a 6-m radius around the center of the field. The samples were air-dried and ground to pass through a 2-mm sieve. Soil particle size distribution was determined using the hydrometer method. From the collected point-data, maps of subsoil textures were generated for each of the 250 × 250-m cells of the 40,000-ha area using indicator geostatistics with and without prior means (Goovaerts 1997), as described in Oberthür et al (1999). The GIS-based map would allow us to classify and map out cells according to their subsoil clay contents.

Results and discussion Rainfall and water table depth Figure 1 displays the extent of rainfall variability during the wet season across years for the Det Udom area. Seasonal rainfall distributions and groundwater table fluctuations during the 1993 and 1994 growing seasons contrasted markedly (Fig. 2). Total rainfall received by the rainfed lowland rice crops was 1,035 mm for direct-seeded rice in 1993 and 751 mm for the transplanted rice crop during the 1994 wet season. In the 1994 WS, 650 mm of rainfall were received from May to late July, when the green manure plants were growing before the rice crop. In 1993, significant rainshowers

Rainfall (mm) 1,600

1,400

1,200

1,000

800

600 AF80 AF50 400 AF20 1993 200 1994

0 Jun Jul Aug Sep Oct Nov Month

Fig. 1. Cumulative rainfall in 1993 and 1994 and at probability of exceedence of 20% (AF20), 50% (AF50), and 80% (AF80) at Det Udom, Ubon Ratchathani, lower northeast Thailand. The frequency analysis was based on 1963-97 rainfall record.

Identifying soil suitability for subsoil compaction . . . 101 Rainfall (mm) Water table depth (cm)

A C = soil compaction B = broadcasting F1–3 = 1st, 2nd, 3rd fertilizer application H = heading HAR = harvest 150 F1 F2 F3 H HAR 100 CB

0 0

–50

–100

–150 –20 0 20 40 60 80 100 120 140 160 180 200 Days after sowing

Rainfall (mm) Water table depth (cm) 150 B TP H HAR 100 TP = transplanting F1 F2 F1–2 = 1st, 2nd fertilizer application 50 H = heading HAR = harvest 0 0

–50

–100

–150 –50 0 50 100 150 200 Days after transplanting

Fig. 2. Rainfall distribution, groundwater table fluctuations, and rainfed lowland rice cropping calendar during the 1993 (A) and 1994 (B) wet seasons in Det Udom District, Ubon Ratchathani Province, lower northeast Thailand.

102 Harnpichitvitaya et al ceased at 116 days after sowing (DAS), apart from a small shower about 1 wk before heading (Fig. 2A). In 1994, rain also stopped about 1wk before heading at 70 DAT (for plots transplanted on 8 August), or 80 DAT (for plots transplanted on 29 July). During the 1994 WS, no dry spells longer than 5 d occurred between transplanting and the last rainshower before heading (Fig. 2B). In 1993, the water table depth stayed at less than 50 cm below the soil surface in most of the fields from 85 to 130 DAS (some 10 d after panicle initiation). Such high groundwater table levels mitigated the effects on the rice crop of the dry spells occurring during that period. During the 1994 WS, the mean groundwater table depth rose to less than 20 cm from the soil surface from 15 d before transplanting and remained at this level until 55 DAT (Fig. 2). The groundwater table remained within the top 40- cm depth up to the heading stage, but fell below 50 cm after flowering. This means that, in the 1994 WS, rice encountered water stress during grain filling.

Effects of subsoil compaction on soil penetration resistance and saturated conductivity In the seven fields for which data were collected, subsoil compaction significantly increased soil penetration resistance to at least 75-cm depth, with the 15–65-cm layer being the most compacted (Fig. 3). Subsoil compaction effectively decreased saturated hydraulic conductivity (by a factor of 2–3), but most of the postcompaction values remained higher than those commonly observed (10–20 cm d–1) in paddy fields (Fig. 4). The degree of subsoil compaction obtained on these very coarse-textured soils, without removing the topsoil, varied extensively, particularly according to the percentage of clay in their subsoils. The observed effects on soil hydraulic conductiv-

Soil depth (cm) Soil penetration resistance (kPa) 0 100 200 300 400 500 0

80 Uncompacted + shallow tillage Compacted + shallow tillage 100 Compacted + deep tillage

120 Fig. 3. Soil penetration resistance profiles (mean ± SE values of seven farmers’ fields) in different subsoil compaction and tillage treatments, 1993 wet season, in Det Udom District, Ubon Ratchathani Province, lower northeast Thailand.

Identifying soil suitability for subsoil compaction . . . 103 Saturated hydraulic conductivity (cm d–1) 600

513 500 Uncompacted 400 Compacted

300

200

100

10 0 1.4 2.1 2.3 2.6 4.4 4.6 10.9 12.0 Subsoil clay content (%) Fig. 4. Relationship between percentage of subsoil clay content and subsoil compaction effects on saturated hydraulic conductivity of seven farmers’ fields and one field within the Ubon Rice Research Center (with subsoil clay content of 12%), Ubon Ratchathani Prov- ince, lower northeast Thailand.

ity suggested that subsoil compaction was not effective in fields with less than 2% clay in the subsoil, where the postcompaction conductivity remained exceedingly high (more than 200 cm d–1). Soil hydraulic conductivity decreased significantly in fields having a subsoil clay content of more than 2% but less than 5%, but not up to an optimum of less than 10–20 cm d–1. On the other hand, where subsoil clay content was higher than 10%, the uncompacted hydraulic conductivity was already low. Where subsoil clay exceeded 10%, the technical, and probably the economic, justification for using subsoil compaction could be questionable, as the benefits of reducing percolation rate were limited. Farmers’ fields with subsoil clay contents between 2% and 10% seemed to be suitable for subsoil compaction. Since data points were few, this relationship requires further study.

Effects of subsoil compaction on field hydrological conditions Ponded water depth and duration. In the 1993 WS, no ponded water was observed in any plot before 82 DAS. Subsoil compaction did not increase the duration of floodwater, and had only a limited effect on floodwater depth during 82–120 DAS in shallow tillage plots (Fig. 5A). In all but one field, the groundwater table was within 20– 30-cm depth from 85 to 130 DAS. Shallow groundwater was a major factor masking the effect of subsoil compaction on floodwater duration in the 1993 WS. In the other field where the groundwater table was mostly below 50–100-cm depth during 80– 130 DAS, the differences in floodwater depth clearly displayed the effect of subsoil compaction on water retention above the soil surface (data not shown). Since the

104 Harnpichitvitaya et al Floodwater depth (cm) 25 A Uncompacted + shallow tillage 20 Compacted + shallow tillage Compacted + deep tillage

0 80 90 100 110 120 Days after sowing Floodwater depth (cm) 50

40 B 30 20 10 0 –10 –20 –30 –40 –60 –40 –20 0 20 40 60 80 100 Days after transplanting Fig. 5. Floodwater depth (positive values) and perched water table (negative values) as affected by subsoil compaction and tillage treatments in (A) 1993 wet season (WS) and (B) 1994 WS, Det Udom District, Ubon Ratchathani Province, lower northeast Thailand. Data were from seven farmers’ fields in 1993 and six farmers’ fields in 1994. water table depth varied with toposequence, the findings highlight the importance of the toposequence for the success of subsoil compaction in prolonging the floodwater duration. In the 1994 WS, because of the generally high groundwater table in most fields, the total number of weeks with water ponded in the rice plots was the same for the C0 and C1 treatments. The effect of subsoil compaction on floodwater duration was clearly displayed only in one field with a lower groundwater table for most of the wet season, where the percolation rate was higher (data not shown). In this field, the duration of soil submergence in the compacted C1 plots (13.0 wk) was twice as long as observed

Identifying soil suitability for subsoil compaction . . . 105 in the uncompacted C0 plots (6.7 wk). In that field, however, no floodwater was observed from 55 and 69 DAT onward for the C0 and C1 treatments, and therefore the rice suffered from severe water stress during the second half of its reproductive cycle. Thus, in both seasons, the effects of subsoil compaction were limited to a few extra centimeters in the level of the perched water table (from –42 to –15 DAT) and depth of floodwater (from –15 to 53 DAT) in C1 compared with C0 plots (Fig. 5B). This result is in contrast with earlier reports (Sharma et al 1995a,b, Trébuil et al 1998), emphasizing the importance of subsoil clay percentage, toposequence position, and groundwater table depth on the response to subsoil compaction. Soil moisture content. In the 1993 WS, volumetric soil moisture measurements made after the disappearance of standing water did not show any significant differences between treatments. In the 1994 WS, the perched water table observed in the C1 compacted plots receded almost as rapidly as in the uncompacted plots at the end of the wet season (Fig. 5B). Similar observations were made in 1993 concerning the lack of differences between treatments for volumetric soil moisture content at the end of the wet season. Such a rapid depletion of the soil profile water content, because of the very sandy soil texture and high infiltration rates, even after subsoil compaction, shows the lack of effectiveness of subsoil compaction to conserve water for the rice crop for use during the last month of its reproductive phase, although this may vary with the soil factors discussed above. These observations show that the understanding of subsurface hydrology at the field and landscape levels must be taken into account when delineating soil suitability for subsoil compaction. Further, the variation in response between on-station and on- farm studies illustrates the importance of on-farm testing.

Effects of subsoil compaction and green manure on rainfed lowland rice productivity Under the conditions of the 1993 WS, mean yields measured in the C1 and C2 compacted plots were 1.24 and 1.18 t ha–1, respectively, compared with 1.31 t ha–1 in uncompacted C0 plots. With standard errors of the yields of about 0.6 t ha–1, no significant differences in rice grain yields between either compaction or tillage depth treatments were observed. In the 1994 WS, with a mean value of 1.3 t ha–1, grain yields obtained in compacted plots were 17% higher than in the C0 uncompacted plots. However, large standard errors of the yields led to a statistically nonsignificant difference in final grain yields between the main plots (Fig. 6). Such large standard errors reflect substantial variation from field to field within each treatment, and with no replication within fields. Green manure before rice generally increased grain yields by only 8% and its effect in uncompacted plots was even more limited. On average, the combination of subsoil compaction and green manure in C1F1 plots increased grain yields by 25% in comparison with the C0F0 treatments, but this difference was also not statistically significant. In fields with at least 2% subsoil clay content, statistically nonsignificant differences in dry matter production at heading were observed, which reflected the positive effects of compaction and green manuring on rice growth (data not shown).

106 Harnpichitvitaya et al Yield (t ha–1) 3.0 Uncompacted + fertilizer 0 Uncompacted + fertilizer 1 2.5 Compacted + fertilizer 0 Compacted + fertilizer 1 2.0

1.5

1.0

0.5

0 1.4 2.1 2.3 2.6 4.4 4.6 Subsoil clay content (%) Fig. 6. Effects of subsoil compaction and green manure on rice grain yields in six farmers’ fields of Det Udom District, Ubon Ratchathani Province, 1994 wet season. Fertilizer 0 = no green manure, fertilizer 1 = with green manure.

Delineating soil suitability for subsoil compaction in the Ubon Ratchathani Land Reform Area Delineation of soil domains for which subsoil compaction could be suitable was achieved by reclassifying the map of clay content in the subsoil. Clay content was categorized using thresholds of 2%, 5%, and 10% clay in the subsoil. Figure 7 identifies the 250 × 250-m cells that have 2–5% and 5–10% of clay in the subsoil on 40,000 ha within the Ubon Ratchatani Land Reform Area. Approximately 40% of the mapped area has clay content between 2% and 10%. This land may be suitable for subsoil compaction. It should be noted that subsoil compaction in this study was based on the clay content of the 30–60-cm layer, whereas the map in Figure 7 was produced from the available data in Oberthür et al (1999), where the subsoil was defined as the 15– 40-cm layer. Figure 3 shows, however, that soil penetration resistance within 15–60- cm depth did not change as much as within 0–15-cm depth. This may be an indication that soil texture did not change very much in the 15–30-cm layer. Figure 3 also shows that the layer between 15- and 65-cm depth was most affected by the compaction. Data for the 15–30-cm layer were thus important in defining soil suitability for subsoil compaction. The mismatch between the requirement for subsoil compaction and the data available from the characterization work illustrates a common problem in identifying suitability classes. There is a need to validate these findings before pro- ceeding further.

Identifying soil suitability for subsoil compaction . . . 107 Clay classified

≤ 2% 50 ha (<1%)

>2% but ≤ 5% 762 ha (2%)

>5% but ≤ 10% 1,365 ha (37%) >10% 22,019 ha (60%)

Fig. 7. Suitability for subsoil compaction in a subregion of the Ubon Ratchathani Land Reform Area. Subsoil compaction may be effective in reducing percolation in areas with clay content between 2% and 10%.

Conclusions Variations in ponded water depth and rice yield in response to subsoil compaction were attributed to differences in subsoil texture and toposequence position, which affected the groundwater depth of the experimental fields. A minimum subsoil clay content of 2% seemed to be necessary to ensure successful subsoil compaction of the coarse-textured soils of the Nam Pong series. Even in such field situations, the infiltration rates (which were largely affected by the observed hydraulic conductivity of the subsoil) in compacted plots were still relatively high; therefore, subsoil compaction would not be expected to impede the production of annual crops such as grain legumes. These findings need further confirmation. The study illustrated a methodology using GIS and available secondary data to delineate areas with appropriate subsoil clay contents for subsoil compaction. Care- ful interpretation of these areas is necessary, however, as the chosen approach to extrapolation did not address several characteristics that might prevent the successful adoption of the technique. One important factor is the characteristic of the soil profiles. On-farm compaction trials were located in areas with deep sandy horizons, although much of the mapped area has compacted illuvial horizons in the 0.3–0.8-m soil depth. Gravelly layers of hardened iron oxide pebbles are also frequently found just below the topsoil; these might prevent successful compaction. Further on-farm

108 Harnpichitvitaya et al experiments are required to investigate the effects of these environmental phenom- ena. The study also illustrated that, apart from subsoil clay content, water table depth also determined the effectiveness of soil compaction technology. To improve the present delineation of this soil suitability, data on groundwater table changes should be integrated. Unfortunately, this type of information is rarely readily available among secondary data. Therefore, more carefully planned on-farm experiments taking into account soil texture heterogeneities (particularly the percentage of subsoil clay content and the clay type), and also groundwater table fluctuation during the wet season at the micro-catchment level, need to be carried out to delineate more precisely the soil suitability for these farming practices, as well as to better understand interfield variability and probability of response. Further investigations on these relatively costly techniques are also needed to assess their economic profitability and social acceptability in such a highly heterogeneous and variable rice ecosystem populated by a great majority of resource-poor farmers.

References Garrity D, Vejpas C, Herrera W. 1992. Percolation barriers increase and stabilize rainfed lowland rice yields on well drained soils. In: Murty VVN, Koga K, editors. Proceedings of the International Workshop on Soil and Water Engineering for Paddy Field Manage- ment, 28-30 Jan. 1992. Rangsit (Thailand): Asian Institute of Technology. p 413-421. Goovaerts P. 1997. Geostatistics for natural resources evaluation. New York (USA): Oxford University Press. Oberthür T, Goovaerts P, Dobermann A. 1999. Mapping soil texture classes using field texturing, particle size distribution and local knowledge by both conventional and geostatistical methods. Eur. J. Soil Sci. 50:459-479. Sharma PK. 1992. In situ water conservation in sandy soils for rainfed lowland rice. Rice Res. J. Bangkok (Thailand) 1:59-64. Sharma PK, Ingram KT, Harnpichitvitaya D, De Datta SK. 1995a. Management of coarse- textured soils for water conservation in rainfed lowland rice. In: Ingram KT, editor. Rainfed lowland rice: agricultural research for high-risk environments. Manila (Philip- pines): International Rice Research Institute. p 167-177. Sharma PK, Ingram KT, Harnpichitvitaya D. 1995b. Subsoil compaction to improve water use efficiency and yields of rainfed lowland rice in coarse-textured soils. Soil Till. Res. 36:33-44. Trébuil G, Harnpichitvitaya D, Tuong TP, Pantuwan G, Wade LJ, Wonprasaid S. 1998. Im- proved water conservation and nutrient-use efficiency via subsoil compaction and mineral fertilization. In: Ladha JK, Wade LJ, Dobermann A, Reichardt W, Kirk GJD, Piggin C, editors. Rainfed lowland rice: advances in nutrient management research. Proceed- ings of the International Workshop on Nutrient Research in Rainfed Lowlands, 12-15 October 1998, Ubon Ratchathani, Thailand. Manila (Philippines): International Rice Research Institute. p 245-256.

Identifying soil suitability for subsoil compaction . . . 109 Notes Authors’ addresses: D. Harnpichitvitaya, G. Pantuwan, and D. Suriya-Arunroj, Rice Research Center, Ubon Ratchathani, Thailand; G. Trébuil, Department of Annual Crops, Centre de coopération internationale en recherche agronomique pour le développement (CIRAD- CA), BP 5035, 34032, Montpellier Cedex 1, France. Seconded to Agronomy, Plant Physi- ology, and Agroecology Division, International Rice Research Institute, Los Baños, Laguna, Philippines; T. Oberthür, International Center for Tropical Agriculture, Cali, Colombia; I. Craig, Land Reform Area Development-AIDAB project for Det Udom District, Ubon Ratchathani Province; T.P. Tuong and L.J. Wade, International Rice Re- search Institute. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

110 Harnpichitvitaya et al Modeling water availability, crop growth, and yield of rainfed lowland rice genotypes in northeast Thailand

S. Fukai, J. Basnayake, and M. Cooper

Characterizing and understanding the nature of the limitations imposed by rainfed lowland environments within a target region are important for developing an efficient rice breeding program. Major factors determining the environment as it influences plant performance are soil properties and water availability. Rainfall is a major determinant of the water environment, but its seasonal variability is high in northeast Thailand. Simulation models are useful in determining many features of the paddy water environment, as they can readily estimate water availability as a probability function by using past rainfall patterns as inputs. The RLRice model was developed for simulating the paddy water balance and growth of cultivar KDML105 for the rainfed lowlands in Thailand. The model was used to quantify the water balance of paddies of the several locations used in the rainfed lowland rice breeding program in northeast Thailand. Yearly variation in simulated yield at any location was related to variation in rainfall during crop growth. However, sensitivity analysis results showed that simulated yield varied greatly with changes in components of the water balance, particularly the deep percolation rate, lateral water movement, and initial water level at transplanting. Simulated yield was generally associated with the time of disappearance of water relative to flowering, and the depth of the free water level at flowering. The simulation results indicate a strong interaction of genotype and water environment through variation in water availability in relation to phenological development of each genotype.

Progress in breeding for high-yielding cultivars for rainfed lowland rice is slow in many countries, partly because genotypic ranking for yield varies greatly in different environments as a result of large genotype-by-environment (G × E) interactions for grain yield. Often, the G × E interaction variance for yield is several times greater than the genotype variance alone in rainfed lowland rice (Cooper et al 1999a). The large interaction for yield is related to the large environmental variation within the

Modeling water availability, crop growth, and yield . . . 111 target area of a breeding program. Wade et al (1999b) have documented the spatial and temporal variability in growing environments and yield in rainfed lowlands within Asia. Variation occurs because of soil type and biotic factors, but the main cause of the large yield variation is often related to variation in water availability in rainfed lowlands. Water availability in lowland fields affects the timing and intensity of drought and submergence development, and also causes reduced nutrient availability (Fukai et al 1999a). Thus, yield is often reduced with drought because plant growth is affected by reduced water and nutrient availability. Simulation models have been used successfully to determine the target environments of a breeding program in some crops (e.g., sorghum, Chapman and Barreto 1997). This area of work is being used to help develop new cultivars that are suitable for target environments. The approach has not yet been applied to the characterization of environments for rainfed lowland rice, partly because of the difficulty of quantifying the water balance for rainfed lowlands, particularly lateral water movement across paddies. This chapter aims to evaluate factors that determine water availability in rainfed lowlands and demonstrate how a change in water balance components may interact with genotypes of different phenology. The RLRice model (Fukai et al 1995) was used to estimate the water balance and grain yield for a range of environments in northeast Thailand. The mean yield in northeast Thailand is 1.7 t ha–1 and popular cultivar KDML105 can produce 4.0 t ha–1 under favorable conditions.

Water balance in rainfed lowlands The position of lowland fields in a toposequence is important in determining water availability (Wade et al 1999b). Lowland fields in the high or upper terraces lose a large amount of water readily, particularly after heavy rainfalls, through surface runoff and underground lateral water movement, while those in the lower terraces may intercept the water that flows in from the paddies in an upper position. This lateral water movement results in different periods of water availability and growing duration, often more than 30 days within a small area. Thus, the upper terraces may be classified into the drought-prone subecosystem and the lower terraces may belong to the submergence-prone or drought- and submergence-prone subecosystem. Most farms are composed of a mixture of these different positions in the toposequence in varying proportions, depending on their location. When selection trials are conducted in a rainfed lowland rice breeding program, the positions of the trials in the toposequence are likely to have large effects on the relative yield of the breeding lines and therefore on the lines selected. The change in soil water content or free water level above or below the soil surface is described by the equation

∆S = R – (E + T + P + L + O)

112 Fukai et al Rainfall Transpiration

Evaporation Runoff

Lateral Percolation movement

Fig. 1. Diagrammatic representation of the components involved in water balance in terraced rainfed lowland paddies.

where ∆S is the change in soil water content, R the rainfall, E the evaporation from standing water surface or soil surface, T the transpiration, P the deep percolation, L the net lateral water movement in the soil (positive means loss of water from the particular field), and O the runoff of water above the bund (Fig. 1). Water availability during growth has a direct effect on grain yield in rainfed lowland rice. Jearakongman et al (1995) found that yield was high when standing water was maintained until flowering, but it decreased sharply when standing water disappeared before flowering. Because of late flowering, late-maturing cultivars were more affected than early maturing cultivars by late-season drought. The interaction between environmental variation in the timing of drought and plant development has contributed to many of the large G × E interactions for grain yield of rainfed lowland rice that are observed in the multienvironment trials conducted in Thailand. The deep percolation rate varies greatly, and this causes large variation in total water loss from lowland fields (Yahata 1976). One way of altering the water balance is by puddling or compacting the soil to reduce the deep percolation rate and this affects grain yield (Sharma et al 1988). The effect of variation in deep percolation rate on grain yield, however, depends on the phenology type of the cultivar used (Fukai 1996).

The RLRice model A rainfed lowland rice model was developed for cultivar KDML105 (Fukai et al 1995). A brief account of this model is given here, concentrating on aspects of the model that are relevant to this study. The model consists of several sections. Daily rainfall, mean temperature, solar radiation, and pan evaporation are weather inputs to the model. The water balance submodel uses the equation mentioned earlier to estimate daily water balance from the time of transplanting to maturity. Both evaporation (E) and transpiration (T) are determined from daily pan evaporation, soil water content, and canopy ground cover,

Modeling water availability, crop growth, and yield . . . 113 which is calculated from the estimated leaf area of the crop. The transpiration rate decreases when soil water content is less than 75% of the total extractable soil water content. The deep percolation rate (P) is a characteristic of the soil type at a location and is a required input for the model. The deep percolation rate is assumed to be 0 when the free water surface is below the effective root zone. The model has two soil layers: the top 10 cm and the subsurface soil layer. Soil evaporation takes place only from the top layer, whereas T can occur from both layers. The water-holding capacity for each soil layer has to be estimated and is an input. Roots are assumed to extend at the rate of 2 cm d–1. Parameters E and T are reduced as soil water content decreases in the top and bottom soil layers, respectively. Lateral water movement (L) depends on rainfall, soil water level, and the position of the lowland field in a toposequence. This is estimated from a coefficient CL and rainfall (R) such that L = R*(CL – 1) if the soil is saturated with water and rainfall is large. Thus, the total water available to the paddy from rainfall is estimated as R*CL; CL >1.0 indicates that the lowland field is located at the lower part of the toposequence and gains water from net lateral water movement. Net loss of water by lateral water movement is estimated as R*(1.0 – CL) where CL <1.0. When CL = 1.0, the lowland field is located in a position where there is no net lateral water movement. Floodwater level at transplanting varies in each crop and is an input. The subsequent water level below or above the soil surface is estimated from R, E, T, P, L , and O throughout growth. Crop growth rate (CGR) is calculated daily from incident solar radiation, the proportion of radiation intercepted by the crop canopy, and radiation-use efficiency (amount of biomass produced per unit radiation intercepted) when soil water does not limit growth. Radiation-use efficiency is affected by soil fertility and is an input for each location. The proportion of radiation intercepted by the crop canopy is estimated from the leaf area index. The leaf area index is estimated from dry matter production, partitioning of assimilates to leaves, and the ratio of leaf area to leaf dry weight. When plants are water-stressed, CGR is calculated as the product of T and transpiration efficiency (amount of biomass produced per unit water transpired). Accurate estimation of CGR is vital, as it determines not only total biomass production and leaf area index but also the yield components directly in the model. Panicle number m–2 is proportional to CGR before panicle initiation, whereas spikelet number panicle–1 is proportional to CGR between panicle initiation and anthesis. The unfilled grain proportion and single grain weight are determined directly by CGR after anthesis. These yield components are, however, also affected by water stress, which occurs near and after anthesis. Thus, the proportion of unfilled grain increases as severe water stress develops near anthesis; the proportion of unfilled grain increases linearly with the number of stress days around anthesis, with the regression coefficient indicating a genotype’s tolerance of or susceptibility to drought near anthesis. Under severe water-stress conditions, the plants may die prematurely. In this case, grain yield is estimated from dry matter produced after flowering and a proportion of assimilates produced before anthesis, which are translocated to fill grains.

114 Fukai et al Seeding date and transplanting date are inputs. If old seedlings are used for transplanting, the time from transplanting to anthesis is shortened and the potential for high grain yield will be reduced. While the model was developed for cultivar KDML105, the phenology of the cultivar can be modified in the model. The date of flowering is assumed to depend on the extent of photoperiod sensitivity, the length of the basic vegetative phase in which the plants do not respond to photoperiod, and the actual date of sowing. Flowering is delayed if severe water stress develops during the vegetative and reproductive stage before flowering. On the other hand, water stress after flowering hastens senescence. The stage of phenological development determines assimilate partitioning among roots, stems, leaves, and panicles, including grain. The proportion of assimilates allocated to the roots and leaves decreases gradually while that allocated to the stem increases after transplanting until about the time of anthesis. Late-flowering genotypes have a longer time for leaf development and, hence, given an adequate water supply, have a higher grain yield potential. The model was calibrated for rainfed lowland conditions using data collected for six different locations in Thailand. The simulated results have been compared with experimental results for KDML105 and give good agreement (Fukai et al 1995). The model was used to estimate the magnitude of drought effects in northeast Thai- land (Jongdee et al 1997), G × E interaction for flowering time and grain yield (Henderson et al 1996), and water and nutrient availability interaction (Fukai et al 1999a,b). The present work estimates how genotypes may interact with variation in various components of the water balance equation to demonstrate G × E interaction caused by variation in water availability.

Simulation procedures Five sets of simulation experiments were conducted. 1. Yearly yield variation due to rainfall for a given condition at transplanting Yearly variation for yield in relation to rainfall during the growth period was estimated from simulations for Ubon Ratchathani using 22 years’ rainfall data from 1975 to 1996. For these simulations, all input values other than rainfall were constant for each year and were those obtained in an experiment conducted in 1993. The standard inputs used for Ubon are as follows: time of seeding, 16 July; age of seedlings at transplanting, 40 d; standing water level at transplanting, 10 mm; lateral water movement coefficient, 1.0; deep percolation rate, 6.0 mm d–1 (Jongdee et al 1997). The time when standing water disappeared from the lowland field was estimated for each year and this time was expressed in relation to flowering time. Free water level at flowering was also used as an indicator of water availability to the crop.

2. Sensitivity analyses of water balance components The location of selection trials within a small area may result in different values of the water balance components, and may have a large effect on water availability and

Modeling water availability, crop growth, and yield . . . 115 hence grain yield. The three major parameters that affect the water balance in rainfed lowlands—standing water at transplanting, lateral water movement coefficient, and deep percolation rate—are altered independently of each other to examine the magnitude of the effect of variation in each parameter value (sensitivity analysis). In these simulation runs, the standard values of the parameters for Ubon, except for the one being examined, are used together with Ubon 1996 rainfall data. The simulation was repeated for Phrae using the standard parameter values for Phrae and the 1993 rainfall data for the location. The standard values were time of seeding, 8 July; age of seedlings at transplanting, 38 d; standing water level at transplanting, 50 mm; lateral water movement coefficient, 1.0; and deep percolation rate, 1.0 mm d–1. The standing water at transplanting was either –200 or 50 mm. The former simulates the case where there is no standing water at transplanting but the lowland field would be moist enough to allow successful transplanting of seedlings. The latter is considered to represent the maximum water depth at transplanting. Values for the coefficient for lateral water movement (CL) used were 0.5, 0.75, 1.0, 1.25, and 1.5. The values of 0.5 and 0.75 would represent the fields at high positions in a toposequence and 1.25 and 1.5 the fields at low positions. Deep percolation rates used were 1, 4, and 6 mm d–1. The common range of this parameter is 1–6 mm d–1 in northeast Thai- land (Fukai et al 1995).

3. Response of genotypes with different phenology to variation in water balance components In this section, five genotypes of different phenology type were used for simulation for Ubon. These were the same as those used by Henderson et al (1996) for simulation of G × E interaction analysis and they cover the range of phenology types common in northeast Thailand. The standard genotype was KDML105, which is photoperiod-sensitive and late-flowering, commonly flowering around 25 October in northeast Thailand. Chiangsaen is strongly photosensitive and later flowering than KDML105. RD23 is almost completely photoperiod-insensitive and early flowering. NSG19 and IR57514-PMI-5-B-1-2 are both mildly sensitive, but NSG19 flowered earlier than IR57514-PMI-5-B-1-2. Only phenology was altered in these simulations and other cultivar characteristics remained the same as for KDML105. The simulations were run for the higher, middle, and lower positions of a toposequence by using the lateral water movement coefficient (CL) of 0.5, 1.0, 1.25 (only for the 16 July simulation in 1994), and 1.5, respectively. The two seeding dates used were 16 June and 16 July using rainfall data for Ubon in both 1994 (development of some water stress) and 1996 (favorable water conditions). For the simulation with CL of 1.25 in the 16 July seeding, the deep percolation rate was altered. The deep percolation rate may be lower in lower toposequence positions as the soil tends to have more clay content (Wade et al 1999b). This was simulated by changing the deep percolation rate from the standard of 6 to 4 and 2 mm d–1. The standard value for water-holding capacity of the soil for Ubon is 55 mm, and the effect of an increase to 75 mm was simulated. Wade et al (1999b) showed variation of about 20 mm along the toposequence in their measurements in Bangladesh.

116 Fukai et al 4. Effects of seedling age at transplanting Another source of G × E interaction is the use of different seedling ages at transplanting in different trials. A genotype’s response to seedling age at transplanting may be different in different water availability conditions. The simulation was conducted for Ubon using 1994 and 1996 rainfall data. Seedling ages used were 25, 40, and 55 d old. The 25-d-old seedling would represent young seedlings used when water is available at the optimum transplanting time. The older seedlings represent cases where water is not sufficient in lowland fields when seedlings are young and transplanting is delayed for 15–30 d.

5. Yield and yield stability of genotypes with different phenology This set of simulations used the same five genotypes with different phenology mentioned above and the standard parameter values obtained from experiments conducted in 1993 at nine locations in Thailand: three from northern Thailand and six from northeast Thailand (Fig. 2). The simulation was conducted using the past 20 years’

Myanmar

Gulf of Lao PDR Tonkin Vietnam SPT PRE SKN PSL CPA KKN

PMI UBN SRN

Bangkok

Cambodia

Gulf of Thailand

Andaman Sea

Malaysia

Fig. 2. Map of Thailand showing nine northern and northeastern locations where cultivar comparisons were made.

Modeling water availability, crop growth, and yield . . . 117 rainfall data for each location. Mean yield and standard deviation of yield were estimated for each genotype at each location.

Results 1. Yearly yield variation due to rainfall for a given condition at transplanting The simulated yield level at Ubon was generally low in these simulations because seeding occurred rather late (16 July), the deep percolation rate was high (6 mm d–1), and soil fertility was low. Figure 3 shows water levels in the field from transplanting to maturity for four contrasting years; 1980 was one of the highest rainfall years, with standing water lasting until flowering, whereas 1986 had periods in early vegetative stages when there was no standing water and standing water disappeared rather early, resulting in yield of 0.93 t ha–1. The third year, 1993, was a very dry year and standing water disappeared well before flowering, resulting in a low yield of 0.58 t ha–1. In 1996, the water level was low immediately after transplanting and standing water also disappeared about 10 d before flowering. Yearly variation in yield was mostly accounted for by the variation in rainfall during the period from transplanting to maturity (Fig. 4A). In years of high rainfall,

Water level (mm) 400 F F

–400

–1 –1 –800 Simulated yield = 1.94 t ha Simulated yield = 0.93 t ha 1980 1986

–1,200

400 F F

–400

–800 1993 Simulated yield = 1.79 t ha–1 –1 Simulated yield = 0.58 t ha 1996 –1,200 40 60 80 100 120 140 40 60 80 100 120 140 Days after sowing Fig. 3. Simulated water levels during growth and flowering of genotype KDML105 for Ubon Ratchathani in four selected years. The letter “F” indicates time of flowering.

118 Fukai et al Grain yield (t ha–1) 3 R2 = 0.67 A Y = –0.49 + 0.0037X

84 2 80 75 8296 87 83 91 76 90 1 86 92 81 78 95 85 93 89 94 88 77 79 0 0 100 200 300 400 500 600 700 800 Rainfall (mm) 3

R2 = 0.42 B Y = 1.47 + 0.020X

84 2 80 96 75 82 83 87 76 91 90 1 8692 81 78 F 95 85 93 8994 88 77 79 0 –60 –40 –20 0 20 Time of disappearance of standing water (days to flowering) 3

R2 = 0.61 C Y = 1.74 + 0.0018X

84 80 2 75 9682 83 87 76 91 1 92 90 86 81 78 95 85 93 8994 77 88

79 0 –1,000 –800 –600 –400 –200 0 200 400 Water level at flowering (mm)

Fig. 4. Relationship between simulated grain yield and (A) rainfall from transplanting to maturity, (B) date of disappearance of standing water relative to flowering (– indicates before flowering and + indicates after flowering), and (C) water levels at flowering for 22 years for Ubon Ratchathani.

Modeling water availability, crop growth, and yield . . . 119 such as 1980 and 1996, yield was high, whereas, in years of low rainfall, such as 1993, yield was low. In years of high rainfall, standing water disappeared from the lowland field later and the water level at flowering was higher (Fig. 4B,C). Yield was below 1 t ha–1 when standing water disappeared more than 20 d before flowering or the water level at flowering was less than 400 mm below the soil surface.

2. Sensitivity analyses of water balance components The standing water level at transplanting showed a large simulated effect on grain yield because water availability to the crop during subsequent growth periods was modified. In Figure 5, the simulated water level throughout growth is shown when the initial water level at transplanting was –200 mm and 50 mm. With the initial water level at 200 mm below the soil surface, there was no standing water for 15 d after transplanting and standing water also disappeared several days earlier. This resulted in a simulated yield of 1.41 t ha–1 versus 1.84 t ha–1 for the 50-mm water treatment with a longer period of standing water. Figure 6 shows simulation results of the effect of a change in the lateral water movement coefficient (CL) in 1996 together with the results in 1994. The period of standing water increased markedly with the increase in CL from 0.5 to 1.5 in both years.

Water level (mm)

–400

–1 –800 Simulated grain yield = 1.41 t ha –200 mm (water level at transplanting) F –1,200

–400

Simulated grain yield = 1.84 t ha–1 –800 50 mm (water level at transplanting)

F –1,200

40 60 80 100 120 140 Days after sowing

Fig. 5. Simulated water levels during the growth of genotype KDML105 for two levels of standing water at the time of transplanting at Ubon Ratchathani. The letter “F” indicates time of flowering.

120 Fukai et al Water level (mm)

UBN 94 (0.5) UBN 96 (0.5) 400 (B,C) (B,C) (E) (D) (A) (E) (D) (A) 0

–400

–800

–1,200

UBN 94 (1.0) UBN 96 (1.0) 400 (E) (D) (B,C) (B,C) (A) (E) (D) (A) 0

–400

–800

–1,200

UBN 94 (1.5) UBN 96 (1.5) 400 (E) (D) (E) (B,C) (A) (D) (B,C) (A) 0

–400

–800 40 60 80 100 120 140 40 60 80 100 120 140 Days after sowing

Fig. 6. Simulated water levels under three levels of lateral water movement coefficients (0.5, 1.0, and 1.5) during the growth of five genotypes: Chiangsaen (A), KDML105 (B), IR57514 (C), RD23 (D), and NSG19 (E) tested at Ubon Ratchathani in 1994 and 1996. Down arrows indicate time of flowering of each genotype. The filled bars indicate the period when KDML105 had died because of severe drought.

Modeling water availability, crop growth, and yield . . . 121 Table 1 shows the results of the sensitivity analyses for simulated change in yield for the coefficient for lateral water movement and water level at transplanting and for the coefficient for lateral water movement and deep percolation rate. The simulation results show that the effect of a change in lateral water movement would be greater with an initial water level of +50 mm than with –200 mm at Ubon 1996 because lateral water movement would be greater with a longer period of standing water and, hence, the soil is saturated with water for a longer period. When the water level at transplanting was reduced to –200 mm, the period of standing water was shorter, and this resulted in a rather small effect on grain yield with the variation in CL. At Ubon, the effect of a change in CL became smaller as the deep percolation rate decreased from 6 to 1 mm d–1. With 1 mm d–1, there was almost no water stress throughout the growth period and hence the effect of water movement was small, whereas, with 4–6 mm d–1, there were some periods with standing water but there were also periods of water stress (Fig. 5 for a deep percolation rate of 6 mm d–1). In this case, there was a large simulated effect of variation in lateral water movement. The results are different for Phrae 1993, where yield potential was higher and simulated yield was more responsive to a change in initial water level and deep per-

Table 1. Simulated grain yield (t ha–1) for KDML105 under different degrees of lateral movement of water and initial water level (mm) at the time of transplanting and deep percolation rate (mm d–1) at Ubon (1996) and Phrae (1993).

Coefficient for lateral movement

Water level at of water (CL) transplanting 0.5 0.75 1.0 1.25 1.50

Ubon –200 1.32 1.34 1.41 1.45 1.52 +50 1.56 1.72 1.84 2.12 2.26 Phrae –200 0.94 1.41 2.25 2.97 3.18 +50 2.08 2.89 3.26 3.49 3.69

Coefficient for lateral movement

Deep percolation of water (CL) rate 0.5 0.75 1.0 1.25 1.50

Ubon 1 2.28 2.46 2.46 2.46 2.46 4 1.78 1.85 2.15 2.37 2.41 6 1.56 1.73 1.79 1.88 2.12 Phrae 1 2.08 2.89 3.26 3.49 3.69 4 0.37 0.61 0.89 1.34 1.81 6 0.35 0.49 0.51 0.76 0.88

122 Fukai et al colation rate. In this case, the larger effect of variation in lateral water movement was found with lower water availability (–200 mm) at transplanting and a lower percolation rate. With the highest percolation rate of 6 mm d–1, the water-stress effect was severe and the effect of variation in CL was small. In these simulations, grain yield was directly related to the date of disappearance of standing water relative to the flowering date for both locations and Figure 7 shows the results for Ubon. Note that relationships obtained from the sensitivity analyses are similar to those obtained from the analysis of yearly variation (Fig. 4B,C).

3. Response of genotypes with different phenology to variation in water balance components For the June planting in 1994 at Ubon, the yield of genotypes with different phenology increased with a delay in flowering until about 10 October with the lateral water movement coefficient of 0.5 and until about 20 October with the coefficients of 1.0 and 1.5 (Fig. 8). However, with seeding later in July, earlier flowering was advanta- geous with a CL of 0.5, 1.0, and 1.25, whereas the phenology had a small effect when the coefficient was 1.5, with a slight advantage with later flowering. It should be pointed out that the order of flowering among genotypes changed between the two seeding dates because of differences in photoperiod sensitivity. For example, the mildly photosensitive IR57514-PMI-5-B-1-2 flowered earlier than KDML105 when planted on 16 June, but they flowered at almost the same time when sown on 16 July. In 1996, the water conditions were generally more favorable and the highest yield was obtained with the KDML105 phenology type for all values of CL in both June and July sowing. In Figure 6, where the water level throughout the growth period for the July sowing is shown, flowering time of each genotype is also shown for both 1994 and 1996. With a CL of 0.5 and 1.0, the three late genotypes flowered well after the standing water disappeared from the lowland field and the yields were lower. With the CL of 1.5, standing water was maintained until the last genotype flowered and the later

Grain yield (t ha–1) 3 3 R2 = 0.73 A B Y = 1.89 + 0.025X 2 2

6 mm d–1 1 50 mm –1 1 R2 = 0.97 4 mm d F –200 mm –1 Y = 1.99 + 0.018X F 1 mm d 0 0 –30 –20 –10 0 10 20 30–30 –20 –10 0 10 20 30 Time of disappearance of standing water (days to flowering) Fig. 7. Relationship between the simulated grain yield and date of disappearance of water relative to flowering time for five genotypes. (A) with two levels of standing water at transplanting and (B) three levels of deep percolation rate at Ubon Ratchathani in 1996.

Modeling water availability, crop growth, and yield . . . 123 Grain yield (t ha–1) 3 UBN 1994

3 UBN 1996

RD23 1 NSG19 IR57514 KDML105 CS 0 13 Sep 23 Sep 3 Oct 13 Oct 23 Oct 2 Nov 12 Nov Flowering date Fig. 8. Simulated grain yield in relation to flowering date for five genotypes seeded at two different times in 1994 and 1996 (June, open symbols; July, closed symbols) under four levels of lateral water movement coefficients ( 1.5; 1.25; 1.0; . 0.5 mm d–1) at Ubon Ratchathani (UBN).

genotypes were not disadvantaged. In 1996, water disappearance was several days later, and this resulted in the later flowering genotypes maintaining yields comparable with those of the early genotypes, even with the CL of 0.5. The simulation results of a change in deep percolation rate for Ubon 1994 where the CL of 1.25 was used suggest that later flowering genotypes may not be disadvantaged at the lower toposequence position if the deep percolation rate was reduced from the standard 6 mm d–1 to 2–4 mm d–1 (Fig. 9). The effect of a change in soil water-holding capacity from 55 to 75 mm was small, however. Yield increased from

124 Fukai et al Grain yield (t ha–1) 3 2 mm

2 4 mm

RD23 1 NSG19 IR57514 KDML105 6 mm CS 0 13 Oct 23 Oct 2 Nov 12 Nov Flowering date

Fig. 9. Simulated grain yield in relation to flowering date of five genotypes tested under a lateral water movement coefficient of 1.25 mm d–1 and three levels of deep percolation rate at Ubon Ratchathani in 1994.

Table 2. Simulated effects of seedling age at transplanting on grain yield (t ha–1) for five rice phenology groups (genotypes) grown at Ubon; 1994 and 1996 rainfall data were used to simulate the grain yield of these genotypes.

Seedling age at transplanting Genotypes 25 d 40 d 55 d

1994 NSG19 1.96 1.17 0.55 RD23 2.02 0.96 0.48 IR57514 1.79 0.52 0.17 KDML105 1.88 0.54 0.19 Chiangsaen 1.38 0.39 0.09

1996 NSG19 1.79 1.55 1.10 RD23 1.83 1.61 1.15 IR57514 1.93 1.79 1.41 KDML105 1.92 1.80 1.39 Chiangsaen 1.60 1.43 1.09

1.0 to 1.3 t ha–1 with a deep percolation rate of 6 mm d–1, but the increase was smaller with a deep percolation rate of 4 mm d–1.

4. Effects of seedling age at transplanting There were large adverse effects for the use of old seedlings for transplanting, and the effects were greater in 1994, when stress developed earlier (Table 2). Crop growth

Modeling water availability, crop growth, and yield . . . 125 decreased with the use of old seedlings because the period from transplanting to flowering in the lowland field was shortened. This was particularly obvious when water stress developed. However, interaction with genotypes was relatively small; it was apparent only in 1994, when early flowering genotypes (NSG19, RD23) were advan- tageous under delayed transplanting. As the sowing was late at 16 July, IR57514- PMI-5-B-1-2 (mildly photosensitive) and KDML105 (strongly photosensitive) had simulated flowering at about the same time, and the effect of delayed transplanting was almost the same in both years.

5. Yield and yield stability of genotypes with different phenology There were genotype-by-location interactions for mean yield in these simulations (Table 3). The highest yield was obtained with later maturing genotypes (KDML105, Chiangsaen) at Surin and Phrae, where yield was generally high, with little adverse effect of late-season drought. The earlier genotypes (NSG19, RD23) produced a higher yield than the others at Phimai and Ubon, where late-season drought was a common problem. These interactions reflect rainfall patterns and the standard water balance parameter values used for the different locations. On the other hand, the standard deviation for yield was commonly higher for later-flowering genotypes. These results indicate the more stable nature of early flowering genotypes compared with late- flowering genotypes. The standard deviation of each genotype differs markedly among locations. Yield obtained at Phimai generally had a larger standard deviation than at the other locations, whereas it was small at Khon Kaen and Surin.

Table 3. Simulated grain yield (t ha–1) (average of 20 years) and their standard deviation (Std) for five phenology groups grown in nine locations in Thailand.

Cultivar

Loca- NSG19 RD23 IR57514 KDML105 Chiangsaen tiona Mean Std Mean Std Mean Std Mean Std Mean Std

CPA 1.82 ±0.68 1.84 ±0.68 2.64 ±1.08 1.86 ±0.74 2.79 ±1.39 KKN 1.74 ±0.24 1.53 ±0.21 2.13 ±0.29 3.23 ±0.25 2.33 ±0.41 PMI 2.50 ±1.29 2.21 ±0.85 2.08 ±0.78 1.64 ±0.79 2.22 ±1.28 PRE 2.36 ±0.36 2.29 ±0.47 3.03 ±1.06 3.02 ±0.47 3.30 ±1.38 PSL 2.29 ±0.27 2.49 ±0.37 3.27 ±0.75 3.23 ±0.74 3.33 ±1.01 SKN 2.36 ±0.58 2.14 ±0.41 2.43 ±0.68 2.32 ±0.80 2.01 ±0.84 SPT 2.19 ±0.64 2.39 ±0.78 2.67 ±1.06 2.73 ±1.06 2.76 ±1.19 SRN 3.26 ±0.35 2.85 ±0.34 4.01 ±0.33 4.33 ±0.35 4.62 ±0.40 UBN 1.33 ±0.31 1.26 ±0.39 1.01 ±0.50 1.03 ±0.53 1.04 ±0.56

aCPA = Chum Phae, KKN = Khon Kaen, PMI = Phimai, PRE = Phrae, PSL = Phitsanulok, SKN = Sakon Nakhon, SPT = Sanpatong, SRN = Surin, UBN = Ubon Ratchathani.

126 Fukai et al Discussion This simulation work shows the importance of maintaining standing water for a long time period during the reproductive to grain-filling stages. This confirms the experimental results in northeast Thailand reported by Jearakongman et al (1995). Two indicators can be used to estimate overall water availability and grain yield in a plant breeding program: time of disappearance of standing water relative to flowering time and free water level at flowering. The latter can be more readily determined if there is standing water at flowering, but this may not be the case if standing water has already disappeared by flowering. Recording the date of disappearance of standing water and dates of flowering of different genotypes should help characterize the water conditions of each selection trial. The water balance in rainfed lowlands is complex and the time of disappearance of standing water relative to flowering is determined by several factors, as demonstrated in the simulation studies here. For a given hydrological and agronomic condition, rainfall between transplanting and flowering affects the relative time of disappearance of standing water. However, deep percolation rate, lateral water movement, and, to a lesser extent, initial water level at transplanting all affect water balance and hence grain yield. Sensitivity analyses of the influence of these factors on grain yield (Table 1) show interactions among these factors as well as with rainfall. These complex relationships would affect the water balance of a lowland field used by a breeding program. Among these interacting factors, rainfall can be determined accurately and initial water level can be readily estimated if there is standing water at transplanting. The deep percolation rate and lateral water movement are important components in determining overall water balance, and a combined rate can be determined readily by monitoring the field water level and using the estimated value of the evapotranspiration rate. A separate estimation of the deep percolation rate and lateral water movement would make a more accurate estimation of water balance possible. This may be particularly important in a toposequence where the lateral water movement component is expected to be large. Lateral water movement is a difficult component of the water balance to determine because it is not constant throughout the growth period. The model estimates the net amount of lateral water movement from the coefficient CL, the amount of rainfall, and soil water level, thus assuming that lateral water movement takes place immediately after a large rainfall event and only when the soil is saturated with water. The value of the coefficient may range from 0.5 to 1.5 as used in the sensitivity analysis in the present work, as this would cause about 30 d of difference in the date of disappearance of standing water in a toposequence. This difference is not uncom- mon in rainfed lowlands (Wade et al 1999b) and cultivars with different maturity groups are often planted at different toposequence positions (Nesbitt and Phaloeun 1997). The development of a method to determine lateral water movement appears important for the characterization of rainfed lowland environments. Water loss through under-bund percolation may also be important in determining the water balance, and may need to be considered in the model (Tuong et al 1994).

Modeling water availability, crop growth, and yield . . . 127 The use of five genotypes with different phenology, some of which correspond to the proposed reference lines for rainfed lowland rice (Wade et al 1999a), shows the G × E interaction due to genotypic phenology differences and water availability, particularly during later growth stages. These interactions are mostly explained by the time of disappearance of standing water relative to flowering of each genotype, thus emphasizing the importance of the water balance in each screening field in determining the ranking of genotypes for yield. The genotypes used for the simulation were assumed to be different only for phenology and not for other characters. Therefore, other important characters, such as potential yield under nonlimiting conditions, drought resistance, submergence tolerance, and tolerance for low nutrient availability and insects and diseases, are not included in the simulation. These factors are known to affect genotypic performance (e.g., Fukai et al 1999a,b, Mackill et al 1999, Ito et al 1999) in rainfed lowland rice, and hence further contribute to the G × E interactions observed in multienvironment trials. In the model, genotypes with a later phenology are assumed to have a longer time for vegetative growth and hence will have a larger biomass at maturity, particularly under favorable water conditions. They generally resulted in higher yield when there were no periods of severe water stress (e.g., Fig. 6, sowing on 16 June and CL of 1.0 or 1.5). This may not be the case in reality, as often later-flowering cultivars have a lower harvest index (Jearakongman et al 1995). Similarly, the potential yield of photoperiod-insensitive cultivars is often higher than that of strongly photoperiod- sensitive cultivars (Mackill et al 1996). These features are not simulated in the present work, although the RLRice model has the capacity to simulate modifications of the variation in potential yield. The present work aimed to identify environmental factors, particularly those associated with water availability, that cause G × E interactions for grain yield. The effects of variation in attributes of genotypes other than phenology, such as drought resistance and potential yield, are the subject of another study. While the initial water level at transplanting is an input to the model, this value is affected by prior rainfall events as well as land preparation. Rainfed lowland breeders often irrigate the field for transplanting, and this practice ensures the use of young seedlings for transplanting. This may, however, affect the selection of genotypes as shown in the simulation results in this study, which favor the selection of later-flowering genotypes. The simulated effect of the use of old seedlings on grain yield was similar to that obtained in the field in our recent experiments in Lao PDR. This simulation study indicated the complex nature of the water balance in rainfed lowlands and suggested how this may interact with the phenology of genotypes in plant breeding experiments. Cooper et al (1999a,b) attributed much of the large genotype-by-site-by-year interactions for grain yield to the effects of environmental variation in the timing and intensity of water deficit relative to the flowering times of the genotypes. Given the strong influence of the water balance component of the rainfed lowland paddy on the relative yield performance of genotypes demonstrated in this study, it is important to consider the characterization of rainfed lowland rice environments sampled at all stages of the plant breeding program if the compli-

128 Fukai et al cations of G × E interactions for yield are to be addressed. Further field studies are required to develop a method for accurately estimating lateral water movement for better characterization of rainfed lowland environments.

References Chapman SC, Barreto HJ. 1997. Using simulation models and spatial database to improve the efficiency of plant breeding programs. In: Cooper M, Hammer GL, editors. Plant adaptation and crop improvement. Wallingford (UK): CAB International and International Rice Research Institute. p 563-590. Cooper M, Rajatasereekul S, Immark S, Fukai S, Basnayake J. 1999a. Rainfed lowland rice breeding strategies for northeast Thailand. 1. Genotypic variation and genotype × environment interactions for grain yield. Field Crops Res. 64:131-151. Cooper M, Rajatasereekul S, Somrith B, Sriwisut S, Immark S, Boonwite C, Suwanwongse A, Ruangsook S, Hanviriyapant P, Romyen P, Porn-uraisanit P, Skulkhu S, Fukai S, Basnayake J, Podlich DW. 1999b. Rainfed lowland rice breeding strategies for northeast Thailand. 2. Comparison of intrastation and interstation selection. Field Crops Res. 64:153-176. Fukai S. 1996. Crop physiological approaches to understanding rice production under water limited conditions. In: Crop research in Asia: achievements and perspectives. Proceed- ings of the 2nd Asian Crop Science Conference, 21-23 August 1996, Fukui, Japan. p 240-245. Fukai S, Rajatasereekul S, Boonjung H, Skulkhu E. 1995. Simulation modeling to quantify the effect of drought for rainfed lowland rice in Northeast Thailand. In: Fragile lives in fragile ecosystems. Proceedings of the International Rice Research Conference. Manila (Philippines): International Rice Research Institute. p 657-674. Fukai S, Inthapanya P, Blamey FPC, Khunthasuvon S. 1999a. Genotypic variation in rice grown in low fertile soils and drought-prone, rainfed lowland environments. Field Crops Res. 64:121-130. Fukai S, Pantuwan G, Jongdee B, Cooper M. 1999b. Screening for drought resistance in rainfed lowland rice. Field Crops Res. 64:61-74. Henderson SA, Fukai S, Jongdee B, Cooper M. 1996. Comparing simulation and experimental approaches to analysing genotype by environment interactions for yield in rainfed lowland rice. In: Cooper M, Hammer GL, editors. Plant adaptation and crop improvement. Wallingford (UK): CAB International and International Rice Research Institute. p 443- 486. Ito O, Ella E, Kawano N. 1999. Physiological basis of submergence tolerance in rainfed lowland rice ecosystem. Field Crops Res. 64:75-90. Jearakongman S, Rajatasereekul S, Naklang K, Romyen P, Fukai S, Skulkhu E, Jumpaket B, Nathabutr K. 1995. Growth and grain yield of contrasting rice cultivars grown under different conditions of water availability. Field Crops Res. 44:139-150. Jongdee S, Mitchell JH, Fukai S. 1997. Modelling approach for estimation of rice yield reduction due to drought in Thailand. In: Fukai S, Cooper M, Salisbury J, editors. Breeding strategies for rainfed lowland rice in drought-prone environments. Proceedings of an International Workshop held at Ubon Ratchathani, Thailand, 5-8 November 1996. p 65- 73.

Modeling water availability, crop growth, and yield . . . 129 Mackill DJ, Coffman WR, Garrity DP. 1996. Rainfed lowland rice improvement. Manila (Philip- pines): International Rice Research Institute. 242 p. Mackill DJ, Nguyen HT, Jingxian Zhang. 1999. Use of molecular markers in plant improvement programs for rainfed lowland rice. Field Crops Res. 64:177-185. Nesbitt HJ, Phaloeun C. 1997. Rice-based farming systems. In: Rice production in Cambodia. Manila (Philippines): International Rice Research Institute. p 31-37. Sharma PK, Datta SK, Redulla CA. 1988. Tillage effects on soil physical properties and wetland rice yield. Agron. J. 80:34-39. Tuong TP, Wopereis MCS, Marquez JA, Kropff MJ. 1994. Mechanisms and control of percolation losses in irrigated puddled rice fields. J. Soil Sci. Soc. Am. 58(6):1794-1803. Wade LJ, McLaren CG, Quintana L, Harnpichitvitaya D, Rajatasereekul S, Sarawgi AK, Kumar A, Ahmed HU, Sarwoto, Singh AK, Rodriguez R, Siopongco J, Sarkarung S. 1999a. Genotype by environment interactions across diverse rainfed lowland rice environments. Field Crops Res. 64:35-50. Wade LJ, Fukai S, Samson BK, Ali A, Mazid MA.1999b. Rainfed lowland rice: physical environment and cultivar requirements. Field Crops Res. 64:3-12. Yahata T. 1976. Physical properties of paddy soils in relation to their fertility. In: The fertility of paddy soils and fertilizer applications for rice. Taipei (Taiwan): Food and Fertiliser Technology Centre for Asian and Pacific Region. p 28-35.

Notes Authors’ address: School of Land and Food Sciences, The University of Queensland, Brisbane, Australia. Acknowledgments: Financial support from the Australian Centre for International Agricultural Research is gratefully acknowledged. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

130 Fukai et al Using reference lines to classify multienvironment trials to the target population of environments, and their potential role in environmental characterization

C.G. McLaren and L.J. Wade

In heterogeneous rainfed environments, cultivar performance interacts with crop management, soil type, topography, and agrohydrology to complicate the task of selecting better-adapted cultivars. To make consistent progress with selection in the presence of these genotype by environment interactions (G × E), it is important to clearly identify the target population of environments and to know how well actual test environments represent this population. This chapter evaluates a methodology for using measurements on a set of reference lines to classify sites according to previously identified response patterns for a target population of environments. Strategies for choosing reference lines, classifying new sites, and deducing their environmental characteristics are examined. The results showed that the reference set was able to capture repeatable G × E patterns provided it contained representatives of all discriminatory genotype groups. The methodology for characterizing new environments on the basis of reference line responses relied heavily on an ability to impute missing values. Although no optimal solution was available, a heuristic solution in the pattern analysis algorithm was satisfactory. Reference lines should be chosen based on how well they match the discriminatory pattern of the genotype group, their agronomic features, knowledge of their physiological responses, and practical issues such as the availability of seed. Based on this analysis, we conclude that a series of small field trials at many locations could be employed to obtain a useful characterization of new environments and allow breeders to weight responses of test lines appropri- ately. If detailed physical and climatic measurements are also made in these environments, the responses can be integrated with geographical information, physiological understanding, and crop modeling to quantify environment frequency, predictability, repeatability, and risk.

Using reference lines to classify multienvironment trials . . . 131 In rainfed agriculture, crop performance is strongly influenced by climatic characteristics and by spatial heterogeneity over soil types, topographic sequences, and agrohydrologic conditions. Cultivar and management interact with these environmental variables, thus complicating the task of identifying improved cultivars. These interactions and the complexity of factors involved make it difficult to adequately define the target population of environments or even to reliably assess cultivar performance over those environments (Wade et al 1995, Cooper 1999). To provide a focus for selection programs, it is essential to clearly identify the target population of environments, their spatial and temporal frequencies, and their characteristics. Wade et al (1996) discussed two broad approaches for characterization of environments: one based on analysis of physical parameters such as soil properties, climate, and hydrology, and the other based on discrimination by reference lines. The first approach, physical characterization, requires access to substantial data sets and a capacity to interpret their implications for crop adaptation. It is hampered by a lack of data and poor understanding of plant responses to complex combinations of environmental factors. Since this is a priority topic for research, however, such relationships will become clearer as data coverage, methodologies, and physiological understanding improve (V.P. Singh et al, this volume). In the second approach, differential responses of reference lines are used as a bioassay for the occurrence of particular conditions (Cooper and Fox 1996, Wade et al 1996). Although this approach requires less environmental data for new test locations, its effectiveness depends on knowledge of the reference lines and their patterns of adaptation. Confidence in the classification is improved and further understanding is developed if key environmental data are collected on the new test environments as well as on cultivar response. The availability of pattern analysis methods permits rigorous analysis of G × E interactions and classification of genotypes and environments into groupings with similar patterns of adaptation or discrimination. Using these methodologies, recent studies have reported a reliable assessment of repeatable G × E interactions in rainfed lowland rice (Cooper et al 1999, Wade et al 1999). Groups of varieties with common patterns of adaptation over environment groups were identified in these studies. Ref- erence lines would be chosen as standard and well-known representatives of these variety groups (Cooper and Fox 1996, Wade et al 1996). The principle is that a few reference lines can be used to represent a wide range of genotypic adaptation. Fur- thermore, discrimination between these lines could be used to classify new test environments into previously identified targets. An improved capacity to classify new test environments is important because breeders need to know how new sites relate to the target population of environments for which they are breeding. Also, performance of a new genotype in a new set of multienvironment trials needs to be assessed relative to the performance of other lines in similar environments. Both these requirements can be met by using a representative set of reference lines since the adaptation of a new genotype may be classified by its similarity in response to the reference line with which it groups and by knowledge of the types of environments in which it and the reference line show simi-

132 McLaren and Wade lar adaptive responses. Thus, the availability of a known reference set of genotypes could greatly assist consideration of whether a new test environment is representative of the selection target and whether or not a new genotype is preferentially adapted to that target. In addition to these immediate advantages for the interpretation of breeders’ evaluations, we propose that reference lines be used to achieve a rapid and integrative assessment of environmental characteristics over space and time and to indicate likely characteristics of new test environments. This would require evaluation of the reference lines in many locations and seasons to sample a broad cross-section of possible environments. For example, locations distributed across the toposequence would be needed in order to assess the pattern of environments represented there. Clearly, integration of physical and biological characterization is the ultimate goal for characterization, as this would lead to an improved assessment of where particular types of environments are present. The frequency and production risk associated with these environments could be assessed using crop simulation (Aggarwal et al 1996, Cooper et al 1999). Since physiological understanding of patterns of genotype adaptation is required for the identification of useful traits conferring an adaptive advantage in particular conditions, widespread use of a set of representative reference lines could provide a first step in this process. This chapter considers the feasibility and methodology for using a set of reference lines to classify sites from multienvironment trials into an identified target population of environments. We consider strategies for selecting reference lines to capture a wide range of G × E interactions for this purpose. We examine methodologies for assessing the characteristics of new sites from multienvironment trials on the basis of their reference line responses. We seek to use these responses to indicate physical and climatic properties of those new environments.

Materials and methods Using data from a series of 36 multienvironment trials with 47 entries conducted between 1994 and 1998 across South and Southeast Asia by the Rainfed Lowland Rice Research Consortium and at the International Rice Research Institute, Wade et al (1999) reported that G × E interactions were repeatable and genotypes and environments with similar patterns could be grouped consistently. The genotypes and environments they studied and their groupings and some class properties are listed in Figures 1 and 2. We seek to identify representative individuals from these genotype groups for use as reference lines and to evaluate whether the G × E classification can be repeated with just the reference lines. This would indicate whether a new environment could be correctly classified based on reference lines alone. In the selection of reference lines, we are concerned with sensitivity of response of lines to environmental characteristics and would not like this sensitivity to be ob- scured by mean genotypic effects (McLaren 1996). Hence, we have used pattern analysis based on mean polished yield response data (genotypes and environments centered with least squares adjustment for imbalance).

Using reference lines to classify multienvironment trials . . . 133 Grain Site Year Code Group yield Sand Clay Silt pH (19xx) (t ha–1) (%) (%) (%)

Faisabad, IND 97 FC 3.10 35 16 49 7 Udorn, THA 95 TD 63 1.90 50 20 30 6 Phimai, THA 95 TE 0.88 43 40 17 5 Phimai, THA 97 TK 3.11 46 38 16 5 2.25 Raipur, IND 94 IA 46 1.26 15 51 35 7 Raipur, IND 97 II 3.44 12 36 52 7 2.35 Rajshahi, BGD 94 BA 2.26 28 44 28 6 Rajshahi, BGD 96 BC 3.11 28 40 32 6 Faisabad, IND 95 FA 2.77 35 16 49 7 Faisabad, IND 96 FB 2.75 35 16 49 7 Jagdalpur, IND 96 IG 62 2.43 31 44 26 6 Raipur, IND 97 IH 0.11 15 50 35 7 Jagdalpur, IND 97 IK 2.63 31 44 26 6 Ubon, THA 94 TA 1.60 74 6 20 4 Ubon, THA 95 TB 2.26 74 6 20 4 Chumphae, THA 95 TC 0.21 60 32 8 4 Chumphae, THA 96 TG 1.48 45 32 22 6 Sakon Nakhon, THA 96 TH 1.50 76 7 17 4 1.93 Rajshahi, BGD 95 BB 2.03 24 36 40 6 Rajshahi, BGD 97 BD 2.06 28 40 32 6 Raipur, IND 95 ID 0.38 18 51 32 8 Raipur, IND 96 IE 57 0.80 16 53 31 8 Jakenan, IDN 96 OB 1.78 26 12 62 6 Ubon, THA 96 TF 1.24 74 6 20 4 Ubon, THA 97 TI 2.29 74 6 20 4 Chumphae, THA 97 TJ 1.78 45 32 22 6 1.54 Jakenan, IDN 96 OA 60 4.62 26 12 62 6 Jakenan, IDN 97 OC 3.50 26 12 62 6 4.06

Tarlac, PHL 94 PA 21 4.00 20 39 42 6

Raipur, IND 95 IC 10 2.55 15 53 33 7

Bilaspur, IND 97 IJ 16 2.51 15 53 32 7

Raipur, IND 95 IB 1.74 30 35 35 7 Guimba, PHL 95 PB 1.89 34 20 46 7 Tarlac, PHL 95 PC 61 2.11 20 39 42 6 Muñoz, PHL 95 PD 2.38 33 47 33 7 Tarlac, PHL 96 PE 0.92 37 21 42 1.81

Fig. 1. Dendrogram of environment groups from pattern analysis of mean polished yield data. IND = India, THA = Thailand, PHL = Philippines, BGD = Bangladesh, IDN = Indonesia. (Reprinted from Wade et al (1999), Field Crops Research 64:35-50, with permission from the publisher.)

134 McLaren and Wade Grain Plant Days to Filled 1,000– Designation Code Group yield height flowering grain grain (t ha–1) (cm) (%) weight (g)

Sabita 03 2.08 127 111 70.3 28.9 Sabita-A 3A 2.15 128 114 65.4 28.3 Sabita-B 3B 2.12 128 113 65.0 28.8 KDML105 04 82 1.76 122 111 62.7 24.7 KDML105-A 4A 1.76 125 111 60.2 23.9 KDML105-B 4B 1.77 124 112 60.1 23.5 IR57546-PMI-1-B-2-2> 21 1.50 122 116 61.2 23.3 1.88 125 112 63.5 25.9 NSG19 02 2.12 126 96 69.8 27.1 NSG19-A 2A 52 2.25 128 96 66.7 27.0 NSG19-B 2B 2.21 126 96 67.8 27.1 2.19 127 96 68.1 27.1 IR58821-23-1-3-3 22 1.56 110 121 54.5 23.1 IR66469-17-5-B 25 78 1.76 113 108 53.6 21.7 IR66516-11-3-B 27 1.64 110 113 51.3 24.4 IR66516-24-3-B 28 1.91 116 114 54.6 24.5 IR66516-37-7-B 29 2.00 113 109 64.1 23.0 1.78 112 113 55.6 23.4 Mahsuri 05 2.00 106 109 70.2 17.4 Mahsuri-A 5A 2.19 109 110 69.1 16.8 Mahsuri-B 5B 79 2.15 110 110 71.2 16.9 IR66883-18-2-B 36 1.76 102 111 62.3 22.9 IR66883-18-3-B 37 1.75 103 111 61.2 24.0 IR66883-44-3-B 38 1.86 108 111 57.4 21.9 1.95 106 111 65.2 20.0 IR52561-UBN-1-1-2 15 1.90 118 97 57.9 21.0 IR54071-UBN-1-1-3-1> 16 2.26 106 101 59.1 24.6 IR57515-PMI-8-1-1-S> 20 2.67 107 103 66.8 27.0 IR66506-5-1-B 26 2.01 99 107 58.5 24.3 IR66879-19-1-B 30 84 2.07 105 105 56.1 20.5 IR66879-2-2-B 31 2.11 103 106 58.4 21.1 IR66879-20-2-B 32 2.18 103 108 57.0 20.5 IR66879-8-1-B 33 2.01 101 108 56.0 20.5 IR66883-11-1-B 35 2.15 111 99 66.0 26.0 2.15 106 104 59.5 22.8 IR20 01 2.03 85 102 64.7 19.8 IR20-A 1A 81 2.04 86 104 63.9 19.5 IR20-B 1B 2.02 86 104 63.9 18.9 CT9993-5-10-1-M 11 1.34 91 95 59.0 22.0 IR66893-5-2-B 39 2.05 104 107 62.7 25.6 IR58307-210-1-2-3-3> 62 2.24 91 109 63.7 18.5 1.95 90 103 63.0 20.7 IR54977-UBN-6-1-3-3> 17 2.26 94 99 62.8 28.1 IR57514-PMI-5-B-1-2 19 2.33 101 107 70.0 24.8 IR62266-42-6-1 23 76 2.26 92 98 64.5 22.4 IR63429-23-1-3-3 24 2.37 110 96 69.9 24.9 IR66882-4-4-B 34 2.29 102 100 63.6 21.7 2.30 100 100 66.2 24.4 PSBRc 14 08 74 2.71 83 89 69.9 21.3 IR64 09 2.58 85 89 73.8 24.1 IR36 10 2.61 74 91 71.6 22.4 2.63 81 90 71.7 22.6 CT9897-55-2-M-3-M 12 2.46 86 97 68.4 23.0 IR64615H 40 85 2.39 92 98 59.3 23.5 IR68877H 42 2.63 87 89 62.9 22.7 2.49 88 95 63.5 23.1

Fig. 2. Dendrogram of genotype groups from pattern analysis of mean polished yield data. (Re- printed from Wade et al (1999), Field Crops Research 64:35-50, with permission from the publisher.)

Using reference lines to classify multienvironment trials . . . 135 Criteria considered for selecting reference lines were a high correlation with the group pattern, a high yield level within the genotype group, a good knowledge of characteristics of the line, seed availability, and, for the purpose of this simulation study, an adequate number of locations where the lines were tested. The ability of a reference set to classify environments in the same way as a full set of lines is assessed by comparing ordination (AMMI analysis) and cluster groupings of the standardized reference line responses with those in the full analysis of Wade et al (1999). Two reference sets, one of eight lines and the other a subset of five lines, are considered. Three algorithms for classifying new environments on the basis of reference line responses are compared: the expectation-maximization (EM) procedure implemented in the MATMODEL AMMI analysis package (Gauch 1990); the AMMI procedure in the IRRISTAT G × E analysis module, which assumes zero interaction for missing values (IRRI 1998); and the pattern analysis procedure in GEBEI, which uses a nearest cluster procedure to estimate missing values (De Lacy et al 1996). To compare the efficiency of the three algorithms, we used data for eight or five reference lines over 36 environments. Yields in each of the 36 environments in the full analysis were omitted, one at a time, with just the reference line yields added back. The change in ordination position for the omitted location was used to assess the ability of the method to accurately characterize a new environment on the basis of reference line responses alone. The resulting classification may be used to indicate likely characteristics of a new site and the types of genotypes that should be preferentially adapted there.

Results and discussion Selecting reference sets Based on their pattern analysis of the full data set, Wade et al (1999) suggested a reference set containing representatives of six of the nine genotype groups detected in their analysis. Their choice was based primarily on knowledge of the characteristics of lines from the major groupings: Sabita or KDML105, NSG19, Mahsuri, IR57514- PMI-5-B-1-2 or IR62266-42-6-1, PSBRc14, and CT9897-55-2-M-3-M. Here, we consider the criteria outlined in the materials and methods section as a basis for identifying entries from the genotype groups to serve as reference lines. In group 82 (Fig. 2), Sabita and KDML105 had similar patterns, but Sabita had a higher yield, so, from a practical point of view, it may be a better reference line. For this study, however, we chose KDML105 because it is widely grown in northeast Thailand, a critical set of environments where drought and low soil fertility are common. There was no representative of group 78 in the original set of six indicated by Wade et al (1999). The pattern for group 78 differed from that for 79 only at the singleton environment group IC-Raipur 1995 (Fig. 3A, B). For this study, line 29 (IR66516-37-7-B) was included to represent environment group 78.

136 McLaren and Wade GGP78 GGP79 2.5 2.5 A B 1.4 1.4

0.3 0.3 GTP38 GTP28 GTP37 –0.8 GTP29 –0.8 GTP36 GTP27 GTP5B GTP25 GTP5A –1.9 GTP22 –1.9 GTP05 Zero Zero GGP78 GGP79 –3.0 63 46 62 57 60 61 IC IJ PA –3.0 63 46 62 57 60 61 IC IJ PA GGP84 GGP81 2.5 2.5 C D 1.4 1.4 GTP35 GTP33 0.3 GTP32 0.3 GTP31 GTP62 GTP30 GTP39 –0.8 GTP26 –0.8 GTP11 GTP20 GTP1B GTP16 GTP1A –1.9 GTP15 –1.9 GTP01 Zero Zero 63 46 62 57 60 61 IC IJ PA GGP84 63 46 62 57 60 61 IC IJ PA GGP81 –3.0 –3.0 Environment group Fig. 3. Interaction response patterns for four genotype groups. Genotype and genotype group codes are identified in Figure 2.

Group 84 was also not represented in the original set, so line 31 (IR66879-2-2- B) was selected on the basis of few missing values and good correlation with the group pattern (Fig. 3C). No representative of group 81 was chosen because this group showed little interaction, low yield, and a weak pattern of discrimination between environments (Fig. 3D). In group 76, line 34 (IR66882-4-4-B) was chosen as the reference line for this study in preference to line 23 (IR62266-42-6-1) because the latter was evaluated at only 27 of 36 sites. Line 34 had a reasonable correlation with the group pattern (Table 1) and a high yield in its group (Fig. 2). There was very little difference between PSBRc14 and IR36 in terms of pattern or yield, with PSBRc14 preferred because it is recommended for rainfed lowlands in the Philippines. Since PSBRc14 was not tested at all sites, however, IR36 was used in this study to represent environment group 74. The eight reference lines, whose characteristics are shown in Table 1, were used in subsequent analyses in this chapter. To assess the effect of size of reference set on reliability of characterization, a reduced set was defined by retaining only one representative (IR66879-2-2-B) from groups 78, 79, and 84—the first to fuse in the structure shown in Figure 2, and one representative (IR36) from groups 74 and 85—

Using reference lines to classify multienvironment trials . . . 137 Table 1. Characteristics of the reference lines chosen to represent the genotype groups shown in Figure 2.

Correlation Genotype No. in Reference with group Range of group group line pattern correlation

82 7 04 KDML105 0.81 0.59–0.96 52 3 02 NSG19 0.92 0.92–0.98 78 5 29 IR66516-37-7-B 0.92 0.92–0.97 79 6 05 Mahsuri 0.95 0.56–0.95 84 9 31 IR66879-2-2-B 0.87 0.51–0.93 81 6 None –0.38–0.85 76 5 34 IR66882-4-4-B 0.83 0.70–0.92 74 3 10 IR36 0.98 0.93–0.98 85 3 12 CT9897-55-2-M-3-M 0.93 0.83–0.93 the next to fuse in the structure. This leads to a reduced group of five reference lines: KDML105, NSG19, IR66879-2-2-B, IR66882-4-4-B, and IR36.

Reliability of reference sets To examine whether reference sets can reliably detect and estimate the same interaction as the full analysis, we repeated the pattern analysis reported in Wade et al (1999) using data for eight reference lines in 36 environments, then for only five reference lines in 36 environments. Figure 4 shows the shift in environment ordination as a result of using data on reference lines alone compared with the full data set. With the full reference set (Fig. 4A), movement is quite limited except for a few environments: IC, IJ, and TD. Sites IC and IJ nevertheless remain in similar positions relative to the remainder of the environments, so discrimination of these extreme sites (Wade et al 1999) is still representative of the full data set. Site TD is more of a concern, since it failed to distinguish between the reference lines at all in this analysis, but discriminated strongly in the full data set. This is due to specific interactions with a few test lines, which cannot be captured by the smaller group. The average absolute shift due to using the reference set of eight in place of the full set of 47 genotypes was 9% of the range in IPCA1 scores and 10% of the range in IPCA2 scores. The use of the reduced reference set of five lines had more severe consequences, as can be seen by the relatively larger shifts in Figure 4B, 13% for IPCA1 and 14% for IPCA2. The clustering procedure on the full reference set of eight lines produced eight environment clusters that could be aligned with the original clusters in Figure 1. Table 2 shows this. Of the 36 environments, 23 clustered into the same groups and 5 into neighboring groups, leaving 8 or 22% that were not well characterized. With the reduced set of reference lines, 30% were not well characterized according to the full analysis. Hence, reference lines are able to reproduce the main features of the environmental classification, but representatives of all the major genotype groupings were

138 McLaren and Wade IPCA2 2.0 A B 1.3 IA PB IA FA PA PA PB FA PD 0.6 TD PD TD OA II OA II TA TH PE PE TA TH PC TK TG OB PC TG BB TCBD TK IE –0.1 TE IBBB TE BB TB IG TI OC TI OC ID FC IDTF FC TO TF –0.8 IK

–1.5 IJ IJ IC IC –2.2 –2 –1 0 1 2 –2 –1 0 1 2 IPCA1 Fig. 4. Environment ordination shifts due to using reference data. Letter codes from Figure 1 indicate environment positions from the full analysis, symbols at the end of the spokes in (A) indicate positions based on analysis of data from eight reference lines and in (B) for analysis based on five reference lines.

Table 2. Comparison of environment clusters formed by analysis of data from 8 reference lines over 36 environments and from the full data set from 37 genotypes over 36 environments.

Cluster tree from analysis Cluster number from full analysis (Fig. 2) of reference lines 63 46 62 57 60 21 10 16 61

70 60 FC, TE, TK 62 IA, BC BD II 63 TD BA, FA, PA FB, IH, TH, TA, 71 69 66 TC, TG 64 IK, TTS BB, RE, OA IB OB,TI, 65 TJ IG ID IJ 61 68 IC 19 67 TF OC PE 59 PB, PC, 51 PD

Using reference lines to classify multienvironment trials . . . 139 required as the quality of the reproduction dropped markedly when only a few groups were represented.

Characterizing new environments In order to check how “new” environments would be characterized on the basis of reference line responses, environments were removed from the data set one at a time. For each removed environment, only responses of the eight reference lines were added back to represent a “new” test environment. Pattern analysis on mean polished data was carried out on each reconstructed data set and the shift in environment ordination for the new environment was recorded. This technique will underestimate the error in characterizing truly new environments because the so-called “new” information had been included in the original analysis. It more accurately shows the effect of not having the responses on the test lines. The two concepts should not be too different, however, as the repeated information in each analysis is only eight responses in a matrix of l,692 cells. Three analysis methods were used and two failed to characterize the new environments at all because their analysis relied heavily on the ability of a technique to impute interaction effects for missing data. The more successful technique was to estimate the missing response of line i in environment j as the mean response in environment j of lines in the genotype group with which line i first fuses in the cluster analysis. The resulting complete matrix is then subject to ordination (De Lacy 1996). There is no suggestion that this strategy is optimal, but it is certainly better than the strategy of assuming zero interaction for missing lines as in the G × E module of IRRISTAT (IRRI 1998), or the proportedly optimal EM strategy used in MATMODEL (Gauch and Zobel 1990), both of which lead to a complete collapse in the environmental characterization, with each replaced site shifting near to the zero interaction position. Figure 5 shows the results of the successful strategy. Shifts in the environment ordination are very modest. Mean absolute shifts were 8% of IPCA1 range and 11%

IPCA2 2.0

1.3 IA FA PA PB 0.6 TD OA PD II TA OB PE TH BA IE IB PC TK TG IH BD BB TE FBBC TI –0.1 TC FC TBIG TI OC TF IK ID –0.8 Fig. 5. Ordination shifts when each environment in turn is represented only by data from refer- –1.5 IC IJ ence lines. Letter codes from Figure 1 indicate environment positions from the full analysis, sym- –2.2 bols at the end of the spokes are the positions –2 –1 0 1 2 based on data from eight reference lines only IPCA1 for that environment.

140 McLaren and Wade of IPCA2 range. The extreme, singleton environments IC and IJ showed large shifts, as did OA and TD, indicating that the characterization of these sites depended critically on lines omitted from the reference sets.

Conclusions The results indicate that a small set of reference genotypes is able to capture a significant amount of the G × E information that is available from a much larger set of test lines. The efficiency of the reference set depends, however, on having good representation of all of the sensitive genotype groups. The statistical methodology for this analysis needs to be improved as it relies on imputation of responses from one environment to another and there is no optimal way to do this at present. Reference lines should be selected first on the basis of how well they match the discriminatory pattern of the genotype group and then by desirable agronomic features, state of knowledge about the physiological responses of the line, and finally practical considerations such as seed availability. The reason for wanting well-known lines is that the ultimate goal of characterization is to understand plant adaptation to different subecosystems in the target population of environments and to use this information to identify important traits that confer preferential adaptation to those target environments. From this knowledge, efficient selection strategies may be devised, but their success requires a detailed understanding of physical and climatic properties at all sites where reference sets are grown. Similarly, detailed information on crop management and crop development are essential to make the ultimate link between adaptation and environmental conditions. We propose that small trials of reference sets, widely grown in the rainfed lowlands, have the potential to provide a rapid and useful characterization of new environments. This can be used immediately to weight selection to the desired target environments and to extrapolate the performance of test lines over wider geographical areas. Integration of this classification with physical and climatic data would provide the link with a geographical characterization. The results can also be used to develop an understanding of the physiological basis of plant adaptation, and, through a simulation analysis using historical climatic data, to complete the characterization in terms of frequencies of occurrence and risks associated with different target environments.

References Aggarwal PK, Kropff MJ, Mathews RB, McLaren CG. 1996. Simulation models to design new plant types and to analyze G × E interactions in rice. In: Cooper M, Hammer G, editors. Plant adaptation and crop improvement. Oxford (UK): CABI. p 403-418. Cooper M, Fox PN. 1996. Environmental characterization based on probe and reference genotypes. In: Cooper M, Hammer G, editors. Plant adaptation and crop improvement. Ox- ford (UK): CABI. p 529-547.

Using reference lines to classify multienvironment trials . . . 141 Cooper M, Rajatasereekul S, Immark S, Fukai S, Basnayake J. 1999. Rainfed lowland rice breeding strategies for Northeast Thailand. 1. Genotypic variation and genotype × environment interactions for grain yield. Field Crops Res. 64:131-151. Cooper M. 1999. Concepts and strategies for plant adaptation research in rainfed lowland rice. Field Crops Res. 64:13-34. De Lacy IH, Basford KE, Cooper M, Bull JK, McLaren CG. 1996. Analysis of multi-environment trials: an historical perspective. In: Cooper M, Hammer G, editors. Plant adaptation and crop improvement. Oxford (UK): CABI. p 39-124. Gauch HG Jr. 1990. MATMODEL Version 2.0: AMMI and related analyses for two-way data matrices. Ithaca, N.Y. (USA): Cornell University. Gauch HG, Zobel RW. 1990. Imputing missing yield trial data. Theor. Appl. Genet. 79:753- 761. IRRI (International Rice Research Institute). 1998. IRRISTAT for Windows tutorial manual. Biometrics Unit, The International Rice Research Institute, Los Baños, Laguna, Philip- pines. McLaren CG. 1996. Methods of data standardization used in pattern analysis and AMMI models for the analysis of international multi-environment variety trials. In: Cooper M, Ham- mer G, editors. Plant adaptation and crop improvement. Oxford (UK): CABI. p 225- 242. Wade LJ, Sarkarung S, McLaren CG, Guhey A, Quader B, Boonwite C, Amarante ST, Sarawgi AK, Haque A, Harnpichitvitaya D, Pamplona A, Bhamri MC. 1995. Genotype by environment interaction and selection methods for identifying improved rainfed lowland rice genotypes. Proceedings of the International Rice Research Conference, 13-17 Feb- ruary 1995, International Rice Research Institute, Los Baños, Laguna, Philippines. p 885-900. Wade LJ, McLaren CG, Samson BK, Regmi KR, Sarkarung S. 1996. The importance of environment characterization for understanding G × E interactions. In: Cooper M, Hammer G, editors. Plant adaptation and crop improvement. Oxford (UK): CABI. p 549-562. Wade LJ, McLaren CG, Quintana L, Harnpichitvitaya D, Rajatasereekul S, Sarawgi AK, Kumar A, Ahmed HU, Sarwoto, Singh AK, Rodriguez R, Siopongco J, Sarkarung S. 1999. Genotype by environment interaction across diverse rainfed lowland rice environments. Field Crops Res. 64:35-50.

Notes Authors’ address: International Rice Research Institute, DAPO Box 7777, Metro Manila, Philip– pines. Acknowledgments: This chapter used data published by Wade et al (1999), which indicated the contributions from scientists in the Rainfed Lowland Rice Research Consortium. These experiments received support from the Asian Development Bank (ADB), Directorate General for International Cooperation (DGIS), and Department for International Devel- opment (DFID). Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

142 McLaren and Wade Biophysical characterization and mapping

Biophysical characterization of rainfed systems . . . 143 144 Amien and Las Biophysical characterization of rainfed systems in Java and South Sulawesi and implications for research

I. Amien and I. Las

Limited suitable land for rice production in the past has forced the development of terraced and bunded rice fields at a higher elevation. Because the rice-cropping system lacks terrestrial water resources, it totally depends on available rainfall. Java, North Sumatra, and South Sulawesi have the largest areas of rainfed lowland rice fields, 796,900, 210,300, and 259,100 ha, respectively. Because of high competition for space, rainfed rice area is steadily declining on Java but is relatively stable outside Java. A high proportion of the rainfed rice area is in the northern part of Java and eastern part of South Sulawesi. The soils of rainfed areas in West Java were formed from acid tuft. In the other provinces of Java and South Sulawesi, the soils were formed from limestone or marine sediments. Less fertile soil that is low in P and K and the low adoption of modern technology in rainfed areas resulted in lower rice yields. Yields in rainfed areas were 10% to 25% less than the average yield in Java and 15% to 20% less than the average yield in South Sulawesi. Yield levels of lowland rice on Java are inversely related to the proportion of rainfed lowland rice areas. With monsoonal rainfall patterns, the rainy season begins in October in West and Central Java, in November in East Java, and in March in South Sulawesi. West Java and South Sulawesi have no distinct dry period, while in Central and East Java the dry period varies from 4 to 5 mo. During El Niño years, the rainy season comes about a month later, with rainfall less than 70% of that of normal years. Dependence on rainfall also makes rainfed rice more susceptible to drought, particularly during El Niño years. This rainfed rice system helps reduce water runoff during the rainy season, but differs with deep-rooted vegetation, because the water retained is mostly lost through evapotranspiration. In the increasingly globalized economy, labor shortages and more competitive markets threaten the sustainability of the rainfed rice system.

Biophysical characterization of rainfed systems . . . 145 As Indonesia’s staple food, rice is strategically important to the agricultural development and economy of the country. The crop is cultivated in diverse environments. Irrigated rice occupies the largest area, followed by rainfed lowlands, uplands, and tidal swamps, with 58%, 20%, 11%, and 11%, respectively. Because of less favorable environments, however, the yield and production of nonirrigated rice are far lower than those of irrigated rice. The increasing demand for rice, along with the increasing population and lack of technological development in rice culture before the Green Revolution, has led to expanded rice cultivation. With limited suitable land for ideal rice culture, new rice fields turned to less favorable environments. Rice fields were developed on sloping land at a higher elevation, often without available water resources except rainfall. Terraced and bunded rainfed lowland rice helps reduce water runoff and prevents flooding in the rainy season. But the system cannot conserve water for the dry season because most is lost through evapotranspiration. When expanded further to even higher elevation and steeper slopes, the rainfed rice system disrupts the hydrological cycle in the watershed and becomes prone to drought in the dry season. In the increasingly globalized economy with rapid progress in telecommunica- tions, transportation, and tourism, the low productivity of rainfed lowland rice, particularly in Java, means less competitiveness. We therefore studied the performance of rainfed lowland rice as affected by the physical characteristics of the environments in Java and South Sulawesi.

Rainfed lowland rice areas and their distribution in Indonesia Figure 1 shows the distribution of rainfed rice area in Indonesia. Rainfed lowland area in Indonesia declined from 2.2 million ha in 1988 to 2.1 million ha in 1995 (BPS 1988, 1995). There has been a slight increase in this area in Kalimantan and Sulawesi, but it is decreasing steadily in Java. The decreasing trend is also observed in Sumatra and Nusa Tenggara. The rate of conversion of rainfed lowland area to other uses is highest in Java, with 13,800 ha annually, followed by Sumatra and Nusa Tenggara, with 6,100 and 2,700 ha per annum, respectively. The largest area of rainfed lowland rice in the Outer Islands is in the province of South Sulawesi, with 259,100 ha or 41.1% of the total rice area (Table 1). High proportions of rainfed lowland rice area in South Sulawesi are found in three districts in the east, Sinjai, Bone, and Wajo, with 49.6%, 52.7%, and 79.6%, respectively, of the total rice area. Although not quite as high in terms of the proportion to the total rice area, rainfed lowland rice area in the three big provinces of Java is among the largest in the country, with a total of 786,800 ha. Central Java has the largest area, with 293,600 ha or 29.4% of the total rice area, followed by East and West Java, with 252,400 and 240,800 ha or 22.0% and 20.9%, respectively, of the total rice area. In Java, a large proportion of the rainfed lowland rice area is in the gently sloping northern coast. The districts of Lebak and Pandeglang in the northwest part of West Java have 43.1% and 45.9% of the total rice area. Grobogan, Blora, Pati, and

146 Amien and Las Rainfed rice area (ha)

Fig. 1. Distribution of rainfed rice in Indonesia.

Table 1. Rainfed rice area in Indonesia.

Rainfed rice area (ha) Proportion Yieldc Region/province Single crop Double crop Drf/totala Rf/totalb (t ha–1) (%)

Sumatra 422,458 153,731 26.7 23.9 4.1 Java 618,846 178,022 22.3 23.7 5.4 West Java 178,305 62,490 26.0 20.9 5.3 Central Java 203,474 90,148 30.7 29.7 5.3 East Java 229,783 22,623 9.0 22.0 5.5 Kalimantan 323,627 52,833 14.0 27.4 2.9 Nusa Tenggara 47,872 6,391 11.8 13.8 4.5 Sulawesi 258,785 39,118 13.1 31.6 4.4 South Sulawesi 225,902 33,222 12.8 41.1 4.8

aRatio of rainfed double crop area to total rainfed area. bRatio of total rainfed area to total rice area. cMean yield of all rice crops.

Rembang districts in the northeast of Central Java have 57.6%, 73.7%, 39.4%, and 60.6% of the total rice area, respectively, in rainfed lowlands. All eight districts in the northern part of East Java, from Bojonegoro and Tuban in the west to Sumenep in the east, have a high proportion of rainfed lowland rice area, which ranges from 41.3% in Lamongan to 82.9% in Gersik.

Biophysical characterization of rainfed systems . . . 147 Soil and climate of rainfed lowland rice areas in Java and South Sulawesi Soils The soils of the rainfed areas in Lebak and Pandeglang districts of West Java were formed from acid tuft. The soils developed from acid tufts in the region are classified as Inceptisols of the Dystropept subgroup or Ultisols of the Tropudults great group. They have low fertility although they have good physical properties (LPT 1967). The soils with a high clay content are able to retain the amount of water required by plants. This is particularly important when the impermeable plow pan in rainfed rice areas is not developed or is destroyed by soil tillage. These soils are low in nutrients such as P and K as well as Ca and Mg. Under high iron and aluminum, the limited soil P is mostly retained by oxides. Although puddled soil tends to release retained soil P, the low P potential in the soil means that high P fertilization is required to sustain high yield. In the other provinces of Java and South Sulawesi, the soils were formed from limestone or marine sediments. Soils developed from these materials are classified as Inceptisol of the Ustropepts great group, Alfisols of the Tropustalfs great group, and Vertisols of the Chromustert great group (Dames 1955, LPT 1975). The smectitic clay minerals in the soils often make them difficult to till, particularly when water is inadequate early in the rainy season. The soils contain a relatively sufficient amount of Ca and Mg, but are low in P. The high bases often make K less available to plants. To attain higher yield, a higher rate of fertilization of P and K is required.

Climate The climate in the rainfed areas, because they are in the lowland tropics, is always warm throughout the year. Differences in temperature are higher between night and day than between seasons. With ample solar radiation, the limiting factor for agricultural production is water availability. In these rainfed areas, the water supply totally depends on rainfall. Rainfall in Java and South Sulawesi is of the monsoon type, with distinct wet and dry seasons (Fig. 2). In a normal year, the rainy season with monthly rainfall of 200 mm or more starts in September in the rainfed areas of West Java and in Novem- ber in northern East Java. During El Niño years from 1955 to 1994, the annual rainfall dropped about 30% with delayed rainy seasons. The rainy season in the eastern part of South Sulawesi starts in March during normal years, but is delayed until May during El Niño. Table 2 presents mean annual rainfall during normal and El Niño years in the regions with data from representative stations. From the number of wet months with mean rainfall of 200 mm or more only in Lebak and Pandeglang, West Java, the water supply is adequate for more than two rice crops in one year, whereas in the other areas wet-season duration is only adequate for one rice crop. As indicated in Table 2, there is a higher proportion of double crops in the rainfed areas of Sumatra and West and Central Java, which have more wet months. During El Niño, water is available for only one crop. In Madura (East Java) and South Sulawesi, rice growing is particularly

148 Amien and Las Rainfall (mm) 600 400 Rangkasbitung, West Java Rembang, Central Java 500 Enso 300 Normal 400 Crop 1 Crop 2 300 200 Crop 3

200 100 100

0 0

300 400 Gersik, East Java Bone, South Sulawesi

300 200

200

100 100

0 0 JFMAMJJA SOND JFMAMJJASOND Month Fig. 2. Monthly rainfall distribution in rainfed lowland rice areas.

Table 2. Annual rainfall, wet and dry months, and month when rainy season begins in rainfed areas of Java and South Sulawesi.

Annual rainfall Wet Dry Month when rainy Area (station) months months season begins Normal Ensoa (mm) NE NE N E

West Java Rangkas 3,705 1,584 9503 September November Central Java Rembang 2,009 1,530 5447 October December East Java Gersik 1,621 1,103 4458 November December Pamekasan 1,811 1,293 5346 November December South Sulawesi Watampone 2,428 1,584 5304 March May aEnso = El Niño southern oscillation, N = normal years, E = years with El Niño.

Biophysical characterization of rainfed systems . . . 149 risky when the crop is planted late. Farmers are usually bound by tradition, and they plant their rice crops as the rainy season begins in a normal year.

Effects of El Niño on rice harvest area Because the water supply totally depends on rainfall, the climate has a very strong effect on the rice harvest in rainfed lowland areas. Rainfed rice in West Java is usually planted at the beginning of the rainy season from September to November and is harvested from January to April. The harvested rice area of West Java (Fig. 3) from 1989 to 1996 indicates that rainfed areas in Lebak and Pandeglang districts suffered most during El Niño of 1991 and 1994. The harvested rice area of the first crop decreased markedly from the normal year of 1990 to El Niño of 1991. In El Niño of 1994, although the decrease in harvested area was not as high as in 1991, the drought also affected the harvested area in 1995. Although the harvested area of the first crop decreased again in 1995, the total harvested area in 1995 increased significantly.

Harvested area (× 1,000 ha) 160 3,000 Lebak and Pandeglang West Java 140 2,000 120 Total 100 First crop 1,000 80

60 0 Harvested area (× 1,000 ha) 300 1,600

1,400 250 Grobogan, Blora, Pati, Central Java and Rembang 1,200 200 1,000 150 800

100 600 89 90 91 92 93 94 95 96 89 90 91 92 93 94 95 96 Year (19..) Fig. 3. Effect of climate on harvested rice areas in West and Central Java.

150 Amien and Las In Central Java, the effects of El Niño of 1994 were more marked than those of El Niño of 1991. In Grobogan, Blora, Pati, and Rembang areas of Central Java (Fig. 3), although the harvested area from 1990 to 1991 for the first crop showed a slight increase, the total harvest that year decreased by almost 5,000 ha. But the decrease in harvested area in 1994 was almost 4 times higher for the first crop compared with the total harvest that year. In the rainfed lowland rice areas of Central Java, further decreases in harvested area occurred after El Niño, as indicated by the reduction in area in 1992 and 1995 for the first crop harvest. The poor harvest of El Niño affected the harvest of the coming year probably because of the delayed planting time and unavailability of seed for planting. In the eight districts of the north and in all areas of East Java, harvested area decreased significantly during El Niño of 1991 and 1994 (Fig. 4). However, the decrease was more pronounced in the eight districts of the north, which have a higher proportion of rainfed lowland area. The reduction in harvested area of the first crop from 1990 to 1991 in the eight districts accounted for 72.5% of the total reduction in 30 districts of East Java. Although it was not as high as in 1991, there was a significant reduction in harvested area in El Niño of 1994, with a further decrease in 1995.

Harvested area (× 1,000 ha) 500 1,600 Total First crop 1,400 East Java 400 1,200 Eight districts of northern East Java 1,000 300 800

200 600

Harvested area (× 1,000 ha) 300 900 Sinjai and Wajo SouthCentral Sulawesi Java 800 700 200 600 500 100 400 300 0 200 89 90 91 92 93 94 95 96 89 90 91 92 93 94 95 96 Year (19..)

Fig. 4. Effect of climate on harvested rice area in East Java and South Sulawesi.

Biophysical characterization of rainfed systems . . . 151 The decrease in harvested area only occurred in the first crop where rainfed lowland rice was planted. The harvested area in the second and third crop increased slightly. South Sulawesi has a different rainfall pattern (Fig. 2). Rainfed lowland rice is commonly planted in March or April and harvested in August to September. Compar- ing the harvested area of the second crop harvested in August to December in Sinjai and Wajo districts with the total harvest of the year showed that most of the crop failure occurred in the second crop. This indicates that rainfed lowland areas suffered most from drought caused by El Niño. The decrease in harvested area of the second crop in the three eastern districts of South Sulawesi from 1990 to 1991 was more than 70% of the total decrease that year (Fig. 4). A decrease in harvested area also took place in 1993. The data from the four provinces showed significant adverse effects of climate on rainfed lowland rice production. Because of the total dependence on rainfall for water supply, a reliable prediction of climate will help farmers plan better planting times. A selection of early cultivars for El Niño years will also reduce the risk of crop failure when the rainy season is shortened by one or more months.

Rice yield The rice yield in districts with a high proportion of rainfed lowland rice area is always lower than the mean yield of the province or region. In Java, the rice yield of the predominantly rainfed districts of Lebak, Rembang, Pati, Tuban, Bangkalan, Sampang, Pamekasan, and Sumenep from 1990 to 1996 was about 83% to 87% of the mean rice yield in Java (BPS 1990, 1991, 1992, 1993, 1994, 1995, 1996). The lowest yield was in Lebak District, with a 7-year mean yield of 4.2 t ha–1 y–1 versus 5.5 t ha–1 in Java, probably because of poor soil conditions. Although rice tolerates acid soil and in puddled soils of paddy rice culture the pH of the soil moves toward neutrality, the high iron and aluminum in the acid soil retain soil P and the P becomes less available to plants. In South Sulawesi, the rice yield from 1990 to 1995 in Sinjai and Wajo districts was about 85% of that in all districts of South Sulawesi. With more adoption of modern rice technology, the mean yield in the province steadily increased over time (Fig. 5). But the mean yield in the districts with a high proportion of rainfed lowland rice, such as Sinjai and Wajo, fluctuated with interannual climate variability. The mean yield during El Niño 1994 was only 82.9% in Sinjai and 83.3% in Wajo of the mean yield in South Sulawesi. When rice yield is plotted against the proportion of rainfed lowland rice area, there is a strong tendency indicating that the higher the proportion, the lower the yield. The 1996 rice yield of about 80 districts in Java plotted against the proportion of rainfed lowland rice area indicated a negative relationship with a linear equation of (Fig. 6)

152 Amien and Las Yield (Qt ha–1) Yield (Qt ha–1) 45 65 Y95 Y92 Yield = 55.963 – 0.137* P Y94 Y91 R = 0.66** 43 60 Y93 Y90

41 55

39 50

37 45

35 Sinjai Wajo South 40 Sulawesi 0 1020304050607080 District Rainfed proportion (%) Fig. 5. Rice yield in South Sulawesi. Fig. 6. Relation between proportion of rainfed lowland rice area to total rice area and rice yield.

Yield (Qt ha–1) = 55.963 – 0.137* proportion (in % of total rice area)

Labor requirement and technology adoption A study in Lampung, Java, and South Kalimantan by Kasryno and Sudaryanto (1994) reported that rice farming in rainfed lowlands required 175 man-days from land preparation to harvest compared with only 157 man-days in irrigated areas and 71 man- days in tidal swamp areas. Compared with irrigated rice, rainfed rice requires more labor for land preparation, seeding, transplanting, and crop care but less for harvest. Land preparation, seeding, and transplanting probably require more labor because of the difficult terrain and harder-to-till soil. Land preparation is particularly difficult in swelling-clay soil types such as Vertisols that require adequate water for land preparation. Being somewhat remote or less accessible because of the distance from economic centers, rainfed lowland rice areas took longer for the adoption of modern rice technology. In 1970, only 8% of the rainfed rice area used modern rice varieties compared with 53% of the irrigated rice area (Kasryno and Sudaryanto 1994). The proportion of modern rice varieties increased to 42% in 1980 and 81% in 1987, whereas in the irrigated rice system it was 91% in 1980 and 98% in 1987. Relatively poor soil conditions like those in rainfed lowland rice areas will require more fertilizer to attain a higher yield. Fertilizer application in rainfed lowland rice in 1987 as reported by Kasryno and Sudaryanto (1994) was only 107 kg ha–1 of a combination of urea, triple superphosphate, and KCl fertilizers, versus 144 kg ha–1 in irrigated rice. Much more fertilizer was applied in rainfed lowland rice

Biophysical characterization of rainfed systems . . . 153 than in tidal swamp rice, for which only 56 kg ha–1 was applied. The slower adoption of modern rice technology coupled with less favorable environments in rainfed lowland rice areas have resulted in a lower rice yield.

Research priority and policy implications Although the trend is declining, rainfed lowland areas in Indonesia, with more than 2 million ha, play an important role in producing staple food and providing employment opportunities for the population. These rainfed lowland areas are commonly situated at a higher elevation in the watershed, but are also scattered in small patches wherever human settlements exist. To better understand the nature and characteristics of rainfed lowland rice, a systematic program to delineate its area is required. The delineation of rice area in Java using remote-sensing technology needs to be further expanded to cover other areas in the country. The lower yield of rice in rainfed lowlands occurs mainly because of the slower adoption of modern rice technology. Less favorable soil conditions can be overcome to some extent by applying the appropriate type and rate of fertilizers. Research on fertilization for rainfed lowland rice and subsequent crops in an appropriate cropping pattern for the region needs to be pursued. In the more humid areas, such as West Java, in a normal year, rainfall is adequate for two rice crops, which can be followed by a less water-demanding secondary crop. In the subhumid areas of Central and East Java, rainfall is adequate at least for one rice crop. The susceptibility of rainfed lowland rice to drought caused by interannual climate variability has affected it more than other rice cultivation systems. With a better prediction of climate, better planning for time of planting and selection of rice cultivars can be achieved to avoid crop failure. Research to improve the accuracy of seasonal climate forecasting is currently being done by national and international agricultural research institutions. This cooperation needs to be further promoted for a better understanding of the dynamics of tropical climate.

References BPS (Central Bureau of Statistics). 1988. Luas dan Penggunaan Lahan di Indonesia. Jakarta (Indonesia): BPS. BPS (Central Bureau of Statistics). 1990. Agricultural survey: production of paddy in Indone- sia. Jakarta (Indonesia): BPS. BPS (Central Bureau of Statistics). 1991. Agricultural survey: production of paddy in Indone- sia. Jakarta (Indonesia): BPS. BPS (Central Bureau of Statistics). 1992. Agricultural survey: production of paddy in Indone- sia. Jakarta (Indonesia): BPS. BPS (Central Bureau of Statistics). 1993. Agricultural survey: production of paddy in Indone- sia. Jakarta (Indonesia): BPS. BPS (Central Bureau of Statistics). 1994. Agricultural survey: production of paddy in Indone- sia. Jakarta (Indonesia): BPS.

154 Amien and Las BPS (Central Bureau of Statistics). 1995. Agricultural survey: production of paddy in Indone- sia. Jakarta (Indonesia): BPS. BPS (Central Bureau of Statistics). 1996. Agricultural survey: production of paddy in Indone- sia. Jakarta (Indonesia): BPS. Dames TWG. 1955. The soils of East Central Java. Contributions of the General Agricultural Research Station, Bogor, Indonesia. LPT (Lembaga Penelitian Tanah). 1967. Reconnaisance soil map of Java. Bogor (Indonesia): LPT. LPT (Lembaga Penelitian Tanah). 1975. Reconnaisance soil map of South Sulawesi. Bogor (Indonesia): LPT. Kasryno F, Sudaryanto T. 1994. Modern rice variety adoption and factor-market adjustments in Indonesia. In: David CC, Otsuka K, editors. Modern rice technology and income distribution in Asia. Boulder, Colo. (USA): Lynne Reinner Publishers, Inc., and Manila (Phil- ippines): International Rice Research Institute. p 107-127.

Notes Authors’ address: Center for Soil and Agroclimate Research, Bogor, Indonesia. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Biophysical characterization of rainfed systems . . . 155 Monitoring rainfed and irrigated rice in Southeast Asia using radar remote sensing

R. Verhoeven, H. van Leeuwen, and E. van Valkengoed

Rice is the main food crop in the Asia-Pacific region and reliable spatial information such as area, yield, crop type, and secondary crops is therefore essential for management of the rice-growing areas and food policy. Space-borne remote sensing provides a synoptic view over extensive areas on a regular basis and has proven to be useful as an independent data source for agricultural statistics. Radar remote sensing can be useful for mapping and monitoring cloud-covered areas and can be used to identify and monitor rice fields. This is demonstrated on the basis of some current and past projects on rice in Southeast Asia. Using an integrated spatial approach, combining information from remote sensing and other spatial data sources in a geographic information system, rice crops can be identified in both the dry and wet season and useful information such as area, number of harvests, and eventually yield can be estimated.

In the European context, European Union-Directorate General I (EU-DG-I) and EU- DG-VI have begun two major rice programs to support food security monitoring in relation to the rice-cropping systems in Asia using a weather-independent and remote sensing-based approach. In the philosophy of the European MARS program (“Moni- toring Agriculture with Remote Sensing”) to produce crop statistics on the national level, the Southeast Asian Rice Radar Investigation (SEARRI) and Satellite Assess- ment of Rice in Indonesia (SARI) programs have been defined for rice mapping and production monitoring using radar remote sensing and national rice statistics from the International Rice Research Institute (IRRI) and the national government of Indo- nesia, respectively. The European Space Agency (ESA) has begun several data user program projects (DUP) using the latest developments in radar remote sensing in relation to rice-wheat-water in Sri Lanka, Europe, and Bangladesh. Synoptics Remote Sensing and GIS Applications BV is involved in some major rice investigation, monitoring, and yield-forecasting projects. The current projects in Southeast Asia where Synoptics plays the role of information facilitator in rice map-

Monitoring rainfed and irrigated rice in Southeast Asia . . . 157 ping and monitoring using spatial techniques (remote sensing and geographic information system, GIS) are the following: 1. SEARRI project (1998-99): The Southeast Asian Rice Radar Investigation project is a demonstration project within the Centre of Earth Observation (CEO) program AS3200 package (CEO, Aschbacher 1995, Division of the JRC-EU, Italy). The SEARRI technical objectives were consolidated in response to the requirements expressed by several Directorate Generals within the European Commission, in particular DG-VI, DG-IX, and DG-I-B. The project outcome is meant to support the DGs’ requirements. The main scope of the project is to map at a regional scale rice cultivated areas in Southeast Asia using space-borne synthetic aperture radar (SAR) and GIS technology. The SEARRI thematic products are intended to support two main objectives: crop production and methane emission estimations. The first theme is related to important socioeconomic issues, rice being the dominant source of food in Southeast Asia; the second is linked to climate and global change issues, methane being one of the important greenhouse gases. 2. SARI project (1998-2001): The Satellite Assessment of Rice in Indonesia project is defined by the Indonesian government and European Union-DG- I-B and the Joint Research Centre (JRC) in Ispra, Italy, to improve the ability to accurately and timely predict rice yields in Indonesia. Furthermore, the SARI project is defined to meet the wish of the Indonesian government to develop an independent and efficient rice production-forecasting system similar to what is being implemented in Europe for a range of crops in the MARS project. The set-up of the project is to adapt MARS techniques to the Indo- nesian situation and conditions. Earth observation satellites using active microwave sensors (radar), such as the current European ERS and future ENVISAT, are unhampered by the presence of clouds. Information from these types of satellites would permit regular observation of the rice-growing cycle and in principle could form the backbone of an independent monitoring system adapted to the Indonesian context. 3. UPRICE project (1998-99): This project is funded by the Dutch National Remote Sensing Programme (NRSP-2) of the Netherlands Remote Sensing Board (BCRS) and aims to provide and upscale rice location/flooding status information from satellite radar images of local fields to the regional scale of Luzon Province in the Philippines. It has been demonstrated in various studies (Kurosu et al 1995, Le Toan et al 1997) that satellite synthetic aperture radar data from the ERS and JERS instruments can be successfully used to monitor rice growth. Two sensors on board the ERS satellite have been used in this project: the SAR sensor and the windscatterometer. The SAR images were used for detailed mapping of rice crops, whereas the windscatterometer gave a coarse representation of the area at a high revisit- ing frequency. Once adequately validated with detailed SAR images, windscatterometer data may provide a handle on the dynamics of flooding

158 Verhoeven et al and, indirectly, methane emission from rice-wetlands at subcontinental scales (Denier van der Gon et al 1999, van Leeuwen 1998).

Classification of rice using radar image time series SAR remote sensing offers favorable capabilities for rice mapping and monitoring. The typical radar response to rice fields at different growing stages is the key factor for discriminating rice from other land-use classes. Le Toan et al (1997) studied the radar response to rice fields at different growing stages. They distinguished three main periods in the crop cycle: (1) sowing-transplanting, (2) growing, and (3) the after-harvest period. They found a bridge-shaped backscatter pattern representing the radar signature of rice during the growing period (Fig. 1). In ERS, the SAR time series classification of rice can take place on the basis of this typical signature. Le Toan demonstrated that detection of changes of 3 dB or more over the growing season indicates rice areas. The methodology of Le Toan works fine for extensive rice areas in flat terrain in the same growing stage. For smaller rice fields, however, this methodology does not work, especially not in mountainous terrain, where the dynamic range of backscatter is somewhat compressed (Verhoeven and van Leeuwen 1999). Also, the presence of small rice fields in different growing stages is difficult to detect using this method. For classification of rice using radar time series, standard methodologies such as maximum likelihood of iso-data clustering can yield unsatisfying results. A major disadvantage of these techniques is that they ignore the typical time sequence of the

Radar backscattering Surface Volume contribution Interaction contribution

Sowing

Tillering

Stem extension Sowing Tillering Flowering Stem extension

Vegetative stage Reproductive stage

Fig. 1. Radar backscatter response over a rice-growing cycle. Source: CEO-EWSE.

Monitoring rainfed and irrigated rice in Southeast Asia . . . 159 radar signature of rice as well as the spatial coherence of neighboring pixels. Classi- fication can be performed more efficiently if nonrice classes such as urban settlements, layover, water, and other crops are masked out beforehand. A methodology applied by Le Toan et al (1997) uses multitemporal changes of 3 dB or more as a rice mask. As this methodology is very sensitive to speckle, a multitemporal speckle reduction filter should preferably be applied to preserve image geometry. The ERS SAR sensor has a spatial resolution of 25 m and acquisitions are separated by 35 days in time, so three to four consecutive ERS scenes cover a typical rice crop cycle. Many rice paddies are small and of the same order of magnitude as the sensor resolution; they are therefore hard to detect. Another effect called “speckle” is typical for SAR imaging and hampers classification on a pixel level. It is clear that, for a proper classification of rice, additional information is necessary. Knowledge of transplanting/sowing date, type of cultivar, land use, soil type, and water supply (irrigated, rainfed) is therefore essential. Here, a GIS could serve the purpose. Validation of the classification results can take place using existing agricultural statistics and field survey data. Though accurate determination of rice on a field level is difficult or even impossible to achieve with most space-borne remote-sensing images, reliable statistics at larger geometric scales (e.g., on a district or provincial level) can be derived. Low-resolution data with a high repetition frequency can cover gaps in time. In spite of a lack of geometric detail, the overall rice signature in homogeneous areas can still be seen at a large scale. Important for classification of rice paddies is the moment of transplanting or sowing. Because of the time gap of 35 d for the ERS SAR sensor, this moment has not always been monitored. Low-resolution data can be of great help, especially here, to improve classification at lower scales. When studying the cropping cycle of rice in, for example, the Philippines (UPRICE project, Denier van der Gon 1996, Denier van der Gon et al 1999), there is for each growing cycle a period of about 4 to 6 wk of flooding and after that the rice crop matures into a full coverage within several weeks. These dynamics can be followed by the radar and can distinguish the rice crop from other crops having different crop management (see Fig. 2). Even low-resolution radar data (up to 50 km!) can result in these signatures on a regional scale when the rice crop (and related hydrological features) is dominantly present in the region. In the projects mentioned, an iterative integrated process has been set up to improve the classification of rice. The basis for a proper classification is a co-registered and speckle-filtered set of ERS SAR images. An optimal image set can be selected on the basis of cropping calendars. Possible rice areas are selected on the basis of their dynamic behavior in time. Stable objects, such as forests, will be excluded in this phase. Vector maps can also be used to mask out nonrice areas (e.g., open water, urban areas). Supervised or unsupervised classification of the remaining areas is now more efficient and the presence of probable mixed classes will be reduced. After classification, analysis of the radar signature as a function of time is used to merge classes that have a similar signature and are geometrically linked to each other. An example of a radar backscatter signature of rice can be seen in Figure 3. The latter can

160 Verhoeven et al ERS windscatterometer dynamics 1995 Radar backscatter (dB) –7

–8

–9

–10

–11 Upscale moments: Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Dry-season rice crop Wet-season rice crop

Fig. 2. The cropping calendar of rice in Luzon Province of the Philip- pines and the rice-specific radar signature in time of low-resolution radar satellite data. The so-called “upscale moments” indicate an almost simultaneous acquisition of ERS SAR and windscatterometer data.

2-cycle rice flooded in May and September dB 0

–5

–10

–15

–20 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Month Fig. 3. Radar backscatter signature for 2-cycle rice in the Mekong Delta, Vietnam. Horizontal axis represents acquisition time (1996), vertical axis is radar backscatter in dB. The vertical bars represent one standard deviation.

Monitoring rainfed and irrigated rice in Southeast Asia . . . 161 be done using automatic or manual clustering techniques, such as grouping of radio- metric and geometrically related areas in the image series. Existing vector maps from a GIS can also be used for grouping similar areas. Ancillary spatial information such as rice maps, rice statistical data, land-use maps, crop calendars, and field data can be of great use for improving and refining the classification. Knowledge about transplanting or sowing date is very important for discrimination of rainfed crops from other rice crops. The SEARRI project demonstrates the integration of rice information derived from radar observations, statistics, expert knowledge, and land-use information in a GIS environment. End-users could have access to the database yearly. The areas covered by radar data combined with the occurrence of rice according to the Hukes’ database (Huke and Huke 1997, IRRI 1995) can be seen in Figure 4. Figure 5 shows the workflow followed in the SEARRI project, where time series of radar (ERS) images are combined with regional statistics, expert knowledge, crop calendars, and meteorological data in a GIS environment. The radar images were all acquired between March 1996 and December 1997. The thematic rice maps can be validated against statistics (Hukes’ database) and land-cover maps in the tailor-made SEARRI GIS application. In total, 261 ERS scenes divided over 52 locations (3 to 9 images per location) were processed. The ERS time series covers almost all major rice-growing areas in Southeast Asia, including Thailand, Cambodia, Vietnam, and Malaysia. In the UPRICE project, crop calendars and rice statistics on the municipality level were available. The radar data consist of a time series of five ERS images of 1995. The time series of ERS images was classified using a supervised classification

Fig. 4. Location of ERS scenes selected for SEARRI in combination with provincial rice crop statistics from the Hukes’ database.

162 Verhoeven et al Hukes’ Administrative ERS SAR image statistics geodataset series 2. RS preprocessing ERS-frames

1. GIS Texture and combination/ speckle-filtered selection images

Cropping calendars Geodatasets on Meteorological data Hukes’ statistics Rice maps 3. RS postprocessing Seasonal land-cover regions Final classified grid Expert 4. GIS knowledge analysis/ presentation

Regional Thematic maps Validation statistics/maps (rice versus nonrice, comments single versus double crop, planting dates)

Fig. 5. SEARRI image-processing workflow using remote sensing and GIS data to identify rice fields. methodology. Areas within the municipalities with near-100% rice cultivation were used as training data. The SARI project (still running at the time of the writing of this chapter) aims at assessing rice yields on a regular basis for all of Indonesia, similar to the European MARS program. Because most paddy fields are smaller than the agricultural areas monitored in the MARS program, an integrated remote-sensing and GIS environment similar to the SEARRI approach will be necessary. To validate biomass and yields, an intensive field campaign runs parallel to the satellite acquisitions. ERS SAR time series, acquired every 35 days, are used to classify rice paddies and the number of crops and to estimate area and biomass. Classification of rice areas is knowledge-driven, using GIS data based on the methodology of SEARRI. Crop growth models, such as ORYZA, use data from different spatial sources (e.g., fertilizer information), biomass estimates, and crop cycle information from the remote- sensing data. Low-resolution imagery such as NOAA/AVHRR (1-km resolution), SPOT VÉGÉTATION (1 km), RADARSAT ScanSAR (100 m), and, in the near future, the ENVISAT ScanSAR (150 m) is being used as well. These tools can deliver useful information on a more regular basis than ERS SAR does and can be used to control the high-resolution classification process as well. An example of the usefulness of low-resolution remote-sensing data is the rice yield–forecasting system SHIERARY, developed by SEAMEO BIOTROP in Indonesia, which provides rice yields on a

Monitoring rainfed and irrigated rice in Southeast Asia . . . 163 Table 1. Example of form used to collect ground information for the SARI project.

LAND PREPARATION + WIND NO GROWTH STAGE R

RAIN NO HOMOGENEITY OF FIELD 3 PESTS/DISEASES NO PLANT ROW DIRECTION NW INTERCROPS NO SURFACE SHAPE OF FIELD –9.00 INTERCROP NAME NA SURFACE SHAPE OF WATER –9.00 LAND USE PADI STRUCTURE OF PLANT 1+4 RICE VARIETY LOKAL SOIL MOISTURE 4 UPLAND NO MAXIMUM PLANT HEIGHT 105.20 IRRIGATED YES AV OF LEAVES PER PLANT 43.00 RAINFED NO AV OF STEMS PER PLANT 29.00 TIDAL/SWAMP NO AV LENGTH OF LEAVES 29.95 SEEDED NO AV WIDTH OF LEAVES 1.39 TRANSPLANTED YES AV NUMBER OF PANICLES 28.80 DATE OF TRANSPLANTING 9-4-99 AV PLANT DENSITY 16.00 YIELD –9.00 AV WATER LAYER –9.00 FERTILIZER YES AV BIOMASS OF STEMS AND 136.20 LEAVES FERTILIZER NAME UREA, SP36 AV WEIGHT OF PANICLES 108.00 provincial level using daily data from the NOAA/AVHRR sensor. Integration of different data sources will lead to a reliable rice-forecasting system. For the SARI project, the acquisition of ERS images is accompanied by radar- dedicated field surveys, in which information about the field, crop(s), soil, hydrological and meteorological conditions, and other parameters is acquired to validate classification results from the ERS image series and to calibrate biomass estimates from the radar data. Table 1 gives an example for field parameters assessed. Stratified sampling of rice fields in Indonesia for suitable and less suitable rice areas is performed. This approach is not synchronized with the radar image acquisitions. Using aerial photos over 1 × 1-km areas and the actual situation in the field, plots with similar phenological stages or land-cover type are being digitized and stored in a GIS. Information from this stratified sampling survey is meant to give another independent rice data source, and can be used to validate crop area from the remote- sensing data (Table 1).

Some results The classification results of the UPRICE project can be seen in Figure 6. The classification accuracy is 78% when comparing the results with the figures of the Provin- cial Bureau of Regional Statistics. The rice map indicates the rice-growing area dur-

164 Verhoeven et al Fig. 6. Supervised classification of ERS time series, Luzon Province, Philippines. Classes are rice (yellow), urban area (red), layover (white), radar shadow (dark gray), rivers/open water (blue), and other (green). ing the wet season. In a second phase of the project, ERS radar data of the dry season are included to detect the dry-season rice areas (irrigated rice). Figure 7 (part of) shows the classification results of the SEARRI project combined with administrative boundaries. Validation is mainly done with the help of the IRRI provincial database on rice of the region (Hukes’ database). Although the Hukes’ database estimates are only valid for the base year not later than 1995, it is the only database available at the required scale (covering Southeast Asia). In general, the total rice area from the radar classification is in accordance with the Hukes’ database (± 70% accuracy). Improvement of the rice map is needed for the occurrence and number of rice cycles throughout the season. Because of the heterogeneous spatial character of many rice-growing areas, it is still difficult to detect all rice cycles using a limited number of radar images. Using the area and number of cycles throughout the season from the classification results, rice production and methane emission maps are generated. Figure 8 shows the yearly rice production per province.

Monitoring rainfed and irrigated rice in Southeast Asia . . . 165 Fig. 7. A part of the SEARRI final radar classification in a GIS layer combined with administrative boundaries. The map is based on radar time series acquired between March 1996 and December 1997. The red area in the middle of the image is the city of Bangkok, Thailand.

Fig. 8. Rice production map for the major rice-growing areas in Southeast Asia (t y–1).

166 Verhoeven et al Conclusions Monitoring of rice crops in the Asia-Pacific region in terms of changes in production and quality can be supported with a combination of radar remote sensing and GIS. Radar remote sensing provides an all-weather monitoring tool and, using time series of radar data, useful information about rice, such as area, growing stage, and cycling of crops, can be assessed over extensive areas and rice can be identified even at large geometric scales. Despite the advantages of radar remote-sensing satellites, in the case of very small fields or mountain slopes, classification of rice will be less accurate. Integration of different spatial data sources using a GIS in combination with crop growth models will improve the identification of rice and assessment of area, biomass, and yield. Low-resolution optical imagery (NOAA AVHRR, SPOT VÉGÉTATION) and radar (ScanSAR sensors) can improve the identification and classification of rice areas at a larger scale.

References Aschbacher J. 1995. Rice mapping and crop growth monitoring, an ERS/SAR demonstration project. Earth Observ. Q. 49:1-3. Denier van der Gon HAC. 1996. Methane emission from wetland rice fields. PhD thesis. Wageningen Agricultural University, Wageningen, Netherlands. 182 p. Denier van der Gon HAC, Janssen L, van Leeuwen HJC, Verhoeven R, van der Wal JT, van der Woerd H. 1999. Upscaling methane emissions from wetland rice fields (UPRICE). Po- sition paper on the feasibility of employing radar data for the Philippines. Interim report BCRS Project 4.2/AP-09. Huke RE, Huke EH. 1997. Rice area by type of culture: South, Southeast, and East Asia, a revised and updated data base. Los Baños (Philippines): International Rice Research Institute. 59 p. IRRI (International Rice Research Institute). 1995. Field variabilities of soil and plant: their impact on rice productivity and their use in modelling of soil kinetics and rice yield. Terminal report 1992-1995. The International Rice Research Institute (IRRI) & Universität Leipzig, Los Baños, Philippines. Kurosu T, Fujita M, Chiba K. 1995. Monitoring of rice crop growth from space with ERS-1 C- band SAR. IEEE Trans. Geosci. Rem. Sens. 33(4):1092-1096. Le Toan T, Ribbes F, Floury N, Wang LF, Ding KH, Kong JA, Fujita M, Kurosu T. 1997. Rice crop mapping and monitoring using ERS-1 data based on experiment and modelling results. IEEE Trans. Geosci. Rem. Sens. 35(1). van Leeuwen HJC. 1998. Feasibility study in using scatterometer data for wetland rice mapping yielding methane emission indicators for global applications. Proceedings of a Joint ESA-EUMETSAT Workshop on Emerging Scatterometer Applications from Research to Operations, 5-7 Oct. 1998, at ESTEC, Noordwijk. ESA-SP-424. Verhoeven R, van Leeuwen H. 1999. Satellite assessment of rice in Indonesia (SARI). Interim report, April 1999.

Monitoring rainfed and irrigated rice in Southeast Asia . . . 167 Notes Authors’ address: SYNOPTICS Remote Sensing & GIS Applications BV, Costerweg 1-k, 6702 AA Wageningen, The Netherlands, Tel: +31 317 421221, Fax: +31 317 416146, E-mail: [email protected], [email protected], [email protected], [email protected]. Acknowledgments: The participating organizations in the Asia-Pacific region, the European Union, European Space Agency (ESA), and Netherlands Remote Sensing Board (BCRS) are kindly acknowledged for subsidizing the abovementioned studies. The SARI project and its staff, Ir. Mubekti in particular, are gratefully thanked for their valuable contributions to this publication and its presentation at the IRRI workshop in Bali. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

168 Verhoeven et al Characterizing soil phosphorus and potassium status in lowland and upland rice-cropping regions of Indonesia

A. Clough, I.P.G. Widjaja-Adhi, J. Sri Adiningsih, A. Kasno, and S. Fukai

Mapping of extractable soil phosphorus (P) and potassium (K) began in Indo- nesia to define soil fertility in lowlands and uplands that were under rice or likely to be brought under cultivation. These maps, however, have not been published and have therefore been largely inaccessible to researchers outside of Indonesia. This chapter reviews the history of soil P and K map production in Indonesian lowlands and uplands, how mapping has been used to improve P and K fertilizer recommendations in mapped regions, and limitations of the present mapping approach in determining accurate P and K fertilizer requirements. The latest edition of soil nutrient status maps produced in 1998 covers all the lowlands of Java, Lampung in southern Sumatra, and some of the smaller outer islands with a scale of 1:250,000 or 1:500,000. The soil P and K status of the lowlands has been defined as low, medium, or high based on the responsiveness of rice crops grown in field trials to fertilizer P and K. Trials and soil surveys have shown that there is a good relationship between expected P and K requirements and actual P and K responses in lowland rice grown in soils with a low P and K fixation capacity. The maps, however, have limited application in lowland areas with high fixation capacity and their low resolution can lead to fertilizer recommendations being inaccu- rate. Despite these limitations, soil P and K mapping in lowlands has resulted in the blanket recommendations given in the 1960s being replaced by recommendations specific for small regions within provinces. Few extractable soil P and K maps have been developed for upland rice- cropping regions. The usefulness of soil P maps as a tool for developing fertilizer recommendations is limited in uplands because many soils have high P sorption capacity, particularly in Lampung. Fertilizer recommendations for upland rice cropping are still broad, especially for K.

Characterizing soil phosphorus and potassium status . . . 169 The Centre for Soil and Agroclimate Research (CSAR) has been conducting a long- term project on mapping of extractable soil P and K in lowlands since the late 1960s. Extensive mapping efforts have been restricted to soil P and K in lowlands because lowland rice is the most economically important crop production system, and P and K are the nutrients most likely to increase grain yields after nitrogen. The work at CSAR has resulted in complete extractable soil P and K maps for all the rice-cropping lowlands of Java, Lampung in southern Sumatra, Bali, and Lombok. Some small upland areas have also been mapped by CSAR for extractable soil P and K as part of land assessment processes conducted for the transmigration programs of the 1980s. Nutrient status maps produced solely by CSAR have not been published and have therefore been largely inaccessible to researchers outside of Indonesia. This chapter aims to introduce the evolution of extractable soil P and K mapping in Indo- nesia to the wider scientific community and demonstrate how the maps have been used to improve P and K fertilizer recommendations. In conjunction with mapping exercises, many fertilizer response trials have been conducted, sometimes in the same location where soil samples were taken for mapping. Results from the field trials have been used with the maps to create fertilizer recommendations for small regions. The review primarily examines the maps produced solely by CSAR but also includes higher resolution maps produced by other institutions to highlight the limitations of low-resolution mapping in improving fertilizer management. This review of the relationship between mapping and fertilizer recommendations is restricted to Java and Lampung because these areas are the predominant ones for rice cropping and they have high variability in P and K status relative to other mapped regions. The relationship between soil P status in the maps and P responsiveness of rice is exemplified through field trials including some trials conducted in 1998 in Lampung. Lampung was selected because the province has been mapped and it produces rice crops that are responsive to P but not always as expected from soil P maps because of P fixation. This is contrary to the situation in Java where rice crops have been overfer- tilized. As a result, rice is largely unresponsive to additional P and responsiveness tends to be closely related to expectations based on soil P mapping.

Soils of Java and Lampung Soil types are one of the elements of soil mapping that are used as a tool for producing fertilizer recommendations. The dominant soil types of Java and Sumatra differ; thus, the two regions need to be discussed separately (Table 1). Alluvials and Latosols are the main soil types in Java used for lowland rice cropping. Under native vegetation, lowland soils in Java are relatively rich in nutrients because Java has a history of greater volcanic activity than the outer islands, especially eastern Indonesia (Amien 1997). The Latosols and Ultisols in the uplands of Java have comparatively high native fertility and are highly weathered, acidic (pH 4 to 5), low in most nutrients (Muljadi 1997), and susceptible to erosion even when terraced (Amien 1997). However, with good fertilizer management, the uplands are productive (Sri Adiningsih et al 1991).

170 Clough et al Table 1. Area and distribution of soils in Indonesia (000 ha).

Soil type Java Sumatra

Red yellow podzolic (Ultisols) 325 15,950 Organosols (Histosols) 25 6,781 Alluvials (Entisols) 2,550 6,238 Latosols (Inceptisols/Oxisols) 2,831 6,788 Mediterranean (Alfisols) 1,625 – Andosols (Inceptisols) 844 2,725 Podsols (Spodosols) – 931 Regosols (Entisols/Inceptisols) 1,431 831 Grumusols (Vertisols) 1,481 – Renzina (Mollisols) 38 394 Complexes (mostly Ultisols) 2,069 6,725 Total 13,219 47,363

Source: Muljadi and Soepraptohardjo (1975).

In Sumatra, upland cropping constitutes the bulk of agricultural land use (88%) and red-yellow podzolic soils are the dominant soil type (Santoso 1991), particularly in areas newly opened for agriculture through the transmigration programs. The major region of agricultural development is the southern-most province of Lampung. The soils of the newly developed regions of Lampung are high in kaolin (>30%) with low organic matter (<1%), low water-holding capacity, low P, N, K, S, Mg, Ca, and Zn status, and high P fixation capacity (Sudjadi 1984, Prasetyo et al 1997). Red- yellow podzolic soils also tend to be acidic (pH 4 to 5) and this can lead to problems such as Ca and Mg deficiencies, readily leached K, high P, S, and Mo sorption, and excessive hydrogen ions. In addition to acidity problems, plant growth can be limited in these same soils through high Fe concentrations occurring within a few centimeters of the soil surface (Prasetyo et al 1997) and/or high Al saturation (Santoso 1991). The problems associated with the use of red-yellow podzolic soils for upland rice production are destined to be of increasing significance as previously uncultivated land is made available to farmers. Organosols, alluvials, and Latosols are also prevalent particularly in the coastal swamp lands in Sumatra that have been developed as part of the transmigration program (Widjaja-Adhi et al 1996). The alluvial soils in Sumatra contain pyrites and are acid sulfate soils. The extent of the various soil types in the outer island’s coastal swamp lands was mapped in 1991-92 on a 1:500,000 scale using land unit maps and soil maps produced by CSAR. Areas with deep peats (>3 m) or high susceptibility to erosion as determined by a set of distinct criteria are considered to be unsuitable for food cropping (Widjaja-Adhi and Karama 1994). Organosols have high N and organic carbon (OC) contents, but productivity on these soils is poor due to subsidence after drainage, the slow release of OC and N, and high susceptibility to erosion through poor soil structure and slow water infiltration rates (Sudjadi 1984). Efforts to overcome these limitations have focused on encour-

Characterizing soil phosphorus and potassium status . . . 171 aging small landholders to farm together as a group and controlling water movement across a series of farms (Widjaja-Adhi and Karama 1994). Newly cleared Organosols intended for the transmigration program tend to have less nutrients than the peats that have been farmed by spontaneous migrants for decades (Ruddle 1987). Soil mapping has helped identify the main soil types in Java and Sumatra and has highlighted the differences between the two regions. Based on mapping and related field trials, fertilizer recommendations have been developed that account for the differences in soil types.

Development of phosphorus and potassium maps and fertilizer recommendations for lowlands Developing soil phosphorus maps and fertilizer recommendations Before 1972, P fertilizer recommendations did not take into account the effects of diversity of soil types, chemical properties of soils, or cropping histories. During the 1960s, a blanket rate of 20 kg P ha–1 for both upland and lowland rice was recommended to all farmers throughout the country. Efforts to improve the specificity of P recommendations for particular regions and soil types have focused on soil mapping and relating those maps to results from fertilizer field trials in the lowlands. Much of the mapping work has been conducted by CSAR with some work being done through collaboration with overseas institutions. Lowland soil P status mapping in Java began before 1970. In the first edition of the soil P maps for Java that were completed in 1971 (Widjaja-Adhi, personal communication), lowland areas were divided into two classes: responsive and nonrespon- sive to P fertilizer application. All soil P values for the first and following series of maps were determined by extraction in 25% HCl. Soils were deemed to be nonre- –1 –1 sponsive if extractable soil P was higher than 88 mg P kg (20 mg P2O5 100 g ). This cutoff point was based on findings that yield response to P fertilizer generally occurred only in soils where the P status was less than 88 mg P kg–1. Based on the initial soil map, there were two recommendations: 0 and 20 kg P ha–1 crop–1. The accuracy of the first map was limited because samples were taken only from research stations where field trials were also held. The relationship between field response trials and soil P status at the research stations was extrapolated to farmers’ fields by assuming that soil P status and crop response were related to soil type. Thus, the first soil P status map was based on soil type. The second edition of soil P maps for lowlands in Java was developed in 1974 (Widjaja-Adhi, personal communication). These maps were also based on soil type but with extensive analyses of soils for reserved P estimations taken from a wide range of locations. The third and fourth editions of the soil P maps of Java were completed in 1984 and 1992, respectively (Widjaja-Adhi, personal communication). Both the third and fourth editions divided the soils into three P status classes: high, medium, and low, which required 10 to 25 kg P ha –1 at varying frequencies (Table 2). The main change between the third and fourth editions of the P status maps was that the number of soil samples used to identify whether an area had low, medium, or high

172 Clough et al High

, medium soil K 166– K soil medium ,

–1

– ( 0) 91,571 (100)

– ( 0) 122,485 (100)

Potassium status

. Low soil K <166 mg kg mg <166 K soil Low .

–1

– ( 0)

Low Medium

, high soil P >176 mg kg mg >176 P soil high ,

–1

High

, medium soil P 88–176 mg kg mg 88–176 P soil medium ,

–1

Phosphorus status

. Data sourced from CSAR (unpublished).

–1

– ( 0) 11,652 (10) 110,833 (90)

LowMedium

1,996 ( 2) 15,521 (17) 74,054 (81)

30,470 (12) 118,180 (45) 115,831 (44) 19,595 ( 7) 139,935 (53) 104,951 (40)

48,224 (16) 128,116 (43) 120,818 (41) 12,071 ( 4) 56,505 (19) 228,582 (77)

17,707 ( 8) 47,453 (22) 147,922 (69) 104,048 (49) 53,824 (25) 55,210 (26)

235,621 (19) 454,396 (37) 523,348 (43) 225,625 (19) 496,250 (41) 491,490 (41)

183,500 (15) 544,945 (43) 531,475 (42) 71,875 ( 6) 345,625 (27) 842,420 (67)

, high soil K >332 mg kg

–1

Region in the Nusa Tenggara Barat Province. Low soil P <88 mg kg mg <88 P soil Low Province. Barat Tenggara Nusa the in Region

West Java West Central Java 123,439 (10) 658,785 (56) 397,120 (34) 175,050 (15) 330,000 (28) 674,294 (57) East Java Aceh North Sumatra 53,440 (10) 301,598 (57) 175,425 (33) 10,135 ( 2) 430,633 (81) 89,695 (17) Jambi West SumatraWest 37,389 (17) 95,983 (43) 91,793 (41) 50,398 (22) 110,711 (49) 64,056 (28) South Sumatra 145,570 (34) 251,981 (59) 32,315 ( 8) 12,910 ( 3) 261,290 (61) 155,666 (36) Lampung South Kalimantan 145,829 (31) 164,206 (35) 155,186 (33) 66,252 (14) 261,333 (56) 137,636 (30) South Sulawesi 115,448 (20) 175,456 (30) 290,116 (50) 26,669 ( 5) 89,070 (15) 465,281 ( 80) Bali Palau Lombok

Table 2. Table Lowland area (ha) classified medium, as and low, high soil P and K status for 13 Indonesian provinces (percentage of total area given in parentheses).

Province a 332 mg kg

Characterizing soil phosphorus and potassium status . . . 173 Table 3. Phosphorus fertilizer recommendations for lowland rice grown in Java based on soil P status as determined by 25% HCl extraction (Hermanto 1995).

Extractable P Recommended P rate (mg kg–1) (kg ha–1)

<88 20–25 every season 88–176 15 every second season >176 10 every fourth season

soil P was increased. In 1998, the fourth edition of lowland maps for Java and the outer islands was digitized by CSAR. The definition of low, medium, and high soil P was derived from P responsiveness field trials conducted throughout the lowlands of Java (Table 3). The lack of response to P fertilizer was demonstrated in 18 and 15 trials conducted by Sri Adiningsih (unpublished) over the 1987-88 wet season and 1988 dry season, respectively. Rice grown in the wet season responded to P at only seven sites and rice grown in the dry season responded at only two sites. The significant yield increases tended to be small (0.5 t ha–1) in terms of actual grain yields. Trials with rice on experimental farms located at Muara in West Java and Ngale in East Java in the wet and dry seasons also produced no response to P applications (Miyake et al 1984). The fourth edition (Widjaja-Adhi and Sri Adiningsih, personal communication) of lowland soil maps for Java (scale 1:500,000) showed that most of the lowlands of Java were classified as having medium to high soil P status (Fig. 1). The dominance of medium to high P status is because the lowlands have received P fertilizer in excess of that required for rice crops for many years. Over the past 30 years, P has been applied to irrigated and rainfed lowland rice fields at an average of 20 kg P ha–1 crop–1. Subsequently, extractable soil P has increased as P has accumulated in the soil and P fertilizer requirements in 1999 are low (Table 2). Mapping data and results from lowland rice trials indicate that P fertilizer need only be applied to most lowland rice in Java at rates equivalent to the amount of P removed in the grain. High-yielding rice cultivars may produce between 5 and 8 t grain ha–1 (Hermanto 1995); thus, P fertilizer needs to be applied at between 12.5 and 20 kg ha–1 crop–1 to avoid removing P from the soil. However, most lowland areas in Java can be cropped for several seasons without a depletion of soil P resulting in a decline in grain yield. Soil P maps (1:500,000) of the lowland areas of Lampung also show that large areas have a high soil P status (Fig. 2), which has been extensively developed for agricultural use through the transmigration programs. However, using the same classification system, other provinces such as South Sumatra are mainly classified as having low to medium soil P status (Table 2). Lowland regions on most of the other islands that have been mapped are deemed to have low to medium soil P status. Ex- ceptions are the soils of Bali and Lombok (Fig. 3), which have medium to high soil P because the islands are volcanic.

174 Clough et al are indicated by red, yellow, and green shadings, respectively). are indicated by red, yellow,

–1

Fig. 1. Extractable soil P map (25% HCl extraction) of the lowlands of West Java (Jawa Barat) produced in 1998 (1:500,000) 1998 in produced Barat) (Jawa Java West of lowlands the of extraction) HCl (25% map P soil Extractable 1. Fig. (<88, 88–176, and >176 mg P kg

Characterizing soil phosphorus and potassium status . . . 175 Fig. 2. Extractable soil P map (25% HCl extraction) of the lowlands of Lampung Province in Sumatra produced in 1998 (1:500,000) (<88, 88–176, and >176 mg P kg–1 are indicated by red, yellow, and green shadings, respectively).

176 Clough et al are indicated by yellow and green and yellow by indicated are

–1

Fig. 3. Extractable soil P map (25% HCl extraction) of the lowlands of Palau Lombok in the province of Nusa Tenggara Barat produced in 1993 (1:250,000) (88–176 and >176 mg P kg P mg >176 and (88–176 (1:250,000) 1993 in produced Barat Tenggara shadings, respectively).

Characterizing soil phosphorus and potassium status . . . 177 Benefits and limitations of soil phosphorus mapping In the 1970s, an undeveloped upland and lowland area of Sitiung (Mimpi Plain) in West Sumatra was forwarded as a potential area for settlement of transmigrants and for upland research coordinated as part of the Benchmark Soils Project conducted by the University of Hawaii (Buurman and Sukardi 1980). The area was mapped at a moderate resolution (1:40,000) for soil type by CSAR by taking soil samples from more than 50 locations scattered throughout the area in 1976-77 (Soil Research Insti- tute 1979). The mapping project demonstrated the benefits of mapping for agricultural development policy and soil classification, and the limitations of low-resolution mapping conducted by CSAR. Some of the sites in Sitiung were classified in detail with samples being taken to a depth of 180 cm. This detail enabled several soils in the region to be reclassified among the classes brown tropical forest, Latosol, and podzolic (Buurman and Sukardi 1980). Soil profiles from 88 sites in the same region were later used to map topsoil properties including KCl-extractable Al, extractable P (25% HCl), sand and clay contents, total P, and pH, and 109 scattered samples were used to map variation in silt content. Actual soil property values were translated into maps by krigging to give 268 points within the sample area (Trangmar et al 1984). The soil P status maps from Sitiung showed that extractable soil P was low to medium throughout the region (mainly 40–160 mg P kg–1) with high P sorption capacity compared with other tropical Oxisols and Ultisols. Mapping showed that native extractable soil P in uplands and lowlands was correlated positively with silt content (r = 0.52**) and negatively with sand content (r = –0.55**) at 0 to 15-cm depth. The relationship between soil P and soil texture in Sitiung demonstrates that CSAR’s original soil P recommendation maps completed in 1971 (Widjaja-Adhi, personal communication), which were based on soil type rather than directly measuring soil P in each field, were well founded. However, the variation in soil P in Sitiung shows that a weakness in mapping conducted by CSAR may be in the resolution since the maps are only 1:250,000–1:500,000 scale. This low resolution may explain why not all field trials gave the responses as expected based on the soil maps. For example, P response trials conducted near Indramayu and Subang in 1987-88 responded to P application despite the area being classified as having high P status (Sri Adiningsih, unpublished). Good correlation between actual yield and expected yield based on the soil nutrient status maps may also be limited because the maps do not distinguish irrigated, rainfed, or tidal lowlands. All the lowland types are presented on one map and recommendations that follow from the mapping exercises and field trials apply to all lowland rice-growing conditions. This may limit the usefulness of the maps as a tool for formulating P recommendations since average grain yields and farmers’ aversion to the risks involved in investing in fertilizers differ between cropping systems. The fourth edition of soil P maps shows that much of the lowlands of Lampung in Southern Sumatra have a high soil P status. This was confirmed by a soil survey of 32 irrigated lowland sites conducted in 1998 before beginning field trials in a program known as the Acid Soils Project (Kasno, unpublished). The survey showed that,

178 Clough et al Table 4. Recommendations for P application on lowlands of Sumatra.

Extractable P Recommended P ratea (mg kg–1) (kg ha–1)

>88 0 66–88 33 44–65 66 22–43 99 <22 132

aRecommendations are for each cropping season. Data sourced from CSAR (unpublished).

based on the low, medium, and high classifications used in Java, only 11 lowland sites were likely to respond to P applications. One of the sites expected to be responsive, Braja Mas, was selected for the Acid Soils Project and was transplanted to rice in the 1998-99 wet season. The soil P for the site was 55 mg P kg–1 (25% HCl extraction). As expected, the rice crop gave a positive response to P with 90% maximum grain yield achieved with 52 kg P ha–1. The amount of P required is close to that recommended by CSAR for all lowlands in Sumatra (Table 4). The recommendations are higher for lowlands in Sumatra than for those in Java because of the need to overcome P fixation in some soils and the soils’ inherent low fertility. Rice grown in farmers’ fields is responsive to P applications and the potential exists to increase soil P status to the point where P applications are no longer required as is the case in most lowlands of Java (Fig. 1). In Lampung, particularly on the red-yellow podzolic soils, high P fixation capacity limits the relevance of soil P status to P recommendations. This was shown in an irrigated field trial conducted at Pringsewu in Lampung as part of the Acid Soils Project in the 1998-99 wet season. The Pringsewu site had very high soil P (>300 mg P kg–1) yet responded to P application, producing 90% maximum rice grain yield (4.2 t ha–1) with 59 kg P ha–1. An additional detrimental effect of lowland soils with high P-fixing capacity is their tendency to fix any P that is not used in the year of application. CSAR recommendations for lowlands in Java are to apply P only every fourth season if soil P is classified as high. However, two irrigated lowland trials in Lampung conducted during the Acid Soils Project showed that P applied in excess of crop requirements was not extractable by 25% HCl at the end of the growing season. This type of field result has led to the recommendation that P be applied every season in Sumatran lowlands.

Validation of soil phosphorus mapping In 1994-95, the maps of East, Central, and West Java compiled by CSAR were used as a baseline for predicting the change in status of extractable soil P after several cropping seasons (Pandutama 1996). Soils for the study were collected from several lowland fields throughout Java covering the main soil types present on the island:

Characterizing soil phosphorus and potassium status . . . 179 Latosol, Regosol, alluvial, and Grumusol (Oxisol, Entisol, Inceptisol, Vertisol). The location of each sampling was based on the fourth edition of soil P maps produced by CSAR in 1992 to ensure that soils with low, medium, and high P status were included in the study. Soil P (25% HCl extraction) for the ten sites selected ranged from 83 to 633 mg P kg–1 (median = 258 mg P kg–1). A series of rice crops were grown in each soil with and without P fertilizer. At each site, the change in soil P over the experimental cropping period was related in a model to clay content, organic carbon content, physical clay activity, initial soil P status (25% HCl extraction), and P application rate. Soil parameters in the model had been selected from 12 potentially influen- tial soil characteristics using multiple regressions. The model was then used to predict the decline in soil P after rice cropping without P application given a particular initial soil P value for each soil type. Predictions by Pandutama (1996) confirmed and refined what CSAR had found based on its nutrient status mapping and fertilizer field trials. Pandutama’s (1996) estimates of P requirements based on mapped P status and rates of decline with cropping predicted that lower rates (10 kg P ha–1) could be applied to achieve yields of up to 4 t ha–1 than the rates (15 to 25 kg P ha–1) recommended by CSAR for soils with medium to high P status. Although the target yield is less than the maximum potential yield of modern varieties, the yield is realistic for Javanese farmers.

Developing soil potassium maps and fertilizer recommendations Soil mapping for extractable K (25% HCl extraction) in the lowlands was conducted using techniques similar to soil P mapping, that is, information was collected over several years and recommendations were created based on results of soil sampling and field trials, although K recommendations are not as well developed as P recommendations. In 1989, an extractable soil K status map of Java, including Madura, was made based on 600 soil samples and some field trials. Like the soil P maps of Java, soil K maps divide the lowlands into three categories: high, medium, and low. The 1989 map indicated that 40% of lowland soils had low K status (<20 mg K kg–1 in 25% HCl extraction) (Didi Ardi et al 1989). In 1996, less than 13% of lowland soils on the island of Java required K fertilizer (Karama et al 1998). In Java, the reduction in the amount of land with low K is due to farmers applying K at about 3 to 4 kg ha–1 since 1978 (Sri Adiningsih et al 1990, O’Brien et al 1990) and farmers adopting the practice of incorporating rice residues, which can contain 40% to 60% of the K taken up by the crop (Dobermann et al 1996a). Conse- quently, many lowland rice crops in Java are no longer responsive to K application (Sri Adiningsih et al 1990, Sri Rochayati et al 1990) and many of Java’s lowland soils are classified as high (Fig. 4). Lack of response to K in the lowlands of Java indicates that K fertilizers need not be applied in many regions, as shown in the soil K maps. The areas of lowland Java that are low in K also tend to be low in soil P, as shown in the P and K maps of Java (Figs. 1 and 4). Some sites, however, are still responsive to K as demonstrated by K trials at the Jakenan Experimental Station in Central Java with rainfed rice (Wihardjaka et al 1998). The two initial consecutive trials at Jakenan

180 Clough et al ed, yellow, and green shadings, respectively). ed, yellow,

are indicated by r

–1

Fig. 4. Extractable soil K map of the lowlands of West Java (Jawa Barat) produced in 1998 (1:500,000) (<83, 83– (<83, (1:500,000) 1998 in produced Barat) (Jawa Java West of lowlands the of map K soil Extractable 4. Fig. 166, and >166 mg K kg

Characterizing soil phosphorus and potassium status . . . 181 gave a positive yield response with application of 75 kg K ha–1. In the third, fourth, and fifth trials, there was no increase in grain yields due to either increases in soil K status (third and fifth trials) or low in-crop rainfall (fourth trial). The prevalence of high soil K status in Java needs to be considered along with the K balance of the soil when formulating K fertilizer recommendations. A positive soil K balance was shown to occur when K was applied to some treatments in Jakenan rainfed rice trials. In the treatments at Jakenan where N and P were added without K, the K balance was negative, whereas adding NPK (120:18:75) gave an increase in soil K (Wihardjaka et al 1998). Trials with irrigated rice in Java (Dobermann et al 1996a), however, showed that K uptake was greater than the amount applied, thus creating a negative K balance in the soil at the end of the season. Cropping systems with a negative K balance are unsustainable in the long term and K fertilizers need to be applied to reduce the risk of K deficiencies. Thus, there is only one recommended rate for the whole of Java: 26 kg K ha–1 (Karama et al 1998). Lowland rice has been shown to be more responsive to fertilizer K in Sumatra than in Java. This reflects the fact that low-K areas are more prevalent in Lampung (Fig. 5). Unlike in Java, however, soil P maps are not related to soil K maps. Results from soil K mapping in Lampung are supported by a soil survey conducted in 1998 for the Acid Soils Project, which found that 25 of 32 lowland sites had low soil K (<50 mg K kg–1) as determined by 25% HCl extraction. Surveys in West and South Sumatra showed that only 10% of the lowland soils had high soil K (>166 mg K kg–1) (Karama et al 1998). Despite the need for K fertilizer to be applied in the lowlands of Sumatra, no K fertilizer recommendations specific to that region have been developed. Developing K recommendations is hampered by the difficulty in relating soil K extracted by 1M NH4Oac to grain yields. Dobermann et al (1996b) concluded that predicting crop K uptake in irrigated rice using a static soil test was not practical because too many soil properties (extractable K, Ca, and Mg combined, CEC, organic matter, clay content) were required to give accurate results. K uptake in irrigated lowland rice has been shown to be related to soil K (r2 = 0.82) as measured using mixed-bed ion exchange resin capsules (Dobermann et al 1996b). Ion exchange resins have the advantage of being able to measure the amount of K that can be extracted from a soil over time rather than measuring K in solution at a particular moment. Dynamic measurements of K are required for intensive cropping systems and where the soils have K-fixing properties. Data used to test the validity of predicting total K uptake at harvest in lowland rice were derived from NPK trials conducted at 11 sites in five countries including Indonesia with K as the only limiting factor (Dobermann et al 1996b). K fertilizer rates varied between sites from 25 to 100 kg K ha–1, soil pH ranged from 5.7 to 8.5, and clay contents ranged from 25% to 57%. This same method could possibly be used to predict K uptake by rainfed lowland rice in locations within Java and Lampung.

182 Clough et al Fig. 5. Extractable soil K map of the lowlands of Lampung Province in Sumatra produced in 1998 (1:500,000) (<83, 83–166, and >166 mg K kg–1 are indicated by red, yellow, and green shadings, respectively).

Characterizing soil phosphorus and potassium status . . . 183 Phosphorus and potassium recommendations for uplands Developing phosphorus recommendations Apart from isolated exercises such as the mapping of Sitiung in the 1970s, very little soil mapping of uplands has been conducted. The lack of work in upland mapping is primarily due to the uplands making only a small contribution to national rice production compared with the lowlands, especially in Java. Phosphorus recommendations are therefore based solely upon responses of rice grown in field or pot trials and tend to still be broad. Recommendations for upland rice and other food crops are 20 to 40 kg P ha–1 in the first crop and 20 kg P ha–1 for each crop thereafter. The difference between actual and recommended P fertilizer rates applied to upland rice in Java is small. As such, the upland area that is low in soil P is less than 20% for Java and getting smaller. In Lampung, the amount of upland rice-growing area that has low soil P is also declining; however, responsive sites are still prevalent. A survey of 11 upland sites conducted when the Acid Soils Project began showed that a response to P could be expected at 10 of the sites (Kasno, unpublished). The upper limits of the recommendations are constrained by farmers’ potential to invest in P fertilizer in a rainfed environment where risk of crop failure is compounded by the presence of other nutritional deficiencies. The need for a holistic approach to fertilizer management in upland rice cropping was shown in a trial conducted in the 1998-99 wet season at the Taman Bogo Experimental Farm as part of the Acid Soils Project in Lampung. The trial at Taman Bogo gave a positive response to SP-36 (16% P as superphosphate) applied at 80 kg P ha–1, but only when applied in conjunction with >80 kg N ha–1 (Clough, unpublished). The validity of these broad P recommendations for upland crops was exemplified, however, by trials with upland rice on newly formed bench terraces with extractable soil P (Bray I) of 6 mg P kg–1. Grain yields in upland rice increased upon applications of 0 to 40 kg P ha–1 in three consecutive years (Schmidt et al 1990). Mean yield across all treatments was 1.88, 2.40, and 2.59 t ha–1 in 1984, 1985, and 1986, respectively. Within treatments, the average yield across the three years was 0.35, 2.45, 2.89, and 3.46 t ha–1 for applications of 0, 10, 20, and 40 kg P ha–1, respectively. In low P-fixing conditions, such as those presented by Schmidt et al (1990), soil P mapping in the uplands of Java may be an effective means of improving P recommendations. In Lampung, however, where the uplands are dominated by P-fixing red- yellow podzolics, soil mapping may be of limited value.

Limitations of using soil phosphorus mapping for developing fertilizer recommendations due to P fixation P fixation capacity is a significant issue for P fertilizer management in the red-yellow podzolic soils of Lampung and other provinces of Sumatra. High P fixation capacity of soils is more likely to occur when several of the following factors are present (Brady and Weil 1999):

184 Clough et al ● acidic pH (< 6) ● low organic matter content ● high clay content, particularly kaolinite, gibbsite, or goethite ● high concentrations of Fe, Al, or Mn ions in soil solution The variability in response of rice to P applications because of P fixation in red- yellow podzolic soils has been demonstrated in several P trials. P response trials with upland rice in Lampung have shown an increase in grain yield with P application at some sites with relatively low soil P, particularly in newly cleared areas (Effendi et al 1982). However, a two-year trial with upland rice in Pekalongan, Lampung, gave no response to P (Makarim 1990), whereas upland rice grown on a clayey Typic Paleudult produced 90–100% maximum grain yields at 20 kg P ha–1 (Palmer and Sudjadi 1984). A trial on a newly cleared site in Central Lampung showed that grain yields could be doubled by applying P fertilizer at rates similar to those recommended for irrigated rice in Java (Effendi et al 1982). The dual effects of extractable soil P and P sorption capacity are exemplified in two upland P trials conducted in Lampung during the 1998-99 wet season as part of the Acid Soils Project. Site selection was based on soil P maps produced by CSAR and 1998 soil surveys. Phosphorus was applied at four rates, the maximum rate being determined through the soils’ P sorption curve to be adequate to give 0.02 µg mL–1 soil P in solution. Rice crops grown at both sites gave a positive response to SP-36 application (P <0.005). The magnitude of the grain yield response was related to the sites’ soil P (25% HCl extraction) before P treatment and P sorption capacity. Soil P was 118 and 191 mg P kg–1 and maximum P fertilizer application was 327 and 96 kg P ha–1 at Buyut Udik and Jagang, respectively. The amount of P fertilizer required to give 90% maximum grain yield was 95 kg ha–1 at Buyut Udik and 55 kg ha–1 at Jagang. This means that Jagang, the site with relatively high initial soil P and low P sorption, required less P fertilizer for the rice to achieve maximum grain yield than Buyut Udik. The diversity of results from response trials reflects the range of extractable soil P values and P sorption capacity found at upland sites in Sumatra. A tactic promoted to reduce the problems of P fixation is to apply rock phosphate (RP) instead of readily available SP-36 or triple superphosphate (TSP). A review of P response studies in upland rice (Partohardjono and Sri Adiningsih 1991) concluded that, although the initial grain yields with RP were lower than with TSP, over several seasons five RP fertilizers from Indonesia and Christmas Island were just as effective as TSP. In 1998, as part of the Acid Soils Project, an upland trial was established to determine whether a blend of readily available phosphate and rock phosphate could reduce P loss through fixation while maintaining yields in the first season. The trial compared locally derived RP (Ciamis), SP-36, and Prolong (supplied by Pivot Ltd.). Prolong is a combination of superphosphate and reactive rock phosphate (P:S:Ca = 9.5:15:23), which is designed to provide a readily available source of P to crops and a source that is slowly released over one or more seasons. Phospho- rus was applied at 0, 20, 40, and 80 kg P ha–1. As expected from a low-P-fixing site with medium soil P status, all P sources gave a significant increase in grain yield (P = 0.005) with 90% maximum grain yield being achieved with 40 kg P ha–1. The re-

Characterizing soil phosphorus and potassium status . . . 185 sponse of the crop to the three P sources differed with Prolong giving significantly higher grain yields than the other two sources (P = 0.007). The rice crop achieved 90% maximum grain yield with 28 kg Prolong-P ha–1 compared to requiring about 60 kg P ha–1 applied as SP-36. However, these findings that show that commonly available P sources are equally as effective over time at increasing yields must be judged relative to the cost of each P source. Economically, RP is more efficient than SP-36 with rice farmers paying about Rp 500 (US$0.07) kg–1 of RP fertilizer (Rp 60 per kg P) and about Rp 2,500 (US$0.38) kg–1 of SP-36 fertilizer (Rp 400 per kg P) in Lampung in July 1999. At present, rice farmers are willing to pay a maximum of Rp 1,500 (US$0.23) kg–1 of fertilizer.

Developing potassium recommendations As stated previously, no intensive effort has been made to create soil K maps from the uplands of Lampung or Java; consequently, K recommendations are broad. Upland rice grown in Lampung has been shown to be responsive to K. Results from field trials were supported by a soil survey conducted in 1998 for the Acid Soils Project, which found that 9 of 11 upland sites had low soil K (<50 mg K kg–1) as determined by 25% HCl extraction. The prevalence of areas with low soil K in the uplands of Lampung is partially due to the dominance of acid soils. Under acidification processes that occur in upland areas of Lampung, adsorbed K+ is replaced by H+ and Al3+. Subsequently, K+ is easily leached down the soil profile. This situation is accentuated by the availability of K being lower in sandy or acid soils or where the clay mineralogy is dominated by kaolinite, as is the case in upland Lampung. High Ca and Mg in a soil, also present in uplands, can further reduce K uptake by plants. All treatments in the lowland and upland trials at the Taman Bogo Experimen- tal Farm conducted as part of the Acid Soils Project received a basal application of KCl at 60 kg K ha–1. The mean K concentration of the upland rice grain was slightly lower (0.15%) than the mean value (0.21%) given for four rainfed K trials at the Jakenan Experimental Station (Wihardjaka et al 1998). The K balance was positive with about 27 kg ha–1 of soil K available at the site in addition to applied K. Total K content in the grain and straw was about 42 kg ha–1 where no N or P was applied and 56 to 73 kg K ha–1 where the crop was given 80 kg P ha–1 and 40–160 kg N ha–1. Other upland rice trials in Lampung have found the K balance to be negative (Gill and Kamprath 1990) with K recovered being greater than 100%. Recoveries greater than 100% in the upland rice were attributed to applied K stimulating root growth and consequently increasing uptake of native K. Additional K uptake even in nondeficient rice crops has been shown to be beneficial for increasing grain yields through being accompanied by a reduction in the incidence of disease (Gill and Kamprath 1990) or iron toxicity (Ismunadji 1990). In upland Sumatra, where soils are inherently low in K, the application rates need to be increased to match K removal and possibly assist in alleviating the detrimental effects of iron toxicity. General K fertilizer recommendations for upland crops were developed by CSAR based on field trials in Lampung and Java. Based on field

186 Clough et al trials, K recommendations for upland rice and maize are 66–100 kg K ha–1 (Sri Adiningsih et al 1991), which are far higher than the rates recommended for irrigated lowland and higher than the amounts applied by upland farmers. The higher recommended rates reflect the poor native fertility of upland soils compared with lowland soils.

Conclusions Mapping soil type and P and K status is a long-term task that has been carried out in Indonesia continuously for more than 30 years. Over that 30-year period, the maps have become more detailed by dividing provinces into three instead of two nutrient status categories and maps have been made more accurate by using large sample sizes to categorize regions. Mapping P and K status in lowland Java has been completed and efforts now focus on mapping lowlands in the outer islands. Extensive progress has been made in mapping Sumatra, Bali, and Lombok. Comparisons between P and K maps show that low P is usually accompanied by low K except in some regions such as Lampung. The combination of mapping and field trials has led to the development of P fertilizer recommendations based on soil P for the lowlands of Java. To this end, the maps for P status in Java are adequate despite the low resolution. Adequacy primarily stems from the fact that overall P status in Java is high due to intensive P fertilizer application. P fertilizer recommendations in Java may be improved by superimposing a rice-cropping system map (irrigated, rainfed, tidal) over the P status maps since P fertilizer has not been evenly applied across the three production systems (Widjaja- Adhi, personal communication). The P recommendations for Java also appear to be applicable in other regions such as Bali and Sumatra where P sorption is not a significant issue. Together, mapping and field trials show that P fertilizer is only required in limited amounts in the lowlands of Java and at rates that are declining in Lampung. Improvements in P recommendations may be acheived in some areas of Lampung by mapping at a higher resolution where current maps show that soil P status varies within small (20 km2) lowland rice production areas. Potassium recommendations are less specific than P recommendations. The mapping, however, is comprehensive, especially in Java, and K status-specific recommendations could be developed from field trials conducted in the region. Trials, mapping, and surveys show that the benefits of K fertilizer applications are underesti- mated by farmers in terms of both direct yield increases and the reduced risk of iron toxicity. Mapping in upland regions is less developed than in lowland regions; this is reflected by less specific P and K fertilizer recommendations being available to upland rice farmers. Fertilizer trials in the uplands show that P and K applications are more likely to be beneficial there than in the lowlands. Phosphorus has the added problem of P sorption in the red-yellow podzolic soils that are prevalent in Lampung. Recommendations specific to the uplands of Lampung need to take this into account by using soil P status and the soils’ sorption capacity in models.

Characterizing soil phosphorus and potassium status . . . 187 References Amien I. 1997. Land resource analysis for agriculture development. Indones. Agric. Res. Dev. J. 19:1-12. Brady NC, Weil RR. 1999. Soil phosphorus and potassium. In: The nature and properties of soils. Upper Saddle River, NJ (USA): Prentice-Hall Inc. p 540-584. Buurman P, Sukardi. 1980. Brown soils, latosols or podzolics? In: Buurman P, editor. Red soils in Indonesia. Wageningen (Netherlands): Centre for Agricultural Publishing and Docu- mentation. p 93-104. Didi Ardi S, Supardi AK, Aldjabri M, Supartini J, Adiningsih S, Sudjadi M. 1989. The status of potassium in lowland soils in Java and Madura. Bogor (Indonesia): Centre for Soil Re- search. Dobermann A, Cassman KG, Sta. Cruz PC, Adviento MA, Pampolino MF. 1996a. Fertilizer inputs, nutrient balance, and soil nutrient-supplying power in intensive, irrigated rice systems. II. Effective soil K-supplying capacity. Nutr. Cycl. Agroecosyst. 46:11-21. Dobermann A, Sta. Cruz PC, Cassman KG. 1996b. Fertilizer inputs, nutrient balance, and soil nutrient-supplying power in intensive, irrigated rice systems. I. Potassium uptake and K balance. Nutr. Cycl. Agroecosyst. 46:1-10. Effendi S, Ismail IG, McIntosh JL. 1982. Cropping systems research in Indonesia. In: Report of a workshop on cropping systems research in Asia. Manila (Philippines): International Rice Research Institute. Gill DW, Kamprath EJ. 1990. Potassium uptake and recovery by an upland rice-soybean rotation on an oxisol. Agron. J. 82:329-333. Hermanto MS. 1995. Rice production technology supporting sustainable rice self-sufficiency in Indonesia. Bogor (Indonesia): Central Research Institute for Food Crops. p 62. Ismunadji M. 1990. N, P, K and S status of food crops in islands outside Java. In: Blair G, Lefroy R, editors. Sulfur fertilizer policy for lowland and upland rice cropping systems in Indonesia, Jakarta, Indonesia, 18-20 July 1989. Australian Centre for International Agricultural Research (ACIAR) Proceedings No. 29. Canberra (Australia): ACIAR. p 83-86. Karama AS, Fagi AM, Rochayati S. 1998. Current use of and requirement for nutrients for sustainable food crops production in Indonesia. In: Johnston AE, Syers JK, editors. Nu- trient management for sustainable crop production in Asia. Wallingford (UK): CAB International. p 291-305. Makarim AK. 1990. S research on upland crops in Indonesia. In: Blair G, Lefroy R, editors. Sulfur fertilizer policy for lowland and upland rice cropping systems in Indonesia, Jakarta, Indonesia, 18-20 July 1989. Australian Centre for International Agricultural Research (ACIAR) Proceedings No. 29. Canberra (Australia): ACIAR. p 91-94. Miyake M, Ismunadji M, Zulkarnaini I, Roechan. 1984. Phosphate response of rice in Indone- sian paddy fields. Tech. Bull. Trop. Agric. Res. Cent. 17:1-78. Muljadi D. 1997. Application of phosphate rock fertilizer based on soil and climatic conditions in sustainable Indonesian agriculture. Indones. Agric. Res. Dev. J. 19:35-42. Muljadi D, Soepraptohardjo M. 1975. Masalah data luas dan penyebaran tanah-tanah kritis. In: Simposium Pencegahan dan pemulihan tanah-tanah kritis dalam rangka pengembangan wilayah, Jakarta, Indonesia.

188 Clough et al O’Brien DT, Blair G, Lefroy R. 1990. The fertilizer subsidy and fertilizer use in Indonesia. In: Blair G, Lefroy R, editors. Sulfur fertilizer policy for lowland and upland rice cropping systems in Indonesia, Jakarta, Indonesia, 18-20 July 1989. Australian Centre for Inter- national Agricultural Research (ACIAR) Proceedings No. 29. Canberra (Australia): ACIAR. p 42-50. Palmer B, Sudjadi M. 1984. Efficient use of fertilizer in upland cropping systems: an experimental approach. In: Bay-Petersen J, editor. Ecology and management of problem soils in Asia. Food and Fertilizer Technology Center for the Asian and Pacific Region, No. 27, Bangkok, Thailand. p 293-302. Pandutama MH. 1996. Predicting change in residual extractable P for site-specific management in paddy soils in Indonesia. PhD dissertation. University of Hawaii, Hawaii. Partohardjono S, Sri Adiningsih J. 1991. Response of food crops to phosphate rocks. Indones. Agric. Res. Dev. J. 13:46-57. Prasetyo BH, Sulaeman, Sri Mulyani N. 1997. Red-yellow soils from Kotabumi, Lampung: their characteristics, classification, and utilization. Indones. J. Crop Sci. 12:37-45. Ruddle K. 1987. The impact of wetland reclamation. In: Wolman MG, Fournier FGA, editors. Land transformation in agriculture. SCOPE Series No. 32. Paris (France): John Wiley and Sons. p 171-201. Santoso D. 1991. Agricultural land of Indonesia. Indones. Agric. Res. Dev. J. 13:33-36. Schmidt DR, Casler MD, Saefuddin A. 1990. Crop response to lime and phosphorus on new bench terraces in the tropics. Agron. J. 82:333-337. Soil Research Institute. 1979. Soil survey of the irrigation project area, Sitiung-S. Jujuhan (West Sumatra). Ministry of Agriculture, Agency for Agricultural Research and Devel- opment, Bogor, Indonesia. Sri Adiningsih J, Santoso D, Sudjadi M. 1990. The status of N, P, K and S of lowland rice soils in Java. In: Blair G, Lefroy R, editors. Sulfur fertilizer policy for lowland and upland rice cropping systems in Indonesia, Jakarta, Indonesia, 18-20 July 1989. Australian Centre for International Agricultural Research (ACIAR) Proceedings No. 29. Canberra (Aus- tralia): ACIAR. p 68-76. Sri Adiningsih J, Semali A, Effendi S, Hadiwigeno S. 1991. Resources and problems associated with the development of upland areas in Indonesia. In: Blair G, Lefroy R, editors. Technologies for sustainable agriculture on marginal uplands in Southeast Asia. Ter- nate, Cavite, Philippines, 10-14 Dec 1990. Australian Centre for International Agricul- tural Research Proceedings, No. 33, Canberra, Australia. p 45-54. Sri Rochayati, Muljadi D, Sri Adiningsih J. 1990. Penelitian efisiensi penggunaan pupuk di lahan sawah. In: Prosiding Lokarkya nasional efisiensi penggunaan pupuk. Pusat Penelitian Tanah dan Agroklimat, Cisarua. p 107-143. Sudjadi M. 1984. Problem soils in Indonesia and their management. In: Bay-Petersen J, editor. Ecology and management of problem soils in Asia. Bangkok (Thailand): Food and Fer- tilizer Technology Center for the Asian and Pacific Region, No. 27. p 58-73. Trangmar BB, Yost RS, Sudjadi M, Soekardi M, Uehara G. 1984. Regional variation of selected topsoil properties in Sitiung, West Sumatra, Indonesia. Hawaii Institute of Tropi- cal Agriculture and Human Resources, University of Hawaii. Widjaja-Adhi IPG, Karama AS. 1994. Problems related with the sustainable development of coastal plains in Indonesia: research needs and related priorities. Indones. Agric. Res. Dev. J. 16:37-47.

Characterizing soil phosphorus and potassium status . . . 189 Widjaja-Adhi IPG, Suriadikarta DA, Karama AS. 1996. Reclamation and management of salt- affected acid sulphate soils for agricultural development. Indones. Agric. Res. Dev. J. 18:37-43. Wihardjaka A, Kirk GJD, Abdulrachman S, Mamaril CP. 1998. Potassium balances in rainfed lowland rice on a light-textured soil. In: Ladha JK, Wade L, Dobermann A, Reichardt W, Kirk GJD, Piggin C, editors. Rainfed lowland rice: advances in nutrient management research. Proceedings of the International Workshop on Nutrient Research in Rainfed Lowlands, Ubon Ratchathani, Thailand, 12-15 Oct 1998. Manila (Philippines): Interna- tional Rice Research Institute. p 127-137.

Notes Authors’ addresses: A. Clough, S. Fukai, School of Land and Food Sciences, The University of Queensland, St. Lucia 4072, Australia; I.P.G. Widjaja-Adhi, J. Sri Adiningsih, A. Kasno, Centre for Soil and Agroclimate Research, Jalan Ir. H. Juanda 98, Bogor 15123, Indone- sia. Acknowledgments: The authors wish to acknowledge the financial support provided by the Australian Research Council (ARC) and Pivot Ltd. for the Acid Soils Project conducted by CSAR and The University of Queensland. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

190 Clough et al Planning and managing rice farming through environmental analysis

K. Borkakati, V.P. Singh, A.N. Singh, R.K. Singh, A.S.R.A.S. Sastri, and S.K. Mohanty

Environmental analysis provides insights into planning and managing rice farming according to the prevailing conditions. This helps in developing situation-specific technologies and in selecting areas suitable for the application of promising technologies, particularly in rainfed conditions. Studies carried out in eastern India characterized and classified the rainfed rice area in different rice ecosystems and subecosystems and elucidated the main problems and opportunities for enhanced productivity in them. A detailed analysis of the principal rice subecosystems provided the periods and quantities of water surplus and deficit and moisture use and recharge patterns along with other climatic variables during the rice-cropping season. This information was used for selecting the technological interventions that were considered suitable for managing rice farming in such areas. The interventions were compared with the farmers’ normal practices by monitoring crop performance at selected locations in two rice ecosystems.

Rice environmental analysis is carried out for various purposes. It provides useful information on what the environments are, how much rice area there is, and how production is distributed across environments. It is supposed to enhance the resource- use efficiency and impact of technologies. It provides insights into planning and managing rice farming and selecting areas suitable for the application of promising technologies. This chapter reviews some of the environmental analysis that has been carried out in the rainfed regions of eastern India. The first part presents the ecosystem characterization from the mega to micro level and the classification of rice areas into broad ecosystem/subecosystem categories. The next part outlines the main factors causing low rice productivity and cropping intensity in the principal rainfed subecosystems and the strategies developed for addressing this and other related issues. The following section presents selected case study results of the on-farm research conducted for developing promising technologies on the basis of environmen-

Planning and managing rice farming through environmental analysis 191 tal analysis for rainfed uplands and rainfed lowlands (drought-prone, submergence- prone, and drought- and submergence-prone). The final section draws conclusions on the planning and management of rice farming in the setting of an agroecological framework.

Ecosystem characterization from the mega to micro level in eastern India An ecosystems analysis for research prioritization at different levels throughout eastern India has been done collaboratively by the Indian Council for Agricultural Re- search, state agricultural universities, departments of agriculture, and the Interna- tional Rice Research Institute (IRRI 1992).

Mega-level analysis India’s rice-growing area occupies 42 million ha. A mega-level analysis indicated that, although rice yields in northern and southern India have increased rapidly in recent years, yields remained basically stagnant in eastern India (except in West Ben- gal, which experienced rapid growth in recent years). The eastern India region, comprising Assam, Bihar, West Bengal, Orissa, and the eastern parts of Uttar Pradesh and Madhya Pradesh, is the largest rice-growing region in the country and accounts for about 67% (26.8 million ha) of India’s rice area. In five of the six eastern states, average rice yield (1.8 t ha–1) is below the national average (2.7 t ha–1). About 80% of rice farming in the region is rainfed. Rainfall is moderate to high, and is limited to a short period. This results in drought in the uplands and flooding in the lowlands. Sastri and Singh (this volume) give a more detailed hydrological account of eastern India. Eastern India is the priority region for research because of its large rice area and low and stagnant rice yields. It is also a priority region because about half of the country’s population of one billion people live here and are largely dependent on rice farming.

Macro-level analysis The macro-level analysis of rice-growing ecosystems in eastern India revealed that only 21.2% (5.69 million ha) of the 26.8-million-ha rice area is irrigated (IRRI 1992). About 16.4% (4.38 million ha) is upland, 47.8% (12.78 million ha) is rainfed lowland (0–50-cm water depth), and the remaining 14.7% (3.95 million ha) is under the deepwater (50–100-cm water depth) or very deep water (>100-cm water depth) ecosystem category. For the rainfed lowland ecosystem, about 83% (10.6 million ha) has a shallow water depth (0–30 cm) and 17% (2.2 million ha) has a medium water depth (30–50 cm) during the rice-growing season. Analysis of drought and flooding patterns, water balance, selected land characteristics, and length of the growing season in the shallow rainfed lowland ecosystem showed that 54.6% (5.7 million ha) is drought-prone, 25.5% (2.7 million ha) is drought- and submergence-prone, 10.3% (1.1 million ha) is submergence-prone, and 9.6% (1.0 million ha) is favorable. The entire area in the medium-depth category of the rainfed

192 Borkakati et al lowlands is submergence-prone. There is a wide fluctuation in the extent of the deepwater ecosystem and the medium-depth category of rainfed lowlands, depending on rainfall pattern and amount, and onset and cessation of the monsoon. Rice yields in all the rainfed ecosystems are low and vary greatly from year to year. Yields in the irrigated areas average 3.2 t ha–1. Yield is 0.6–1.5 t ha–1 in the uplands, 0.9–2.4 t ha–1 in the rainfed lowlands, and 0.9–2.0 t ha–1 in the deepwater and very deep water areas. The ecosystem with first priority in eastern India is the rainfed lowlands because of its area, larger dependent population, and potential for yield increase.

Meso-level analysis A meso-level analysis of rainfed rice ecosystems was conducted in several districts of eastern India, such as Bahraich (Singh and Pathak 1990, IRRI 1992) and Faizabad (IRRI 1992) of Uttar Pradesh, Hazaribagh (IRRI 1993) of Bihar, and Raipur (IRRI 1998, Singh et al 1999) of Madhya Pradesh. Characterization of the rice environments of Faizabad District (total area of 451,100 ha and rice area of about 181,000 ha) was done using satellite remotely sensed data, selective field checks, and auxiliary data (Singh and Singh 1996, Singh 1996). Maps (1:250,000 scale) were prepared to delineate physiographic units, land-use patterns, soils, flooding, and drought. In- formation on climate, groundwater, irrigation sources, landholding, and input use was integrated with the maps. The classification of rainfed rice environments showed that about 40% of the area in Faizabad District (Fig. 1) is favorable rainfed lowland, 51% is drought-prone

Rainfed shallow drought-prone Rainfed shallow submergence-prone Rainfed shallow favorable Pond/lake/river District headquarters

0 10 20 30 km

Fig. 1. Rainfed lowland rice-growing subecosystems in Faizabad, Uttar Pradesh, India.

Planning and managing rice farming through environmental analysis 193 lowland, 2% is submergence-prone lowland, and 4% is submergence- and drought- prone lowland. Apart from drought and submergence, soil sodicity was identified as another priority research area in the district. Similarly, a detailed meso-level analysis was done in the Masodha block of Faizabad District (IRRI 1992). The block covers about 21,000 ha of total land area and has about 8,000 ha of rice area. Rice-growing environments in terms of physiography, land use, soils, flooding, drought, groundwater, and irrigation were studied in detail using remote sensing and conventional data. The major part of the block is classified as a shallow favorable rainfed lowland rice subecosystem. This analysis showed that about 14% of the block area is affected by flooding, 10% by sodicity, and 2% by waterlogging. Only 32% of the groundwater potential has been developed so far. Recharge-draft analysis showed that about 16,000 ha-m of groundwater is still available for irrigation.

Micro-level analysis Within each of the upland, rainfed lowland, and deepwater ecosystems in eastern India, target environments were characterized at the micro level to set research priorities within and among the dominant farming systems. More than 100 sites were analyzed. Rapid rural appraisal techniques, which included agroecosystems mapping and diagnostic surveys, were employed at all sites (IRRI 1990). The analysis focused on spatial, temporal, resource flow, and decision patterns. The methodology involved a two-tier training program for researchers on how to set priorities using agroecosystems analysis. The problem diagnosis and research prioritization at this level were conducted by multidisciplinary teams, with continuous involvement and interactions from groups of farmers. At all sites, the static factors studied were land types, land use, source of water supply, and soil properties, as described in Singh et al (1993). The dynamic factors were rainfall and field-water depth; cropping pattern and crop calendars, crop yields, varieties and management practices, insects, diseases, and weeds; production costs and returns; and labor supply pattern, income distribution, landholding size, and demography by social class and gender. The geographic area was zoned into agroecosystems and the problems and opportunities elucidated in each major agroecosystem (Singh et al 1993). The highest priority was given to the agroecosystem with the largest rice area. The research programs were then prioritized on the basis of the physical extent (coverage); number of affected households; complexity, severity (crop-loss estimates), and frequency of problem occurrence; importance of the affected enterprise in the farming system; and the farmers’ perceptions of the problem. All site studies within each subecosystem and ecosystem were pooled and compared to identify the commonality of problems and opportunities. This provided an empirical picture of the entire ecosystem, which served as a basis in formulating a need-based research agenda for developing appropriate technologies for the specific

194 Borkakati et al situations and the allocation of resources at the national, regional, and zonal level (IRRI 1992).

Constraints to rice productivity and strategies for addressing the major issues The constraints to rice productivity in eastern India vary from state to state and even from area to area. The major constraints to higher rice productivity in different rice ecosystems are related to hydrology (moisture stress and flooding), soil and nutrient management, the availability of situation-specific improved varieties and high-quality seed, insect, disease, and weed management, crop establishment, and other specific technologies. These constraints can be listed as follows: 1. Moisture stress due to erratic and often inadequate rainfall, high runoff, poor soils, and lack of facilities for rainwater and soil moisture conservation/ supplementary (life-saving) irrigation (upland and drought-prone rainfed lowlands). 2. Intermittent moisture stress due to low and erratic rainfall; poor soils as in Madhya Pradesh, Orissa, and some parts of Uttar Pradesh; and flash floods and waterlogging/submergence due to poor drainage, low-lying physiography, and high rainfall in submergence-prone lowlands, as in Assam, West Bengal, and north Bihar. Accumulation of toxic decomposition products in ill-drained soils and soil reduction, encouraging problems of iron toxicity in Assam. 3. Continuous use of traditional varieties because of the nonavailability of seeds and farmers’ lack of awareness about high-yielding varieties (uplands, rainfed lowlands, and deepwater areas). 4. Low soil fertility due to soil erosion, leading to losses of soil nutrients and low and imbalanced use of fertilizers in uplands, and to the nonavailability of a suitable method for applying the fertilizer in standing water in rainfed lowland, semideep, and deepwater areas. 5. Heavy infestation of weeds and insect pests such as blast and brown spot and poor attention to their timely control (uplands and rainfed lowlands). 6. Poor crop stand establishment because of broadcast seeding, resulting in uneven germination (upland and direct-seeded lowlands); and delay in monsoon onset, often leading to delayed and prolonged transplanting and subop- timum plant population (mostly in rainfed lowlands). 7. Poor adoption of improved crop production technologies because of technology inappropriateness and economic backwardness of the farmers (uplands and lowlands). Strategies to increase rice productivity and cropping intensity in eastern India mainly included the following: 1. Adoption of runoff rainwater management practices suited to the conditions of individual farm holdings as well as the watershed as a whole, thus moti- vating farmers to provide life-saving irrigation to the crop during long dry spells.

Planning and managing rice farming through environmental analysis 195 2. Emphasis on balanced use of plant nutrients along with the popularization of integrated nutrient management approaches and methods of applying required nutrients in standing water, such as deep placement of urea super granules. 3. Ensuring timely and adequate availability of inputs such as seeds, fertilizers, and credit to farmers. In this regard, multiplication of seeds of promising high-yielding varieties for specific ecologies, such as the rainfed uplands and drought-prone or submergence-prone lowlands with different water depth and deep water, by the seed-producing agencies and making them available to farmers can play a significant role in enhancing the productivity of rainfed rice. 4. Promotion of an integrated pest management approach for the control of insect pests, diseases, and weeds. 5. In upland rice areas, line sowing may be popularized through suitable seeding devices to establish the desired level of plant population, ease in weed control, and the application of other management technologies. 6. Encouraging the use of improved farm implements for effective and timely field operations. 7. Organization of field demonstrations of improved technological packages in specific situations and training of farmers for effective transfer of newly developed crop production technologies. Suitable technological packages for different ecosystems by states in eastern Indian are described in Singh and Singh (2000).

Rainfed rice-farming systems technologies The main environmental stresses identified in eastern India are drought, submergence, flash flooding, and stagnant deepwater situations. In each of these conditions, technology development aspects considered cropping systems, improved genotypes, water, nutrients, weeds, and other insect pest management options in an integrated manner. In addition, possibilities were also explored for developing water resources, groundwater and rainwater management for drought-prone areas, and drainage options and water recycling in submergence-prone areas. Such strategies were adopted throughout eastern India by all the participating centers in upland, rainfed lowland, and deepwater ecosystems.

Technologies for rainfed uplands Integrated development of technologies for rainfed rice-farming systems in drought- prone uplands. The agroecological analysis of upland rice areas in Hazaribagh, Bihar, indicated that drought was one of the major causes of low and unstable rice yields and low cropping intensity. To develop drought management strategies, drought was characterized according to duration and severity by analyzing meteorological data from 1913 to 1987 from the India Meteorological Department and the Soil Conservation Research and Demonstration Farm at Hazaribagh. The mean dates of onset and termination of effective monsoon, duration of the monsoon period, and the probability

196 Borkakati et al estimation of drought were determined from the daily rainfall data from these 75 years. The detailed analysis of drought was done for a 25-y period by following the Universal Hydrologic Equation to arrive at the weekly water balance (rainfall-potential evapotranspiration). Because of the lack of data, however, the analysis did not take into account the field-stored moisture. The drought analysis was done for the period between the 22nd and 43rd wk of the year (28 May to 28 Oct, 154 d duration), as upland rice crop cultivation and the monsoon in this region are confined to this period. During the rice cultivation period, the duration and probability of drought occurrence were analyzed separately for three groups of 7 wk each: 22nd-28th wk, 29th-35th wk, and 36th-42nd wk, termed as initial, intermediate, and terminal stages of rice growth. Based on drought occurrence and severity analysis, on-station and on- farm experiments were conducted on adjusting seeding dates to avoid drought in the latter part of the growing season, selection of short-duration rice cultivars (similar to traditional ones) suitable for drought situations, and weed management and sowing methods. Table 1 presents the results of the analyses, which indicate a higher probability of drought occurrence at the initial and terminal stages than at the intermediate stage of the crop. Drought affecting the crop at the initial and terminal stages of growth,

Table 1. Selected monsoon and drought characteristics at Hazaribagh, Bihar, India.

Monsoon characteristicsa Amount or time

Normal annual rainfall (mm) 1,299 Normal seasonal rainfall (mm) 1,169 Probable annual rainfall (mm) At 99% probability 759 At 20% probability 1,385 Probable seasonal rainfall (mm) At 99% probability 666 At 20% probability 1,385 Monsoon onset (date) 18 June (earliest 2 June, latest 6 July) Monsoon termination (date) 11 Oct (earliest 9 Sept, latest 28 Oct) Monsoon duration (d) 117 (shortest 77, longest 137) Seasonal drought durationb (wk) Initial stage (28 May to 3.75 15 July) Intermediate stage (16 July 2.00 to 17 Sept) Terminal stage (18 Sept 4.55 to 28 Oct)

aPeriod of analysis from 1913 to 1987 using rainfall data. bPeriod of analysis from 1972 to 1991 using rainfall and evapotranspiration data.

Planning and managing rice farming through environmental analysis 197 however, depended on sowing time and the duration of the varieties used. Almost all farms in the region experienced drought at the intermediate stage. Farmers in this region normally do final land preparation and sow rice in the 3rd week of June after the onset of effective monsoon. This results in an avoidable loss of initial monsoon moisture to some extent and in delayed sowing in years of continuous rains because of poor workability of the soil. In such cases, crops are more likely to suffer from drought at the terminal stages of growth. Advance sowing before the onset of monsoon was therefore considered as one of the strategies for avoiding or minimizing the effects of terminal drought, which is of longer duration and has a higher probability of occurrence (Table 1). The other strategy in this respect was to use the selected rice cultivars of short maturity duration (90–100 d) that had shown consistently better performance in previous on-station and on-farm experiments. The results of advance sowing, done on around 7 June before the onset of rains, indicated significantly higher grain yield, number of panicles m–2, number of fertile spikelets panicle–1, and taller plants in all the genotypes tested than sowing done around 22 June after the onset of rain (normal practice) (Table 2). Among the genotypes, Brown Gora and Kalinga III were inferior to RR167-982 (Vandana) and RR165-1160 in all respects including panicle weight and panicle length (Table 2). The early sown plots, however, had a higher weed infestation at the early stages of plant growth owing to the simultaneous emergence of weeds with rice upon the first rain showers. In normal sown plots, the weeds had germinated before rice seeding and were destroyed with the additional harrowing. Therefore, weed management was followed along with advance sowing in successive experiments that also included seeding method and rice plant population studies. With advance sowing, grasses dominated the weed flora in the initial stages of crop growth followed by dicotyledons in the latter stages. The bulk of the weed flora constituted Cyperus rotundus, Cyperus iria, Echinochloa colona, Cynodon dactylon, Setaria glauca, Commelina benghalensis, Aeschynomene indica, and Brachiaria ramosa, which competed with rice during all its growth stages. The population of Ageratum conyzoides was severe at the reproductive stage only. Losses in rice grain yield (difference between a weed-free and weedy plot) were as high as 77%, and were higher in drilled than broadcast-seeded crops. However, when manually weeded at 20 and 40 days after sowing, the drilled crop produced significantly higher yield (2.4 t ha–1) than the manually weeded broadcast crop (2.0 t ha–1). Application of butachlor with one handweeding was as effective as two handweedings in both systems. The initial 4-wk period of crop growth was crucial for weeding in the case of early maturing tall genotypes, such as Kalinga III, having poor early vigor in contrast with the semitall genotypes, such as RR167-982 (Vandana) and Brown Gora, that have higher early vigor and a suppressive effect on weeds. Plant population, attained using a seeding rate of 500 seeds m–2 or 112.5 kg ha–1, and closer row spacing (20 cm) also had a suppressive effect on weeds, as reflected by lower weight of total weed dry matter. Rice grain and straw yields were higher under these practices than using a higher or lower seeding rate (400 or 600 seeds m–2) and sowing by broadcast or with wider row spacing (30 cm).

198 Borkakati et al Table 2. Effect of date of sowing, weed management, seed rate, and sowing method on the growth and yield of upland rice, Hazaribagh, Bihar, India.

Treatment Grain yield Plant height Panicles Fertile spikelets (t ha–1) (cm) (no. m–2) (no. panicle–1)

Date of sowing (variety RR167-982)a 7 June 2.2 87.1 325 72.4 22 June 1.8 77.9 309 65.4

Cultivarsb Grain Plant Panicle Panicle Filled 1,000-grain (sown on 7 June) yield height length weight grains (no. weight (g) (t ha–1) (cm) (cm) (g) panicle–1)

Brown Gora (farmer’s var.) 1.2 111 16.4 1.49 41 28.5 Kalinga III 1.5 106 18.9 1.50 55 22.7 RR167-982 (Vandana) 2.5 117 19.5 2.17 76 24.2 RR165-1160 2.1 106 23.8 2.50 97 21.3

Weed managementc (variety RR167-98, sown on 7 June) Grain yield (t ha–1)

Farmer’s method in broadcast 1.02 Weedy check broadcast 0.75 Weedy check drilled 0.54 Weed-free broadcast (handweeded) 1.97 Weed-free drilled (handweeded) 2.37 Butachlor + 1 handweeding (30 d after sowing) 2.01

Seed rateb Grain yield Sowing method Grain yield (t ha–1) (variety RR167-982, early sown on 7 June) (t ha–1)

400 seeds m–2 2.00 Broadcast 1.8 (90 kg ha–1) 500 seeds m–2 2.25 Sowing behind plow 2.2 (112.4 kg ha–1) (20-cm rows) 600 seeds m–2 2.10 Sowing behind plow 1.7 (135 kg ha–1) (30-cm rows) aMean of three kharif (wet) seasons, 1989-91. bMean of two kharif seasons, 1991-92. cMean of three kharif seasons, 1990-92.

Sowing in furrows at 20-cm spacing behind the plow using a rate of 112.5 kg seed ha–1 was superior to broadcast not only for ease in weed control but also for combating drought effects, especially at the initial stages. Furrow sowing allowed seeds to be placed at 3–5-cm depth, thus encouraging better root development and exploitation of soil moisture. Soil moisture retention and release characteristics also support this and indicate that the upper 5-cm layer of these soils dried up quickly even when their subsoil layers had adequate moisture. Sowing in furrows also protected seeds from desiccation and bird damage and required a lesser amount of seed than that used by farmers in broadcasting (150–200 kg ha–1).

Planning and managing rice farming through environmental analysis 199 Several farmers in the on-farm study villages have shifted to furrow sowing from broadcast and have improvised the “country plow” by removing its iron shear to open only 5–6-cm-deep furrows. They have also replaced traditional cultivars with RR167-982 (Vandana) and have opted to advance seeding to before the onset of monsoon. Managing rainwater for stable rice yields and improved cropping intensity. Rainfed uplands in eastern India cover several million ha and are contiguous in the Chotanagpur plateau region. Because of low and concentrated rainfall (1,130 mm from June to October), these lands are traditionally cultivated only in the wet season to grow a crop of millet-blackgram-rice-fallow in a 4-y crop rotation cycle. These areas generally remain fallow during the dry season because of the lack of soil moisture and unavailability of irrigation water. Analysis of the existing farming systems and hydrological mapping of the area indicated higher potentials for rainwater collection in valley check reservoirs that can be used to supplement the water requirement of rice during drought spells and for irrigating short-duration dry-season crops. The farming systems analysis included land, labor, and cash resources; land types and their use pattern; and crop calendars with respect to rainfall pattern and labor availability and use. Hydrological mapping included studies on available water resources, surface and subsurface hydrology, water losses and recharge characteristics of the soil, occurrence and severity of drought, and climatic factors (Paul and Tiwari 1994). The possibility of on-farm rainwater collection and management was explored at Handio village in Hazaribagh. Starting with two farmers as pilot cases, villagers constructed 16 rainwater-holding structures (valley check reservoirs and dug ponds) ranging from 1.75 ha in the uppermost toposequence with a large runoff catchment area to 0.25 ha in the lower lands along the slope (Singh et al 1993). The construction of tanks was done through family and communal labor on an exchange basis during the summer months as sufficient labor was available in this period. The catchment to storage ratio in terms of land surface area varied from 10:1 to 6:1 depending on the general topography and micro relief of the area. On average, the volume of earth dug and moved amounted to 0.6 to 1.0 m3 capita–1 d–1. The dug soil was placed on the dikes and compacted by the respective farm families themselves. The volume of available water in these structures varied from 9,100 m3 to 16,900 m3 for 7 to 10 mo (July to April). In addition to the rainwater-holding structures, the farmers have also dug and constructed 37 cement wells in their respective landholdings with government support. The water from tanks is distributed for irrigation by gravity through cement culverts placed at different depths from the water surface and connected to earthen channels at different elevations. This allows farmers to draw water at different levels in the tank without cutting the dike or siphoning. It also reduces maintenance of dikes. With these practices, cropping intensity in the village has increased from <100% to 160% in the upper toposequence and from 150% to almost 250% in the lower lands (IRRI 1992). The increase in cropping intensity was achieved through vegetable cultivation during the dry season that includes table pea-potato + coriander and chillies

200 Borkakati et al from October to February and a summer crop of tomato for some farmers (Singh et al 1993). All farmers continue to grow rice, blackgram/pigeonpea, and some vegetables during the wet season with stable yields. Of the various crop sequences, rice-table pea or niger-potato + coriander-tomato are the most common in the village. With the availability of water, some farmers have started growing fish (common carp) at a low stocking density (6,000 ha–1). Results with older fingerlings of heavier weight (50 g piece–1 or more) are reported to be more promising. With the increase in cropping intensity, farmers in the village have also started using animal manure in vegetable production and stall feeding of livestock, a practice not followed earlier because of difficulties in dung collection from open land grazing at distant places. Managing groundwater resources to alleviate drought effects. Drought is a major cause of low and unstable yields and low cropping intensity in Hazaribagh, Bihar. To alleviate drought effects, farmers in this area adopt various cropping and agronomic practices and explore ways to provide supplemental irrigation. They dig shallow wells and construct rainwater-holding structures (on-farm reservoirs and tanks). The water- supplying capacity of the wells, however, is highly variable in the district, depending on the local recharge capacity and the draw of water. Some wells remain productive throughout the year, whereas others dry up after a few months. It is also not possible to have productive wells in all places. The State Remote Sensing Application Center (IRRI 1993) prepared a groundwater potential map for Hazaribagh covering a geographical area of 1,116,500 ha and divided the district into four groundwater zones: poor, poor to moderate, moderate, and moderate to good. The map provided only a qualitative assessment and could not be used directly for further groundwater exploitation purposes because no information on cropped area was available on the map. The local unit of the Central Ground- water Board had calculated the net groundwater recharge and draft in the district, taking the block as a unit, but it did not provide spatial information on the promising areas for groundwater exploitation within the block. A study was therefore carried out using satellite remote sensing to generate information on rice and other crop lands and integrate it with the available groundwater information and block and village boundary maps on a geographic information system (GIS) to identify the promising cropland areas for further groundwater exploitation. A crop area map of the district was prepared based on the interpretation of Indian Remote Sensing Satellite (IRS-IA) images of three different years (1988, 1991, and 1992). Cropped area analysis showed that approximately 48% of the district area is under agriculture and about 49% is under forest. Nearly all agricultural land (532,550 ha) is cropped during the kharif (rainy) season. About 25,620 ha of land are cropped during the rabi (dry) season. Rice is the dominant kharif crop. Cropping intensity in the district varies from 105% to 116%. Integration of the crop-land map, groundwater zone map, and block boundary map resulted in a map showing the distribution of groundwater potential in crop lands of different blocks. Moderate and moderate-to-good groundwater potential zones were merged, since these were considered promising from the viewpoint of groundwater

Planning and managing rice farming through environmental analysis 201 Sargawan

Domchanch Markacho Hunterganj Chauparan

Partabpur Chatra Itkhori Barhi

Simaria Ichak Hazaribagh Bishungarh

Keredari Churchu Barkagan Tandwa

Mandu

Crop lands with moderate to good groundwater potential Block boundary Patratu Gola

Ramgarh

Fig. 2. Promising area for groundwater exploitation in crop lands under different blocks of Hazaribagh District.

exploitation. Poor and poor-to-moderate groundwater zones were excluded due to their low potential. The final map (Fig. 2) shows promising areas for groundwater use in crop lands of different blocks. The extent of these areas was then calculated (Table 3). The calculations showed that, out of 24 blocks of the district, 11—Chatra, Barhi, Churchu, Patratu, Ramgarh, Gola, Mandu, Barkatha, Jainagar, Markacho, and Kodarma—had more than half of their crop-land area as promising for groundwater exploitation. The remaining 13 blocks showed less than half of their crop lands as promising. The data on annual groundwater recharge, draft, and available balance in different blocks (Table 3) showed that the annual recharge varies from 8.5 million m3 in Jainagar to 51.0 million m3 in Simaria. The net groundwater draft is low, ranging from 0.8 million m3 in Churchu to 7.1 million m3 in Hunterganj. The draft is high, however, in Hazaribagh (12.1 million m3). Since the geographical area of different blocks varied considerably, from 228 km2 in Jainagar to 976 km2 in Simaria, the available groundwater balance was averaged by calculating the balance in each km2- area of the block. Overlaying this information on the promising groundwater area

202 Borkakati et al Table 3. Promising areas for groundwater exploitation in different blocks of Hazaribagh District, Bihar, India (1992-93).

% Net annual Total crop land recharge Net draft Groundwater Block crop land with available (million m3) balance (ha) promising (million (m3 km–2) groundwater m3)

Pratappur 18,073 13 28.5 1.2 40,700 Hunterganj 21,171 1 28.8 7.1 42.300 Chatra 21,751 58 36.4 2.5 53,100 Simaria 26,432 43 51.0 2.3 50,200 Tandwa 19,659 13 19.6 1.3 48,800 Keredari 14,916 26 20.1 1.0 44,000 Barkagan 18,385 19 17.1 1.5 34,900 Katkamsandi 21,107 26 17.7 3.5 30,600 Itkhori 25,317 16 27.1 2.8 45,600 Chauparan 24,433 40 33.8 4.2 44,333 Barhi 20,603 98 27.2 1.9 52,300 Ichak 17,038 10 17.2 1.6 39,600 Hazaribagh 15,961 2 17.3 12.1 16,400 Churchu 17,716 74 20.2 0.8 46,500 Patratu 16,412 82 13.0 2.0 35,100 Ramgarh 22,396 97 11.7 5.2 21,300 Gola 25,580 80 18.5 2.6 47,300 Mandu 22,634 96 28.6 3.9 57,200 Bishungarh 25,524 2 25.0 2.8 42,700 Barkatha 20,432 97 14.2 1.3 30,000 Jainagar 19,391 100 8.5 1.8 29,400 Markacho 17,828 95 14.7 1.0 43,200 Kodarma 18,546 74 19.6 4.0 33,900 Satgawan 6,873 nil 13.5 1.6 39,300

map showed that 10 of the 11 promising blocks (except Ramgarh) had a net balance of more than 30,000 m3 km–2 and are suitable for further groundwater exploitation in crop-land areas as identified. Overlaying village boundaries on the block map led to the identification of promising villages for groundwater exploitation. Possibilities for developing surface irrigation sources were indicated in the remaining 14 blocks of the district.

Technologies for the drought-prone rainfed lowland ecosystem At Raipur, which represents the drought-prone lowland ecosystem, the duration of humid (rainfall (R) > potential evapotranspiration (PE)) and moist (PE>R>PE/2) periods is 151 d (Sastri and Singh, this volume). This suggest that, if an early or medium-duration (110–125 d) rice crop is grown, especially in heavy soils, a second crop of chickpea or lathyrus or linseed can be grown with conserved moisture. These crops, unlike rice, can also thrive even in a submoist (PE/2>R>PE/4) period.

Planning and managing rice farming through environmental analysis 203 Table 4. Average data of the double-cropping experiment.

Crop sequence Mean grain yield (t ha–1)

Wet Winter Wet- Winter- Total valuea of season season season season produce (Rs ha–1) crop crop

Early rice Chickpea 2.3 0.25 16,200 (Poornima) Lathyrus 2.3 0.12 14,400 Linseed 2.3 0.05 14,460

Medium rice Chickpea 3.2 0.13 20,760 (Mahamaya) Lathyrus 3.2 0.07 19,980 Linseed 3.2 0.02 19,800

aPrices for rice: Rs6,000 t–1; lathyrus: Rs8,000 t–1; chickpea: Rs10,000 t–1; linseed: Rs15,000 t–1.

Table 5. Economics of rice-chickpea crop sequence at Raipur, Madhya Pradesh.

Mean yield (t ha–1) Gross Year Crop sequence incomeb Rice Chickpea (Rs ha–1)

1995-96 Early rice-chickpea 3.1 0.42 23,040 Medium rice-chickpea 4.5 0.25 29,616

1996-97 Early rice-chickpea 1.6 0.00a 9,720 Medium rice-chickpea 2.5 0.00a 14,820

1997-98 Early rice-chickpea 2.3 0.27 16,500 Medium rice-chickpea 3.4 0.06 20,940

1998-99 Early rice-gram 2.1 0.33 15,660 Medium rice-gram 2.5 0.22 17,760

Mean Early rice-gram 2.3 0.26 16,230 Medium rice-gram 3.2 0.13 20,784

aCrop could not be established because of inadequacy of moisture. bPrices for rice Rs6,000 t–1 and chickpea Rs10,000 t–1.

With this hypothesis, on-farm experiments were conducted in nine farmers’ fields in black soils from 1995-96 to 1998-99 (4 years) in Tarpongi village about 30 km north of Raipur. Second crops of chickpea, lathyrus, and linseed were grown after early (var. Poornima) and medium-duration (var. Mahamaya) rice varieties. Tables 4 and 5 show the results of the four experiments, which indicate that

204 Borkakati et al 1. There are possibilities of growing a second crop after rice with the conserved moisture and nutrients. 2. Though the productivity of a second crop (chickpea or linseed or lathyrus) is higher in short-duration rice fields than in medium-duration rice fields, the overall rice equivalent productivity of the crop sequence (total cash value of produce from both the crops, based on prevailing price) is higher in the case of medium-duration rice followed by a second crop, indicating that this system is more suitable. 3. The rice equivalent productivity (cash value) of chickpea produce is higher than that of either lathyrus or linseed. Table 5 indicates that in general chickpea yield was higher in the short-duration rice-chickpea crop sequence, but the average total income in this sequence is Rs16,230, whereas, in the medium-duration rice-chickpea crop sequence, the average total income is Rs20,784. It is therefore recommended that, in view of the higher production potential, the medium-duration rice-chickpea system should be followed in heavy (black) soils of the drought-prone lowland rice ecosystem. Year-to-year fluctuations in the productivity of both crops occur, however. During 1996-97, which happened to be a severe drought year, the chickpea crop could not be sown due to dry soil conditions and the rice yields were the lowest of the four years. The period 1995-96 was a good rainfall year and both the chickpea and rice yields were the highest of the four years. A risk analysis on the cropping systems over time is essential in such cases. In drought-prone lowland ecosystems, the ill effects of drought can be allevi- ated with timely and proper crop management practices. In the on-farm experiments conducted at Raipur during 1995-96 to 1998-99, the highest, lowest, and median rice productivity across fields and its standard deviation are shown in Table 6. Table 6 shows that, even in bad (drought) years such as 1996, the highest yields of rice varieties Mahamaya and Poornima were 3.6 and 2.7 t ha–1, respectively, whereas their lowest yields during 1996 were 1.8 and 0.4 t ha–1, respectively. Such a variation in yields across fields is a clear indication of crop management differences, which in this case were mainly related to the timeliness of crop establishment, “beushening” (dry-sown crop then plowed and laddered 30–35 d after sowing with 10–15 cm of standing water), and weed control operations. The beneficial effects of proper management can also be realized even in good rainfall years such as 1995 (Table 6). An analysis of the dry weeks (weeks receiving less than 50 mm rainfall) indicates that 1999 had a continuous dry spell of 10 wk from 27 August to November, that is, from the 35th to the 44th standard meteorological week (SMW). This resulted in severe drought conditions. In light soils, the rice crop failed completely. In heavy soils, some farmers provided a small supplementary life-saving irrigation from the water collected in roadside ditches, which benefited those farmers considerably (Table 7).

Technologies for the submergence-prone rainfed lowland ecosystem The agroecological analysis of rainfed lowland rice areas in Jorhat District of Assam indicated that submergence proneness was one of the major causes of low and un-

Planning and managing rice farming through environmental analysis 205 Table 6. Rice productivity variations using two varieties at Raipur, Madhya Pradesh.

Rice Productivity (t ha–1) Standard Year variety deviation Highest Lowest Median (t ha–1)

1995 Mahamaya 5.8 3.3 4.5 1.1 Poornima 3.7 2.5 3.1 0.5

1996 Mahamaya 3.6 1.8 2.4 0.7 Poornima 2.7 0.4 1.6 0.8

1997 Mahamaya 4.4 2.5 3.4 0.7 Poornima 2.7 2.1 2.3 0.3

1998 Mahamaya 3.2 1.1 2.6 0.9 Poornima 2.8 0.8 2.1 0.9

Table 7. Rice grain yield (t ha–1) of a short- and medium-duration rice variety under severe drought conditions with and without supplementary irrigation (SI) in a heavy soila, Raipur, India, 1996.

Rice variety Grain yield (t ha–1)

Medium duration – with SI 2.57 – without SI 2.02

Mean 2.48 Early duration – with SI 1.87 – without SI 0.41 Mean 1.62

aThere was no grain yield in light soils for any of the rice varieties under any of the treatments. stable yields of rice. Farmers in these situations normally practice rice monocropping, mainly in the sali (winter) season. The two-rice cropping pattern is also followed in some villages, but to a limited extent. For the sali crop, seedbed preparation starts in June with the onset of monsoon and transplanting is done in July. But, because of floods in June, transplanting is delayed, which, on the recession of floodwater, continues up to August. Because of late planting, however, the flowering stage coincides with low temperature. As a result, grain filling is not completed, which leads to lower rice yields. Also in these situations, farmers use very low rates of fertilizer (15-5-0 N, –1 P2O5, K2O ha , respectively).

206 Borkakati et al mm mm

ABSoil moisture use Soil moisture accumulation Percolation Surplus water Standing water Rainfall Floodwater 450 PE 140 120 350 100 250 80 60 150 40 50 20 0 0 JFMAMJJASOND Jun Jul Aug Sep Oct Month Month Fig. 3. Climate water balance (A) and distributions of surplus water (B) at Jorhat, Assam, India.

To develop management strategies, a detailed analysis of climatic water balance was done for 18 years (1980-97). The depth of surface water in 80 well-distributed paddy fields in four selected villages (20 fields in each village) representing the district was measured twice a week from 2 June to 27 November (178 d duration) in 1997, as sali rice crop cultivation and the monsoon in this region are confined to this period. Sastri and Singh (this volume) give details of an agroclimatic inventory in Assam. Figure 3 shows the climatic water balance of bunded rice fields using average rainfall and PE values as inputs. This was computed using the bookkeeping procedure of Thornthwaite and Mather (1955). The following assumptions were made: 1. The water-holding capacity of the soils up to saturation is 300 mm. 2. From the surplus water that accumulates in the rice fields, percolation losses occur at the rate of 4 mm d–1, or 120 mm mo–1. 3. After percolation, the excess water is stored in the fields as standing water up to 50-mm depth in rice fields. 4. Any water depth above 50 mm is considered as floodwater. The soil moisture recharge in Jorhat, Assam, starts on 18 March and continues till mid-June (Fig. 3). From mid-June onward, percolation losses continue and standing water up to 50-mm depth remains till 20 October. The flood period begins at Jorhat after 15 June and ends by mid-August; its peak under normal conditions is around mid-July. The average amount of floodwater in rice fields (above standing water of 50 mm) is 112 mm. In all the villages, the water started accumulating from 4 June (24th SMW). Its depth fluctuated after 4 June and peaked on 13 August (33rd SMW) (Table 8). Then it declined and reached a minimum in October. The surface-water depth was directly related to rainfall, which, during the peak water-depth period, was the maximum (307.6 mm) of the season. Hydrology varied widely within the villages. Based on the sur-

Planning and managing rice farming through environmental analysis 207 face-water depth, rice subecosystems were delineated in each village. Mainly, three different hydrological situations were identified from the data with maximum water depth of 0–10, 10–30, and 30–50 cm during the crop season. The rainfed shallow, submergence-prone (10–30 cm) situation occupied more area than the other two hydrological (water depth) situations. Crop production strategies for such cases were adjusted for planting time, selection of rice varieties to accommodate two rice crops, and selection of rice varieties and fertilizer management according to the water-depth classes even within subecosystems. Increasing rice-cropping intensity. To increase cropping intensity, on-farm trials were conducted to accommodate two rice crops in the cropping pattern by selecting relatively higher-yielding and short-duration (early maturing) promising ahu (autumn) rice varieties for preflood cultivation. In this effort, some local varieties collected from different localities and some high-yielding varieties were evaluated against the popularly grown rice variety Rangadoria as the local check. Among the varieties tested, only Culture-1 matured earlier (88 d vs 96 d for the local check) and had around a 20% yield advantage over the local check variety (Table 9). Within the same maturity group (about 90 d), varieties such as Chilarain, Luit, and Lachit, developed by Assam Agricultural University, have been found to be very promising. Their average yield surpassed 4 t ha–1. They are now becoming popular among the farmers. Using these promising varieties, ahu (autumn) rice can generally be broadcast/ transplanted in March-April and harvested in June-July before floodwater enters the

Table 8. Surface-water deptha (in cm) in paddy fields in different hydrological situations, Assam, India, 1997.

Hydrological situation class Hydrological situation class Standard (cm) Standard (cm) meteorological meteorological week 0–10 10–30 30–50 week 0–10 10–30 30–50

24 1.2 2.8 10.8 37 6.2 8.5 34.3 25 1.2 2.8 17.2 38 4.3 6.2 31.8 26 1.8 3.4 18.2 39 2.3 4.8 24.6 27 2.1 4.2 19.6 40 0.0 2.1 22.5 28 2.2 4.4 20.2 41 0.0 2.0 15.2 29 2.0 4.0 21.4 42 0.0 2.5 15.8 30 2.3 4.5 22.5 43 0.0 1.8 12.8 31 4.5 6.8 28.5 44 0.0 1.0 8.2 32 8.8 12.5 40.6 45 1.2 0.0 6.2 33 8.6 14.2 43.5 46 0.0 0.0 5.8 34 6.5 11.3 40.2 47 0.0 0.0 2.0 35 6.2 12.2 40.8 48 0.0 0.0 0.0 36 7.4 9.6 42.2 49 0.0 0.0 0.0 aValues are means of 46 measurements (twice a week) from a total of 80 paddy fields from four villages (20 in each village).

208 Borkakati et al Table 9. Yielda of preflood ahu (autumn) rice (t ha–1) in the flood- prone ecosystem, Assam, 1998.

Variety Days to maturity Grain typeb Grain yield (t ha–1)

Gunilahi 100 SB 1.7 Megli 106 SB 2.4 Isahajay 100 SB 2.1 Culture-1 88 LB 2.3 Pusa 2-21 115 LB 2.3 Annada 115 LB 2.7 Rangadoria (local check) 96 SB 1.9

aMean of 8 replicates. bSB = short bold, LB = long bold. rice fields. After harvesting of ahu rice, sali (winter) rice can easily be accommodated from July-August and harvested in November-December. Management of inputs in different subecosystems (hydrological classes). Ex- periments were conducted with a semidwarf variety (Ranjit) and a tall variety (IET- –1 10016) with farmers’ usual fertilizer dose (15-5-0 kg N, P2O5, K2O ha , respectively), –1 and the recommended fertilizer dose (40-20-10 kg N, P2O5, K2O ha , respectively) under different hydrological conditions (0–10, 10–30, and 30–50-cm water depth). The results revealed that, in general, variety Ranjit was superior to IET-10016, producing about 8% higher yield. Application of the recommended level of fertilizer showed a definite yield advantage of about 39% over the farmers’ usual practice in all the hydrological conditions (Table 10). However, it was more in the case of semidwarf variety Ranjit than tall variety IET-10016. This clearly indicates that the inadequate application of fertilizer is one of the major causes of lower rice productivity. Depth of surface water also exerted a significant influence on grain yield of rice under both levels of fertilizer application and for both varieties (Table 10). Although an increase in surface-water depth from 0–10 to 30–50 cm had a positive effect on the grain yield of IET-10016, the same increase in water depth had a negative effect on the grain yield of Ranjit. For both varieties, however, such hydrological effects were more pronounced at water depths exceeding 30 cm, particularly at the recommended levels of fertilizer application. Up to 30-cm water depth, there was a slight change in the grain yield of both varieties, but at water depths exceeding 30 cm the change in grain yield was substantial. Although both varieties responded to the recommended level of fertilizer application, they behaved differently in surface hydrology class (Fig. 4). Ranjit yielded highest in the 0–10-cm water-depth class, whereas IET-110016 yielded the highest in the 30–50-cm depth class. Both yielded similarly in the 10–30-cm water-depth class with the farmers’ fertilizer level. These results indicate that, to raise the yields of rainfed lowland rice, fertilizer-responsive, high-yielding rice varieties are needed according to the specific hydrological situation (surface-water-depth class).

Planning and managing rice farming through environmental analysis 209 Table 10. Grain yielda (t ha–1) of rice varieties as affected by fertilizer management at different water depths, Assam, 1998.

Variety (V) Factor Ranjit IET-10016 Mean

–1 Fertilizer level (F) in kg N, P2O5, K2O ha Farmers’ fertilizer levelb 3.2 3.2 3.2 Recommended fertilizer levelc 4.8 4.2 4.5 LSD (0.05) F = 0.262 V × F = 0.371

Surface water depth (SWD) in cm 0–10 cm 4.7 2.9 3.8 10–30 cm 4.3 3.5 3.9 30–50 cm 3.0 4.7 3.8 LSD (0.05) SWD = 0.457 V × SWD = 0.647

Fertilizer × water depth × variety 0–10 cm: Farmers’ levelb 3.7 2.7 Recommended levelc 5.8 3.1

10–30 cm: Farmers’ levelb 3.5 3.1 Recommended levelc 5.1 3.9

30–50 cm: Farmers’ levelb 2.4 3.9 Recommended levelc 3.5 5.5 LSD (0.05) V × F × SWD = 0.643 Mean of varieties 4.0 3.7 LSD (0.05) V = 0.335

Fertilizer level on surface-water depth Farmers’ fertilizer Recommended level level 0–10 cm 3.2 4.4 10–30 cm 3.3 4.5 30–50 cm 3.1 4.5 LSD (0.05) 0.456

a b –1 c –1 d Mean of 8 replicates. 15, 5, 0 kg N, P2O5, K2O ha . 40, 20, 10 kg N, P2O5, K2O ha . At recommended level of fertilizer.

Selection of suitable rice varieties. The results of the environmental analysis have been used to help farmers select rice varieties according to ecosystem suitability in 20 villages of Milkipur block in Faizabad District, Uttar Pradesh. A micro-level rice ecosystem analysis was first carried out in these villages. Aerial photographs were used to delineate landforms, land use, and hydrological conditions in these villages. Within the culturable agricultural lands (including presently uncultivated but potentially cultivable), the percentage of area under different land types was uplands 12%, midlands 78%, and lowlands 10%. The meso-level (district-level) analysis of the area had shown these villages to be under the rainfed shallow (0–30-cm water

210 Borkakati et al Grain yield (t ha–1)

Farmers’ level (Ranjit) Recommended level (Ranjit) Farmers’ level (IET-10016) 6 Recommended level (IET-10016)

0 0–10 cm 10–20 cm 30–50 cm Water-depth class Fig. 4. Grain yield (t ha–1) of rice genotypes Ranjit and IET-10016 in three surface-water-depth classes with farmers’ level (15, 5, 0) and –1 recommended level (40, 20, 20) of N, P2O5, K2O (kg ha ) application. depth) drought-prone lowland subecosystem. The micro-level analysis showed about 88% of the rice-growing area under rainfed shallow drought-prone lowlands and the remaining 12% under the rainfed medium-deep (30–50-cm water depth) waterlogged subecosystem (Table 11). Subecosystem characterization also included information on drought and flooding pattern, timing, and duration in the rice-growing season. Based on this knowledge of rice environments, the farmers were provided with a rice variety (NDR-118) that is an early (90 d flowering), fine-grained variety, ex- perimentally found to be very suitable for rainfed shallow drought-prone lowlands. The performance of this variety was compared with that of the farmers’ grown varieties prevalent in the area (IR36, Sarju-52, and Pant-4), which are of medium duration (100 d flowering) and generally suitable for irrigated or rainfed shallow favorable lowlands. Mahsuri, a variety of medium to long duration (110–120 d flowering) and tall plant type, was being grown by farmers on medium-deep waterlogged lowlands. Seeds of these varieties were also provided to the farmers. During the wet season of 1994, despite a normal total rainfall in the area, terminal drought in rice occurred because of no rain after 20 September. Because of this drought, some farmers could not get any yield and the crop failed completely. Data collected from 385 farmers from these villages showed such total failure to the extent of only 7% of farmers’ fields with NDR-118. The other varieties used by farmers showed total failure to the extent of 39% for farmers using IR36, 18% for Mahsuri, and 11% for Sarju-52 (Table 12). Data collected from two selected villages in subsequent years showed a significant increase in rice area coverage under NDR-118, the variety found suitable for the dominant subecosystem of the area.

Planning and managing rice farming through environmental analysis 211 Table 11. Area under different rice ecosystems in Milkipur study area, Faizabad District, India.

Rice subecosystems and other Area % geographical land-use class (ha) area

Rainfed shallow drought-prone lowlands 4,447 62.7 Rainfed medium-deep waterlogged lowlands 583 8.2 Other crops 736 10.4 Forests and orchards 472 6.7 Uncultivated 502 7.1 Ponds/lakes 193 2.7 Habitation 155 2.2

Total 7,088 100.0

Table 12. Rice varieties tested, ecosystem suitability, and percentage total crop failure in farmers’ fields in Milkipur area, Faizabad District, India (kharif 1994). Total number of farmers = 385.

Duration and ecosystem % farmers Rice variety suitability reporting total crop failure

IR36 130 d, irrigated/rainfed shallow favorable lowlands 39 Sarju-52 140 d, irrigated/rainfed shallow favorable lowlands 11 Mahsuri 145 d, irrigated/rainfed shallow favorable lowlands/medium- deep waterlogged 18 NDR-118 90 d, rainfed shallow drought-prone lowlands/rainfed uplands 7

This shows that the knowledge of environmental characterization helps in selecting appropriate varieties, whose use can help farmers increase production and reduce risk in uncertain rainfed environments.

Conclusions The results of the studies carried out in eastern India indicate that the performance of the rice crop is strongly influenced by the hydrological conditions prevailing in the growing season. These hydrological influences can be mitigated considerably if crop management strategies are developed on the basis of the knowledge of environmental conditions. Likewise, other appropriate technologies can be developed using such knowledge and can easily be applied by farmers in rainfed situations.

212 Borkakati et al References IRRI (International Rice Research Institute). 1990. Program report for 1989. Los Baños, La- guna (Philippines): IRRI. 302 p. IRRI (International Rice Research Institute). 1992. Program report for 1991. Los Baños, La- guna (Philippines): IRRI. 322 p. IRRI (International Rice Research Institute). 1993. Program report for 1992. Los Baños, La- guna (Philippines): IRRI. 316 p. IRRI (International Rice Research Institute). 1998. Program report for 1997. Los Baños, La- guna (Philippines): IRRI. 175 p. Paul DK, Tiwarim KN. 1994. Rainwater storage systems for rainfed rice lands of eastern India: results from research in Hazaribagh district. In: Bhuiyan S, editor. On-farm reservoir systems for rainfed rice lands. Manila (Philippines): International Rice Research Insti- tute. 164 p. Singh VP. 1996. Agroecological analysis for sustainable development of rainfed environments in India. J. Indian Soc. Soil Sci. 44(4):601-615. Singh VP, Pathak MD. 1990. Rice growing environments in Baharaich district of Uttar Pradesh, India. Tech. Bull. No. 1. Uttar Pradesh Council of Agricultural Research, Lucknow, India. 56 p. Singh VP, Singh AN. 1996. A remote sensing and GIS-based methodology for the delineation and characterization of rainfed rice environments. Int. J. Remote Sensing 17(7):1377- 1390. Singh VP, Singh RK, editors. 2000. Rainfed rice: a sourcebook of best practices and strategies in Eastern India. Makati City (Philippines): International Rice Research Institute. 292 p. Singh VP, Singh RK, Singh RK, Chauhan VS. 1993. Developing integrated crop-livestock-fish farming systems for rainfed uplands in Eastern India. J. Asian Farm. Syst. Assoc. 1:523- 536. Singh VP, Singh RK, Sastri ASRAS, Baghel SS, Chaudhry JL. 1999. Rice growing environments of Eastern India: an agroclimatic analysis. IGAU and IRRI Pub. 76 p. Thornthwaite LW, Mather JR. 1955. The water balance. In: Climatology 11:1-14. Drexel Insti- tute of Technology, Centerton, N.J. (USA).

Notes Authors’ addresses: K. Borkakati, Department of Soil Science, Assam Agricultural University, Jorhat, Assam, India; V.P. Singh, International Rice Research Institute, Los Baños, Phil- ippines; A.N. Singh, Uttar Pradesh Remote Sensing Applications Center, Lucknow, In- dia; R.K. Singh, Central Rainfed Upland Rice Research Station, Hazaribagh, India; A.S.R.A.S. Sastri, Indira Gandhi Agricultural University, Raipur, Madhya Pradesh, India; S.K. Mohanty, Central Rice Research Institute, Cuttack, Orissa, India. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Planning and managing rice farming through environmental analysis 213 Agroclimatic inventory for environmental characterization of rainfed rice-based cropping systems of eastern India

A.S.R.A.S. Sastri and V.P. Singh

Climate is an important component of environment. In environmental characterization for developing strategies to improve productivity, an agroclimatic inventory is a prerequisite. In eastern India, comprising the states of Orissa, Bihar, West Bengal, eastern Madhya Pradesh, eastern Uttar Pradesh, and the northeastern states, rice is grown mostly under rainfed conditions in upland, lowland, and flood-prone ecosystems. As this region comes under the influence of the southwest monsoon from June to October, crop productivity depends entirely upon the vagaries of monsoon. Also, because of monsoon activity, the sky is mostly overcast and radiation becomes a limiting factor. In this study, the moisture regime was analyzed using simple measures such as amount of rainfall and number of rainy days and derived parameters such as moisture availability and stable rainfall periods at different probability levels. We found that the average rainfall in eastern India matches mostly with 30% or, in some cases, 40% probability levels, indicating that any average rainfall-based strategy for improving rice productivity is successful only once in three years. Therefore, a concept of stable rainfall period has been developed by defining it as a period when weekly rainfall exceeds 50 mm. Stable rainfall periods even at a 60% probability level are very short in some parts of the region, suggesting a need to develop viable location-specific water management practices. The thermal regime in this region is not generally a limiting factor, except on a few occasions. Because radiation is a limiting factor during the active monsoon months, however, we need to identify rice varieties with higher energy-use efficiency.

Climate and its components such as temperature and rainfall are the factors that most affect agricultural productivity and at the same time are the least modifiable (Frankel 1976). Rainfall is the major climatic element that affects crop growth and development, particularly where rainfed farming is widely accepted. Also, in comparison with other climatic factors, data on rainfall are generally available for an extensive

Agroclimatic inventory for environmental characterization . . . 215 network (Sastry 1976). Temperature conditions in a particular geographic area represent decision criteria for the natural establishment of plants. All the physiological processes take place within certain tolerance ranges of temperature. The temperature has to reach a certain minimum value at which biological processes start and their activity reaches its peak at optimum temperatures. At too high temperatures, the biological processes will cease (Morison and Butterfield 1990). Yield potential primarily depends on the amount of solar radiation, especially during the reproductive and maturity stages (Yoshida and Parao 1976). The climatic adaptability of the rice crop, which is adapted to a broad latitudi- nal belt ranging from 40°S in central Argentina to 53°N in northeast China, is different from that of other crops. The rice crop is very sensitive to water deficit during the reproductive to heading stage, resulting in high sterility (Seshu 1989). On the other hand, extreme temperatures are destructive for rice crop growth and development. The critical low and high temperatures for rice, normally below 20 °C and above 30 °C, vary among growth stages (Yoshida 1981). According to Yoshida (1973), higher temperatures result in increased tiller number when light is adequate and low temperatures may result in the production of more tillers under low light conditions. Nishiyama (1976) stated that 9–16 °C and 33 °C are the lower and higher critical temperatures for tiller production and the 25–31 °C range is optimum for tillering. In eastern India, temperatures are mostly within this optimum range and hence temperature does not affect tiller production. Chaudhary and Ghildayal (1970), while working in eastern India, stated that 10 °C is the lowest critical temperature for tiller production. Owen (1972), however, reported that a temperature range of 20–25 °C is optimum for flower bud initiation in cultivar IR8, whereas 15 °C prevented floral initiation. Such conditions often arise in eastern Uttar Pradesh, Bihar, and some parts of the northeastern states. Murata (1976) developed a regression equation between grain yield (Y) and dry weight at heading (Wo) and average solar radiation during ripening and opined that Wo is a better measure than solar radiation to assess yield potential. Yoshida and Parao (1976) obtained a positive correlation between solar radiation and biomass production during the reproductive stage. In eastern India, during good rainfall years, solar radiation becomes a limiting factor for increased biomass and thereby grain yield. Thus, for the rice crop, all three components of climate—moisture, thermal, and light regimes—individually and combined affect growth, development, and grain yield. Besides climate, other environmental factors such as soil physical and chemical properties, weed flora, and other soil microorganisms vary. All these other environmental factors in some way or another are influenced by climatic factors. It is therefore necessary to make a thorough inventory of these climatic factors for the characterization of different rainfed rice-based cropping systems, which helps in planning agricultural development in any given region.

216 Sastri and Singh The eastern India scenario In eastern India, consisting of Bihar, Orissa, eastern Madhya Pradesh, eastern Uttar Pradesh, West Bengal, and the northeastern states (Fig. 1), rice is the major crop grown in favorable as well as adverse and fragile environments. Except in the irrigated area, which is very small, most of the region is monocropped and rice is grown under rainfed conditions. The rainfed rice ecosystem in eastern India is again divided into different subecosystems such as upland, lowland, and flood-prone. The lowlands can be further subdivided into favorable and drought-prone. These subecosystems, in relation to other cropping and production systems, form the overall agricultural scenario of eastern India. In lowlands, a typical cultivation system called the “broadcast beushening” system is adopted, under which rice seeds are broadcast in preplowed fields immediately after the onset of monsoonal rains around 15–30 June. After about 30–35 d, when sufficient water is impounded in the diked rice fields, the fields are plowed in the standing rice crop. This operation is called beushening. Beushening helps (1) reduce the plant population, (2) weed the fields, and (3) create a semipuddled condition to minimize percolation losses to some extent. In irrigated areas and also in a small portion of rainfed areas, transplanting is being practiced. In uplands and in light soils, direct seeding through drilling in lines is also practiced. The broadcast beushening method, however, is widely adopted. Sowing, beushening, and later crop operations depend entirely on the vagaries of monsoonal rains. In addition, winter conditions in certain states of eastern India set

Bahraich Dibrubarh LUCKNOW Gorakhpur Darjiling Tezpur Jorhat Kanpur Muzaffarpur Darbhanga GUWAHATI Fatehpur Allahabad PATNA Cherrapunji Gaya Silchar IMPHAL Daltenganj Hazaribagh Agartala Ambikapur Burdwan Ranchi Aizawl Jabalpur Purulia Jamshedpur Raigarh CALCUTTA Raipur Sambalpur Cuttack BHUBANESHWAR Titlagarh Jagdalpur Gopalpur

Fig. 1. Rainfed rice-growing areas of eastern India.

Agroclimatic inventory for environmental characterization . . . 217 Temperature (°C) and solar radiation (SR) Rainfall (mm) 70 700 A B C 50 500

30 300

10 100 0 0 JFMAMJJASOND JFMAMJJASOND JFMAMJJASOND Month

Max. temp. (°C) Min. temp. (°C) SR Rainfall (mm)

Fig. 2. Temporal dynamics of rainfall, maximum and minimum temperature, and solar radiation in coastal areas (Cuttack, Orissa, A), higher latitude areas (Patna, Bihar, B), and high-rainfall areas (Silchar, Assam, C). in early, causing damage from low temperature, such as high sterility in the rice crop. As all of eastern India comes under the influence of the southwest monsoon (June- October), solar radiation also becomes a limiting factor during the vegetative stage and early parts of the reproductive stage of the rice crop (mostly in July and August).

Temporal dynamics Figure 2A-C shows the temporal dynamics of rainfall, maximum and minimum temperatures, and solar radiation for three representative stations—coastal, higher latitude, and high-rainfall areas—in eastern India. In coastal areas (Fig. 2A), the maximum temperature is always near or above 30 °C and the minimum temperature is always above 15 °C. In higher latitudes such as in Patna, the minimum temperature falls sharply from October onward (Fig. 2B) and this, as stated earlier, may create problems in grain filling, if these conditions start early. On the other hand, in high- rainfall areas such as Silchar in the northeastern states, with higher rainfall from May to September, solar radiation is lower than in other places (Fig. 2C). The minimum temperature falls below 15 °C only in November. As the rainfall starts early with higher amounts and intensity, submergence or flood is the main problem in this area, especially in June and July, besides lower amounts of solar radiation. In view of the importance of these climatic factors, which limit the optimum productivity of the rice crop, this chapter attempts to analyze three agroclimatic factors—moisture, thermal, and radiation regimes—in eastern India during different stages of crop growth.

Materials and methods The climatic data required for the analysis are collected from the India Meteorologi- cal Department (IMD). The rainfall probabilities are collected from the data published by IMD (1995). The solar radiation data required for the analysis are estimated

218 Sastri and Singh from cloud cover data based on Penman’s (1948) equation. The desirable rainfall periods are graphically extrapolated as per the criteria given in the text. The maps are initially prepared by cartographic method and then computerized through scanning. The computerized maps are analyzed through GIS software IDRISI.

Moisture regime Rainfall amount. In eastern India, a major portion of rainfall is received during the southwest monsoon season (June-October). In all the states, however, especially in the northeastern states, rainfall also occurs because of premonsoon thunderstorm activity during April and May. Figure 3 shows the mean annual rainfall pattern in eastern India. Annual rainfall varies from more than 3,000 mm in the northeastern states to less than 1,000 mm in the southern parts of Orissa and eastern Uttar Pradesh. The higher amounts of rainfall cause floods in the northeastern states, whereas lower amounts of rainfall cause acute drought conditions in Bihar, Orissa, eastern Madhya Pradesh, and eastern Uttar Pradesh. Using a simple measure of annual rainfall, the rainfed rice environments can be characterized. For example, eastern India can be divided into eight rainfall zones according to rainfed rice cultivation (Table 1).

Gonda

Hazaribagh Ranchi Aizawl Ambikapur Jabalpur Champa Jamshedpur Calcutta

Balasore Sambalpur Raipur Cuttack Titlagarh >3,000 mm 1,400–1,600 mm Extremely high rainfall Medium rainfall Berhampore 2,000–3,000 mm 1,200–1,400 mm Very high rainfall Moderately medium rainfall Gopalpur 1,800–2,000 mm 1,000–1,200 mm High rainfall Low rainfall 1,600–1,800 mm <1,000 mm Moderately high rainfall Extremely low rainfall

Fig. 3. Mean annual rainfall pattern in eastern India.

Agroclimatic inventory for environmental characterization . . . 219 Table 1. Characterization of rainfall zones in eastern India.

Annual rainfall (mm) Classification zone

>3000 Extremely high 2,000–3,000 Very high 1,800–2,000 High 1,600–1,800 Moderately high 1,400–1,600 Medium 1,200–1,400 Moderately medium 1,000–1,200 Low <1,000 Extremely low

The area in the first three zones is flood- and submergence-prone for longer periods, whereas the area in the last three zones is drought-prone, with either intermittent or terminal drought conditions. The other two zones—moderately high and medium rainfall—are often favorable for rice cultivation. Even the transplanting method of cultivation is followed in this area under rainfed conditions, as mentioned earlier. Rainy days. Rainy days, in a broader sense, determine crop duration as well as cropping intensity. In eastern India, the number of rainy days varies from more than 130 in the northeastern states to less than 50 in eastern Uttar Pradesh (Fig. 4). In a major portion of eastern India, the annual number of rainy days varies from 50 to 80. With intermittent dry spells, the crop-growing season is always longer than the rainy season. The length of the growing season depends not only on the number of rainy days or duration of the rainy season but also on moisture availability periods and soil physical characteristics such as water-holding capacity. Thus, to assess the possibility of increasing cropping intensity, moisture availability periods are a better measure than just rainy days. To analyze moisture availability periods, data on the temporal distribution of rainfall and potential evapotranspiration (PET) are needed. PET values for 44 meteorological stations in eastern India are computed using Penman’s (1948) equation. Using PET values, the moisture availability periods for different stations in eastern India are worked out as described below. Moisture availability periods. Cocheme and Franquin (1969) designated different moisture availability periods using rainfall (R) and PET values. Moisture availability periods are categorized as follows:

Moisture availability period Status of rainfall (R) and PET

Humid R > PET Moist PET > R > PET/2 Submoist PET/2 > R > PET/4

220 Sastri and Singh Bahraich Gonda Lucknow Boghlera Kanpur Gorakhpur Darbhanga Dhuburi Fatehpur Patna Allahabad Baghalpur Sabour Safna Gaya Hazaribagh Aizawl Daltenganj Burdwan Purulia Uherte Ambikapur Ranchi Midnapur Jabalpur Champa Pendro Jamshedpur Calcutta Raigarh Balasore Raipur Sambalpur

Cuttack Kanker Titlagarh 100–120 60–70 Berhampore 120–130 70–80 Jagdalpur >130 <80

Gopalpur <50 <90 50–60 90–100

Fig. 4. Annual number of rainy days in eastern India.

For rainfed rice cultivation, the submoist period is not suitable and hence the duration of the humid and moist periods alone is discussed in this chapter. The submoist period is helpful, however, for the second crop after rice, which is mostly lathyrus, linseed, or chickpea sown as a relay crop in heavy soils with higher moisture retention capacity. The moist period preceding and succeeding the humid period is designated as the moist I and moist II period, respectively. The moist I period is the period at the beginning of the rainy season when PET values are higher than rainfall. Most crop establishment operations are done during this period. The moist II period is the period at the end of the rainy season. It is a suitable period for rice crop maturity and harvesting. Also, a relay crop of lathyrus, linseed, or chickpea is sown during this period in the standing rice crop. Based on these criteria, the moisture availability periods for the 44 meteorological stations in eastern India have been worked out based on the graphical interpolation of temporal dynamics of rainfall and PET. Considerable variability exists in the moisture availability periods across the stations in each state. In the northeastern states, the humid period is as high as 278 d in Cherapunji (Meghalaya), whereas in Kanpur (Uttar Pradesh) it is as low as 81 d. Similarly, the moist I period, which is suitable for field preparation and sowing, starts as early as 2 February in Assam to as late as 9 July

Agroclimatic inventory for environmental characterization . . . 221 in eastern Uttar Pradesh under normal conditions. Table 2 shows the variations in sowing periods (moist I period) in different states of eastern India. In general, the entire sowing operation in eastern India should be completed within 20–25 d. Within a classified rainfall zone, significant variations occur in the total duration of moisture availability periods across locations (Table 3). Within the rainfall zone of 1,200–1,400 mm, the humid period is 95 d in Gaya and 128 d in Sabour (Bihar). Similarly, in the rainfall zone of 1,400–1,600 mm, the humid period is 153 d in Cuttack (Orissa) and only 117 d in Raipur (Madhya Pradesh). This clearly implies that varietal duration should be based on the duration of the moisture availability period but not on the amount of rainfall.

Table 2. Range of the beginning of the sowing period (beginning of moist period I) in different states in eastern India.

Sowing period State Variation Duration (d)

Eastern Madhya Pradesh 31 May–21 Jun 21 Eastern Uttar Pradesh 11 Jun–9 Jul 28 Bihar 5 Jun–7 Jul 32 Orissa 19 May–9 Jun 20 West Bengal 28 Apr–9 Jun 42 Assam 2 Feb–28 Mar 54 Meghalaya 14 Feb–6 Mar 20 Mizoram 22 Mar–18 Apr 27

Table 3. Variation in moisture availability periods of two different rainfall zones.

Moisture availabilitya Annual Site rainfall Moist I Humid Moist II (mm) DP D P D P

Gaya, Bihar 1,200–1,400 17 15 Jun–10 Jul 95 11 Jul–13 Oct 22 14 Oct–10 Nov

Sabour, Bihar 1,200–1,400 16 7–22 Jun 128 23 Jun–28 Oct 18 29 Oct–15 Nov

Raipur, 1,400–1,600 14 16–29 Jun 117 30 Jun–24 Oct 18 25 Oct–11 Nov Madhya Pradesh

Cuttack, 1,400–1,600 18 30 May–17 Jun 153 18 Jun–17 Nov 18 18 Nov–5 Dec Orissa aD = duration in days, P = period.

222 Sastri and Singh Table 4. Range of moisture availability periods in different states of eastern India.

Range of moisture availability periods (d) State Moist I Humid Moist II

Madhya Pradesh 10–16 103–147 12–21 Uttar Pradesh 12–18 81–123 17–22 Bihar 12–18 95–139 17–22 Orissa 12–24 131–165 15–18 West Bengal 16–28 130–208 11–22 Assam 25–33 191–270 17–22 Other northeastern states 21–28 224–278 9–17

Table 4 shows the range of different moisture availability periods in each state of eastern India. The duration of the moist I period varies from 10 to 33 d in different states, whereas that of the moist II period varies from 9 to 22 d. The humid period, which is very important for the rice crop’s growth and development, varies from 81 to 123 d in Uttar Pradesh to 224 to 278 d in the northeastern states. The higher number of days under the humid period also includes the submergence (flooding) period in some states. Those areas with a longer humid period duration are suitable for increasing cropping intensity. The kind of crop to be grown again depends on the soil type and thermal regime during the winter period (November-March). Rainfall probabilities. The discussion so far is based on average rainfall conditions. But to improve the productivity of rice in eastern India, especially under rainfed conditions, it is more realistic if the rainfall analysis is carried out at different probability levels and interpreted accordingly. The India Meteorological Department (IMD 1995) published the probabilities of different quantities of weekly rainfall for all the meteorological stations in India. The data for the period 20th to 45th standard meteorological week (May to October) have been used to analyze and characterize rainfed rice environments in different states in eastern India. The average weekly rainfall for most of the stations in eastern India matched with either 30% or 40% probability. This implies that any agricultural planning on the basis of average rainfall data has a probability of success once in three years only and, in the other two years, rainfall is always below average. This may affect rice crop growth and development, especially in upland and lowland drought-prone ecosystems. This clearly indicates that, in eastern India, rainfed rice-growing environments can be characterized in a better way by considering rainfall probabilities, especially desirable rainfall probabilities. Desirable rainfall for rainfed rice cultivation. The average daily water losses by potential evapotranspiration and percolation in rice fields of eastern India account for approximately 3 to 4 mm each. Therefore, the total water requirement for water stress-free rice cultivation under rainfed conditions can be assured at a minimum of 7 mm d–1. Also, by considering a minimum value of 7 mm d–1, it is imperative that the

Agroclimatic inventory for environmental characterization . . . 223 fluctuations/variations in rainfall above this threshold value do not influence crop growth and development. Thus, considering desirable rainfall for rainfed rice cultivation to be a minimum of 7 mm d–1, the duration and starting periods of desirable rainfall for average and different probabilities are worked out. The important features of the analysis follow. Figure 5 shows the duration of desirable rainfall under average conditions in eastern India, which varies from less than 60 d in a small pocket in eastern Uttar Pradesh to more than 200 d in Meghalaya. In a major portion of eastern India (excluding the northeastern states), the desirable rainfall period is around 80 to 100 d, while in the northeastern states the average duration varies from 140 to 200 d. Thus, the problem in eastern India (excluding the northeastern states) is water stress at one stage or another, whereas in the northeastern states excess moisture or submergence limits rice productivity. In Berhampore District of West Bengal, the desirable rainfall period occurs in two splits, implying that in some parts of West Bengal submergence and drought can occur in the same cropping season. Such factors need to be thor- oughly examined for developing improvement strategies for rice productivity at the district or microregional level, such as tehsil or blocks.

Dibrubarh Bahraich Barchie Jorhat Gorakhpur Gonda Tezpur LucknowLUCKNOW GorkhoreMuzaffarpur Darjiling Kanpur Dhuburi GUWAHATI Fatehpur Darbhanga PATNA Cherrapunji Allahabad Baghalpur Gaya Silchar Gaya IMPHAL SetnaSafna Daltenganj Sabour Berthampore Umaria Battonganj Purulia Uherte Ambikapur Burdwan Jabalpur Hazaribagh Jabalpur Midnapur Pendro Champa Raigarh Ranchi CALCUTTACalcutta Jharsuguda RaipurRalpur Sambalpur Belasore Cuttack KankerKanker Titlagarh >100 60–70 160–180 BHUBANESHWAR Jagdalpur 90–100 >60 140–160 80–90 >200 120–140 Gopalpur 70–80 180–200 >120

Fig. 5. Duration (in days) of desirable rainfall periods in eastern India.

224 Sastri and Singh A knowledge of the starting of desirable rainfall periods, besides their duration, helps in developing crop calendars to suit a particular location, which help in planning different crop operations in the growing season. Hence, the starting dates for average and different probabilities of desirable rainfall periods have been worked out and are discussed below. The analysis of average dates of starting periods of desirable rainfall in eastern India revealed that, in the northeastern states, desirable rainfall starts as early as the end of March in Manipur and from 1 to 5 April in a major portion of the northeastern states. In Assam, the desirable rainfall period starts on or just after 15 April. In other eastern states, the starting period varies from 15 June to 5 July. It has a direct relation with the onset of the southwest monsoon, but with a slight lag of about 10 to 15 d. The average dates of the start of the desirable rainfall period are around 15 June in Orissa and around 5 July in the western parts of eastern Uttar Pradesh. The early start of the desirable rainfall period in the northeastern states clearly shows the dependability of rice cultivation on the premonsoon thunderstorm rainfall for crop operations. Analysis of desirable rainfall periods at different probabilities: a summary by states. The results of the analysis of desirable rainfall periods for all the stations in each state are discussed in detail below. In Orissa, the average duration of desirable rainfall varies from 33 d at Gopalpur to 120 d at Balasore. At Gopalpur, the average duration is not only less but also not continuous. This implies that in this area rice cultivation without supplemental irrigation is not possible. At 50% and higher probability, duration is either too low or nil at all the places in Orissa except Sambalpur, where duration of the desirable rainfall period is 81, 65, and 51 d at 50%, 60%, and 70% probabilities, respectively. This indicates that there are few opportunities for rainfed rice cultivation without water stress in most of these places. For eastern Madhya Pradesh, the desirable rainfall period at 60% probability is more than 60 d except at Raipur and Satna. This is a clear indication of favorable conditions for rainfed rice cultivation. In these districts with proper water management practices, rice productivity can be improved considerably because other conditions are favorable. In Bihar, with considerable area under uplands, the average duration of desirable rainfall periods varies from 67 d in Baghalpur to 97 d in Ranchi. At Baghalpur, the desirable rainfall period starts later (1 July) than in other areas. At 50% probability, there is no desirable rainfall period at this station, whereas at other stations duration varies from 36 to 73 d with a discontinuity in between at Darbhanga and Gaya. Ranchi has 54 d of desirable rainfall at 60% probability, indicating a better scope for improving rice productivity by adopting better water management practices. In eastern Uttar Pradesh, the average duration of the desirable rainfall period varies from 63 d at Allahabad to 93 d at Bahraich. At Bahraich, however, the duration of desirable rainfall decreases sharply at 40% and 50% probability. This shows that Bahraich has a higher variability of rainfall and less dependability; hence, rice productivity is very low at Bahraich compared with other parts of Uttar Pradesh. Also,

Agroclimatic inventory for environmental characterization . . . 225 the desirable rainfall period at 50% probability at all stations in eastern Uttar Pradesh ends by mid-August. Thus, terminal drought is a recurring feature in eastern Uttar Pradesh. In West Bengal, the average duration of the desirable rainfall period is around 100 d but it starts earlier than in other eastern states. Duration is less, however, when compared with eastern Madhya Pradesh and Orissa. At 50% probability, duration varies from 39 to 57 d in different parts of West Bengal. In this state, however, the intensity of rainfall is higher, as the rains are received because of cyclonic activity in the Bay of Bengal, resulting in flooding in rice fields; hence, deepwater rice is more prevalent. For the northeastern states, the desirable rainfall situation is quite different from that of the other states of eastern India. The average duration is high and starts about 50 to 60 d earlier than in other states. At 60% probability, duration varies from as high as 145 d at Silchar to as low as 3 d at Guwahati. Besides duration, the intensity of rainfall is also high and the terrain is undulated in this area. Therefore, floods and sometimes flash floods are a recurring feature. It is therefore necessary to analyze rainfall and other agroecological characteristics of the northeastern states with a different approach. The strategies for improving rice productivity in these states must be different from those of other states of eastern India.

Thermal regime In eastern India, in general, the thermal regime (both the maximum and minimum temperatures) is favorable for rice cultivation. Higher day temperatures at the initial stages and lower night temperatures during the anthesis period, however, sometimes limit the crop’s growth and development. In some high-rainfall areas, rice is grown under rainfed conditions from September-October either after the floodwater recedes or after the harvest of the first crop sown in April-May. For this crop, the temperatures in the winter (November-March) season are also important. In view of this, the maximum and minimum temperatures of both the rainy (June-October) and winter (No- vember-Dececember) seasons are analyzed and discussed below. Maximum temperatures. The maximum temperature during the rainy season ranges mostly between 30 and 34 °C. Higher day temperatures prevail in the western- most part of eastern Uttar Pradesh, where the onset of monsoon is relatively late and therefore, in some years with its late onset (after 1 July), higher day temperatures during the sowing/germination time may affect the crop. In the rest of the region, maximum temperatures are favorable during the rainy season. Minimum temperatures. The minimum temperature during the rainy season varies little throughout eastern India. In eastern Uttar Pradesh, Bihar, West Bengal, and coastal Orissa, it is more than 25 °C, whereas, in a majority of the area in the northeastern states and in eastern Madhya Pradesh, it varies from 23 to 25 °C. In a small area in eastern Madhya Pradesh and Meghalaya, it is less than 23 °C. The thermal regime in general is favorable for rice cultivation in eastern India. As mentioned earlier, however, temperature becomes a limiting factor only when

226 Sastri and Singh Temperature (°C) 50 Maximum (normal) Maximum (1985) 40 Minimum (normal) Minimum (1985) 30° 30° 31° 30 22° 20° 20° 20

10 Panicle Germination Vegetative initiation Grain filling Maturity Seedling

0 0 2 46810121416 18 20 Weeks from sowing

Fig. 6. Pattern of weekly averages of maximum and minimum temperatures in a normal year and in 1985 during the rice-growing season compared with the optimum values in different growth stages. winter conditions set in early and the night temperature falls below 15 °C at the grain- filling stage, as occurs in Bihar and eastern Uttar Pradesh. Figure 6 shows the average weekly temperature pattern at Raipur and cardinal temperatures at various phenological stages. The maximum temperature during the vegetative stage is slightly below the optimum temperature limits (Mavi 1976) . Similarly, during the grain-filling stage, the minimum temperature, which limits grain filling, is above the threshold values (Fig. 6) under average conditions at Raipur. Even under such favorable conditions, the minimum temperature sometimes falls below the threshold limits, thus causing sterility problems. Variations also occur in varietal interaction with such low night temperatures. A detailed analysis is needed to examine varietal interaction with low night temperatures.

Radiation regime As mentioned earlier, radiation is a limiting factor for rice in the vegetative and early reproductive stages of crop growth as eastern India comes under the influence of the southwest monsoon. Even during a dry period, radiation values are lower in this region as the sky is mostly overcast. Radiation is therefore a very important climatic component for rice cultivation in this region during the rainy season. In India, the network of radiation measuring stations is sparse. Data on cloud amount in octos measured twice a day, however, are readily available. Therefore, radiation values were estimated for all 44 meteorological stations in eastern India using Penman’s (1948) equation.

Agroclimatic inventory for environmental characterization . . . 227 The radiation values range from less than 12.5 to more than 16.5 MJ m–2 d–1 in January and they start increasing thereafter until May. In May, the radiation values reach their peak at more than 25.0 MJ m–2 d–1 in eastern Uttar Pradesh. In the northeastern states, however, the peak values are reached in April and decrease thereafter, as premonsoon and then monsoonal rains start. The solar radiation values in eastern India are lowest in July and August and range from less than 8.5 to slightly more than 12.5 MJ m–2 d–1 (Fig. 7A,B). For better crop growth and development, a minimum value of 12.0 MJ m–2 d–1 is needed. Figure 2 also shows the temporal dynamics of solar radiation at three representative stations. At Cuttack, where the monsoon starts earlier (mid-June), the solar radiation is less than 12.0 MJ m–2 d–1 from July to September, whereas, in Patna, where the monsoon starts late, in the first week of July, solar radiation values are less than 12.0 MJ m–2 d–1 only in August and September. In the rest of the months, the values are more than 12.0 MJ m–2 d–1. In the case of Silchar, however, the solar radiation values are lower than 12 MJ m–2 d–1 from June to September. This is a clear indication that, in higher rainfall areas, solar radiation constrains the rice crop’s growth and development.

Summary and conclusions To develop strategies for improving rice productivity, a thorough inventory of the three components of climate—moisture, thermal, and radiation regimes—is needed. In eastern India, rice is grown mostly under rainfed conditions in three subecosystems: rainfed lowland, upland, and flood-prone. In this region, rainfall or moisture regime is an important climatic component because most of the area is under upland and rainfed lowland drought-prone ecosystems. Rainfall amount and rainy days are simple measures for characterizing rainfed rice environments. Using rainfall amounts, eastern India is divided into eight zones, of which three are under flood/submergence- prone conditions and three are under drought-prone conditions. Only two zones come under favorable conditions. In both the flood-prone and drought-prone areas, better water management practices need to be developed. Rainy days, though readily available, are not a good measure for assessing the growing season as the growing season is always longer than rainy days due to intermittent dry spells. Therefore, moisture availability periods are a relatively better measure for assessing the crop-growing season and cropping pattern. The humid period (P > PE) is the best period for the rice crop’s vegetative and reproductive phases; therefore, varietal duration needs to be assessed based on the humid period. The rice- based second crop in the winter season can be grown in areas with a longer duration of humid and moist II periods and in soils with a higher moisture retention capacity. In spite of such detailed analysis, in eastern India average rainfall matches with only 30% or 40% probability, implying that any strategy based on average rainfall can be successful only once in three years. Hence, a concept of desirable rainfall period has been developed by defining a desirable rainfall period as the period when daily rainfall is 7 mm or more. The rainfall fluctuations above this threshold value do

228 Sastri and Singh 30°N A

Lucknow Gonda

Jabalpur Pendro Calcutta Champa Midnapur Raipur Sambalpur 20 Balasore Cuttack >12.5 8.5–10.5 Jagdalpur Berhampore 10.5–12.5 <8.5 Gopalpur

30°N B

Jabalpur

Raipur Sambalpur 20 Cuttack

Titlagarh

Gopalpur

15 75 80 85 90 95 100°E

Fig. 7. Solar radiation (MJ m–2 d–1) in July (A) and August (B) over eastern India.

Agroclimatic inventory for environmental characterization . . . 229 not affect the growth and development of rainfed rice. The analysis of the desirable rainfall periods at different probabilities, however, indicates that the duration of the desirable rainfall period at 60% or higher probability is very low in most states in eastern India. This suggests the need to develop better water management practices to improve rice productivity. An analysis of the thermal regime indicated that, in general, it is favorable for rainfed rice cultivation in eastern India under normal conditions. In some years, however, when winter conditions set in early, low night temperature may lead to sterility conditions in rice. In eastern India, radiation is a limiting factor during the vegetative and reproductive stages. With a threshold value of 12.0 MJ m–2 d–1 as the lower limit (approximately 300 calories m–2 d–1), radiation is more constant in high-rainfall areas than in low-rainfall areas of eastern India. Thus, a thorough inventory of the three climatic components revealed that, in most of the places, rainwater management in both flood-prone and drought-prone areas is the most important strategy for increasing rainfed rice productivity. Water management practices include better drainage in flood-prone areas and rainwater harvesting and recycling in drought-prone areas. The thermal regime is more or less favorable, but, for radiation, it is necessary to identify suitable varieties with higher energy-use efficiency.

References Chaudhary TN, Ghildayal BP. 1970. Influence of submerged soil temperature regimes on growth, yield and nutrient composition of the rice plant. Agron. J. 62:282-285. Cocheme J, Franquin P. 1969. An agroclimate survey of a semi-arid area in Africa-South of Sahara. FAO/WMO Technical Bulletin 86. Frankel OH. 1976. The IRRI phytotron: science in the service of human welfare. In: Climate and rice. Los Baños (Philippines): International Rice Research Institute. p 3-9. IMD (India Meteorological Department). 1995. Weekly rainfall probability for selected stations of India. Vol II. Pune (India): IMD. 517 p. Mavi HS. 1976. Introduction to agrometeorology. New Delhi (India): Oxford & IBH Publish- ing Co. 237 p. Morison JLL, Butterfield RE. 1990. Cereal crop damage by frosts; spring 1990. Weather 45(8):308-317. Murata Y. 1976. Productivity of rice in different climatic regions of Japan. In: Climate and rice. Los Baños (Philippines): International Rice Research Institute. p 449-470. Nishiyama I. 1976. Effect of temperature on the vegetative growth of rice. In: Climate and rice. Los Baños (Philippines): International Rice Research Institute. p 159-185. Owen PC. 1972. Effects of night temperature on growth and development of IR8 rice. Exp. Agric. 8:213-218. Penman HL. 1948. Natural evaporation from open water bare soil and grasses. Proc. R. Soc. (A) 193. 120 p. Sastry PSN. 1976. Climate and crop planning with particular reference to rainfall. In: Climate and rice. Los Baños (Philippines): International Rice Research Institute. p 51-63.

230 Sastri and Singh Seshu DV. 1989. Impact of major weather factors on rice production. In: Agrometeorological information for planning and operation in agriculture with particular reference to plant protection. Geneva (Switzerland). p 41-63. Yoshida S. 1973. Effect of temperature on growth of the rice plant (Oryza sativa L.) in controlled environment. Soil Sci. Plant Nutr. 19:299-310. Yoshida S. 1981. Fundamentals of rice crop science. Los Baños (Philippines): International Rice Research Institute. 269 p. Yoshida S, Parao FT. 1976. Climatic influence on the yield and yield components of lowland rice in the tropics. In: Climate and rice. Los Baños (Philippines): International Rice Research Institute. p 471-494.

Notes Authors’ addresses: A.S.R.A.S. Sastri, Indira Gandhi Agricultural University, Raipur, India; V.P. Singh, International Rice Research Institute, Los Baños, Philippines. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Agroclimatic inventory for environmental characterization . . . 231 Agrohydrologic and drought risk analyses of rainfed cultivation in northwest Bangladesh

A.F.M. Saleh, M.A. Mazid, and S.I. Bhuiyan

Drought is a common problem in the northwest region of Bangladesh, where the monsoon season (June-October) receives only about 1,000 mm of rainfall. For drought characterization, long-term (1961-93) weather data were analyzed and the impact of drought on rice establishment and farmers’ management practices was studied for two wet seasons (1994-95). The probabilities of two and three consecutive 5-d droughts occurring during the grain- filling period of transplanted rice (TPR) are 73% and 53%, respectively. In an average year, rainfall may be adequate for transplanting by mid-July. But, twice in ten years, the required rainfall may not be available by 15 August and transplanting may be delayed. The average seasonal relative water supply (RWS) from rainfall is 0.79. Because of late transplanting, the RWS is expected to be only about 0.51, twice in ten years, and can be very detrimental to crop yield. Dry-seeded rice (DSR) may be established early, by the first week of June in an average year, and, twice in ten years, by the third week of June. DSR yields are similar to those of TPR, but DSR matured 1–2 wk earlier than TPR and left a better soil-water regime for the subsequent nonrice crop.

In Bangladesh, more than 50% of the rice areas are rainfed lowland (IRRI 1993) and most of these areas suffer from either flood or drought. In flood-prone areas, farmers minimize yield losses by selecting crops in accordance with flood depth, duration, and timing. In drought-prone areas, however, farmers often ignore the possibilities of drought and sometimes suffer significant yield losses. MPO (1985) estimated that the average yield reduction between irrigated and rainfed (drought-prone) situations for modern variety (MV) aus (premonsoon) rice was 23% and for MV aman (monsoon) rice was 31%. Drought effects are most severe in the northwest region of the country, especially in the Barind Tract, which covers about 0.16 million ha, of which nearly 0.1 million ha is rainfed. The average seasonal rainfall of about 1,000 mm during the five monsoon months (June-October) is the lowest in the country (Manalo 1977) and can

Agrohydrologic and drought risk analyses of rainfed cultivation . . . 233 be classified as low for a rainfed ecosystem (Garrity et al 1986). The situation is exacerbated by drought spells of 1 or 2 wk within a month, which, depending on their timing, can drastically affect crop yield. As both technical and economic constraints limit further irrigation development in Bangladesh, the majority of the farmers in rainfed systems will certainly continue to remain in the drought-prone environment in the foreseeable future. Tech- nologies for drought alleviation are now available and are gradually being accepted and adopted by farmers in rainfed lowlands. Drought alleviation is possible by switch- ing from the traditional transplanting method of crop establishment (TPR) to dry seeding (DSR) in which rice seeds are sown on dry-tilled unsaturated fields early in the season. These fields may subsequently become flooded with the onset of the monsoon rains. Studies on DSR in the Philippines (Saleh and Bhuiyan 1995, Lantican et al 1999) have shown that DSR uses rainfall more efficiently and suffers less drought risk than TPR. On-farm reservoirs for supplementary irrigation have also been used effectively for drought alleviation in the Philippines (Moya et al 1986) and in India (Paul and Tiwari 1994). But, before planning and recommending such interventions for drought alleviation in a rainfed ecosystem, a systematic and quantitative analysis of drought is imperative. This study is an attempt in that direction with the specific objectives of (1) characterizing the area in terms of agrohydrology, (2) studying the nature, extent, and frequency of droughts during the aman season from long-term rainfall data, and (3) studying the relative merits of dry seeding over transplanting of rice in terms of drought alleviation.

Materials and methods Agrohydrologic setting The field study was carried out at Rajabari Union of Godagari Thana of Rajshahi District, in the Barind Tract of northwest Bangladesh. The Barind Tract has an undulating topography with gray terrace soil and an average elevation of 43 m above sea level. The soil texture varies from silt loam to silty clay loam and is poorly drained with a 6–8-cm thick plow pan at 9–11-cm depth (Mazid et al 1998). The soil is low in organic matter (0.8–1.2%) and is acidic (pH from 5.5 to 6.5). Annual rainfall at Godagari averages about 1,300 mm and about 80% of it is confined to the monsoon months of June to September. The average daily evaporation at Rajshahi, located centrally within the Barind Tract, varies from 2.3 mm in January to 6.3 mm in April. The four months when the rainfall exceeds evaporation are June to September. The maximum and minimum temperatures at Rajshahi average about 39 °C and 10 °C, respectively, and occur in April and January. The predominant cropping pattern in the area is transplanted aman rice–fallow. In some areas, chickpea, linseed, barley, and other crops are grown either as a monocrop or as a mixed crop during the rabi (dry) season.

234 Saleh et al Analytical procedure The nature and extent of droughts in the area were studied by using the water balance method to find out the numbers of 5-d water-deficient periods and their continuity (consecutive 5-d periods) during the aman season. In these periods, the rainfall was inadequate to meet the crop water requirement. Since rainfall during one day may be adequate for the entire period considered, the chances of a drought within the period are higher if the period is longer than 5 d. As soil water may be available to the rice roots 2 to 3 d after standing water has disappeared, a shorter duration was not considered. The methodology followed is similar to Thornthwaite’s method of water balance (Steenhuis and Van der Molle 1986, Paul and Tiwari 1992). In this analysis, however, the seepage and percolation loss (S&P) was also taken into account along with evapotranspiration (ET), as S&P is an integral part of the water requirement for lowland rice. The 5-d water balance is written as follows:

Ht = Rt + Ht–1 – ETt – (S&P)t (1)

where H is bund storage, R is rainfall, and subscripts t and t–1 denote time in 5-d time steps (present and previous 5 d, respectively). R, ET, and S&P are all expressed in mm d–1 (in each 5-d time step). Thus, the incidence of either drought or adequacy of water supply was determined by the following criteria:

If (Rt + Ht–1 – ETt ) < 0, then (S&P)t = 0 and Ht = 0; there is drought (2) If ETt < (Rt + Ht–1) < ETt – (S&P)t and then Ht = 0; there is no drought (3) If Rt + Ht–1 – ETt – (S&P)t > 0 and Ht > 0; there is no drought (4) If Rt + Ht–1 – ETt – (S&P)t > Hmax and Ht = Hmax = 20 cm; there is no drought (5)

Water availability during the crop growth season was determined by using the concept of relative water supply (RWS), which is defined as the ratio of water supply to demand by the crop. Thus, RWS for a given period t can be written as

RWSt = [(R/(ET + S&P)]t (6)

An RWS value greater than 1 indicates that the water supply is abundant, whereas a value less than 1 indicates drought. For lowland rice, pan evaporation data are good indicators of crop evapotranspiration (Tomar and O’Toole 1980). Rainfall probability and drought occurrences were analyzed by using gamma distribution (Thom 1968). For drought analysis, 31 years of daily rainfall data (1963-93) of Godagari station, situated about 15 km northwest of the study area, were collected from the Bangladesh Water Development Board as long-term rainfall data were not available for the study site. Daily rainfall and U.S. Class A pan evaporation data for the study seasons were collected from a nearby temporary weather station. Past crop yield data were collected from the Bangladesh Bureau of Statistics (1979-83) and the Depart-

Agrohydrologic and drought risk analyses of rainfed cultivation . . . 235 ment of Agricultural Extension (1985-93) for studying the impact of droughts on crop yield.

Field experiment and farmers’ survey The field study was carried out during the 1994 and 1995 aman (July-November) season. Crop production practices of 50 randomly selected farmers were surveyed during the study period using a predesigned questionnaire to (1) gain qualitative information on farmers’ practices in crop establishment, crop management, and production limitations and (2) collect quantitative information on timings of farming activities, input use, and productivity of the aman crop. Field water status and water use of 30 farms were closely monitored by install- ing 100-cm-long PVC tubes (50 cm perforated) 20 cm above the ground surface and some below the ground surface. The water-level readings in the PVC tubes (standing or perched water level) were taken every other day. Changes in standing-water levels during rainless days were used to estimate S&P loss after deducting pan evaporation. Field experiments on effect of crop establishment method (DSR and TPR) and seeding/transplanting date on crop yield were carried out at Rajabari in a split-plot design. For transplanting, seedlings were raised in the wet bed on the same day of seeding of DSR and then 30-d-old seedlings were transplanted in puddled soil. The transplanting schedule was delayed by 15 d in 1994 because of inadequate rainfall for puddling.

Results and discussion Farming activities in relation to rainfall Figure 1 shows the timing of farming activities for 50 transplanted rice-growing (the traditional practice) farmers sampled during the 1994 and 1995 aman seasons. Farm- ers began the first plowing and sowing activities earlier in 1994 because of the 60- mm rainfall in April. But transplanting could not be carried out until July, when there was enough rainfall for land preparation. In 1995, the first plowing and seeding started

Farmers (%) Rainfall (mm) 100 500 A B 80 First plow 400 Seeding 60 300 Transplant Harvest 40 200

20 100

0 0 Apr May Jun Jul Aug Sep Oct Nov Dec Apr May Jun Jul Aug Sep Oct Nov Dec Month Fig. 1. Progress of farming activities during the 1994 (A) and 1995 (B) aman season at Rajabari, Godagari, Rajshahi.

236 Saleh et al late because of low rainfall at the beginning of the season. Heavy rainfall from the beginning of June resulted in quick completion of farming activities and transplanting was completed by the end of July. The average time difference between the first plowing and transplanting was 62 d in 1994. The duration was only 31 d in 1995 due to more favorable rainfall conditions. The average time difference between seeding and transplanting was 45 d in 1994 and 40 d in 1995. In both years, 50% of the farmers completed transplanting by mid-July but harvesting was completed by mid-November in 1994 and the end of November in 1995. Heavy rain in mid-November (50 mm as shown in Figure 1) delayed harvesting in 1995. The survey data showed that, for 65% of the farmers, the preferred times of transplanting and harvesting are mid-July and mid-November. Thus, the 120 d between 15 July and 15 November were considered as the preferred field duration of the crop in further analysis. The average crop field duration was 119 d in 1994 and 130 d in 1995. During the two months required from first plowing to transplanting by 50% of the farmers (15 May to 15 July), the amount of rainfall in 1994 was 380 mm. During the same period in 1995 (in this case 19 June to 20 July), rainfall was 391 mm. Hence, it was assumed in further analysis that, in the study area, the farmers required at least 400 mm of rainfall to complete land preparation before transplanting. About 87% of the farmers grew modern rice varieties, whose average yield was 3.7 t ha–1 and range was 1.8 to 4.7 t ha–1. The average use of fertilizer was high: 100 kg N ha–1, 17 kg P ha–1, and 16 kg K ha–1. Nitrogen application in two splits was most common (44%), followed by three splits (36%). About half the farmers applied organic fertilizer (manure) during land preparation at an average rate of 3.5 t ha–1. About one-third of the farmers grew a second crop following the harvest of aman rice. The major second crop was chickpea, which was grown by about two-thirds of the farmers.

Drought analysis Drought in the study area was characterized through probability analysis of rainfall during the crop field duration (15 July to 15 November). The 50% (rainfall equal to or exceeding the specified amount in 5 out of 10 years) and 80% dependable (rainfall equal to or exceeding the specified amount in 8 out of 10 years) crop field duration rainfalls are 745 mm and 597 mm, respectively. Water balance with 5-d time steps was carried out for the crop field duration using equations 1–4 and with 31 years of rainfall data. The field duration was divided into three stages: vegetative, reproductive, and grain filling, each stage being 40 d long. The average pan evaporation during the season was 3 mm d–1 and the average S&P loss was 7 mm d–1. The numbers of 5-d drought periods and consecutive 5-d drought periods during each growth stage for the past 31 years are given in Tables 1 and 2. They are most frequent during the grain-filling stage. On average, about seven 5-d droughts are expected to occur during crop field duration. But, as these are not continuous, they are not expected to seriously affect yield. About three consecutive

Agrohydrologic and drought risk analyses of rainfed cultivation . . . 237 Table 1. Number of 5-d droughts during crop field duration.

Cycle Cycle Year Total Year Total Vege- Repro- Grain Vege- Repro- Grain tative ductive filling tative ductive filling

1963 11131980 0314 1964 02241981 0358 1965 3 3 5 11 1982 4 3 5 12 1966 14491983 1135 1967 2 3 5 10 1984 1124 1968 2 3 5 10 1985 4 4 2 10 1969 11351986 2158 1970 33281987 1157 1971 11351988 2349 1972 2 4 4 10 1989 2125 1973 40371990 1236 1974 22371991 1056 1975 2 5 4 11 1992 2237 1976 10561993 2024 1977 2215 1978 2248Average 1.7 2.0 3.4 7.1 1979 1247

Table 2. Number of consecutive 5-d droughts during crop field duration.

Cycle Cycle Year Seasonal Year Seasonal Vege- Repro- Grain maximum Vege- Repro- Grain maximum tative ductive filling tative ductive filling

1979 1 2 4 7 1979 1 2 4 7 1963 0 0 0 0 1980 0 0 0 0 1964 0 0 0 0 1981 0 0 5 6 1965 0 3 5 5 1982 3 2 5 7 1966 0 3 3 6 1983 0 0 3 3 1967 0 2 5 7 1984 0 0 2 2 1968 0 3 5 5 1985 4 3 0 4 1969 0 0 3 3 1986 0 0 5 5 1970 3 2 0 3 1987 0 0 5 6 1971 0 2 3 3 1988 0 3 4 4 1972 0 4 4 4 1989 2 0 2 2 1973 2 0 2 2 1990 0 2 3 3 1974 0 2 3 3 1991 0 0 5 5 1975 2 2 2 2 1992 2 0 2 2 1976 0 0 5 5 1993 0 0 0 0 1977 0 0 0 0 1978 2 0 4 4 Average 0.7 1.1 2.9 3.4 1979 0 0 4 4

238 Saleh et al Probability (%) Probability (%) 100 Vegetative Vegetative 100 Reproductive 80 Reproductive Grain filling 80 Grain filling 60 60 40 40 20 20 0 2345 2345 2345 0 Consecutive 5-d drought periods 12345 12345 12345 Number of 5-d drought periods

Fig. 2. Probability of 5-d droughts during the Fig. 3. Probability of consecutive 5-d droughts crop season. during the crop season.

5-d droughts can be expected during the grain-filling stage compared with none during both the vegetative and reproductive stages. Figures 2 and 3 show the probability analysis of 5-d droughts and consecutive 5-d droughts. The probability of two 5-d droughts (not consecutive) during the grain- filling stage is more than 80%. The probability of 10-d and 15-d droughts (two and three consecutive 5-d droughts) during the grain-filling stage is 73% and 53%, respectively. Thus, a 2-wk period without rain is expected once in two years and can be detrimental to crop yield. Figures 4 and 5 show a scatter diagram of seasonal rainfall (15 July-15 Novem- ber) and total numbers of 5-d droughts and consecutive 5-d droughts. There is a fair correlation (r = 0.56 and significant at the 1% level) between total seasonal rain and total number of 5-d droughts. Seasonal rainfall and total number of consecutive 5-d droughts, however, are uncorrelated. The 50% and 80% dependable 5-d rainfalls for May-November were determined and are shown in Figure 6, which shows that in the study area little or no rainfall is expected after early October. Since early October is the beginning of the grain-filling stage, the crop is vulnerable to damage by drought. Figure 7 shows the probability of getting the 400 mm of rainfall required for transplanting at different times at the beginning of the crop season. In an average year (50% probability), the 400 mm of rainfall required for transplanting can be expected by 15 July. Twice in ten years (80% probability), however, this amount may not be available before 15 August and transplanting may be delayed by one month. A water adequacy analysis was carried out for crop field duration by calculating the relative water supply using equation 5 in each 5-d period. The average seasonal RWS value at 50% probability, when transplanting is completed by 15 July (average year), is 0.79. This means that, even in an average year, the water supply from rainfall is inadequate to meet the crop water requirement. If transplanting is delayed because of inadequate rainfall at the beginning of the season (completed after

Agrohydrologic and drought risk analyses of rainfed cultivation . . . 239 Seasonal rainfall (mm) Seasonal rainfall (mm) 1,400 1,400 1,200 Y = 1,083 – 45.17X 1,200 (r = 0.56; significant 1,000 at 1% level) 1,000 800 800 600 600 400 400 200 200 0 0 02468101214161820 02468101214161820 Total 5-d drought periods Total consecutive 5-d drought periods

Fig. 4. Scatter diagram of seasonal rainfall and Fig. 5. Scatter diagram of seasonal rainfall and total 5-d droughts. total consecutive 5-d droughts.

Rainfall (mm) 50 50% 40 80% 30

0 5 May 5 Jun 5 Jul 5 Aug 5 Sep 5 Oct 5 Nov 5-d periods by month Fig. 6. 5-d rainfall at 50% and 80% probabilities.

Cumulative rainfall (mm) 800 700 50% 600 80% 500 400 300 200 100 0 Jun Jul Aug Sep 5-d periods by month Fig. 7. Cumulative rainfall at 50% and 80% probabilities.

240 Saleh et al 15 August), which may happen twice in ten years, the RWS value drops to 0.51. In such cases, only about 50% of the crop water requirement would be available from rainfall and crop yield would be seriously affected.

Yield-rainfall relationship The average yields obtained in 1994 and 1995 were 3.6 and 3.5 t ha–1 and the corresponding crop field duration rainfalls were 672 and 790 mm, respectively. Both the yield and rainfall in 1994 were below the average for the study area by 0.1 t ha–1 and 73 mm, respectively. In 1994, the October rainfall of 204 mm was exceptionally high (probability of 15%) and probably helped keep yield close to the average. Drought during the grain-filling stage in 1995 was mainly responsible for the decrease in yield. The total rainfall during October was only 10 mm (probability of 5%). However, 195 mm of rainfall during the last five days of September meant that all the fields had standing water until the middle of October and water stress developed only in the second half of the month. Thus, serious drought did not occur in October and yield was not seriously affected. Total rainfall during crop field duration and the corresponding yields for 1979- 93 are plotted in Figure 8. The poor correlation between the two is not unexpected as the total rainfall during crop field duration can be high, but droughts can still occur during the critical stages of the crop. Also, changes in yield depend not only on water but also on other inputs. Changes in input use are expected because of changes and uncertainties in water availability from year to year. Moreover, the farmers also changed varieties every 3 to 4 years, and the most common reason cited was to increase or maintain yields. Droughts in the study area are expected during October. Hence, the yields (Y) were plotted against the corresponding October rains (R0) to obtain a nonlinear regression of the form (Fig. 9)

Yield (t ha–1) Yield (t ha–1) 5 5

4 4

3 3 0.4 Y = 1.6774 + 0.326 R0 (r2 = 0.47; significant at 1% level) 2 2

1 1

0 0 0 200 400 600 800 1,0001,2001,400 1,600 0 50 100 150 200 250 300 July-October rainfall (mm) October rainfall (mm) Fig. 8. Scatter diagram of yield and total sea- Fig. 9. Scatter diagram of yield and October sonal rainfall. rainfall.

Agrohydrologic and drought risk analyses of rainfed cultivation . . . 241 0.4 Y = 1.674 + 0.326 R0 (6) with a coefficient of determination (r2) of 0.47 and significant at the 1% level. This signifies the importance of rainfall in October for rice yield in the study area.

Drought alleviation: potentials of dry-seeded rice From the drought analysis, it can be concluded that the study area is drought-prone at two stages: at the beginning of the season, which can cause a delay in transplanting, and at the beginning of the grain-filling stage, which can drastically reduce crop yield. It has already been mentioned that DSR can be established early and can be very effective in drought alleviation. With DSR, drought at the end of the season is not expected to affect yield because the earlier established crop would be near the harvesting stage by the time drought set in in October. Moreover, earlier harvested dry- seeded rice would leave a favorable soil-water regime for the subsequent nonrice crop. A 3-y experimental study (1994-96) on comparative productivity of DSR and TPR at Rajabari by Mazid et al (1998) showed that DSR yields were similar to or slightly better than those of TPR for all seeding/transplanting dates (Table 3). The experiments also showed that DSR matured about 1–2 wk earlier than TPR. Because of the better soil-water regime, the yield of chickpea planted after DSR was higher than that after TPR. Studies in the Philippines have shown that only about 150 mm of rainfall is required for crop establishment through DSR compared with about 600 mm for TPR (Saleh and Bhuiyan 1995). Rainfall probability analysis has shown that, if 150 mm of rainfall is required for establishment of DSR (as in the Philippines with a similar soil

Table 3. Effect of time of seeding (dry-seeded rice, DSR) and transplanting (transplanted rice, TPR) on grain yield (t ha–1) of high-yielding aman rice at Rajabari, Rajshahi, 1994-96.

Method of Time of seeding/transplanting establishment 1 June 16 June 1 July 16 July 1 August 15 August

1994 DSR – 3.11 3.50 3.45 2.78 – TPR – – – – 2.44 2.77

1995 DSR 2.60 2.75 2.85 2.46 – – TPR – – 2.52 2.45 2.93 2.37

1996 DSR 3.43 3.93 – – – – TPR – – 3.40 3.81 – –

242 Saleh et al texture), then once in two years (50% probability) crop establishment may be possible by the first week of June. Twice in ten years (80% probability), however, rainfall may be inadequate for crop establishment before the third week of June. Present DSR yield data (Table 3) show that even such a delay is not expected to have any effect on crop yield.

Conclusions Droughts commonly occur in the study area at the beginning of the season prior to transplanting and from the beginning of the grain-filling stage. In an average year, about 80% of the crop water requirement is available from rainfall. In two out of ten years, rainfall supplies only about 50% of the total water requirement because of late transplanting due to early drought. The probabilities of at least one 10-d and 15-d drought occurring during the grain-filling period are 73% and 53%, respectively. Al- though the total seasonal rainfall is not related to crop yield, there is a fair correlation between rainfall in the grain-filling stage and crop yield. Experimental studies on drought alleviation through dry seeding have been encouraging and the productivity of the dry-seeded rice–chickpea cropping pattern has been higher than that of the traditional practice. Further research on biophysical and socioeconomic constraints that inhibit the wider adoption of dry-seeded rice for drought alleviation as a substitute for traditional transplanted rice is needed.

References Garrity DP, Oldeman LR, Morris RA, Lenka D. 1986. Rainfed lowland rice ecosystems: characterization and distribution. In: Progress in rainfed lowland rice. Manila (Philippines): International Rice Research Institute. p 3-23. IRRI (International Rice Research Institute). 1993. World rice statistics. Manila (Philippines): IRRI. Lantican MA, Lampayan RM, Bhuiyan SI, Yadav MK. 1999. Determinants of improving productivity of dry-seeded rice in rainfed lowlands. Exp. Agric. 35:127-140. Manalo EB. 1977. Agro-climatic survey of Bangladesh. Bangladesh Rice Research Institute and International Rice Research Institute. 360 p. Mazid MA, Mollah MIU, Mannam MA, Elahi NE, Wade LJ. 1998. Increasing productivity through rainfed rice-chick pea cropping system in high Barind tract of Bangladesh. Pa- per presented at RLRRC Planning and Review Meeting, Bangladesh Rice Research In- stitute, Gazipur, Bangladesh. Moya TB, de la Vina WC, Bhuiyan SI. 1986. The potential of on-farm reservoir use in increasing productivity in rainfed areas. Philipp. J. Crop. Sci. 2(2):125-132. MPO (Master Plan Organization). 1985. Crop production limitations in Bangladesh. Technical Report No.1, Ministry of Irrigation, Water Development, and Flood Control, Govern- ment of Bangladesh. Paul DK, Tiwari KN. 1992. Agricultural drought analysis for Hazaribagh, Eastern India. Int. Rice Res. Newsl. 17(6):32-33.

Agrohydrologic and drought risk analyses of rainfed cultivation . . . 243 Paul DK, Tiwari KN. 1994. Rainwater storage systems for rainfed ricelands of eastern India: results from research in Hazaribagh District. In: Bhuiyan SI, editor. On-farm reservoir systems for rainfed ricelands. Manila (Philippines): International Rice Research Insti- tute. p 127-139. Saleh AFM, Bhuiyan SI. 1995. Crop and rain water management strategies for increasing productivity of rainfed lowland rice. Agric. Syst. 48(3):259-276. Steenhuis TS, Van der Molle WH. 1986. The Thornthwaite-Mather procedure as a simple engineering method to predict recharge. J. Hydrol. 84:221-229. Thom HCS. 1968. Direct and inverse tables of gamma distribution. Environmental Data Ser- vice, U.S. Department of Commerce. Tomar VS, O’Toole JC. 1980. Water use in lowland rice cultivation in Asia: a review of evapotranspiration. Agric. Water Manage. 3:83-106.

Notes Authors’ addresses: A.F.M. Saleh, Professor, Institute of Flood Control and Drainage Research, Bangladesh University of Engineering and Technology, Dhaka 1000, Bangladesh; M.A. Mazid, Principal Agronomist and Head, Rajshahi Regional Station, Bangladesh Rice Research Institute, Rajshahi 6212, Bangladesh; S.I. Bhuiyan, IRRI Liaison Scientist, GPO Box 64, Ramna, Dhaka 1000, Bangladesh. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

244 Saleh et al Characterizing biotic stresses 246 Jahn et al Characterizing biotic constraints to production of Cambodian rainfed lowland rice: limitations to statistical techniques

G.C. Jahn, Pheng Sophea, Pol Chanthy, and Khiev Bunnarith

Because rainfed lowland rice (RLR) accounts for 86% of the rice-growing area of Cambodia, it is important to investigate the biotic constraints to rice in this ecosystem. Over 3 years, a total of 73 RLR fields in Cambodia were studied to determine the effect of pests on destabilizing Cambodian RLR yields, which pests and pest combinations affect RLR yields, and how cropping practices affect pest levels. Pests included insects, weeds, and diseases of rice. Pesticides were not used in any of the fields in this study. Information on crop characteristics and biotic constraints was gathered at four crop development stages: tillering, booting, milk, and maturity. The incidence of damage caused by insects and diseases was recorded from 10 hills chosen randomly from each field. Weed infestation was recorded as the percentage weed cover in three 1-m2 areas. The rice yield of each field was estimated from an average of three randomly selected 10-m2 areas. Corre- spondence analysis was used to characterize the patterns of cropping practices, pest infestations, environmental conditions, and yields. The results of this analysis generated testable hypotheses about the factors contributing to pest problems. Fields of early duration rice tended to have low levels of hispa and high levels of Pentatomids, while late-duration fields had high hispa and low Pentatomid levels. Fields without standing water had higher than average levels of weeds, cutworm, hispa, and Pentatomids. Brown spot and narrow brown spot were the only diseases observed frequently enough to make inferences about their relation to cropping practices. As a component of systems research in rice plant protection, this study assisted in predicting the effects of cropping practices on pest infestations. These techniques have several limitations, however: the danger of drawing false conclusions, difficulties in interpreting results, insufficient attention to the soil type and relative rates of fertilizer, the inability to capture time adequately as a variable, the lack of information on the relative contribution of pests to variation in yield data, the incomplete coverage of pests, and the fallibility of observers. Each limitation is discussed in detail. The results of this study contributed to an assessment of relative pest importance and thereby helped prioritize research to develop integrated pest management recommendations.

Characterizing biotic constraints to production . . . 247 Rainfed lowland rice (RLR) accounts for 86% of the rice-growing area of Cambodia (Javier 1997). Cambodian rice productivity is the lowest in Asia due to the high percentage of RLR. The reasons for low RLR productivity include erratic rainfall, poor soils, low seed quality, socioeconomic constraints, and pest damage (Javier 1997). Cambodian rice farmers often cite poor water availability and pests as the major causes of low and unstable yields (Jahn et al 1997). This is understandable since poor soils, low seed purity, and socioeconomic factors (such as lack of farm machinery) are ever-present constraints to yield that will not completely destroy a crop. In contrast, droughts, floods, and pest outbreaks can visibly reduce the yields of a farmer’s field compared to previous years. From 1994 to 1996, a total of 73 RLR fields in Cambodia were studied to determine the effect of pests on RLR yield stability, which pests and pest combinations affect yields, and how cropping practices affect pest levels. Systematic survey data on pest losses are an important part of the information needed to quantify the risk of pest injuries (Cohen et al 1998). The Cambodia-IRRI-Australia Project (CIAP) Integrated Pest Management (IPM) Program uses this information to determine which pests are the most important, which helps prioritize research. We also use this information to predict the effect of cropping practices on pest infestations and develop IPM recommendations. Using databases of categorical information to characterize the patterns of cropping practices, pest infestations, environmental conditions, and yields is an important component of CIAP’s systems research in rice plant protection (CIAP 1998).

Materials and methods Pest constraints to RLR yields were characterized based on the methods of Savary et al (1995, 1996) developed at the International Rice Research Institute (IRRI). “Pests” included insects, weeds, and diseases. The data collection methods of Savary et al (1996) were revised to reflect the conditions and pests of the Cambodian RLR ecosystem. The observations were restricted to certain pests. The selection of pests included in the study was based on their perceived importance according to farmers, extension workers, and scientists. Farmer interviews (Jahn et al 1997), pest collections, and practical considerations (e.g., ease of recognizing the pest or the damage) were used to determine which pests to include in the study of pest constraints to RLR production. Over 3 years, 1994 to 1996, a total of 73 RLR farmers’ fields (>0.3 ha each) were surveyed in Phnom Penh, Svay Rieng, and Takeo provinces. All surveys were conducted in the wet season. Pesticides were not used in any of the fields in this study. Information on the crop characteristics and the biotic constraints was gathered at four crop development stages: tillering, booting, milk, and maturity. The yield of each field was estimated by crop cuts from an average of three randomly selected 2 m × 5 m areas at ripening stage, converted to t ha–1 and adjusted to 14% moisture. The incidence of pests was recorded from 10 hills chosen haphazardly from each field. Weed infestation was assessed on the basis of the percentage weed cover in three 1-m2 areas that included sampling hills number 3, 6, and 9. Rat damage was not

248 Jahn et al included in these studies. Later studies, however, indicated the importance of rats and they were included in follow-up studies from 1997 to 1999 (CIAP 1998, 1999). The levels of gall midges and stem borers were measured by the number of infested tillers per hill. Damage by whorl maggots, leaffolders, hispa, and cutworms was measured by the number of damaged leaves per hill. Planthoppers, rice bugs, and other sucking insects were directly counted from 10 hills, whereas leafhoppers and beneficial insects were counted from 5 sweep-net samples per field per visit. Each sweep-net sample consisted of 10 sweep-net strokes. Rice bugs were counted directly rather than assessed on the basis of unfilled or partially filled grains. This type of damage could be caused by other pests or even by thermal damage (Satake and Yoshida 1978, Sheehy et al 1998). Rice diseases were quantified by counting the number of damaged leaves, tillers, or panicles per hill depending on the symptoms of the particular disease. A team of two to three trained observers took approximately 1 h to make (and record) all of the necessary observations on each visit to a farmer’s field.

Data analysis The analysis proceeded in five steps: (1) determination of average injury levels of each pest for each crop stage, (2) categorization of the variables (i.e., pest, cropping practices, and yield) into classes, (3) testing for independence of paired variables (i.e., pest levels, cropping practices, and yields) in contingency tables, (4) clustering of cropping practices and pest profiles, and (5) development of contingency tables and correspondence analysis. Data were analyzed using the computer programs Ex- cel® and STAT-ITCF®. Average injury levels and selection of peak period. The percent damage in- flicted by each particular type of pest was averaged over the fields at successive crop development stages and graphed (Figs. 1 to 7) to indicate changes in injury and pest levels over time. Only data from the peak average injury level for each pest were analyzed thereafter (Savary et al 1996). Transformation of quantitative variables into classes. All levels within a class contained a similar number of rice fields, i.e., classes were made as balanced as possible. Yields and injury levels were placed into categories of low, medium, and high so that qualitative and quantitative variables could be analyzed simultaneously in

Weed infestation rating 1

0 Tillering Booting Milky Ripening Crop stage Fig. 1. Average weed infestation in 73 rainfed lowland rice fields. All weeds were below the rice canopy; ■ = average level of weeds below rice canopy.

Characterizing biotic constraints to production . . . 249 Average % tillers affected 4 Gall midge 3 Deadhearts Whiteheads 2

0 Tillering Booting Milky Ripening Crop stage Fig. 2. Average levels of tiller damage by gall midges and stem borers in 73 rainfed lowland rice fields. Deadhearts and whiteheads are both symptoms of stem borer damage.

Average % leaf affected

Whorl maggot Leaffolder 20 Hispa Cutworm Leaf-feeding insects 15

0 Tillering Booting Milky Ripening Crop stage Fig. 3. Average levels of leaf damage by whorl maggot, leaffolder, hispa, cutworm, and other leaf-feeding insects in 73 rainfed lowland rice fields.

Average insect number 2 Planthoppers Rice bugs Other sucking insects

0 Tillering Booting Milky Ripening Crop stage Fig. 4. Average levels of planthoppers, rice bugs, and other sucking insects in 73 rainfed lowland rice fields.

250 Jahn et al Average number of insects per sweep 40 Leafhoppers 30 Pests Natural enemies 20

0 Tillering Booting Milky Ripening Crop stage Fig. 5. Average number of leafhoppers, pests, and natural enemies per sweep-net sample of 40 rainfed lowland rice fields.

Average % leaf affected

Brown spot 40 Bacterial leaf streak Narrow brown spot 30 20 10 0 Tillering Booting Milky Ripening Crop stage Fig. 6. Average incidence of foliar diseases at different crop stages in 73 rainfed lowland rice fields.

Average % tiller or panicle affected 1 Sheath blight Sheath rot

0 Tillering Booting Milky Ripening Crop stage Fig. 7. Average percentage of tiller damaged by two fungal diseases at different crop stages in 73 rainfed lowland rice fields. contingency tables. Tables 1 to 3 summarize the classes, boundaries, and frequency distribution of the qualitative and quantitative variables. Tests of independence for relationships among pests, cropping practices, and yields. Tests for independence were performed only on variables with balanced classes. The cells of contingency tables (particular pest × particular cropping practice) were filled in with the number of fields that matched each pair of classification criteria. Chi-square tests were used to interpret relations between any paired variables. Rela-

Characterizing biotic constraints to production . . . 251 Table 1. Categorization of cropping practices into classes.

Variable Label Class boundaries Individuals (no.)

Crop duration CD1 Early 21 CD2 Medium 34 CD3 Late 18 Fertilization F1 0–30 kg ha–1 of mineral fertilizer (MF) 17 F2 >30–150 kg ha–1 of MF 24 F3 >150 kg ha–1 of MF 19 F4 Manure 13 Yield Y1 Very low yield (0–1.85 t ha–1)19 Y2 Low yield (>1.85–2.51 t ha–1)17 Y3 Medium yield (>2.51–3.31 t ha–1)19 Y4 High yield (>3.31 t ha–1)18 Water status WST1 No or little water (1–7 cm) 32 WST2 Adequate water (>7–8 cm) 19 WST3 Too much water (>8 cm) 22

tionships with P <0.05 are summarized in Table 4. Pests without any detectable relation to any cropping practice were omitted from the table. Categorize cropping practices and pests into clusters. Four clusters of cropping practices (CP) and pest profiles (PE) were generated by STAT-ITCF using cluster analysis (Tables 5 and 6). Only pest variables related to at least one of the CP clusters were included in the analysis. Contingency tables and correspondence analysis. Contingency tables were constructed to show the frequency distribution of RLR fields as a function of yield and cropping practices, and as a function of yield and pest profiles (Table 7). The two (4 × 4) contingency tables (PE × Y and CP × Y) were linked to produce a 4 × (4 + 4) data matrix for correspondence analysis and graphed on factorial axes (Savary et al 1996) (Table 8, Fig. 8). Two factorial axes were sufficient to display the positions of row and column points. The display could be completed with three axes, but the interpretation is more complex.

252 Jahn et al Table 2. Categorization of weed and arthropod data into classes.

Variable Class boundaries Individuals (no.)

Weeds (WB) Very low (score 0–0.2) 25 Medium (>0.2–0.4) 19 High (>0.4–2.0) 29 Gall midge (GM) Low (0–0.2%) 25 Medium (>0.2–1.0%) 23 High (>1–59.3%) 25 Deadhearts (DH) None (0%) 42 Low (>0–0.5%) 14 Medium (>0.5–4.1%) 17 Whiteheads (WH) Low (0–0.3%) 24 Medium (>0.3–1.0%) 26 High (>1–4.9%) 23 Whorl maggot (WM) Low (0–1.1%) 25 Medium (>1.1–2.9%) 24 High (>2.9–14%) 24 Leaffolder (LF) Low (0–0.2%) 25 Medium (>0.2–0.7%) 23 High (>0.7–4.2%) 25 Hispa (HP) Low (0–0.9%) 25 Medium (>0.9–4.8%) 24 High (>4.8–12.8%) 24 Cutworm (CW) Low (0.8–8.1%) 25 Medium (>8.1–15.4%) 24 High (>15.4–51.1%) 24 Leaf-feeding insects Low (0–0.1%) 25 (LFI) Medium (>0.1–2.5%) 23 High (>2.5–32%) 25 Planthoppers (PH) Low (0–0.1 PH hill–1)33 Medium (>0.1–0.4) 21 High (>0.4–14.8) 19 Rice bug None (0 RB hill–1)56 Low (>0–0.1) 9 Medium (>0.1–0.6) 8 Other sucking insects Low (0–0.1 SI hill–1)25 (SI) Medium (>0.1–0.9) 24 High (>0.9–16.3) 24 Leafhoppers (LH) Low (3.1–14.4 LH sweep net–1)14 Medium (>14.4–31.1) 13 High (>31.1–104.5) 13 Pest (PS) Low (3.1–8.8 PS sweep net–1)14 Medium (>8.8–16.8) 13 High (>16.8–65.7) 13 Natural enemies (NE) Low (6.2–15 NE sweep net–1)14 Medium (>15–19.8) 13 High (>19.8–51.3) 13

Characterizing biotic constraints to production . . . 253 Table 3. Categorization of disease data into classes.

Variable Class boundaries Individuals (no.)

Brown spot (BS) Low (0–12.3%) 25 Medium (>12.3–32.1%) 24 High (32.1–93.7%) 24 Bacterial leaf streak (BLS) None (0%) 44 Low (>0–5.5%) 15 High (>5.5–30%) 14 Narrow brown spot (NBS) None (0%) 27 Low (>0–4.2%) 23 High (>4.2–35.9%) 23 Sheath blight (SHB) None (0%) 61 Low (>0–0.6%) 6 Medium (>0.6–2.7%) 6 Sheath rot (SHR) None (0%) 51 Low (>0–0.5%) 11 Medium (>0.5–4.1%) 11

Table 4. Relationship of management practices to pests in rainfed lowland rice.a

Pentatomids Narrow Management (stink bugs Whorl Brown brown practice Weeds Cutworm Hispa and maggot spot spot black bugs)

Cultivar duration Late – ? + – NNN Medium NNNNNNN Early + N – + N – N Water level No water + + + + N ? ? 1–5 cm NN–– N?? >5 cm ––NNN?? Chemical fertilizer 0–30 kg ha–1 ? – N N N?N 31–150 kg ha–1 ? N N N N+N >150 kg ha–1 ?+?N +–– a + = increases pest population, – = decreases pest population, N = no effect, ? = unknown, i.e., more data required.

254 Jahn et al Table 5. Properties of individual clusters of cropping practices (CP).

Clusters Crop duration Fertilizer Water status

CP1 (25 fields) Late varieties and some No to low amount of Little water medium varieties mineral fertilizer

CP2 (25 fields) Medium varieties and High mineral fertilizer Good water some early varieties application and manure management

CP3 (11 fields) Medium rice varieties Medium to high levels of Too much water mineral application CP4 (12 fields) Early varieties No to low amount of Variable mineral fertilizer

Table 6. Properties of the pest profile (PE) clusters.

Pest profile Characterized by high levels of these pests

PE1 (29 fields) Weeds, whorl maggots, hispa, narrow brown spot, and sucking insects PE2 (22 fields) Narrow brown spot PE3 (13 fields) Brown spot, gall midges, and sucking insects PE4 (9 fields) Gall midge, cutworm, brown spot, and weeds

Table 7. Contingency table showing the number of fields matching each yield (Y) profile and each cluster of cropping practices (CP) and pest profiles (PE).

Yield profiles Cluster Y1 Y2 Y3 Y4

CP1 13 5 3 4 CP2 1 5 9 10 CP3 3 2 4 2 CP4 2 5 3 2 PE1 9 5 9 6 PE2 1 7 6 8 PE3 4 3 2 4 PE4 5 2 2 0

Characterizing biotic constraints to production . . . 255 Table 8. Numerical output of the correspondence analysis of Table 7.

Eigen values (variances on principal axes) Contribution to total variances Cumulated percentage

0.1844 79.7% 79.7% 0.0269 11.6% 91.3%

Principal axis 1 Principal axis 2

Columns Squared Relative Squared Relative Coordinates cosine contribution Coordinates cosine contribution (%) to axis 1 (%) to axis 2

Y1 +0.702 0.992 69.5 –0.041 0.003 1.6 Y2 –0.102 0.109 1.3 +0.280 0.824 67.9 Y3 –0.227 0.467 7.3 –0.174 0.272 29.1 Y4 –0.405 0.789 21.9 –0.038 0.007 1.3

Principal axis 1 Principal axis 2

Rows Squared Relative Squared Relative Coordinates cosine contribution Coordinates cosine contribution (%) to axis 1 (%) to axis 2

CP1 +0.588 0.935 32.1 +0.048 0.006 1.4 CP2 –0.550 0.930 28.0 –0.143 0.063 12.9 CP3 +0.039 0.021 0.1 –0.184 0.487 9.5 CP4 –0.116 0.065 0.6 +0.367 0.656 41.0 PE1 +0.107 0.277 1.2 –0.159 0.615 18.7 PE2 –0.488 0.900 19.5 +0.159 0.095 14.1 PE3 +0.077 0.088 0.3 +0.083 0.102 2.3 PE4 +0.738 0.926 18.2 +0.006 0.000 0.0

Simultaneous representation of rows (observations) and columns (variables) 0.4 CP4 B

D Y2

Axis 2 A PE2 PE3 CP1 PE4 0 Axis 1 –0.8 0 C 0.8 1.6 Y4 Y1 PE1 CP2 Y3 CP3

–0.4 Fig. 8. Graphical display provided by correspondence analysis of the data matrix in Tables 7 and 8. The cropping practice (CP), yield (Y), and pest (PE) clusters can be grouped into four domains described in Table 9.

256 Jahn et al Results Relations among pests and cropping practices Late-duration rice fields tended to have high levels of hispa and low levels of Pentatomids, whereas early duration fields were the opposite. Fields without standing water at the four key crop stages had higher than average levels of weeds, cutworm, hispa, and Pentatomids. Brown spot and narrow brown spot were the only diseases observed frequently enough in the RLR production systems to make inferences about their relation to cropping practices. Other diseases, such as sheath blight, are associated with intensive and high-input production systems (Cu et al 1996). Fields receiving more than 150 kg fertilizer ha–1 had lower than average levels of brown spot and narrow brown spot (Table 4). The fertilizer associations would depend, of course, on the type of fertilizer used, not only the rates. Later experiments addressed the association between type of fertilizer and pest levels (CIAP 1998, 1999).

Correspondence analysis The first and second axes of the graphical display of the correspondence analysis (Fig. 8) accounted for 79.7% and 11.6% of total inertia, respectively. Therefore, the first two axes provided a good overall view of the numerical output of the correspondence analysis (Table 8), as 91.3% of the total inertia was represented. Axis 1 represents the gradient of decreasing levels of rice yields. It involves the contributions of Y1, Y4, CP1, CP2, PE2, and PE4 (Tables 5, 6, and 7). Axis 2 involves the contributions of Y2, Y3, CP3, CP4, PE1, and PE3. Four domains were generated from Figure 8—domain A (Y1, CP1, PE4), domain B (Y2, CP4, PE3), domain C (Y3, CP3, PE1), and domain D (Y4, CP2, PE2)—by correspondence analysis using STAT-ITCF (Savary et al 1996). Table 9 summarizes the characteristics of each domain. Correspondence analysis of Table 7 indicates that CP1 (Table 5) tends to have lower yields, whereas CP2 tends to have medium to higher yields. Fields with a pest profile matching PE2 (Table 6) are more frequently associated with high yields, but fields matching PE4 more frequently have low yields (Table 8, Fig. 8).

Discussion The effects of cropping practices on the levels of pest infestation (Table 4) are based on chi-square tests of independence. While this type of analysis indicates whether a particular practice (or condition) tends to be associated with high or low levels of particular pests, it does not measure the relative strength of that association (or possibly effect). The interactions between combinations of cropping practices, multiple pests, and yields (Fig. 8, Table 9) are based on correspondence analysis. In this case, the relative strength of these relationships is indicated by the distance between clusters in Figure 8.

Characterizing biotic constraints to production . . . 257 Table 9. Characteristics of the domains derived from Figure 8.

Domain Clusters Cropping practices and yields Pest constraints

A Y1, CP1, PE4 Late-duration rice varieties Weeds, gall midge, cutworm, and some medium varieties and brown spot Low application of mineral fertilizer Diverse water depths Very low yield B Y2, CP4, PE3 Early rice varieties Brown spot, gall midge, Low application of mineral sucking insects fertilizer Diverse water depths Low yield C Y3, CP3, PE1 Medium rice varieties Weeds, whorl maggot, High application of mineral hispa, sucking insects, fertilizer and gall midge Too much water Medium yield D Y4, CP2, PE2 Medium cultivars and some Narrow brown spot early varieties High application of mineral fertilizer and manure Adequate water High yield

Examples of predictions based on findings Numerous predictions were generated by the rice pest constraint study. Table 4 summarizes some of these predictions; other examples include: ● Fields of medium-duration varieties with high mineral fertilizer (>150 kg ha–1) and manure applications and good water management have lower than average levels of pests in general, except narrow brown spot disease. These fields tend to have yields higher than 3.3 t ha–1. ● Fields of late-maturing varieties with 0 to 30 kg ha–1 of mineral fertilizer and insufficient water have higher than average levels of weeds, cutworm, gall midge, and brown spot. These fields tend to have yields less than 1.9 t ha–1 (Table 9). ● Leaf-feeding insects (taken as a single group) do not constrain yields under any combination of cropping practices when less than 33% of the leaves are damaged. ● Tall, late-maturing rice varieties compete better against weeds than short, early maturing rice varieties in the RLR ecosystem.

Assessing the accuracy of predictions The results of these analyses were converted into hypotheses that were tested in on- station or on-farm experiments as part of the systems approach to rice plant protection research (Fig. 9). For example, the relationship between rice cultivar duration

258 Jahn et al Start here Gather pest distribution data from collections, Interview farmers to farmers, extension document their Collect and identify flora and fauna of the rice services, news reports, knowledge, practices, etc. and perceptions about ecosystems rice pests Verify reports Characterization of biotic constraints to yield Experiments to evaluate Build database of pest effectiveness of distribution in time farmers’ practices Generation of testable and space hypotheses

Experiments to improve on farmers’ pest Create pest distribution maps management Conduct experiments to test hypothetical relationship between cropping practices, pests, Forecast likely pest problems Test integration of pest and yields prevention and control techniques with farmers

Project likely future pest problems in relation to changes in agriculture, e.g., new cultivars, fertilizers IPM options and recommendations

● Farmer field schools ● Extension ● Farmer participatory research ● Demonstrations ● Media ● Farmer experiments

Adoption and adaptation of IPM recommendations by farmers

Desired outcomes: Goals: ● Stable yields ● Increased net household ● Increased average yields income ● Less costly pest management ● Improved food security ● Safer pest management ● Sustainable pest management

Fig. 9. Schematic diagram of the systems research in rice plant protection developed and used by the IPM Program of the Cambodia-IRRI-Australia Project. IPM = integrated pest management.

Characterizing biotic constraints to production . . . 259 and weeds (Table 4) suggests that late rice varieties will compete against weeds better than early rice varieties in the RLR ecosystem. An experiment testing this hypothesis led to the conclusion that CAR8, a late-duration rice, is a better competitor against weeds than IR66, an early duration rice, in the RLR ecosystem (CIAP 1998). In another experiment, cutting off different amounts of rice leaves at different stages indicated that, before the booting stage, leaf-eating insects do not constrain rice yields (Khiev et al 1999), supporting the conclusions of the pest constraint study. The results of the correspondence analysis (Fig. 8, Table 9) suggest that the proper combination of rice cultivar, fertilizer application, and water management can minimize pest problems and thereby increase and stabilize yields (by removing the lower extremes of the yield variation). To test this hypothesis, we measured pest levels and yields in farmers’ fields treated with a factorial combination of two rice varieties and three fertilizer rates, for a total of six plots per field. We collected data from 157 fields in 1997 and from 104 fields in 1998 (CIAP 1998, 1999). Multiple regression analysis of the resulting database reveals which pests contribute to variations in yield under different combinations of soil types, fertilizers, rice cultivars, and water levels. Some general conclusions that apply to all RLR in Cambodia can also be elucidated from this database. For example, tillering-stage insect pests had little impact on yields in RLR, whereas rats were an important source of yield loss (CIAP 1998, 1999).

Assessing pest importance and prioritizing research We assess the importance of pests based on farmers’ perceptions, the impact of the pest on yield, the farmers’ ability to manage the pest, and the estimated risk of damage to the crop (Table 10). These factors are used to prioritize IPM research, taking into account whether the farmers’ pest management practices are sustainable (Fig. 10).

Limitations of the techniques While the pest and injury survey techniques described in this chapter are useful for predicting the effects of cropping practices on pest infestations, these techniques have several limitations: the danger of drawing false conclusions, difficulties in interpreting results, insufficient attention to the soil type and relative rates of fertilizer, the inability to capture time adequately as a variable, the lack of information on the relative contribution of pests to variation in yield data, the incomplete coverage of pests, and the fallibility of observers. Each of these limitations is discussed in detail below. These limitations do not render the statistical analysis useless, but they must be considered when interpreting the results, deriving hypotheses, and planning further research based on the results. It is important to note that the characterization of biotic constraints to rice production through independence testing and cluster analysis is a means of generating testable hypotheses rather than an end in itself. Although these techniques are power- ful for ruling out factors as major constraints to rice production, they only reveal associations and correlations of factors rather than cause and effect. Researchers must

260 Jahn et al No

No No

Yes

Yes?

cultivars)

available?

(allelopathic rice)

No (competitive Yes?

No No No

Yes

practices?

Do farmers’ Resistant or

No No

Yes

Yes Yes Yes

Yes

Major

Major Major

Major

Minor

Minor Minor Minor

Minor

High

Low

High

Low

High

High High

Low

Low to medium Minor

High

is not managed

Medium Medium

High

Low

High Medium High

Medium Medium Low

High

High High

High

High Medium Low

Low

Medium Low

Low

High

Potential yield Probability of Farmer

No No No

Prone loss in indivi-yield loss in assessment Farmers attempts to competitive

in Cambodia or field field if the pest endemic area manage it to unsustainable

to outbreaks dual seedbed an individual of problem in know how to manage it lead rice varieties

A Yes

C Yes

C C C

C & A No

C & A Yes

Status

A = acute, C = chronic.

Stem borer Gall midge

Table 10. Information used to prioritize research and assess opportunitiesTable for gains.

Pests

Brown planthopper

Grasshopper

Tungro

Rats Crabs

Rice bug

Thrips Caseworm

Brown spot Blast Sheath rot

Armyworm

Leaffolder Cricket Weeds Insects Lythraceae Cyperus difformis C. iria Echinochloa colona E. crus-galli Diseases Others a

Characterizing biotic constraints to production . . . 261 Do farmers consider this a major pest?

No Yes

Does it Does it constrain rice constrain rice production? May need to reevaluate production? farmers’ views and effect of pest on yield

No Yes No Yes

Determine why farmers hold this view Are farmers’ attempts to manage it unsustainable?

Address farmers’ Address scientists’ misconceptions through misconceptions farmer field schools, through publications extension, etc. No Yes

Low research priority No further High research research priority

Fig. 10. Process for prioritizing research to develop sustainable pest management practices. proceed with caution when interpreting the results of such analyses. An illustration of this is provided by the data on cutworm. Higher than average levels of cutworms were associated with an absence of water in rice fields (Table 4). Naturally, fields without water have low yields, so it is not surprising that the cluster analysis (Fig. 8) places the lowest yielding fields and cutworm-infested fields in the same domain (Table 9). To conclude from Table 9 that cutworms are a major constraint to RLR production, however, would be rash.

262 Jahn et al Interpreting the results of independence tests and cluster analysis also presents difficulties. While independence tests indicate that late-maturing varieties discourage weeds better than early maturing varieties (Table 4), correspondence analysis resulted in the formation of domain A (Table 9), which contains traditional varieties, low yields, and higher than average weed levels. In other words, fields with traditional varieties and high weed levels had lower yields, but this does not mean that traditional rice varieties encourage weeds. On the contrary, laboratory screening has revealed that at least 11 traditional Cambodian rice cultivars exhibit allelopathic properties against awnless barnyard grass, Echinochloa colona (L.) Link, while the IR varieties tested do not exhibit such traits (Pheng et al 1999). This assessment of pest constraints to RLR production did not include data on the soil types, nor did we analyze the data in terms of relative amounts of NPK or other minerals applied. The effect of soil nutrient stress on pest damage may be indirect, but profound. For example, nutrient stresses may delay crop development (Kirk et al 1998), resulting in crops that are out of synchrony with crops in the surrounding area, and therefore at greater risk of pest damage. Detailed soil and fertilizer variables were recorded and taken into account in the follow-up study designed for multiple regression analysis, described above (CIAP 1998, 1999). Another limitation of this technique arises from the assessments of injuries or pest levels over four successive development stages. There are three ways to deal with the large data set of multiple pests spread over time. First, the entire data set could be analyzed: carrying out the analysis for each pest for each stage. Not only would this process be time-consuming, but the results would be difficult to interpret, since different pests affect the crop at different stages. For instance, an analysis of tillering-stage pests may reveal the relationship of gall midge and stem borer to yield, but not rice bugs, which are primarily pests at the milk stage. Analyzing all of the data by stage would not reveal whether rice bugs or gall midges represent a greater constraint to RLR production. A second method would be to compact the data over time: creating a new variable based on the area under a curve or an average of the pest/injury levels at each crop stage (Jahn 1992, Savary et al 1996, Teng and Bissonnette 1985). Taking the area under the curve would be quite misleading for some of the pest variables, however. Some types of damage are retained from one stage to the next. Tillers damaged by gall midge at the tillering stage will still be visible at the booting stage, but not visible by the milk stage (Fig. 2). Taking the area under the gall midge curve would lead to an exaggeration of the amount of gall midge damage, since most of the galls recorded at tillering would be recorded again at booting. Other types of injury, such as whitehead resulting from stem borer damage, only appear at a certain stage. The area under the curve for whitehead (Fig. 2) would include the large area created by connecting the data point, i.e., zero, at booting to the data point at the milk stage, when in fact no whitehead is visible before the milk stage. An additional problem is that the area under the curve would depend on the graphed distance from one stage to another. Since the study included numerous rice varieties, the number of days between stages varies greatly. Attempting to incorporate that information into the analy-

Characterizing biotic constraints to production . . . 263 sis would make it unwieldy. While taking an average of the pest/injury levels across stages would avoid the problem of gauging the time between crop stages and analyzing areas under the curve, the problem of counting some damage twice (or more) would remain. A third approach to the problem would be to analyze data from the peak period of average injury. This is the approach that we chose. Like the creation of a new variable, analyzing peaks results in a considerable reduction in the number of variables to examine, but with the added advantage of avoiding the issues of the time between crop stages, areas under curves, and recounted observations. Analyzing peak data has the additional advantage that the pest/injury variables are compared with each other when they cause the greatest harm and are the easiest to observe. The chief disadvantage of this approach is that it cannot distinguish between a field with high levels of a pest at a single stage and another field with similarly high levels of the same pest but over several stages. A field that is continuously defoliated from tillering to the milk stage might be expected to have a lower yield than a field that is only defoliated at the booting stage, even if the two fields have similar levels of defoliation at the booting stage. Keeping this limitation in mind, we were careful not to make inferences about the relationship between specific pests and yields when we were testing for independence of variables (Table 4). Cluster analysis does not reveal the relative contribution of variables (e.g., pest species, rice cultivar, fertilizer rates) to the variation in yield data. Multiple regression analysis might provide this information; however, this experiment was not designed for multiple regression analysis. Some of the variables in the database were always associated with each other (e.g., certain rice cultivars were found only at one location, and all without fertilizer), making it impossible to distinguish their relative contributions to variations in yield. The factorial combination experiments conducted in 1997 and 1998 (described above) were designed for analysis by multiple regression and corrected for this limitation (CIAP 1998, 1999). Finally, the selection of key pests to include in the survey presents several limitations. First, this type of study characterizes pest constraints to RLR production based on the fields already growing rice. Pests (e.g., rats) that reach such an intensity that they discourage farmers from growing rice at all could not be included in this survey. Instead, we gather information (on pests that prevent crops from being grown) by interviewing farmers (Jahn et al 1997). Although water may be adequate to grow an early duration variety in the early wet season, farmers are generally loath to do so because of the extreme pest damage that these crops suffer. A second problem is that only certain key pests are recorded in the study, which automatically excludes rare but potentially important pests from the database. The CIAP IPM Program, however, has ongoing studies of RLR arthropod community structure and biodiversity as part of an attempt to classify all of the flora and fauna of the Cambodian RLR ecosystem (CIAP 1997, 1998, 1999). Recently introduced pests, such as the golden apple snail and the rice leaf weevil, have been discovered and monitored as a result of these biological inventories (Jahn et al 1998, CIAP 1996). A third problem is that pest damage often does not permit identification of the pest to the species, or even genus,

264 Jahn et al level. Again, we augment our database through separate field collections. For example, rat damage is easily recognized, but only by trapping specimens have we been able to determine which rat species are attacking RLR in Cambodia (Jahn et al 1999). Fourth, the quality of the data must also be considered. It is often difficult to distinguish different types of pest damage. All of the cooperators joining us in data collection have undergone extensive training in identifying, quantifying, and recording pests/ injuries of RLR. Still, there are bound to be mistakes when dealing with such a large database. All the findings and predictions must be considered hypotheses that are subject to more rigorous testing.

Conclusions Nonparametric statistical techniques (Savary et al 1996) can be valuable tools for generating testable hypotheses (but not conclusions) on how cropping practices are related to pest levels, and how those pest levels are related to yield constraints. Non- parametric analysis not only reveals associations and correlations but also helps rule out factors as major constraints to rice production over a large area. For example, leaf-feeding insects did not constrain yields under any combination of cropping practices if less than 33% of the leaves were damaged at the booting stage. The results of such analysis, however, are only a partial contribution to understanding biotic yield constraints. Relations among pests and cropping practices cannot be adequately described or even quantified simply based on statistical techniques that do not reveal causality. Nor can these nonparametric statistical techniques determine the impact of pests on yields at the field or farm level, where research results must ultimately be applied. Statistical descriptions of pest constraints should be interpreted in the context of the biology and ecology of the pest, the physical environment (e.g., soil type), economic and social conditions, the experience of rice farmers, and the manner in which farmers react to perceived pest problems. The value of generating testable hypotheses by statistical techniques must be evaluated by the cost of gathering such data, the reliability of such data, and the ultimate application of the analysis.

References CIAP (Cambodia-IRRI-Australia Project). 1996. Annual research report 1995. Phnom Penh (Cambodia): CIAP. 185 p. CIAP (Cambodia-IRRI-Australia Project). 1997. Annual research report 1996. Phnom Penh (Cambodia): CIAP. 177 p. CIAP (Cambodia-IRRI-Australia Project). 1998. Annual research report 1997. Phnom Penh (Cambodia): CIAP. 181 p. CIAP (Cambodia-IRRI-Australia Project). 1999. Annual research report 1998. Phnom Penh (Cambodia): CIAP. 205 p.

Characterizing biotic constraints to production . . . 265 Cohen MB, Savary S, Huang N, Azzam O, Datta SK. 1998. Importance of rice pests and challenges to their management. In: Dowling NG, Greenfield SM, Fischer KS, editors. Sustainability of rice in the global food system. Davis, Calif. (USA): Pacific Basin Study Center, and Manila (Philippines): International Rice Research Institute. p 145-164. Cu RM, Mew TW, Cassman KG, Teng PS. 1996. Effect of sheath blight on yield in tropical, intensive rice production system. Plant Dis. 80(10):1103-1108. Jahn GC. 1992. Effect of neem oil, monocrotophos, and carbosulfan on green leafhoppers, Nephotettix virescens (Distant) (Homoptera: Cicadellidae) and rice yields in Thailand. Proc. Hawaiian Entomol. Soc. 31:125-131. Jahn GC, Cox P, Mak Solieng, Chhorn Nel, Tuy Samram. 1999. Rat management in Cambodia. In: Singleton G, Hinds L, Leirs H, Zhibin Zhang, editors. Ecologically-based rodent management. Canberra (Australia): Australian Centre for International Agricultural Re- search. Jahn GC, Pheng S, Khiev B, Pol C. 1997. Pest management practices of lowland rice farmers in Cambodia. In: Heong KL, Escalada MM, editors. Pest management practices of rice farmers in Asia. Manila (Philippines): International Rice Research Institute. p 35-51. Jahn GC, Pheng S, Khiev B, Pol C. 1998. Pest potential of the golden apple snail in Cambodia. Cambodian J. Agric. 1:34-35. Javier E. 1997. Rice ecosystems and varieties. In: Nesbitt HJ, editor. Rice production in Cam- bodia. Manila (Philippines): International Rice Research Institute. p 39-81. Khiev B, Jahn GC, Pol C, Chhorn N. 1999. Simulating rice pest damage to determine effects on yield. Cambodian J. Agric. 2(1):29-32. Kirk GJD, Dobermann A, Ladha JK, Olk DC, Roetter R, Tuong TP, Wade L. 1998. Research on natural resources management: strategic research issues and IRRI’s approach to addressing them. IRRI Discussion Paper Series No. 27. Manila (Philippines): International Rice Research Institute. 28 p. Pheng S, Adkins S, Olofsdotter M, Jahn GC. 1999. Allelopathic effects of rice (Oryza sativa L.) on the growth of awnless barnyard grass [Echinochloa colona (L.) Link]: a new form for weed management. Cambodian J. Agric. 2(1):42-49. Satake T, Yoshida S. 1978. High temperature induced sterility in indica rice at flowering. Jpn. J. Crop Sci. 47:6-17. Savary S, Madden LV, Zadoks JC, Klein-Gebbinck HW. 1995. Use of categorical information and correspondence analysis in plant disease epidemiology. In: Advances in botanical research. Vol. 21. London: Academic Press Limited. p 213-240. Savary S, Elazegui FA, Teng PS. 1996. A survey portfolio for the characterization of rice pest constraints. IRRI Discussion Paper Series No. 18. Manila (Philippines): International Rice Research Institute. 32 p. Sheehy JE, Mitchell PE, Beerling DJ, Tsukaguchi T, Woodward FI. 1998. Temperature of rice spikelets: thermal damage and the concept of a thermal burden. Agronomie 18:449-460. Teng PS, Bissonnette HL. 1985. Developing equations to estimate potato yield loss caused by early blight in Minnesota. Am. Potato J. 62:607-618.

266 Jahn et al Notes Authors’ address: Cambodia-IRRI-Australia Project, P.O. Box 1, Phnom Penh, Cambodia. Acknowledgments: We thank Dr. Serge Savary and Dr. Paul Teng for introducing us to the data collection and analysis techniques used in this study. We are grateful to Dr. Graham McClaren for advising us on the statistical analysis, to Dr. Harry Nesbitt and Dr. Peter Cox for reviewing the manuscript, and to the rice farmers of Cambodia for sharing their cropping practices with us and allowing us to collect data from their fields. Financial support for this research was provided by AusAID. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Characterizing biotic constraints to production . . . 267 Weed communities of gogorancah rice and reflections on management

H. Pane, E. Sutisna Noor, M. Dizon, and A.M. Mortimer

Developing strategies to protect rice yields in the long term involves understanding the structure and dynamics of weed species in response to management. Where weed control is imperfect and farmers change rice crop establishment methods and control tactics, weed shifts can occur. Further- more, in rainfed rice, the inherently complex abiotic nature of the cropping environment may result in variation in weed composition. A survey of the weed communities remaining after farm weeding practices was conducted during booting of gogorancah rice (dry-seeded bunded rice) in rainfed lowland areas of Pati and Rembang, Indonesia. Counts were made of all weed species present in four randomly placed 1-m2 quadrats at low, mid, and upper points of the land toposequence in fields at each of 25 farm sites. In addition, soil nutrient status (pH, N, P, K, and organic matter) at each site was measured. Fifty-six weed species covering 18 families were recorded. The average total weed density was 175 plants m–2, with the greatest number of species occurring in upper toposequence locations. Weed communities remaining after farmer weeding at the upper and mid positions of the toposequence were broadly similar in species composition (Lindernia species, Echinochloa colona, Fimbristylis miliacea, and Murdannia nudiflora). These differed from those at the base of the toposequence, which was dominated by Ammannia baccifera, E. colona, F. miliacea, and Leptochloa chinensis. Cyperus species were also abundant across the toposequence, but differed in relation to position. Cyperus tenuispica at the top was replaced by C. iria in the middle, and, at the lower points, C. difformis was predominant. L. chinensis, a competitive grass weed, was also abundant in sites at the bottom of the toposequence. Canonical correspondence analysis was used to examine interrelationships among sampling sites based on species composition and nutrient status. Sites at the base of the toposequence were delineated sharply from those in the mid and upper positions, in which there was greater similarity in weed flora. Multivariate analysis showed that sample sites differed

Weed communities of gogorancah rice and reflections on management 269 in soil nutritional status, especially for P and pH, which in turn was reflected in species composition. The results indicate that, under current weed management practices, the residual weed flora is strongly governed by hydrological factors with respect to toposequence in addition to soil factors. Future research and weed management options are discussed in the context of this baseline characterization of weed communities.

A defining characteristic of the agroecosystem in which rainfed lowland rice is grown is the occurrence of noncontinuous flooding of land of variable depth and duration (Zeigler and Puckridge 1995). The Jakenan region in Central Java, Indonesia, with an average of <1,500 mm of annual rainfall, falls into the shallow drought-prone subecosystem of the rainfed lowlands (Khush 1984). Governed by erratic seasonal rainfall, a sequence of cropping is practiced. Dry-seeded bunded rice (gogorancah) is grown from the onset of the wet season (October-November) through to January- February, followed by minimum-tillage transplanted rice (walik jerami) through to May, with a secondary crop (palawija) such as mungbean, maize, soybean, or cowpea grown for the remainder of the dry season. Walik jerami and palawija crops remain at risk from drought (Fagi 1995) and the use of on-farm water reservoirs (embung) has been actively promoted (Syamsiah et al 1994). Rice is grown in bunded fields on sloping lands with up to a 30-degree slope and farmers manage cropping in relation to the toposequence (Fujisaka et al 1993). As in other rainfed environments, considerable small-scale spatial variability occurs in the cropping environment with respect to hydrologic processes (driven by topography, rainfall, and soil texture) and soil fertility (Wade et al 1999a). Moreover, the weed flora in these areas is potentially very diverse since the crop habitat may exhibit soils that range from aerobic through saturated to fully flooded for varying parts of the crop cycle. Typically, during the first three to four weeks (depending on rainfall) of gogorancah, rice grows as an upland crop in moist soil only to be flooded as rainfall intensity increases, as a result of impounded water. The time to flooding and its depth and duration in relation to field surface topography are well-known hydrologic determinants governing germination and seedling establishment of rice weeds and these act interspecifically as a sieve in the recruitment of weeds into the growing crop (Pons 1982, Pane and Mansor 1994). In rainfed lowlands, these weed species are often loosely described as “semiaquatic” in that they possess life history characteristics that enable establishment in moist or saturated conditions and later survival under flooded conditions, for example, many sedge species. The weed flora of walik jerami often exhibits species in common with gogorancah but typically includes obligate aquatic species. Weed control in gogorancah and walik jerami rice continues to rely on manual weeding. In gogorancah, weeding with a small hoe begins at the three-leaf stage of rice about 2 wk after emergence (WAE) with further weedings at 5–6 WAE and some-

270 Pane et al times at 7–8 WAE, 80 labor-days ha–1 being the estimated effort (Fagi 1995). In walik jerami, the crop is usually handweeded only twice, 2–4 wk after transplanting (WAT) and at 6–7 WAT, needing 48 labor-days ha–1. The extent to which weeds are a constraint to yields has not been accurately determined by on-farm yield gap studies in gogorancah rice but a rapid rural appraisal of fields (Fujisaka et al 1993) suggested that weed infestations were higher in gogorancah than in walik jerami and severest in the low-lying areas. Recently, from three seasons of research-station trials, Bangun et al (1998a,b) estimated that grain yield reductions due to weeds in gogorancah and walik jerami were about 76% and 45%, respectively, when comparing unweeded plots with weed-free checks. Improving weed management practices for gogorancah rice invokes many of the same issues associated with direct-seeded rice production systems in other agroecosystems (Mortimer et al 1997, Mortimer and Hill 1999). In the context of prevailing socioeconomic production domains (Pandey 1998) and the heterogeneous environment of rainfed rice, the appropriate integration of agronomic and water management practices to ensure rapid crop establishment of a competitive crop stand and fertilizer management and weed control interventions after crop establishment remain important adaptive research issues. Equally essential is an understanding of existing farm weed management systems and the agronomic and economic constraints experienced by farmers. Documentation of the efficacy of farm weed control practices and of the abundance and diversity of weeds throughout crop growth contributes to a baseline in the analysis of the impact of existing weed control systems and to the ex ante assessment of proposed changes in management. Given a priori knowledge about the relative competitiveness and ecology of individual weed species, this baseline provides a precursor to the design of on-farm yield gap trials, which are expensive to conduct (Moody 1993). Of particular importance is knowledge of the weed flora remaining after normal farmer weeding practices. Apart from general inventories of the weed species present in rainfed rice in Indonesia (e.g., Soerjani et al 1987), no systematic surveys have been conducted to characterize the on-farm weed flora of gogorancah rice. The objectives of this study were therefore to quantitatively describe the weed flora in dry direct-seeded gogorancah rice in Central Java from extensive farmer field surveys, to analyze variation in this flora in relation to toposequence and soil characteristics, and to review implications for improved weed management.

Materials and methods Survey sites Twenty-five farm sites were chosen in subdistricts within 50 km of the Jakenan experimental station of the Central Research Institute for Food Crops in the districts of Pati and Rembang, Central Java, Indonesia. Farms were identified as being in areas of intensive rainfed lowland rice production by the presence of a water reservoir in the local vicinity and all showed sloping lands (5–30°). At each farm, three positions (upper, mid, and lower) on the toposequence were identified for census of the weed

Weed communities of gogorancah rice and reflections on management 271 flora. The range in elevation across the toposequence at all study sites did not exceed 10 m. Prior to data collection, farmers were briefly interviewed to confirm that rice weeding had been finished and to ascertain nutrient management practices.

Weed flora During February 1998, crops were inspected and the weed flora assessed when rice was at the booting stage, approximately 60–70 d after sowing (DAS). Four 1-m2 quadrats were placed randomly within the crop at each toposequence position and the density of individual weed species (plants m–2) enumerated by destructive removal of all plants beyond the small seedling stage. Identification to the species level was not achieved for the genus Lindernia because of the lack of floral states. Consequently, Lindernia spp. refers in the text to several species within the genus and is likely to include the annuals L. anagallis, L. antipoda, and L. cillata but excludes the perennial L. crustacea, which was readily identifiable.

Soil At each toposequence location, bulk samples were taken by pooling soil from each individual quadrat used for weed sampling. Two replicate sets were obtained in a like manner. Soil parameters, including pH, were measured following standard laboratory practices as described by Hidayat (1978). Percent total organic carbon in soil was measured spectrophotometrically, % total N by semimicro Kjeldahl, soluble phosphorus (mg kg–1) and exchangeable K (meq 100 g–1) using Bray’s method, and cation exchange capacity, CEC (meq 100 g–1), by Schollenberger’s semimicro percolation method. Farmyard manure was the principal source of fertilizer although some farmers referred to the use of inorganic phosphate and potassium.

Data analysis Univariate analysis of variance was used to explore variation in weed densities and in soil nutrient status. Since sites were randomly chosen, and toposequence positions within sites were fixed, a mixed model was used, partitioning variation among toposequence positions within individual sites (SAS 1986). A range of quantitative approaches was employed to explore weed community structure in relation to the environmental variables measured at each site. Log rank abundance curves (Ludwig and Reynolds 1989) were used to evaluate overall community composition in relation to toposequence, pooling data over sites. These were then compared with rankings given by the summed dominance ratio (Kent and Coker 1992). Ordination techniques (ter Braak 1985) were applied to quantify the weed community structure in relation to soil parameters and position on the toposequence. Simple ordination methods (sensu Bray and Curtis 1957) provide graphical representations of community structure based on either the similarity of sampled sites with respect to relative abundance of constituent species (site ordinations) or the similarity of species with respect to co-occurrence at the same sampling sites (species ordinations).

272 Pane et al As such, they provide a method of indirect gradient analysis in that interrelationships between species and sites can be simultaneously quantified and may be interpreted in the light of what is known, or can be inferred, about environmental gradients reflected by relative species abundance across the range of sites. A contrasting approach is direct gradient analysis by regression methods (Whittaker 1967), which seeks to explicitly examine a species response to chosen environmental variables or to calibrate sites with respect to environmental variables. Indirect gradient analysis is argued to have several advantages for heterogeneous environments (Jongman et al 1995) in that (1) species compositions at sites are readily identifiable and plant species are intrinsically “phytometers” of their environment, (2) environments are difficult to characterize exhaustively in terms of both biotic and abiotic variables, and (3) the actual occurrence of an individual species may be too unpredictable to discover its relationship with environmental regimes directly and general patterns of coincidence of several species may be of greater use in detecting species-environment relations. Nevertheless, indirect gradient analysis remains intrinsically a correlative method and can only suggest hypotheses of causal factors. Canonical ordination techniques combine both ordination and regression into multivariate direct gradient analysis to simultaneously interpret the response of many species to many environmental variables. We used correspondence analysis (CA) and canonical correspondence analysis (CCA) (ter Braak and Smilauer 1998) for indirect and direct gradient analysis, respectively, assuming species response to environmental variables on a unimodal (Gaussian) model. Data were individual species counts per m2 averaged over replicates at each sampling location and associated soil parameters after standardizing to comparable scales. Log-linear regression was used to relate species abundance to environmental scores derived from ordination using the predictive function 2 exp(b0 + b1x + b2x ), with the constraint b2 <0.

Results Soil Tables 1 and 2 give soil site characteristics. Significant intersite variation was found in mean levels of potassium, phosphate, organic carbon, and CEC, the latter being significantly correlated with all variables except phosphate. No differences in nitrogen status were found among sites or among locations within sites. In 22 out of the 25 sites, statistically significant differences in soil pH were detected among toposequence positions. Soils from the upper positions were always more acid, with the average difference within a site being 0.65 of a unit from upper to lower position. Contrastingly, potassium, phosphate, organic carbon, and CEC varied among locations at every farm site and differences were not correlated with toposequence position. Levels of these nutrients fell within reported ranges for other rainfed lowland environments, sites being characteristically nutrient-deficient (Wade et al 1999b).

Weed communities of gogorancah rice and reflections on management 273 )

–1

)(%)(meq 100 g )(%)(meq

–1

2.05 0.20 2.65

) (mg kg

–1

0.375 94.80 2.25 29.38

0.020

K (meq 100 g

pH % N Exchangeable P Organic C CEC

5.1 0.07 (n)0.066 30.07 0.48 3.87

7.4 0.10 (n)0.048 42.52 1.00 6.18

7.8 0.13 (n)0.171 6.72 1.53 24.14

7.4 (n) 0.10 (n) 0.113 10.75 0.95 13.90

7.6 0.13 (n)0.078 4.33 1.29 10.66

6.7 0.10 (n)0.033 14.97 0.99 9.32

5.8 0.10 (n)0.130 68.32 0.76 8.84 7.0 (n) 0.11 (n) 0.073 31.87 1.27 12.72 7.5 0.11 (n)0.046 25.75 1.22 14.30

7.0 0.10 (n)0.026 33.84 1.06 10.67 6.3 0.11 (n)0.088 22.97 0.95 11.19

6.3 0.11 (n)0.106 9.50 1.11 15.06 7.2 0.15 (n)0.048 4.71 0.98 18.15 6.8 0.12 (n)0.073 9.99 1.40 12.80 7.5 0.10 (n)0.051 29.78 0.88 14.66 7.4 0.11 (n)0.158 21.29 1.34 15.15 8.1 0.12 (n)0.228 29.06 1.12 22.73

5.0 0.22 (n)0.055 13.38 0.56 3.76

7.8 0.13 (n)0.171 27.22 1.55 25.63

6.1 0.08 (n)0.055 17.91 0.54 7.31

ecorded 4.9 0.52

Maximum value r Minimum value recorded 8.4 0.07

CEC = cation exchange capacity

Table Table 1. Soil characteristics of the sample sites. Data are averaged over toposequence position and replicate. Means followed by (n) indicate no significant variation among positions within sites.

Site Village/subdistrict

number

1 Grawan/Sumber 2 Pragu/Sulang 3 Jadi/Sumber 4 Megulung/Sumber 5 Bogorejo/Sumber 7.7 0.12 (n)0.176 3.99 1.34 24.97 6 Seren/Sulang 7 Sindangsari/Lasem 8 Samaran/Pamotan 9 Pragen/Pamotan 10 Lambangan Wetan/Bulu 6.9 0.13 (n)0.211 7.43 1.05 14.62 11 Kembang/Bulu 12 Warugunung/Bulu 6.8 0.08 (n)0.083 12.85 0.53 11.04 13 Sekarsari/Kaliori 14 Mateseh/Kaliori 15 Maguan/Kaliori 16 Banyuurip/Gunem 17 Panohan/Sulang 18 Gunem/Gunem 19 Bamban/Pamotan 20 Gembol/Pamotan 21 Palemgede/Pucukwangi 22 Karangrejo/Jaken 6.6 0.09 (n)0.073 75.31 0.68 8.29 23 Sidomulyo/Jaken 24 Harumanis/Jaken 7.5 (n) 0.12 (n) 0.266 6.25 1.12 26.74 25 Mantingan/Jaken a

274 Pane et al Table 2. Correlation (product moment) matrix of soil characteristics pooled over sites and toposequence position. Data in bold are significant (P ≤ 0.05). CEC = cation exchange capacity.

pH C N P K

C 0.7252 N 0.5723 0.6151 P –0.2343 –0.2572 –0.2871 K 0.4421 0.3680 0.2719 –0.1623 CEC 0.6982 0.6778 0.5222 –0.2636 0.6190

Weed abundance Fifty-six weed species (Table 3) in total from 18 familes (Amaranthaceae, Araceae, Asteraceae, Boraginaceae, Commelinaceae, Convolvulaceae, Cyperaceae, Euphorbiaceae, Lythraceae, Marsileaceae, Molluginaceae, Onagraceae, Poaceae, Pontederiaceae, Portulacaceae, Rubiaceae, Sphenocleaceae, Scrophulariaceae) were recorded in the census. The developmental stages of species varied considerably, most being vegetative and below the crop canopy height. Species commonly observed in the flowering stage were Eclipta alba, Echinochloa crus-galli, E. colona, Fimbristylis miliacea, and Cyperus difformis and C. rotundus. The overall mean total weed density was 175 plants m–2 (Fig. 1) and this did not differ significantly in relation to toposequence when averaged across sites. Positions were ordered lower < middle < upper in terms of mean number of species. At low positions, the variance in densities was least and density distributions were highly skewed in the mid and upper regions. Very high densities (above 400 plants m–2) were always dominated by a single species. Frequency distribution of counts per m2 varied among species and between toposequence positions, both Poisson and log- normal distributions being evident in the data set. For the most abundant species, distributions were typically strongly skewed but in some instances apparently dis- junct (e.g., Ammannia baccifera, lower toposequence) (Fig. 2). Figure 3 illustrates the rank order abundance of species in each toposequence, based on pooled counts across sites. While all communities were structured geometrically, common taxa had a different rank in their relative abundance in relation to ≤ toposequence (Spearman’s Rs, P 0.03). In the upper and mid toposequence, communities were dominated by a similar group of species with Lindernia spp. the most abundant. At the low toposequence, Lindernia was replaced by Ammannia baccifera and in low and mid positions Leptochloa chinensis was more abundant than in the upper toposequence. Both Echinochloa colona and Fimbristylis miliacea achieved high rankings across all toposequence positions. Ranking species by summed dominance ratio (averaged relative abundance and relative frequency) promoted relatively rare species overall (e.g., Phyllanthus niruri and Digitaria ciliaris) to a much higher rank (Fig. 4), especially evident at the low toposequence. Figure 5 shows the biplot ordination from CA of species and sites according to toposequence. In the analysis, data were logarithmically transformed and the influ-

Weed communities of gogorancah rice and reflections on management 275 Table 3. Weed species recorded and numerical code used in Figure 3.

Codes Species Codes Species

1 Ageratum conyzoides L. 28 Fimbristylis miliacea (L.) Vahl 2 Alternanthera philoxeroides 29 Hedyotis biflora (L.) Lam. (Mart.) Griseb. 30 Hedyotis corymbosa (L.) Lam. 3 Alternanthera sessilis (L.)31 Heliotropium indicum L. R. Br. ex Roem. & Schult. 32 Hymenachne acutigluma (Steud.) 4 Amaranthus dubius Mart. Gilliland 5 Ammannia baccifera L. 33 Ipomoea triloba L. 6 Borreria latifolia (Aubl.) Schum. 34 Ischaemum rugosum Salisb. 7 Brachiaria mutica (Forssk.) Stapf 35 Jussiaea repens (L.) Hara 8 Commelina benghalensis L. 36 Leptochloa chinensis (L.) Nees 9 Commelina nudiflora (L.) Brenan 37 Lindernia crustacea (L.) F. Muell. 10 Cynodon dactylon (L.) Pers. 38 Lindernia spp. 11 Cyperus compressus L. 39 Ludwigia hyssopifolia (G. Don) Exell 12 Cyperus difformis L. 40 Ludwigia octovalvis (Jacq.) Raven 13 Cyperus halpan L. 41 Marsilea crenata L. 14 Cyperus iria L. 42 Mollugo pentaphylla L. 15 Cyperus kyllingia Endl. 43 Monochoria vaginalis (Burm. f.) 16 Cyperus rotundus L. Presl 17 Cyperus tenuispica Steud. 44 Murdannia nudiflora (L.) Brenan 18 Dactyloctenium aegyptium (L.) . 45 Panicum maximum Jacq. Willd 46 Paspalum distichum L. 19 Digitaria ciliaris (Retz.) Koel. 47 Phyllanthus niruri Webster 20 Digitaria longiflora (Retz.) Pers. 48 Phyllanthus virgatus Forst. f. 21 Echinochloa colona (L.) Link 49 Polytrias amaura (Buse) O.K. 22 Echinochloa crus-galli (L.) P. Beauv. 50 Portulaca oleracea L. 23 Eclipta alba (L.) Hassk. 51 Scirpus juncoides Roxb. 24 Eleocharis palustris (L.) R. Br. 52 Sphaeranthus indicus L. 25 Eleusine indica (L.) Gaertn. 53 Sphenoclea zeylanica Gaertn. 26 Eragrostis tenella (L.) P. Beauv. 54 Tridax procumbens L. ex Roem. & Schult. 55 Typhonium trilobatum (L.) Schott 27 Euphorbia hirta L. 56 Vernonia cinerea (L.) Less.

Plants m–2 Number of species 800 30

600 20

400

10 200

0 0 Upper Middle Lower Fig. 1. Box plots of the number of weed species and total weed density in relation to toposequence. Means are shown as horizontal lines within boxes. Data are averaged over farm sites.

276 Pane et al Frequency 32 16 28 Echinochloa crus-galli n = 75 14 Ammannia baccifera— n = 25 high toposequence 24 12 20 10 16 8 12 6 8 4 4 2 0 0 32 16 28 Fimbristylis miliacea n = 75 14 Ammannia baccifera— n = 25 mid toposequence 24 12 20 10 16 8 12 6 8 4 4 2 0 0 10020 30 40 50 60 70 80 90 100 >100 16 14 Ammannia baccifera— n = 25 low toposequence 12 10 8 6 4 2 0 10020 30 40 50 60 70 80 90 100 >100 Number of plants per m2 Fig. 2. Frequency distributions of the abundance of selected weed species. Distributions for Echinochloa crus-galli and Fimbristylis miliacea are based on pooled data from all toposequence positions. Data for Ammannia baccifera are given for each toposequence position. Note the different scales on abscissae.

ence of rare species removed by discarding species with a proportional overall abundance (Fig. 3) of less than 0.1%. The ordination accounted for 34% of the variance in the species data and 57% of the variance in the site data. Sites from the lower part of the toposequence were clearly separated from those in the mid and upper positions. Species were delineated on the first dominant axis (eigenvalue = 0.240, total inertia = 1.91), typically those with high positive scores being obligate aquatic species (e.g., Monochoria vaginalis and Marsilea crenata) and terrestrial species (e.g., Hedyotis biflora and Eleusine indica) having negative ones. Consequently, the first axis may be hypothesized to reflect the water status of sites during the cropping season. Spe- cies locations in relation to axis 2 are not readily interpretable from simple auteco- logical observations. Figure 6 gives the response curves to the toposequence environment quantified by axis 1 scores for selected species together with the location of sites according to axis score. High densities of Lindernia spp. were predicted for negative scores, which

Weed communities of gogorancah rice and reflections on management 277 Proportional abundance (log scale) % 100 Lindernia spp. 13.6 A Cyperus tenuispica 9.1 Fimbristylis miliacea 8.4 10 Lindernia crustacea 7.2 Ammannia baccifera 7.1 Echinochloa colona 6.9 Murdania nudiflora 5.7 1 Cyperus iria Monochoria vaginalis Cyperus difformis 0.1 Ludwigia octovalvis Leptochloa chinensis Phyllanthus niruri 0.01 Cyperus rotundus Scirpus juncoides Hedyotis corymbosa 0.001 Sphaeranthus indicus 0 102030405060Digitaria ciliaris Paspalum distichum Order of species by code: Eclipta alba 38,17,28,37,5,21,44,14,43,12,40,36,47,16,51,30,52,19,46,23,55,41,39,7,24,10,26,3,2, 5,18,35,4,15,8,50,56,27,34,45,20,11,22,2,13,48,31,29,33,53,54,1. Typhonium trilobatum

100 Lindernia spp. 24.1 B Echinochloa colona 14.5 Fimbristylis miliacea 11.8 10 Murdania nudiflora 6.3 Cyperus tenuispica 6.3 Cyperus iria 6.0 1 Leptochloa chinensis 5.8 Cyperus difformis Ammannia baccifera 0.1 Digitaria ciliaris Cynodon dactylon Hedyotis biflora 0.01 Cyperus rotundus Monochoria vaginalis

0.001 0 102030405060 Order of species by code: 38,21,28,44,17,14,36,12,5,19,10,29,16,43,26,52,41,47,15,23,39,51,34,50,46,18,1,25, 8,40,3,7,45,22,49,55,13,42,31,2,30,24,11,9,56,4,27,33,48,6.

100 Ammannia baccifera 27.5 C Echinochloa colona 9.5 Fimbristylis miliacea 8.6 10 Cyperus difformis 8.3 Leptochloa chinensis 8.1 Marsilea crenata 1 Cyperus halpan Cyperus iria Eleocharis palustris 0.1 Monochoria vaginalis Cyperus rotundus Ludwigia octovalvis 0.01 Lindernia spp. Mollugo pentaphylla Paspalum distichum 0.001 Eclipta alba Ischaemum rugosum 0 102030405060Eragrostis tenella Order of species by code: Phyllanthus niruri 5,21,28,12,36,41,13,14,24,43,16,40,38,42,46,23,34,26,47,8,22,3,31,10,49,50,33,32, 1,8,51,19,7,39,48,45,27,29,25. Fig. 3. Log rank abundance curves of weed communities in relation to toposequence, pooling over sites: (A) upper, (B) middle, (C) lower. The species list to the right includes all those occurring at greater than 1%, and species with attached data greater than 5%. Rank order by species code (Table 3) is given below each abscissa.

278 Pane et al SDR (%) 20 Upper Middle Lower 18 16 14 12 10 8 6 4 2 0 spp. spp.

Cyperus iria Cyperus iria Cyperus iria

Lindernia Lindernia

Digitaria ciliaris Digitaria ciliaris Cyperus halpan

Marsilea crenata Phyllanthus niruri Phyllanthus niruri Phyllanthus niruri Cyperus rotundus Cyperus rotundus Cyperus difformis Cyperus rotundus

Murdania nudiflora Murdania nudiflora Cyperus tenuispica Ludwigia octovalvis Cyperus tenuispica Echinochloa colona Echinochloa colona Echinochloa colona Mollugo pentaphylla Fimbristylis miliacea Fimbristylis miliacea Ammannia baccifera Ammannia baccifera Spaeranthus indicus Ammannia baccifera Fimbristylis miliacea Leptochloa chinensis Ludwigia hyssopifolia Monochoria vaginalis Leptochloa chinensis Fig. 4. Abundance ranked by summed dominance ratio (following Kent and Coker 1992) of weed species present in rice at rice booting stage. Ranks are presented for the first 12 weed species. Absolute density (Di ) = total number of plants for species i in all sample quadrats. ∑∑∑ ××× Relative density (Rdi) = Di/( ∑∑i Di ) ×× 100. ××× Absolute frequency (Fi) = (The number of quadrats in which species i occurs)/4 ×× 100. ∑∑∑ ××× Relative frequency (Rfi ) = Fi/∑∑ i Fi ×× 100. Summed dominance ratio (SDR) = (Rdi + Rfi )/2.

in turn were associated predominantly with middle and upper sites. The converse was true for A. baccifera, for which predicted densities increased monotonically with increasing score (although note that regression was not significant, P ≤0.096). Echinochloa colona and F. miliacea had similar maximum densities, the latter having a more restricted distribution range. Regression analysis predicted more dense stands of C. difformis at sites in the lower toposequence, the distribution of Leptochloa chinensis being largely similar. CCA (Fig. 7) was performed using a restricted set of soil parameters. N content, which did not differ among sampling locations, and CEC, which was highly correlated with other variables, were excluded. Seventy-three percent of the species-environment relationship was explained by the first two axes of the analysis. The first axis was strongly correlated with soil pH (r = –0.95, recorded range 4.9–8.4) and the second axis with available phosphate (r = 0.69, recorded range 2.05–94.8 mg kg–1). Spe- cies were delineated noticeably in relation to axis 1, the inference being that soil pH may govern species distribution. Most species (left of the diagram) had low scores,

Weed communities of gogorancah rice and reflections on management 279 1.0

Echi_col Cype_iri

Moll_pen

Erag_sp Euph_hir Alte_ses Cype_hal Eleu_ind Lept_chi Digi_cil Phyl_nin Ludw_hys Isch_rug Comm_ben Lind_spp Amma_bac Ecli_alb Brac_mut Pasp_di Axis 1 Fimb_mil Echi_cru Murd_nud Cype_rot Hedy_cor Typh_sp Cype_dif Ludw_oct Lind_cru Mono_vag Port_ole Cyno_dac Scir_jun Mars_cre Spha_ind Hedy_bif Cype_ten Axis 2 –1.0 –1.0 1.0 Fig. 5. Biplot diagram from correspondence analysis ordination diagram of sites and species. Species locations are indicated by small circles and sites by large ones. = lower, = mid, = upper sites. which were correlated with neutral or mildly alkaline soils. On the other hand, Scirpus juncoides, Digitaria ciliata, and Lindernia spp. had high scores associated with acidic sites. Differential responses to phosphate may be postulated. Of those species common to continuously flooded conditions, Leptochloa chinensis and Marselia crenata were more abundant at sites containing high phosphate (positive correlation between axis 2 and soil phosphate content) than Cyperus iria and Murdania nudiflora. Species in the lower left quadrant were characteristic of communities reported to occur in productive rainfed rice habitats (Moody 1983) and were associated with high soil organic carbon.

Discussion Implications for yield Wade et al (1999b) reported that rice yields in weed-free researcher-managed trials of gogorancah rice varied between 3 and 5 t ha–1 depending on water availability and that improved nutrient management was essential to raising yields. The highest yields

280 Pane et al Predicted density (plants m–2) 50

Lindernia spp. 40 Ammannia baccifera

30 Fimbristylis miliacea Echinochloa colona 20 Cyperus difformis Leptochloa chinensis 10

0 –2 –1 0 1 2 Axis 1

Lower Middle Upper Site distribution

Fig. 6. Log-linear regression of density of selected species per quadrat on the site scores of the first axis of correspondence analysis (Fig. 5) assuming a Gaussian distribution function. Axis scores themselves were derived from the site ordination and are not those shown in Figure 5, which have been rescaled for biplot presentation. Re- gressions gave significant (P ≤ 0.05) fits to abundance distributions for all species except Ammannia baccifera.

were found in lower toposequence positions with differences of up to 1 t ha–1 being recorded. Average on-farm yields in the same locality, however, have been reported to be significantly lower and typically less than 3 t ha–1 (Syamsiah et al 1994). While confirming the well-known variability in soil characteristics in Java, this study illustrates that surprisingly dense, diverse weed communities persisted in gogorancah rice during the latter stages of crop development. Within the restrictions of this study and by inference only, these may be strongly structured by spatial variability in soil pH, by nutrient availability governed by soil pH, and by hydrology through toposequence position. The literature is sparse on data describing species distributional ranges in relation to soil characteristics in Indonesia. It is interesting to note, however, that Echinochloa crus-galli achieved a mid score on axis 1 (Fig. 7) and that this species is reported to prefer neutral soils (Soerjani et al 1987). Figure 3 indicates that only seven species were common (>5%) in the upper and mid toposequence with five in the lower toposequence. Acidic, nutrient-poor sites in the upper toposequence were dominated by Lindernia spp., which are commonly considered to be ruderal species of open sites. Habitat specificity was reflected within Cyperus, with C. tenuispica being replaced by C. difformis in lower site locations, where

Weed communities of gogorancah rice and reflections on management 281 +3.0

Echi_cru

Ecli_alb Lept_chi Mars_cre Mono_vag

P Scir_jun

Moll_pen Echi_col Ludw_hys Cype_iri Murd_nud Isch_rug Erag_ten Typh_tri Digi_cil K Pasp_dis Cype_hal Eleu_ind

Comm_ben Cype_ten Axis 1 Lind_spp Hedy_cor Fimb_mil pH Cype_rot Ludw_oct Phyl_nir Cyno_dac C Alte_ses Cype_dif Euph_hir Brac_mut Amma_bac Axis 1 Axis 2 Port_ole pH –0.954 –0.277 Hedy_bif C –0.616 –0.661 P –0.028 0.694 Lind_cru K –0.516 0.156

Spha_ind Axis 2 –2.5 –2.0 +3.0 Fig. 7. Biplot diagram for species and selected soil variables from canonical correspondence analysis. Arrows indicate the direction of change of environmental variables (pH, P, K, and C) and their relationship with ordination axes. Correla- tions between measured environmental variables and axis scores are inset. Spe- cies names are truncated (see Table 3).

Ammannia baccifera was dominant. Both Echinochloa colona and Fimbristylis miliacea showed a broad adaptation to toposequence conditions (Fig. 6), although the latter was more abundant in mid slope. These communities are likely to have comprised plants of varying age, since mature individuals may have escaped manual weeding operations earlier in the life of the crop or be younger by virtue of recruitment after weeding has ceased. Assuming that the last weeding was approximately 40–50 days after emergence and booting was one month later, then 30 days may have elapsed between the completion of manual weeding and the census. This time period is of sufficient duration for populations of Echinochloa colona, Fimbristylis miliacea, and Lindernia spp. to develop, and for species with strong developmental plasticity to complete the life cycle and contribute seed to future weed infestations.

282 Pane et al The relatively high densities of weeds and the diversity of weed species present may simply reflect the fact that farmers in the region perceive little economic benefit in weed removal during the later stage of crop development, particularly as further manual weeding may damage the crop. Moody (1993), however, has pointed out that additional weeding close to rice panicle initiation will often increase final yield. Clearly, without more detailed experimentation, it is only possible to hypothesize as to likely yield reductions resulting from these observed levels of weed infestation. The following comments are in consequence purely speculative. Field observation that most weeds were below canopy height and the anticipated limited growth period postweeding suggests that weed biomass per unit area may be low and that interfer- ence with photosynthetically active radiation capture by rice will be limited. Weed species capable of strong plasticity in growth and late development, however, such as Ammannia baccifera, may pose a competitive threat to yield during grain filling. Anecdotally, this species is reported to increase in height and biomass as the crop matures and as fields drain. Competition for nutrients during late crop development may also occur from high densities of weeds lower in the canopy. Soils in this region are characteristically phosphate- and potassium-limited (Clough et al, this volume) and all sites surveyed in this study were nutrient-poor. While underlying heterogeneity in soil pH may restrict the distribution of weed species, Figure 7 suggests that there may be differential species preferences for phosphate. This, in turn, raises the hypothesis that competition for this nutrient may occur particularly in low toposequence positions in which weeds of irrigated rice persist. Monochoria vaginalis, Marselia crenata, and Scirpus juncoides are all species that have been reported to compete strongly with rice for nutrients (Soerjani et al 1987) as has Eclipta alba in upland rice (Lee and Moody 1989). The extent to which Lindernia spp. may compete for nutrients with rice remains open to question (not least because of the lack of formal iden- tity). However, the distribution of this group of species was correlated with acidic, nutrient-poor soils in which rice yields themselves are likely to be low. The extent to which overall productivity governs the nature and intensity of competitive interactions for nutrients deserves further study.

Implications for weed management The wide diversity of weed species present within farm localities strongly indicates the potential for transient temporal and longer-term shifts in relative abundance of weed species in relation to changes in agronomy and water and weed management. Studies of soil fertility in rainfed lowlands have clearly pointed to the importance of fertilizer use in increasing rice yield. Improvement in crop nutrition and deployment of new varieties may result in suppression of many weed species simply through enhanced crop vigor by improved agronomic practices. Many authors (e.g., Cooper et al 1999), however, have pointed to the difficulty of breeding widely adapted rice cultivars for rainfed environments tolerant of abiotic stresses throughout the life of the crop. The presence of a wide spectrum of weeds encompassing species common to intensive irrigated production systems as well as rainfed and upland environments

Weed communities of gogorancah rice and reflections on management 283 indicates the need to deploy broad-spectrum weed control practices that are either effective across all toposequence positions or location-specific. At present, time-hon- ored manual weeding satisfies either option, although it may be insufficiently employed in the late season. However, replacement of manual weeding by chemical means will be contingent on improved water management techniques, which may only be feasible in the lower toposequence. Water management, in turn, will depend critically on drainage and supply infrastructures that may be beyond the economic and practical reach of farmers. Conversely, in the upper toposequence, rapid infiltration rates and lack of water supply from lower slope reservoirs may prohibit the easy use of water in weed control. The existence of grass weeds such as Echinochloa crus-galli, Leptochloa chinensis, and Ischaemum rugosum, and of the sedge Cyperus difformis among other sedges within the agroecosytem clearly poses a significant threat to rice intensification and underlies the importance of effective early weed control. From on-farm trials in the midtoposequence, Bangun et al (1998a,b) concluded that the use of oxadiazon (0.5 kg ha–1 a.i. applied 1 DAS) was more effective in controlling Leptochloa chinensis and Cyperus iria than manual weeding at 21 DAS and also suppressed the growth of Dactyloctenium aegyptium and Eleusine indica by 30 DAS. By 60 DAS, however, the abundance of Echinochloa colona did not differ between weed control treatments and additional manual weeding was required. In the same trial with IR64, no significant effects of changing crop spacing were detected in weed suppression. Bangun and others noted, in addition, that, in the following walik jerami crop, weed biomass was lower than in gogorancah but with an increase in grass weeds including L. chinensis, E. colona, Paspalum distichum, Ischaemum rugosum, and Isachne globosa. Fagi (1995) concluded that, while improved weed control in gogorancah rice would necessarily be achieved by the use of broad-spectrum herbicides in both crops, significant gains could also be made by improved land preparation prior to cropping. Typically, land for walik jerami is rapidly prepared after gogorancah harvest by straw incorporation into wet soil and rice transplanted soon after to minimize risk of late- season drought. In consequence, the opportunities for the imposition of a stale seedbed with a nonselective herbicide or repeated tillage cycles are precluded. This points to the potential value of short-duration rice cultivars for use in gogorancah and walik jerami to enable a window of opportunity for improved land preparation for weed control either at the start of gogorancah or between rice crops. Delayed time of planting of gogorancah, however, may elevate the risk of rain damage to grain quality and harvesting difficulties. The requirement for abiotic stress tolerance in such varieties remains fundamental.

Methodological and analytical implications At first sight, the application of ecological methods in the detailed analysis of weed communities may seem an unnecessary adjunct to developing weed management systems in which the overall goal is simply the reduction of weed biomass that interferes with crop production. However, the inherent heterogeneity of rainfed rice ecosystems and the temporal and spatial variation that occurs in rice yield, together with the

284 Pane et al imperfections of existing weed control technologies, call for a clear understanding of the role played by weeds in limiting yields and the ecological and agronomic factors governing their persistence. Summed dominance ratio (SDR) has been widely used in the past in describing the structure of weed communities but the approach is increasingly being replaced by alternative analytical techniques. Comparisons of Figures 3 and 4 lead to different conclusions about the relative rankings of individual species, for example, Phyllanthus niruri and Digitaria ciliaris. The SDR is often an average of two indices: one based on the likelihood of occurrence (presence/absence in a sampling unit) and the other based on the absolute abundance of individual species. In consequence, the agglom- erated index is sensitive to sampling variation in two different criteria and significant departures from an underlying Poisson frequency distribution. Correspondence analyses of the type employed here are specifically designed to analyze sparse data sets (high preponderance of zeros) and to enable downweighting of the influence of rare species. Such multivariate analyses are also designed to detect underlying environmental gradients through the analysis of multiple species responses (Jongman et al 1995). Their interpretation and the construction of hypotheses of explanatory processes, however, require considerable care, as evidenced by Ammannia baccifera in this study. Indirect gradient analysis (Fig. 5) suggested that this species was associated with low toposequence regimes and species of permanently flooded fields (e.g., Cyperus difformis). Direct gradient analysis (Fig. 7), independent of toposequence, links this species with others with which it does not frequently co-occur in the field. This paradox is more apparent than real and is readily explained by the developmental growth response referred to earlier and the absence of an appropriate index capturing the dynamics of water regimes over the crop cycle (e.g., disappearance of ponded water, Jearakongman et al 1995) for use in analysis.

Conclusions As in much of rainfed rice agriculture, advances in weed management in gogorancah rice will be closely linked to improvement in crop management practices through both the deployment of improved lines (Wade et al 1999b) and integrated nutrient and water management (Tuong et al 1995). While adaptive research into the opportunities for early postemergence chemical control practices as a replacement for, or in addition to, handweeding is important, the very nature and subtlety of rainfed environments will require a much better understanding of the interaction so often described as “water-tillage-weeds.” It is here that knowledge of the effect of agronomic processes governing early recruitment of weed species is essential in relation to the inherent variance observed across the toposequence. Especially in the instance of the rainfed lowlands, farmers’ fields are fundamentally important laboratories in which to work and ones in which on-farm yield gap trials together with the analysis of farmers’ perceptions and practices in weed control remain essential adaptive research tools to protect rice yields from weeds.

Weed communities of gogorancah rice and reflections on management 285 References Bangun P, Pane H, Jatmiko SY, Moody K. 1998a. Studies of weed dynamics and their management in rainfed lowland rice. Rainfed Lowland Rice Research Consortium Phase II. Final Report 1994-97, Jakenan Experimental Station. Central Research Institute for Food Crops and International Rice Research Institute. p 59-66. Bangun P, Pane H, Jatmiko SY, Moody K. 1998b. Weed dynamics and weed management alternatives in gogorancah and walik jerami rices. Rainfed Lowland Rice Research Con- sortium Phase II. Final Report 1994-97, Jakenan Experimental Station. Central Research Institute for Food Crops and International Rice Research Institute. p 67-79. Bray JR, Curtis JT. 1957. An ordination of the upland forest communities of southern Wiscon- sin. Ecol. Monogr. 27:325-349. Cooper M, Fukai S, Wade LJ. 1999. How can breeding contribute to more productive and sustainable rainfed lowland rice systems? Field Crops Res. 64:3-12. Fagi AM. 1995. Strategies for improving rainfed lowland rice production systems in Central Java. In: Ingram KT, editor. Rainfed lowland rice, agricultural research for high-risk environments. Los Baños (Philippines): International Rice Research Institute. p 189- 200. Fujisaka S, Moody K, Ingram K. 1993. A descriptive study of farming practices for dry seeded rainfed lowland rice in India, Indonesia and Myanmar. Agric. Ecosyst. Environ. 45:115- 128. Hidayat A. 1978. Methods of soil chemical analysis. Framework Report of the Indonesia-Japan Joint Food Crop Research Program. Japan International Cooperation Agency, Bogor, Indonesia. Jearakongman S, Rajataserrkul S, Naklang K, Romyen P, Fukai S, Skulkhu E, Jumpake B, Nathabutr K. 1995. Growth and grain yield of contracting rice cultivars grown under different conditions of water availability. Field Crops Res. 44:139-150. Jongman RHG, ter Braak CJF, Van Tongeren OFR. 1995 Data analysis in community and landscape ecology. Cambridge (UK): Cambridge University Press. 299 p. Kent M, Coker P. 1992. Vegetation description and analysis. London (UK): CRC Belhaven Press. 363 p. Khush GS. 1984. Terminology for rice growing environments In: Terminology for rice growing environments. Los Baños (Philippines): International Rice Research Institute. p 5- 10. Lee HK, Moody K. 1989. Nitrogen fertilizer level on competition between upland rice and Eclipta prostrata (L.) L. In: Proceedings of the 12th Asian-Pacific Weed Science Soci- ety Conference, Seoul, Korea. p 187-193. Ludwig JA, Reynolds JF. 1989. Statistical ecology. New York (USA): Wiley. 337 p. Moody K. 1993. Weed management in rice. In: Pimentel D, editor. Handbook of pest management in agriculture. Boca Raton, Fla. (USA): CRC Press Inc. p 301-328. Mortimer AM, Lubigan R, Piggin C. 1997. Constraints and opportunities for weed management in rainfed lowland rice. Brighton Crop Protection Conference (1997) 2:191:196. Mortimer M, Hill JE. 1999. Weed species shifts in response to broad spectrum herbicides in sub-tropical and tropical crops. Brighton Crop Protection Conference (1999) 2:425-437. Pandey S. 1998. Nutrient management technologies for rainfed rice in tomorrow’s Asia: economic and institutional considerations. In: Ladha JK, Wade L, Dobermann A, Reichardt W, Kirk GJD, Piggin C, editors. Rainfed lowland rice: advances in nutrient management research. Manila (Philippines): International Rice Research Institute. p 3-28.

286 Pane et al Pane H, Mansor M. 1994. The ecology of Leptochloa chinensis (L.) Nees and its management. In: Sastroutomo SS, Auld BA, editors. Appropriate weed control in Southeast Asia. Kuala Lumpur (Malaysia): CABI, CAB International Regional Office for Asia. p 52-63. Pons TL. 1982. Factors affecting weed seed germination and seedling growth in lowland rice in Indonesia. Weed Res. 22:155-161. SAS. 1986. SAS user’s guide, statistics. Cary, N.C. (USA): SAS Institute Inc. Soerjani M, Kostermans AJGH, Tjitrosoepomo G. 1987. Weeds of rice in Indonesia. Jakarta (Indonesia): Balai Pustaka. 716 p. Syamsiah I, Suprapto, Fagi AM, Bhuiyan SI. 1994. Collecting and conserving rainwater to alleviate drought in rainfed ricelands of Indonesia. In: Bhuiyan SI, editor. On-farm reservoir systems for rainfed ricelands. Manila (Philippines): International Rice Research Institute. p 141-152. ter Braak CJF. 1985. Canonical correspondence analysis: a new eigenvector technique for multivariate direct gradient analysis. Ecology 67:1167-1179. ter Braak CJF, Smilauer P. 1998. Canoco reference manual and user’s guide to Canoco for Windows: software for canonical community ordination. Ithaca, N.Y. (USA): Micro- computer Power. 352 p. Tuong TP, Ingram KT, Siopongco J, Confesor RB, Boling A, Singh U, Wopereis MCS. 1995. Performance of dry seeded rainfed lowland rice in response to agrohydrology and N fertiliser management. In: Ingram KT, editor. Rainfed lowland rice, agricultural research for high-risk environments. Los Baños (Philippines): International Rice Research Insti- tute. p 141-156. Wade LJ, Fukai S, Samson BK, Ali A, Mazid MA. 1999a. Rainfed lowland rice: physical environment and cultivar requirements. Field Crops Res. 64:199-210. Wade LJ, Amarante ST, Olea A, Harnpichitvitaya D, Naklang K, Wihardjaka A, Sengar SS, Mazid MA, Singh G, McLaren CG. 1999b. Nutrient requirements in rainfed lowland rice. Field Crops Res. 64:91-107. Whittaker RH. 1967. Gradient analysis of vegetation. Biol. Rev. 49:207-264. Zeigler RS, Puckridge DW. 1995. Improving sustainable productivity in rice-based rainfed lowland systems of South and Southeast Asia—feeding 4 billion people: the challenge for rice research in the 21st century. GeoJournal 35:307-324.

Notes Authors’ addresses: H. Pane, E. Sutisna Noor, Research Institute for Rice, Sukamandi, Subang 41256, West Java; M. Dizon, A.M. Mortimer, International Rice Research Institute, Los Baños, Philippines. Acknowledgments: We are grateful for discussions with Dr. Sunendar Kartaatmadja (CRIFC) and Dr. Mahyuddin Syam, Dr. T.P. Tuong, Dr. L. Wade, and Mr. R. Lubigan (IRRI). Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Weed communities of gogorancah rice and reflections on management 287 Socioeconomic characterization 290 Joshi and Suresh The role of characterization in ex ante assessment of research programs: a study in the rainfed rice production system

P.K. Joshi and Suresh Pal

In a scarce and dwindling research resource scenario, characterizing production environments is a prerequisite for allocating resources more efficiently and developing a demand-driven research agenda for wider impact. This study intends to illustrate how characterization of a production system can facilitate efficient allocation of research resources. The objectives are twofold: (1) characterize the rainfed rice production system and identify existing and potential production constraints, and (2) assess research programs in an ex ante framework to allocate resources more efficiently and meet national objectives. The study has been undertaken in the rainfed rice production system, which is largely confined to eastern India. This production system lagged far behind the Green Revolution belt in agricultural development. It has now been recognized that future sources of agricultural growth lie in the rainfed rice production system. Therefore, investment in agricultural resources should be able to tap the potential of this production system. It has been characterized based on its agroclimatic features and the economic contribution of important enterprises. Characterization has aimed to delineate a homogeneous production environment to better understand common production constraints, identify technological options to alleviate these constraints in a target domain, and accelerate adoption to increase the impact of research resources. Five criteria were used to assess potential technological options in the ex ante framework: efficiency, household food security, gender issues, crop diversification, and sustainability of natural resources. These criteria have been considered important for their contribution to meeting the socioeconomic and environmental objectives in the rainfed system. The impact of efficiency has been measured using the economic surplus approach and quantifying the net present value and internal rate of returns for each technology option. Other criteria have been assigned ranks between 1 and 5 depending on their contribution. The ex ante assessment of various programs and technological options in the rainfed production system noted that the research agenda was biased in favor of a few commodities and ignored some important and potential activities. Reallocation of research resources has been proposed to develop demand-driven technological choices. The characterization of the production system also helped to identify niches for disseminating potential technological options in the target domain.

The role of characterization in ex ante assessment . . . 291 Characterization of production environments in agriculture is gaining considerable importance for identifying the major constraints to production and technology adoption (ICRISAT 1998, CRIDA 1998). Characterization of production environments is usually undertaken to understand agroclimatic environments, resource endowments, and production and socioeconomic constraints to be able to identify and prioritize research programs. Delineating homogeneous production environments makes it possible to assess research capacity and gaps. This helps in designing need-based technologies, which are expected to reduce research and adoption lags. This enhances agricultural research efficiency and accelerates the impact of investments in research. In the past, several characterization studies attempted to understand the target research domains, mainly with respect to climate, insect pests, soils, etc. Although characterization information was largely used to delineate homogeneous agro- ecoregions and production environments, and to a lesser extent design technologies for alleviating one or more production constraints, it was not further applied to assess the feasibility of research programs. With shrinking research resources and increasingly complex problems in agriculture, there is a need to use characterization information more rigorously to assess research programs and improve research efficiency. In the past, increasing food-grain production to meet self-sufficiency was the major target of agricultural research. The new set of problems has broadened the agricultural research focus to address the sustainability of natural resources, the conservation of biodiversity, and other areas. Publicly funded research has to meet multiple goals (e.g., equity, sustainability, food and nutritional security, diversification) in addition to efficiency issues for social welfare. The new paradigm therefore calls for effectively using characterization information to better target and plan agricultural research programs. This chapter attempts to use characterization information for an ex ante assessment of research programs to assist in the judicious use of scarce research resources, keeping in view the social objectives. The study focuses on the rainfed rice production system in India, a system characterized by low productivity, slow and poor dissemination of new technologies, a large concentration of poor people, high degradation of natural resources, including biodiversity, and poor infrastructure. Although this system is lagging far behind the irrigated and other favorable regions, considerable potential and opportunities exist, as it possesses fairly good soils, high precipitation, enough human resources, and a large cattle population (Joshi et al 1999). The study is based on the projects submitted to the rainfed rice production system in the rainfed ecoregion under the National Agricultural Technology Project (NATP). The World Bank approved a sum of US$239.70 million in 1998 to strengthen agricultural research in India. The World Bank and the government of India are sharing the cost of the project. Its principal objective is to address the key constraints that currently limit, and which if not addressed would in the future increasingly restrict, the efficient use of the public resources that India devotes to the generation, assessment, and transfer of agricultural technology. To facilitate the process, the country is divided into five ecoregions: arid, coastal, hills and mountains, irrigated, and rainfed. The present exercise was done for the rainfed ecoregion.

292 Joshi and Suresh Pal Approach The following steps were taken to assess research programs in an ex ante framework using characterization information:

Step I: Delineation of the rainfed rice production system To undertake more focused research in the rainfed rice production system, it was necessary to identify the research domain, which represents the predominant rainfed rice area. For this purpose, the data (1990-94 series) generated in a project on “Sus- tainable Rainfed Agricultural Research and Development” were used (ICRISAT 1998). The rainfed rice production system in India is delineated into different ecoregions as follows: 1. Agroecological subregions from 3 to 13 (delineated by the National Bureau of Soil Survey and Land Use Planning) were included because the remaining subregions fall under different ecoregions, such as arid, hills and mountains, irrigated, and coastal. This step identified 280 districts. 2. Districts having less than 40% irrigated area were selected in the second stage. This step reduced the number of districts to 152. 3. Districts having rice area more than 20% of the gross cropped area were retained to focus on rainfed rice. This yielded a list of 50 districts. 4. To maintain contiguity of districts, three (two in Uttar Pradesh and one in Maharashtra) were eliminated. This step confined the cluster to 47 districts, characterized as the rainfed rice production system. The districts identified in step 5 cover about 85% (about 10 million ha) of the total rainfed rice area in the country. The average yield of these districts is nearly 1 t ha–1.

Step II: Identification of constraints and technologies Ideally, the following steps should be adopted: 1. Identify and prioritize production constraints based on yield loss, extent of the constraints, and probability of their occurrence. 2. Identify possible technologies to alleviate constraints. 3. Develop research programs in case location-specific technologies are not available. This step is largely based on the information generated during characterization. Based on available information and scientists’ perceptions, constraints were identified in the rainfed rice production system.

Step III: Compilation of minimum data set The costs of research were estimated. To assess the potential benefits as a result of technology intervention, the following information was compiled: A. Technical information i. Existing yields of crops under study ii. Expected yields as a result of technology intervention

The role of characterization in ex ante assessment . . . 293 iii. Unit cost reduction iv. Change in resource-use pattern v. Characteristics of improved technology (i.e., technology traits) vi. Base level of area and production of major crops in the target domain vii. Fallow land B. Socioeconomic information i. Input and output prices ii. Demand and supply elasticities. These are important for assessing how farmers and consumers respond to changing prices. iii. Population below poverty line iv. Percentage of literate farm women C. Research process i. Research lag. This is the time difference between the starting year of the research project and when the research output (i.e., technology) is identified. ii. Probability of success. This is the probability of success expected in achieving the objectives set in the target period. iii. Adoption of technology. This is the expected adoption and ceiling level of adoption in the target domain. Data sets A and B were based on the information generated to characterize the production system, and the research team. Data set C was collected from the research team involved in developing technology. The information supplied by the research teams was discussed with specialists and extension agents, and some modifications were made based on their past experiences. More discussion focused on the probability of success, which largely depends on the strength of the research station in terms of facilities and human resources.

Step IV: Assessment of benefits The economic surplus approach was used to assess the potential benefits generated as a result of the technology intervention. The approach is used with the assumption that technology intervention would improve supply, reduce the unit cost of production, and benefit consumers and producers. Figure 1 gives a simple, conventional, comparative-static partial equilibrium model of supply and demand in a commodity market. DD is the demand curve for the commodity under study. S0S0 is the supply curve of the commodity under study before the technology intervention. With DD demand for a commodity and S0S0 supply of the commodity, the equilibrium price would be P0 and the quantity Q0. With technology intervention, the production function would shift upward and the unit cost of production would come down. Under this scenario, the supply curve of the commodity would shift to the right-hand side. The new supply curve would be S1S1. With the new supply function, the equilibrium price would be P1 and the quantity Q1. Prices would fall and quantity supply and demand would be higher at the new equilibrium. If prices fell, consumers would always be the gainers. But producers would be losers as a result of the fall in prices, but gainers due to the

294 Joshi and Suresh Pal Price S D 0

S1 a

P0 b P1 d c S 0 D

0Q0 Q1 Quantity Fig. 1. A framework for measuring producer and consumer surpluses.

increase in supply. The net gain to producers would depend on whether the increase in production compensates for the fall in prices. Yield-enhancement or cost-reduction technology intervention would result in a change in the consumer’s surplus equal to the area P0abP1. Similarly, the producer’s surplus would be represented by the area P1bS1 – P0aS0 (approximately P1bcd). The total economic surplus would be equal to the area S0abS1. This economic surplus is adjusted with the expected adoption and probability of success. Economic surplus is estimated for each research program in the rainfed rice production system. This information is used to assess the net present value (NPV) of each research program as follows:

n ∑ i NPV = [(ES * Ps * ADi – RCi]/(1 + r) i = 1

where ES is the economic surplus, Ps is the probability of success, ADi is the adoption of the technology in the ith year, RCi is the research cost in the ith year, r is the discount rate, and i is the time period. The internal rate of return (IRR) of each project was also computed. To compute NPV and IRR, the following assumptions were made: ● Supply and demand elasticities of different commodities were used from Kumar (1997). These are important for understanding the response of farmers and consumers in the event of changing prices and supply. ● Adoption of improved technologies as a result of research under the NATP was considered up to 2020 with technology degeneration at a linear rate after reaching the ceiling level.

The role of characterization in ex ante assessment . . . 295 ● 1997-98 was used as the base year (i.e., start of the research project) for the target domain of the improved technologies and output prices. ● The target domain of the improved technologies was assumed to be the ecological region of the research station/center(s). ● The economic surplus approach in a closed economy model was used to estimate the total economic surplus and NPV.

Step V: Assessment of other indicators In addition to the net present value (which represents the efficiency indicator), five other criteria were used: sustainability, household food security, gender, trade, and crop diversification. Since quantitative information for these criteria was not available, these were subjectively assigned on a scale of 1–5 (with 1 the lowest and 5 the maximum) based on the contribution of the research program.

Step VI: Prioritizing research programs A composite index of all the criteria (efficiency, sustainability, household food security, gender, trade, and crop diversification) was developed by assigning appropriate weights to the selected indicators. The efficiency indicator was assigned a weight of 0.50, food security 0.20, and gender issues, sustainability, and crop diversification 0.10 each. Weights were decided in consultation with the scientific community and were based on the significance of the criteria in the target domain. The purpose of developing composite indices was to rank all the projects of the proposed rainfed rice research program in view of their expected contribution toward multiple objectives.

Step VII: Prioritization across commodities at the production systems level The above steps were followed to prioritize research programs to be implemented in the target domain. This is a bottom-up approach to prioritize research programs. Prioritization across commodities at the production systems level was also done to match the national and regional priorities. For this, the congruence approach was used, along with three criteria: (1) efficiency (measured as the value of crop output), (2) equity (measured as illiterate farm women in the target domain), and (3) sustainability (fallow land was taken as the proxy for degradation of natural resources).

Prioritization of research programs Twenty research projects in the rainfed rice program under the NATP were assessed in an ex ante framework. NPV and IRR were computed using the economic surplus approach. To compute NPV and IRR, data pertaining to yield and cost of cultivation of the existing best technology and of proposed research were collected from the research teams. Table 1 contains the results of the efficiency indicator based on the ex ante assessment. The expected IRR and NPV of all the research programs were very high, which suggested a high potential for research on generating economic surplus. The top five projects generating the highest economic surplus as a result of research suc-

296 Joshi and Suresh Pal Table 1. Ex ante assessment of research programs of the rainfed rice production system.

NPVa IRRb Research project Constraint (million (%) Rank US$)

Improve crop yield ceiling Low yield 639.30 114 1 Rainwater management for Drought 277.55 184 2 drought alleviation Weed management Weeds 241.65 190 3 Control of parasitic diseases Diseases 182.80 197 4 Sustainable livestock Nutrient deficiency117.85 98 5 production systems Integrated plant nutrient Low organic matter, 114.48 142 6 management N and P deficiency Soil quality and degradation Erosion, nutrient 85.10 252 7 deficiency Integrated pest management Pest damage 77.88 79 8 Managing excess water Excess water 63.33 125 9 INMc in vertisols and alfisols Low N and P efficiency54.08 197 10 Soil tillage guidelines Crusting 53.65 109 11 INM in fish cultivation Manure unavailability41.30 167 12 Nutrient management of Low yields 39.98 141 13 hybrid rice Vegetable-based production Fallow lands 36.50 237 14 systems Impact of tank irrigation Lack of assured water 28.33 183 15 Restoration of degraded Degradation, runoff 20.35 90 16 watersheds Bioinoculants Low rhizobium 17.63 103 17 Characterization of rainfed Inappropriate diagnosis 16.95 112 18 rice system Dynamics of socioeconomic Inappropriate diagnosis 8.63 77 19 changes Crop management strategies Fallow lands 2.45 46 20 to increase cropping intensity

aNPV = net present value converted into US$ using the exchange rate of US$1 = Rs 40. bIRR = internal rate of return. cINM = integrated nutrient management. cess addressed low yields, drought management, weed management, diseases in livestock, and nutrient deficiencies in the livestock system. The next five projects focused on integrated nutrient management (INM), soil degradation, integrated pest management (IPM), management of excess water, and INM in vertisols and alfisols. In the next stage, the efficiency indicator was complemented by food security, equity/gender issues, sustainability, and crop diversification. The projects were assessed on the basis of their expected contribution to improving food security, meeting the needs of women in agriculture, enhancing the sustainability of natural resources,

The role of characterization in ex ante assessment . . . 297 cost cost

continued on next page

Ranked byResearch Cumulative

index index (million US$)

cation

diversifi- EfficiencyComposite

issues

ability

IRR Food Sustain- Gender Crop

NPV

US$)

deficiency

nutrient deficiency

runoff

efficiency

unavailability

Drought 277.55 184 5 4 3 2 2 2 0.3605 0.5520

Diseases 182.80 197 4 3 3 4 4 3 0.3603 0.9123

Excess water 63.33 125 4 5 3 4 8 5 0.1985 1.5178 Weeds 241.65 190 3 3 3 1 3 6 0.3925 1.9103 Erosion, 85.10 252 4 5 3 2 6 7 0.9275 2.0030

Degradation, 20.35 90 5 5 4 1 15 8 0.9900 2.1020

Pest damage 77.88 79 4 5 4 1 7 9 0.9028 3.0048 Fallow lands 2.45 46 4 5 3 4 18 10 0.4680 3.4728

Low yields 39.98 141 5 2 2 1 13 11 0.0640 3.5365 Fallow lands 36.50 237 3 3 3 4 14 12 0.2370 3.7735 Manure 41.30 167 3 2 2 4 12 13 0.1013 3.8748

Fallow lands 2.20 48 3 3 3 4 20 14 0.7100 3.9455 Low N and P 54.08 197 3 4 2 1 10 15 0.6875 4.0143

Constraint (million (%) security

esearch programs based on multiple goals.

oduction Nutrient 117.85 98 4 4 3 4 9 4 0.4070 1.3193

drought alleviation

systems

increase cropping intensity

fish culture

Table 2. Assessment of r Table

Project

Improve crop yield ceiling Low yield 639.30 114 4 3 3 1 1 1 0.1915 0.1915 Rainwater management for

Control of parasitic diseases Sustainable livestock pr

Managing excess water Weed management Weed Soil quality and degradation

Restoration of degraded watersheds

Integrated pest management Crop management strategies to

Nutrient management of hybrid rice Vegetable-based production systems Vegetable-based Integrated nutrient management in

Vegetable cultivation and storage Vegetable INM in vertisols and alfisols

298 Joshi and Suresh Pal cost cost

Ranked byResearch Cumulative

index index (million US$)

cation

On a scale of 1–5, where 1 = the lowest and 5 = the highest. the = 5 and lowest the = 1 where 1–5, of scale a On

diversifi- EfficiencyComposite b

issues

ability

IRR Food Sustain- Gender Crop

NPV

US$)

deficiency

water

Lack of assured 28.33 183 1 2 1 2 19 20 0.2000 5.0685

Low OM, N 114.48 142 2 4 2 1 5 16 0.1768 4.1910

Low rhizobium 4.63 59 2 4 2 2 17 18 0.2833 4.7613 Crusting 53.65 109 2 3 2 1 11 19 0.1070 4.8685

Low rhizobium 17.63 103 2 4 2 2 16 17 0.2873 4.4780

Constraint (million (%) security

management and P

NPV = net present value, IRR = internal rate of return, OM = organic matter. Exchange rate used: US$1 = Rs 40. Rs = US$1 used: rate Exchange matter. organic = OM return, of rate internal = IRR value, present net = NPV

Table 2 continued. Table

Project

Integrated plant nutrient

Bioinoculants Appropriate inoculants Soil tillage guidelines Impact of tank irrigation a

The role of characterization in ex ante assessment . . . 299 and improving crop diversification. Table 2 gives the results of this analysis. The ranking of projects changed when all indicators of national and regional priorities were considered. Increasing yield levels and improving drought management retained the same priorities, but the two projects dealing with the livestock system (diseases and nutrient management) were ranked higher because of their contribution toward sustainability and diversification. Similarly, a project such as the restoration of degraded watersheds, which was ranked 16th by efficiency criteria, moved to 8th because of the contribution of research outputs toward sustainability. This is the project that proposed to develop technologies to use fallow lands through legumes in the rice-fallow system. Although legumes contribute too little in terms of profits, their role in improving soil fertility and conserving soil and water resources is well recognized. The project was elevated because of the contribution of legumes to improving sustainability, encouraging crop diversification in fallow lands, and achieving food security. The ranking of research projects and cumulative research costs provided useful information for decision-making. The total budget requirement of all these projects under the NATP was estimated to be US$5,224,000 (or Rs 208.96 million at an exchange rate of Rs 40 = US$1). This suggested that only the top 11 projects should be supported in case $3,750,000 was available for the rainfed rice production system. The information also supports the justification for more research funds. For example, in case the last four projects are not allocated resources (which are about $877,500), the potential benefit sacrificed would be about $104,250,000. This is valuable information for research management, which suggests that raising the research budget would generate a huge economic surplus.

Role of characterization Characterization played a vital role in reallocating research resources to better target research to areas expected to contribute relatively more in various dimensions. In the present production system, if the entire research program is supported by the NATP, the budget distribution is as follows: 79% for crop production activities, including natural resource management, diagnostic surveys, and socioeconomic studies; 16% for animal husbandry; and 5% for horticultural research. Based on the contributions of different enterprises, the extent of poverty in the target domain, and the degradation of natural resources, there is a need to reallocate resources to maximize private and social goals. The analysis suggested that, within the rainfed rice production system, crop activities should receive about 61% of resources, followed by 24% for fruits and vegetables, 12% for dairy enterprises, and 3% for small ruminants. The composition of the research resource allocation changes in two subproduction systems, depending on the importance of different activities (Table 3). It will not be desirable to allocate all available research resources (61%) in the rice production system to the rice crop. This is relevant because some other crops are also of economic importance to the farming community, and these should also receive some resources, depending on

300 Joshi and Suresh Pal Table 3. Research resource allocation (%) within the rainfed rice ecosystem.

Rainfed rice Rainfed rice Aggregate Activitywith fruits and with animal allocation in vegetables husbandryrainfed ecosystem

Crop production 58 62 61 Fruits and vegetables 35 16 24 Dairy 18 5 12 Small ruminants 2 4 3

Table 4. Research resource allocation to different crops in the rainfed rice production system.

Rainfed rice Rainfed rice Rainfed rice Crop with fruits and with animal production vegetables husbandrysystem

Rice 52 49 50 Maize 2 1 1 Wheat 2 3 2 Pigeonpea 1 0 1 Rape and mustard 1 0 2 Sesamum 0 3 2 Groundnut 0 6 3 their significance within the production system. An exercise on research resource allocation across crops suggested that half of the total available resources for research on the rainfed ecosystem should go to rice research, and about 11% to other crops (Table 4). Rice research should receive the bulk of the resources in the rainfed rice production system. The crop is grown in diverse environments. According to the ecological distribution of rice, research resources for lowland rice should be about 30% (Table 5). Upland rice research should receive 15% of the total research resources available in the rainfed rice ecosystem and 5% of the total should be earmarked for deepwater rice. The information generated during the characterization of the production system provided a valuable input for setting research priorities based on the existing production constraints, and for linking this information with the aggregate-level priority setting, which includes national objectives as well.

Conclusions This study focused on an ex ante evaluation of research programs in the rainfed rice production system under the NATP. The information generated to characterize the

The role of characterization in ex ante assessment . . . 301 Table 5. Research resource allocation (%) to rice research in different ecologies.

Research Rice ecologySubecologya resource allocation

Lowland rice Shallow-water rice 20 Intermediate-water 10 rice Upland rice Upland rice 15 Deepwater Semideepwater 3 rice rice Deepwater rice 2

aShallow-water rice = 0–30 cm, intermediate water = 30–50 cm, semideepwater = 50–100 cm, and deepwater rice = >100 cm. production system was used to assess research programs for better targeting of research resources to meet the multiple goals of society: efficiency, food security, meeting the needs of women in agriculture, sustainability of natural resources, and crop diversification. The analysis revealed that there is a need to reallocate research resources to maximize research efficiency and meet social and environmental objectives. The information generated for characterization also suggested gaps in the existing research programs. Linking the information from characterization and the ex ante assessment of research programs provided several opportunities that ought to be explored through better-targeted research.

References CRIDA (Central Research Institute for Dryland Agriculture). 1998. District-based research prioritization in the rainfed rice-based production system. Hyderabad (India): CRIDA. ICRISAT (International Crops Research Institute for the Semi-Arid Tropics). 1998. Sustain- able rainfed agricultural research and development project: database development, ty- pology construction, and economic policy analysis. Module I. Patancheru (India): ICRISAT. Joshi PK, Suresh Pal, Vittal KPR. 1999. Research prioritization of the rainfed rice production system. In: Suresh Pal, Joshi PK, editors. New paradigms of agricultural research management. New Delhi (India): National Centre for Agricultural Economics and Policy Research. Kumar P. 1997. Supply demand projections. New Delhi (India): Indian Agricultural Research Institute.

302 Joshi and Suresh Pal Notes Authors’ address: National Centre for Agricultural Economics and Policy Research, New Delhi 110 012, India. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

The role of characterization in ex ante assessment . . . 303 Constraints to the adoption of modern varieties of rice in Bihar, eastern India

A. Kumar and A.K. Jha

The introduction of high-yielding modern varieties (HYVs) and seed-fertilizer technology in agriculture during the mid-1960s has led to a marked increase in the growth of agricultural output and has been instrumental in transform- ing traditional household agriculture into modern, commercial agriculture in some agriculturally developed states. The rainfed rice production system, which is largely confined to the eastern part of India, lagged far behind the Green Revolution belt in agricultural development and prosperity. Of late, it has been recognized that future sources of agricultural growth lie in this region. The adoption of modern varieties and associated technologies seems to offer an opportunity to increase output and income substantially. But even now, the pattern and pace of adoption of modern rice varieties and other component technologies have had only partial success in the rainfed rice system. Several biotic and abiotic stresses and socioeconomic and institutional constraints limit their adoption in the rainfed rice system. This study aims to analyze constraints to the adoption of modern varieties and other component technologies in the rainfed lowland ecosystem of Bihar. Both cross-sectional primary data and time-series secondary data were used. The bulk of rice is harvested in the kharif season, which suffers from the vagaries of the monsoon characterized by drought or floods and vulnerability to pests and diseases. The technical constraints to the adoption of modern varieties and component technologies are highly variable according to soil-water relationships, which differ from one ecosystem to another. The lack of tolerance for submergence, insect pests such as gall midge and brown planthopper, and diseases such as tungro, sheath blight, and bacterial blight are likely to be important obstacles inhibiting adoption in the rainfed lowland rice ecosystem. Several studies have reported the unavailability of modern varieties as the major limitation to their adoption in the rainfed rice production system. Bihar also suffers from outdated tenurial relations that adversely affect adoption. A related problem in the entire eastern region is the fragmentation of holdings. There is a need to undertake tenurial reforms in a large way and simultaneously consolidate holdings.

Constraints to the adoption of modern varieties of rice . . . 305 In addition to the unavailability of modern varieties, their poor threshability and thatchability, the requirement for additional capital and labor, excessive weed infestation, the lack of timeliness or availability of fertilizer, and lack of credit along with knowledge and cultural practices may be instrumental to their slow adoption in this region. The adoption of component technologies such as fertilizer and pesticides seems to have been affected by the adulteration prevalent in the market. The input delivery system is very weak in Bihar. The region also lacks research and development and appropriate extension services. Poor roads, communications, and market infrastructure are also important constraints. Addressing some of these constraints through appropriate research and policy intervention could have a large impact on the adoption of modern varieties and their component technologies.

The introduction of high-yielding modern varieties popularly known as seed-fertilizer technology in Indian agriculture during the mid-1960s has led to a marked increase in the growth rate of agricultural output and has been instrumental in trans- forming traditional household agriculture into modern, commercial agriculture in some of the agriculturally developed states. The rainfed rice production system, which is largely confined to the eastern part of India, lagged far behind the Green Revolution belt in terms of agricultural development and prosperity. Of late, it has been recognized that future sources of agricultural growth lie in this region. The adoption of modern varieties (MVs) and associated technologies seems to offer an opportunity to increase output and income substantially. But even now, the pattern and pace of adoption of MVs of rice and other component technologies have met with only partial success, particularly in the rainfed rice system. For instance, the rice area under high-yielding varieties (HYV) in Bihar is 60%, whereas it is 94% in Punjab. The conventional wisdom is that constraints to the rapid adoption of innovations involve factors such as the lack of credit, limited access to information, aversion to risk, inadequate farm size, inadequate incentives associated with farm tenure arrangements, insufficient human capital, the absence of equipment to relieve labor shortages (thus preventing timeliness of operations), the chaotic supply of complementary inputs (such as seed, chemicals, and water), and inappropriate transportation infrastructure (Feder et al 1985). The purpose of this chapter is to provide an analysis of constraints to the adoption of MVs of rice in eastern India on the basis of a survey of various studies that have attempted to explain these constraints and component technologies for MVs as well as a study of farm-level primary data obtained from eight villages, two each belonging to Pusa and Kalyanpur blocks of Samastipur District in the north Bihar plains and Pali and Bihta blocks of Patna District in the south Bihar plains, to make the study more comprehensive and meaningful. Among the major rice-producing states in eastern India, Bihar was selected because, even after more than three decades of

306 Kumar and Jha the Green Revolution, the coverage of HYV rice area is merely 60%, which is ahead of only Assam (50%). This chapter has two main parts. The first section briefly reviews some of the relevant past work. Because the volume of such published research is overwhelming, we have attempted simply to review some representative work rather than to present an exhaustive discussion of all work. The second section attempts to identify and analyze constraints to the adoption of MVs and other component technologies in the rainfed lowland ecosystem of Bihar. To summarize the vast amount of empirical literature on adoption constraints systematically, we organized the review of this work according to the key constraints to adoption.

Biophysical constraints The bulk of rice in the rainfed ecosystem is grown in the rainy (kharif) season, in which the vagaries of the monsoon, characterized by drought or floods, as well as the reoccurrence of other biotic and abiotic constraints are remarkable. The hot and humid climate supports the outbreak of several harmful pests and diseases. Since it is not possible to manipulate nature (climate), we need to incorporate sufficient tolerance and adaptability into the rice varieties to be developed for the rainfed ecosystem. The lack of tolerance in rice varieties for abiotic constraints such as flash floods and frequent drought as well as biotic constraints such as gall midge, brown planthopper, tungro, sheath blight, and bacterial blight seems to affect the adoption of modern rice varieties and their component technologies in the rainfed rice ecosystem. Tripathi (1977) identified susceptibility of HYVs to diseases and pests, low germination, inef- fectiveness of dry seed treatment, nonavailability of irrigation water in summer for the nursery, and serious waterlogging in the wet nursery as important constraints in coastal Orissa, whereas Gowda and Jolihal (1969) have reported the unsuitability of MVs for late planting as an important constraint. The lack of irrigation is also one of the most limiting factors affecting the adoption of MVs and their associated technologies. Roy (1976) found inappropriate irrigation facilities as the most important constraint to the adoption of HYVs in the kharif season besides disease incidence and the lack of suitable varieties in West Bengal. Shakya and Flinn (1985) examined factors influencing the adoption of MVs and fertilizer in the Tarai of southeastern Nepal and stated that adoption of MVs is highest where irrigation exists. Because rice lands in eastern India will remain rainfed in the foreseeable future, a greater spread of MVs into these adverse environments will probably depend on new varieties being bred that are specifically adapted to these environments.

Constraints to the adoption of modern varieties of rice . . . 307 Socioeconomic constraints Farm size Farm size is one of the important factors on which the empirical adoption literature focuses (Feder et al 1985). Several studies have tried to highlight the association between farm size and the adoption of HYV technology (Bhati 1976, Palmer 1976, Asaduzzaman 1979, Ahmad 1981, Misra et al 1986, Sarap and Vashist 1994) and have found a positive relationship. Contradictory evidence, however, is also not un- common. Hayami (1981), from Barker and Herdt’s (1978) study of 30 villages in five Asian countries, mentioned that the relationship between the adoption of modern rice varieties and absolute farm size for a cross-country pooled sample is negative. Muthia (1971), Schluter (1971), and Sharma (1973) found that small and medium-sized farms in India adopted HYVs on a larger proportion of area than did large farms. Fuglie (1992) found a similar result in the rainfed rice-farming system in northeast Thailand. In general, larger farms, because of their income, economic power, social pres- tige, and links with local political leadership, have a more assured supply of modern inputs including credit facilities necessary for fruitfully using the potential of new technology (Rehman 1983, Sarap and Vashist 1994). Moreover, farm size is a prerequisite for several factors such as access to credit, inputs, and information. As such, large farmers enjoy preferential treatment in obtaining such inputs (Mitra 1971, Thorner 1964, Sarap 1990, Sarap and Vashist 1994).

Risk and uncertainty Innovations in most cases face risk and uncertainty, which may affect adoption. Hazell (1982), however, found that variance in cereal production increased over time because of other factors and not because of the adoption of HYVs. On the other hand, Walker and Ryan (1990) reported the adoption of HYVs of sorghum and pearl millet to be a major contributing factor to the increased production variability of these crops. Singh and Byerlee (1990) found variability in wheat yield, measured by the coefficient of variation, to have decreased over time mainly as a result of an expansion in irrigated area. Pandey et al (2000) found that an increase in the percentage of HYV area has a stabilizing effect on yield variability, which is contrary to the conventional wisdom that HYVs lead to greater instability. Although one would expect irrigation to have a stabilizing effect, this may not hold true in the case of eastern India because rice is only partially irrigated in this region and the irrigation supply is not reliable.

Credit constraints In spite of the view that a lack of credit alone does not restrict the adoption of technologies that are scale-neutral (Schutjer and Van der Veen 1977, von Pischke 1978), differential access to capital is often cited as one of the factors in differential rates of adoption (Feder et al 1985). Several studies found that a lack of credit significantly limits the adoption of HYV technology even where fixed pecuniary costs are not large (Bhalla 1979, Frankel 1971, Wills 1972, Khan 1975, Behra and Sahoo 1975, Shakya and Flinn 1985, Sarap 1990, Fuglie 1992, Green and Ng’ong’ola 1993, Sarap

308 Kumar and Jha and Vashist 1994). There is a greater likelihood that credit constraints may create obstacles for the expansion of HYV area and the use of an optimum dose of inputs. Since credit is given by the lender on the basis of collateral provided by the borrower (Eswaran and Kotwal 1986, Sarap 1990), a poor farmer may not be in a position to raise a sufficient amount of working capital for the complete adoption of new technology. Thus, despite the availability of infrastructure, a lack of investible cash may retard the adoption of MVs and their component technologies.

Tenure Certain forms of tenurial arrangements may also affect the adoption of MVs. Parthasarathy and Prasad (1978) observed that tenants had a lower tendency to adopt HYVs than owners did. In contrast, some empirical studies do not find a clear relationship between tenure and adoption. Vyas (1975) cites studies referring to HYV wheat adoption in India that show that tenants were not only as innovative as landowners but sometimes used more fertilizer per hectare than owners did. It has been pointed out that a distinction should be drawn between pure tenants and tenant-owners because the latter can be expected to be more receptive to innovations. Schutjer and Van der Veen (1977) have opined that any observed changes were possibly due to the discriminatory access of the tenants to credit, input markets, product markets, and technical information. If these variables vary in different sociocultural environments, empirical results may seem to be in conflict if the underlying factors are not considered directly. In a backward agricultural area, however, where the landlord is an important source of finance, the landlord may discourage the tenant from adopting the MVs and associated technologies or provide inputs at his own convenience lest the tenants’ dependence on the landlord decrease. As such, tenancy may be negatively associated with the adoption of MVs.

Supply constraints The unavailability of complementary inputs is an important constraint to the adoption of MVs. In most cases, the high yield potential of the seed can be realized only if at least some fertilizers are applied (Feder et al 1985). Other inputs are also complementary to different degrees. Several studies reported the unavailability of modern varieties and associated technologies as one of the major constraints to the adoption or slow adoption of the same (Behra and Sahoo 1975, Herdt and Wickham 1978, Gupta and Kurkute 1978, Sarup and Pandey 1982). Satyanarayana and Kiresur (1990) concluded that the adoption of HYVs was proportional to the availability of complementary inputs. This issue is more relevant in a state such as Bihar where the input delivery system is very weak and input use, including that of fertilizer and credit, is low.

Labor availability HYV technology generally requires more labor inputs, so labor shortages may prevent adoption. Moreover, new technologies may increase the seasonal demand of labor, so that adoption is less attractive for those with limited family labor or those operating in areas with less access to labor markets. Hicks and Johnson (1974) have

Constraints to the adoption of modern varieties of rice . . . 309 found that a higher rural labor supply leads to greater adoption of labor-intensive rice varieties in Taiwan. Harris (1972) has found that shortages of family labor explain nonadoption of HYVs in India. Several studies, however, have not found the availability of labor to be a major determinant of the adoption of new technologies (Shakya and Flinn 1985). The above list of adoption constraints is not exhaustive and it may differ from one ecosystem to another. Conflicting conclusions can also emerge from studies from different regions or countries because of different social, cultural, and institutional environments apart from pure economic factors. It is therefore essential to provide comprehensive information about the interactions among the various factors that generate the observed adoption patterns.

Analysis of constraints Bihar accounts for nearly 12% of the rice area of the country but its share of total rice production is only around 9%. In Bihar, rice occupies around 50% of the gross cropped area; however, about 70% of the rice area in Bihar is rainfed. The rainfed lowland rice ecosystem constitutes the highest area (40%), followed by the rainfed upland rice ecosystem (20%) and deepwater rice ecosystem (10%). Because the rainfed lowland ecosystem has the most rice area, this study was designed to identify and analyze constraints to the adoption of MVs of rice and their component technologies in this ecosystem.

Data and methodology The study used both secondary and farm-level primary data. Time-series secondary data were collected and used to analyze trends in area, production, and yield of rice in Bihar from 1970 to 1998 (Table 1). The trends in adoption of HYVs of rice and fertilizer consumption in the state were also analyzed. Compound annual growth rates were estimated to analyze the trends. To study the decade-wise performance, the span of 28 years was divided into three periods: (1) 1970-71 to 1979-80, (2) 1980-81 to 1989-90, and (3) 1990-91 to 1997-98.

Table 1. Area, production, and yield of rice in Bihar.a

Area Production Yield Years (000 ha) (000 t) (kg ha–1)

1970-71 5,134 4,631 903 1980-81 5,339 4,496 838 1990-91 5,156 5,791 1,117 1997-98 5,027 6,899 1,372

aData are for triennium ending average.

310 Kumar and Jha Farm-level primary data were obtained from 80 randomly selected farmers, from eight villages belonging to the four development blocks in two each of the north (Samastipur) and south Bihar (Patna) plains. Ten farmers from each of the sample villages were selected randomly from lists of the farmers having rainfed lowland rice area. The information on existing farming practices, varieties, yields, input use, different adoption constraints, etc., was obtained from the sample farmers with the help of a comprehensive questionnaire from the selected sample farms. In addition, a comprehensive list of adoption constraints was given to them and they were asked to assign a value of one to the most limiting adoption constraint, two to the next important one, and so on. Then the rank values were averaged across the villages and a composite score was obtained on the basis of which the top ten and five constraints were identified for the adoption of MVs and fertilizer application, respectively. This information was cross-checked with the scientists working in the area to make the data more reliable.

Results and discussion Annual compound growth rates in rice area, production, and yield were computed for different periods in the state (Table 2). It is evident from Table 2 that Bihar did not show significant growth in rice production and yield during 1970-79. There was a conspicuous change in rice production in the state during the 1980s, however, when production increased at an annual growth rate of 4.05%. Production gained further momentum and increased at an annual rate of 5.36% during the 1990s. The increased rice production in the state during both subperiods (1980-89 and 1990-97) came almost entirely from yield increments (about 93% and 98%, respectively). Taking all the periods together (1970-97), a significant growth of 1.65% in yield and 1.38% in production was observed. The trend in this chronically stagnant zone with low productivity was therefore positive.

Area under high-yielding varieties of rice Table 3 shows the coverage of area under HYVs of rice from 1970-71 to 1996-97 in the state. Analysis reveals that high-yielding rice varieties are gaining wider acceptability in the state in recent years. Rice area under HYVs was only about 7% of total Table 2. Growth patterns of area, production, and productivity of rice in Bihar.

Growth rate (%)a Item 1970-79 1980-89 1990-97 1970-97

Area 0.51 0.25 0.09–0.27* Production 0.294.05* 5.36* 1.38** Yield –0.21 3.80** 5.27* 1.65***

aCoefficient of the semi-log regression. ***,**,* = significant at the 1%, 5%, and 10% level, respectively.

Constraints to the adoption of modern varieties of rice . . . 311 Table 3. Rice area under HYVs from 1970-71 to 1997-98 in Bihar.

Area under Area under Area of rice Annual Year HYVs rice under HYVs growth ratea (million ha) (million ha) (%) (%)

1970-71 0.38 5.13 7.4 – 1980-81 1.32 5.34 24.7 10.34* 1990-91 1.64 5.16 31.8 5.02 1997-98 3.13 5.03 62.2 35.14* 1970-97 – – – 8.98*

a* = significant at 1% level. Source: Fertilizer Statistics (various issues), Fertilizer Association of India, New Delhi, Agriculture, CMIE, September 1999.

Table 4. Fertilizer consumption in Bihar (kg ha–1).

Years N P2O5 K2O NPK

1970-71 7.47 1.66 0.83 9.96 1980-81 13.47 2.73 1.27 17.47 1990-91 40.70 11.07 4.40 56.17 1997-98 62.43 12.77 4.93 80.13

Source: Fertilizer Statistics (various issues), Fertilizer Association of India, New Delhi, Agriculture, CMIE, September 1999. rice area during 1970-71, but increased to 62% during 1997-98. The coverage of HYV rice area thus increased at a compound annual growth rate of 9%. The adoption of HYVs of rice in the state during the 1990s (especially after 1992-93) recorded an unprecedented annual compound growth rate of 35%. This transformation was possible due to the joint efforts of ICAR-DOAC in taking the improved technologies from the shelf to the farmers through the compact block frontline demonstration—a unique experiment launched in 1990-91. Since 1990-91, 340 demonstrations have been conducted with the active support and cooperation of agricultural universities and departments of agriculture. This experiment disseminates knowledge and facilitates large-scale farmer participation, enabling farmers to choose varieties and a production package suited to their farm situations (Siddiq 1999).

Fertilizer consumption Table 4 shows the consumption of chemical fertilizers in Bihar. Fertilizer consumption was a mere 10 kg ha–1 in 1970-71, but increased to 17 kg ha–1 in 1980-81 and to a high of 80 kg ha–1 in 1997-98, recording an annual compound growth rate of 7.86% during 1970-97. Fertilizer consumption witnessed a spectacular increase, especially in the 1990s (consumption increased by 150% in 7 years). Despite such an impressive jump in the use of chemical fertilizers, the state is still far below the level of fertilizer consumption in states such as Punjab (178.6 kg ha–1), Haryana (141.6 kg ha–1), Uttar

312 Kumar and Jha Pradesh (108 kg ha–1), West Bengal (120 kg ha–1), Tamil Nadu (152 kg ha–1), and Andhra Pradesh (153 kg ha–1). Moreover, the use of fertilizer nutrients was not balanced in the state.

Area under traditional and modern varieties of rice on sample farms Table 5 shows the area under traditional and modern rice varieties on selected farms. An inquiry into the descriptions of land area among the farmers growing rice in the rainfed lowland ecosystem in sample districts of the north and south Bihar plains clearly reveals that rice area in Samastipur District was more than that of Patna Dis- trict. However, the rice area under HYVs was considerably higher in Patna than in Samastipur. This conforms with the district-level secondary data (Jha and Viswanathan 1999). The reason for such a big difference in the adoption of MVs between the two districts might be the difference in existing irrigation potentials. The actual irrigated area in north Bihar is lower and the available sources of irrigation are costly, which compels farmers to cultivate traditional varieties.

Adoption of traditional and modern rice varieties on sample farms Table 6 shows the position of farmer-adopters in both districts. It is clear from the table that, out of 40 sample farmers in Patna, only two farmers grew MVs exclu- sively, whereas in Samastipur 25 farmers grew only MVs. This shows that improved modern varieties are more popular in Patna and MVs were replacing traditional varieties (TVs) at a higher rate in this district. The replacement of TVs by MVs was meager in Samastipur. However, MVs are gaining wide acceptance along with TVs as 47.5% of the farmers adopted MVs along with TVs. In spite of MVs gaining popularity among the farmers, the predominance of TVs was uninterrupted. For adopters and nonadopters of MVs in the Bihar plains (taking both districts together), the number of farmer-adopters was apparently high. More than 70% of the farmers identified were either cultivating MVs only (33.75%) or MVs and TVs both (37.5%). The percentage of farmers cultivating only TVs was 28.75%.

Table 5. Area under traditional and modern rice varieties on sample farms.

Area under Area under No. of Net Area under modern traditional Sample zones farms cultivated ricea varietiesb varietiesb area (ha) (ha) (ha)

North Bihar plains 40 86.2 42.3 14.3 28.0 (49.1) (33.8) (66.2) South Bihar plains 40 78.8 32.7 27.4 5.3 (41.5) (83.7) (16.3) Total 80 165.0 75.0 41.7 33.3 (45.5) (55.6) (44.4)

aNumbers in parentheses indicate area under rice as a percentage of total cultivable land area. bNumbers in parentheses indicate area under modern and traditional rice varieties as a percentage of area under rice.

Constraints to the adoption of modern varieties of rice . . . 313 Rice varieties in farmers’ fields and recommended varieties Table 7 lists different varieties that were released for cultivation in Bihar under different situations of the rainfed lowland ecosystem. It is obvious from Table 7 that there are only 10 recommended MVs for such a vast lowland area (17.84 million ha) in the state. Table 6 reveals that, in spite of the presence of MVs, several TVs are popular among farmers in the study area. Even with MVs, varieties such as Sita and Jaya are being cultivated in some areas in shallow rainfed conditions although these varieties are not suitable for rainfed cultivation. This clearly reflects the poor technical knowledge about the varieties among the farmers because of insufficient technical guidance and poor transfer of technology. This is probably one of the most important reasons

Table 6. Adoption of traditional and modern rice varieties on sample farms.

No. of No. of No. of Sample Total no. farmers farmers farmers district of farmersa growing MVs growing growing and TVs both MVs only TVs only

Samastipur 40 192 19 (north Bihar (100) (47.5) (5.0) (47.5) plains) Patna 40 11 25 4 (south Bihar (100) (27.5) (62.5) (10.0) plains) Total 80 30 27 23 (100) (37.5) (33.8) (28.8)

aNumbers in parentheses indicate percentage of sample farmers in respective sample zones.

Table 7. Rice varieties recommended for the rainfed lowland ecosystem in Bihar.

Land situation Modern varieties Varieties cultivated released by farmers

Shallow lowland Jayshree, Rajshree, Sita, Jaya, BR34, Mahsuri, Vaidehi, Bakol, Ramraji, Radha, Pankaj, Panjabi, Patania, BR3 Permal Intermediate lowland Jayshree, Radha, Rajshree, BR34, Pankaj, BR8 Radha, Bakol, Punjabi, Hathijhuln, Chenab, Latisail, Patania, Permal Semideep Janki, Sudha Janki, Sudha, Hathijhuln, Bakol, Dermi, Pankaj

314 Kumar and Jha that create negative impressions about the MVs among the farmers as the injudicious selection of varieties often results in large failures. This also compels the farmers to continue cultivating low-yielding but to some extent reliable TVs.

Yield gap between modern and traditional rice varieties Table 8 shows the average yield gaps between modern and traditional rice varieties in Samastipur and Patna districts. The average yield of MVs on the sample farms of Samastipur was 2.6 t ha–1, whereas the average estimated yield of the TVs was 2.2 t ha–1. Although the average yield of MVs in Patna was higher (2.7 t ha–1) than that of Samastipur, the yield of TVs was comparatively lower in Patna (2.1 t ha–1).

Constraints to the adoption of modern varieties and their component technologies The rice area under HYVs and fertilizer use in Bihar have registered a significant increase in recent years. This progress, however, has not yet reached the expected level and the poor adoption of modern rice varieties under the rainfed lowland ecosystem in the Bihar plains is cited as one of the main reasons for the lower productivity of rice in Bihar. Farmers have several constraints to the adoption of MVs and their associated technologies. Table 9 presents different constraints that affect the adoption of modern rice varieties and that many farmers in the study area often face. The nonavailability of MVs or unavailability in time has emerged as the most important constraint to their adoption in the rainfed lowland ecosystem of Bihar. This reaffirms the poor condition of the state’s input delivery mechanism. Cereal farms in developing countries such as India often have three major sources of seed: seed purchased from a formal seed industry, seed obtained from other farmers, and seed retained from the previous year’s grain crop (Tetlay et al 1991). More than 85% of the seed consumed in India was produced by the farmers themselves (Banerjee 1984). The next important source was fellow farmers. The share of the organized seed sector was meager on account of the high price of certified seeds and nonavailability at proper places in time (Sidhu 1999). The gap between the seed requirement and seed actually distributed in India for paddy was about 52% in 1992-93 (Sidhu 1999). This figure was expected to be higher in Bihar. The large gap between the requirement of certified/quality seeds and their dis-

Table 8. Yield gap between modern and traditional rice varieties in the rainfed lowland ecosystem in Bihar (kg ha–1). Yield figures are in terms of unhusked paddy.

Av yield of Av yield of Sample modern traditional Yield gap district varieties varieties

Samastipur 2,575 2,212 363 (14.1) Patna 2,690 2,056 634 (23.6)

Constraints to the adoption of modern varieties of rice . . . 315 Table 9. Constraints for low adoption of modern varieties in Bihar plains.

Composite Constraint score Rank

Unavailability of seeds/nonavailability of seeds in time 1.34 1 High input costs 1.83 2 Susceptibility to drought/flood 3.23 3 Prone to insect pests and diseases 4.33 4 Low profitability 4.88 5 Poor threshability and thatchability 5.55 6 Unavailability of credit 6.67 7 Bad tenurial system 7.11 8 Poor extension service 7.60 9 Scarcity of labor 7.91 10 tribution in the country is a matter that requires serious attention for increased seed production. The high input cost associated with the cultivation of MVs has also been perceived as the next important constraint to MV adoption although a huge amount is being spent on input subsidies. The MVs being cultivated in the area are susceptible to either drought or flood or both and these are also more prone to insect pests and disease infestation (CRRI 1997). Although partial solutions are available in the form of resistant varieties and diversified cropping patterns, their adoption is often delayed by the slow pace of technology transfer (Siddiq 1996). These factors limit the adoption of MVs of rice in the study area considerably. Farmers in this area urgently need a variety that can withstand water stress as well as submergence to some extent. Low profitability (relatively) has also emerged as an important constraint. As pointed out by Dr. Norman E. Borlaug, people would adopt technology to increase yield only if it gives returns of 200% or more (Hindu Survey, Agriculture, 1999). He disagrees with the notion that it is difficult to bring about changes. Such a psychological hurdle can be removed by the effective demonstration of technology. Poor threshability and thatchability also restrict the adoption of MVs as they limit the use of straw as fodder and for other domestic purposes. Few credit facilities are available to farmers in most of the eastern states, especially in Assam, Orissa, and Bihar, and they are often less than what a farmer in Punjab, Haryana, and Kerala has. Further, the farmers’ access to credit is frustrated by financial institutions, which adopt complicated procedures for granting and recovering agricultural loans. A review and necessary improvement of the system will help in getting the benefits and facilities provided by the government to the farmers in full measure. Bihar also suffers from outdated tenurial relations, which also adversely affect the adoption of MVs in the state. A related major problem is the fragmentation of holdings. There is a need to undertake tenurial reforms in a large way and simultaneously to consolidate holdings. Institutional development must also receive high priority, as it would facilitate the adoption of MVs and their compo-

316 Kumar and Jha Table 10. Factors affecting the adoption of chemical fertilizers on sample farms.

Composite Factors/constraints score Rank

High fertilizer cost and unprofitable use 1.291 Unavailability of fertilizers in time 1.78 2 Scarcity of capital 2.77 3 Operational difficulty 3.33 4 Adulteration 3.88 5 nent technologies. Another problem facing this region is the lack of appropriate extension services. Measures have to be devised to acquaint the farmers with modern production technology through adequately trained personnel and to provide them with facilities for soil testing and proper advice on upgrading production technology. More- over, the communication gap between researchers and farmers should be bridged.

Adoption of chemical fertilizer Modern rice varieties are highly responsive to chemical fertilizers; however, most farmers do not adopt recommended doses of chemical fertilizers in their fields. Table 10 presents the factors that affect the application of chemical fertilizers to rice in the rainfed lowland production system. All the farmers responded that high fertilizer costs make their application unprofitable in the rainfed lowland ecosystem because dependence on monsoon and inefficient water management techniques reduce the efficiency of chemical fertilizers. A large portion of chemical fertilizers becomes unavailable to the plant either by leaching or by being fixed in the soil. Besides the high fertilizer cost, the problem of unavailability of chemical fertilizer during the peak season is one of the most limiting factors for the application of fertilizer in rainfed lowland rice. Therefore, the timely supply of good-quality fertilizers, especially in remote areas, is equally important. This requires a revamping of the entire infrastructure for an effective fertilizer supply distribution system. Farmers also reported that the lack of capital in their command areas made the purchase of chemical fertilizer economically unaffordable; thus, they apply the minimum possible doses. Farmers also face difficulty in applying fertilizers because their fields were unbunded and thus they believed that fertilizer application would be ineffective. The prevailing adulterated fertilizer in markets also makes farmers wary of its possible adverse effect on the crop.

Conclusions and policy implications Agriculture in the rainfed lowland ecosystem has performed well in recent years. Production will probably never match that in the most productive irrigated areas because of inferior agroclimatic conditions, but growth potential still exists (Bhalla et al 1999). The higher adoption of MVs and other component technologies can become a

Constraints to the adoption of modern varieties of rice . . . 317 vehicle for realizing the untapped growth potential in this region. Hence, a twin pillar strategy of varietal improvement and appropriate improved production technologies addressing these constraints would be ideal. The higher adoption of new technologies in the rainfed rice ecosystem has to be matched with suitable policy initiatives for desired results. Some critical areas for intervention would be (1) improving the availability of seed/planting material of HYVs, (2) strengthening the input (seed, fertilizer) delivery system, (3) radically reorienting credit policies and procedures, (4) continuously transferring technology through assessment and refinement as well as improving and strengthening the existing extension system, (5) mechanizing small farms, (6) promoting water harvesting for judicious use of water, (7) achieving reforms in tenancy and associated laws, (8) extending crop insurance schemes to provide safe- guards against risk and uncertainty, and (9) ensuring remunerative prices and improving marketing infrastructure. Moreover, there is a crucial need for a macro policy for creating a favorable environment for the better flow of information, investment, inputs, and technology, although the macro policy in itself is not sufficient to deliver the expected results. This is largely due to the characteristic features of Indian agriculture. This underscores the necessity of decentralized or micro planning if the adoption of modern rice varieties and their component technologies is to be raised substantially.

References Ahmad I. 1981. Technical changes and agrarian structure: a study of Bangladesh. Geneva (Swit- zerland): International Labour Office. Asaduzzaman M. 1979. Adoption of HYV rice in Bangaladesh. Bangladesh Dev. Stud. 7(3):23- 52. Banerjee SK. 1984. Seed in Indian agriculture during 1980s and beyond. Indian Agric. 28(2): 115-120. Barker R, Herdt RW. 1978. Equity implications of technology changes. In: Interpretive analysis of selected papers from Changes in Rice Farming in Selected Areas of Asia. Los Baños (Philippines): International Rice Research Institute. Behra C, Sahoo MS. 1975. Impact of national demonstration on adoption of agricultural practices. Indian J. Ext. Educ. 11(1-2):32. Bhalla GS, Hazell P, Kerr J. 1999. Prospects for India’s cereal supply and demand to 2020. Food, Agriculture, and the Environment Discussion Paper 29. Washington, D.C. (USA): International Food Policy Research Institute. Bhalla Surjit S. 1979. Farm and technical changes in Indian agriculture. In: Berry R, Cline W, editors. Agrarian structure and productivity in developing countries. Baltimore, Md. (USA): Johns Hopkins University Press. Bhati UN. 1976. Some social and economic aspects of the introduction of new varieties of paddy in Malaysia: village case study. Geneva (Switzerland): United Nations. CRRI (Central Rice Research Institute). 1997. Perspective plan. Cuttack (India): CRRI. Eswaran M, Kotwal A. 1986. Access to capital and agrarian production organization. Econ. J. 96(382):482-498.

318 Kumar and Jha Feder G, Just RE, Zilberman D. 1985. Adoption of agricultural innovations in developing countries: a survey. Econ. Dev. Cult. Change 33(2):263-283. Frankel FR. 1971. India’s green revolution: economic gains and political costs. Princeton, N.J. (USA): Princeton University Press. Fuglie KO. 1992. The adoption of new agricultural technology in a rainfed rice farming system in northeast Thailand. Unpublished thesis, Humanities and Social Sciences, University of Minnesota, Minneapolis, Minn., USA. Gowda Channe MB, Jolihal KA. 1969. Adoption of recommended varieties and seedling practices of paddy by farmers. Indian J. Ext. Educ. 10(1&2):70. Green DAG, Ng’ong’ola DH. 1993. Factors affecting fertilizer adoption in less developed countries: an application of multivariate logistic analysis in Malawi. J. Agric. Econ. 44(1):99-110. Gupta DS, Kurkute KD. 1978. New rice technology and small farmers. Eastern Econ. 71(14):666- 668. Harris B. 1972. Innovation adoption in Indian agriculture: the high yielding variety program. Modern Asian Stud. 6. Hayami V. 1981. Induced innovation, green revolution, and income distribution: comment. Econ. Dev. Cult. Change 30:169-176. Hazell PBR. 1982. Instability in Indian foodgrain production. Research Report 30. Washing- ton, D.C. (USA): International Food Policy Research Institute. Herdt RW, Wickham TH. 1978. Exploring the gap between potential and actual rice yields: the Philippine case. In: Economic consequences of the new rice technology. Los Baños, Laguna (Philippines): International Rice Research Institute. p 3-29. Hicks WH, Johnson R. 1974. Population growth and the adoption of new technology in Tai- wanese agriculture. Working Paper in Economics No. 1974-E6. Columbia, Mo. (USA): University of Missouri. Jha TN, Viswanathan. 1999. Problems and prospects of agricultural development in Bihar. Occasional Paper-10. Mumbai (India): National Bank for Agriculture and Rural Devel- opment. Khan MH. 1975. The economics of the Green Revolution in Pakistan. New York (USA): Frederick A. Misra B, Mukhopadhyay SK, Flinn JC. 1986. Production constraints of rainfed lowland rice in eastern India. In: Progress in rainfed lowland rice. Los Baños, Laguna (Philippines): International Rice Research Institute. p 25-52. Mitra A. 1971. Delivery and distribution of inputs of agriculture to small farmers in India. A Background Paper for the Raman Magsaysay Award Foundation Seminar held at Manila. Muthia C. 1971. The Green Revolution: participation by small versus large farmers. Indian J. Agric. Econ. 26:25-34. Palmer I. 1976. The new rice in Asia: conclusion from four country studies: Report of the United Nations Research Institute of Social Development, Geneva. Pandey S, Singh HN, Villano RA. 2000. Rainfed rice and risk-coping strategies: some microeconomic evidence from eastern Uttar Pradesh. In: Pandey S, Barah BC, Villano RA, Pal S, editors. Risk analysis and and management in rainfed rice systems. Limited Proceedings of the NCAP/IRRI Workshop on Risk Analysis and Management in Rainfed Rice Systems, 21-23 Sep 1998, National Centre for Agricultural Economics and Policy Research, New Delhi, India. Los Baños (Philippines): International Rice Research Insti- tute. p 115-134.

Constraints to the adoption of modern varieties of rice . . . 319 Parthasarathy G, Prasad DS. 1978. Response to the impact of the new rice technology by farm size and tenure: Andhra Pradesh. Los Baños (Philippines): International Rice Research Institute. Rehman RI. 1983. Adoption of HYV: role of availability and the supply side problems. Bangladesh Dev. Stud. 11(4). Roy GL. 1976. Reasons for low spread of HYV paddy during the pre-kharif and kharif season in West Bengal. Indian J. Ext. Educ. 12(1-2):50. Sarap K. 1990. Factors affecting small farmers’ access to institutional credit in rural Orissa, India Dev. Changes 21(2):281-307. Sarap K, Vashist DC. 1994. Adoption of modern varieties of rice in Orissa: a farm level analysis. Indian J. Agric. Econ. 49(1):88-93. Sarup S, Pandey RK. 1982. Investigation into non-adoption of rice varieties in Orissa: an application of discriminant function techniques. Margin 15(1):71-77. Satyanarayana M, Kiresur VR. 1990. An investigation into partial adoption of HYVs of kharif rice in India. Agric. Situation India 45(5):339-344. Schluter M. 1971. Differential rates of adoption of the new seed varieties in India: the problem of small farms. U.S. Agency for International Development, Occasional Paper No. 47. Ithaca, N.Y. (USA): Cornell University. Schutjer W, Van der Veen M. 1977. Economic constraints on agricultural technology adoption in developing countries. U.S. Agency for International Development, Occasional Paper No. 5. Washington, D.C. (USA): USAID. Shakya PB, Flinn JC. 1985. Adoption of modern varieties and fertilizer use on rice in the eastern Tarai of Nepal. J. Agric. Econ. 36(3):409-419. Sharma AC. 1973. Influence of certain economic and technological factors on the distribution of cropped area under various crops in the Ludhiana District. J. Res. Punjab Agric. Univ. 10:243-249. Siddiq EA. 1996. Rice: development opportunities. Hindu Surv. Agric. Siddiq EA. 1999. Rice: not a distant dream. Hindu Surv. Agric. Sidhu MS. 1999. Impact of intellectual property rights on the Indian seed industry. Indian J. Agric. Econ. 54(3):370-379. Singh AJ, Byerlee D. 1990. Relative variability in wheat yields across countries and over time. J. Agric. Econ. 41:21-32. Tetlay KA, Heisey PW, Ahmed Z, Munir A. 1991. Farmers’ sources of wheat seed and wheat seed management in three irrigated regions of Pakistan. Seed Sci. Technol. 19(1):123- 138. Thorner D. 1964. Agricultural cooperative in India: a field report. Bombay (India): Asia Pub- lishing House. Tripathi A. 1977. A study of technology gap in adoption of new rice technology in coastal Orissa and constraints responsible for the same. Unpublished PhD thesis. Division of Agricultural Extension, Indian Agricultural Research Institute, New Delhi. von Pischke J. 1978. When is smallholder credit necessary? Dev. Dig. 26:6-14. Vyas VS. 1975. India’s high yielding varieties programme in wheat. 1966/7-1971/2. Mexico City (Mexico): Centro Internacional de Mejoramiento de Maíz y Trigo. Walker TS, Ryan JG. 1990. Village and household economies in India’s semi-arid tropics. Baltimore, Md. (USA): Johns Hopkins University Press. Wills IR. 1972. Projection of effects of modern inputs on agricultural income and employment in a C.D. Block, U.P., India. Am. J. Agric. Econ. 54(3):452-460.

320 Kumar and Jha Notes Authors’ address: National Centre for Agricultural Economics and Policy Research, Pusa, New Delhi-12, India. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Constraints to the adoption of modern varieties of rice . . . 321 Appendix. Agroclimatic zones of Bihar The state of Bihar is divided into six agroclimatic zones based on rainfall and temperature, soil type, and physiographic features: (I) northwest alluvial plains, (2) northeast alluvial plains, (3) south Bihar alluvial plains, (4) central and northeastern plateau, (5) western plateau, and (6) southeastern plateau. Of these, the first three form the segment of the middle Gangetic plains. The climate is dry to moist subhumid and the soil type is heavy-textured sandy loam to clayey, medium acidic. This subzone receives more than 1,200 mm of rainfall annually. The last three form part of the eastern plateau and hills. The climate of this subzone is moist subhumid to subhumid and the soil is red sandy, loamy, red, and yellow. It receives a higher rainfall of around 1,300–1,400 mm but irrigation development is very poor as only 8% to 10% of the cultivated area is irrigated. The Rajendra Agricultural University, Pusa, however, classified Bihar into four comprehensive agroclimatic zones: (1) northwest alluvial plain, (2) east alluvial plain, (3) southwest alluvial plain, and (4) plateau region. The selected districts, Samastipur and Patna, come under the northwest alluvial plain and southwest alluvial plain, respectively. The southwest alluvial plain has about 33% of the state’s total production of rice with around 24% area. The northwest plains contribute about 26% to rice production in the state. The east plains contribute about 18%. The rest is contributed by the plateau region. Across different zones, the southwest zone displayed higher yield (1.58 t ha–1) than the northwest plains (1.08 t ha–1).

322 Kumar and Jha Rainfed rice, risk, and technology adoption: microeconomic evidence from eastern India

H.N. Singh, S. Pandey, and R.A. Villano

Risk and risk aversion are often believed to be important factors that constrain the adoption of technology by poor farmers. The nature and magnitude of risk and how farmers manage risk, however, have been poorly studied in the context of rainfed rice. This chapter provides an analysis of variability in rice production and farmers’ coping strategies using farm-level panel data from two villages of eastern Uttar Pradesh, India. The results indicate that crop diversification is an important farmer strategy for dealing with risk. Through crop and income diversification, farmers have been able to reduce the effect of risk in rice production on the variability of total household income. The economic cost of risk in rice production was found to be quite low, implying that yield-increasing rather than yield-stabilizing rice technologies are likely to be more appropriate in these environments. The adoption of modern varieties was conditional on irrigation, farmers’ education, farmers’ age, and their wealth status as represented by farm size. Implications of these results for technology design and policy reforms are discussed.

Uncertainties involved in rainfed rice farming are manifold and farmers face considerable challenges in dealing with these. Of all the elements of risk, the most important for agriculture is rainfall. Risk under rainfed conditions generally tends to be high as variability in rainfall can lead to wide swings in yield and output. Poor farmers, when faced with high levels of risk, may respond by adopting practices that reduce risk even if they entail a reduction in income on average. To the extent that risk and risk aversion constrain technology adoption, the study of risk and farmers’ mechanisms for coping with risk is important for technology design and policy reforms. Farmers cope with risk by developing various strategies over time. These strategies can be classified into ex ante and ex post depending on whether they help reduce risk during the production process or reduce the impact of risk after a production shortfall has occurred. Because of the absence of efficient market-based mechanisms for diffusing risk, farmers modify their production practices to provide self-insurance

Rainfed rice, risk, and technology adoption: . . . 323 such that losses are reduced during unfavorable years (Pandey et al 1998). Ex ante strategies can be broadly grouped into two categories: those that reduce risk by diversification and those that do so by maintaining flexibility. Spatial diversification of farms, diversification of agricultural enterprises, diversification of crops and varieties, and diversification of income among farm and nonfarm activities are some examples (Siddiq and Kundu 1993, Singh et al 1995). Similarly, maintaining flexibility is an adaptive strategy that allows farmers to switch resources among activities as required to reduce risk. Loss management or ex post strategies are designed to prevent a shortfall in consumption when the family income drops below what is necessary for maintaining consumption at its normal level. These include seasonal migration, consumption loans, and asset liquidation (Jodha 1978). However, these strategies are often inadequate to prevent a reduction in consumption (Pandey et al 2000). The objective in this chapter is to characterize the nature and magnitude of risk in rainfed rice areas of eastern Uttar Pradesh, India. Farmers’ ex ante risk-coping strategies and the extent to which risk and risk aversion may be constraints to technology adoption are also analyzed.

Methodology and data The study of risk and risk management strategies requires temporal data as it is the uncertainty in production and income over time that is of concern to individual farmers. Such data also permit an assessment of the dynamic nature of risk and adjustments that farm households undertake in response to risk. Accordingly, the study is based on panel data collected from two villages in eastern Uttar Pradesh for the period 1994-98. Ninety respondent farmers were randomly selected, 45 farmers from each village, and detailed data on rice production practices and other farm and nonfarm activities of farm households were collected using the interview method. Pro- duction data were collected at the plot level. The two villages were selected to provide a contrast in production systems. The village Mungeshpur has better access to irrigation, has more lowland fields, and the adoption of modern varieties is greater. In contrast, Itgaon has a poorer irrigation infrastructure, has a relatively higher proportion of upland fields that are more easily drained, and the adoption of modern varieties is lower (Table 1). Thus, Mungeshpur provides a “benchmark” against which the production strategies of farmers in the more risk-prone Itgaon could be assessed. A comparison of rice production practices in these two villages provided the basis for assessing farmers’ responses to differential risks. Our analytical method consisted of computing the variability of rice yield, input use, and income at both the plot and farm level. Data were analyzed by pooling the cross-sectional data with temporal data. To remove farmer-specific effects in plot- level data, a fixed-effects model was specified using farmer-specific dummy variables. The estimating equation was of the form

324 Singh et al Table 1. Basic characteristics of the study area.

Characteristics Mungeshpur Itgaon

No. of respondents 45 45 Irrigated area (%) 85 37 Average operational 0.7 1.4 holding (ha) Proportion of land type (%) Upland 34 55 Medium land 14 24 Lowland 52 21 Average years of 5 6 schooling of household head Average household 7 8 size (no. of members)

Yijt = a + bFj + eijt (1)

where Yijt is the yield of the ith field for the jth farmer in year t, Fj is the dummy variable for the jth farmer, and e is the random error term. The residuals of this dummy variable model were used to estimate the variance of yield and net returns from rice. The availability of panel data also permitted the estimation of variances of rice income and total household income for each household over time. Estimates of these temporal variances and coefficients of variation were then used to assess the potential effect of uncertainty in rice production on the variability of household incomes. Crop diversification is a potentially risk-reducing strategy. If crop yields across different crops are poorly or negatively correlated, crop diversification can reduce the instability in total output. The extent of crop diversification was computed by an index using equation (2):

2 D = 1 – Σ (aij/Ai) (2) j

where aij is the area planted to the jth crop by the ith farmer and Ai is the total cropped area of the ith farmer. The index D ranges from zero for complete specialization in one crop to the maximum value of unity indicating a high degree of diversification. If farmers choose activities that generate a lower average income over time in an attempt to reduce risk, the “cost of risk” is the mean income forgone. The proportional risk premium is the cost of risk expressed as a proportion of the total mean income. Under the simplifying assumption that the net returns are normally distributed, the proportional risk premium can be approximated by the following equation (Pandey et al 1999):

Rainfed rice, risk, and technology adoption: . . . 325 2 2 P = 0.5 R [a Cr + 2 a (1 – a) g Cr Cy](3)

where R is the coefficient of relative risk aversion, Cr is the coefficient of variation (CV) of rice income, a is the share of rice income in total income, Cy is the CV of nonrice income, and g is the correlation coefficient between the rice and nonrice income. The proportional risk premium measured in equation (3) provides an estimate of the cost of risk currently borne by farmers relative to the situation in which the variability of rice income is completely eliminated. As there will always be some variability of rice income that cannot be eliminated, the estimate obtained from equation (3) can be considered to be an upper bound value. In addition, the determinants of adoption of modern varieties were identified using the probit model (Greene 1997). The factors determining the adoption of modern varieties on individual plots were identified using plot-level data. Regressors used for this analysis were a set of household-specific factors and a set of plot-specific factors. The probit coefficients were used to calculate the marginal probabilities of adoption associated with various regressor variables. The use of the probit model permitted the simultaneous consideration of the effects of biophysical and socioeconomic factors in the adoption of modern varieties1.

Results Table 2 shows summary statistics regarding rainfall from a location near the study villages. These data indicate that rainfall variability is high at both the initial stage of land preparation/rice establishment (June/July) and during the later stage of crop growth (October). Compared with 1994 and 1997, rainfall in June/July was much lower in 1995 and 1996. Of the two drought years, 1995 and 1996, the early season drought in 1995 was more severe than in 1996. Based on the distribution of and the amount of rainfall, 1994 may be classified as a “normal” year, 1995 and 1996 as “drought” years, with drought being more severe in 1995, and 1997 and 1998 as “good” years. The surveyed years thus provide a range of rainfall scenarios for investigating farmers’ responses.

1As panel data are used to estimate the probit model, a brief digression on the econometric issues involved is in order. Such data permit the specification of farmer-specific effects, which can be treated as fixed or random effects (Greene 1997). A simple way of incorporating fixed effects is through the use of farmer-specific dummy variables. However, with a large number of cross-sectional units (60 in the present data set), the estimates of individual fixed effects are likely to be somewhat intractable (Greene 1997, p. 897). Estimation of the random effects model is a possibility but at the added cost of computational burden. Instead of these more sophisti- cated methods, we have treated the panel data as a single cross section by ignoring the farmer-specific effects. While such a simplification is likely to lead to parameter estimates that are statistically less efficient, any bias on the estimated slope coefficients is likely to be minimal due to the low correlation of the independent variables used with the excluded farmer-specific dummy variables. Nevertheless, this caveat should be kept in mind while interpreting the coefficients of the probit model estimated.

326 Singh et al Table 2. Kharif-season rainfall for Kumarganj, Faizabad, in 1994-98.

June July August September October Total

Year No. of Rain No. of Rain No. of Rain No. of Rain No. of Rain No. of Rain rainy (mm) rainy (mm) rainy (mm) rainy (mm) rainy (mm) rainy (mm) days days days days days days

1994 8 145 15 224 16 341 6 153 1 2 46 865 1995 8 109 7 60 23 355 10 395 0 0 48 919 1996 11 99 5 133 18 337 14 168 3 165 51 902 1997 4 121 19 485 15 264 12 294 2 43 52 1,207 1998 4 29 19 337 16 276 6 110 2 47 47 799

CRS, Masodha, Faizabad (1967-98) Long-term 145 314 300 141 74 1,088 mean rainfall (mm) Coefficient 87 45 40 56 115 22 of variation of long-term rainfall (%)

Table 3. Measures of probability distribution of plot-level rice yield, net returns, and gross returns in rice (1994-98).a

Item Mungeshpur Itgaon

Yield Mean (t ha–1) 2.2 1.7 CV (%) 52 60 Skewness 2.88 0.79 Gross returns Mean (US$ ha–1) 210 140 CV (%) 62 63 Skewness 4.67 0.94 Net returns Mean (US$ ha–1) 177 112 CV (%) 73 84 Skewness 3.72 0.77

aAll monetary values are at 1994 constant prices. The coefficient of variation (CV) of yield was calculated using the fixed-effects model specified in equation (1). Values in local currency were converted using US$1 = Rs 31.38.

Variability of rice yield and net returns The average rice yield in Mungeshpur is about 0.5 t ha–1 higher than in Itgaon (Table 3). The variability of yield, as measured by the coefficient of variation, however, is lower in Mungeshpur. A higher mean and a lower CV are indications of a more favorable environmental condition for rice production in Mungeshpur. For both Mungeshpur

Rainfed rice, risk, and technology adoption: . . . 327 and Itgaon, however, the plot-level CVs of yield are quite high by conventional stan- dards. The CVs of net returns (defined as gross returns minus cash costs) are even higher and are indicative of the high levels of variability that farmers in rainfed environments have to deal with. The large difference between the CV of yield and net returns also indicates that yield variability can severely underestimate risk if the farmers’ objective is to obtain stable net returns. Analysis based on yield variability alone can hence be misleading.

Crop diversification and risk The diversification indices (Table 4) indicate that the cropping pattern in Itgaon is more diversified than in Mungeshpur for both the kharif (rainy) and rabi (postrainy) seasons. In Itgaon, farmers grow rice in proportionately smaller areas than in Mungeshpur and rely more on pulses and maize, which grow well under rainfed conditions. Mixed cropping and intercropping are also more common in Itgaon. Of the five years, the kharif season diversification indices are higher during the drought years (1995 and 1996). These results indicate that crop diversification may be an important risk-reducing strategy, especially to farmers of Itgaon. The average CV of income from all rainy-season crops is substantially lower than the CV of rice income (Table 5). Using a multiple regression analysis of the CV of rainy-season crop income on farm size and diversification index, Pandey et al (1999) show that more diversified farms in Itgaon do have a lower variability of crop income. The stabilization effect is

Table 4. Crop diversification index.

Mungeshpur Itgaon Year Kharif Rabi Kharif Rabi

1994 0.47 0.56 0.72 0.71 1995 0.53 0.55 0.90 0.77 1996 0.58 0.57 0.82 0.67 1997 0.50 0.53 0.75 0.68 1998 0.45 0.53 0.78 0.62

Table 5. Coefficient of variation (CV) of income from different sources, 1994-98.a

Item Mungeshpur Itgaon

CV of rice income (%) 30 76 CV of all rainy-season crop income (%) 28 23 CV of all crop income (%) 10 9 CV of all income 9 12

aThese coefficients of variation are the average values (across households) of the coefficient of variation for each household calculated using temporal data for that household.

328 Singh et al even stronger if the CV of income from all crops (both the kharif and rabi seasons) is considered (Table 5). The variability of plot-level net returns from rice reported in Table 3 was found to be higher than the variability of total net returns from all plots of rice cultivated by a household (Table 5). Low correlation among plot-level net returns may have led to a reduction in CV of total net returns when rice is grown in several fields with varying soil and hydrological characteristics. Farmers’ land portfolios that include these varying field characteristics may thus help stabilize income even though there are likely to be some efficiency losses in managing different field types that are often spatially scattered.

Importance of rice in farmers’ income Table 6 shows the relative share of income from rice (i.e., the value of rice produced minus the cash costs) as a proportion of the total household income. It is interesting to observe that rice accounts for only a small proportion of total household income, although the share of rice increases as farm size increases in Mungeshpur. Farmers’ incomes are diversified. Rabi crops and nonfarm income contribute 30% and 46%,

Table 6. Percentage share of different sources of income in total household income, 1994-98.

Item Mungeshpur Itgaon Both villages

Small farms Rice 13 6 11 Rabi cropsa 23 19 22 Nonfarm 50 65 54 Othersb 14 9 13 Medium farms Rice 17 3 10 Rabi cropsa 37 27 32 Nonfarm 28 63 46 Othersb 19 7 12 Large farms Rice 22 6 8 Rabi cropsa 37 35 35 Nonfarm 13 43 39 Othersb 28 16 18 All farms Rice 15 5 9 Rabi cropsa 30 30 30 Nonfarm 37 52 46 Othersb 17 13 15

aRabi crops are mostly wheat intercropped with mustard. bIncludes income from other kharif and summer crops, income from livestock, and income as a farm laborer.

Rainfed rice, risk, and technology adoption: . . . 329 respectively, to farmers’ income. Land-use intensity is almost 100% during the rabi season in both villages with farmers growing several cash crops such as wheat, sugarcane, vegetables, and pulses. The availability of tubewell irrigation has facilitated rabi-season cropping in this part of eastern India. Proximity to the town of Faizabad and the city of Lucknow has also led to an expansion of nonfarm income opportunities in these villages. These villages thus do not agree with the stereotype of an eastern Indian village where rice production is the major economic activity. A diversified income portfolio that is less dependent on rice production has evolved over time as a result of commercialization of production systems. Naturally, variability in rice production will have a small effect on the variability of total household income of farmers in such systems, even though farmers continue to grow rice for their food security.

Seasonal adjustments In this section, we analyze the responses over time in terms of rice area planted, method of crop establishment, and changes in rice varieties (Table 7). In both Mungeshpur and Itgaon, rice area decreased in 1995 and 1996 in comparison with 1994 and 1997, with the reduction being greater in Itgaon. Delayed rainfall in 1995 forced many farmers in Itgaon to forego rice completely, whereas farmers in Mungeshpur were able to maintain the rice area by using irrigation. Large fluctuations in area planted to rice in rainfed environments can be a major source of variability in rice production unless there are offsetting movements in yields. A variance decomposition analysis indicated that in Itgaon the variability in area sown accounted for about half the total production variation (Pandey et al 1998). Temporal variations in area under each of the crop establishment methods are higher in Itgaon than in Mungeshpur. Similarly, the proportionate area planted to modern varieties also shows some temporal variations. Open-ended interviews with the farmers indicated that they adjust area under modern varieties and crop establishment methods from year to year to cope with climatic risk. When rains were insufficient or delayed at the time of land preparation and crop establishment, transplanting was not possible in fields without supplemental irrigation due to inability to grow seedlings and prepare land. In this situation, farmers preferred to establish rice by dry seeding even though the anticipated yields were lower. Similarly, the use of traditional varieties was more common in such years. These temporal adjustments are manifestations of flexible decision-making through which farmers attempt to reduce production and income losses in adverse years. Table 7. Tactical adjustments to weather uncertainty.

Mungeshpur Itgaon Item 1994 1995 1996 1997 1998 1994 1995 1996 1997 1998

% area transplanted 78 76 86 88 96 27 39 50 54 68 % area under modern varieties 89 93 94 91 97 67 64 54 61 68 % area under rice 60 53 54 60 63 33 10 23 36 28 % area fallowed 16 20 14 15 14 33 48 33 26 28

330 Singh et al Table 8. Cost of instability in rice income.

Item Mungeshpur Itgaon

CVa of rice income (%) 30 76 CV of nonrice income (%) 15 15 Ratio of rice to total income (%) 15 5 Average cost of risk (% of mean income) if correlation between rice and nonrice income = 0 0.3 0.1 if correlation between rice and nonrice income = 0.2 0.4 0.3

aCV = coefficient of variation.

Cost of risk The cost of risk was estimated using equation (3) by assuming the value of R to be equal to 2 (Anderson 1995) and using the sample estimates of the CV of income from rice and nonrice activities. The correlation coefficient between rice and nonrice income in these villages that have diversified income sources was assumed to be 0.2. The estimated cost of risk is less than 1% of the mean income (Table 8). In other words, risk benefits from the stabilization of income from rice are negligible. The main reason for the relative unimportance of variability in rice income is the low share of rice income in total income. Farmers have already diversified their income away from rice. Thus, stabilization of rice yield and rice income will have little impact on the stabilization of total income. Using a simulation model, Pandey et al (1999) report that even a full stabilization of rice income will lead to a reduction in the CV of total income by only 2% in Itgaon. Naturally, in areas such as the study villages, crop production technologies that raise mean income are likely to be more beneficial than those that reduce the variability of returns only. In areas where the share of rice income in total income is high, however, the benefits from stabilization of rice yield are likely to be higher (Fig. 1).

Proportional risk premium (%) 20 15 10 5 0 0 5 10 15 20 25 30 35 40 45 50 Share of rice in total income (%) Fig. 1. Relationship between risk premium and share of rice income. Drawn using the assumptions in Table 8 and varying the share of rice in total income.

Rainfed rice, risk, and technology adoption: . . . 331 Factors determining the adoption of modern varieties The extent of adoption of modern varieties in Mungeshpur is quite high (more than 90% of area). Although the adoption rate is lower in Itgaon (60% of area), it is nevertheless substantial. The major modern varieties grown are Sarjoo-52 and NDR varieties (Table 9). Sarjoo-52 was released in 1980 and is a 130-d variety that performs well when supplemental irrigation is available. Its use has spread, however, even to areas under rainfed conditions and it has replaced Mahsuri, which is of slightly longer duration. Over time, Sarjoo-52 has become quite popular and this variety alone now covers almost 40% of the rice area in Itgaon. Similarly, another popular modern variety, NDR-97, was released in 1991. Its duration is 95 d and it is grown mostly under upland fields that are proportionately more in Itgaon. Sarjoo-52 is not only higher yielding but also less risky than other modern varieties and some traditional varieties. Using a stochastic dominance analysis, Pandey and Pal (2000) found that Sarjoo-52 outyielded popular traditional varieties in all of the sampled years. The cost/returns analysis shown in Table 10 also indicates that, overall, modern varieties currently being grown not only have higher mean net returns (defined as the value of output minus the cash cost) but also lower coefficients of variation relative to traditional varieties. The usual notion that modern varieties are more risky than the traditional ones is not supported by the data from these study villages. The question still remains as to why the adoption rate remains lower in Itgaon than in Mungeshpur, and among small farmers compared with large ones. Obviously, factors other than risk must constrain the adoption of these seemingly less-risky modern varieties.

Table 9. Percentage area under major modern varieties (MV) at the study sites (1994-98).a

Variety Mungeshpur Itgaon

Sarjoo-52 35 40 Mahsuri 20 3 NDR-118 7 18 Saket-4 6 6 Indrasan 6 2 NDR-80 5 7 Pant-4 4 2 NDR-40032 2 NDR-359 2 Ashwati 2 Pant-10 2 3 NDR-97 2 16 Other NDRs 2 Other MVs 6 2

All MVs 100 100

aNumbers have been rounded to nearest whole number.

332 Singh et al Table 10. Cost and return analysis of rice by variety, 1994-98.a

Modern Traditional Item Mean CV (%) Mean CV (%)

Yield (t ha–1) 2.7 14 1.5 34 Cost Material costs ($ ha–1)30531860 Labor costs ($ ha–1) 15591056 Total cash cost ($ ha–1)45512758 Gross returns ($ ha–1) 264 6 132 18 Net returns ($ ha–1) 219 22 105 34

aAll values are at constant 1994 prices. All local currency is converted using the 1994 exchange rate of US$1 = Rs 31.38. CV = coefficient of variation.

Table 11. Probit model estimates of modern variety adoption in Faizabad, eastern India (1994-98).

Standard Marginal Variables Coefficients error probabilities (%)a

Farm size (ha) 0.178* 0.064 3.53 Age of household head (years) –0.019* 0.004 –0.38 Education of household head 0.038* 0.014 0.75 (years) Nonfarm income (000 Rs) 0.001 0.005 0.02 Dummy for uplandb 0.212 0.114 4.20 Dummy for irrigationc 0.521* 0.138 10.35 Dummy for saline/sodic soilsd –0.042 0.229 –0.84 Time trende 0.073* 0.003 1.46 Constant 0.745 0.301 Sample size 1,057 Log-likelihood ratio 109.00** Percent of correct predictions 80

*and ** indicate significance at 5% and 1% levels. aMarginal probability evaluated at mean values of continuous independent variables with all dummy variables set equal to unity. b1 if upland and 0 otherwise. c1 if irrigated and 0 otherwise. d1 if soil is saline/sodic and 0 otherwise. e1 for 1994, 2 for 1995, 3 for 1996, 4 for 1997, and 5 for 1998.

Several biophysical and socioeconomic factors may condition the adoption of modern varieties. A major set of biophysical factors is related to soil fertility and field hydrology. Both these factors are controlled to a certain extent by the location of the field in the toposequence and the soil type. Table 11 presents the results of the probit model estimated using plot-level data. The results indicate that farm size, age and education of the farmer, and availability of irrigation are the major variables deter-

Rainfed rice, risk, and technology adoption: . . . 333 mining whether or not modern varieties will be grown in a particular field. The effect of farm size is positive, indicating that, for a given set of conditions, large farmers are more likely to adopt modern varieties. As discussed by Feder et al (1985), farm size may be a proxy for several factors such as wealth status, extent of risk aversion, financial capability to acquire complementary inputs such as fertilizers, and access to information and technology. Operators of large farms are better placed in all these aspects; hence, a greater adoption of modern varieties. An increase in farm size by 1 ha raises the probability of adoption of modern varieties by approximately 4%. Al- though land reform to redistribute land may be justified on various grounds, the adoption of modern varieties could decline if land redistribution is not accompanied by improved access to technology and credit by small farmers. Farmer age is negatively correlated with adoption, indicating that younger farmers may be more enterprising and more willing to experiment with modern varieties than older farmers. House- holds with more years of education are more likely to adopt modern varieties. Despite the statistical significance of these two farmer attributes, the associated marginal probabilities are quite small, indicating the limited relevance of these variables for technology targeting. The probability of adoption is 10% higher in irrigated fields than in rainfed fields, other things remaining the same. The positive and significant coefficient of time trend indicates that adoption has increased over time. The coefficient of the dummy variable for uplands is statistically significant at the 10% level (but not at the 5% level), indicating some weak preference of farmers for growing modern varieties in upland field types. Soil type did not have a statistically significant effect. In terms of the size of the marginal effect on the adoption of modern varieties, irrigation has by far the most dominant effect.

Discussion The results of the study show that the variability of plot-level yield and net returns of rice in the study villages is high even in Mungeshpur, where rice is now grown mostly under irrigated conditions. In Itgaon, where irrigation is more limited, the variability of rice yield and area over time is quite high. Although agricultural researchers have focused their energy on addressing the problem of yield variability, area variability in rainfed environments can often be an important source of production variability. Tech- nologies that help stabilize rice area planted can help reduce variability in rice production. One of the major causes of area variability is failure in timely crop establishment. Varieties that can be established late and crop management technologies that facilitate rapid establishment when environmental conditions are most appropriate are needed to reduce the effect of area variability. One of the major ex ante strategies to deal with risk in rice production is crop diversification. The coefficient of variation of income from crops grown during the rainy season was found to be inversely related to the extent of crop diversification in Itgaon. In addition, crop diversification increased during years of low and/or late rains at the beginning of the cropping season. Although further expansion of tubewell irrigation is likely to reduce the importance of crop diversification as a risk-reducing

334 Singh et al strategy in the future (Ballabh and Pandey 1999), rice technologies that facilitate crop diversification will complement farmers’ coping strategies. An example of such technologies may be improvements in direct-seeding methods to reduce the high labor demand associated with transplanting. This will help relax labor constraints that may limit crop diversification. Another important risk-coping mechanism of farmers is the maintenance of flexibility in their decision-making processes. The results show that farmers adjust crop establishment methods and varieties over time depending on the amount of rain received at the beginning of the cropping season. Improved varieties that perform equally well under direct-seeding or transplanting methods will contribute to flexibility. Similarly, an accurate forecast of the early season rainfall pattern and dissemination of forecasts to farmers will help adapt rice production methods to match rainfall. High intensity of land use for production of cash crops during the rabi season is an important feature of the study villages. A substantial proportion of household income is generated from the rabi crops. In addition, the proximity of study villages to urban centers has expanded the nonfarm employment opportunities available to farming families in the study villages. As a result, the share of rice income in the total household income in these villages is below 10%. Even though the rice income is highly unstable, farmers have been able to stabilize total household income through diversification of income away from rice. This strategy seems to have been quite effective in stabilizing total household income, even in Itgaon, which has a higher proportion of rainfed area. Policies such as the development of infrastructure and rural industrialization that facilitate income diversification can thus play an important role in reducing the effect of risk in rice production. A low proportion of rice in total household income also means that risk in rice production per se is relatively unimportant in these villages. The estimated cost of risk was found to be below 1% of mean income. This implies that breeding programs that sacrifice some yield gain in the pursuit of yield stability are likely to be less attractive than those that aim to improve the average yield. Similarly, farmers’ demand for crop insurance to stabilize income from rice is likely to remain low. A cau- tionary note is in order here. These conclusions may not be applicable to other parts of eastern India where the share of rice in total income is higher due to the lack of income diversification opportunities. The results of the study also show that the modern varieties currently being grown are not necessarily more risky than the traditional varieties. The conventional wisdom that modern varieties increase risk is not supported by the data. The spread of tubewell irrigation in the study villages contributed to risk reduction by creating more favorable growing conditions for modern varieties. However, the fact that traditional varieties rarely had higher yields than modern varieties over the study period is an indication of a lower risk associated with these more recent modern varieties. Despite the apparent effectiveness of farmers’ coping mechanisms in reducing risk associated with rice production, the spread of modern varieties is still constrained by the unavailability of irrigation and limited farm size. Although access to irrigation has improved in more recent years because of increasing investments in tubewell

Rainfed rice, risk, and technology adoption: . . . 335 irrigation schemes in the study area, further investments are needed to make groundwater more widely available. Access to groundwater among small farmers can also be improved by encouraging the development of a more effective groundwater market. The development of varieties more suitable to lowland fields can similarly improve the chance of adoption of modern varieties in these field types.

Concluding remarks Farmers’ risk-coping mechanisms were reasonably efficient in preventing income shortfall in the study region. Crop and income diversification away from rice has been the major strategy of farmers for dealing with risk. Because of the low importance of rice in total household income, risk related to rice production is relatively unimportant in constraining technology adoption. Accordingly, the trade-off that may exist between yield gain and stability needs to be carefully considered in designing breeding programs as farmers in the study villages have been able to reduce risk through income diversification. Although methods and tools for microeconomic analysis of risk are generally available, a lack of temporal farm-level data covering enough periods remains a problem for the analyst. While an ingenious use of cross-sectional data and weather-driven crop growth models can help in this, temporal data such as the ones collected in this study are essential for conducting a more complete analysis at the farm-household level. Collection of such data would facilitate in-depth analysis of risk and farmers’ coping mechanisms.

References Anderson JR. 1995. Confronting uncertainty in rainfed rice farming: research challenges. In: Fragile lives in fragile ecosystems. Proceedings of the International Rice Research Con- ference. Los Baños (Philippines): International Rice Research Institute. p 101-108. Ballabh V, Pandey S. 1999. Transitions in rice production systems in eastern India: evidences from two villages in Uttar Pradesh. Econ. Polit. Wkly. March 27, 1999. p A11-A16. Feder G, Just RE, Zilberman D. 1985. Adoption of agricultural innovations in developing countries: a survey. Econ. Dev. Cult. Change 33:255-298. Greene WH. 1997. Econometric analysis. 3rd edition. Princeton, N.J. (USA): Prentice Hall. Jodha NS. 1978. Effectiveness of farmers’ adjustment to risk. Econ. Polit. Wkly. 13(25): A38- A48. Pandey S, Pal S. 2000. The nature and causes of changes in variability of rice production in eastern India: a district-level analysis. In: Pandey S, Barah BC, Villano RA, Pal S, editors. Risk analysis and management in rainfed rice systems. Limited Proceedings of the NCAP/IRRI Workshop on Risk Analysis and Management in Rainfed Rice Systems, 21-23 September 1998, National Centre for Agricultural Economics and Policy Research, New Delhi, India. Los Baños (Philippines): International Rice Research Institute. p 73- 96. Pandey S, Behura DD, Villano RA, Naik D. 2000. Economic cost of drought and farmers’ coping mechanisms: a study of rainfed rice systems in eastern India. Discussion Paper Series No. 39. Los Baños (Philippines): International Rice Research Institute. 35 p.

336 Singh et al Pandey S, Singh HN, Villano RA. 1999. Rainfed rice and risk coping strategies: some microeconomic evidences from eastern India. Selected paper for presentation at the Annual Meeting of the American Agricultural Economics Association, 8-11 August 1999, Nash- ville, Tennessee, USA. Pandey S, Singh HN, Villano RA. 1998. Rainfed rice and risk coping strategies: some microeconomic evidences from eastern Uttar Pradesh. Paper presented at the NCAP/IRRI Workshop on Risk Analysis and Management in Rainfed Rice Systems, 21-23 Septem- ber 1998, New Delhi, India. Siddiq E, Kundu DK. 1993. Production strategies for rice-based cropping systems in the humid tropics. In: Buxton DR et al, editors. International Crop Science 1. Madison, Wis. (USA): Crop Science Society of America. Singh HN, Singh JN, Singh RK. 1995. Risk management by rainfed lowland rice farmers in eastern India. In: Fragile lives in fragile environments. Proceedings of the International Rice Research Conference. Los Baños (Philippines): International Rice Research Insti- tute. p 135-148.

Notes Authors’ addresses: H.N. Singh, Narendra Dev University of Agriculture and Technology, Faizabad, India; S. Pandey, R.A. Villano, Social Sciences Division, International Rice Research Institute, DAPO Box 7777, Metro Manila, Philippines. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Rainfed rice, risk, and technology adoption: . . . 337 Using gender analysis in characterizing and understanding farm-household systems in rainfed lowland rice environments

T. Paris, A. Singh, M. Hossain, and J. Luis

This chapter uses gender analysis in characterizing and understanding farm- household systems in typical rainfed lowland rice villages in Faizabad district in eastern Uttar Pradesh, India. It uses different methods of data collection such as household surveys (structured and unstructured interviews), the participatory rural appraisal, and focused group interviews. Results of the study reveal that small farming households from the lower caste tend to exploit their female family members to meet competing labor demands between farm and home-based activities. Women from the lower caste provide 60% to 80% of the total labor input in rice production. They participate in almost all rice operations, except in land preparation and application of chemicals. When valued, the labor contributions of female members on their own farms and through exchange arrangements make up about 20% of the total labor costs in rice production per hectare. Women’s labor is also crucial to nonrice crops and livestock, which are integral in rainfed rice farming systems. Because of the significant contributions of poor women in farming, their roles and needs should be considered in technology development and dissemination. Efforts are now being made to provide women farmers with access to new information and new seeds by involving them in the early evaluation of rice genotypes through participatory rice varietal selection in drought and submergence rice environments in eastern India.

In recent decades, greater attention has been given to rice research on rainfed lowlands, which cover 48 million hectares in the humid and subhumid tropics of South and Southeast Asia. Farmers in these ecosystems face adverse biophysical, socioeconomic, and cultural constraints to increasing rice productivity. Because of the seasonally variable and erratic rainfall pattern, heterogeneous land types, and diverse socioeconomic groups with limited resources, average rice yields are approximately 2.3 t ha–1 and are as low as 1.3 t ha–1 (IRRI 1997). Risk under rainfed conditions usually tends to be high as variability in rainfall can lead to wide swings in yield and output (Pandey et al 1998). Because of this uncertainty, farmers do not have the in-

Using gender analysis in characterizing and understanding . . . 339 centive to invest in cash inputs and devote more time to crop management. Poor management has led to low yields, no or low marketable surplus, and consequently low income from rice. Farmers are thus caught in a vicious circle of poverty. Despite the riskiness in rice production, farmers grow rice as the major crop to sustain household food security. Their primary concern is to meet their basic food requirements and fodder needs for their livestock through their own production. To reduce risk, they diversify their income sources, one of which is for male members to migrate to other cities or other highly productive agricultural areas. This requires allocation of family labor to various livelihood activities according to the gender roles prescribed by the household and community, degrees of labor specialization by family members, and opportunity costs of family labor. While there has been greater awareness and recognition of the vital roles that poor women play in rice-based farming systems, their unpaid labor contribution in rice farming is seldom valued. This has often led to their exclusion as cooperators in on-farm research and as recipients of training and extension programs. With the increasing male migration (seasonal or semipermanent), other family members, particularly the female members, are left behind to manage the farm (crops and livestock) aside from their daily household and child-care responsibilities. The changing gender roles and responsibilities will have far-reaching implications not only for crop production but also for the social organization aspects of the rice household economy. It is important for biological and social scientists to understand the emerging changes that will shape the nature of rice production systems and their implications for poor households and gender relations. This important understanding will help prioritize research issues for technology and policy interventions that can improve the well- being of members of farming households, especially the women. Thus, an analysis of gender roles and relations should be integral when characterizing target recommendation domains for technology development. This chapter discusses the objectives of gender analysis in on-farm research, and provides a conceptual framework for the livelihood systems. It also presents a case in the rainfed lowland rice ecosystems in eastern Uttar Pradesh, India, which demonstrates how gender analysis can be used in characterizing and understanding the farm household and its environment.

Gender analysis and its objectives “Gender” refers to a social rather than biological construct, whereas “sex” refers to the biological differences between men and women. Gender describes the socially determined attributes of men and women, including male and female roles. As a social construct, gender roles are based on learned behavior as a response to socioeconomic and environmental pressures and conditions and are flexible and variable across and within cultures. Gender is a useful socioeconomic variable to analyze roles, responsibilities, constraints, opportunities, and incentives of the people involved in research and development efforts (Poats 1990). It is relational in focus, that is, it is concerned with women and men in relation to each other. Gender analysis is an ana-

340 Paris et al lytical tool used to identify or distinguish the actual “doers” of tasks, decision makers and the potential users of technologies. A primary social mechanism by which men and women are bound into a relation of interdependence is the gender division of labor (GDOL) in different tasks in farm activities. This will answer the questions ● Who does specific crop and livestock operations? ● Who influences a particular activity to be improved, changed, or eliminated? ● Whose resources (e.g., labor) will be displaced/increased with the proposed change? ● Who has the incentive to accept the technology or will directly benefit from it? Knowing the user and beneficiary has both equity and efficiency implications. This increases the efficiency of farming systems research through targeting and specification, and takes into account patterns and activities of resource use (Feldstein et al 1989). An improved understanding of gender roles means that women who were often overlooked will be recognized while their needs, constraints, and productive opportunities can be addressed by agricultural research and extension. Gender analysis enables scientists to target women’s needs better and predict the impact of planned interventions on women.

The conceptual framework for farm-household and livelihood systems Through gender analysis, women’s roles are not seen in isolation; rather, their roles and responsibilities in relation to men are analyzed within the context of the complex interaction between the farm household and the environment (biophysical, social, economic, and cultural). The complexity of rainfed lowland rice production systems can be best understood by using a holistic conceptual framework for the livelihood system (adapted from Norman et al 1983). A livelihood system includes the farming system and off-farm and on-farm linkages. The farming system consists of a complex interaction of several independent components (Fig. 1). These components can be divided into two elements: technical and human. The technical element determines the type and physical potential of crop and livestock enterprises, and includes the physical and biological factors that can be modified through technology development. For example, the crop calendar could be adjusted to avoid drought or flood by growing early maturing varieties or by changing the crop establishment method from transplanting to direct seeding. Consequently, these changes will have a major effect on gender roles in rice farming. The human element is characterized by exogenous (community structure, ex- ternal institutions, etc.) and endogenous factors that can be controlled by the households. At the center of this interaction is the farm household, which is often treated as a “black box” or homogeneous unit represented by the male head of the household with members having shared and equal access to resources and benefits from production. This assumption ignores the differing roles and sometimes conflicting interests in resource use within a household that affect technology adoption or that in turn can negatively affect a specific user group (Feldstein et al 1989). There is often inequity

Using gender analysis in characterizing and understanding . . . 341 Socioeconomic and Physical and biological conditions cultural circumstances

● Community structures, Physical Biological Soil and topography norms, beliefs – Climate – Pests ● Kinship (nuclear – Rainfall – Diseases Soil type Slope extended) – Occurrence – Weeds ● Caste system of drought ● Land tenure or flood ● Infrastructure – Land type ● Markets ● Extension services ● Wages ● Prices

Male and female household members

Resource allocation

Land Water Capital Labor Human capital – Education – Technical knowledge

Livelihood system

Cropping Livestock Off-farm Nonfarm system system activities activities

Goal

Fig. 1. Conceptual framework for understanding the interrelationship between the environment and farm-household and gender relations in rainfed lowland rice-based farming systems. in access to and control of resources between male and female members within the household and their communities as a result of norms and traditions that dictate the appropriate behavior and activities of men and women. Studies show that there is an unequal allocation of resources within the households, and that adult female members and daughters are the most disadvantaged, especially in South Asia (Agarwal 1998). It is also assumed that household labor is homogeneous and thus is freely sub- stitutable across all household tasks, from household to off-farm wage employment.

342 Paris et al However, household members have certain degrees of specialization and time allocation. Members receiving the highest wage offers and employment opportunities will naturally specialize in market work. Since men often receive better education and training, they develop more skills, receive higher wages, and have greater access to opportunities in the labor market than women. Thus, women tend to do more unpaid and home-based activities, which are often undercounted and undervalued. If they are hired as seasonal agricultural laborers, they receive lower wages than men. Thus, in a particular farming system, even where the household is a unit of analysis, the patterns of activities, resources, and participation in decision making are important information and must be determined by investigation, not assumption (Feldstein et al 1989). The purpose of doing on-farm research is to generate more appropriate technologies under farmers’ conditions, to raise the welfare of farm families, and to enhance society’s goals. In conducting on-farm research, it is necessary to understand the operation of the small farm within the wider context (village, district, regional, and global). Solving farmers’ problems requires an interdisciplinary team of scientists and active participation of the farmers, including women. The social scientist plays an important role at the beginning of the research process, in characterizing the socioeconomic and cultural components of the livelihood systems, in identifying constraints and opportunities that are consistent with the needs of the clients and users (men and women) of technologies, and in analyzing the ex ante and ex post impact of technologies.

Gender analysis in on-farm research On-farm research with a systems perspective is an iterative, overlapping, and dynamic process. However, any on-farm research should follow a process and can be summarized into these main activities: ● Characterizing/describing the local farming systems and practices and diag- nosing farmers’ problems and constraints to productivity, ● Selecting and improving existing technologies and techniques to overcome these constraints, and ● Testing and adapting gender-responsive technologies with farmer participation. Using the same framework, gender analysis and women’s concerns can be integrated from the beginning of the research process and not only during the impact analysis. The research can be divided into two phases: ● Identifying gender roles in the household, in on-farm and nonfarm activities, and in farming practices; quantifying the labor contributions of men and women in major farm activities; assessing gender differences in access to and control of resources; identifying constraints, potential, and needs; and identifying options for improvement. ● Designing, testing, and adapting the best available technologies to increase women’s productivity and income and to reduce work burden.

Using gender analysis in characterizing and understanding . . . 343 Women and not men only should be included as cooperators in on-farm research. Their perceptions, needs, knowledge, and skills should be considered in the design, testing, and evaluation of proposed technologies. Opportunities or options that can enhance women’s roles in terms of increasing productivity of labor and land, providing income opportunities, reducing their work burden, or enhancing their technical knowledge or skills should be addressed in any research project on rice-based farming systems dealing with food security and poverty alleviation.

Application of gender analysis in rainfed lowland rice ecosystems: a case study in eastern Uttar Pradesh, India Selection of research sites and respondents This case study was conducted in major rice-growing villages (Chandpur, Mungeshpur, and Sariyawan) in Faizabad District in eastern Uttar Pradesh. These villages are sites of the Rainfed Lowland Rice Research Consortium coordinated by IRRI and the In- ternational Fund for Agricultural Development (IFAD)-funded IRRI-ICAR (Indian Council for Agricultural Research) Collaborative Project on Rainfed Rice in Eastern India. This study is being conducted in collaboration with the Narendra Deva Univer- sity and Agricultural Technology (NDUAT) in Kumarganj, Faizabad District. These villages differ in proximity to the major market (Faizabad City), agroecology, access to supplementary irrigation, and amenities (Table 1). Chandpur is near Faizabad City, whereas Mungesphur and Sariyawan (adjacent villages) are far from it. Farmers have heterogeneous land types—lowland, medium land, and uplands. More than half of the total cultivated area in Chandpur and Mungeshpur is lowland and medium land. In contrast, half of the total cultivated land in Sariyawan is under uplands. The population to land ratio is higher in Chandpur than in the other two villages. Farming households in Mungeshpur and Sariyawan have greater access to supplementary irrigation for rice and nonrice crops such as wheat, potatoes, and vegetables than in Chandpur. A socioeconomic household survey was conducted in 1995-96 based on total enumeration of the population. This survey covered 431 farming and landless households. Different data collection methods were used starting with informal interviews and a participatory rural appraisal (PRA) followed by household surveys and focused group interviews. Within a household, the principal male or female was interviewed depending upon the questions being pursued. However, for the labor data in rice production, the principal male and female household members who were actively involved in rice farming were interviewed together. The following section describes the environmental (social, cultural, agroclimatic, physical) characteristics for which gender roles and gender relations were analyzed.

The social and cultural environment that affects gender relations Any attempt to assess the problems of Indian women in agriculture and their constraints to improving the performance of farming systems is incomplete without a look at the social structure, cultural norms, and value systems that define and deter-

344 Paris et al Table 1. Village characteristics of the research sites, 1995.

Descriptors Chandpur Mungeshpur Sariyawan

Distance to Faizabad City (km) 3 km (near) 28 km (far) 28 km (far) Agroecology Shallow lowland and Shallow lowland Upland and submergence-prone and drought-prone drought-prone Total cultivated area (ha) 54 62 70 Land types (%) Upland 27 34 50 Medium land 15 14 17 Lowland 50 52 33 Total number of households 200 150 120 Total population 1,244 802 700 Male 643 429 364 Female 601 373 336 Male/female ratio 107/100 115/100 108/100 Population/land (persons ha–1)23 13 10 No. of primary schools 1 1 1 No. of preparatory schools 1 – 0 No. of tube wells 7 15 12 No. of shops 20 10 5 mine the roles of men and women (Mukhopadhyay 1984). Patriarchal ideology, dowry during marriage, caste structure, and combined families are the social and cultural practices that influence gender roles and gender relations, which in turn affect the intrahousehold resource allocation. Patriarchal ideology. The predominant force in the social organization of In- dian society is patriarchy. Until 40 years ago, the legal status of Hindu women in India was based on laws dictated by the ancient Hindu lawgiver, Manu (first and second centuries AD). Briefly, the essence of Manu’s thesis was that women are supposed to continuously remain under some male authority, first under that of the father, then of the husband, and finally of the son. He prohibited widow remarriage, insti- tuted childhood marriage for girls, established the concept of dowry, and disinherited women from their husband’s and father’s property. Though the Hindu Civil code of 1956 has tried to eliminate many of these disabilities, the women’s movement in India still has a long way to go to rid the society of its oppressive customs (Ghosh 1987). Land ownership. In India, the acquisition, ownership, and transfer of property are through the male members of the family. Women have little access to ownership of land or other productive assets because of Hindu Inheritance Law, which entails patrilineal transmission of property. Although the right to inherit property in postindependence India had been assured to female members by law, the law itself is a compromise with the traditional position, which does not recognize a female’s right to ownership of property. Furthermore, the socialization of girls within the partrilineal form of social organization ensures that women will not be in a position to claim their legal rights. A woman’s lack of education, lack of legal knowledge, and dependence

Using gender analysis in characterizing and understanding . . . 345 on the men in the family will prevent her from claiming them (Mukhopadhyay 1984). The bias against women in terms of land inheritance is also true in the Philippines and other countries that are patrilineal societies. Quisumbing’s (1995) study in the Philip- pines revealed that better-educated fathers also favor daughters in terms of education, whereas mothers with more land tend to favor sons. Without a title to land as collateral, women have also been excluded from institutional credit and are thus unable to secure tools and capital for self-employment except through the more costly informal credit system. The loss of land inheritance rights for widows leads to destitution and dependency on other people, especially the in-laws (Agarwal 1998). Dowry during marriage. In India, giving material wealth (referred to as dowry) along with the bride in marriage is a customary practice among the higher caste Hin- dus, especially among the families with land. The bride’s family gives the bride clothes and jewelry and also presents the groom’s family with costly gifts, household goods, cash, and, in some instances, property. Greater economic wealth has resulted in an increase in the amount of dowry among groups traditionally practicing it. It has also adversely affected women’s status through its institution in groups that formerly did not subscribe to it. Traditionally, the scheduled castes did not practice dowry. Instead, the institution of bride-wealth was the custom, in which the groom’s family usually gave the bride’s family some gifts and cash at the time of marriage. This tradition died out many years ago when dowry became a status symbol and a reflection of economic wealth (Ghosh 1987). This practice negatively affects those with more female members as they are forced to give up their valuable resources in agriculture such as land and livestock. Households with daughters are forced to sell their livestock to raise the dowry requirements imposed by the parents of the groom during marriage. Thus, having daughters becomes a liability to a poor household as social norms dictate that they should get married at a certain age. The pressure of raising the dowry is borne by the parents of the bride. Despite the existence of a law that prohibits the dowry system, this practice is deeply embedded, even in very poor families with limited assets. Caste structure. The caste structure forms the basic foundation of India’s social structure. Caste is the official social stratification, which is defined since birth. The lower castes are officially classified as the backward and scheduled castes. They are considered the most deprived and underprivileged in terms of access to resources and social status. To reduce the gap between the upper and lower caste, they are registered in a special governmental schedule and are now entitled to certain statutory measures of positive discrimination, such as reservation of seats in school and colleges, and job reservation. Women among the lower castes are also given reserved seats in the local political organization. Within the lower castes, the backward castes have a higher status than the scheduled castes on the social ladder. Caste is further subdivided into subcaste groups. The predominant upper castes are the Pandeys and Singhs. Among the backward castes, subcastes are the Bhuj, Chauhan, Gupta, Kohar, Vishwakarma, Barber, and Yadavs. The scheduled castes include subcastes such as Harijans, Kori, and Pasi. The caste classification explains the occupation of the households in the village. For example, the Yadavs of the backward caste in Chandpur are known for

346 Paris et al taking care of dairy cattle. On the other hand, the Harijans are the marginal smallholders and landless workers whose main livelihood depends on agricultural labor. People of the same caste tend to cluster themselves in the village. This is due to caste relations based on the principle of impurity, a principle that permeated all aspects of life, whether it was food, occupation, or other intercaste relations. The upper castes were deemed more pure and the lower castes less pure or even impure. There- fore, the upper castes scrupulously observed marriage and commensal restrictions and avoided physical contact with the lower castes. People could eat and drink with members of the same caste above them, but would not do so with the lower caste as it was thought to be “polluting.” Similar restrictions affect the sharing of utensils. Seg- regation of castes is almost complete in matters of residence. The extent of female participation in production in India is determined by a nexus of class/caste hierarchy and norms of patriachal ideology. Women from the upper castes stay in seclusion or “indoors” and do not engage in manual work to maintain their social status. Women from the lower castes have more freedom to work on their own farms and outside their homesteads to earn a living. Combining families into one household. A household constitutes more than one family or a group of persons normally living together and taking food from a common kitchen. The male head of the household is the principal male member. A typical nuclear household consists of a husband and wife with their own children, whereas a combined family includes grandparents, the older children, and grandchildren. Strong kinship ties provide safety nets for farming households, especially during periods of stress. The combined households still exist in India, particularly among the upper castes that own large farm holdings. Families pool their resources, particularly labor, and jointly manage their farms. For example, one hectare of land may be subdivided for the sons while all the earnings are given to the head of the household. One advantage of the combined family is that, when one family member is in need, all the other members help. However, because of the increasing population pressure and subdivision of lands and family conflicts, combined households are breaking down into nuclear units. The further subdivision of lands through inheritance has resulted in a small and marginal size of landholdings, especially among the poor. In India, it is customary for the bride to live with her in-laws, who manage a combined family consisting of sons, daughters, their in-laws, and their grandchildren. Wolf (1996), as cited in Kandiyoti (1991), argues that the key to the reproduction of classic patriarchy lies in the operations of the patrilocally extended households, which are commonly associated with the reproduction of the peasantry in agrarian societies. Under classic patriarchy, girls are given away in marriage at a very young age into households headed by the husband’s father. There, they are subordinate not only to all the men but also to the more senior women, especially their mother-in-law.

Characteristics of the households Farming is the major source of livelihood in the two remote villages. The proportion of farming households is highest in the remote villages Sariyawan (100%) and Mungeshpur (89%) and lower in the nearer village, Chandpur (76%). The majority of

Using gender analysis in characterizing and understanding . . . 347 Table 2. Socio-demographic indicators by village (percentage and mean), 1995.

Indicators Chandpur Mungeshpur Sariwayan (near) (far) (far)

Total number in household 200 150 81 Type of household (%) Farming/landed 76 89 100 Landless 25 11 – Caste (% of households) Upper 10 8 7 Backward 48 50 53 Scheduled 42 42 40 Kinship (% of households) Nuclear 47 61 30 Extended 54 39 70 Household head (% of households) Male-headed 75 81 69 De facto female-headed 19 11 27 De jure female-headed 7 8 4 Average household size by caste Upper 8 7 6 Lower 9 6 7 Average age (years) Male operator 45 42 52 Wife 41 37 48 Average years in school Male operator 7 7 4 Wife 2 1 1 households belong to the lower caste having marginal and small landholdings. Com- bined or extended households are still prevalent in Chandpur and Sariyawan (Table 2). A majority of the farm households are headed by males. However, the proportion of de facto female heads of households is higher in Sariyawan (27%) than in Chandpur (19%) and Mungeshpur (11%) (Table 2). A wife may act as the de facto head of the household when she makes most of the decisions in the household and on the farm. This happens when the husband is an invalid or sick or when he works outside the village for seasonal or permanent employment. It is a common practice in both villages for husbands to leave the village to find seasonal employment during the slack periods on the farm (i.e., usually in December and January after sowing of wheat) and come back in June during land preparation of rice. The male farm operators in Chandpur and Mungeshpur are generally in their forties, but not in Sariyawan, where they are in their fifties. Wives are younger than their husbands. The average educational level of the wife is lower (1 to 2 years) than that of her husband (4 to 7 years).

348 Paris et al The biophysical environment The performance of the farming system in these typical rainfed villages is affected by unpredictable and erratic rainfall resulting in drought, waterlogging, or submergence; nutrient deficiencies and toxicities; and weeds, insects, and disease pressure (Singh 1996). Erratic rainfall pattern. The problem with the rainfall pattern is that it is erratic, sometimes too much and sometimes none, and hence the floods and droughts that occur are major sources of crop loss (Widawsky and O’Toole 1990). Rice suffers from stress during the vegetative stage because of too little water in June or too much water during the panicle initiation stage in August. These uncertainties pose particular hardships for the poor, who face chronic vulnerability in terms of access to resources (Hossain 1995). In fact, the lives of the poor in India have been characterized by the almost total absence of security (Dreze and Sen 1989). The years 1995 and 1996, when this study was conducted, were drought years in Faizabad District, with drought being severe in 1995 (Pandey et al 1997). Heterogeneous land types and moisture stress. Aside from the unpredictable rainfall that makes farming very risky, the rainfed lowland farms suffer from different types of stress. Results using geographic information systems showed that, of the total rainfed area of Faizabad, 35% is drought-prone, 40% is shallow favorable, whereas the rest has problems of drought or submergence, or both, as well as sodicity. Chandpur represents a shallow and submergence-prone area, which is favorable rainfed during the years of low rainfall, whereas Mungeshpur and Sariyawan represent a drought- prone area that is favorable rainfed during the years of high rainfall (Singh 1992). Moreover, farmers have heterogeneous land types (lowland, medium, and upland) that affect rice varietal choice and performance. Thus, improved crop establishment methods and the introduction of short- and medium-duration rice varieties that depend on the land types are some of the technologies that have to be introduced to these villages to minimize the risks of growing rice during the kharif season. Within a farming systems framework, rice improvement must focus on increased tolerance for the predominant abiotic and biotic stresses such as late-season drought, submergence for less than 10 days, and blast (Sarkarung 1996).

Cropping patterns Depending on rainfall distribution, the crop year is divided into three growing seasons: the kharif or monsoon season (June-July to November-December), the rabi or winter season (November-December to March-April), and the zaid or summer season (March-June). In the lowlands and medium lands, rice-wheat and rice-wheat mixed with mustard were the predominant cropping patterns. Rice is planted in June-July and harvested in October or November, depending on the growth duration of the varieties used. To take advantage of residual moisture from the soil, farmers broadcast wheat immediately after harvesting rice in November. Most of the farmers broadcast mustard seeds after sowing wheat. In wheat + mustard intercropping, mustard is usually harvested early, in the first or second week of March. Wheat is harvested from

Using gender analysis in characterizing and understanding . . . 349 Rainfall (mm) 350 Rice Wheat/wheat + mustard Medium + 300 lowland Pea/gram/lentil/berseem

250 Rice Upland and Pigeonpea medium 200 irrigated Curbi (local fodder) 150 Sweet potato 1995 Potato 1996 100 Lahi (oilseed) Vegetable 50 Sudan chari/maize Sugarcane 0 Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Fig. 2. Rainfall and cropping pattern in Chandpur and Mungeshpur.

the last week of March until the second week of April. If rice is harvested late, wheat is also sown late and, consequently, wheat is not mixed with mustard. A few farmers grow wheat and mustard in separate fields because they think that mixing these crops lowers wheat yields. Two types of mustard seeds are grown. The first type is Lahi or Toria (60–80-day variety), which is grown after rice or maize in the uplands or after Curbi (green fodder crop). This variety is planted in October and harvested in Decem- ber as a sole crop. The second type is Varuna (120-day yellow variety), sown with wheat, gram pea, berseem, and potato. Mustard provides household oil needs aside from being a high-value feed (mustard oil cake) for livestock. It is also an important source of cash and income. For thinning purposes, farmers remove mustard as green fodder for animals and also as a green leafy vegetable earlier in the season. Dry mustard straw is also used for roofing and fuel (Fig. 2). In the medium lands and lowlands, pea, gram, and lentil are planted in October after rice and harvested in March. Farmers who have access to supplementary irrigation and raise livestock also grow berseem, a fodder crop, during this period. In the uplands, farmers grow sugarcane and pigeonpea throughout the year. Thus, farmers maximize the use of small plots by combining wheat and mustard to meet their food, fuel, oil, and animal feed requirements and for other housing materials.

Seasonal calendar and gender division of labor The seasonal calendar provides a format for analyzing activities by season and by gender. This seasonal calendar identifies the busy and slack months and the patterns of activities by male and female labor. It identifies “who does what,” particularly as this relates to the agricultural year and other seasonal patterns (Table 3). There are gender-specific tasks or degrees of specialization in rice production. Males are exclu- sively responsible for preparing the land, broadcasting seeds, and applying chemical

350 Paris et al M

M>F

F FF

M>F

MMMMM

M>F

M M>F

F>M F

F FFFFFF

F>M F>M

M>F

FFFF F F

F>M F>M F>M F>M F>M F>M F>M F>M F>M F>M F>M F>M

F>M

harvesting oilseeds

F>M means that female participation is more than male participation.

Table 3. Seasonal calendar and gender division of labor, 1995. 3. Seasonal calendar and gender division of labor, Table

Major activities June July Aug Sept Oct Nov Dec Jan Feb Mar April May

Preparing land for nursery seedbed Applying farmyard manure on fields Irrigating rice fields Pulling seedlings and transplanting rice Weeding rice fields Weeding Harvesting rice Threshing rice Hand-pounding rice

Land preparation for potato, oilseeds, pea Sowing potato, oilseeds, pea, and berseem (fodder) Land preparation for wheat Sowing wheat Irrigating and applying fertilizer on potato fields,

Harvesting potato and mustard Harvesting pea and mustar Harvesting wheat, pigeonpea, gram Threshing rabi crops

Making cow dung cakes Nonfarm work Collecting animal fodder Making baskets/storage bins Kharif (July/July-Nov/Dec) Rabi (Nov/Dec-Mar/April) Other activities Animal care (milking, feeding) a

Using gender analysis in characterizing and understanding . . . 351 fertilizers, whereas women are predominantly responsible for transplanting, weeding, harvesting, hand-threshing, hand-pounding, and cooking rice. Starting in June, male farmers start preparing the land and the nursery seedbed for rice production. July is the busiest month for women, when they are involved in applying farmyard manure on the rice fields, pulling rice seedlings from the seedbeds, and transplanting them on puddled fields. In August, women spend most of their time weeding rice fields. They continue these tasks in September. In October and November, women harvest rice and thresh until December. During the rabi season from September to October, men prepare the land for potato, Lahi, and pea depending on the time of harvesting of rice. During this period, women start to make dry cow dung cakes for fuel. In December, after sowing wheat, men begin to leave the villages in search of work. Women are busy weeding plots planted to vegetables and spices, collecting fodder for animals, hand-pounding rice, making storage bins out of clay, making cow dung cakes, and making baskets from local materials. In January, men irrigate the potato fields and tend the other rabi crops before they leave for nonfarm jobs. Women collect animal fodder, harvest mustard, and continue to make cow dung cakes for fuel. Women harvest potato and mustard in February and continue to make cow dung cakes and collect animal fodder. In March, women are engaged in harvesting potato, gram, and mustard from the wheat fields. May is the only month wherein women are not involved in field activities.

Rice production Rice is the major crop grown in the three villages. It occupies 60%, 73%, and 57% of the total cultivated land in Chandpur, Mungeshpur, and Sariyawan, respectively. Other crops such as sugarcane, pulses, pigeonpea, vegetables, and fodder crops are also grown on upland fields during the kharif season. About 10% of the total land is left fallow. There is more diversity in crops grown after rice during the rabi season (Table 4). In these villages, the adoption of improved varieties is high, ranging from 82% to 93%; however, despite their high adoption, average yields are low, ranging from 1.9 to 2.4 t ha–1, and only slightly higher than that of the local varieties, which average less than 1.5 t ha–1 (Table 5). Pandey et al (1998) in a similar study in eastern Uttar Pradesh revealed that yield is variable from plot to plot due to differences in soil type and management practices. Farmers, especially in Mungeshpur, continue to grow traditional varieties because of their tolerance for submergence and drought compared with the improved varieties. Sarkarung (1996) mentioned that the majority of improved rice cultivars developed on-station failed to perform under farmers’ conditions. This implies that the newly released cultivars could not compete with native and traditional cultivars under adverse conditions where water and fertility are not controlled. Aside from the heterogeneity in land types and management, farmers have different uses, needs, and preferences for rice varieties based on their socioeconomic differences, which affect varietal adoption. Thus, farmers’ criteria for rice varietal selection in the rainfed rice environments have to be well understood by both plant breeders and social scientists.

352 Paris et al Table 4. Crops grown during the kharif and rabi seasons, 1995.

Chandpur Mungeshpur Sariyawan

Crops Total % of Total % of Total % of area total area total area total (ha) area (ha) area (ha) area

Kharif Rice 24 60 40 73 32 56 Pigeonpea 3 8 3 5 – – Vegetables 1 3 2 4 3 5 Curbi 4 10 3 5 – – Sugarcane 4 10 2 4 3 5 Pulses – – – – 8 14 Pulses + vegetables/others – – – – 2 3 Others – – – – 5 9 Fallow 4 10 5 9 4 7 Total 40 100 55 100 57 100 Rabi Wheat + mustard 27 44 40 62 25 44 Wheat + pulses – – – – 8 14 Wheat 7 11 2 3 1 2 Pulses –– – 10 18 Pigeonpea 5 8 8 12 – – Mustard (oilseed) 3 5 – – 1 2 Green fodder (berseem) 2 3 – – – – Vegetables 4 7 3 5 2 4 Spices – – – – 1 1 Sugarcane 4 7 6 9 3 6 Pea + mustard 5 8 4 6 – – Oilseed + vegetables/spices/others – – – – 1 1 Vegetables + spices/others – – – – 4 6 Others – – – – 1 2 Fallow 4 7 2 3 – – Total 61 100 65 100 57 100

Table 5. Rice yield and adoption of rice varieties, 1995.

Actual Normal % of total Variety yield yield rice area (t ha–1) (t ha–1)

Chandpur All improved 2.0 2.5 83 All local 1.5 2.1 17 Mungeshpur All improved 2.4 3.0 82 All local 1.1 2.3 18 Sariyawan All improved 1.9 2.5 91 All local 1.3 1.9 9

Using gender analysis in characterizing and understanding . . . 353 Labor use in rice production and gender division of labor Rice production is a very labor intensive activity that employs both men and women in production until postharvest activities. Average labor days per hectare on rainfed lowland environments from key sites in South and Southeast Asia range from 95 to 270 d depending on the variety used, levels of technology, management, etc. (Table 6). Female labor participation in rice production can vary by country, agroecosystem, class/caste, availability of male labor, mother’s stage in the life cycle, and other factors. Compared with that of males, female labor participation is highest in Lao PDR and India, more than half in Thailand, Vietnam, and Nepal, and less than half in Indo- nesia and the Philippines. A closer look into labor days per hectare in four villages in Faizabad District reveals a higher proportion of female participation in the villages near the cities (Chandpur and Khanpur) than in the remote villages (Mungeshpur and Sariyawan) where farming is the major source of livelihood. Women are not a homogeneous group; rather, they belong to different socioeconomic categories such as class and caste. Caste, which is positively correlated with farm size, determines the extent of female labor participation in rice production in eastern Uttar Pradesh. A focus interview from households sampled in the socioeconomic surveys was conducted to quantify the labor inputs in rice production of adult males and females from different sources, by operation and by social status. Among the upper caste, female family members do not provide labor on their own farms, but instead hire women from the lower caste to substitute for their labor. However, within the confines of their homesteads they select and store seeds and feed the animals. The upper castes follow a very strict system of seclusion (purdah) (Table 7). In Chandpur,

Table 6. Labor inputs in rainfed rice production (d ha–1), 1995.

Country Villages Total Male Female (d ha–1) (%) (%)

Indonesia Jakenan, Central Java 161 54 46 Sumber, Central Java 178 59 41 Thailand Ban Sai Khram, South Thailand 104 45 55 Ban Don Paw Daeng 102 46 54 Philippines Carosucan, Sta. Barbara 133 83 17 Tampac, Nueva Ecija 188 68 32 Cambodia Kandal and Takeo 167 54 46 Vietnam He Thu District 105 45 55 Lao PDR Khok Nghai, Xaythani District 110 24 76 Ak-sang, Phonethong District 117 38 62 Nepal Naldung, Nagarkot (midhills) 269 42 58 Mohana, Rantnagar (lowland) 101 50 50 Baghmara, Rantnagar (lowland) 95 45 55 India Chandpur, Faizabad District (near) 187 16 84 Khanpur, Faizabad District (near) 210 24 76 Mungeshpur, Faizabad District (far) 132 33 67 Sariyawan, Faizabad District (far) 211 45 55

Sources: IRRI (1990, 1992).

354 Paris et al Table 7. Labor input (person-days ha–1) in rice production, Chandpur, Faizabad District, eastern Uttar Pradesh. Use of exchange labor is not practiced in this village, 1995.

Family Hired Total Caste/operation Total M F M F M F labor

Upper caste (n = 11) Prepare land 1.0 0.0 2.2 0.0 3.1 0.0 3.1 Pull seedlings/transplant 0.0 0.0 5.8 62.7 5.8 62.7 68.5 Broadcast seeds 2.0 0.0 0.7 0.0 2.8 0.0 2.8 Apply fertilizer 2.4 0.0 1.7 0.3 4.1 0.3 4.3 Weed 0.0 0.0 0.0 38.9 0.0 38.9 38.9 Irrigate 2.1 0.0 0.5 0.0 2.6 0.0 2.6 Harvest 1.4 0.0 2.3 26.6 3.6 26.6 30.2 Thresh (manual) 0.0 0.3 0.5 29.4 0.5 29.7 30.3 Total 8.9 0.3 13.7 157.9 22.5 158.2 180.7 Percentage of total labor 4.9 0.2 7.6 87.4 12.5 87.5 100.0

Backward caste (n = 51) Prepare land 6.3 0.0 0.4 0.0 6.8 0.0 6.8 Pull seedlings/transplant 4.2 14.3 0.7 39.0 4.9 53.3 58.1 Broadcast seeds 0.9 0.0 0.2 0.0 1.1 0.0 1.1 Apply fertilizer 6.8 3.6 0.0 0.0 6.8 3.6 10.5 Weed 2.7 24.1 0.8 30.3 3.4 54.4 57.8 Irrigate 0.4 0.1 0.0 0.0 0.4 0.1 0.5 Harvest 9.7 17.9 0.1 7.1 9.8 25.0 34.7 Thresh (manual) 0.6 29.8 0.0 7.3 0.6 37.1 37.7 Total 31.6 89.8 2.2 83.7 33.8 173.5 207.2 Percentage of total labor 15.2 43.4 1.0 40.4 16.3 83.7 100.0

Scheduled caste (n = 31) Prepare land 4.4 0.0 2.2 0.0 6.5 0.0 6.5 Pull seedlings/transplant 3.5 19.9 0.8 34.3 4.3 54.2 58.5 Broadcast seeds 0.3 0.0 0.7 0.2 1.0 0.2 1.1 Apply fertilizer 8.2 6.4 0.1 0.1 8.3 6.5 14.8 Weed 1.9 36.0 0.0 21.5 1.9 57.5 59.4 Irrigate 6.5 2.2 0.1 0.0 6.6 2.2 8.7 Harvest 7.4 24.8 1.0 5.0 8.4 29.8 38.2 Thresh (manual) 2.4 35.0 0.7 4.3 3.0 39.3 42.3 Total 34.6 124.3 5.6 65.4 40.0 189.7 229.5 Percentage of total labor 15.1 54.2 2.3 28.5 17.4 82.6 100.0

Using gender analysis in characterizing and understanding . . . 355 female family members among the upper caste households contribute less than 1% of labor. In contrast, female members among the backward and scheduled castes contribute 43% and 54%, respectively. These results are similar to the study of Kelkar (1992) on the roles of women in Bihar, which are influenced by class and caste factors. In this village, the contributions of upper and lower caste male family members to total labor in rice production are 5% and 15%, respectively. The low participation of male family members in farming is due to their participation in nonfarm employment in Faizabad City. Most male members commute every day to the city for work and leave most of the farm work to the female family members. Similarly, in Mungeshpur, female family members from the upper castes do not work on their own farms. In contrast, the female family members from the lower castes contribute about one half of the total labor input in rice production. Among the upper caste, the male family members contribute a low 7%, whereas the men from the lower caste contribute about one-fourth (Table 8). Although upper caste women are not supposed to provide physical labor in crop production activities outside the homesteads, our findings show that women from the upper caste in Sariyawan village break the norms, out of economic necessity. Female family members from the upper caste contribute about 10% of labor in rice production (Table 9). These women are either widows or de facto heads of households who are left to manage their own farms. Exchange labor is still commonly practiced in Mungeshpur and Sariyawan, as a strategy for managing production requirements when cash is scarce. About 20 households from the lower caste organize themselves so that they can exchange labor. As found by Chen (1990), households from the same kinship and caste group often bor- row, loan, pool, or exchange productive assets, including labor, pump sets, farm implements, and bullocks.

Value of unpaid male and female labor in rice production Women’s unpaid work in rice production can be made “visible” by imputing the value of male and female family labor using the prevailing market wage rates. An estimation of noncash costs in rice production (Table 10) reveals that female family labor (exchange and hired) contributes about 25% of the total costs. In contrast, male family members contribute less than 10%. On a per-hectare basis, the imputed value of the unpaid labor of female family members is about US$35.71 per hectare. These results indicate the importance of female family labor in saving the costs of hiring labor in rice production. However, the work burden of women from the lower caste, small, and marginal farming households increases without necessarily having a com- pensatory improvement in their standard of living. Beteille (1985, cited in Ghosh 1987) has observed that most Western scholars associate emancipation of women with the right to work. But there is also another side to the picture. In agrarian societies, work is regarded more often as a hardship than as a privilege, because agricultural work entails arduous physical work, low status, and low pay or no pay as in the case of female family members (Ghosh 1987).

356 Paris et al Table 8. Labor input (person-days ha–1) in rice production, Mungeshpur, Faizabad District, eastern Uttar Pradesh, India, 1995.

Family Exchange Hired Total Caste/operation Total M F M F M F M F labor

Upper caste (n = 5) Prepare land 0.0 0.0 0.0 0.0 3.4 0.0 3.4 0.0 3.4 Pull seedlings/transplant 0.3 0.0 0.0 0.0 17.3 32.4 17.6 32.4 50.0 Broadcast seeds 0.5 0.0 0.0 0.0 0.3 0.1 0.8 0.1 0.9 Apply fertilizer 4.1 0.0 0.0 0.0 0.8 0.0 5.0 0.0 5.0 Weed 0.1 0.0 0.0 0.0 8.1 27.6 8.3 27.6 35.8 Irrigate 2.6 0.0 0.0 0.0 0.0 0.0 2.6 0.0 2.6 Harvest 2.2 0.0 0.0 0.0 7.6 20.2 9.8 20.2 30.0 Thresh (manual) 1.0 0.0 0.0 0.0 4.7 16.3 5.6 16.3 21.9 Total 10.8 0.0 0.0 0.0 42.2 96.6 53.1 96.6 149.6 Percentage of total labor 7.2 0.0 0.0 0.0 28.3 64.5 35.5 64.5 100.0

Backward caste (n = 33) Prepare land 9.6 0.0 0.0 0.0 0.6 0.0 10.1 0.0 10.1 Pull seedlings/transplant 6.3 16.2 1.6 4.3 6.2 15.7 14.0 36.1 50.1 Broadcast seeds 0.4 0.1 0.0 0.0 0.0 0.0 0.4 0.1 0.5 Apply fertilizer 3.5 2.7 0.0 0.0 0.1 0.0 3.6 2.7 6.2 Weed 3.0 14.4 0.2 2.6 1.4 5.1 4.6 22.1 26.7 Irrigate 3.1 1.0 0.0 0.0 0.2 0.0 3.3 1.0 4.3 Harvest 5.5 15.6 0.5 3.3 1.2 3.8 7.3 22.7 29.9 Thresh (manual) 3.2 20.9 0.0 2.3 0.8 2.3 4.0 25.6 29.5 Total 34.6 70.9 2.3 12.5 10.5 26.9 47.3 110.3 157.3 Percentage of total labor 22.0 45.0 1.4 7.9 6.6 17.1 30.0 70.0 100.0

Scheduled caste (n = 28) Prepare land 8.4 0.0 0.0 0.0 0.8 0.0 9.2 0.0 9.2 Pull seedlings/transplant 8.1 20.1 0.8 3.6 4.1 14.2 13.0 37.9 50.9 Broadcast 0.4 0.1 0.0 0.0 0.0 0.0 0.4 0.1 0.4 Apply fertilizer 4.4 5.3 0.0 0.0 1.0 0.0 5.3 5.3 10.6 Weed 7.7 22.9 0.0 1.3 1.0 6.3 8.7 30.5 39.2 Irrigate 3.9 2.5 0.0 0.0 0.0 0.1 3.9 2.6 6.5 Harvest 8.3 18.5 0.0 0.8 0.9 4.1 9.2 23.4 32.6 Thresh (manual) 5.6 22.2 0.0 0.8 0.1 0.9 5.7 23.9 29.6 Total 46.8 91.6 0.8 6.5 7.9 25.6 55.4 123.7 179.0 Percentage of total labor 26.1 51.1 0.4 3.7 4.4 14.2 31.0 69.0 100.0

Using gender analysis in characterizing and understanding . . . 357 Table 9. Labor input (person-days ha–1) in rice production, Sariyawan, Faizabad District, eastern Uttar Pradesh, India, 1995.

Family Exchange Hired Total Caste/operation Total M F M F M F M F labor

Upper caste (n = 4) Prepare land 6.4 0.3 0.0 0.0 5.3 0.0 11.6 0.3 11.9 Pull seedlings/transplant 2.5 1.3 0.0 0.0 13.8 47.8 16.3 49.0 65.3 Apply fertilizer 3.6 0.0 0.0 0.0 0.0 0.0 3.6 0.0 3.6 Apply FYM 7.6 0.6 0.0 0.0 4.6 0.0 12.3 0.6 12.9 Weed 6.5 5.0 0.0 0.0 1.0 13.4 7.5 18.4 25.9 Irrigate 5.1 0.9 0.0 0.0 0.0 0.0 5.1 0.9 6.1 Harvest 6.6 5.6 0.0 0.0 5.6 24.4 12.3 30.0 42.3 Thresh 5.4 4.7 0.0 0.0 4.3 13.9 9.6 18.6 28.2 Total 43.7 18.4 0.0 0.0 34.6 99.5 78.3 117.8 196.2 Percentage of total labor 22.3 9.4 0.0 0.0 17.6 50.7 39.9 60.1 100.0

Backward caste (n = 9) Prepare land 15.7 0.8 0.0 0.0 1.2 0.0 16.9 0.8 17.7 Pull seedlings/transplant 14.7 17.2 0.7 4.2 2.5 10.4 18.0 31.8 49.8 Apply fertilizer 5.1 0.0 0.0 0.0 0.0 0.0 5.1 0.0 5.1 Apply FYM 5.7 3.9 0.0 0.0 1.0 0.6 6.7 4.5 11.2 Weed 16.9 19.0 0.0 4.0 2.5 13.6 19.3 36.5 55.9 Irrigate 1.3 0.1 0.0 0.0 0.0 0.0 1.3 0.1 1.4 Harvest 12.6 18.2 0.0 1.2 1.0 3.1 13.6 22.6 36.2 Thresh (manual) 7.5 15.5 0.1 0.3 0.2 2.5 7.8 18.2 26.0 Total 79.5 74.7 0.8 9.7 8.4 30.2 88.7 114.5 203.3 Percentage of total labor 39.1 36.8 0.4 4.7 4.1 14.8 43.7 56.3 100.0

Scheduled caste (n = 27) Prepare land 20.1 0.4 0.0 0.0 0.0 0.0 20.1 0.4 20.5 Pull seedlings/transplant 17.8 17.9 0.0 5.4 0.0 0.0 17.8 23.3 41.1 Apply fertilizer 5.4 0.0 0.0 0.0 0.0 0.0 5.4 0.0 5.4 Apply FYM 4.7 2.9 0.0 0.0 0.0 0.0 4.7 2.9 7.6 Weed 23.6 27.5 0.0 5.6 0.0 2.7 23.6 35.7 59.4 Harvest 13.6 19.6 0.0 1.7 0.0 0.0 13.6 21.2 34.9 Thresh 6.3 14.5 0.0 0.0 0.0 0.0 6.3 14.5 20.8 Total 91.5 82.8 0.0 12.7 0.0 2.7 91.5 98.0 189.7 Percentage of total labor 47.6 43.1 0.0 6.6 0.0 1.4 49.0 51.0 100.0

358 Paris et al Table 10. Cost and returns of rice production, Faizabad District, eastern Uttar Pradesh, 1995.

Chandpur Mungeshpur Sariyawan

Variables US$ % US$ % US$ %

Cash inputs Seeds 7.57 5 7.37 5 6.89 5 Fertilizer 31.94 19 26.43 18 28.86 20 Tractor rent 22.17 13 8.57 6 6.49 5 Animal rent 1.20 1 1.14 – 2.26 2 Irrigation 30.06 18 28.91 20 17.71 12 Hired male labor 0.29 – 2.26 2 2.51 2 Hired female labor 18.00 11 4.66 3 7.14 5 Total paid-out cost (TPC) 111.23 67 79.34 54 71.86 51

Noncash inputs Farmyard manure 6.60 4 2.60 2 2.91 8 Family adult male 8.17 5 12.06 8 11.29 8 Exchange adult male – – 1.71 1 2.00 1 Family adult female 30.46 19 27.00 18 26.11 18 Exchange adult female 3.23 2 8.63 6 6.54 5 Own animal 3.97 2 11.09 8 10.31 7 Exchange animal 0.71 1 3.71 3 2.89 2 Total noncash cost (TNC) 53.14 33 66.80 46 62.05 49 Total cost (TC) 164.37 100 146.14 100 133.91 100 Gross returns (GR)a 274.57 214.26 238.97 Net income (GR – TPC) 163.34 134.92 167.11 Surplus (GR – TC) 110.20 68.12 105.06 Average yield (t ha–1) 2.2 1.7 1.6 Average rice area (ha) 0.23 0.29 0.34

aIncludes byproducts.

Animal systems Across all the study villages, animals constitute an integral part of the mixed farming systems. Farming households raise a mix of a small number of bullocks, cattle, and buffalo. Bullocks and male buffaloes are used as draft power for plowing and harrowing the fields, and transporting farm products. Animal manure is used as organic fertilizer for crops and converted into dung cakes for household fuel. On the other hand, the biomass and by-products of crops and residues from the fields are fed to the livestock. Crop biomass includes straw of rice and wheat, green sugarcane tops, and pigeonpea and gram straw. Rice and wheat straw are also used as bedding, particularly during winter, as roof thatch, and as storage insulators, and are mixed with clay for making storage bins. Rice straw usually lasts for 3 months. It is also a source of cash income when sold during times of fodder scarcity. Farmers who grow sugarcane use bullocks to crush the sugarcane stalks. Female buffaloes are raised for milk, curd, and ghee, which are parts of the daily diet of Indian families.

Using gender analysis in characterizing and understanding . . . 359 Ownership of livestock allows farmers to fend off risks of drought and to maintain capital in the form of animals as insurance against bad times, particularly during drought. Livestock are also raised to cover large expenditures such as medical expenses and payment for dowry during marriage. Farmers also use livestock in exchange for another resource. For example, a farmer with a tube well can provide water to the land of another farmer who, in turn, prepares his land with his pair of bullocks. The landless laborers who own livestock but do not have crops and crop residues work on a sharing basis and provide inputs to the land they cultivate as long as they get rice and wheat straw for their animals. For poor women, raising goats provides them with independent income, security, and instant cash during times of emergency. Gupta (1991) stressed that, in a rainfed economy, it is not the crops but the livestock that are the main anchor of household survival in the dry regions. Once this is recognized, the primacy of fodder (whether from grasslands, trees, or crop residues) vis-à-vis grain becomes clear.

Animal population A higher proportion of the farming households in Chandpur raises cattle and buffaloes, whereas farming households in Mungeshpur and Sariyawan raise more bullocks. Women from poor and landless households raising goats is a more popular activity in Mungeshpur and Sariyawan. A few households raise pigs as a means of security and instant cash (Table 11). Because of the proximity of Chandpur to the market, household members, particularly those belonging to the backward castes (Yadavs), raise dairy cattle for consumption and for sale. According to key informant interviews, the bullock population in Chandpur has been declining because farmers are shifting to the use of tractors for land preparation.

Table 11. Percentage of farming households that own animals by village, 1995.

Chandpur Mungeshpur Sariyawan (n = 151) (n = 133) (n = 81) Animal No. of No. of No. of households % households % households %

Cattle 89 59 60 45 16 20 Buffalo 101 67 42 32 36 44 Bullock 13 9 38 29 44 54 Goat 12 8 31 23 22 27 Pig – – 3 2 2 2 Chicken – – 2 2 – –

360 Paris et al Gender roles in animal husbandry When one visits the villages in eastern Uttar Pradesh, it is common to see women carrying a huge headload of green fodder home to where the animals (buffaloes or cows) are kept tethered in the courtyard. Since women are responsible for milking and taking care of milch animals, they bear the responsibility for collecting and cutting green fodder for the animals. Twice a day, they feed and milk the animals, and also clean the shed. With limited grazing lands, women on average spend half a day walking long distances to collect grasses and weeds, particularly in Chandpur, where more dairy cattle require green fodder every day. Thus, women are the worst affected when drought occurs because this means that they have to walk farther and spend longer hours to collect animal fodder for their livestock (Paris et al 1998). Poor women also volunteer to weed the fields without wages as long as they can take home the weeds for their animals. During the summer season (November to February), women spend 4 to 6 hours per day making cow dung cakes for household fuel. Making dry dung cakes is an income- or expenditure-saving activity wherein women save approximately US$20 per year. In Chandpur, five to six households sell cow dung cakes for fuel. Because of poverty, women minimize the use of purchased inputs while maximizing the exploitation of residues, by-products, and their own labor. A decline in the animal population will result in increasing demands on women’s schedules, as they will have to travel farther into the forests in search of fuel. On the other hand, when more animals are raised, women’s work burden is greater. Thus, technologies that can increase the availability of fuel and animal fodder will directly benefit the female members of farming households.

Diversifying sources of income Despite the importance of rice as a staple food and in terms of land area, rice contributed only a small (less than 12%) proportion of the total income in 1995-96 (Table 12). Thus, farming households diversify their income sources. Wheat and other crops (pulses, oilseeds, sugarcane, vegetables, etc.) are the major crops in terms of their share in income, particularly in the remote villages. Farmers, particularly the lower castes with small landholdings, rely more on nonfarm income. Sales of livestock products such as milk are also an important source of cash income among the farming households. For women, taking care of goats is one strategy for securing an independent source of income. They use their earnings to buy their own saris (clothes), bed sheets, food, and medicine for their families. Family members from the lower caste, particularly the female members, derive cash by working as agricultural laborers on other farms. Thus, the higher the cropping intensity within the village, the greater is the employment and income opportunity for the poor and landless women. Because of the lower wages received by women, however, the proportion of off-farm income to total income is quite low.

Using gender analysis in characterizing and understanding . . . 361 Table 12. Percentage share of different sources of income of farming households, by village and caste, 1995-96.

Chandpur Mungeshpur Sariyawan Income source Upper Lower Upper Lower Upper Lower (n = 13) (n = 97) (n = 12) (n = 121) (n = 6) (n = 75)

Rice 7 5 9 11 9 4 Wheat 5 4 11 13 8 9 Other crops 7 5 21 9 42 46 Livestock 35 13 14 15 13 7 Farm by-products 2 1 10 2 – 4 Rent of machine – 1 2 – 20 8 Farm labor – 3 – 5 – 11 Nonfarm 44 68 33 45 8 11 Total 100 100 100 100 100 100 Av (US$ y–1) 2,090 1,100 1,260 600 860 200

Males Females Household

Animal management

Nonfarm activities

Off-farm activities

Farm activities

020406080

Percentage of total time used Fig. 3. Time allocation of principal male and female family members, Mungeshpur village, Faizabad. Time allocation Men and women have different uses of time depending on their degrees of specialization and opportunity costs of their labor. To better understand the differences in time use, 20 principal males and females from lower caste farming households in Mungeshpur were interviewed with regard to the number of hours they spend per day and per month in major activities. The relative contribution of the principal females is consistently higher than that of the principal males in all activities, except in nonfarm employment (Fig. 3). The principal females spend 62% in farm activities, 61% in off-

362 Paris et al farm activities, 25% in nonfarm activities, 75% in animal management, and 76% in household activities. This indicates that women’s work burden is greater than that of their male counterparts in almost all activities except nonfarm work. Most often women have to combine reproductive and productive activities to meet the competing demands for their time.

Resource endowments Farm size, rice area, land use, and rice diversity by social status Although India has a rich agriculture, with a huge mass of land, farming is generally dominated by small landholders from the lower castes, whereas the large landholders belong to the upper castes (Gopalan 1992). This is evident in these study villages where farm size and rice area are negatively related to social status (Table 13). The small size and fragmentation of landholdings are major constraints to increasing the efficiency of rice productivity, particularly in the allocation of water, land preparation with the use of tractors, and proper management of plots. Farming households from the lower castes, however, tend to maximize their land use throughout the year as reflected by the cropping intensity indices (CII). Moreover, the lower castes tend to

Table 13. Indicators of land use and rice diversity, 1995.

Average Rice Village/caste farm size area CIIa CDI RVI (ha) (ha)

Chandpur Upper 1.01 0.49 161 0.76 0.40 Backward 0.33 0.22 197 0.60 0.18 Scheduled 0.24 0.17 198 0.60 0.13

Mungeshpur Upper 1.32 0.56 175 0.77 0.44 Backward 0.60 0.38 186 0.68 0.27 Scheduled 0.26 0.22 196 0.60 0.13

Sariyawan Upper 3.56 1.01 150 0.73 0.44 Backward 1.56 0.40 150 0.66 0.38 Scheduled 0.93 0.33 150 0.64 0.44

aCII (crop intensity index) indicates the extent of land use. An index of 200 indicates full use of land. CDI (crop diversification index) indicates the diversity of crops grown. This index ranges between zero and one with higher values indicating a greater degree of diversification. RVI (rice diversity index) indicates the number of rice varieties grown. This index ranges between zero and one with higher values indicating a greater degree of diversification.

Using gender analysis in characterizing and understanding . . . 363 meet their food and fodder requirements from their own limited land. Farming households from the upper castes that have a larger size of landholdings tend to grow more than one crop and more than one variety of rice as shown by the crop diversification (CDI) and rice varietal (RVI) indices.

Access to supplementary irrigation A majority of the farmer-operators across all villages and castes own the lands they are cultivating; however, they suffer from a lack of an assured source of water for rice production. This causes delays in crop establishment and drought stress at some stages of crop growth. In low-lying areas, the onset of heavy monsoon, accumulation of rainwater, and slow and inadequate drainage cause delays in crop establishment and damage to the standing crop from flooding. These conditions result in decreased rice yields as well as low overall farm productivity (Singh 1996). Farmers in the three villages obtain supplementary irrigation through their own and rented tube wells (Table 14). There is disparity by caste in terms of access to irrigation facilities. Of the total farming households in Chandpur, 23% and 13% of the upper and lower caste, respectively, invested in pump sets and tube wells. In Mungeshpur, a higher proportion (67%) was able to afford investing in supplementary irrigation facilities. Only 9% of the lower caste households in this village have their own tube wells and pump sets. Thus, they rent from the richer farmers. Farmers from the lower caste are often at the mercy of upper caste households who have more access to water. Those who do not own irrigation facilities suffer from crop failure. Because of the high cost of supplementary irrigation, farmers in Mungeshpur use their own irrigation facilities not only

Table 14. Access to land and supplementary irrigation by village and caste (%), 1995.

Chandpur Mungeshpur Sariyawan

Farm characteristic Upper Lower Upper Lower Upper Lower caste caste caste caste caste caste (n = 13) (n = 138) (n = 12) (n = 121) (n = 6) (n = 75)

% of land (tenure status) Owned 88 81 98 94 100 97 Share tenant 12 9 2 6 – – Leasehold – 10 – – – 3

% of land irrigated by Own tube well 21 40 66 17 34 34 Rented tube well 55 40 34 80 66 62 Canal irrigation/ 24 20 – 3 – 4 none

% of farmers with 23 13 67 9 33 35 own irrigation facility

364 Paris et al for rice but also for cash crops such as vegetables sometimes during the kharif and often during the rabi season. Access to supplementary irrigation enables farmers to grow rice, wheat, berseem, and vegetables for consumption and for the market. Poor farmers, particularly the widows who can’t afford to pay the rental fees of water for irrigation, exchange their labor for water but are often taken advantage of by the tube well owners, who pay them lower than normal wage rates.

Human capital A major factor influencing women’s productivity is the extent to which they have access to education, training, and extension. There is general agreement that education increases productivity and a substantial literature exists documenting the positive effects of women’s education on human capital development, paid labor force participation, and agricultural production (Cloud 1985). There is a wide disparity in access to education by caste and gender of adult workers (Fig. 4A). Among all the adult females of the upper castes in Chandpur and Mungeshpur, about one-fourth have not gone to school. Illiteracy rates among the lower caste female adults, however, are very high at 80% in Chandpur and 91% in Mungeshpur. In both villages, most of the upper caste adult males are literate. In contrast, among the lower caste adult males, 37% in Chandpur and 58% in Mungeshpur were not able to go to school. Literacy rates among children 15 years old and below, especially for girls, are higher in Chandpur than in Mungeshpur among the lower caste (Fig. 4B). This trend is similar in Sariyawan where female illiteracy is higher than male illiteracy.

Literacy rate (%) 100 A Chandpur Mungeshpur 80 60 40 20 0 100 B Chandpur Mungeshpur 80 60 40 20 0 Males Females Males Females UC LC UC LC UC LC UC LC Illiterate Literate

Fig. 4. Literacy rates of (A) adult family members of farming households (above age 16) and (B) males and females (15 years and below). UC = upper caste, LC = lower caste.

Using gender analysis in characterizing and understanding . . . 365 According to Bennett (1989), the higher illiteracy rates of women than men are common in eastern India. Five populous states (Andhra Pradesh, Bihar, Madhya Pradesh, Rajasthan, and Uttar Pradesh), wherein rice farming is predominantly rainfed, account for more than half of India’s illiterate females. These five states contain 89% of India’s districts where the rural literacy rate is below 5%, 83% of those with rates of 5–9%, and 67% with between 10% and 14%. Illiteracy is widespread among rural women of the scheduled castes and scheduled tribes. The reasons for the low literacy rates among women and girls are social, cultural, and economic. Among poor farming and landless households, the need for child labor within and outside the home is a major reason for boys and girls not to attend or to drop out of school. Girls are expected to help with the domestic chores, substitute for their mothers in taking care of younger siblings, and help in field activities while boys help tend animals after school. Another reason for the women’s lack of access to education is the greater limitation parents put on a girl’s freedom of movement, which may prevent her from going to school after a certain age. According to Mukhopadhyay (1984), girls who have reached the age of puberty are withdrawn from schools because of the “social dangers” associated with male school teachers and students. Thus, socialization, gender roles, and sexual mores all play important roles in depriving girls of formal education (Bennett 1989). Girls are married off at an early age, thus confining them to the status of daughter-in-law, which curtails their freedom of movement, association, and communication even further. It is also traditionally believed that sons are more important because a daughter will leave her mother’s home and join her husband’s family after marriage. Sons are expected to take care of their parents in their old age (whole life) and after death, when they will perform the last rituals. Another factor contributing to low education levels for girls is the small return anticipated from girls’ schooling. While boys’ education is viewed as an investment in families’ socioeconomic status and as old-age security for parents, girls are destined to be married into other families and hence yield no returns to their parents (Bennett 1992). Girls will be mothers and workers in occupations that require little formal education. Investment in boys’ education is likely to pay more dividends in the future in terms of increased chances of employment and consequent support of the family. In the dowry system, males who have higher education can request a higher dowry price. In addition to these reasons, the direct costs of education also deter families from sending their girls to school. Although school education in India is entirely free, expenses on books and learning materials, uniforms, and transport can be a heavy burden on poor families.

Access to agricultural-related information A focused survey of male and female farmers from the lower castes in the three villages was conducted to determine their sources of agricultural information (Table 15). A majority of the men and women interviewed obtained their technical knowledge from their neighbors. In both villages, households of the same caste cluster together and obtain information through socialization. This information indicates the importance of social networks and kinship in disseminating information and identify-

366 Paris et al Table 15. Access to agricultural information of males and females, 1995.

Village/source Male % Female %

Chandpur Neighbors 18 60 13 65 Extension staff 5 17 3 15 Research institute 4 13 4 20 Radio 3 10 3 15 Television 6 20 3 15 Spouse and relatives 12 40 6 30

Mungeshpur Neighbors 23 76 15 75 Extension staff 7 23 2 10 Research institute 6 20 5 25 Radio 2 6 Television 2 6 Spouse and relatives 15 50 3 10

Sariyawan Neighbors 21 70 15 75 Extension staff 10 33 2 10 Research institute 8 26 3 15 Radio 4 13 2 10 Television 3 10 1 5 Spouse and relatives 5 16 8 40 No. of respondents 30 20 ing the key persons or “shining stars” who can serve as agents of change. A low percentage of the female respondents receive information from extension agents. One reason for this is the general lack of female extension workers who can directly interact with women farmers. FAO (1991) revealed that women constitute a mere 0.59% of India’s agricultural extension workers. Since almost all extension workers are men, women’s roles and skills are usually overlooked, even in areas where they do most of the work.

Summary and conclusions This chapter demonstrated the use of gender analysis as an analytical tool in characterizing and understanding farm-household systems in rainfed lowland rice villages. The analysis showed that, within farm households, there are gender-specific roles and responsibilities and gender differences in access to resources that have to be considered by scientists and extension and development workers. Although poor women in rainfed lowland rice environments play vital roles in sustaining food security and alleviating poverty, they face several constraints that limit their potential for increasing farm productivity. These barriers to productivity are a lack of access to education,

Using gender analysis in characterizing and understanding . . . 367 training, extension, new seeds suitable to their environments, animal fodder, and equipment to ease their workload and overcome drudgery in performing their farm tasks. One strategy for enhancing their productivity and income is to provide them with access to new seeds suitable to their specific adverse rainfed environments. Efforts are now being made under the System-wide Initiative on Farmer Participatory Plant Breeding and Gender Analysis to develop methodologies that involve both male and female farmers in rice variety development in rainfed environments (Paris et al 1998). Other potential research areas for enhancing women’s roles will be in producing animal fodder within the cropping systems and developing agricultural/mechanical implements/tools to reduce their drudgery, increase their labor efficiency, and explore ways to optimize the use of rice by-products and home-based technologies. Gender analysis will also be replicated to complete the socioeconomic and cultural characterization of major rainfed lowland rice environments in South and Southeast Asia and to provide a gender-related database for policymakers in addressing gender issues in agriculture.

References Agarwal B. 1998. Disinherited peasants, disadvantaged workers: a gender perspective on land and livelihood. Econ. Polit. Wkly. p 2-14. Bennett L. 1992. Women, poverty, and productivity in India. EDI Seminar Paper Number 43. Washington, D.C. (USA): Economic Development Institute of the World Bank. Bennett L. 1989. Gender and poverty in India: issues and opportunities concerning women in the Indian economy. Washington, D.C. (USA): World Bank. 153 p. Chen M. 1990. Coping with seasonality and drought. New Delhi (India): Sage Publications. 247 p. Cloud K. 1985. Women’s productivity in agricultural systems. In: Overholt C, Anderson MB, Cloud K, Austin J, editors. Gender roles in development projects: a case book. West Hartford, Conn. (USA): Kumarian Press. p 57-78. Dreze J, Sen A. 1989. Hunger and public action. Oxford (UK): Clarendon Press. Duvvury N. 1989. Women in agriculture: a review of the Indian literature. Econ. Polit. Wkly. 28 October 1989. FAO (Food and Agriculture Organization). 1991. Most farmers in India are women. New Delhi (India): FAO. 20 p. Feldstein HS, Poats SV, Cloud K, Noreem R. 1989. Intra-household dynamics and farming systems research and extension: conceptual framework and worksheets. In: Feldstein HS, Poats SV, editors. Gender and agriculture: case studies in intra-household analysis. West Hartford, Conn. (USA): Kumarian Press. Ghosh H. 1987. Changes in the status of north Indian women: a case study of Palitpur villages. Working Paper No. 141. Michigan State University, East Lansing, Mich. (USA). Gopalan S. 1992. Sectoral policies on agriculture and related macro-economic policies and their gender-responsiveness. New Delhi, India. (In mimeo.) Gupta AK. 1991. Reconceptualizing development and diffusion of technology for dry regions. In: Prasad C, Das P, editors. Extension strategies in rainfed agriculture. New Delhi (In- dia): India Society of Extension Education. p 322-356.

368 Paris et al Hossain M. 1995. Recent developments in the Asian rice economy: challenges for rice research. In: Evenson R, Herdt R, Hossain M, editors. Rice research in Asia: progress and priorities. Manila (Philippines): International Rice Research Institute and Center for Agriculture and Bioscience International. p 59-70. IRRI (International Rice Research Institute). 1990. Gender analysis in rice farming systems research: Does it make a difference? Report (unpublished) of the Women in Rice Farm- ing Workshop held in Indonesia, 4-8 June 1990. IRRI (International Rice Research Institute). 1992. Proceedings of the international workshop on gender concerns in rice farming, Chiang Mai, Thailand, 22-25 October 1992. Manila (Philippines): IRRI. IRRI (International Rice Research Institute). 1997. Sustaining food security beyond the year 2000: a global partnership for rice research. Manila (Philippines): IRRI. Kandiyoti D. 1991. Bargaining with patriarchy: social construction of gender. New Delhi (India): Sage Publications. p 114-118. Kelkar G. 1992. Women, peasant organizations and land rights: a study from Bihar, India. Occasional paper. Gender and Development Studies, Asian Institute of Technology, Bangkok, Thailand. 50 p. Mukhopadhyay M. 1984. Silver shackles: women and development in India. Oxford (UK): OxFam. Norman D, Simmon EB, Hays M. 1983. Farming systems in the Nigerian Savanna: research and strategies for development. Boulder, Col. (USA): Westview Press. Pandey S, Singh HN, Villano R. 1998. Rainfed rice and risk-coping strategies: some macroeconomic evidences from Eastern Uttar Pradesh. Paper presented at the NCAP-IRRI Workshop on Risk Analysis and Management in Rainfed Rice Systems. 21-22 Sep 1998, New Delhi. Paris T, Singh A, Hossain M, Luis J. 1998. Incorporating gender concerns in rice varietal improvement and germplasm conservation: preliminary results in eastern Uttar Pradesh, India. Paper presented at the 2nd International Seminar of the CGIAR System-Wide Program on Participatory Research and Gender Analysis (SWP PRGA) held in Quito, Ecuador, 6-9 September 1998. (Forthcoming.) Poats S. 1990. Gender issues in the CGIAR system: lessons and strategies from within. Paper presented at the 1990 CGIAR Mid-Term Meeting, 21-25 May 1990, The Hague, The Netherlands. Quisumbing A. 1995. Women in agricultural systems. In: Strommoquist N, editor. Women in the Third World: an encyclopedia of contemporary issues. New York (USA): Garland Publishing, Inc. p 262-271. Sarkarung S. 1996. Breeding rice cultivars suitable for rainfed lowland environmetns: a farmer participatory approach in eastern India. In: Eyzaguire P, Iwanaga M, editors. Participa- tory plant breeding. Proceedings of a Workshop on Participatory Plant Breeding, 26-29 July 1995, Wageningen, The Netherlands. Rome (Italy): International Plant Genetic Resources Institute. Singh VP. 1992. Institutionalization of a farming systems approach to development. In: Pro- ceedings of technical discussions. Rome (Italy): Food and Agriculture Organization. Singh VP. 1996. Monitoring and assessing the impact of a participatory research for the development of sustainable production systems: a decade of experience in rainfed situations of Eastern India. Manila (Philippines): International Rice Research Institute.

Using gender analysis in characterizing and understanding . . . 369 Widawsky D, O’Toole JC. 1990. Prioritizing the rice research agenda for Eastern India. In: Evenson RE, Herdt RW, Hossain M, editors. Rice research in Asia: progress and priorities. Manila (Philippines): IRRI and CAB International.

Notes Authors’ addresses: Thelma Paris, Affiliate Scientist-Gender Specialist, Social Sciences Divi- sion, IRRI; Abha Singh, Sociologist, Kumarganj, Faizabad District, Eastern Uttar Pradesh; Mahabub Hossain, Agricultural Economist and Head, Social Sciences Division, IRRI; Joyce Luis, Assistant Scientist, Social Sciences Division, International Rice Research Institute, DAPO Box 7777, Metro Manila, Philippines. Acknowledgments: The authors wish to thank Dr. R.K. Singh, former Director of Research of the Narendra Deva University and Agricultural Technology (NDUAT) in Kumarganj, Faizabad District, Eastern Uttar Pradesh, and Dr. V.P. Singh, Agronomist, APPA Divi- sion, IRRI, for their suggestions. We are also grateful for the assistance in data analysis provided by Ms. Josie Narcisco and Ms. Gemma Belarmino of IRRI. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

370 Paris et al Agricultural commercialization and land-use intensification: a microeconomic analysis of uplands of northern Vietnam

Nguyen Tri Khiem, S. Pandey, and Nguyen Huu Hong

Low food security, high population growth, and environmental degradation are some of the major problems for agricultural intensification in the uplands of Vietnam. Although Vietnam has now become a major rice exporter, food production in these remote upland environments is insufficient to meet the needs of the country’s growing population. To increase farmers’ incomes and their access to food, the government of Vietnam has encouraged production of cash crops in uplands through investments in marketing infrastructure and institutional reform. Using farm-level survey data, this chapter examines the effects of improved access to markets and improvements in lowland productivity on land-use intensity, labor productivity, and food security in the uplands. The results indicate that these changes have reduced the intensification pressure on uplands and generally improved the food security of upland households.

Two-thirds of Vietnam’s natural area is classified as uplands, where 25 million people (one-third of the country’s population) are living. The uplands in Vietnam, as in other parts of the developing world, are characterized by heterogeneous and fragile ecosystems, a high incidence of poverty, severe deforestation, and soil degradation. The increased population pressure caused by natural population growth as well as migration of lowlanders has compounded these problems. A major factor affecting the upland systems of Vietnam in recent years is a shift in the outlook of the government toward these areas based on the recognition that upland systems are an important component of the overall economy. As a result, the government is undertaking additional investment to build rural infrastructure. In addition, policy and institutional reforms are being undertaken to improve the welfare of people in these uplands. For example, new kinds of policies on property rights such as stewardship contracts are being promoted to encourage more sustainable use of land at the forest margin. Policies to discourage shifting cultivation and forest clearing for upland rice cultivation (MARD 1998) are being introduced. These efforts

Agricultural commercialization and land-use intensification: . . . 371 have improved market access in the uplands, encouraged a diversification of upland production systems, and led to increasing commercialization of agriculture. The process of commercialization and diversification of the upland systems has also been facilitated by rapid increases in the productivity of lowland areas. In the late 1980s, Vietnam began a process of decollectivization, market reform, and trade liberalization in combination with investment in water control and promotion of short- duration high-yielding rice varieties. These reforms stimulated a rapid expansion in rice production from 1986 to 1989 and Vietnam has now become the third largest rice-exporting country (Khiem and Pingali 1995, Minot and Goletti 1998). Improve- ments in food grain productivity in the lowlands have encouraged diversification in the uplands as food needs are increasingly being met from the lowlands, thus releas- ing upland areas for more profitable uses, especially where marketing facilities are also better. These changes in the lowland rice economy and the impact of new policy initiatives being undertaken to develop upland areas may increase or reduce the pressure for intensification of upland system use. It is not possible to predict a priori the nature of adjustments in the upland systems that these changes might trigger. Some studies have indicated that policy changes in the form of land allocation and more stable land tenure in the northern mountain region have led to an increase in crop yields and reforestation of formerly barren hills (Dovonan et al 1997, Tachibana et al 1998). An increase in lowland productivity can be expected to reduce the pressure for food production in the uplands (Coxhead and Jayasuriya 1994, Tachibana et al 1998). But it may also increase land-use intensity by encouraging cash crop production and may lead to further exploitation of marginal lands (Barbier and Bergeron 1998, Hardaker et al 1993). There is evidence that as farmers in Vietnam have substituted high-yielding maize for upland rice in more favorable upland areas (Sikor and Truong 1998), upland rice production has been pushed to the more marginal land. In addition, in areas where marketing institutions are not well developed, a shift to nonfood crops can increase the vulnerability of small farmers because of the uncertainty associated with the price of cash crops (Dewey 1981). On the other hand, commercial production can increase total household income and trigger further multiplier effects by encouraging the adoption of improved technologies (von Braun and Kennedy 1994). The manner in which upland households respond to improvements in market access and to changes in economic opportunities is a critical factor in determining the merits of agricultural policies. The present study aims at examining the impact of changes in market access and population pressure on food supply at the household level, cropping intensity in uplands, the extent of cash crop production, the composition of household income, food security, and the productivity of land and labor. The analysis is based on an intensive survey of 980 households in 33 communes of 6 provinces in the northern mountainous highlands during the crop years of 1997-98 and 1998-99.

372 Khiem et al Conceptual framework Population pressure has been considered to be a major factor leading to intensification of agriculture (Boserup 1965, 1981). Although the initial response to a population increase may be to expand the area, the closure of land frontiers ultimately will force more intensive use of land as households attempt to satisfy their food needs from the shrinking land base per capita. This type of intensification will reduce labor productivity as fallow periods are reduced and more and more labor is applied to a given landholding. This situation can lead to a downward spiral of population growth- intensification-poverty unless technological change arrests a further decline in labor productivity or a massive emigration reduces the population pressure. Intensification may also result from an expansion of marketing opportunities. The additional demand created by marketing opportunities initially provides incentives for an expansion of area and ultimately for more intensified land use. Labor supply is less likely to be a constraining factor as improved access to markets reduces the cost of labor-substituting technologies such as mechanical tillage and herbicides. Better access to inputs such as fertilizers also improves the returns to land as well as labor, thus further reinforcing the intensification process. The downward spiral resulting from population-driven intensification can be avoided when intensification is market-driven. Theories of agricultural intensification and induced institutional innovation (Boserup 1965, Ruthenberg 1980, Hayami and Ruttan 1985) and agricultural household behavior in the context of incomplete markets (de Janvry et al 1991) provide conceptual frameworks to study the impacts of commercialization (Pender et al 1998, von Braun et al 1991, Barbier 1998). The effects of commercialization and policy measures on rural household income and consumption are mediated through complex relationships. An expected increase in income and production capacity will motivate households to enter the exchange economy and become more commercialized. The most important exogenous determinants of commercialization are population change, availability of new technologies, improved seeds or agronomic practices, investment in infrastructure, macroeconomic policies, wages and employment opportunities, and direct government action (Fig. 1). Household responses to population growth and commercialization are reflected in their decisions on natural resource management and household resource allocation, that is, mainly family labor, including intensification use of labor and capital per unit of land, reduction of fallow periods, adoption of technologies reflected in land-use patterns, and expenditures on food and nonfood products. Because of the complexity of modeling the interrelationships within rural households and limited data, reduced-form models are used in this study to evaluate the effects of population pressure and commercialization on land-use intensification, product choice, labor productivity, and food sufficiency.

Agricultural commercialization and land-use intensification: . . . 373 Population, Household resource, endowment demographic change (land, labor, capital)

Technologies, new crops Resource allocation Infrastructure Prices, wages, risks

Macroeconomic policies, government Home goods Off-farm Agricultural investment, programs production work production

Commercialization effects

Marketed Cash income surplus Subsistence food Nonfood expenditure Food consumption

Fig. 1. Determinants and consequences of commercialization at the household level. Recent trends of production systems in the northern uplands Because of their rotational pattern and the difficulty in collecting information, estimates of cultivated uplands vary widely among the sources. The National Institute for Agriculture Planning and Projection (NIAPP 1993) estimated that, out of 2.7 million ha of cultivated land, about 1.4 million ha currently comprise swidden fields. If the fallow area is included, the total area under rice-swidden has been estimated to be 3.5 million ha (Sam 1992) under shifting cultivation where upland rice is usually the first crop planted after slash-and-burn agriculture. Arraudeau and Xuan (1995) estimated total upland rice area in Vietnam at 0.45 million ha with the total area under rice- swidden at approximately 8 million ha. According to the World Bank (1995), the upland rice system in the northern mountain region comprises mainly sedentary shifting cultivators who stay in one place but shift cultivation sites, affecting an area of about 1 million ha. Another 0.2 million ha is under itinerant cultivation practiced by a few ethnic groups. Rapid population growth, driven by both high birth rates and in-migration of lowlanders, has brought about drastic changes in the economy of the northern uplands. Poverty, environmental degradation, increased pressure on resources, and social marginalization are interacting to create a downward spiral that is currently reaching crisis proportions (Jamieson et al 1998). The northern uplands experienced an increase in population of more than 300% between 1960 and 1984. The incidence of poverty is much higher in the uplands than in other agroecological regions. The eight poorest provinces are all located in the northern uplands (Minot 1998). The average incidence of poverty in the northern mountain region is 31% compared with the na-

374 Khiem et al tional average of 18%1, and most of these poor households are found among the ethnic groups (MARD 1998). Total food crop production in the region from 1985 to 1997 was barely adequate to meet the increased demand brought about by population growth. The per capita staple food crop output of the region remains at 250 kg despite a slight increase in rice, cassava, and sweet potato output. Investment in small-scale irrigation in some lowland fields helped expand wetland rice area by 1.3% per annum and yield increased 2% per annum from 1985 to 1997. The most remarkable change during the more recent years has been the expansion of area planted to improved maize or hybrid maize from 150,000 to 250,000 ha (Fig. 2). There was a rapid increase in area planted to fruit trees and industrial crops; area growth rates of these crops were 8.3% and 2.5% per annum, respectively, from 1985 to 1997. The climate of the northern mountain region is suitable for a wide range of valuable fruits such as apricot, persim- mon, plum, jujube, tangerine, lychee, and others. A more favorable land tenure policy and improvement in access to markets have also encouraged farmers to plant fruit

170

160 227,000 ha Rice 150 Horticultural crops and fruit trees Industrial crops 82,000 ha 140 Maize

130 59,000 ha 120

110

100 824,000 ha

70 1985 1987 1989 1991 1993 1995 1997 Year

Fig. 2. Index of areas planted to various crops in the northern mountain region (1985 = 100). Horticultural crops include vegetables and beans. Industrial crops include sugarcane, groundnut, soybean, tea, coffee, and rubber.

1A poor household is defined as one having a monthly per capita income equivalent to 13 kg of milled rice or lower (about US$50 per annum).

Agricultural commercialization and land-use intensification: . . . 375 trees and other perennial crops. However, price fluctuations and unstable markets caused by undeveloped postharvest facilities and marketing institutions are preventing the stable development of fruit trees in the region. In many cases, farmers have been forced to cut down one type of tree in order to grow another tree species that produces more marketable products or they found the harvesting cost higher than the market price of the product (Tuyen 1995, Dang 1993). Although the area planted to upland rice is declining, it is the main food crop for the millions of poor people and the ethnic minorities. Upland rice is grown alone or in diverse mixtures in shifting or permanent fields under a wide range of conditions of climate, slope, and soil type. Upland rice area during 1980-85 expanded rapidly because of food scarcity as the productivity of lowland rice was low. However, after the decollectivization of agriculture, the area of upland rice has declined, especially since 1990. According to official statistics, the total area planted to upland rice in the northern mountain region is reported to be between 100,000 and 120,000 ha. Current state policy, however, is to discourage shifting cultivation and limit the areas open for upland rice. Therefore, the reported upland rice area tends to be severely biased downward. On the other hand, conducted surveys show that villagers failed to report fields in remote areas or decreased the area under cultivation that is subject to taxation. The use of remote imagery in a selected commune in Son La Province found that the actual worked area is much larger than the reported area (Sikor and Truong 1998).

Ethnicity and production systems in the northern mountainous region The production systems in the northern uplands are generally determined by the different agroecological conditions and cultural and food preferences of the diverse ethnic groups. There are 31 ethnic groups belonging to seven language groups living in the northern uplands. The six largest groups are the Tay (1 million), the Thai and the Nung (0.6 million each), and the H’Mong, Muong, and Dzao (0.5 million each). Sev- enteen groups have populations under 10,000. Each group usually has its own dis- tinctive customs and traditions, socioeconomic characteristics, and community structures. However, many ethnic groups are also quite diverse internally. For example, although sharing a common language, the H’Mong are divided into several distinct subgroups (e.g., Red H’Mong, Black H’Mong, Flowered H’Mong). Many ethnic groups live intermixed with one another within the same delimited territory. Of the 109 districts and towns in the northern mountain provinces, 59 districts have ten or more ethnic groups. Residential segregation by ethnic group is common only at the hamlet level (Khong Dien 1996, Vien 1997). The H’Mong live on high slopes and mountain ridges, usually at altitudes above 800 m. They cultivate maize or monocropped rice on swidden fields in combination with wet rice on terraces. The H’Mong of Meo Vac and Dong Van in Ha Giang Prov- ince inhabit high-altitude cold regions, and rely only on maize monocropping. The sequence of crops after slash-and-burn agriculture commonly practiced by the H’Mong is sticky rice–nonsticky rice–maize–barley–Job’s tears (Coix lacryma) or cassava.

376 Khiem et al The Dzao live in medium-altitude areas; practice rotation of upland rice, maize, and cassava; and have experience in forestry and traditional garden crops. They practice swidden cultivation by clearing fields from scrub vegetation and cultivating for 2 to 4 years, followed by a fallow period of 10 years or more. They are specialists in constructing terraces on upland slopes for paddy rice. The Thai, Tay, Nung, and Muong live in low areas, intermontane valleys, and river basins. They practice intensive wet rice agriculture. In growing food crops for their own consumption, the Thai prefer sticky rice, whereas the Tay and Nung prefer nonsticky rice. The Nung, Tay, Giay, and others inhabit the lowland areas, at the base of hills and in valley bottoms; paddy rice is their main crop. They have a remarkable irrigation technology for paddy production in terraced land; paddy rice output, however, is not sufficient to satisfy food requirements and they need to plant root crops and maize on slopes to supplement their diet.

Survey design and the data set Using the a priori information on the upland systems of the northern mountain region, a stratified sampling design was carried out to generate a data set covering a wide range of population density, market access, ethnicity, relative proportion of upland and lowland areas, and the extent of crop diversification. In the first step, 12 districts in five mountain provinces were selected. The second step consisted of selecting in each district two to three communes that differ in ethnicity and degree of market access. Communes were classified as having good or poor market access by contrasting with other communes in the same district based on their relative degree of access to the main transportation route (provincial and national roads) and district or provincial center markets in terms of physical distance in kilometers, accessibility by vehicles, and existence of a local market. In the final step, 30 to 50 households in each commune were randomly selected for structured interviews. In total, 980 households in 33 communes were included in the survey. Comparative analysis was carried out at the household, commune, and district levels. Figure 3 shows the selected sites. Table 1 shows the distribution of selected sites and their corresponding demographics, crop diversification, physical distance to paved roads and district markets, and market access characteristics. Table 2 summarizes the general characteristics of households in areas with poor and good market access. The average farm size is 1.72 and 1.32 ha for the two groups of households, respectively. Households in locations with good market access are generally found at lower altitude, cultivate land with a lower slope, and have more lowland field area per capita. On average, households in these areas have 0.31 ha of lowland compared with 0.22 ha in the poor market access area. The extent of irrigation is greater in the low-slope areas. As a result of irrigation, cropping intensity of lowlands in these areas is higher than in lowlands of the upper slopes where only one crop of rainfed rice per year is planted. In market-accessible areas, the major cash crops are improved maize and horticultural crops. The proportionate area devoted to cash crops is usually higher in communes with better access to markets than other-

Agricultural commercialization and land-use intensification: . . . 377 Ha Giang Cao Bang Lao Cal Lai Chau Tuyen Quang Lang Sori Yen Bal Bac Thai Quan Ninh Vinh Phu Ha Bac Sorr La Pia Noi Ha TaiHal Hung Hoa Binh Thai Binh Nim Binh Nam Ho Thanh Hoa

Nghe An

Ho Tinh

Fig. 3. Selected sites for socioeconomic studies in the northern mountain region. wise. For example, 32% of the upland area is planted to cash crops (fruit trees, sugarcane, beans, peanuts, medicinal crops) in areas with good market access, whereas they occupy only 13% of the upland area in locations with poor market access. Despite a smaller landholding, the rice output per household in areas with better access to markets is 32% higher, indicating a more favorable environment for rice production in these areas. A higher proportion of the lowland in total landholding, a higher yield of lowland rice, and a higher intensity of rice production in the lowland account for this difference. For areas with poor market access, households are much more dependent on upland rice. Their upland rice output accounts for about 50% of total rice production, while this percentage for households in areas with good market access is only 16% (Table 2). Cropping systems prevailing in the study area are extremely labor-intensive where upland rice, lowland rice, and maize are the main food crops, occupying more than 70% of total cultivable uplands. Table 3 summarizes the total labor input per hectare for upland rice, lowland rice, and maize, which is further grouped into local and improved varieties. On average, farmers devote about 345, 344, and 250 person- days per hectare to the cultivation of upland rice, lowland rice, and maize, respectively. Land preparation, weeding, and harvesting account for almost 95% of these labor inputs. Weeding labor occupies the largest share in the planting of upland rice and maize, and accounts for about 40% of total labor input. Average returns to land and labor in the production of upland rice, lowland rice, local maize, and hybrid maize are also presented in Table 3. Returns to both land and labor are highest for lowland rice and improved maize. Because of high returns to

378 Khiem et al Dominant

Low Thai

continued on next page

a a

2 Yes Low H’mong

2 Yes High Thai

5 Yes/no

5 No Low H’mong

2 Yes High Thai

1 Yes High Thai

7 No Low H’mong

4 Yes High H’mong

2 20 No Low H’mong

5 60 No Low Dzao

5 20 No Low H’mong

5 10 Yes High Dzao

2 60 Yes High Tay

8 10 No Low H’mong

5 60 No Low H’mong

) (km) road (km)

–2

39 10 30 No Low Dzao

47 45

34 50 50 No Low H’mong

27 70 15 No Low Thai

43 10

39 30 20 No Low Thai

50 12

202 15

Sampled Population Distance Distance

communes density to district to Access by Market ethnic

households) km

(no. of sampled (persons market provincial vehicle access group

Ban Cong (33) 15 Hat Luu (34)

Che Cu Nha (34)

Kim Noi (34)

Muong Giang (12)

Co Ma (12) Chieng Khoang (13) 143 30 20 Yes High Thai

Chieng Chung (13) 48 40 12 Yes/no

Ta Hoc (12) Ta Phieng Pan (12)

Hong Ngai (33) 38 Phien Ban (34) 77

Yen Phong (33)Yen 50 Yen Phu (34) Yen

Tram TauTram (33) Tau Tram

Ngan Son Thuan Mang (38)

Mai Son Chieng Mung (13) 164 15

Bac Yen Xua (33) Ta

(ha)

Upland

Table 1. Selected districts and communes and sampled households. Table

Cao Bang 3,000 Nguyen Binh Kim (39) Tam

Son La 28,000 Thuan Chau Chieng Pha (13) 290 12

Ha Giang 3,000 Bac Me Cuong (33) Yen 31

Yen BaiYen 5,000 Mu C. Chai Mo De (33)

Province rice area District

Agricultural commercialization and land-use intensification: . . . 379 Dominant

High H’mong

8 Yes/no

6 No Low H’mong

1 Yes High Thai

6 10 No Low H’mong

) (km) road (km)

–2

20 9 12 No Low Dzao

72 10

Sampled Population Distance Distance

communes density to district to Access by Market ethnic

households) km

(no. of sampled (persons market provincial vehicle access group

Na Hoi (50) 215 5 5 Yes High Tay

Vinh Yen (40)Vinh Yen 62 3 2 Yes High Tay Nghia Do (30) 116 15 1 Yes High Tay

Sung Phai (30) 53

Nam Loong (40) 48

Pu Nhung (33) 47 15 Phing Sang (34) 44 25 20 No Low H’mong

Bao Yen Son (30) Yen 101 18 3 Yes High Dzao

Tuan GiaoTuan Quai Nua (33)

(ha)

Upland

Accessible only by two-wheel vehicles up to village center.

Table 1 continued. Table

Lao Cai 5,000 Bac Ha Bao Nhai (50) 76 15 20 No Low Dzao

Lai Chau 25,000 Phong Tho La Nhi Than (30) Province rice area District a

380 Khiem et al Table 2. General characteristics of the sampled households (1998-99).

Item Poor market access Good market access

Number of sampled households 598 382 Average household size 7.84 (0.15)a 6.81 (0.13) Annual average per household (kg) Rice production 1,323 (45.34) 1,533 (52.29) Lowland rice 660 (37.18) 1,289 (50.13) Upland rice 663 (24.12) 244 (15.43) Rice purchase 275 (13.99) 248 (16.97) Average distance from markets (km) 30 10 Per capita rice supplyb (kg) 210 (4.42) 261 (5.76) Average farm size (ha) 1.72 (0.07) 1.32 (0.08) Lowland 0.22 (0.02) 0.31 (0.01) Upland 1.50 (0.06) 1.01 (0.07) Lowland rice area/lowland area 1.14 1.40 Upland rice area/upland area 0.42 0.27 Upland maize area/upland area 0.29 0.26 Garden and orchard area/upland area 0.13 0.32 aNumbers in parentheses are standard errors. bSum of rice production and net purchase.

Table 3. Total labor use, yield, and returns to land and labor for upland rice, lowland rice, and maize.

Item Upland Lowland Local Improved or rice rice maize hybrid maize

Sample size 697 672 507 207 Total person-days ha–1 345 344 250 281 Labor time allocation (%) Land preparation 27 26 39 27 Seeding/crop 9 19 8 11 establishment Weeding 42 16 30 35 Fertilizer application 1 8 – 2 Pest control – 1 – – Irrigation – 9 – – Harvesting 20 21 22 26 Yield (t ha–1) 1.56 3.15 1.50 3.30 Average returns to Land ($ ha–1) 197 393 133 282 Labor ($ person-day–1) 0.57 1.10 0.65 1.00

Agricultural commercialization and land-use intensification: . . . 381 land, lowland fields are almost invariably planted to rice. Although returns to land are higher for upland rice than for local maize, the high labor requirement for upland rice, especially for weeding, lowers the returns for labor. Both labor and land productivity are higher for hybrid maize than for upland rice. The higher productivity of hybrid maize has induced a rapid expansion of its area. Upland rice is usually the crop planted in the first one or two years after slash- and-burn agriculture, followed by maize and cassava when the soil becomes less fertile. While wet rice from lowland fields becomes the main source of rice supply, farmers usually maintain supplementary upland rice on swidden fields, mostly sticky rice, to satisfy household needs. In the more favorable upland environment with good market access and where farmers have sufficient access to lowland rice cultivation, there is an increasing trend toward substitution of hybrid or improved open-polli- nated variety maize for upland rice. Most of the hybrid maize output is used in the household as feed for domestic animals and a small portion is sold to markets. Rice is the main food crop of the uplands. Although production of maize, especially improved maize, is mainly used for domestic animals, some ethnic minorities still depend on maize as their main diet. Other staples such as cassava, sweet potato, and canna (Canna edulis, a root crop) are consumed when rice and maize are in short supply. Most of the households in the study areas reported that they were unable to meet the family food requirement from their own production of rice and maize. On average, 88% of rice and maize consumption is from own production. This proportion of subsistence food production is almost equal among the two groups of households in the poor and good market access areas. The frequencies of reported food shortage are presented in Figure 4. On average, the extent of food shortage is greater in areas with poor access to markets (2.4 mo) than in areas with good market access (2.1 mo). The chi-square test for the difference in the two distributions gave a value of 37.17 (9 degrees of freedom), which is statistically significant at 1%. It is expected that households in areas with good market access having more diversified sources of income would depend more on the

Percentage of households 35 30 Poor market access Good market access 25 20 15 10 5 0 0123456789 Number of months of food shortage Fig. 4. Distribution of households reporting food shortage by number of months of food shortage in 1997-98.

382 Khiem et al markets for their food needs. With the marketing system still being undeveloped, however, the day-to-day concern of households is how to produce sufficient food. Figure 5 presents the relationship between the number of reported months of food shortage in the crop year 1997-98 and the average landholding per capita of the households reporting a food shortage. The incidence of food shortage seems to depend on the per capita availability of both uplands and lowlands. Households that reported a food shortage in more than 6 mo a year have half the size of both lowland fields and upland fields in comparison with those that suffered a food shortage only 1 mo or none (Table 4). About 5% and 8% of the sampled households reported having more than 6 mo of food shortage in poor and good market access areas, respectively. The

Landholding Percentage of households (m2 capita–1)

Percentage of households 30 Upland holdings 1,600 Lowland holdings 1,400 25 1,200 20 1,000 15 800 600 10 400 5 200 0 0 012345678 Number of months of food shortage

Fig. 5. Relationship between the number of reported months of food shortage in 1998 and the average of landholding per capita of the households reporting a food shortage.

Table 4. Landholding, annual cash income, and rice purchase of households facing food shortage of selected frequencies in 1997-98.

Poor market access Good market access

Months of reported More than 1 mo or More than 1 mo or food shortage 6 mo none 6 mo none

Percentage of householdsa 532 846 Upland area (ha) 0.56 1.22 0.29 0.53 Lowland area (ha) 0.14 0.27 0.15 0.37 Cash income (US$ household–1) 69 192 383 276 Rice production (kg household–1) 506 90 491 63 Rice purchase (kg household–1) 706 1,829 508 2,112 aThe percentages are calculated separately for the two groups of households, 598 in poor market access areas and 382 in good market access areas. Only households that reported a food shortage in more than 6 mo or in only 1 mo or none during crop year 1997-98 are included in the table.

Agricultural commercialization and land-use intensification: . . . 383 Table 5. Cash and noncash income per household and source of cash income by market access.

Poor Good Item market market access access

Total income (US$) 517 638 Share of cash income (%) 35 45 Source of cash income (US$) Animals 72 128 Rice 4 4 Maize 5 13 Other crops 29 30 Off-farm work 28 74 Forest products 6 6 Garden and orchard products 4 29 Noncash income (US$) 369 354 cash income of this group of households in poor market access areas, however, is much lower than that of their counterparts in the good market areas ($69 versus $383). They are the most food-insecure group of upland households. The average cash income derived from sales of home garden and orchard products of households in the good market access areas is seven times higher than that of households in the poor market access areas (Table 5). The former’s total cash income is about 1.5 times higher, whereas noncash income of the two groups is almost equal. Overall, total annual income per household averaged at the district level varies from $340 in Bac Me District to $800 in Thuan Chau District. Contrary to the widely held belief that the upland system in the northern mountain region is highly subsistence- oriented, income derived from exchange with markets contributes to 30–60% of total household income. The sale of domestic animals is the highest single source of cash income, contributing to about 40% of cash income for both groups of households.

Hypotheses and model specification Rice is the major staple crop in Vietnam. For household consumption purposes, rice from lowlands and uplands is an almost perfect substitute. Thus, an improvement in the productivity of lowland paddy can be expected to reduce the pressure for intensification of food production in the uplands. Empirical testing of this hypothesis is done by parameter estimation of a reduced-form model in which intensification is specified as a function of population density, the proportion of lowland area, the average slope of upland fields, and market access. The model is estimated by ordinary least squares. Following Ruthenberg (1980), the intensification index is calculated as the ratio of growing period in years to the sum of the growing period and the fallow period. Farmers responsed to rising population density and unavailable land for fur-

384 Khiem et al ther “extensification” in the uplands by reducing fallow periods and prolonging the cropping cycles of food crop production, which resulted in a higher intensification index. Thus, in calculating the index, the land used for annual food crop production only is considered. Upland fields with higher slope can be considered to be of poorer quality due to their greater susceptibility to soil erosion. Dummy variables for different ethnic groups were specified to examine the effect of ethnicity on the dependent variable. In the model, a positive coefficient of the population density variable would support the hypothesis of population pressure-driven intensification. The expected sign of the ratio of lowland to total landholding is negative as food production in the lowlands substitutes for food production in the uplands. The coefficient of slope is expected to be negative due to the land quality constraint to intensification. An improved access to markets is expected to reduce the pressure to intensify uplands for food production as income generated from commercial crops can be used to purchase food. The second model attempts to explain the variations in the importance of upland rice across households. The importance of upland rice is measured by the ratio of upland rice area to upland area. Since upland rice is mainly a subsistence crop, this ratio is also a proxy for the degree of subsistence orientation in the use of uplands. Other variables are as for the first regression. As with the first regression, population density is expected to have a positive effect and the proportion of lowland area and better access to markets are expected to have negative effects on the dependent variable. Since the dependent variable is censored, the model is estimated by Tobit regression. In the third model, labor productivity in upland agriculture is specified to depend on farm size, household capital, labor endowments, and market access. Except for the market dummy, all the variables in the model are expressed in logarithm. The model is estimated using ordinary least squares. Following the Boserupian argument, an increase in population pressure (or a reduction in farm size per capita) is expected to reduce labor productivity, ceteris paribus. Thus, the expected sign of the coefficient of farm size, which is a proxy for population pressure, is positive. Household capital endowment is also expected to have a positive effect on labor productivity. Improved market access is similarly expected to have a positive effect by encouraging a switch to high-value cash crops. The fourth model attempts to determine the factors explaining the frequencies of food shortage, which is defined as the number of months in a year that the household encountered a short supply of food from its own production. The number of months of food shortage in any particular year depends on several factors (Pandey and Minh 1998). First, production may not be adequate due to adverse weather conditions. Second, the incidence of food shortage depends on family size, which determines both demand for food as well as labor-supplying capacity. The demand for food grain depends on total family size, whereas labor-supplying capacity depends on the number of adult family members. Thus, the incidence of food shortage is likely to be higher among households that have a lower proportion of adults. Third, the area

Agricultural commercialization and land-use intensification: . . . 385 of land operated and soil quality determine production potential. Households with smaller land areas or poorer soils are more likely to suffer from food shortage. Since the dependent variable is censored, the model is estimated by Tobit regression.

Model results The regression results are presented in Table 6. In the first model, all coefficients except that of slope are statistically significant. The positive coefficient associated with population density supports the Boserupian hypothesis. The results also indicate that improved access to markets and increased productivity of lowlands can reduce the pressure for intensification of uplands. Thus, investments in research to improve

Table 6. Estimation of the models.a

Land-use Proportion Labor Reported Variable intensity of upland productivity months of indexrice area in uplands food shortage

Sample size 710 980 960 980 Constant 0.71 0.36 1.08 0.77 Population density 0.14** 0.04 1.37** Household size 0.08** Farm size 0.32** Upland area –0.36** Lowland area –1.58** Proportion of lowland –0.20** –0.32** Average upland slope –0.12 –0.16 Capital 0.13** Market access dummy –0.17** –0.01 0.36** –1.00** H’mong ethnic dummy –0.23** 0.13** 0.81** Dao ethnic dummy –0.09** 0.22** 0.85** Thai ethnic dummy –0.14** 0.20** 0.26 R-square 0.29 0.12 Pseudo R-square 0.16 0.14 F-value 40.08** 40.93** Log-likelihood function –623.2 –1,890.7 % censored 20.6 27.1

aModels 1 and 3 are estimated by ordinary least squares. Models 2 and 4 are estimated by Tobit regression. *, ** denote statistical significance at 5% and 1% level, respectively. The dependent variables are Model 1: Land-use intensity index defined as number of years of cropping of food crops divided by the sum of years of cropping of food crops and years of fallow period. Model 2: Proportion of upland area planted to upland rice. Model 3: Logarithm of gross returns per day in thousand dong of family labor used for agricultural and home goods production. Model 4: Months of reported food shortage in 1997-98. Variables are defined as population density = population density of communes in number of persons per ha; household size = number of persons; farm size = logarithm of farm size measured in ha; area of lowland and area of upland are measured in ha; proportion of lowland = proportion of lowland area in total landholding; average upland slope = average of slope of all upland parcels cultivated by the household measured in hundredth degrees; market access = dummy variable, poor market access = 0, good market access = 1; capital = logarithm of total value of farm tools and animal stock per adult person in million dong; base ethnic dummy = the Tay.

386 Khiem et al the productivity of lowland rice can have a positive environmental impact by reducing the incentive to intensify food production in the fragile uplands. Improvements in access to markets can generate a similar effect through income enhancement. Nega- tive statistically significant coefficients of ethnic dummies reflect location-specific effects. Compared with the H’Mong, Dao, and Thai, the Tay, who live in lower areas and are used as a base dummy, practice more intensive cropping on their upland plots. In the second model, the proportion of lowland area and the ethnic variables are found to have a statistically significant effect in determining the proportionate area allocated to upland rice. This result supports the hypothesis that an improvement in the productivity of lowland paddy will reduce pressure to produce rice in the uplands. In doing so, it will free up uplands for other crops that may be cash crops. Whether such a switch will have a positive environmental effect is uncertain since it will depend on several other factors such as the types of cash crops chosen and the institutional arrangements for growing such crops. If upland rice is substituted by more erosive annual cash crops, the environmental effect could be more detrimental. In fact, this seems to be the case in some parts of Vietnam where upland rice is being replaced by maize. As farmers plow the fields more intensively for maize than for upland rice, soil erosion problems in the maize fields, which also have poorer canopy cover, may increase. The third model indicates that labor productivity in upland agriculture is positively related to farm size (which is inversely related to population pressure). The parameter estimates suggest that an increase in population pressure (i.e., a reduction in farm size per capita) by 1% will reduce labor productivity by 0.32%. Thus, at the current rate of population growth of 3% per year in the uplands of Vietnam, labor productivity will decline by 1% per year. Thus, productivity improvement at the rate of 1% per year is needed just to maintain labor productivity. The signs of all other variables are as expected. In the fourth model, access to markets and farm size are found to have statistically significant negative effects on the frequencies of food shortage. A higher absolute value of the coefficient associated with the size of lowland holdings reflects the higher food productivity of the lowlands in food production. Investment in improvement of lowland rice productivity therefore helps enhance the status of food security of upland households. Population density as a proxy for land scarcity and family size has a significant positive effect on the frequency of food shortage. The incidence of food shortage depends on family size, which determines both demand for food as well as labor-supplying capacity. The positive and statistically significant effect of family size on incidence of food shortage implies a stronger demand effect for food grain than the labor-supplying capacity effect.

Conclusions and policy implications Improvement in market access and policy reforms for land tenure affect the allocation of land-use patterns in the northern uplands. Secondary data show that areas planted to fruit trees and horticultural crops have increased dramatically in the past

Agricultural commercialization and land-use intensification: . . . 387 ten years, whereas the increase in area planted to rice and other staple food crops has been just enough to maintain the region’s level of per capita food output. The results of the study indicate the important role that access to markets can play in arresting and reversing the Boserupian decline in labor productivity as population-driven intensification of land use occurs. Both land and labor productivity were higher in areas with better market access. The improvement in productivity and income resulted mostly from an expansion in cash crop production. Improvements in market access also reduced the need to intensify food production in the uplands. These factors may have also generated positive environmental benefits. These positive effects of market access, however, probably would not have materialized unless the productivity of lowlands increased to improve the food security of farmers. Unlike in other countries, Vietnamese upland farmers also have some lowland fields in river basins and intermontane valleys. A rapid improvement in the productivity of lowland fields as a result of policy changes during the mid-1980s was instrumental in relaxing food supply constraints to the diversification of land use in the uplands. In fact, upland rice area, which had expanded rapidly during the 1970s when lowland rice productivity was low, started to decline after the mid-1980s. Im- provements in market access alone without changes in the productivity of rice in the lowlands would probably not have resulted in more commercial production in the uplands. These results highlight the importance of taking measures to assure food security as a prerequisite for a move toward more commercialized production systems in the Asian uplands. In other countries and regions where farmers do not have access to lowlands to secure their food supplies, additional food production must come either from improvements in the productivity of uplands or through stable marketing channels. Since marketing institutions in upland areas of most Asian countries are poorly developed, improvements in the yield of food crops in uplands through agricultural research and policy reform are needed to encourage changes in land use toward income-generating activities.

References Arraudeau M, Xuan VT. 1995. Opportunities for upland rice research in Vietnam. In: Denning G, Xuan VT, editors. Vietnam and IRRI: a partnership in rice research. Manila (Philip- pines): International Rice Research Institute and Hanoi (Vietnam): Ministry of Agricul- ture and Food Industry. p 191-198. Barbier B. 1998. Impact of market and population pressure on production, incomes and natural resources in the dryland savannas of West Africa: bioeconomic modeling at the village level. EPTD Discussion Paper No. 21. Washington, D.C. (USA): International Food Policy Research Institute. Barbier B, Bergeron G. 1998. Natural resource management in the hillsides of Honduras: bioeconomic modeling at the micro-watershed level. EPTD Discussion Paper No. 32. Washington, D.C. (USA): International Food Policy Research Institute. Boserup E. 1965. The conditions of agricultural growth. New York (USA): Aldine Publishing Co.

388 Khiem et al Boserup E. 1981. Population and technology. Oxford (UK): Blackwell. Coxhead I, Jayasuriya S. 1994. Technical change in agriculture and land degradation in developing countries. Land Econ. 70:20-37. Dang BV. 1993. Economic and cultural changes in the northern mountain provinces. Hanoi (Vietnam): Social Science Publisher. de Janvry A, Fafchamps M, Sadoulet E. 1991. Peasant household behavior with missing market: some paradoxes explained. Econ. J. 101:1400-1417. Dewey KG. 1981. Nutritional consequences of the transformation from subsistence to commercial agriculture in Tabasco, Mexico. Human Ecol. 9:151-187. Dovonan D, Rambo AT, Fox J, Cuc LT, Vien TD, editors. 1997. Development trends in Vietnam’s northern mountain region. Hanoi (Vietnam): East-West Center and Center for Natural Resources and Environmental Studies. Hardaker JB, Fleming E, Tin HN. 1993. Economic aspects of environmentally endangered upland farming systems in the Asia-Pacific Region. In: Proceedings of the Workshop on Upland Agriculture in Asia, 6-8 April 1993, Bogor, Indonesia. Hayami Y, Ruttan VW. 1985. Agricultural development: an international perspective. Balti- more, Md. (USA): The Johns Hopkins University Press. Jamieson N, Cuc LT, Rambo AT. 1998. The development crisis in Vietnam’s mountains. Hono- lulu, Haw. (USA): East-West Center Special Report (6). Khiem NT, Pingali PL. 1995. Supply responses of rice and three food crops in Vietnam. In: Denning G, Xuan VT, editors. Vietnam and IRRI: a partnership in rice research. Manila (Philippines): International Rice Research Institute and Hanoi (Vietnam): Ministry of Agriculture and Food Industry. p 275-290. Khong Dien. 1996. Socio-economic characters of the ethnic minorities in the northern mountain region. Hanoi (Vietnam): Social Science Publishing House. (In Vietnamese.) MARD (Ministry of Agriculture and Rural Development). 1998. Socioeconomic development strategies for the Northern Mountain Region 2000-2010. Hanoi (Vietnam): MARD. 35 p. Minot N. 1998. Generating disaggregated poverty map: an application to Vietnam. MSSD Discussion Paper No. 25. Washington, D.C. (USA): International Food Policy Research Institute. Minot N, Goletti F. 1998. Export liberalization and household welfare: the case of rice in Viet- nam. Am. J. Agric. Econ. 80:738-749. NIAPP (National Institute for Agricultural Planning and Projections). 1993. Bare lands in Viet- nam. Hanoi (Vietnam): NIAPP. Pandey S, Minh DV. 1998. A socio-economic analysis of rice production systems in the uplands of northern Vietnam. Agric. Ecosyst. Environ. 1373:1-10. Pender J, Place F, Ehui S. 1998. Strategies for sustainable agricultural development in the East African highlands. Paper presented at the International Conference on Strategies for Poverty Alleviation and Sustainable Resource Management in the Fragile Lands of Sub- Saharan Africa. Entebbe, Uganda. Ruthenberg H. 1980. Farming systems in the tropics. 3rd edition. Clarendon (UK): Oxford Press. Sam DD. 1992. National background paper on shifting agriculture in Vietnam presented at the Workshop on “Shifting agriculture in Laos and Vietnam, its social, economic and environment values to alternative land uses,” Chiang Mai, Thailand, August 1992.

Agricultural commercialization and land-use intensification: . . . 389 Sikor T, Truong DM. 1998. Sticky rice, collective fields: community-based development among the Black Thai. Hanoi (Vietnam): Center for Natural Resources and Environmental Stud- ies. Tachibana T, Trung NM, Otsuka K. 1998. From deforestation to reforestation through tenure reforms: the case of the Northern Hill region of Vietnam. Tuyen BC. 1995. Community-based natural resources management in Lam Dong province. In: Rambo AT et al, editors. The challenges of highland development in Vietnam. Honolulu, Haw. (USA): East-West Center. Vien TD. 1997. Ethnic culture and farming systems in northern Viet Nam. UNESCO Work- shop on “Cultural aspects of natural resources management.” Hanoi, Vietnam. von Braun J, de Haen H, Blanken J. 1991. Commercialization of agriculture under population pressure: effects on production, consumption, and nutrition in Rwanda. Research Re- port No. 85. Washington, D.C. (USA): International Food Policy Research Institute. von Braun J, Kennedy E. 1994. Agricultural commercialization, economic development and nutrition. Baltimore, Md. (USA): Johns Hopkins University Press. World Bank. 1995. Vietnam: environmental program and policy priorities for a socialist economy in transition. Agricultural and Environment Operations Division. Washington, D.C. (USA): The World Bank.

Notes Authors’ addresses: Nguyen Tri Khiem, Can Tho University, Vietnam, and Social Sciences Division, International Rice Research Institute; S. Pandey, Social Sciences Division, International Rice Research Institute; Nguyen Huu Hong, Thai Nguyen University, Viet- nam. Acknowledgments: The authors wish to acknowledge comments and generous support from Professor Hermann Weibel of Hanover University. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

390 Khiem et al Economics of intensive rainfed lowland rice-based cropping systems in northwest Luzon, Philippines

M.P. Lucas, S. Pandey, R.A. Villano, D.R. Culannay, and T.F. Marcos

Farming in the rainfed lowlands of Ilocos Norte, Philippines, is highly intensive, diversified, and commercialized. The cropping system is predominantly rice-based in the wet season and high-value cash crops are grown during the dry season. The profitability of cash crop production has encouraged farmers to use high levels of purchased inputs such as chemical fertilizers and pesticides. Concerns are being raised about the long-run sustainability of such intensive systems. This chapter assesses the sustainability of such systems using a total factor productivity analysis. The trend in total factor productivity was positive (1992-95) initially but then became negative (1996-97). However, the total factor productivity estimates for six years do not show any clear negative trend. Groundwater pollu-

tion, particularly with NO3-N, has occurred as a result of the excessive use of fertilizers on dry-season crops. If the effect of this negative externality were also captured in the total factor productivity estimates, the decline for recent years could have been sharper. Although fluctuations in total factor productivity within the short period analyzed may be attributed mainly to climatic factors, technological interventions for improving input-use efficiency are nevertheless needed to reduce both the input cost and the contamination of groundwater.

Farming in the rainfed lowlands of northwest Luzon, Philippines, is highly intensive, diversified, and commercialized. As in other rainfed environments, weather conditions are erratic with interspersed dry spells. The cropping system is predominantly rice-based in the wet season (WS) from May to October and high-value crops are grown during the dry season (DS) from November to April. Garlic (Allium sativum L.), maize (Zea mays L.), mungbean (Vigna radiata L. Wilczek), sweet pepper (Cap- sicum annum L. var. annum), and tomato (Lycopersicon esculentum L.) are among the most common DS crops grown after rice (Fig. 1). Some farmers maximize land- use intensity by growing two or three cash crops. These crops depend on irrigation water from tube wells.

Economics of intensive rainfed lowland rice-based cropping systems . . . 391 Rainfall (mm) 600

Crops planted after

500 Other

Tobacco 400 Onion

300 Sweet pepper

Tomato 200 Mungbean

Maize 100 Rice Garlic 0 May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Month Fig. 1. Cropping calendar and average monthly rainfall in Ilocos Norte, Philippines. Source: MMSU-PAGASA-PCARRD Agromet Station, MMSU Batac, Ilocos Norte, 1976-98.

The rainfed lowlands of Ilocos Norte demonstrate a case of high cropping intensity and input use and can serve as a model for other rainfed areas that are being intensified. Farmers in Ilocos Norte use high levels of fertilizer and pesticides because of the economic benefits derived from high-value crops. These high-input intensive systems, however, may be unsustainable in the long run. This chapter presents an economic analysis of intensive rice-based production systems in Ilocos Norte, the northernmost province in the Philippines. The possible sustainability or unsustainability of the system is also addressed. The province of Ilocos Norte has a total land area of 362,000 ha, of which 25% is devoted to agriculture and forestry. There are 39,000 ha of rainfed areas, of which 66% is rainfed lowlands. Agriculture and forestry directly employ 47% of the labor force. In 1998, the province reached a sufficiency level of 256% for rice, 145% for yellow maize, 124% for root crops, 2,277% for garlic, 3,463% for onion, and 170% for mungbean. In 1998, rice area declined slightly due to natural calamities such as El Niño and La Niña and two super typhoons that passed through the province. In spite of these, however, the province still produced a surplus of rice. The cropping intensity increased from 130% in 1995 to 151% in 1998. The cropping intensity is expected to increase further to 180% by 2001. This increase is expected to be achieved mainly through irrigation development projects.

Trends in area planted and yield of rice and nonrice crops The area planted to rainfed rice in Ilocos Norte decreased in the past three years (Fig. 2), whereas the area planted to irrigated rice has generally increased. The decrease in

392 Lucas et al Area (ha) 50,000

40,000 Irrigated

30,000

20,000

10,000 Rainfed

0 1991 1993 1995 1997 1992 1994 1996 1998 Year

Fig. 2. Area planted to rice in 1991-98, Ilocos Norte, Philippines. Source: Bureau of Agricultural Statistics (1998).

Yield (t ha–1) 4.0 3.5 Irrigated 3.0 2.5 Rainfed 2.0 1.5 0 1991 1993 1995 1997 1992 1994 1996 1998 Year Fig. 3. Average yield of rice in 1991-98, Ilocos Norte, Philippines. Source: Bureau of Agricultural Statistics (1998). area under rainfed rice was attributed to the prolonged drought experienced in the province. The area under irrigated rice increased, however, with the expansion of irrigation facilities. The yield of irrigated rice increased rapidly with a growth rate of 6.3% per year, which was higher than that of rainfed rice (Fig. 3). The area planted to dry-season crops in the province varied widely over time (Fig. 4). Maize area increased rapidly between 1994 and 1996 and stabilized. Garlic and mungbean areas increased from 1996 to 1998, while the area planted to sweet pepper remained highly variable. Tomato area remained stable because most of the farmers are contract growers obtaining a fixed quota from the National Food Corpo- ration tomato paste processing plant. Except for tomato and sweet pepper, yields of dry-season crops did not vary much over time (Fig. 5).

Economics of intensive rainfed lowland rice-based cropping systems . . . 393 Area index 350 Maize 300 Garlic Mungbean 250 Sweet pepper Tomato 200

150

100

0 1991 1993 1995 1997 1992 1994 1996 1998 Year Fig. 4. Index of area planted to cash crops in Ilocos Norte, Philippines, 1991-98.

Yield (t ha–1) 18 16 Maize 14 Garlic 12 Mungbean Sweet pepper 10 Tomato 8 6 4 2 0 1991 1993 1995 1997 1992 1994 1996 1998 Year Fig. 5. Average yield of major dry-season crops in rainfed areas in Ilocos Norte, Philippines, 1991-98. Source: Bureau of Agricultural Statis- tics (1998).

Methodology Study area and sampling design Ilocos Norte is divided into four major regions: northern coastal, central lowlands, southern coastal, and eastern interior (PPDO 1995). Most of the agricultural activities are in the central lowlands composed of ten municipalities: Bacarra, Laoag City, San Nicolas, Dingras, Batac, Paoay, Sarrat, Currimao, Badoc, and Pinili. The monitoring of rice and nonrice production practices in farmers’ fields began in the 1991 WS and continued until the 1998 DS. Panel data from 100 randomly selected farmers were collected.

394 Lucas et al The economic characterization and initial sustainability analysis were already presented in an earlier publication (Lucas et al 1999). This chapter updates the earlier paper by including more recent data that cover the period from the 1994 WS to 1998 DS. The analysis on the possible sustainability or unsustainability of the cropping system, however, is based on the data for 1992 to 1997.

Measuring sustainability Sustainability is defined, for the purpose of this chapter, as an improvement in the productive performance of a system without depleting the natural resource base upon which future performance depends (Pandey and Hardaker 1995). Unsustainability may result from on-site and/or off-site effects of agricultural land use. On-site effects include adverse changes in the physical, chemical, and biological properties of the soil-water-plant complex that reduce farm productivity. For example, in intensified irrigated rice systems of tropical Asia, reduced availability of nutrients to plants because of changes in soil properties could lead to unsustainability (Cassman and Pingali 1995). Off-site effects, which are also called externalities, refer to those effects that are not normally valued in the market place. Common examples are adverse health effects of groundwater contamination and pesticide use, and damage to irrigation infrastructure from soil erosion. Two commonly used economic indicators of sustainability are trends in partial factor and total factor productivities. Partial factor productivity is defined as the average productivity of a factor of production. Total factor productivity (TFP) is defined as the ratio of the aggregate quantity of all outputs produced within a given time period (usually a year) to the aggregate quantity of all inputs applied during the same time period. Suitable weights based on prices or output and input shares are used for aggregating physical quantities of various outputs and inputs. Economists consider TFP to be a more meaningful concept than partial factor productivity for assessing sustainability (Lynam and Herdt 1989, Harrington 1993). As all inputs and outputs are accounted for, a declining trend in TFP is an indicator of possible degradation of the resource base, or unsustainability. Although the definition requires the inclusion of all inputs and outputs, data limitations and valuation problems mean that only those inputs and outputs that can be easily measured and valued are generally included. The externalities, such as environmental pollution, which are difficult to value, are often excluded from TFP calculations. Similarly, changes in prices of inputs and outputs can affect TFP values over time, despite the use of methods that attempt to correct for such price effects (Rayner and Welham 1995). Despite some of these practical limitations to calculating TFP, the trend in TFP (not its level) is considered to be a useful indicator of (un)sustainability and has been widely used (Capalbo and Antle 1988). Following the method suggested by Rayner and Welham (1995), we used the Tornqvist-Theil method to calculate the TFP index. The Tornqvist-Theil method is considered to be theoretically superior to other methods since it is consistent with a flexible production function that does not arbitrarily constrain the substitution possibilities between inputs. The input index I(X)t is computed as

Economics of intensive rainfed lowland rice-based cropping systems . . . 395 Σ I(X)t = I(X)t – 1 exp[1/2 (sit + si, t – 1) ln xit – ln xi, t – 1)] (1) i

where xit = quantity of input i in period t, sit = share of input i in total cost in period t;

witxit s = , i,k = 1, ... n, and (2) it Σ wktxkt wit, wkt = actual prices of inputs in period t. Similarly, the output index is computed as

I(Q) = I(Q) exp[1/2Σ (r + r ) ln q – ln q )] (3) t t – 1 j jt j, t – 1 jt j, t – 1

where qjt = quantity of output j in period t, rjt = share of output j in total revenue in period t,

pjtqjt r = ; i,j = 1, ... m, and (4) jt Σ pitqit pjt, pit = actual prices of outputs in period t. Finally, the TFP index is obtained as the ratio of I(Q)t /I(X)t. The computation of these indices follows the following procedure. For each year, all crop outputs produced and all inputs used were included in the calculation. Analysis was limited to the period 1992-97 as the data for 1991 and 1998 were incomplete. A combination of province-level and farm-level data was used in the analysis as farm-level data for 1992-94 did not include all outputs for all farmers. On the output side, provincial-level data on yield and farm-gate prices for 1992-97 were used. For the cost of production, data for individual crops for each year were not available at the provincial level. Hence, farm-level data were used for this purpose. Indices of the value of outputs and inputs were subsequently obtained using equations (1) and (3).

Results and discussion Cropping patterns Rice is grown in most of the farmlands during the WS, except for small areas in the higher toposequence that are planted to vegetables such as beans, eggplant, and tomato. Most of the farmers plant modern rice varieties. In the 1996 and 1997 WS, IR64 and PSB Rc14 were the most popular varieties planted, occupying 24% and 25% of the rice area, respectively (Table 1). BPI Ri10 remained a common variety among farmers. Other IR varieties covered almost 25% of the total area planted in the last two years. During the DS, farmers plant high-value cash crops such as garlic, onion, tomato, sweet pepper, and tobacco. These crops are entirely dependent on irrigation from shallow tube wells. Vegetables such as eggplant, bitter gourd, bottle gourd, squash,

396 Lucas et al Table 1. Rice varieties planted (1994-97), Ilocos Norte, Philippines.

Variety 1994 1995 1996 1997 % area

IR64 18 26 24 15 BPI Ri10 30 27 22 20 PSB Rc14 – 2 13 26 Other IR varietiesa 45 28 23 25 Other PSB Rc varietiesb 2 11 15 10 Other varietiesc 5 6 3 4

aOther IR varieties include IR36, IR42, IR58, IR60, IR66, IR68, and IR78. bOther PSB Rc varieties are PSB Rc8, PSB Rc10, PSB Rc12, PSB Rc18, PSB Rc22, PSB Rc28, and PSB Rc34. cOther varieties include UPL Ri 4, UPL Ri 5, c4-137, C22, and glutinous. and cowpea are also planted in smaller parcels. Some farmers can even grow a third crop.

Economics of rice production Rice is planted predominantly during the wet season. There is an increasing use of modern rice varieties. Traditional rice varieties are rarely found. Farming activities begin immediately after rains occur. However, farmers usually experience unexpected dry periods after a seemingly continued downpour. This forces them to resort to supplemental irrigation. Land preparation is becoming more mechanized using hand tractors. This has decreased the labor requirement for land preparation by 25%. Water buffaloes, however, are still a common sight during land preparation. Rice plants are established by transplanting, with the average seeding rate being 101 kg ha–1 (Table 2). Farmers still use more than the recommended seeding rate of 40 kg ha–1. The average fertilizer rate was 129-31-21 kg NPK ha–1. Fertilizers were applied in two splits—2 wk after transplanting and 5 wk after transplanting. Basal application was seldom practiced. The rate of application of insecticides and herbicides was low. Labor use for crop management is generally low and accounts for only about 14% of the total labor input. The average grain yield of rice from 1996 to 1997 decreased by 16% compared with the mean yield for 1994 to 1995 (Table 3). The decrease in yield is attributed to the El Niño phenomenon. The average returns above cash costs per hectare were $450. About 40% of the total production cost was accounted for by material inputs. Despite a decrease in total labor ha–1, labor cost increased because of a higher wage rate.

Economics of high-value cash crops Various dry-season cash crops are planted after rice. Over the years, maize occupied almost one-third of the total area in rainfed areas. Tobacco covered more area (7%) than tomato (5%) and sweet pepper (3%) from 1996 to 1998. This may be attributed to the increasing economic benefit from tobacco unlike in the past years. Area de-

Economics of intensive rainfed lowland rice-based cropping systems . . . 397 Table 2. Average material and labor inputs for rice (1994-97), Ilocos Norte, Philippines.

Categories 1994 1995 1996 1997 All years

Material inputs Seed (kg ha–1) 111 (76) 107 (68) 103 (53) 93 (42) 101 (57) Nitrogen (kg ha–1) 155 (87) 143 (99) 121 (41) 116 (39) 129 (64) Phosphorus (kg ha–1) 30 (25) 39 (38) 30 (22) 30 (19) 31 (25) Potassium (kg ha–1) 19 (15) 22 (20) 19 (18) 25 (19) 21 (18) Insecticide (kg ai ha–1) 0.03 (0.09) 0.05 (0.11) 0.09 (0.25) 0.03 (0.09) 0.05 (0.17) Herbicide (kg ai ha–1) 0.01 (0.08) 0.07 (0.17) 0.12 (0.22) – 0.05 (0.15) Fuel for land preparation 7 (28) 6 (14) 6 (21) 31 (29) 14 (27)

Labor inputs (person-days ha–1) Land preparation 8 (9) 8 (7) 7 (6) 6 (6) 7 (7) Crop establishment 28 (21) 41 (30) 35 (22) 26 (18) 32 (23) Crop managementa 6 (6) 12 (14) 8 (10) 18 (21) 11 (13) Harvesting and threshing 38 (32) 45 (29) 37 (20) 26 (19) 35 (25) Total labor 80 (49) 106 (57) 87 (58) 76 (43) 85 (47)

aIncludes fertilizer, chemical application, and weeding; ai = active ingredient. Numbers in parentheses are standard deviations.

Table 3. Average grain yield, costs, and returns for rice (1994-97), Ilocos Norte, Philippines.

Categories 1994 1995 1996 1997 All years

Yield (t ha–1) 3.95 (2.4) 3.56 (1.4) 3.30 (1.3) 2.88 (1.2) 3.31 (1.5)

Material inputs (US$ ha–1) Seed 24 (16) 23 (15) 29 (15) 25 (12) 26 (14) Fertilizer 107 (58) 128 (72) 76 (27) 74 (26) 89 (48) Insecticide 6 (12) 7 (13) 3 (5) 3 (6) 4 (9) Herbicide 0.42 (2) 1.2 (3) 0.10 (1) 0.30 (1) Power/fuela 36 (47) 39 (36) 24 (30) 46 (27) 36 (35)

Labor inputs (US$ ha–1) Land preparation 24 (26) 24 (23) 20 (17) 44 (15) 33 (19) Crop establishment 59 (41) 82 (85) 94 (60) 94 (50) 103 (60) Crop managementb 16 (18) 33 (40) 28 (33) 5 (32) 10 (32) Harvesting and threshing 106 (89) 40 (87) 100 (53) 33 (52) 42 (70) Total material costs 174 (95) 198 (78) 131 (45) 147 (46) 155 (66) (US$ ha–1) Total labor costs (US$ ha–1) 206 (122) 179 (123) 243 (110) 177 (98) 197 (114) Total costs (US$ ha–1) 380 (190) 377 (137) 352 (127) 324 (120) 351 (140) Gross returns (US$ ha–1) 998 (604) 899 (341) 569 (229) 549 (229) 688 (384) Returns above paid-out 855 (563) 799 (351) 277 (229) 249 (386) 450 (458) costs (US$ ha–1) Net returns (US$ ha–1)c 618 (526) 522 (334) 194 (231) 225 (222) 336 (356) aFor land preparation and irrigation. bIncludes fertilizer, chemical application, and weeding. cNet of cash cost and imputed cost of family labor; 1US$ = P25. Numbers in parentheses are standard deviations.

398 Lucas et al Table 4. Yield and input use of major dry-season crops (1994-98), Ilocos Norte, Philippines.

Categories Maize Garlic Mungbean Sweet pepper Tomato

No. of fields 184 233 135 54 89 Yield (t ha–1) 3.15 (2.9) 0.84 (0.73) 0.41 (0.37) 6.0 (4.8) 33 (19.5)

Material inputs Seed (kg ha–1) 21 (15) 264 (119) 31 (20) 1.3 (1) 0.70 (0.59) Nitrogen (kg ha–1) 102 (73) 136 (71) 6 (30) 305 (138) 126 (51) Phosphorus (kg ha–1) 23 (25) 49 (35) 2 (9) 85 (69) 67 (42) Potassium (kg ha–1) 27 (30) 41 (40) 1.3 (8) 78 (69) 111 (73) Insecticide (kg ai ha–1) 0.09 (0.25) 0.21 (0.44) 0.18 (0.39) 1.5 (1.3) 0.58 (0.65) Fungicide (kg ai ha–1) 0.02 (0.15) 0.80 (1.6) 0.10 (0.37) 2.10 (3) 1.71 (2) Herbicide (kg ai ha–1) – 0.08 (0.44) – 0.07 (0.48) – Fuel (L ha–1)a 16 (20) 32 (32) 16 (19) 81 (52 ) 32 (23)

Labor inputs (person-days ha–1) Land preparation 3 (4) – 3 (6) 3 (6) 4 (7) Crop establishmentb 14 (19) 20 (20) 6 (5) 43 (42) 26 (20) Crop managementc 4 (4) 15 (13) 2 (3) 9 (9) 6 (5) Irrigation 4 (5) 6 (4) 3 (3) 13 (13) 4 (4) Weeding 1.3 (2) 10 (13) 1 (5) 14 (15) 2 (5) Harvesting and threshing 14 (16) 12 (14) 17 (22) 23 (30) 30 (30) Total labor 40 (35) 63 (39) 32 (37) 105 (65) 73 (50) aFor land preparation and irrigation. bIncluding mulching in garlic. cIncludes fertilizer, chemical application, and weeding. Numbers in parentheses are standard deviations. voted to sweet pepper remained the lowest. Fertilizer rates applied to dry-season crops usually exceeded the recommended rates. The average fertilizer rate applied to sweet pepper was 305-85-78 kg NPK ha–1 and for garlic was 136-49-41 kg NPK ha–1 (Table 4). The recommended rates are 170-57-163 kg NPK ha–1 for sweet pepper and 90-26- 50 kg NPK ha–1 for garlic. However, from 1996 to 1998, about a 30% decline and 10% decline in fertilizer application to sweet pepper and garlic were observed, respectively. These declines could be partly due to the increasing awareness of farmers regarding apparent pollution of groundwater. Some degree of consciousness may have occurred after information generated from several research studies in sweet pepper- growing areas started to spread (e.g., Gumtang et al 1999, Shrestha and Ladha 1998, Tripathi et al 1997). However, pesticides were applied at high rates. For example, sweet pepper is sprayed with pesticides weekly. Material inputs account for 40% to 50% of the total production costs in dry- season crops (Table 5). The high marginal profitability of sweet pepper, tomato, and garlic may have encouraged farmers to apply higher doses of inputs. Economic returns from dry-season crops showed that sweet pepper was the most profitable, followed by tomato and garlic with net returns of $1,096, $861, and $842 ha–1, respectively. High yields from sweet pepper and its higher price are factors that contributed to higher returns. Low returns from garlic are attributed to the dramatic drop in farm- gate prices (Fig. 6) due to the poor quality of output in 1997.

Economics of intensive rainfed lowland rice-based cropping systems . . . 399 Table 5. Costs and returns of major dry-season crops (1994-98), Ilocos Norte, Philippines.

Categories Maize Garlic Mungbean Sweet pepper Tomato

Material inputs (US$) Seed 36 (29) 607 (343) 32 (27) 109 (93) 93 (77) Fertilizer 67 (51) 134 (109) 5 (32) 239 (149) 155 (76) Insecticide 7 (16) 16 (32) 10 (19) 110 (115) 77 (64) Fungicide 2 (13) 12 (23) 2 (5) 43 (63) 54 (64) Herbicide 1 (7) 1 (1) 1 (5) Power/fuela 54 (38) 27 (15) 52 (64) 95 (33) 65 (25)

Labor inputs (US$) Land preparation 18 (24) 17 (28) 23 (31) 23 (34) Crop establishment 20 (31) 60 (59) 21 (41) 53 (37) 53 (49) Crop management 6 (8) 7 (7) 1 (3) 26 (31) 18 (14) Irrigation 14 (15) 23 (21) 11 (36) 73 (99) 21 (21) Weeding 7 (7) 61 (70) 2 (15) 130 (134) 12 (18) Harvesting and threshing 69 (64) 49 (126) 67 (67) 124 (102) 146 (411) Total material costs (US$) 156 (77) 786 (404) 80 (58) 547 (236) 383 (181) Total labor costs (US$) 133 (95) 200 (156) 118 (105) 429 (211) 272 (411) Total costs (US$) 289 (124) 986 (450) 198 (134) 976 (343) 655 (504) Gross returns (US$) 654 (608) 1,828 (2,021) 310 (285) 2,057 (2,047) 1,517 (941) Net returns (US$)b 366 (577) 842 (1,962) 112 (294) 1,096 (2,134) 861 (936) aFor land preparation and irrigation. bNet of cash cost and imputed cost of family labor. 1US$ = P25. Numbers in parentheses are standard deviations.

Price index

Maize 250 Rice Mungbean 200 Sweet pepper Tomato Garlic 150

100

0 1991 1992 1993 1994 1995 1996 1997 1998 Year Fig. 6. Index of real farm-gate prices of major crops, Ilocos Norte, Philippines, 1991-98. Source: Bureau of Agricultural Statistics (1999).

400 Lucas et al TFP, output and input indices 200

160

120

80 Input index 40 Output index TFP 0 1992 1993 1994 1995 1996 1997

Fig. 7. Trend in total factor productivity (TFP), input indices, and output indices for Ilocos Norte, Philippines, 1992-97.

Trend in total factor productivity The index of input use generally increased but seemed to have leveled off in 1996 and 1997 (Fig. 7). The output index, which initially showed an increasing trend, declined after 1995. The decline in output index can be partly attributed to the prolonged drought and strong typhoons experienced in the province in the past three years. The decline in output may also be due partly to the decline in the yield of high-value crops such as tomato and garlic. The time period covered in the analysis is too short to detect any long-term trend in total factor productivity. The fluctuations observed in total factor productivity likely reflect mainly the climatic variation. Contamination of groundwater with NO3-N in intensively cultivated areas was observed. Results of groundwater monitoring showed that, in areas where rice-sweet pepper is practiced, NO3-N concentration was consistently high in both the WS and DS (Alibuyog et al 1999). A high NO3-N concentration was also found in areas planted to garlic, tobacco, and tomato. In addition to NO3-N contamination of groundwater, there were some indications of salinity intrusion because of excessive pumping. Elec- trical conductivity (EC) was observed to be generally high. Although a decline was observed at the start of the WS, the EC again increased toward the end of the DS. The increase in EC could be due to possible intrusion of saline water in the area as an excessive quantity of irrigation water was used during the DS (Alibuyog et al 1999). In sweet pepper, irrigation interval is one week with a total irrigation depth of 397 mm per season. Garlic is irrigated 4 to 5 times per season with a total irrigation depth of 235 mm, while tobacco, tomato, maize, and eggplant are irrigated 2 to 3 times per season with depths of 129, 214, 187, and 92 mm, respectively (Alibuyog et al 1999).

Conclusions The trend in total factor productivity was positive (1992-95) initially but became negative afterwards (1996-97). Indications of contamination of groundwater with NO3- N and excessive drawdown of groundwater were also apparent. If these negative externalities continue to increase, production systems in Ilocos Norte may not be sus-

Economics of intensive rainfed lowland rice-based cropping systems . . . 401 tainable in the future even though total factor productivity did not show any clear negative trend over the study period. Without appropriate interventions, the intensive cropping systems in Ilocos Norte may become unsustainable in the long run. A proper understanding of N and irrigation requirements of crops as well as timing of application is necessary to avoid excessive losses of N and to maintain sustainability and environmental quality (Bucao et al 1999). Farmers should be encouraged to consider other crops such as N catch crops (e.g., maize + indigo, mungbean) as alternatives to continuous rice-cash crop. The maize + indigo intercrop during the dry-to-wet transition has been effective in capturing residual soil N (Alam 1999). Extensive dissemination of information on the negative effects of high NO3-N concentration in the groundwater should be carried out. Research and extension programs that improve the quality of farm management practices so that nutrient-use efficiency is increased and input costs are reduced are needed to ensure the sustainability of the production system.

References Alam M. 1999. Increasing yield and nutrient use efficiency through improved fertilization and integrating of catch crop in a rice-vegetable cropping system. Unpublished PhD thesis, University of the Philippines Los Baños. Alibuyog NR, Bucao DS, Agustin EO, Tuong TP. 1999. Annual report. Rainfed Lowland Rice Research Consortium. Batac, Ilocos Norte (Philippines): Mariano Marcos State Univer- sity. Bucao DS, Gumtang RJ, Alibuyog NR, Agustin EO, Tuong TP, Obien SR. 1999. Crop diversification and intensification impact on groundwater resource. Paper presented during the Commodity Review held at CONDORA, Damortis, La Union, Philippines, 27-28 May 1999. Bureau of Agricultural Statistics. 1998. Production survey of rice and major dry season crops in the province of Ilocos Norte. Bureau of Agricultural Statistics, Laoag City, Ilocos Norte, Philippines. Bureau of Agricultural Statistics. 1999. Monthly average farm gate prices of selected agricultural commodities, Province of Ilocos Norte, 1991-1998. Laoag City, Ilocos Norte, Phil- ippines. Capalbo S, Antle JM. 1988. Agricultural productivity measurement and explanation. Washing- ton, D.C. (USA): Resources for the Future. Cassman KG, Pingali PL. 1995. Intensification of irrigated rice systems: learning from the past to meet future challenges. GeoJournal 35:299-305. Gumtang RJ, Pampolino MF, Tuong TP, Bucao D. 1999. Groundwater dynamics and quality under intensive cropping systems. Exp. Agric. 35:153-166. Harrington LW. 1993. Interpreting and measuring sustainability: issues and options. In: Harrington LW, Hobbs PR, Cassaday KA, editors. Methods of measuring sustainability through farmer monitoring: application to the rice-wheat cropping pattern in South Asia. Proceedings of the Workshop. Mexico: CIMMYT, IRRI and NARC. Lucas M, Pandey S, Villano R, Culannay D, Obien SR. 1999. Characterization and economic analysis of intensive cropping systems in rainfed lowlands of Ilocos Norte, Philippines. Exp. Agric. 35:211-224.

402 Lucas et al Lynam JK, Herdt RW. 1989. Sense and sustainability: sustainability as an objective in international agricultural research. Agric. Econ. 3:381-398. Pandey S, Hardaker JB. 1995. The role of modelling in the quest for sustainable farming systems. Agric. Syst. 47:439-450. PPDO (Provincial Planning Development Office). 1995. Comprehensive land use plan. Prov- ince of Ilocos Norte (Philippines): PPDO. Rayner AI, Welham SJ. 1995. Economics and statistical considerations in the measurement of total factor productivity (TFP). In: Barnett V, Payne R, Steiner R, editors. Agricultural sustainability: economic, environmental and statistical considerations. New York (USA): John Wiley. p 23-28. Shrestha RK, Ladha JK. 1998. Nitrate in groundwater and integration of a nitrogen catch crop in intensive rice-based cropping systems to reduce nitrate leaching. Soil Sci. Soc. Am. J. 62:1610-1619. Tripathi BP, Ladha JK, Pandey S. 1997. Economic feasibility, production potential and nitrogen behavior in intensively cultivated rice-based cropping systems in Northern Luzon, Philippines. Philipp. J. Crop Sci. 22:39-48.

Notes Authors’ addresses: M.P. Lucas, D.R. Culannay, T.F. Marcos, Mariano Marcos State Univer- sity, Batac 2906, Ilocos Norte, Philippines; S. Pandey, R.A. Villano, Social Sciences Division, International Rice Research Institute, DAPO Box 7777, Metro Manila, Philip- pines. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Economics of intensive rainfed lowland rice-based cropping systems . . . 403 Integrating biophysical and socioeconomic characterization 406 Garcia et al Socioeconomic and biophysical characterization of rainfed versus irrigated rice production in Myanmar

Y.T. Garcia, M. Hossain, and A.G. Garcia

Myanmar’s rainfed rice land covers 79% of its total cultivated rice area, which translates into about 4 million hectares, one of the largest in the world. Because of a lack of irrigation facilities, rice is generally grown once a year during the monsoon season. A study in the Ayeyarwardy Delta, a major rice- producing area in southern Myanmar, was conducted to characterize the different production systems of rice farmers. The study also investigated the operations of the output and input (i.e., land, labor, and capital) markets and how farmers’ access to these markets shaped the distribution of income and resources in the village economy. The biophysical and socioeconomic factors leading to the adoption of new rice-based technologies, such as summer rice, double monsoon rice, the use of high-yielding varieties, and rice-fish culture, were identified. Fur- thermore, biotic and abiotic factors that significantly constrained rainfed rice production were evaluated in terms of reduction in rice yield and productivity. Alternative management practices and policy options were proposed to help minimize the adverse effects of these constraints.

Myanmar, one of the last frontiers for increasing world rice production, is located in Southeast Asia. Thailand, Lao PDR, China, India, and Bangladesh surround its borders from east to west. Its land area of 68 million hectares spans 2,092 kilometers in length from north to south and about 925 kilometers from east to west at its widest parts. In 1997, the total population was 46 million, with 75% living in the rural areas. The country’s economy basically depends on agriculture, with about 15% of its area currently used for crop production. Unlike many countries in Asia, it has a large potential for increasing the cropping area by opening untapped culturable wastelands, which are approximately 12% of its total land area. Rice, the main staple food crop of the country, is grown extensively, covering 53% of the country’s total area sown to crops. Rice production accounts for 34% of the gross domestic product and 47% of the total agricultural exports. It is also one of the principal sources of foreign exchange, contributing 57% of the total export earnings.

Socioeconomic and biophysical characterization of rainfed . . . 407 Increased rice production had always been the government’s major thrust in crop production. Rice policies were geared toward meeting not only the local rice demand but also providing surplus production for export. In the early 1990s, the government launched a nationwide campaign to boost rice production. To achieve these objectives, the Ministry of Agriculture and Irrigation focused on the following strategies: (1) expand cultivated areas by opening new frontiers, (2) explore potential water resources for irrigation facilities, (3) intensify the use of modern rice varieties, and (4) adopt new cropping practices to enhance production, such as summer rice, double monsoon rice, and rice-fish culture. The success of these strategies depends very much on how farmers respond to the government’s policies. Farmer responses, however, depend to a large extent on the biophysical and socioeconomic factors that influence farmers’ adoption of the new rice-based technologies. Rice production can be increased if major biotic and abiotic constraints can be properly identified and appropriate technologies to solve them can be recommended. The main objectives of this chapter are (1) to document the new farming practices that evolved out of recent developments in the rice sector and (2) to identify the biophysical and socioeconomic opportunities for and constraints to the adoption of the new rice-based technologies, such as summer rice, double monsoon rice, the use of high-yielding varieties, and rice-fish culture. The chapter also investigates the operations of the output and input (i.e., land, labor, and capital) markets and their effects on the distribution of income and resources in the village economy. Based on the findings, the chapter proposes alternative management practices and policy options that could help alleviate these constraints.

Methodology, data sources, and surveys Most of the data at the national level used in this study were from the Myanmar Agriculture Service (MAS) of the Ministry of Agriculture and Irrigation. A survey was also conducted in 1996 to study the rural household economy and document farmers’ responses to the government’s rice production programs. The study covered four selected villages in Nyaungdong Township located in the Ayeyarwardy Delta. The Ayeyarwardy Delta, located in southern Myanmar, is the largest rice-producing region and contributes about 34% of the country’s total rice production. Approxi- mately 1.8 million hectares of its land are allocated to rice production with an average yield of 3.2 t ha–1. The villages were chosen according to the presence of new rice-based cropping systems practiced by farmers in the area, namely, (1) double rice cropping (monsoon rice followed by summer rice), (2) triple rice cropping (two rice crops in the monsoon season and one summer rice crop), (3) rice-fish farming, and (4) one rainfed rice cropping during the monsoon season. The sample was generated through total enumeration consisting of 739 households, of which 40% were landless. Only 22% of the total farm households were observed to adopt the new rice-based technologies promoted by the government, that is, double cropping with summer rice as a second crop

408 Garcia et al (13%), triple rice cropping (3%), and rice-fish farming (6%). Hence, the remaining 78% of the total farming households were all rainfed farms, which grew only one rice crop during the monsoon season. Despite the low percentage of adopters of the newly established rice-based technologies, their experiences provided the study with valuable insights regarding the incentives, constraints, and problems they faced in the process of adoption and adaptation of these technologies. Farmer interviews were carried out to identify biophysical and socioeconomic factors that led to the adoption of the new rice-based technologies. A statistical approach and cost and return analysis were used to characterize the impacts of recent developments on the village rice economy and farmers’ production systems. Like- wise, Gini ratios were estimated and Lorenz curves were fitted to describe the income distributions found in the sample villages representing the different rice ecosystems in the study area. To identify the major rice production constraints, data were collected through a series of surveys in 30 selected townships from five divisions and one state (Ayeyarwardy, Bago, Yangon, Sagaing, and Mandalay divisions, and Shan State) in upper, central, and lower Myanmar. These sites were selected since approximately 80% of the total rice crop in Myanmar is grown in these locations. Major biotic and abiotic factors that significantly constrained rice production were evaluated and ranked in terms of yield reduction brought about by these factors. Problems associated with these constraints were then prioritized, which allowed us to identify alternative management practices that can help minimize their adverse effects.

Recent developments in the Myanmar rice economy Sown area The crop area planted to rice was less than 5 million hectares from 1962 to 1990 (Table 1). Since 1991, however, area planted to rice grew annually by 4%, reaching 6 million hectares in 1995. Government efforts to expand and open new lands increased the net sown area by one million hectares over five years. In the crop year 1995-96, total rice area for the whole Union of Myanmar was 6 million hectares, wherein 4.83 million hectares were devoted to monsoon (wet-season) rice and 1.21 million hectares for summer (dry-season) rice. After 1996, summer rice area started to decline, reaching only 0.89 million hectares in 1998, thus decreasing the total rice sown area. The summer rice area decreased because of the limited supply of diesel fuel for irrigation pumps and commercial fertilizer (both related to foreign currency limitations), hence only the more suitable rice areas were targeted for planting.

Production and yield Rice production changed drastically from 1962 to 1995 (Table1). It increased from about 8 million metric tons in the 1960s to about 16 million metric tons in the early 1990s. Average rice yields also almost doubled from 1.6 to 3.0 t ha–1 within the period. The production growth rate was about 1% in the 1960s and reached 5% in the 1970s. The increase in rice yield was the major factor that induced production growth

Socioeconomic and biophysical characterization of rainfed . . . 409 Table 1. Total rice area, average yield, total production, and fertilizer consumption from 1962 to 1998 for the Union of Myanmar.

Total area Av yield Production Fertilizer Urea No. Year (million ha) (t ha–1) (million t) consumption consumption (t) (t)

1 1962-63 4.64 1.64 7.63 8,800 5,456 2 1963-64 4.86 1.59 7.75 8,841 5,865 3 1964-65 4.96 1.71 8.47 13,727 7,647 4 1965-66 4.83 1.66 8.02 6,962 4,187 5 1966-67 4.50 1.47 6.61 8,615 4,537 6 1967-68 4.69 1.65 7.73 12,245 7,971 7 1968-69 4.75 1.68 7.99 40,286 28,607 8 1969-70 4.65 1.71 7.95 26,179 18,926 9 1970-71 4.79 1.70 8.12 21,781 16,590 10 1971-72 4.75 1.71 8.14 47,197 29,782 11 1972-73 4.51 1.62 7.32 67,223 43,965 12 1973-74 4.86 1.76 8.56 68,905 47,202 13 1974-75 4.87 1.76 8.54 73,775 60,167 141975-76 5.01 1.83 9.16 86,959 65,255 15 1976-77 4.89 1.90 9.27 89,629 76,268 16 1977-78 4.85 1.94 9.42 105,891 85,659 17 1978-79 4.99 2.10 10.48 160,519 131,407 18 1979-80 4.43 2.35 10.40 173,531 129,840 19 1980-81 4.78 2.77 13.25 205,315 147,098 20 1981-82 4.79 2.94 14.08 225,732 169,134 21 1982-83 4.55 3.15 14.30 281,831 198,440 22 1983-84 4.64 3.06 14.22 331,969 228,842 23 1984-85 4.58 3.09 14.19 304,571 211,396 24 1985-86 4.64 3.07 14.25 324,972 229,041 25 1986-87 4.65 3.02 14.06 304,317 208,704 26 1987-88 4.47 3.04 13.57 202,815 162,390 27 1988-89 4.51 2.90 13.10 153,565 123,373 28 1989-90 4.72 2.91 13.74 131,888 102,307 29 1990-91 4.74 2.93 13.90 109,098 86,647 30 1991-92 4.56 2.88 13.14 99,802 77,025 31 1992-93 5.04 2.93 14.77 149,745 122,685 Monsoon rice 4.71 2.93 13.83 Summer rice 0.32 2.87 0.93 32 1993-94 5.47 3.05 16.68 248,423 194,672 Monsoon rice 4.65 2.97 13.81 Summer rice 0.82 3.51 2.87 33 1994-95 5.76 3.18 18.35 298,488 226,841 Monsoon rice 4.69 3.07 14.39 Summer rice 1.07 3.69 3.96 341995-96 6.04 2.94 17.77 305,109 199,690 Monsoon rice 4.83 2.84 13.71 Summer rice 1.21 3.36 4.06 35 1996-97 5.77 3.02 17.49 328,020 213,251 Monsoon rice 4.92 2.93 14.44 Summer rice 0.85 3.60 3.05 36 1997-98 5.44 3.07 16.74 149,922 120,048 Monsoon rice 4.55 2.97 13.53 Summer rice 0.89 3.60 3.21

Source: Myanma Agriculture Service (1997).

410 Garcia et al during this period, brought about by the widespread adoption of modern rice varieties and increased fertilizer use. In the 1980s, however, the growth rate plunged to less than 1%. This production decline was mainly due to (1) scarcity of fertilizer inputs, which was attributed to the weakening of the country’s economy since 1985, and (2) restricted access to foreign exchange, which severely hampered the importation of agricultural imports. The early 1990s, on the other hand, saw dramatic changes in production growth rates, which soared to an average of 12% annually. This was brought about by the government’s program on summer rice production, which started in 1992, resulting in a significant increase in the total rice output of the country.

Farming practices adopted by rice farmers Rainfed lowland rice-cropping system During the crop year 1997-98, the total area planted to rainfed rice was approximately 4.6 million hectares, which was estimated to be 79% of the total cropped area of the country. It was generally grown under several conditions: normal plains (68%), deepwater area (7%), and saline area (4%). Under the rainfed lowland ecosystem, only one rice crop was grown annually since production was totally dependent on rainfall as the main water source. Land preparation normally began during the onset of the rainy season, mostly from May to June when enough water had soaked the field. During field preparation, rice seeds were sown in seedbeds to grow seedlings. A majority of the rainfed lowland farmers used draft animals (a pair of bullocks) as a power source for land preparation and spent about 36 man-animal-days to finish one hectare. Transplanting was the most common method of crop establishment used by farmers, normally done in rows with the use of strings as guides. Farmers transplanted rice seedlings that were about 30 to 45 d old between June and July. Older rice seedlings that were taller were generally preferred by farmers, especially those who did not have water control in their fields. Taller seedlings have a higher survival rate during heavy rains when fields become totally submerged in water. Rice harvesting started in October to December. Farmers normally left the cut straws to dry in the field before threshing. Harvesting done during the late monsoon when rainfall was still high posed a big problem in grain drying. After the rice crop was harvested, the fields were left idle and became grazing grounds for livestock during the dry season. Threshing, on the other hand, was done mostly with animals or by renting a mechanical thresher and paying a fixed amount per unit volume of rice threshed.

Summer rice Rice planted between September and February and harvested between December and April was classified as the summer rice crop. As with rainfed rice, the most common method of crop establishment was transplanting. More and more farmers, however, were learning to use direct seeding (done through either broadcasting or line seeding) to reduce the time constraint brought about by double cropping and to maximize the

Socioeconomic and biophysical characterization of rainfed . . . 411 use of water by avoiding seedbed preparation for rice seedlings. Generally, farmers practicing summer rice cultivation planted high-yielding varieties and applied higher rates of fertilizers; hence, summer rice yields were often higher than those of the monsoon rice (3.6 vs 3.0 t ha–1) if the water supply was adequate. Large-scale promotion of summer rice cropping, which started in 1992, was considered as one of the biggest achievements of the government in boosting rice production in the country. With an initial area of 0.32 million hectares and total production of 0.93 million metric tons, summer rice production had peaked at 4.0 million tons in 1995, covering a total area of 1.2 million hectares (Table 1). In 1997, total sown area declined to 0.89 million hectares, thereby reducing total production to 3.2 million tons. The current target area for summer rice production is 1.6 million hectares with a total production of 5.0 million tons. To achieve this target, present irrigation facilities were being expanded since summer rice can only be planted where irrigation water is available. As an incentive for farmers to grow summer rice, the government waived the production quota (12 baskets acre–1 or 593 kg ha–1) that they needed to sell to the government’s cooperative for this season.

Two rice crops during the monsoon season Another innovation for increasing rice production in Myanmar was the introduction of two rice crops during the wet season, which was implemented in 1992. Rice grown from May to September was considered the first monsoon crop, whereas rice grown from September to January was considered the second monsoon crop. This production system allowed farmers to achieve three rice crops per year, that is, one summer rice and two monsoon rice crops. Growing two monsoon rice crops, however, was only possible under good water control and drainage in the field, especially during the height of the monsoon rains when flooding normally occurred. The common method of crop establishment in the first monsoon crop was transplanting rice seedlings in rows. For the second monsoon rice crop, planting was normally done immediately after harvesting the first rice crop. Because of time constraints, farmers either practiced direct seeding or purchased rice seedlings for transplanting since they had no time to grow their own seedlings. Harvesting the first monsoon rice crop was normally done between September and October, which often coincided with heavy rainfall such that farmers commonly experienced problems in threshing and grain drying. These postproduction problems resulted in low yield and poor grain quality. Higher grain yields, on the other hand, were normally reported during the second monsoon rice crop. Labor shortage, a common problem of rice farmers in Myanmar, was further aggravated under this technology since harvesting and other postproduction activities in the first monsoon crop coincided with land preparation for the second monsoon crop. Peak harvesting for the second rice crop was done in early December, which also overlapped with land preparation for the summer rice crop. Because of time and labor constraints, few farmers adopted triple cropping of rice.

412 Garcia et al Rice-fish farming A new technology introduced by the Ministry of Agriculture in the early 1990s was the rice-fish technology. It was implemented in the low-lying areas of the Ayeyarwardy Delta with an allocated area of 1,600 ha. Each rice-fish pond consisted of a 2-ha plot with 1.2 ha in the middle used for paddy production, 0.4 ha surrounding the paddy field as pond area to rear fish, and 0.4 ha for embankment. Catfish, common carp, and prawns were usually grown in the pond. Vegetables, flowers, banana, and perennial fruit trees (guava, papaya, citrus, drumstick, etc.) were planted in the embankment. The rice-fish technology was envisioned to enable double cropping of rice with water from the pond serving as irrigation during the dry summer months. The rice-fish ponds were constructed by the government and sold to interested parties who wished to practice the rice-fish technology, such as landless households with financial resources, retired government personnel, and key officials in the local government. Each pond was sold at US$500 in 1992. In most instances, the buyers of the ponds obtained a medium-term loan from the government banks payable in three years with an interest rate of about 1.5% per month or 18% per annum. The government targeted 120 ponds to be built in the study area for a period of 5 years. The site was called Kan Taya (hundred ponds), where 112 ponds were actually built. The number of ponds owned by operators ranged from 1 to 11 normally located side by side. The project became very popular in the pilot areas such that private ponds were voluntarily constructed and operated by local farmers. It was estimated that about 5,300 ha or 33% of the total rice-fish ponds in the country were constructed under this system. The rice-fish area increased rapidly from 1991, reaching maximum adoption in 1993. In 1995, the government phased out the construction of rice-fish ponds. Sev- eral reasons were cited for the phase-out of the project: the enormous cost (both implicit and explicit costs) of pond construction and the implementation of a new rule that allowed multinational companies to develop large tracts of land (as much as 400 ha) for rice production. As an incentive for these companies, 50% of their rice produce can be exported privately to generate export earnings needed to finance their importation of raw materials and agricultural inputs. The existing ponds are still operational but plagued with problems. During the monsoon season, heavy rain normally flooded the rice-fish area, which equalized the water level inside and outside the pond embankment. This made monsoon rice production and seeding of the pond with fingerlings impossible. Only when the water level receded could rice production be started, often late in the monsoon season. The bigger problem of the rice-fish technology lay in the fish culture. At the height of the monsoon rains, the seeded fingerlings escaped with the floodwater, which significantly reduced the fish population in the pond. This brought tremendous losses to the operators. Moreover, the cost of feed meal for the fish was often unaffordable for the small operators. Hence, the original technology of rice-fish culture was altered by the pond operators. Every monsoon season, a majority of the pond owners opened their embankments and waited for the nearby rivers to overflow in their ponds. The flood-

Socioeconomic and biophysical characterization of rainfed . . . 413 water often brought with it a variety of fish population that grew with minimum care and feeding. When the floodwater receded (in November to December), water from the ponds was pumped out to irrigate the rice fields in the middle of the plot for summer rice cultivation. This allowed the harvesting of the wild fish trapped in the ponds. Small operators find this practice viable since they can earn as much as $1,600 from the sale of the wild fish harvest, on top of their earnings from summer rice production. On the other hand, big-time operators of the rice-fish ponds altered the technology differently. They totally stopped rice production and concentrated on fish culture. Some actually dug up the middle plot for a larger pond area. Since the owners were originally businessmen and retired government officials, they found rice growing too tiresome and laborious. Instead, they simply seeded their ponds with fingerlings and diligently fed their fish stock until they were ready for harvest, which normally took about 1 1/2 to 2 years. Earnings from this practice ranged from $2,500 to $3,000 per pond, which were higher than the earnings from rice-fish culture per year with less labor involved.

Biophysical and socioeconomic characteristics of the village rice economy Climate and rainfall Myanmar has two distinct seasons: (1) the dry season, which lasts from mid-October to mid-May, and (2) the wet season from mid-May to mid-October. During the dry season, there is a cold spell from December to February, after which the warm weather begins. Humidity is normally high from April to December. In the Ayeyarwardy Delta, the temperature is relatively less variable during the different seasons, ranging from 16 to 37 °C compared with the northern parts of the country. Agriculture depends highly on the southwest monsoon, which often occurs from mid-May to mid-October. During this period, rainfall provides sufficient moisture for growing crops except in the dry regions of northern Myanmar. This period also serves as the sole source of water to replenish dams and reservoirs. The amount of precipitation, however, varies significantly depending on the location. Higher annual rainfall is recorded in the coastal and deltaic regions, whereas low annual rainfall is observed in the dry and mountain regions.

Topography and irrigation Vigorous public investments in irrigation infrastructure, such as the construction of small to large dams and installation of pumping stations along the banks of major waterways, increased the area under irrigation in all of Myanmar from 12% of the total cultivable land in 1992 to 18% in 1995 (Table 2). This significantly increased cropping intensity in the areas covered by irrigation development especially in the delta area. About 80% of the country’s total irrigated area was devoted to rice production. Despite the government efforts, however, about 79% of the total rice area was still planted to rainfed rice during the monsoon season when rainfall was the only water source.

414 Garcia et al Table 2. Total irrigated area from 1961 to 1995, Union of Myanmar.

Net sown area Irrigated Year (000 ha) area Percentage (000 ha)

1961-62 7,136 5348 1971-72 7,933 887 11 1981-82 8,383 1,040 12 1991-92 8,308 995 12 1992-93 8,683 1,106 13 1993-948,706 1,332 15 1994-95 8,962 1,639 18

Source: MAI (1995).

Table 3. Distribution of land area by source of irrigation, Nyaungdon Township, 1996.

Irrigated

Rainfed Lower terrace Upper terrace Deepwater Source of irrigation Total area % Total area % Total area % Total area % (ha) area (ha) area (ha) area (ha) area

Not irrigated 145 99 182 64 112 75 276 69 Pump from canals – – 91 31 23 16 48 12 Tube wells – ––––––– Creeks/streams/ 1 1 13 5 14 9 76 19 rivers Total 146 100 286 100 149 100 400 100

Topography in the study area was characterized by gentle rolling plains, which can be divided into three elevation zones: upper terrace, lower terrace, and deepwater areas. These topographical zones were used to classify the irrigated villages at the study site. About 25% to 36% of the total sample farms were covered by irrigation (Table 3). Water was generally drawn using lift pumps from irrigation canals, creeks, rivers, and streams. Some of the irrigated farms (12% to 31%) drew water by pumping from irrigation canals that were constructed manually by the community of farmers in the villages under the leadership of the village chairman.

Ownership and characteristics of land resources Table 4 presents the distribution of landholdings in the sample villages. The average landholding size ranged from 1.8 to 2.5 ha but many farming households (35% to 49%) owned less than 2 ha of land. On the other hand, the percentage of landless ranged from 19% to 51% of the total households. The lower percentage of landless in the deepwater villages was attributed to the existence of common lands that were characterized by open access. These marginal lands were located in the deepest por-

Socioeconomic and biophysical characterization of rainfed . . . 415 Table 4. Distribution of farm households by size of landholding, Nyaungdon Township, 1996.

Irrigated

Landholding (ha) Rainfed Lower Upper Deepwater terrace terrace (% of HHa)

No cultivated land 48 44 51 19 0.01–1.0 16 13 15 24 1.01–2.0 21 22 22 25 2.01–3.0 6 46 10 3.01–4.0 6 8 5 9 4.01–10.0 3 8 1 11 10.0 and above 0 1 0 2 Total 100 100 100 100 Average size of 1.77 2.53 1.71 2.48 holding (ha)

aHH = households.

Table 5. Distribution of farm households by tenure status, Nyaungdon Township, 1996.

Irrigated

Tenure status Rainfed Lower Upper Deepwater terrace terrace (% of HHa)

Owned farms 98 93 99 95 Tenant – 2 – 1 Leaseholder 2313 Rented out – 2 – 1 Total 100 100 100 100

aHH = households. tion of the village and were generally submerged throughout the monsoon season. During the dry season, however, residual moisture in the soil allowed cropping of rice, groundnuts, and pulses. About 93–98% of the farmers reported that they owned the land they farmed either legally or de facto as in the case of common lands (Table 5). Absolute land ownership in the country was vested in the state. Farmers, however, were granted the right to cultivate the land and reap its benefits. This right was supported by a certifi- cate that could be transferred from generation to generation. Since land ownership was not absolute, transferability through buying and selling was restricted. This ren- dered the market for agricultural lands nonfunctional. However, illicit buying and selling of land were reported to exist in the villages at the rate of $60 ha–1. A very

416 Garcia et al small percentage of the land (1–3%) was cultivated under a leasehold arrangement, mostly between families who lent and borrowed parcels of land for a given season. Land leasing and tenancy arrangements were not regularly practiced in the study area and elsewhere. During the height of the monsoon season, low-lying areas, that is, deepwater and lower terrace areas, were usually submerged. Floodwater normally rose to at least 90 cm, covering an area of 69% and 66%, respectively (Table 6). In the rainfed village, however, flooding was not a big problem. Only 35% of the land area was affected by flood, of which 25% fell under 30 cm of water. Flooding also occurred in the upper terrace areas, but with an even distribution of floodwater. Farmers’ assessments of soil quality conditions in their fields were solicited. Categories such as good, average, poor, and very poor were given as choices. Most of the farmers (82%) in the rainfed village claimed that soil quality on their farms was good (Table 7). In contrast, many of the farmers owning irrigated farms characterized their lands as having poor soil quality (52–66%). This was expected since extensive cropping of the field can easily exhaust soil fertility. On average, about 14% of the households in all four villages considered their land to be of average quality. Cases of very poor soil quality were reported in the deepwater area, but were of negligible proportion.

Table 6. Distribution of land area (in ha) by depth of flooding, Nyaungdon Township, 1996.

Irrigated

Depth of Rainfed Lower terrace Upper terrace Deepwater flooding Total area % Total area % Total area % Total area % (ha) area (ha) area (ha) area (ha) area

Not flooded 9465 21 7 57 38 4110 Up to 30 cm 37 25 8 3 26 17 12 3 30–90 cm 5 3 68 2431 21 74 18 >90 cm 11 7 190 66 36 24274 69

Table 7. Distribution of land area (in ha) by soil quality, Nyaungdon Township, 1996.

Irrigated

Rainfed Lower terrace Upper terrace Deepwater Soil quality Total area % Total area % Total area % Total area % (ha) area (ha) area (ha) area (ha) area

Good 120 82 51 18 53 35 97 24 Average 17 11 47 16 19 13 66 16 Poor 8 6 188 66 78 52 236 59 Very poor 1 1 – – – – 2 1

Socioeconomic and biophysical characterization of rainfed . . . 417 Soil types found in the villages were classified into sandy loam, clay loam, clay, and silty. Many of the households (50–59%) claimed that clay loam was the predominant soil type in the four villages (Table 8), followed by sandy loam (25– 35%). There were reports of clayey soil (15–23%) but mostly in the irrigated areas. Clayey soil was not conducive to rice production since the soil was deemed too hard for land preparation. This could have accounted for the farmers’ perception that soil quality was generally poor in the irrigated areas. Incidence of silty soil was reported only in the deepwater area, but was likewise negligible in proportion.

Rice varieties (traditional vs modern) The adoption of high-yielding modern varieties was another important strategy that the government actively pursued in conjunction with the introduction of intensive rice cultivation. Table 9 presents the distribution of farmers using modern and traditional varieties in the survey area during the monsoon and summer seasons. Among farmers who engaged in rice double cropping, that is, rice-fish and summer rice, the use of modern varieties was normally higher in the dry season (95%) than in the wet season (34%). The wet-season crop was normally planted with traditional varieties (66%). This was generally attributed to the geographical location of most summer rice fields, where excessive flooding occurred during the height of the monsoon season, hence the higher adoption of the traditional tall varieties. The traditional variet-

Table 8. Distribution of land area (in ha) by soil type, Nyaungdon Township, 1996.

Irrigated

Soil type Rainfed Lower terrace Upper terrace Deepwater

Total area % Total area % Total area % Total area % (ha) area (ha) area (ha) area (ha) area

Sandy loam 50 35 72 25 42 28 37 34 Clay loam 86 59 149 52 76 51 199 50 Clay 9 6 65 23 32 21 62 15 Silty – – – – – – 2 1

Table 9. Percent adoption of modern versus traditional varieties by farming practices and season, Nyaungdon Township, Myanmar, 1996.

Rice- Two Irrigated fish monsoon (double Variety Rainfed farming rice cropping) Average

WSa WS DS WS1 WS2 WS DS WS DS

Traditional 22 65 2 15 0 67 8 34 5 Modern 78 35 98 85 100 33 92 66 95

aWS = wet season, DS = dry season.

418 Garcia et al ies also produced more rice straws, which were often sold or used as livestock feed. The preference for traditional varieties was further reinforced by their desirable qualities: better eating quality, higher rice volume when cooked, and higher price in the market. Harvests from traditional varieties were normally kept for home consumption or as a ready source of cash. On the other hand, farmers who practiced triple rice cropping used more modern varieties in all three seasons because of the short turn-around time needed to make three croppings possible. Hence, shorter-duration varieties (a characteristic of modern varieties) were more preferred. Likewise, rainfed farmers planted more modern varieties (78%) than traditional varieties to maximize yield for the only season they grew rice. Farmers used a wide range of varieties in the study area for both traditional and HYVs (16 and 19 varieties, respectively).

Farm mechanization Because of the rapid expansion in cultivated area and increased potential for double cropping brought about by irrigation infrastructure development, farm mechanization became an integral part of the government’s strategy for boosting agricultural production. Labor shortage was one of the biggest problems of farmers, especially during planting and harvest, when labor demand peaked. To solve this problem, the Agricultural Mechanization Department (AMD) put up 25 tractor stations in selected townships all over the country where needed machines, such as tractors, power tillers, high-lift pumps, threshers, seeders, and harvesters, were supplied to farmers at a fixed rent. The government also encouraged private hiring of farm machinery. To boost local participation in the marketing of farm machinery, private traders were allowed to import farm machinery at zero tariff, which made the price of farm machines effectively low. Moreover, incentives and assistance were afforded to local manufacturers of power tillers, seeders, weeders, dryers, and threshers to make farm machinery more accessible to farmers. Despite the presence of public and private machinery stations, few farmers mechanized their farm operations. The national average for mechanized land preparation was estimated to be 9% of the total rice area, and only about 1% of the total harvest was threshed mechanically. Ownership of agricultural machines in the study area was limited to the irrigation pump, power tiller, and rice mill (Table 10). The most common farm machinery found in the villages was the irrigation pump. About 12–16% of the households owned at least one pump for rent or personal use. Irrigation pumps in the rainfed village were being rented to farmers from the nearby irrigated villages. Power tillers and rice mills were only present in the villages practicing intensive rice cultivation. The percentage of households owning such machines, however, was very low. The only farm operations observed to be mechanized were land preparation and threshing. The percentage of farmers using tractors for land preparation and mechanical threshers for threshing, however, was relatively small compared with the percentage of farmers using manual operations. This was attributed to the high cost of rental fees for these machines. Tractors were normally rented together with an opera-

Socioeconomic and biophysical characterization of rainfed . . . 419 Table 10. Ownership of agricultural machinery and equipment, Nyaungdon Township, 1996.

Irrigated Agricultural machines Rainfed Lower Upper and equipment lowland terrace terrace Deepwater (% of HHa owning)

Tractor – – – – Power tiller – 6 2 4 Irrigation pump 12 12 15 16 Threshing machine – – – – Winnower/blower – – – – Rice huller – – 2 3 Animal cart 60 41 35 20 Fodder cutter – – – –

aHH = households. tor for $5 per man-machine day. On the other hand, a pair of bullocks with an operator can be hired for only $1.65 per man-animal day. The rent for the thresher was determined by the volume of paddy threshed. The average rental fee was $0.05 per basket or $2 per ton of paddy threshed. Irrigation pumps were rented on a daily basis at the rate of $5 per day with the cost of diesel ($2.50 per gallon) being shouldered by the farmers.

Fertilizer use The fertilizer supply in the country was mostly procured through importation. Rice production consumes 95% of the total available fertilizer (MAS 1996). Prior to 1988, the government directly controlled the importation and distribution of agrochemicals, including fertilizers, to the farmers. Because of the country’s constraints in foreign exchange earnings, however, fertilizer importation had always been low and insufficient. Farmers commonly used urea, triple superphosphate (TSP), and muriate of potash (MP). About 72% of the total fertilizer imports were in the form of nitrogen fertilizer (urea). Total fertilizer use increased steadily from 1.9 kg ha–1 in 1963 to a peak of 72 kg ha–1 in 1984. This dramatic growth in fertilizer use was due to the government’s massive fertilizer subsidy that artificially lowered fertilizer prices by 50%. As a result, a significant increase in rice yield was achieved during this period. After 1985, however, fertilizer use continuously declined, up to a low of 21 kg ha–1 in 1992, because of the foreign exchange constraint that prevented the government from supplying enough fertilizer to the farmers. In 1993, however, the government privatized the marketing and distribution of fertilizer, which made it more available to farmers. Hence, fertilizer use started to pick up and reached 50 kg ha–1 in 1996. Nevertheless, fertilizer was still grossly underused on most farms due to its limited access and high cost. It was estimated that farmers were actually applying only 40% of the recom-

420 Garcia et al mended level of fertilizer use (120 kg urea, 60 kg TSP, and 60 kg MP ha–1 or 55 kg N, –1 27 kg P2O5, and 36 kg K2O ha (MAI 1995). In the study area, fertilizer use among farmers practicing intensive rice cultivation was higher than that of the rainfed farmers, but still far below the recommended level. On average, farmers applied a similar amount of fertilizer N (about 32 kg ha–1) to both traditional and modern varieties (Table 11). This amount fell below the recommended N rate by 40%. Farmers using modern varieties, however, applied more P2O5 and K2O in the form of triple superphosphate and muriate of potash, respectively. The use of farmyard manure (FYM), on average, was higher among rainfed farmers (3,068 kg ha–1) than among the progressive farmers (2,331 kg ha–1). This implied that the rainfed farmers tried to compensate for the deficiency in commercial fertilizer by using more organic fertilizer because of its lower price.

Credit Credit in the villages was obtained both formally and informally. Since no banks existed in the villages, formal loans could be obtained only from the township’s Ag- ricultural Development Bank. Crop loans and other medium- to long-term loans were available to farming households only. Loans were extended to farmers in groups (five

Table 11. Fertilizer use (kg ha –1) under modern versus traditional varieties by farming practices and season, Nyaungdon Township, Myanmar, 1996. a

Rainfed Irrigated (one rice cropping) (double cropping)

Fertilizers used Wet season Wet season Summer rice

Traditional Modern Traditional Modern Traditional Modern

Urea 35 (16) 28 (13) 41 (19) 51 (24) 62 (29) 65 (30) Triple superphosphate 5 (2) 30 (13) – 71 (32) – 65 (29) Muriate of potash 16 (9) 13 (8) – – – 31 (19) Farmyard manure 3,272 2,863 2,045 3,272 818 2,045

Double monsoon rice Rice-fish culture

Wet Wet Wet Dr y Fertilizers used season 1 season 2 season season

Traditional Modern Traditional Traditional Modern Modern

Urea 80 (37) 76 (35) 111 (51) 94 (43) 79 (36) 115 (53) Triple superphosphate 25 (11) 30 (14) 65 (29) 62 (28) 92 (41) 89 (40) Muriate of potash – – – 41 (25) – 72 (43) Gypsum –––– – 124 Farmyard manure 2,863 3,272 5,726 818 1,636 818

a –1 Numbers in parentheses are in kg N, P2O5, or K2O ha .

Socioeconomic and biophysical characterization of rainfed . . . 421 in a group) at the rate of $60 per ha in areas under intensive cropping and $20 per ha in rainfed areas for a duration of 6 months. On average, the amount of the crop loans observed in the study area ranged from $19 to $24 per farmer (Table 12). On the other hand, the amount of medium- to long-term loans ranged from $20 to $812. These government-extended loans were designed to help farmers procure their farm inputs (e.g., fertilizer and seeds) and rent farm equipment such as tractors and irrigation pumps. The interest rate for government loans was remarkably low at 1.5% per month or 18% per annum. Among the landless households, loans were available only from informal sources such as moneylenders, traders, shopkeepers, relatives, and friends. A majority of the loans were obtained from moneylenders (56%). Friends and relatives were also common sources of credit (41%). Interest rates varied considerably from 5% to 12% per month or 144% per annum. Landholders also obtained loans from informal sources since government loans could only be used to finance farm-related expenses. Hence, credit for personal needs was obtained mostly from informal channels. Occasionally, the school’s cooperative operated by the PTA (parent-teacher association) extended loans for personal needs of the villagers. Larger loans were observed among farm households belonging to the irrigated villages. This can be related to intensive rice cultivation, which required more cash inputs. Such loans were obtained from both formal and informal sources depending on the size of the farmer’s landholding. Generally, the larger the farmer’s landholding, the larger the loanable amount, since landholding size can be a good indicator of the farmer’s capacity to pay back loans. Notably, in the deepwater village where rice- fish farming was promoted by the government, more credit was extended to farmers to enable them to invest in rice-fish culture. The average loans of landless households ranged from $2.50 to $67. A majority of the loans taken out by landless households were spent to finance businesses (about 70–75%) involving home gardening, livestock raising, and other nonfarm activities (Table 13). Among the personal expenses of the landless that were often financed by borrowing were health care (6–14%) and religious ceremonies (5–9%). The same pattern of use of loans obtained informally was observed among the farming households. Despite the presence of government credit, a majority of the informal loans (53%) were spent to finance agricultural investments. This indicates that credit extended by the government to farmers was still inadequate.

Marketing and prices of rice output Before 1987, the government had always strictly regulated the rice market in Myanmar. Aside from home consumption and seed requirements, all of the farmers’ rice output had to be sold to the state economic enterprises and cooperatives at predetermined prices. Immediately after the 1988 crisis, the production quota was removed and farmers were able to sell their produce privately at prevailing market prices. In 1990, however, the production quota was revived but was reduced to 593 kg ha–1. Surplus production was privately marketed. In depressed areas, the quota was lower depending on the existing rice yields in the locality. Only the harvest from the monsoon rice crop

422 Garcia et al (US$) (%)

of loans rate

Landowner

Irrigated

(US$) (%)

of loans rate

Landless

(US$) (%)

of loans rate

Landowner

Rainfed

(US$) (%)

Landless

amount interest HH amount interest HH amount interest HH amount interest

of loans rate

– – – 75 19 1.8 – – – 51 241.4

32.5 0 – – – – – – 1 3,396 6.0

3 17 0.8 1 38 0.8 – – – – – –

% of Average Average % of Average Average % of Average Average % of Average Average

HH = households.

Formal/banks Crop loan Medium-term loan – – – 1 20 1.5 1 671.5 2 812 1.2 Long-term loan – – – – – – – – – 2 319 4.4 School cooperative Informal Money lender 29 48 12.0 8 32 7.9 53 61 10 17 100 7.8 Traders Friends/relatives 66 42 12.3 15 50 6.2 5 404.9 27 198 7.5

Table 12. Sources and costs of loans by size of landholding in Natpay (rainfed) village. Sources a

Socioeconomic and biophysical characterization of rainfed . . . 423 Table 13. Pattern of loan use (% of loan spent), Nyaungdon Township, 1996.

Rainfed Irrigated

Uses Landless Landowner Landless Landowner

Formal Informal Formal Informal Formal Informal Formal Informal

Investment Agric. fixed investment – – –––– –12 Agric. current expenses – 26 27 15 75 9 13 2 Livestock business – 15 5 16 – 55 83 64 Non-agric. business – 29 56 39 – 11 – 17 Consumption Food 100 6 4 7 25 4 2 1 Education/health – 14 8 14– 6 – 1 Housing – 3 – – – 6 – 1 Social/religious – 6 – 9 – 5 – 1 Others –––– 5 2 1

was subjected to the production quota system. The farmers could privately market all of the harvest from summer rice cropping. This policy was implemented in 1992 to encourage farmers to plant summer rice and boost rice production in irrigated areas. The government’s procurement price had always lagged behind the market price, from 38% in 1963 to a maximum of 65% in 1986 (Table 14). As retail prices started to soar in the 1990s, the government was forced to adjust the procurement price, which narrowed the gap to about 17% until 1995. At the village level, however, results of the study showed that the government’s procurement price lagged behind the market price by as much as 69% in the same year (Table 15). In 1996, the government again abolished the production quota, which allowed farmers to sell all their produce privately. After just two months of implementation, however, the government decided to put back the quota system but adjusted the procurement price by 300% (from $30 t–1 to $125 t–1). Despite this effort, the government’s procurement price still lagged behind the market price by as much as 62%, i.e., the current market price was $333 t–1. The quota system of the government in effect created an indirect tax that penal- ized the rice farmers. Results of the study showed that about 27% of the farmers’ total harvest was sold to the government as a production quota. The average tax per farmer was therefore valued at $85, assuming that the quota could be sold privately at market prices. In turn, this implicit tax was estimated to be 7% of the farmers’ total farm income. The tax burden of the production quota became more acute for farmers with smaller landholdings, for whom the proportion of the production quota to total harvest can go as high as 63%. In some unfortunate circumstances, farmers needed to buy paddy from the market to fulfill their quota for the cropping season.

424 Garcia et al Table 14. Government procurement price and market price of rice from 1963 to 1995, Union of Myanmar.

Government Market price Price gap Difference Year procurement price (kyats t–1) (kyats) (%) (kyats t–1)

1963 239 385 146 38 1971 276 432 156 36 1976 695 1,17447941 1979 695 1,286 591 50 1984695 1,645950 58 1985 732 1,753 1,021 58 1986 732 2,082 1,350 65 1987 1,7742,04 0 266 13 1988 3,075 3,717 642 17 1989 4,331 4,380 49 1 1990 4,391 4,638 247 5 1991 4,198 5,371 1,173 22 1992 5,9248,64 2 2,718 31 1993 9,473 12,164 2,691 22 199412,555 15,563 3,008 19 1995 14,676 17,746 3,070 17

Sources: Data from 1963 to 1992, Hossain and Marlar Oo (1995). Data from 1993 to 1995 (Statistical Yearbook 1995, Ministry of National Planning and Economic Development (1995).

Table 15. Marketing and disposal of rice production, Nyaungdon Township, 1996.

Irrigated Average Variables Rainfed Lower Upper Deepwater all terrace terrace villages

Total amount produced (baskets) 208.68 220.66 217.17 278.94231.36 Amount sold to government (baskets) 56.01 51.21 60.77 82.53 62.63 Price from sale to government (kyats) 74.02 73.06 73.45 72.98 73.38 Amount sold to market (baskets) 69.21 86.65 141.46 211.81 127.28 Price from sale to market (kyats) 221.31 245.56 232.00 246.30 236.29 Average distance to government sales 5.92 5.05 5.55 5.36 5.47 center (km) Average distance to market center (km) 1.61 1.82 1.69 1.75 1.72

Paddy output was often transported to the government’s production camps (about 2 km from the villages) using bullock carts and tractors. Other means of transport were trucks, boats, bicycles, and carrying on the head. On the other hand, paddy output marketed privately was generally picked up by rice traders at the farm gate.

Socioeconomic and biophysical characterization of rainfed . . . 425 Labor and employment Table 16 presents the occupational distribution of the economically active population in the villages from 16 to 59 years old. A majority of the people (64–69%) were engaged in agriculture as farm operator, hired laborer, or livestock holder. The second largest population group was composed of dependents, that is, housewives and mi- nors (18–25%). Students represented 3–5% of the population. About 5% of the population was engaged in trading and market vending. The group of government employ- ees working as teachers, clerks, and extension workers was less than 1%. Job opportunities in rural industry, construction, and transport services were very low, averag- ing about 1% for each employment group. Household members with a farm operator as head of the household with extra time from their farms also worked in a secondary occupation as hired laborers. Women kept themselves busy after working on their farms by tending vegetables, flowers, and betel leaves in their homegardens to augment family income. Taking care of small livestock such as chickens, ducks, and pigs also provided a secondary livelihood. Table 17 presents the labor participation of children 10 to 15 years old in the four villages. The proportion of children who were participating in economic activities at an early age ranged from 25% to 34%. A majority (66–75%) of the children were still attending school in this age bracket. Notably, a higher proportion of female children quit school early to participate in economic activities (32–40%) than male children (15–35%) because of the scarcity of intermediate schools in the villages. Hence, only male children who reached high school were often sent to the township to study. After finishing elementary school, female children were generally commis- sioned to help their parents in their livelihood. Labor for seasonal agricultural work was generally supplied by the landless households. They spent about 160–190 days per year working as farm laborers (Table 18). Although landowners participated in the agricultural labor market, the average number of days they spent as laborers was 66% lower than that of the landless. The participation rate decreased as the size of the landholding increased. Notably, the demand for agricultural laborers was 15% higher in the villages where intensive rice cultivation was being practiced than in the rainfed village. This was due to the short turnaround time required in intensive rice cultivation; hence, more laborers were needed to finish the farm tasks. Working as farm help also provided the landless with a steady job. Generally, young boys between 12 and 16 years old were hired as farm help by landowners to take care of their livestock. In the rice-fish area, many landless laborers worked as caretakers of fishponds and as seasonal laborers during fish harvesting. Transport operation (pedicabs and boats), collection of fuelwood, personal service, and shopkeeping also kept the landless households busy, especially during lean months when farm work was scarce. Industrial jobs were not available in the villages. Indus- trial workers often needed to seek employment in the townships or cities. The number of days spent by the landless for all types of jobs was generally higher than that of the landowning households. Because family members of farm households were often

426 Garcia et al (kyats)

Deepwater

– 264 26,400

– 158 7,012

– 180 4,500

(kyats)

Upper terrace

Irrigated villages

– 90 2,250 300 9,900

(kyats)

ownship, 1996.

Lower terrace

– 288 17,280 –

(kyats)

Rainfed

om nonfarm activities, Nyaungdon T

–

172 11,550 99 9,392 148 19,267 165 13,483

382 28,000 120 6,000 191 36,973 182 14,310

158 9,386 205 16,796 109 15,880 241 83,835

318 33,900 134 8,757 –

248 23,211 222 17,923 282 35,965 170 48,750 142 15,756 105 12,167 85 6,144 192 19,792

257 7,980 395 10,427 360 9,839 381 23,750

277 18,239 329 22,068 360 26,172 292 19,989

465 14,055 –

203 22,900 190 15,191 229 23,95420441,329

days of annual days of annual days of annual days of annual

Average Average Average Average Average Average Average Average

employment income employment income employment income employment income

forestry

Table 16. Employment and income fr

Nonfarm sources

Government employment Rental income Pension/donation

Agricultural labor Industrial labor Collection of firewood/fuel/ 121 13,038 400 15,800 –

Animal husbandr Fishing Trading/shopkeeping Transport operation Construction labor Personal service

Socioeconomic and biophysical characterization of rainfed . . . 427 Table 17. Incidence of child labor (age group 10–15 years), Nyaungdon Township, 1996.

Irrigated

Child population Rainfed Lower Upper Deepwater terrace terrace

Males Total number 40 100 83 66 Attending school 80% 65% 76% 85% Participating in economic activities 20% 35% 24% 15% Females Total number38 7474 73 Attending school 66% 68% 60% 66% Participating in economic activities 34% 32% 40% 34% All Total number 78 174157 139 Attending school 73% 66% 68% 75% Participating in economic activities 27% 34% 32% 25%

Table 18. Average per capita employment in nonfarm activities (no. of days per year), Nyaungdon Township, 1996.

Rainfed Irrigated Nonfarm sources Landless Landowner Landless Landowner

Agricultural labor 160 58 190 61 Collection of fuel 57 90 166 – Animal husbandry 214270 45 100 Fishing 86 – 75 51 Industry 74 60 84 46 Trading 113 33 116 58 Transport 142 53 127 57 Construction 79 47 95 31 Personal service 154174 48 86 Government employment 288 300 329 268 Rental service – 191 106 70 busy with works on their farms, nonfarm activities could be done only during the off- season.

Wages and incomes Table 19 presents the composition of annual household income. Agricultural income contributed 51% to total household income while the remaining 49% came from off- farm sources. Income from agricultural labor, about $0.50 per man-day, contributed 10% of the total income. Other agricultural income generated by households mostly came from farming, home gardening, and raising small livestock in the backyard.

428 Garcia et al Table 19. Composition of household annual income (% of total income), Nyaungdon Township, 1996.

Rainfed Irrigated Sources Average Landless Landowner Landless Landowner

Agriculture Farm/garden 7 66 437 28 Agricultural labor 11 3 1410 10 Animal husbandry 6 3 1 1 3 Fishing 6 – 9 4 5 Forestry 7 2 9 – 5 Total 37 7437 52 51 Nonagriculture Industry 8 2 9 10 7 Trading 143 1416 12 Transport 15 1 9 4 7 Construction 9 2 11 2 6 Personal service 12 10 2 1 6 Gov’t employment 5 46 4 5 Rental service – 49 4 4 Transfer – – 4– 1 Total 63 26 64 41 49

A majority of the income (64%) of the landless households was generated from nonfarm activities. Trading, transport operation, personal service, and construction labor were the major sources of nonfarm income, with shares of about 9–15% each in total income. On the other hand, a larger proportion (63%) of the total income generated by landowner households came from agricultural activities. The higher income of the irrigated farm households compared to the rainfed households was attributed to the intensive rice cultivation in these villages. Income from agricultural labor contributed a higher percentage in the irrigated villages (10% vs 3%) because of the increased opportunity for farm work brought about by the new rice-based production systems. Trading, transport operation, and personal service were also the major sources of nonfarm income for the farm households, which contributed 34% to total income. A higher share of nonfarm income in the farming households belonging to the irrigated villages (41% vs 26%) was observed, implying that family members were able to engage in more high-paying off-farm activities since farm operators tended to hire more agricultural laborers to do the farm tasks.

Income distribution To determine the pattern of income distribution in each village, the households were ranked with respect to per capita income and the corresponding income shares of

Socioeconomic and biophysical characterization of rainfed . . . 429 Cumulative proportion of income 1.0 45 degree line Rainfed 0.8 Lower terrace Upper terrace 0.6 Deepwater All villages

0.4

0.2

0.0 0.0 0.1 0.2 0.3 0.40.5 0.6 0.7 0.8 0.9 1.0 Cumulative proportion of population Fig. 1. Lorenz curves of per capita total income.

Table 20. Gini ratios for total income and its components, Nyaungdon Township, 1996.

Irrigated Distribution Rainfed Lowland Upland Deepwater All

Income Agricultural labor income 0.69 0.68 0.67 0.75 0.70 Farm income 0.59 0.72 0.79 0.59 0.68 Nonfarm income 0.80 0.75 0.81 0.840.81 Total income 0.35 0.41 0.48 0.47 0.44 Ownership of resources Land 0.66 0.68 0.68 0.56 0.65 Education 0.25 0.28 0.22 0.240.25 successive decile groups were estimated. Figure 1 shows a series of Lorenz curves1 depicting the pattern of income distribution in the villages where different rice production systems were being used. To infer the degree of inequality in income distribution, the Gini concentration ratio2 was estimated based on per capita household income. Table 20 presents results of the analysis. The Gini ratio of income generated from all sources, such as farm,

1The Lorenz curve shows the percentage share of income received by different population groups. It shows the degree of equality and inequality in income distribution. The greater the departure of the Lorenz curve from the 45° line, the more unequal the distribution of income. 2The Gini concentration ratio is the estimated area inside the Lorenz curve. The value of the ratio ranges from 0 to 1. The closer the value of the Gini ratio to 1, the more unequal the distribution of income.

430 Garcia et al agricultural labor, and nonfarm income, was 0.44 for all the villages. The rainfed village estimate (0.35) was lower than that of the irrigated villages (ranging from 0.41 to 0.48). This indicates that the distribution of income in the rainfed village was less unequal compared with that of the other villages with access to irrigation facilities, which enabled intensive cropping. The estimated Gini ratio for farm income registered an average of 0.68 for all villages. The ratio was expected to be significantly higher than the aggregate income due to the highly unequal distribution of land ownership in the villages. The Gini concentration ratio for land ownership was estimated at 0.65. Notably, in the deepwater area where common lands were still available for public cultivation, the Gini ratio was lower (0.56). The distribution of farm income at the village level showed interesting Gini coefficients. The rainfed and deepwater areas registered lower ratios (both 0.59) than the villages located in the upper and lower terraces (0.79 and 0.72, respectively). This observation further emphasized that the presence of irrigation facilities contributed to widening the disparity of farm income among landowners. The average Gini ratio for agricultural labor income in all four villages was estimated to be 0.70. However, the deepwater village showed an even higher ratio (0.75) than the other villages (ranging from 0.67 to 0.69). This showed that the opportunity for family labor (especially for the landless) to work in the fields as hired laborers was hampered by excessive flooding in the area during the monsoon season. The highest Gini ratio was registered by nonfarm income (0.81 for all villages). This indicates that greater income inequality resulted from the differential access of household members to nonfarm jobs. This may be due to fewer opportunities for nonfarm activities available in the villages or a lack of the skills required in most nonfarm jobs. Gini ratios for the distribution of education were likewise generated to determine how individuals in the villages differed in per capita access to education. Estimates showed that the ratios were generally low (0.22 to 0.28), indicating that individual members of the active population had relatively the same level of education across villages (at most 7 years of schooling).

Cost and return analysis of different rice production systems To determine the profitability of the different rice-based technologies promoted by the government in the study area, a cost and return analysis was done for each farm category per season (Table 21). Values for cost and return components were expressed on a per hectare basis. The construction of the cost accounts in the analysis was based on the imputation of existing wage rates and rental fees for all activities using family labor and self-owned resources. At the same time, paddy prices (both government and market prices) were used to value the total harvest. The net return should therefore be seen as an indicator of the farmer’s ability to recover both cash and noncash costs of rice production instead of the conventional idea of profit. Results of the study showed that net gain from the different rice production systems ranged from $30 to $95, whereas net loss ranged from $48 to $135. How- ever, if production cost were computed on the basis of pure cash costs (i.e., not valu-

Socioeconomic and biophysical characterization of rainfed . . . 431 Rice-fish

3.75 –

593 –

DS WS DS

3.75 –

Double cropping

593 –

–

Double monsoon

WS1 WS2 DS

3.75 3.75 –

10.00 11.40 11.10 11.80 11.10 11.80 10.45 12.00

593 593 –

5,105 12,270 14,788 16,707 15,226 16,707 16,449 22,447

2,518 769 3,333 1,724966 1,7241,117 2,963

3,731 (13,486) 11,380 (14,696) (16,219) (14,696) (16,143) (575)

Rainfed

13,483 12,655 14,162 21,170 30,320 21,170 14,198 20,916 18,588 24,925 28,950 37,877 15,094 37,877 30,647 43,363

11,846 (7,240) 19,065 (5,849) (8,032) (5,849) (9,095) 3,505

y season.

eturn 1996. analysis (values in kyats) of rice cropping under different production systems, Nyaungdon Township,

–1

WS = wet season, DS = dr

Sale from rice straws 845 448 3,334 2,838 1,154 2,838 607 2,056 Total returns Total 22,219 11,439 40,330 23,181 14,101 23,181 14,504 37,612

Table 21. Cost and r

Item

Material cost Labor cost Total cost Total

Marketed Amount Price Quota Amount Price Costs Returns Net returns season Noncash costsReturn over cash cost a 7,755 6,246 7,685 8,847 8,187 8,847 7,048 9,256

432 Garcia et al ing self-owned resources), net gain resulted in a wider range ($29 to $160) but with a higher frequency of occurrence since net losses were minimized. This observation can be attributed to the significant proportion of noncash costs with respect to total cost (29%). In general, rice monocropping in the rainfed village generated positive net returns compared with intensive rice cultivation. The lower net returns in the irrigated villages was due to the high cost of irrigation for dry-season cropping, that is, the high cost of diesel fuel and pump rental. Likewise, farmers engaged in intensive cultivation used more fertilizers, hence the higher production costs. Despite the higher fertilizer use observed on intensive rice farms, however, yields did not increase enough to cover the high production costs since fertilizer use was still inadequate. Another important factor that must be considered in the farmers’ failure to recover their production costs was their inability to cope with the technological and economic demands of the new rice-based cropping systems in terms of time, labor, and capital resources associated with intensive rice cultivation. Note that among the three programs advocated by the government (summer rice cropping, double monsoon cropping, and rice-fish farming), only the summer rice production showed sustainability in farmer adoption. Double monsoon cropping was totally abandoned after one year of implementation. Likewise, adopters of rice-fish farming drastically modified the system in favor of fish culture.

Major production constraints To summarize the impacts of biophysical and socioeconomic factors on the rice production systems in the country, major production constraints (both biotic and abiotic) were identified and their effects quantified using the yield gap method. Through this technique, normalized yield losses associated with these constraints could be assessed and evaluated to determine their relative importance. Results of the study showed that the abiotic constraints significantly affected a larger percentage of the total rice area and the estimated yield losses were higher compared with those of the biotic constraints (Tables 22 and 23).

Biotic constraints to rice production The biotic constraints commonly reported by farmers in the survey area can be divided into three major categories: genetic problems, insect pests, and diseases. Ge- netic problems included varietal degeneration, stunted growth, and lodging. Results of the study showed that the major genetic problem was varietal degeneration. The estimated normalized yield loss associated with this problem was about 31 kg ha–1, resulting in a projected total production loss of 181,327 tons or 1% of the total national production. The primary reason for this problem was the continuous use of seeds coming from the farmer’s own harvest without proper knowledge of the techniques for seed selection. For insect pests, seven species were identified by farmers in the survey area: rice caseworm, rice hispa, rice stem borer, rainball, gall midge, armyworm, and brown

Socioeconomic and biophysical characterization of rainfed . . . 433 Table 22. Estimated rice yield and production losses due to biotic constraints in Myanmar, 1997.

% of total % Estimated Normalized Estimated % loss with Biotic rice area occurrence yield yield loss production respect to constraints affected in 10 years loss (kg ha–1) loss (Mt) national (kg ha–1) production

Genetic Varietal degeneration 8.59 80 454 30.97 181,327 1.05 Stunted growth 1.41 73 707 7.24 42,410 0.25 Lodging 0.72 80 363 2.08 12,1740.07 Total 235,910 1.36

Insect pests Caseworm 3.76 84251 7.92 46,3640.27 Rice hispa 0.85 31 468 1.23 7,200 0.04 Stem borer 0.7469 251 1.29 7,552 0.04 Rainball 0.27 55 311 0.46 2,693 0.02 Gall midge 0.21 15 482 0.15 878 0.01 Armyworm 0.21 28 492 0.28 1,639 0.01 Brown planthopper 0.03 10 1,3740.04 2340.00 Total 66,560 0.38

Diseases Bacterial leaf blight 2.80 56 398 6.20 36,295 0.21 Blast 0.28 55 1,749 2.65 15,513 0.09 Bacterial leaf sheath 0.47 65 246 0.75 4,391 0.03 Total 56,198 0.32

planthopper. Among these insect pests, the rice caseworm was considered the most destructive, resulting in an estimated yield loss of 8 kg ha–1. This figure, however, translates into only 0.27% of total national production. The total damage caused by all these insects combined was estimated to be less than 1% of the total national production. Although insect pests were commonly observed on rice farms, their occurrence and infestation were still low and considered to be insignificant. Because of the limited supply and high cost of agrochemicals, most farmers did not apply insecticides. This resulted in the nondestruction of beneficial organisms, which actually helped in the control of harmful insects. Rice disease in the country is a new phenomenon and was detected only in the late 1980s. The major diseases that were increasingly affecting Myanmar rice production were bacterial leaf blight, blast, and bacterial leaf sheath. Bacterial leaf blight contributed the highest damage incidence in terms of total national production loss (0.21%) with a normalized yield loss of 6 kg ha–1. Based on the survey, the farmers reporting problems of rice diseases were mostly users of modern varieties, especially in the dry season. Farmers were generally un- aware of the causes and treatment of these diseases. Stubbles in the infected fields

434 Garcia et al 5.78

0.58 0.40

0.46

1.84

0.21 0.02

0.09

0.19

0.29

0.73

0.01 1.15

9.19

3.20

0.13 0.20

0.05

0.50

1.37 0.63

% loss

18.21

16.06

21.26

production

to national

2,986

2,225

8,4

(Mt)

69,955

80,083

36,588

16,391

16,040

33,661

49,876

21,835 34,422 10,6540.06

86,756

100,455

317,755

125,275

201,495

553,475

236,816 109,437

3,151,325

1,000,097

2,777,652

1,589,829

3,677,380

)

–1

2.7 0.4

6.3 0.5

8.5

2.8

5.8

1.5

5.9 1.8

13.7

17.2 12.0

54.3

94.6

14.8

40.5 18.7

170.8

474.5

271.6

(kg ha

) yield loss production loss with respect

–1

666

233 945

487

486

796

628 777

897

404

860

929

668

7943.7

812 595

1,169

1,199

1,814

1,192 1,177

loss (kg ha

70 21

19 25

27 13

38 42 51 541,192

33 57

2.73

3.90

0.86 0.58

5.20 0.30

4.15

3.99

0.80

1.25 1.17 0.30 0.22

46.80

10.50

19.10

49.40

oduction losses due to abiotic constraints in Myanmar, 1997. oduction losses due to abiotic constraints in Myanmar,

area affected

% of total rice % occurrence Estimated yield Normalized Estimated national

vices 6.20

tilizer 63

city

of fertilizer

Total

Low adoption of new technology Seasonal labor scar Limited technical extension ser Irrigation facility deterioration Inadequate supply of good seed

Lack of credit facility

Table 23. Estimated rice yield and pr

Abiotic constraints

Insufficient use of chemical fer Poor weed management

Flood

Acidic soils Sandy soils Low fertility

High cost of fertilizer Untimely and inadequate supply

Poor water management

Drought at vegetative stage Drought at seeding Drought at anthesis Cold at anthesis

Alkalinity Salinity Agronomic Use of old seedlings Climatic/environmental Soil-related Socioeconomic

Socioeconomic and biophysical characterization of rainfed . . . 435 were often burned for disease control. The use of fungicides was almost nil due to the farmers’ limited knowledge of their use and the high cost of these chemicals.

Abiotic constraints to rice production The abiotic constraints reported by farmers in the survey area were categorized into the following: agronomic problems, climatic and/or environmental factors, soil-related problems, and socioeconomic factors. Four major agronomic problems in rice production were identified: insufficient use of chemical fertilizer, poor weed and water management, and the use of old seedlings. Of the four problems, the most serious was insufficient use of chemical fertilizer. It was reported in 63% of the total rice area surveyed, with a normalized yield loss of 474 kg ha–1. This translates into a total production loss of about 2.8 million tons or 16% of the total national production. Because of the high cost of commercial fertilizer, farmers were observed to apply less than half of the recommended rate of nitrogen fertilizer and almost none of the phosphorus and potassium fertilizers. Other agronomic constraints caused by poor weed and water management resulted in normalized yield losses of 95 and 40 kg ha–1, respectively, with estimated national production losses of 3% and 1%. Weed problems were often attributed to farmers’ inability to prepare and level rice fields properly, which made weeding difficult. Also, the practice of broadcasting as a method of crop establishment aggravated the weed problems. At present, the use of herbicides is minimal due to the farmers’ lack of knowledge about the use of chemical weed control and the high cost of herbicides. On the other hand, poor water management was often associated with farmers’ inability to control the water supply because of insufficient water from rainfall and irrigation systems. In central and upper Myanmar, irrigation facilities were constructed to supplement rainfall for the wet-season crop since water was always insufficient in the dry zone. However, despite the government’s effort to boost water supply in these areas, water stress from drought still posed an important constraint in the dry zone. Another important abiotic constraint was brought about by socioeconomic factors. The two most important socioeconomic constraints reported by more than 80% of the township surveyed were the lack of credit facilities (87%) and the high cost of commercial fertilizers (83%). These constraints were reported to bring about normalized yield losses of 171 and 272 kg ha–1, respectively. Production losses were estimated to be 6% and 9% of the national production, respectively. These problems were related, however, since credit was normally used to purchase agricultural inputs, especially chemical fertilizer. With the high cost of fertilizer coupled with the lack of credit facilities, most farmers could barely afford to purchase the much-needed amount of fertilizer. Other socioeconomic constraints reported by farmers were low adoption of new rice technologies, seasonal labor scarcity, limited technical extension services, deterioration of irrigation facilities, and inadequate supply of good seed material. The effects of these constraints on national production, however, were minimal, reaching a maximum of only 2%. Likewise, stresses brought about by climatic/environmental

436 Garcia et al and soil-related problems contributed low production losses and were both estimated to be about 1% of total national production.

Conclusions Myanmar has a substantial potential for increasing rice production. A wide range of policy options is available to policymakers to tap this potential. The implementation of the summer rice (dry-season) program in 1992, coupled with irrigation infrastructure development, successfully showed that this potential could easily be exploited. Unfortunately, given the present coverage of irrigation facilities (only 26% of the total rice area), this potential has not been maximized, despite the availability of abundant water sources for irrigation. More infrastructure development is needed to move the entire country into rice double cropping. Low-cost irrigation structures such as shallow tube wells (a technology still unknown to Myanmar farmers) can be har- nessed to provide the much-needed water for dry-season production. Furthermore, reclamation of the culturable wastelands and fallow lands in the delta area and elsewhere proved to be a strategic policy for extending the land frontier for rice and other crop production. The recent land development policy of the government for the large corporations operating in the country, which aimed to encourage agricultural production and exports, may prove to be ideal along this line. However, extreme caution should be exercised in monitoring the activities of these corporations to ensure that increased agricultural production is indeed achieved, instead of simply rechanneling the existing domestic output to international markets. Increasing yield is also one of the potential sources of rice output growth in Myanmar. One important factor that constrained rice yields, however, was the inadequate use of fertilizer. It was observed that farmers applied only about 40% of the recommended fertilizer rate for both modern and traditional varieties. This was primarily due to the high price of commercial fertilizer and its limited supply in the villages. With approximately 54% of the total rice area grown to modern and improved rice varieties, the potential yield of these varieties can only be achieved if appropriate amounts of fertilizer are applied. The use of FYM helped augment the deficiency in the use of chemical fertilizer. Despite the use of FYM, however, the total available NPK for efficient rice production is still sadly inadequate. There are two possible ways to approach the fertilizer problem in the country. First, fertilizer importation should be deregulated to encourage businessmen to procure it externally and make it more available to local markets at lower prices. Second, more credit facilities should be extended to the farmers to enable them to buy and apply the fertilizer needed to increase their yields. Proper incentives for farmers can also help increase rice production. The implicit tax burden of the production quota can be reduced if the government procurement price can be regularly adjusted to reflect market prices. Landlessness is a pervasive problem among the Myanmar people despite the country’s vast land area. This is a critical issue about which policymakers have opted to remain silent. Land distribution in the sample villages was highly unequal, with the

Socioeconomic and biophysical characterization of rainfed . . . 437 landless population reaching 50–60%. This had a significant effect on the inequality of income distribution among the village population. Without land, the landless had limited sources of livelihood. Their income-generating capacity was severely hampered; as a result, poverty among this group was most pronounced. On the other hand, the landless population provided the much-needed labor for rice production, especially during the peak seasons. Average farm size ranged from 0.8 to 2 ha; hence, farm laborers were commonly hired. The potential for mechanization to solve seasonal labor shortages was encouraging. The cost of buying and/or renting farm machinery, however, was still prohibitive for farmers. With the right government policies, incentives, and direction, Myanmar may be able to regain its position as one of the world’s major rice exporters. The potential for increasing rice production can be achieved with (1) rice planting in the fallow lands (1.3 million ha) coupled with irrigation infrastructure development, (2) the expansion of cultivable land area coming from the available 8.0 million ha of culturable waste- land, and (3) double rice cropping on rainfed lands where abundant water sources can be tapped for irrigation. To address the major rice production constraints, Myanmar scientists should give high priority to research on varietal improvement, integrated nutrient management, integrated pest management, and appropriate agronomic management. Such research is expected to develop production technologies that will overcome present and anticipated production constraints.

References Hossain M, Marlar Oo. 1995. Myanmar rice economy: policies, performance and prospects. Paper presented at the Final Workshop on Projections and Policy Implications of Me- dium and Long-Term Rice Supply and Demand Project, Beijing, China. MAI (Ministry of Agriculture and Irrigation). 1995. Information on Myanmar agriculture. The Government of the Union of Myanmar. Ministry of National Planning and Economic Development. 1995. Statistical yearbook 1995. The Government of the Union of Myanmar. Myanma Agricultural Service. 1996. Myanmar agriculture in progress. Ministry of Agriculture and Irrigation, The Government of the Union of Myanmar. Myanma Agricultural Service. 1997. Rice cultivation situation of Ayeyarwady Division for 1996-97. Ministry of Agriculture and Irrigation, The Government of the Union of Myanmar.

Notes Authors’ addresses: Y.T. Garcia, Assistant Professor, Department of Economics, College of Economics and Management, University of the Philippines at Los Baños, Philippines; M. Hossain, Head, Social Sciences Division, International Rice Research Institute, DAPO Box 7777, Metro Manila, Philippines; A.G. Garcia, Representative and Agronomist, IRRI Representative Office, International Rice Research Institute, Yangon, Myanmar.

438 Garcia et al Acknowledgments: The authors wish to express their sincerest gratitude to the key officials of the Myanmar Agriculture Service and Department of Agricultural Planning for their generous support in the conduct of the village survey. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Socioeconomic and biophysical characterization of rainfed . . . 439 Integration of biophysical and socioeconomic constraints in rainfed lowland rice farm characterization: techniques, issues, and ongoing IRRI research

C.M. Edmonds and S.P. Kam

This chapter reviews research incorporating socioeconomic and biophysical variables into analysis to characterize rainfed rice environments. The use of geographic information systems (GIS) as a tool for integrating these two types of data is highlighted. GIS starts as a useful tool for organizing and display- ing geo-referenced social and economic data. The increasing ease of use of GIS software and improvements in geographic positioning systems and remote-sensing technology facilitate the generation of more accurate spatially referenced data that can be integrated into analyses of agricultural practices and outcomes. GIS software, combined with other software, provides analytical techniques for converting distinct types of biophysical and socioeconomic data to a common scale amenable to analysis as an integrated database. These techniques are outlined. A final area of GIS application in socioeconomic analysis involves linking GIS analysis with other modeling techniques. Econometric modeling and linear programming models using spatially referenced biophysical and socioeconomic data are examples of this type of research. In the second part of the chapter, we review research that the Interna- tional Rice Research Institute is carrying out in the Mekong River Delta of Vietnam in collaboration with the Institute of Agricultural Sciences in Ho Chi Minh City. The research applies geo-informational techniques and methodologies, and review of it enables consideration of the material covered in part one. Here, we develop a spatial economic model of crop choice and land-use intensity to provide an analytical framework for empirical examina- tion. Empirical estimates provide insight into the roles that biophysical and socioeconomic constraints play in explaining changes in land use and productivity, and enable exploration of the interrelation between biophysical and socioeconomic production constraints. Random effects probit estimates show which factors influence farm land use. This estimator uses the panel structure of the data, and provides robust estimates. Ordered probit and multinomial logit estimates of cropping intensity and cropping pattern adopted were carried out using single years of the survey to enable estimation of the effect of time invariant characteristics on these outcomes. Findings from these estimates are reviewed and extensions and future applications of GIS econometric integration are considered.

Integration of biophysical and socioeconomic constraints . . . 441 Both biophysical and socioeconomic constraints influence land-use decisions and limit the production and income of rice-farming families in Asia. Accounting for the biophysical environment in agricultural research is essential in understanding production outcomes and the decisions of farm operators. Without proper characterization of the constraints imposed on farms by climate, topography, soils, and similar factors, efforts to understand outcomes in order to design and disseminate new technologies to bring about outcomes are likely to fail. An understanding of the socioeconomic environment in which farms operate is equally important in agricultural research, technology development, and agricultural extension. Socioeconomic constraints such as the availability of agricultural inputs and presence of buyers for agricultural output, or noneconomic constraints such as cultural sanctions against certain castes or against women engaging in particular production activities, exert significant influence over farm households. Characterizing both types of constraints and understanding their relation to each other is essential to the development of technologies and policies to increase rice production and the incomes of rice farms. Integration of traditional econometric techniques with data organized in a geographic information system (GIS) offers a promising method for modeling both types of constraints. This paper is divided into two parts. The first part provides an overview of the general theme of geo-informational techniques combining biophysical and socioeconomic variables in the characterization of rainfed areas. We describe some of the techniques applied in carrying out such integration, and discuss some of the problems faced in such research. To make these points more concrete, in part two we review research applying these methodologies that the International Rice Research Institute (IRRI) and the Institute of Agricultural Sciences (IAS) of Vietnam are carrying out in the Mekong River Delta.

Geo-informational techniques in the characterization of rainfed environments Geo-informational techniques, which include remote sensing, GIS, and related technologies, offer a useful framework for integrating spatially referenced socioeconomic data and biophysical information. GIS facilitates capturing the spatial dimension and spatial analysis of the effects of these factors and their interactions.

Characterizing rainfed lowland ecosystems: Which characteristics are important? In work characterizing rainfed areas, we view the objective of research to be to identify biophysical constraints that limit agricultural production rather than to provide a holistic description of the environment. Agricultural scientists are accustomed to thinking about biophysical constraints to agricultural production and to using maps of biophysical characteristics in designing in-field experiments or explaining variation in farm outcomes across seasons. Because the biophysical environment provides many of the essential inputs to plant growth, the link between biophysical conditions and agricultural outcomes is relatively clear. Relative to biophysical characterization, socioeconomic characterization is less developed, and features a much greater range of

442 Edmonds and Kam opinion regarding what socioeconomic characteristics are important. Agricultural scientists are generally less accustomed to thinking about socioeconomic constraints to agricultural production. Because socioeconomic constraints involve human behavior, individual decision making, and social norms and institutions, the task of describing the constraints appears more daunting. The greater difficulty of socioeconomic characterization stems not only from the fact that it involves a description of human behavior, but also because in many cases we cannot directly observe socioeconomic constraints. Instead, we must make inferences about constraints from observable social, cultural, and institutional characteristics. For example, access to farmland is obviously an essential prerequisite to farm production and land scarcity is considered to constrain the farming activity of many small farms in population-dense South and Southeast Asia. Fail- ure of land markets is commonplace in developing Asian economies for several reasons. However, we can only observe the actual level of land the families use in their farming activities—when what is truly of interest is whether the household has enough land to carry out its farming activities efficiently or is the family land constrained. If a household carries out its agricultural activities using only land it owns and does not engage in any land rental transactions, this could suggest several possible situations. Usually, researchers must rely on other indicators such as observed land-labor ratios applied on-farm combined with the observed land rental activity (or lack of activity) to make inferences about the performance of land rental markets.

The importance of location in socioeconomic characterization Several instances can be cited where location and spatial relationships are of importance in socioeconomic characterization and analysis. One broad spatial characteristic—accessibility—is described here. Other examples of instances where the location of farm characteristics is important to consider in characterizing farms include neighborhood effects, production externalities that cover a geographic area, localized social capital, and spatial autocorrelation (Edmonds and Kam 1999). For example, in developing-country agriculture, neighborhood effects can explain the adoption of new technologies, rural-to-urban migration patterns, and the presence of informal support networks between farms. Geo-referencing farms provides a precise way of defining neighbors, whereas spatial statistics provide tools that allow flexible definitions of neighborhood distances. A related topic is the area of spatial econometrics. Whenever carrying out statistical estimates using data for which the responses of observations vary systematically depending upon their spatial location vis-à-vis other observations or central points, estimation procedures that fail to take into account the systematic variation will have inefficient parameter estimates. This systematic variation over space is referred to as spatial autocorrelation. The statistical issues encountered with spatially autocorrelated data are similar to the more common problem of serial correlation encountered in time series data. Statistical procedures and software packages for di- agnosing spatial autocorrelation and for estimating models with spatial correlation in observations are available (Anselin 1988).

Integration of biophysical and socioeconomic constraints . . . 443 Applications of geo-referenced biophysical and socioeconomic data—levels of analysis Geo-information techniques provide a range of tools for creating, manipulating, and analyzing geographically referenced data. These tools may be deployed at different levels of sophistication in application to socioeconomic studies. Description and exploration of spatial patterns in variables. The simplest application of GIS with socioeconomic information is to display the geographic distribution of these data. The task of collecting and processing data capturing the complete spatial distribution of socioeconomic characteristics is cumbersome and costly, but once accomplished the display of the information on maps generated from a GIS is uncomplicated and can be rewarding. Visualizing such information in a spatial and graphical manner helps make clear the spatial pattern—if one exists—in observed characteristics. As an example, consider Figure 1. IRRI’s GIS facility generated the figure from a GIS data set on rice production in northeast Thailand. It shows several characteristics and spatial patterns of interest. The pie charts on the figure show the rice production level (indicated by the size of the pies) by variety for 19 provinces. It is evident that the main rice-producing provinces are in the south. Here, the predominant rice variety planted is KDML105, which is a commercial, premium-quality nonglutinous fragrant rice. Rice in the northern provinces is not only less important in terms of volume produced, but in this area it is used mainly for domestic consumption (as indicated by the relatively higher proportion of glutinous varieties). The figure also shows that KDML105 yields in the southern provinces are the lowest among provinces in northeast Thailand. As suggested by this example, superimposing maps of socioeconomic outcomes and biophysical conditions can suggest research hypotheses and foster understanding of the relation between the two types of characteristics. Generation of geo-referenced socioeconomic information—defining information at a common scale. A ubiquitous problem in GIS research with socioeconomic data relates to the scale and comprehensiveness in geographic area addressed in the research. In particular, socioeconomic research is often asked to characterize a large area while data available are often limited to much smaller areas. Unfortunately, most socioeconomic studies cover narrow geographic areas because of the difficulty of carrying out detailed socioeconomic surveys over large areas as the result of financial and logistical constraints. Recent advances in modern, satellite-based remote-sensing and global-positioning system (GPS) technology promise to partly alleviate short- comings in the availability of geo-referenced data. Land use and land cover, distribution of human settlements, and location of infrastructure (roads, public facilities, elec- tricity, irrigation canals) are examples of the types of socioeconomic information that are discernible using remote-sensing data. GPS technology is useful for geo-positioning survey points, such as village and farm locations, and even in tracing field boundaries. The second problem in integrated GIS and socioeconomic research is that the required data are often not available at the proper level of resolution. Commonly, aggregate data are available whereas more detailed information for small areas or individual economic agents is required. Furthermore, data obtained from different

444 Edmonds and Kam 1995-96 Rice production (t) Tv Rd6 Rd15 Rd21 Rd23 Rd-nps Rd-ps Sb60 Sb90 Kdml105 Ps60-1,2 lv-nps lv-ps Bm 1995-96 KDML105 yield (kg rai–1) <200 kg/rai 201–250 251–300 301–350 351–400 >400 kg rai–1

60 0 60 km

Fig. 1. Rice production by variety in provinces in northeast Thailand (1995-96). Source: Social Sciences Division (GIS), IRRI. 1 ha = 6.25 rai. sources are often defined at different spatial resolutions, and need to be reconciled and integrated into a consistent data analysis framework. This requires changing the spatial reference scale of the data. There are three principal types of scale transformation of data (Edmonds and Kam 1999): (1) aggregation—combining units of observation to a higher scale, (2) reaggregation—changing the boundaries of observation while maintaining the same spatial scale or resolution, and (3) disaggregation—moving data from a higher scale to a lower scale. Invariably, analyses based on transformed data have inherent biases and errors. Deichmann (1999) reviews three problems related to aggregation bias: the modifiable areal unit problem, an error-in-variables statistical problem, and the ecological fallacy problem.

Integration of biophysical and socioeconomic constraints . . . 445 A noteworthy exception to the generalization that socioeconomic data are available for only restricted geographic areas is the information collected in national censuses. In a national census, socioeconomic information is collected for all regions of the country. Unfortunately, although census data provide geographically complete socioeconomic information, few variables are collected and these generally provide an incomplete picture of the socioeconomic characteristics and economic outcomes of individuals taking part in the census. Data collected in censuses generally cover little more than the basic demographic characteristics of families and limited information about living conditions at their place of residence. The development of techniques to use detailed information obtained from sample surveys in combination with extensive census data for generating data sets with more comprehensive geographic coverage to generate poverty maps is an active area of research (Hectschel et al 1997). Development of a similar approach using data collected from remote sensing to extrapolate results from detailed farm surveys to broader areas is under way at IRRI. Modeling spatial processes—integrating GIS with other analytical methods. Moving beyond description, data generation, and univariate analysis, GIS integration with other quantitative analytical methods used outside geography seeks to explain the processes underlying the observed spatial distribution of variables of interest by incorporating the spatial distribution of farm characteristics in modeling farm outcomes. One means of achieving this is to link GIS with standard econometric analysis of farm survey data, that is, to use outputs from GIS as inputs into econometric models, or to map the outputs of econometric analysis and use them for further spatial characterization. We turn our attention to research in Vietnam that applies this approach in the section “Land-use dynamics in the Mekong River Delta: an illustration of integration.”

Accessibility and transportation costs Accessibility to markets and market intelligence concerning the prices of agricultural inputs and outputs influences the agricultural production system pursued by a farm and its outcomes. One direct influence of accessibility is on the ease, speed, and cost of transportation of agricultural inputs and farm outputs. These factors influence farmers’ decisions regarding (i) how much of a particular agricultural commodity to produce or purchase, and (ii) the type(s) of crop(s) to produce. Influence of transportation costs on farm household decisions on production. Discussion of a simple supply and demand model displayed in Figure 2 illustrates case (i) above. The convergence of supply and demand at the economy-wide level defines the market-clearing price for a good (Pm). Each farm can be thought to have internal household supply and demand functions for goods, which depend upon the farming resources they have available, their production efficiency, and their consumption needs. As the price of a good increases, the household will demand less of the good from the market and be willing to produce more of the good for market sale. The intersection of a household’s internal supply and demand schedules defines a household’s “shadow price” for the good (Fig. 2, point c).

446 Edmonds and Kam Farm-gate price of agricultural good Internal farm demand for good

Internal farm supply of good Pp Shadow price of c output to farm Pm

Qs = Qd Output purchased/sold by farm Fig. 2. Supply and demand of agricultural output with transportation costs. (Adapted from de Janvry et al 1991.)

In the absence of transportation or other costs associated with conducting business in the market (as either a buyer or seller of a good), whether a household pur- chases a good or sells a good is defined by the relationship between the internal- household equilibrium price and the market price. A household whose shadow price is below the market price will produce enough of the good to meet its demand as well as provide for surplus production for sale on the market. Households for which the household shadow price lies above the market price find it better to produce a portion of the good on-farm and to obtain the rest of the good by buying it at the market. The presence of transportation costs in bringing goods to and from the market changes this situation. Transport costs create a “price band” around the market price. The upper bound of the band (Pp) gives the purchase price of the good when the cost of transporting the good from the market to the farm is added. The lower bound of the band (Ps) gives the net price of sale of the good by reducing the market price by the cost of transporting the good from farm gate to market. In terms of the effect on farm behavior, it is unimportant whether the transport costs are paid directly by the farmers or indirectly through the purchase price offered by intermediaries purchasing the good at the farm gate. The higher the transport cost, the wider the price band. The bands have the effect of lessening the amount of good bought or sold by a household. Households selling the good will produce and sell less of it in response to the price band (the price difference between Ps and Pm). Similarly, households that are purchasers of the good will produce more of the good themselves, purchase less of the good from the market, and consume less of the good overall. When a household’s shadow price lies within the price band, it will produce enough of the good to meet its demand but will not engage in market transactions for the good. Transportation costs, a spatially defined farm characteristic, thus explain the pattern of subsistence and commercial farm pro-

Integration of biophysical and socioeconomic constraints . . . 447 duction in a rural area. The predictions of this simple framework are consistent with the higher incidence of subsistence agriculture in remote areas. Calculation and interpretation of accessibility indicators. The conventional way of estimating transportation time or costs is from questioning of farmers or key infor- mants. This approach often faces problems of subjectivity. Alternative ways of computing quantitative accessibility indicators using less subjective means for estimating travel distance or travel cost/effort are desirable, and provide a means of verifying the data solicited from questionnaires. Because accessibility is largely determined by geography, that is, the availability, density, and quality of infrastructure (i.e., networks of roads, canals), and by the nature of the terrain, geographic information systems lend themselves naturally to the computation of accessibility indicators and generation of accessibility maps. Several indicators of accessibility may be computed, and they fall into two broad categories: 1. Accessibility from the supply perspective, for example, a service area from the point of view of a facility, such as the serviceable area of a tube well drawing water from an underground aquifer. 2. Accessibility from the demand perspective, for example, the ease of reaching or accessing services, or economic and social opportunities for the user (household, farm, and village), the travel distance or time to the nearest facility or a selection of facilities, or how many markets are within a given travel time or travel effort. The ensuing discussion focuses on accessibility from the demand perspective, with particular emphasis on the issue of physical accessibility as a measure of the degree of market integration and its influence on the economics of agricultural production. Accessibility is also a useful measure of the degree of integration of a given location with respect to social opportunities, such as educational, health, and other public services, or, conversely, a measure of isolation and deprivation. Three indicators are available to measure the physical accessibility of rural farms to markets (UN Statistics Division 1997): (1) the equity index, which is the distance or travel time/cost to the nearest markets; (2) the covering index, which is the number of markets that are accessible within a given travel time/cost; and (3) the average travel time/cost index, which is the average time taken or cost/effort required to get to a number of the target market(s). A fourth accessibility indicator, the potential accessibility index, is a more general measure of the degree of interaction or integration of the location of interest with target locations, and is based on a distance- and size- weighted sum of the distances to target locations. For example, one can compute an index by summing the population of towns in the vicinity of a village, weighing each town’s population by an inverse measure of its distance from the village. Such an index would indicate the extent of access of the village to contacts and social opportunities, and to informal information networks and market intelligence. All indices require reliable estimates of travel distances (dij) along available travel routes between an origin location i and a target location j. These distances can be computed by using the network analysis capability of vector-based GIS by sum-

448 Edmonds and Kam ming the distance of all line segments constituting the route between the origin and target locations. This requires an infrastructure map that is properly formatted as a topologically consistent network connecting the points of origin (farms/villages) with the target or destination points (markets, schools, etc.). In remote areas, not all locations are served by roads, or sometimes details of tracks are not provided in small- scale road maps. Auxiliary information such as topography, hydrography, and land cover can be used to model the “least-cost” pathway access from a remote location to existing mapped roads. The least-cost pathways identified in a GIS can also be used to supplement sparse road networks and provide travel distance data for better estimation of accessibility in areas with poorly developed transportation infrastructures. Where the quality of transport infrastructure varies across the network and enables different ease of transit across it (e.g., dirt roads must be traveled at slower speeds than well-maintained paved roads, so that the dirt roads have a higher “imped- ance”), accessibility measures based solely on travel distances are poor indicators of accessibility. One way to incorporate differences in the ease of transport along different segments of the network is to “tag” the travel speed or travel times to these segments, and compute the total time taken to traverse along the network from the origin to the target location. The term “travel cost” is used to refer to the distance, time, or financial cost to get from an origin to a target location. The ease of transporting goods from the farm gate to the market should reflect the overall ease of transport between the farm and all the available local markets, rather than the distance or time to transport goods to the market that happens to be used by the farm at one time. Accessibility indicators calculated using several reference markets are preferable because they reflect the exogenous availability of markets in a particular area rather than the endogenous outcome of a farm’s choice to buy or sell goods/factors at a particular market. GIS-based analysis allows for easy computation of quantitative indicators that reflect overall accessibility from the farm (or village) to multiple markets. The average travel cost index is an example of this type of indicator. The average travel cost index Di for an origin location i is the mean travel cost between the origin location (e.g., village) and a number (J) of target locations (e.g., markets):

J Σ Di = dij /J j = 1 where Di = average distance (in km) between farm i and the J target market(s) and dij = length/distance of the line segment k (in km) between the village and market j. Closely related to the effect transport costs play in determining the level of farm production and the extent of subsistence orientation of farms, transport costs also determine what crops farms cultivate and the intensity of land use on the farm. For example, farms in remote areas may be prevented from economically cultivating fresh produce by the cost and speed required to transport output to the market. Due to the ease of storing rice, the commodity is often particularly favored in remote areas. Agricultural marketing channels in remote areas often focus on rice. As a result, even

Integration of biophysical and socioeconomic constraints . . . 449 farms that have suitable conditions for growing higher value crops may nonetheless cultivate rice because of transport concerns. We return to this hypothesis shortly, when we review an accessibility analysis for road and river transport systems in the Mekong River Delta.

Land-use dynamics in the Mekong River Delta: an illustration of integration The increase in rice production in Vietnam during the 1990s represents a success story in Asian agricultural development. Increases in rice production in the Mekong River Delta, which supplies about half of Vietnam’s total rice production, averaged about 6.3% per year during the nineties, according to government statistics (Govern- ment Statistical Office 1998). These increases took the country from having a large deficit between rice demand and supply to becoming the third largest rice exporter worldwide. This expansion contributed to Vietnam’s high rate of GNP growth by providing urban areas with cheap food and generating foreign exchange earnings. Although the national statistics on rice production in Vietnam are widely known, there have been few studies of the farm-level changes in rice production techniques and land-use changes that have led to production increases. This research provides insight into the farm-level changes in agricultural production that, when aggregated, caused the production increases. We make use of previously collected longitudinal farm-household survey data and existing biophysical characterization of the Mekong River Delta. Farm-level changes in rice output with production, land use, and supply estimates are examined by using data from longitudinal household survey data that solicited farm production information for each year between 1994 and 1997. The survey covered about 150 farms from 8 villages in the Mekong River Delta and 2 villages from river basin areas in Dong Nai Province. Sampled villages represent a range of agroecological and production situations. Because of nonreporting of some villages, and to a lesser extent farm attrition from the survey, the sample size varies over time. The data were collected by the Institute of Agricultural Sciences of Viet- nam.1 Because of developments in water management infrastructure in the area, it offers an ideal context in which to examine the effects of changes in water availability on agricultural production at the farm level. Different villages taking part in the longitudinal survey can be taken to represent different stages of development between rainfed rice agriculture and irrigated farming. We can trace farm-level changes accompanying the transition from a rainfed to irrigated agricultural rice system using the data. Another major development in the study area during the 1990s was the deep- ening and geographic extension of market reforms begun in 1989. These changes make the area and time period ideal for economic study.

1The data were originally collected for a research project involving the Institute of Agricultural Sciences of Vietnam and the Faculty of Agricultural Sciences of Gembloux, University Mons-Hainaut, Belgium.

450 Edmonds and Kam Description of the study area We completed an extensive descriptive analysis of the Mekong River Delta using available survey, secondary-source, and GIS data. In this review, we considered characteristics such as the following biophysical variables: location and accessibility of rice producers and markets, soils, rainfall and temperatures, and the seasonal flooding situation and problem of saltwater intrusion on farmland. We relied on GIS data compiled by IRRI and its collaborators for most of this analysis. We also examined survey and secondary price data to consider changes in agricultural policies, real prices of rice inputs and outputs, and market development in the Mekong River Delta in the 1990s, and to characterize the demographic characteristics and resource endowments of surveyed farms. Extensive work had previously been carried out using the survey data to describe the evolution of rice agriculture, and in particular the evolution of costs and revenues accruing to rice farmers in the region during the 1990s (IAS 1997, 1998). Here we report only the small portion of this work that relates directly to the GIS techniques explained earlier. Table 1 reports selected descriptive statistics from the data set used in the study. Figure 3 superimposes land use reported by farms in the eight surveyed villages on a land-use map for the Mekong Delta (circa 1996). The figure helps to highlight the benefits of integrating GIS with farm survey data. The base map provides a complete characterization of land use, while information from the farm-level survey adds a time dimension and provides detailed data on farm resources and activities. The figure shows the high level of correspondence between land use captured from remote sensing presented on the map and that reported by farms completing the longitudinal survey. We generated rainfall indicators for each of the villages from which households were selected for the longitudinal survey using weekly rainfall data from 24 weather stations in the Mekong River Delta. In this initial analysis, we worked with aggregate annual rainfall as a measure of fresh water available to nonirrigated rice plots. The amount of rainfall is important not only as a water source for rainfed crops, but also because of its influence on flood levels and saltwater intrusion in the dry season. Across the Delta, rainfall is heaviest in the far southwest coast of the peninsula and tends to decline as one goes from the Ca Mau area to the northeast (toward Ho Chi Minh City). The highest total annual rainfall in 1996 (a heavy rainfall year) was about 3,100 mm, whereas the lowest rainfall reported that year was about 1,650 mm. We used linear spatial interpolation to generate rainfall measures for the eight localities where the survey was conducted. The technique takes the weighted average of reported rainfall at all the weather stations surrounding the surveyed villages, with the weight assigned to each weather station being inversely proportional to the distance between the station and the selected village. Our description of the study area included an accessibility analysis of the eight surveyed villages in the Delta. Figure 4 shows the main transportation routes in the area. The principal routes used by the farms interviewed for the longitudinal survey are shown in green (surface-water routes) and black (road and ferry routes). The accessibility indicators calculated for the surveyed farms in the Mekong River Delta

Integration of biophysical and socioeconomic constraints . . . 451 0 1

1 85

0 1

0 1 0 1

0 1

21 85

continued on next page

0.51 – 0.10 –

42.3 19.2 53.0 15.4

0.21 –

0.329 –

0.577 –

0.040 –

1.8 0.79 1.9 0.69 1.9 0.76 2.1 0.58 0 3

1.3 0.69 1.5 0.79 1.1 0.45 1.0 0.10 1 5

– – 5,459 5,901 5,541 5,486 5,405 5,106 0 32,060

1994 (n = 89) 1995 (n = 149) 1996 (n = 122) 1997 (n = 105) All years

0.36 0.31 0.40 0.52 0.34 0.45 0.40 0.48 0 3.96

0.91 0.63 1.22 0.79 1.15 0.70 1.07 0.64 0.13 4

0.88 0.55 1.00 0.61 0.92 0.62 0.79 0.41 0.125 3.5

Mean S.D. Mean S.D. Mean S.D. Mean S.D. Min. Max.

3,841 1,382 5,288 1,490 5,670 1,073 5,023 1,377 1,053 9,000

–1

worker

–1

no.

y y 0/1 dummy

0/1 dummy

ha no. Quality-adj. ha 1.14 1.11 1.47 1.92 1.30 1.81 1.34 1.91 0.04 17.84 0/1 dummy 0/1 dummy

kg ha ha

ow crop 0/1 dummy 0.12 – 0.55 – 0.59 – 0.56 –

a oduction in 1,000 MT – – 10,529 5,430 13,792 3,615 11,539 5,939 0 18,032

m during year kg

b b

in area

member (primary)

(secondary)

(postsecondary) household

winter

province

saltwater

Cultivated nonrice/nonr Paddy sold by far

Table 1. Descriptive statistics from survey, secondary, and GIS data. secondary, 1. Descriptive statistics from survey, Table

Variable Unit

Number of reporting villages 6 of 10 10 of 10 8 of 10 9 of 10

Years since family settled Years Age of head of household Most educated household

Most educated member

Most educated member Total persons residing inTotal Individuals 4.7 1.52 5.8 1.79 5.8 1.71 5.6 1.66 2 13

Land-labor ratio in household

Total land owned by family Total Farming plots cultivated by family Quality-adjusted landholding size Alluvial soil

Paddy yield during winter-spring Area cultivated to rice autumn-

Total yearly rice pr Total

Rice-cropping intensity

Medium-slightly acid sulfate soil Household demographic characteristics Landholding and biophysical characteristics Saline soils with dry-seasonRice production, marketing, and land use 0/1 dummy

452 Edmonds and Kam 0 1

0 1

0 1 0 1

0 1 0 1 0 1

0 1

51 255

12 28 11 28

Soil, water availability, and accessibility measures were derived Soil, water availability,

0.28 –

0.12 – 0.13 –

0.11 – 0.10 – 0.61 –

19.8 4.6 19.5 4.7

0.26 –

130.2 70.2

Calculated as the weighted average (by quantity of rice sold) sale price of rice reported by surveyed farms.

160 87 160 74 169 71 149 73 0 533

1994 (n = 89) 1995 (n = 149) 1996 (n = 122) 1997 (n = 105) All years

0.13 0.03 0.17 0.04 0.14 0.02 0.15 0.02 0.09 0.30

0.14 0.01 0.18 0.02 0.15 0.01 0.14 0.01 0.12 0.20

Mean S.D. Mean S.D. Mean S.D. Mean S.D. Min. Max.

0.244 0.028 0.236 0.022 0.248 0.016 0.196 0.023 0.123 0.337 0.355 0.009 0.302 0.005 0.232 0.014 0.187 0.005 0.175 0.369

–1

–1 –1

vey that did not include all households later interviewed for the longitudinal survey, and (2) interpolation or overlay of values generated values of overlay or interpolation (2) and survey, longitudinal the interviewedfor later households all include not did that vey

–1

kg ha

’97 US$ kg

0/1 dummy 0.20 – 0.35 – 0.37 – 0.43 –

0/1 dummy 0.52 – 0.40 – 0.34 – 0.24 –

’97 US$ kg

0/1 dummy

mm 1,251 66 1,616 216 1,863 107 1,513 145 1,174 2,076

0/1 dummy 0/1 dummy 0/1 dummy 0/1 dummy 0/1 dummy

km km

min

0/1 dummy

b b

ucture b b b

ea ’97 US$ kg

, practices, and inputs

) water >6 mo

ried out on-farm 0/1 dummy 0.38 – 0.28 – 0.29 – 0.38 –

–1

year

–1

local markets

year

during year

nonglutinous rice

(yearly av)

village

nearestmarket

drying court

Numbers reported come from (1) a baseline sur baseline a (1) fromreported come Numbers

Table 1 continued. Table

Variable Unit

Av sale price of paddy during Av

Av local market paddy price Av

Traditional nonglutinous riceTraditional 0/1 dummy 0.18 – 0.11 – 0.11 – 0.07 – Modern short-duration

Modern medium-long-duration

Urea ha Price of urea (weighted yearly av)

Local market price of ur

No. mechanized tractors used

Land leveling car Dike construction on farm 0/1 dummy 0.48 – 0.36 – 0.34 – 0.39 – 0 1 Interpolated annual rainfall at

Flooding 0.5–1 m lasting 3 mo

Distance to nearest local market Accessibility index—distance to

Accessiblity index—time to all

from GIS coverages available in Mekong Delta provinces only. from GIS coverages available in Mekong Delta provinces only.

Whether homestead owns

from GIS coverages. Number of observations for particular variables can vary from general sample sizes reported.

Brackish (>4 g L Rainfed farm (no irrigation)

Limited irrigation available Agricultural technology management infrastr Water Reliable irrigation on-farm Market accessibility and travel distances a

Integration of biophysical and socioeconomic constraints . . . 453 Fig. 3. Land-use map in the Mekong River Delta in the 1990s. Land-use map and land use reported by surveyed farms. Source: V.Q. Minh, Soil Science Department, College of Agricul- ture, Can Tho University, Can Tho, Vietnam.

454 Edmonds and Kam Cambodia

Fig. 4. Main transport routes between surveyed villages and main markets. Source: IAS-IRRI study.

Integration of biophysical and socioeconomic constraints . . . 455 apply the measures based on travel distance and time between single markets (the nearest local market to the farm and Ho Chi Minh City) and average distances/times for transport between the farm and all surrounding markets. The accessibility analysis and calculation of accessibility indicators were already discussed, so we merely report the results of the analysis in Table 2.

Analytical model of land use To explore the relationship between biophysical and socioeconomic characteristics, to derive hypotheses, and to form an estimation equation that can be tested using available data, we develop a spatial land-use model following standard microeconomic analysis. Spatial economic models assert the importance of the spatial location of agents to market centers, economic infrastructure, and to one another in determining the economic activities pursued by the agents. They offer a good analytical framework for considering the effects of biophysical characteristics of a parcel of land and socioeconomic characteristics of the farm operator(s) on land use. Building upon the insights of von Thünen (1826) and more recent work of Chomitz and Gray (1995), our model examines the effect of travel distances between farms and markets on cropping patterns and land-use intensity of farms. It is beyond the scope of this chapter to review the model in detail, but we outline its structure and development. We begin by assuming that land will be used for the activity that gener- ates the highest rent given the physical characteristics of the plot (local climate, basis of land tenure, labor available for farming) and that farm-gate input and output prices depend upon the cost of transport.2 We define a revenue function for each alternative use of the plot. In notation, we define

Rik = PikQik(Pik,Cik,Zi) – CikXik(Pik,Cik) + uik (1)

where Rik gives the rent on plot/point i in use k, Pik is the price of output/crop k at plot/ point i (farm-gate price of k), Cik is a vector of prices of inputs needed for production of crop k at plot/point i, Zi is a vector of fixed characteristics of the plot that influence the land’s production efficiency in use k, Xik is the optimal input level for production of crop k per unit of land at point i, Qik is the potential output of crop k at plot/point i (potential production), and uik is a random disturbance term.

2Several other assumptions also underpin the model. The model assumes that land use is reversible—that once land has been applied to a particular crop it is possible to alter the use of the land in the future. The basis of land tenure affects only the profit the farm operator obtains from cultivation of each crop. This assumption can be captured in the model by defining the farm-gate price for each crop to be net of any rental costs that the farm operator must pay. The model also requires strong assumptions concerning the farm operator’s expectations about future prices of goods that can be produced on the land and of inputs required to produce those goods.

456 Edmonds and Kam 51 36 29.0

1 2 (km)

Accessibility index

181 109 47.9

188 124 69.2

Time Time Distance

o,53 47 37.0

rang 255 246 44.0

, My Tho 126 82 37.6

To all local markets To

Rach Gia, Can Tho, Sa Dec

Minh City

Vi Thanh

Ca Mau

Demand/supply markets

Bac Lieu, Ca Mau, Soc T

Duc Hoa Town, Ho Chi Minh City Duc Hoa Town,

Time of

1 2

travel (min)

60.6 165 120 An, Ho Chi Minh City Tan

39.7 92 69

To Ho Chi Minh City To

Distance

41.4 67 46

Time of

travel (min)

1 2

To nearest local market To

market (km) Time Time (km) Time Time

Name of Distance

via road 22.0 90 60 via canal 29.6 118 89

via road 17.2 30 30 294.6 455 308 via canal 16.9 180 180 294.3 605 458

(bridge)

Phung Hiep 28.5 114 69 206.6 420 265 Can Tho, Phung Hiep, Rach Gia,

Tan AnTan 12.4 30 30 57.5 97 75 My Tho An, Ho Chi Minh City, Tan 65 51 29.8

Duc Hoa Town to HCM Town

Chi to HCM

Duc Lap Ha to

Duc Lap Ha to Cu

Duong Xuan Hoi Thuan My An Tan

Thanh Quoi Thot Not 22.7 30 27 194.0 298 216 Thot Not, Thanh Thang, Lap V

Vinh My A Bac Lieu

Hoa Khan Tay Duc Hoa Town 17.9 36 27 46.8 94 70 An, Ho Chi Tan Duc Hoa Town, 108 64 41.2

Duc Lap Ha Duc Hoa Town 16.6 34 25

Thanh Xuan Cai Rang Town 21.2 85 51 196.0 378 240 Cai Rang, Can Tho, Phung Hiep,

Hoa An

Table 2. Accessibility indicators for surveyed villages—distance and travel times. Table

Village (min)

Integration of biophysical and socioeconomic constraints . . . 457 The prices of inputs and outputs in the revenue equation depend upon the distance of the land from the market. Prices of inputs increase at an increasing rate as farms move farther away from markets. Similarly, the prices farms can obtain for output they sell are assumed to decrease at an increasing rate as the farm is more distant from markets. There are two components to this distance—the distance between farm plots and the homestead, and the distance between the homestead and the reference market(s). In the Mekong River Delta, transport of paddy and rice inputs is largely by boat. Settlement patterns in the area usually focus on canals—with homesteads usually bordering a canal or river. We specify a functional relationship between the level of input applied to farming and the amount of output produced by the farm. The level of output produced depends upon input levels, agroclimatic conditions, and other fixed land characteristics. Using the production function and the expressions for net revenue associated with cultivation of each crop, we can derive relationships between the factors specified as determining net revenue and production, and the demand for inputs by the farm. The demand for inputs for crop k cultivated at location i is a function of the cost of the inputs, the farm-gate price of the output, the characteristics of the plot, and the efficiency of production of crop k on the plot. Next, using the expressions for input demand, production, and the effect of travel distances on revenues, we define an expression for the net returns associated with cultivation of crop k on parcel i, which incorporates the effects of travel cost and the production technology of the farm. Two travel distances are considered in the model. Di is the distance between the homestead and the farming plot or plots operated by the family, whereas Ti is the average distance between the homestead and the input/output market(s) accessible to the farm. Both distances are relevant in the model since various inputs used in farming (e.g., labor, fertilizer, seed, etc.) and the outputs produced are transported between homesteads, farm plots, and markets over the course of a production season. The resulting expression establishes the hypothesis that the likelihood that a plot will be applied to cultivation of a particular crop and the intensity of use will fall as the distance between the plot and the output/input market increases. At the extreme, very distant plots will not be cultivated, whereas plots closest to markets would be expected to be used for intensive commercial farming. We form an expression for net revenue from cultivation of crop k on plot i amenable to estimation from earlier equations:

ln(Rik) = a0k + a1kln(Di) + a2kln(Ti) + a3kln(z1i) + a4kln(z2i) + … + aNkln(zLi) + uik (2)

If we add technical assumptions concerning the distribution of error terms (uik) and the correlation of errors, the probability of any crop k being cultivated on plot i is distributed according to the multinomial logit distribution. This provides the basis for using the multinomial logit model in empirical tests of the model. If we can rank the alternative land uses—as is possible when the sample is limited to farms cultivating

458 Edmonds and Kam rice and the model is applied to explain rice-cropping intensity—the model can be modified to take the form of an ordered logit model. Under the model, the coefficients on distances (Di and Ti) are expected to be negative, whereas those on productivity-enhancing land characteristics (sik) are expected to be positive. The magnitude of the estimation coefficients will depend upon per unit costs of transportation of different crops and the relevance of a particular land characteristic to the production of a particular crop. Whether the crop being cultivated on the plot is destined for commercial or subsistence use will also affect the influence of distance on the likelihood that a particular crop is produced. Subsistence crop production is less influenced by distance. Why develop an analytical model? The model provides a framework for empirical estimates of the determinants of land use in the Mekong Delta, but is not intended to provide a complete description of reality. Actual decision making of farms regarding land use is extremely complex—much more complex than the simple static model just outlined. For example, farmers’ land-use decisions incorporate considerations of the dynamic effects of using land for a particular purpose in this period on its productivity in future years under alternative uses. The role of agricultural income in the broader income of the household also influences land use, as do farmers’ concerns about the risk and expected revenue from alternative land uses. However, given the available data with their measurement error, missing data problems, and relatively small sample size, it is necessary to simplify the number of variables included in the analysis and the number of issues the analysis considers. The model provides this focus. To facilitate interpretation of estimation results, modeling the relationships between variables thought to be important is essential. The model establishes hypotheses that can be tested in the estimates and be used to evaluate whether the structure developed by the model is consistent with available data. This parameterization of the relationship between variables and of the estimation comes at the cost of restricting the number of variables and possible relationships between variables that can be considered in the analysis.

Empirical model and hypothesis tests The analytical model just reviewed provides the basic framework we apply in analyzing farm survey data, and establishes the multinomial logit and ordered probit estimators as appropriate for estimating land use and cropping intensity. The form of the estimation equation is given by equation (2). The key variables of interest in estimates are the terms detailing the distances between the homestead and the farming plots, and the distances between the homestead and markets accessible to the farm. The exogenous or predetermined z variables in equation (2) are other household or farm characteristics expected to influence household land-use decisions, and include characteristics of the biophysical environment where farms are located, family characteristics, and variables capturing market conditions in surveyed villages. Standard microeconomic analysis of production and supply also guides the selection of variables and our expectations regarding their signs in estimations, but we do not review

Integration of biophysical and socioeconomic constraints . . . 459 these, and production and supply function estimation results are not reported. The particular set of right-hand-side variables employed in our empirical estimates varies depending upon the relevance of variables to the left-hand-side variable we are seek- ing to explain. In some cases, we had to reduce the number of right-hand-side variables included in the estimation equation in order for the estimator to find a maximum. This resulted from difficulties caused by missing data and the relatively small sample size available (vis-à-vis the estimation procedures applied). Empirical estimates use both cross-sectional and panel data-based estimation procedures. Panel data estimation procedures provide more robust estimates because they can control for the effect of variables that cannot be observed but are known to have a strong effect on estimated outcomes (e.g., household preferences and motiva- tion). Because panel data estimators make use of the full panel of data (rather than single years of the survey), they can measure more precisely the effect of changes in explanatory variables that can explain change in the variable of interest. Despite these advantages, we make use of cross-sectional data-based estimators in our empirical analysis for two principal reasons. First, many of the right-hand-side variables of interest—including our accessibility indicators and most of the variables capturing aspects of the biophysical environment in which farms operate—could be observed only a single time during the years of the survey. Panel data estimators cannot accommodate the use of time-invariant right-hand-side variables in estimation equations. Second, for technical reasons, certain estimation procedures (e.g., the multinomial logit estimator) have not been developed for panel data. This makes it necessary for us to define our land-use categories so that only two alternative land uses are available to apply a panel probit estimator, and to estimate land use defined over more than two categories using single years of the survey with cross-sectional estimators. In our estimates, we categorize cropping patterns and land uses by intensity rankings (e.g., mono-cropping, double-cropping, etc.) and according to type of crop cultivated. Crops are divided into broad categories: (1) rice, (2) upland row crops (e.g., sugarcane, potato, vegetables), and (3) fruit trees or perennial fruit crops (e.g., dragon fruit) or trees maintained by farms for wood (e.g., eucalyptus). We cannot review all our estimation results here, but we do review panel and cross-sectional estimates of rice-cropping intensity in Tables 3 and 4 as examples of the analysis we performed. We discuss the results from the panel data estimators first. Three models are estimated using the random effects probit estimation procedure: (1) whether the farm reported cultivating one or more than one crop per year, (2) farm cultivation of nonrice crops, and (3) whether the farm reported growing fruit trees or other perennial crops. We used a common set of right-hand-side (explanatory) variables in these estimates. Because of the number of parameters that need to be estimated to account for household-specific error terms, we faced significant constraints in the number of right- hand-side variables that could be considered in panel estimates. An additional constraint to the variables used in panel data-based estimators is that they be observed over time.

460 Edmonds and Kam Table 3. Summary of estimates of land use using the random effects probit estimator.

Cultivates Cultivates Farm cultivates Left-hand-side dependent variable more than nonrice fruit or other one crop y–1 crop(s) trees Estimation coefficient 1994-97a 1994-97 1994-97 (estimated standard error) (n = 436) (n = 436) (n = 436)

Land-labor ratio on farm 1.204** –1.058* 0.056 (ha per household laborer) (0.559) (0.578) (0.591) Age of head of household 0.009* –0.050*** 0.014** (0.005) (0.014) (0.007) Farm used short-duration 1.438*** –0.131 –0.018 modern variety seed(s) (0.343) (0.670) (0.452) Farm used medium- or long-duration –0.218 –2.221*** –0.334 modern varieties (0.317)*** (0.775) (0.364) Farm invested in land leveling 0.664 –0.116 –0.902** or other soil improvement (0.462) (0.467) (0.450) Farm invested in construction 0.031 0.646* –0.769* of dikes or irrigation (0.297) (0.381) (0.449) Rho 0.703*** 0.787*** 0.970 (0.153) (0.225) (0.081) Goodness of fit diagnostics: Pseudo R2: Cragg-Uhler 0.085 0.130 0.447 Maddela 0.046 0.064 0.317 McFadden 0.060 0.098 0.309 Likelihood ratio (X2) test 20.653*** 28.693*** 166.549*** [degrees of freedom] [1] [1] [1] Percentage correctly predicted 0.812 0.842 0.672

Actual/predicted 0 1 total 0 1 total 0 1 total

0 0 82 82 342 17 359 107 126 233 1 0 354 354 52 25 77 66 137 203

total 0 436 436 394 42 436 173 263 436 a*** = estimated coefficient is statistically significant at 99% confidence level, ** = estimated coefficient is statistically significant at 95% confidence level, * = estimated coefficient is statistically significant at 90% confidence level.

The variables considered are the land-labor ratio on the farm (area per full-time equivalent family worker), the age of the head of household, a series of dummy variables indicating the main rice variety cultivated (the omitted variety classes are glutinous and traditional nonglutinous rice varieties), and dummy variables indicating farm investment in dikes or land leveling. It is expected that households with lower land-labor ratios will adopt more labor-intensive land use. It is argued that older farm operators are more traditional and hesitant to adopt new technologies and change their agricultural practices. Their reluctance may also reflect their lower levels of education.

Integration of biophysical and socioeconomic constraints . . . 461 Table 4. Summary of estimates of rice-cropping intensity using the ordered probit estimator for cross-sectional data.

Left-hand-side/dependent Rice-cropping Rice-cropping Rice-cropping Rice-cropping variable intensity intensity intensity intensity Estimation coefficient in 1994a in 1995 in 1996 in 1997 (estimated standard error) (n = 60) (n = 114) (n = 114) (n = 77)

Constant 7.2873 –17.884*** 16.599*** –34.870*** (5.559) (4.217) (4.678) (11.088) Average distance between 0.0011 –0.131 –0.062 –0.025 homestead and plot or plots (0.118) (0.138) (0.139) (0.185) Average travel time to all 0.0323 –0.041*** 0.020*** –0.060*** accessible local markets (0.033) (0.011) (0.007) (0.027) Land-labor ratio on farm 0.0308 0.689 0.356 –0.100 (ha per household laborer) (0.992) (0.521) (0.464) (0.834) Years since family settled in 0.0064 –0.007 –0.003 –0.012 current place of residence (0.011) (0.007) (0.007) (0.014) Maximum educational attain- 0.0020 0.311 0.071 0.526 ment of any family member (0.532) (0.296) (0.294) (0.461) Whether farm served by good- 2.2076 2.053*** 4.266*** 4.704*** quality irrigation system (1.513) (0.466) (0.843) (1.668) Annual precipitation where –0.0091 0.013*** –0.011*** 0.025*** farm is located (0.006) (0.003) (0.003) (0.008) Mu (threshold parameter for 0.3267** 2.208*** 1.801*** 4.765** first stratum) (0.149) (0.330) (0.309) (2.182) Goodness of fit diagnostics: Pseudo R2: Cragg-Uhler 0.432 0.633 0.521 0.768 Maddela 0.361 0.555 0.460 0.654 McFadden 0.248 0.386 0.288 0.556 Likelihood ratio (X2) test 26.862*** 92.256*** 70.189*** 81.707*** [degrees of freedom] [7] [7] [7] [7] Percentage correctly predicted 0.800 0.719 0.632 0.805

Actual/ 0 1 2 total 0 1 2 total 0 1 2 total 0 1 2 total predicted 0 37 0 0 37 21 12 0 33 18 13 0 31 6 6 0 12 1 6 0 1 7 4 45 6 55 5 35 11 51 2 38 5 45 2 5 0 11 16 0 10 16 26 0 13 19 32 0 2 18 20

total 48 0 12 60 25 67 30 114 23 61 30 114 8 46 23 77 a*** = estimated coefficient statistically significant at 99% confidence level, ** = estimated coefficient statistically significant at 95% confidence level, * = estimated coefficient statistically significant at 90% confidence level.

The rice variety planted by farms clearly influences the number of rice crops it is possible for the farm to cultivate. The two dummy variables used in the reported estimates define farms growing short-duration modern varieties and medium- or long- duration modern varieties. Rice variety choice is endogenous with the choice of cropping pattern, so estimates are open to endogeneity bias under the present specification. Unfortunately, neither the data nor the estimation procedures available enable

462 Edmonds and Kam estimation of a system of multinomial equations using panel data. The estimates also include farm-specific error estimates and a parameter (Rho) that provides an indicator of the significance of farm-specific error estimates. A statistically significant estimate of Rho supports the inclusion of the farm-specific error terms and use of the random effects estimator. The three models were each highly statistically significant. At the bottom of Table 3, we report several measures of the overall performance of the models in explaining land use. The likelihood ratio test can be interpreted as testing the null hypothesis that the model, as a whole, cannot explain the variation in the left-hand-side variable of interest. All three of the models rejected this null hypothesis at a high level of statistical significance. The psuedo-R2 measures report nonlinear measures of the proportion of variation in the land-use variables that is explained by the overall model. Psuedo-R2 measures vary between 44.7% and 4.6% across measures and models. Lastly, Table 3 reports the share of land-use categories correctly predicted by each model, and the distribution of actual versus predicted land use. T-tests of the null hypothesis that each right-hand-side variable has no effect on a farm’s chosen land use provide the basis for testing the significance of each variable. These are computed using the estimated coefficients and standard errors. The random effects probit estimator is nonlinear, which makes it difficult to interpret estimation coefficients directly. We must approximate the marginal effect of a change in a right-hand-side variable on the probability that a farm chose a particular land use at the mean values of the right-hand-side variables using an approximation algorithm. The estimates of whether the farm cultivated more than a single crop during the agricultural year show that the land-labor ratio, the head of household’s age, and the use of modern short-duration rice varieties each increased the likelihood that a farm multiple-cropped its land. Construction of water control dikes and carrying out land leveling were also associated with an increased likelihood of multiple cropping, but these effects were not statistically significant. The estimated marginal effect of a 1% increase in the land-labor ratio of farms is to increase by 13.3% the likelihood that the farm cultivated more than a single crop per year. An increase of 10 years in the age of the household head was associated with only a 0.1% increase in the likelihood of multicropping on-farm. Farm use of short-duration modern rice was associated with a 15.9% increase in the likelihood that the farm engaged in multicropping. Interpreta- tion of other models reported in Table 3 follows this analysis, but we forego discussion because of space constraints. The model estimates reviewed provide insight into the characteristics associated with the land-use choices made by farms. We find that farm size and, more particularly, the relative abundance or scarcity of family agricultural labor in relation to the land operated by the farm play an important role in driving farm land use. Farms with relatively abundant labor compared with their farm size are less likely to engage in multicropping. The choice of modern rice varieties and the growing duration of varieties chosen are closely related to the broader land-use choice of farms. Finally, investments in farm or plot improvements to water management were also clearly linked to land-use choices. One of the benefits of dike construction, for example,

Integration of biophysical and socioeconomic constraints . . . 463 appears to be the opportunities it creates for farms to cultivate nonrice crops. In the absence of such investments, farms appeared to adopt land-use options (i.e., fruit trees and other perennial crops) more immune to the effects of poor water management. Lastly, the significance of the parameter Rho emphasizes the importance of unobservable farm/household characteristics on land-use options and underscores the complexity of the land-use choices of farms. Our measures of market accessibility and variables characterizing biophysical conditions in the surveyed villages were fixed over time or observed only once, so we are unable to examine the principal hypotheses of our analytical model through the panel estimators. Accordingly, we used cross-sectional estimates of cropping patterns and rice-cropping intensity to examine our hypotheses concerning the effect of accessibility of land use. Rice-cropping intensity is a categorical variable for which the categories have a natural ordering, so an ordered probit estimator is used. The estimates of rice-cropping intensity are significant overall in each of the four years according to the goodness of fit measures reported at the bottom of Table 4. The key variables of interest from our analytical model are the measures of the distance between the villages where farms are located and the average travel time to all local markets accessible to the farm and the distance between the homestead and plot or plots operated by the farm. The greater these distances, the lower the likely rice-cropping intensity to be adopted by the farm. Estimation results provide limited confirmation of the model’s hypotheses. The distance between the farm plot and markets had a negative and statistically significant effect on rice-cropping intensity in model estimates in 1995 and 1997. According to 1995 results, each 10 km of distance between the homestead and accessible markets causes a 4.4% reduction in the probability that the farm cultivated two rather than a single rice crop, and a 6.7% reduction in the probability of cultivating three rather than two rice crops. Similar marginal effects were estimated in 1997. Greater distances between farm homesteads and plots were associated with a reduced probability of intensive rice cultivation by the farm— but estimated parameters were not statistically significant. According to 1995 estimation results, a 10-minute increase in the average travel time between the farm and plots was associated with a 14% and 21% decrease in the probability of cultivating two and three crops during the year, respectively. Across all the land-use models estimated, the distances between the farm and markets and between the homestead and farm plots had the effect of reducing the cropping intensity on-farm. The effect of the distance between the farm and markets was greater in the case of rice-cropping intensity, whereas distances between the homesteads and farming plots caused greater reductions in the general cropping intensity (results not shown). The study estimated other land-use, production, and rice supply functions implied by our land-use model or basic microeconomic theory. These included the following: general cropping intensity, farm cropping pattern, and rice production and farm supply of rice to the market. Together, the estimates provide a clear indication of the factors driving farm land-use, production, and marketing decisions. We summarize the major conclusions that emerge from these estimates.

464 Edmonds and Kam The availability of low-saline irrigation water to farms had a positive and statistically significant effect on the intensity of land use applied by surveyed farms across all our land-use estimates. The magnitude of the effect of high-quality irrigation on cropping intensity was much greater than the effects of other explanatory variables included in the model. Rainfall levels had mixed effects on the cropping intensity of surveyed farms. In years with normal to high rainfall, increased rain was associated with increased cropping intensity. In 1996, however, rains were particularly heavy and higher rainfall in that year was associated with significantly reduced levels of cropping intensity among surveyed farms. The size of families’ landholdings relative to their available labor had mixed signs across estimates and years, but generally supported the hypothesis that the relative scarcity of land to labor leads to more intensive land use. Results showed that rice variety selection was clearly linked to cropping intensity, with the adoption of modern short-duration rice varieties playing a key role in enabling more intensive rice cultivation. Farm-level investments in land leveling or dike construction were shown to increase the likelihood that farms adopted intensive rice agriculture in the panel data-based land-use estimates. Other variables such as the level of education in the household, the age of the household head, or the farming experience of the family did not have consistent statistically significant effects on land use in our estimates. Rice production and supply estimates were able to explain most of the observed variation in the levels of rice output and marketed surplus across surveyed farms. Adjusted R2 coefficient estimates across the production and supply models ranged between 0.70 and 0.93, with panel data estimators performing better than estimators using only single years of the survey data. Variables estimated to have positive and statistically significant effects on the level of output included farm size, the cropping intensity pursued by farms, the amount of hired labor used in crop cultivation, and the level of seed application. The use of modern versus traditional varieties did not have a consistent positive effect on rice production. The principal effect of the use of modern (usually short-duration) varieties appeared to be to enable farms to pursue more intensive rice production. Other variables included in the production estimates such as amount of fertilizer or pesticide applied on the crop had positive and statistically significant effects on rice production in only a few of the production estimates. The estimated price elasticity of supply ranged between 0.145 and 0.319 in year-aggregate supply estimates. The rice price had a consistent positive and statistically significant effect on rice marketed surplus, as did the quantity of urea applied to the crop. When estimates included the price of urea, occasional positive and statistically significant coefficient estimates were obtained. The increasing price of urea during the years of the survey appears to have been dominated by a broader trend toward increasing urea use among surveyed farms during these years.

Simulation model for evaluation of investments The implications of model estimates can be better understood by using them to for- mulate a simulation model to assess the effect of policy changes or investments in infrastructure on land use and rice production. The results of a simulation model

Integration of biophysical and socioeconomic constraints . . . 465 derived from our empirical estimates are summarized in Tables 5 and 6. Table 5 shows the distribution of rice mono-, double-, and triple-cropping among surveyed farms across the years of the survey. The actual distribution of farms in each of the four years is shown, along with the projected distribution (using model estimates reported in Table 4) under alternative scenarios. One scenario involves improvements in travel networks between surveyed villages and local markets. The second considers the effect of land transport improvements or land consolidation that brings homesteads and farm plots closer. The third contemplates extension of water management infrastructure to an additional 10% of the surveyed farms. We use results from production function estimates, combined with the implied changes in the share of farms double- or triple-cropping rice, to calculate an implied change in aggregate rice output across farms under the different scenarios. The production estimates provide a measure of the average change in annual rice yield associated with mono-, double-, or triple-cropping of rice. The changes in total rice production from the scenarios are detailed in Table 6. The simulation model shows the large effect of investments in irrigation extension on rice production levels, and the more moderate effects obtained from improvements in the transportation system or land consolidation. It shows how the results of land-use and production estimates based on the observed behavior of farms can be used to assess the likely effects of different investments on rice production levels. The integration of behavioral parameters from econometric estimates is broadly applicable and can provide a needed empirical basis for larger simulation models. In- corporating the estimates obtained in this research with other linear programming or general simulation models would be an important application of this research.

Conclusions Review of the IAS-IRRI research on land use in the Mekong River Delta illustrates many of the geo-informational techniques proposed for integrating analysis of biophysical and socioeconomic constraints to rainfed rice production. The study combines farm survey and GIS data and goes beyond description in exploring the causal relationships between the two types of factors. This study relied on existing sources of data that were originally collected for a cost-price accounting study or for general characterization of biophysical conditions in the Mekong River Delta. Because of this, we encountered severe data constraints in carrying out our analysis. Despite these limitations, the research has provided insights into the farm-level changes in land use and production systems that enabled rice production to increase in the Mekong River Delta in the 1990s. The quantitative analytical techniques developed in this research offer promise for applications in future research.

466 Edmonds and Kam system

system

if 10% increase in

Predicted distribution

good-quality irrigation

Predicted distribution

good-quality irrigation

reduced by 1 km

if distance between no. of farms served by

by 10 minutes

to market is reduced home and plot(s) is

if average travel time

Predicted distribution Predicted distribution

to market is reduced home and plot(s) is

if average travel time

Predicted distribution Predicted distribution if 10% increase in

0.3 3.0 5.2 0.0 –0.9 1.0 –2.1 0.4 7.6 4.9 31.5 2.6

a a

Actual distribution Actual

44 85 61 51 44 86 64 51 43 86 60 51 47 87 75 52

1994 1995 1996 1997 1994 1995 1996 1997 1994 1995 1996 1997 1994 1995 1996 1997

total production

Numbers in table refer to number of farms.

Columns may not sum to total rice production due to rounding error.

Table 5. Simulation of effects rice-cropping of investments on distribution of farm intensity. Table

Rice-cropping pattern

Mono-cropping 37 33 31 12 36 21 26 12 40 29 33 12 15 13 0 12 Double-cropping 7 55 51 45 7 62 51 45 6 57 51 44 8 67 52 38 Triple-croppinga 16 26 32 20 17 31 37 20 14 28 30 21 37 34 62 27

Table 6. Simulation of effects of infrastructure investments on rice productionTable (in metric tons) among surveyed farms.

Rice-cropping pattern

Mono-cropping 25 22 11 7 24 14 10 7 27 20 12 7 10 9 0 7 Double-cropping 5 41 24 28 5 47 24 28 5 43 24 28 6 50 24 24 Triple-cropping 13 22 26 16 14 26 30 16 12 23 25 17 31 28 51 21 Total rice production Total Percentage change in a

Integration of biophysical and socioeconomic constraints . . . 467 References Anselin L. 1988. Spatial econometrics: methods and models. Boston, Mass. (USA): Kluwer Publishing. 300 p. Chomitz K, Gray D. 1995. Roads, lands, markets, and deforestation. Policy Research Working Paper (WPS1444). Washington, D.C. (USA): The World Bank. Deichmann U. 1999. Spatial scale and resolution in the analysis of socioeconomic and demographic data. In: Kam SP, Hoanh CT, editors. Scaling methodologies in ecoregional approaches for natural resources management. Paper presented at the Workshop on Scal- ing Methodologies in Eco-Regional Approaches for Natural Resources Management, 22-24 June 1998, Ho Chi Minh City, Vietnam. Limited Proceedings No. 1. Manila (Phil- ippines): International Rice Research Institute. p 11-19. de Janvry A, Fafchamps M, Sadoulet E. 1991. Peasant household behavior with missing markets: some paradoxes explained. Econ. J. 101:1400-1417. Edmonds CM, Kam SP. 1999. Geo-informational techniques in socioeconomic characterization. Paper prepared for presentation at the Workshop on Geo-Informational Techniques in Agricultural Research, 22-24 May 1999, at the Uttar Pradesh Remote Sensing Appli- cations Center, Lucknow, India. Government Statistical Office (Integrated Statistics and Information Department). 1998. Socio- economic statistical data of 61 provinces and cities in Vietnam. Hanoi (Vietnam): Statis- tical Publishing House. Hectschel J, Lanjouw JO, Lanjouw P, Poggi J. 1997. Combining survey data with census data to construct spatially disaggregated poverty maps: a case study of Ecuador. Preliminary draft. IAS (Institute of Agricultural Science). 1997. Competitiveness of rice channel in Mekong Re- gion. Research report on “Competitiveness of rice production in Mekong River Delta project.” Ho Chi Minh City (Vietnam): IAS. 103 p. IAS (Institute of Agricultural Science). 1998. Annual report: comparative analysis of economic efficiency in rice production 1994-1997. Research report of the “Competitiveness of rice production in the Mekong River Delta project.” Ho Chi Minh City (Vietnam): IAS. 54 p. UN Statistics Division. 1997. Accessibility indicators in GIS. Report of the Department for Economic and Social Information and Policy Analysis. 24 p. von Thünen JH. 1826. Der Isolierte Staat in Beziehung der Landwirtschaft und Nationalökonomie. Translated in Peter Hall, ed. Von Thünen’s Isolated State. Oxford: Pergamon Press. 1966.

468 Edmonds and Kam Notes Authors’ address: Social Sciences Division (GIS ), International Rice Research Institute, DAPO Box 7777, Metro Manila, Philippines. Acknowledgments: This research would not have been possible without the support and assistance of several other individuals: (from the Institute of Agricultural Sciences of Viet- nam) Prof. P.V. Bien, H.T. Quoc, H.C. Viet, and T.T. Khai; (from IRRI) C.T. Hoanh, T.P. Tuong, and L. Villano; (from Can Tho University) V.Q. Minh, and (from the Sub-Insti- tute for Agricultural Planning and Projection) Dr. N.V. Nhan. Nonetheless, errors and omissions are solely the responsibility of the authors. The research was funded in part by the Rockefeller Foundation Social Science Research in Agriculture program. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

Integration of biophysical and socioeconomic constraints . . . 469 Regional land-use analysis to support agricultural and environmental policy formulation

B.A.M. Bouman, R. Roetter, R.A. Schipper, and A.G. Laborte

Characterizing rainfed environments is not a goal in itself, but it depends on the type of information to be generated. It needs to be based on a sound understanding of the prevailing biophysical and socioeconomic processes at the field, farm, or regional level. Conflicts in land-use objectives such as food security, farmers’ income, employment, and environmental protection are one reason for local governments to initiate active land-use policies. Charac- terizing socioeconomic and biophysical conditions is an essential part of methodologies for land-use studies that address policy design, formulation, and implementation. Methods and data requirements depend on whether information is needed for identifying current land-use problems, projections of current trends, land-use explorations under changed policies, or feasible interventions to achieve identified objectives. In this chapter, we introduce regional characterization as required for resource assessment and description of production activities in methodologies for future-oriented land-use studies. This is illustrated by two different case studies: (1) an exploratory study on Ilocos Norte Province, Philippines, and (2) a predictive study on the northern Atlantic Zone, Costa Rica. Both studies use optimization models with quantified input-output relations of production activities as input. All calculations refer to land units considered as homogeneous in biophysical and socioeconomic conditions. Within the modeling framework, (multiple-goal) linear programming techniques, technical coefficient generators, and geographic information systems are applied. In the exploratory study for Ilocos Norte, the focus is on opportunities for increasing food security and income if water constraints could be partially removed by either water sharing or expansion of irrigated areas. The aim of the northern Atlantic Zone study is to predict land-use changes as affected by the introduction of policy measures that stimulate forest conservation and reduction of biocide use. Both types of study aim at identifying options and quantifying trade-offs among conflicting goals. The type, accuracy, and spatial resolution of required input data differ considerably, however, according to agroecological diversity and different study objectives. Both exploratory and predictive land-use studies have in common

Regional land-use analysis to support agricultural and environmental . . . 471 that they synthesize fragmented agricultural knowledge and integrate data on resources over time and space. In rainfed rice areas, the high temporal and spatial variability of production resources complicates the analysis. Farm- ers’ diverse responses to climatic and economic risks must be taken into account, which eventually demands stronger links between on-farm research and operational research for meaningful policy formulation and implementation.

Biophysical and socioeconomic characterization of land is a first step in support of agricultural policy formulation for regional land use and development. In general, policymakers face four major questions (Bouman et al 2000): (1) What are current problems and bottlenecks to development? (2) What are future problems and bottlenecks to development if the current trends in land use are projected forward? (3) What are options for land use in the future based on expected technological change? and (4) What are effective policy measures and interventions to satisfy certain policy goals? Traditionally, issues in the debate on the development of the agricultural sector centered on food security, income by food producers, and labor employment (Timmer et al 1983, Pinstrup-Andersen and Pandya-Lorch 1995). More recently, concerns of sustainability and environmental protection have entered the debate (Rabbinge and Van Latesteijn 1992, Kuyvenhoven et al 1995). Therefore, to help policymakers address the broad questions raised above, tools and methodologies are required that are capable of quantifying trade-offs that occur between various development objectives in general and between economic-, sustainability-, and environmental-related ones in particular (Crissman et al 1997, Roetter et al 1998). Moreover, such tools and methodologies should be specifically geared toward answering well-defined policy questions such as the ones formulated above. The first policy question, that is, current problems and bottlenecks to regional development, is often addressed through farming systems analysis (e.g., Lucas et al 1999) or regional characterization. Examples of the latter are given by various authors in this volume (Singh VP et al, Van Nguu Nguyen, Amien and Las, Borkakati et al, Saleh et al, this volume). The answer to the second question, that is, likely future problems and bottlenecks to regional development under ceteris paribus conditions, can be derived from trend extrapolation using projective land-use models. An example of a projective land-use model is CLUE: conversion of land use and its effects (Veldkamp and Fresco 1997). In this model, the current distribution of land use is explained by biophysical and socioeconomic land-use drivers on the basis of statistical regression analysis. Examples of land-use drivers are climate, soil, population, and degree of urbanization. Likely future land use patterns are then generated by changing the values of land-use drivers through extrapolation of past trends or according to expected changes (e.g., changes in population growth). The projection of land use based on current relationships between land use and its drivers only allows looking into the future up to a certain extent. Discontinuities in

472 Bouman et al trends (e.g., technological progress, expansion of infrastructure) cannot be taken into account. Therefore, to answer the third policy question, that is, options for land use in the future, exploratory land-use models are required that do not rely on the past as the sole measure for the future. Recently, concepts for exploratory land-use analysis have been developed that combine political and societal desires with technical possibilities of development (Rabbinge and Van Latesteijn 1992). Building on such concepts, methodologies and tools have been developed that allow (1) new land-use systems that are currently not practiced to enter the analysis, (2) effects of possible changes in resource availability and infrastructure to be evaluated against the achievement of policy objectives, (3) explicit optimization of land use toward well-defined policy objectives, and (4) quantification of trade-offs among the various dimensions of sustainability and socioeconomic parameters (Rabbinge and Van Latesteijn 1992, Bouman et al 1999, Roetter et al 1998). Typically, these methodologies make use of linear programming techniques for optimization and geographic information systems (GIS), and expert systems, simulation models, and so-called technical coefficient generators to calculate inputs and outputs of land-use systems. Finally, to answer the last type of policy question, that is, effective agricultural policies, again requires a different approach. The effects of policy measures can be evaluated with predictive land-use analysis methodologies that specifically address the behavior of the ultimate decision-makers in land-use—the farmers. Recently, predictive methodologies have been developed that combine the tools of exploratory land-use analysis as mentioned above with econometric farm household modeling (Kruseman et al 1995, Kuyvenhoven et al 1995). In the optimization model, a utility function is optimized that describes individual farmers’ behavior. Though individual farm models can be aggregated to obtain a regional model, such techniques are cumbersome and presume that the number of farms and farm types within a region remains constant over time (e.g., Roebeling et al 2000). Another approach uses the same tools and techniques, but models aggregate behavior of a region instead of that of single farms (Schipper et al 2000). Characterization of socioeconomic and biophysical conditions is an essential part of methodologies for land-use studies that address policy design, formulation, and implementation. Such studies integrate biophysical and socioeconomic data to explore the opportunities for and major constraints to adoption of field- and farm- level research on sustainable production systems. In this chapter, we describe the requirements of regional characterization for resource assessment and description of production activities in methodologies for future-oriented land-use studies. This is illustrated for exploratory and predictive land-use modeling on a regional scale. First, we present a generic framework and discuss the role of information derived from regional characterization. Then, we give examples of application of the framework for a case study on land-use explorations in Ilocos Norte Province, Philippines, and for a predictive case study in the Atlantic Zone of Costa Rica. Finally, we discuss possibilities for applying the framework to typical rainfed lowland areas.

Regional land-use analysis to support agricultural and environmental . . . 473 Framework for exploratory and predictive land-use analysis The generic methodology for exploratory and predictive regional land-use analysis consists of a (multiple-goal) optimization model, an “engine” to calculate inputs and outputs of land-use systems, and a GIS (Fig. 1). This methodology was named SOLUS (sustainable options for land use) by Bouman et al (1999a) and Schipper et al (2000). The same elements also form the building blocks of the land-use planning and analysis systems (LUPAS) developed for exploratory studies in four regions of Asia (Roetter et al 1998). Regional characterization or land evaluation in the widest sense includes the assessment of resource availability and land suitability and quantification of the input-output relations of the various land-use systems (Laborte et al 1999).

Optimization The optimization model is constructed using linear programming techniques. It selects land-use systems for the area under consideration by optimizing toward a specific goal. This goal may be the maximization of economic surplus in the agricultural sector, maximization of employment, minimization of certain environmental effects, or maximization of some household utility function. The optimization model may also be of the multiple-goal type in which subsequent optimizations are performed toward different goals. An optimization goal in linear programming models is implemented by a so-called objective function. Optimizations are performed under constraints, which may be absolute, for example, no more land can be used than is available in the area, or normative, for example, minimum threshold boundaries may be imposed on the production of a given crop or on certain sustainability parameters. The optimization of a certain objective function under a set of coherent constraints, and using a specific set of land-use systems to choose from (see below), is called a scenario. Trade-offs between economic and sustainability objectives are quantified by running the model for different scenarios.

Geographic Geographic Maps data information system

Technical Problem Nongeographic coefficient definition data generator

Linear programming Tables, Scenarios model reports

Analysis and user interaction

Fig. 1. Schematic illustration of the land-use exploration framework SOLUS. Boxes are models/tools, ovals are data, solid lines are flow of data, dotted lines are flow of information.

474 Bouman et al Land-use systems Land-use systems are a combination of a specific land-use type (roughly a specific manner of cultivating a certain crop type) with a specific land unit (a piece of land). Land-use systems are fully specified by their technical coefficients, that is, their inputs and outputs such as yield, costs, labor use, and sustainability indicators. Every use of land is associated with effects on its resource base and on the environment. Examples are soil loss by erosion and mining of the soil nutrient stock (resource base), and emissions of nutrients, greenhouse gasses, and biocides (environmental pollution). Such effects are quantified by so-called sustainability indicators. For a scenario analysis to be meaningful, the optimization model should be able to select from a large number of alternative land-use systems. Therefore, an “engine” is required to calculate the technical coefficients of many alternatives, often called a technical coefficient generator (Hengsdijk et al 1999). Two types of land-use systems can be distinguished: actual ones and technologically potential ones. Technical coefficients of actual systems are derived from farm surveys and describe actual systems being practiced in the area under consideration (Jansen and Schipper 1995). Techni- cal coefficients of technologically potential systems can be computed using the so- called target-oriented approach (Van Ittersum and Rabbinge 1997): production targets are predefined and all required inputs are subsequently calculated. The calculations can be done under predefined boundary conditions with respect to sustainability parameters. For example, fertilizer inputs required to obtain a predefined production target may be calculated under the restriction that the soil nutrient stock is not being mined. Also, the technology used to realize the production targets is fully specified, and may incorporate novel designs derived from experimentation or prototyping. Thus, the notion of “technological progress” is incorporated, which allows breaking with the past/present (as quantified by actual land-use systems) and which makes exploratory and predictive land-use analysis possible. Also, the incorporation of new technological options allows ex ante analysis of these options in an exploratory manner (Bouman et al 1999b).

The role of regional characterization Regional characterization is a prerequisite first step in the implementation of SOLUS/ LUPAS for any concrete land-use study. It is required for (1) scenario building and (2) to supply the required input data. A scenario should reflect relevant development problems or issues of the area under consideration and be based on attainable land- use systems as determined by biophysical and socioeconomic boundary conditions. The derivation of this information is typically the domain of regional characterization (Borkakati et al, Saleh et al, Joshi and Pal, this volume). Input data can be divided into nongeographically referenced and geographically referenced data. Most nongeographic data are used in the calculation of technical coefficients and include such items as crop characteristics, prices, and labor requirements for specific operations. Sometimes, some of these data may also be site- specific (e.g., product prices in relation to distance to markets), and then they should be geo-referenced. Geographically referenced data characterize the region under study

Regional land-use analysis to support agricultural and environmental . . . 475 and quantify the spatial distribution of input data. Examples are resource endowments such as land, water, and labor that figure as absolute constraints in the optimization model. GIS is an important tool in archiving and analyzing geo-referenced input data and in presenting spatial output results of the optimization model, such as land-use allocations.

Exploratory case study: Ilocos Norte Province, Philippines Site description and problem definition Ilocos Norte Province in northwestern Luzon, Philippines, with a population of nearly 0.5 million people and a total land resource of nearly 0.34 million ha, is a region with large forest resources (46% of total area). About 38% of the total area is classified as agricultural land. Agriculture in the province basically consists of rice-based production systems. Rice is cultivated in the wet season between June and October, whereas, during the dry season, diversified cropping is practiced: tobacco, garlic, onions, maize, sweet peppers, and tomatoes, all supported by groundwater irrigation. The province can be divided into four subregions, northern coastal, central lowlands, southern coastal, and eastern interior. Agricultural activities are most intensive in the central lowlands. Major environmental problems are soil erosion on sloping land in the eastern interior and groundwater pollution in the central lowlands (Lansigan et al 1998, Tripathi et al 1997). Mean annual rainfall in the province ranges between 1,700 mm in the southwest to above 2,400 mm in the eastern mountain ranges. On average, 6–7 typhoons per year cross the province, mostly between August and November, with considerable adverse effects on agricultural production. Soils are developed from very diverse parent materials. In the lowlands, sandy loams developed from alluvial deposits are predominant. Rice is the most common crop. In 1993, the province had a surplus of 100,000 t above demand (113,000 t). A well-developed marketing system has facilitated the establishment of intensive rice–cash crop production systems (Lucas et al 1999). In response to the recent economic downturn in the Philippines and the Asian region, the provincial government of Ilocos Norte formulated the “Sustainable Food Security Action Plan (SFSAP) and Agro-Fishery-Industrial Modernization Frame- work (AGRIMODE).” This framework, released in February 1999, outlines the market, technological, and policy direction and action plans to be taken in the next 3–6 years. The current major constraints to agricultural development as identified in consultative meetings are the low levels of agro-fishery productivity and income. Some of their major causes include relatively low cropping intensity, underdeveloped irrigation systems, and low average farm size. The specific objectives of the policy document are to design a food security action plan that would increase effective area for crop production by optimizing cropping intensity and increasing the level of irrigation development. The province is envisioned to become food-secure and an agro- industrial center (Provincial Government of Ilocos Norte, 1999, Final Report, Vol. 1—Main Document). The report is anchored on Philippine President Joseph Estrada’s

476 Bouman et al Framework for Agricultural Development, Food Security, and Poverty Alleviation, Republic Act 8435.

Scenario analyses using LUPAS The Systems Research Network for Eco-Regional Land Use Planning in Tropical Asia (SysNet) was established in late 1996 to develop methodologies for exploring alternative options for agricultural land use and rural development. The methods and tools that are needed to analyze various scenarios of future land use in order to guide policy changes are operationalized in a land-use planning and analysis system (LUPAS) that has three main methodology components: (1) land evaluation, (2) scenario construction, and (3) multiple-goal linear programming (MGLP) (Hoanh et al 1998, Laborte et al 1999). Land-use planning consists of various steps. The SysNet methodology aims at exploring alternatives for agricultural land use and development to assist in strategic planning. In an interactive process with stakeholders, SysNet methods and tools are then tailored to local conditions. The four study regions of SysNet are Haryana State, India; Kedah-Perlis Region, Malaysia; Ilocos Norte Province, Phil- ippines; and Can Tho Province, Vietnam (Roetter et al 1998). Results of scenario analyses consist of options for optimum land use under a given set of goals and constraints and the associated goal achievements. Results further indicate required policy changes and the scope for new agricultural production technologies that can satisfy the multiple goals for a given region. In the next two sections, we present results for the Ilocos Norte case study and show how LUPAS was applied to examine the effects of resource sharing and expansion of irrigated areas on rice production and farmers’ income. For Ilocos Norte Province, LUPAS was implemented by the Philippine and IRRI SysNet teams (Francisco et al 1998). Based on constraints analysis and policy views given above, examples from an exploratory study are presented with emphasis on two different development goals: (1) increased food security and (2) increased income from agricultural activities. We examined the future possibilities for increasing rice production and regional income and the trade-offs between these goals. For this, we analyzed various possible improvements in water availability and their effects on the major objectives. We considered five scenarios: (1) without water-sharing, i.e., the available water is restricted to each land unit (base scenario); expansion of irrigated areas (2) by 110% and (3) by 140%; (4) with water-sharing, i.e., land units connected by the current irrigation network can share water; and (5) no constraint on water, i.e., there is sufficient water in all of the land units in the province to support any land-use type. The current implementation is based on biophysical and socioeconomic databases updated in November 1999. Resource limits estimated for 2010 such as land devoted to agricultural production (119,850 ha), water resources, and available labor for agricultural activities for the entire province and for each administrative unit were determined. Provincial demands, targets, or market ceilings for agricultural products were not considered in this study but are dealt with elsewhere (Roetter et al 1999).

Regional land-use analysis to support agricultural and environmental . . . 477 Seventeen different agricultural products were considered and 23 land-use types that are currently practiced in the province were included in this study, of which 20 are rice-based (e.g., rice-maize, rice-tobacco, rice-rice). A total of 200 land units were defined by overlaying biophysical characteristics: irrigated areas, annual rainfall and distribution, slope and soil texture; and administrative units comprising 23 municipalities (Lansigan et al 1998). Two production technology levels were considered to choose from: average farmers’ practice and “best farmers’ practice,” both being actual land-use systems. The data for the input-output tables were derived from 1998- 99 farm surveys in the province consisting of 1,844 farms in the wet season and 2,164 farms in the dry season. The values for the input-output relations for the current technology were derived from the average values for these farms. For the best farmers’ practice, data were derived by taking the mean of the values with a yield level between the 90th and 95th percentile. Based on the major development goals, 11 objectives were identified in consultation with stakeholders in the province (Francisco et al 1998). Here, we will discuss results for two major objectives.

Food security Under the base scenario (i.e., no water-sharing with constraints on land, labor, and water imposed), the maximum rice production that can be achieved by the province is 0.53 million t, but with hardly any nonrice production. This can be realized if 95% of the farmers adopt the technology corresponding to “best available practice.” The resulting land-use allocation will require 8.9 million labor days and result in a total farmer income of US$91 million. When the irrigated area is doubled, rice production will increase by 23% as more land will be allocated to intensive triple and double rice systems. Employment and income will increase by about the same percentage (Fig. 2). When the irrigated area is further expanded (to cover approximately 90% of available agricultural land),

Index

Max. rice production 1.50 Max. agricultural employment Max. farmers’ 1.25 income

1.00

0.75 1.0 1.5 2.0 2.5 Irrigated areas (index)

Fig. 2. Effect of expansion of irrigated areas on goal achievements.

478 Bouman et al rice production increases by 42%, total farmers’ income by 45%, and employment by 34% (Fig. 2). In the next scenario, we assumed that water can be shared among land units belonging to the same irrigation system without expanding the irrigated area. Maxi- mum rice production amounts to 0.73 million t. This can be achieved if all the farmers adopt the technology “best available practice.” It requires 12.0 million labor days and total farmers’ income will amount to US$123 million. With available land as the only constraint, rice production, employment, and income will still increase considerably.

Income Under the base scenario, maximizing income leads to a total farmer income of US$953 million, which is more than 10 times that of the corresponding optimization for rice production. This is associated with 0.24 million t of rice and 2.6 million t of nonrice production (tomatoes, root crops, mungbeans). This requires 11.5 million labor days. When the irrigated area is doubled, maximum income will increase by 26%, and, when it is further extended, by 45% (Fig. 2). Associated rice production, nonrice production, and employment will also increase, however, to a slightly lesser extent. More land will be allocated to profitable, labor-intensive rice-tomato systems. This can be achieved if 90% of the farmers adopt the technology “best available practice.” Table 1 shows the results of the optimizations for the two major goals under the water-sharing scenario. Goal achievements could be further increased if land were the only constraint. Results show that a 140% increase in irrigated area could raise rice production by 42% and total farmers’ income by 45%. Water-sharing would lead to a 40% increase in rice production, but only to an 18% increase in income, as relatively less land would be allocated to the most profitable cropping systems (such as rice-tomato) with high water demands than to the less water-demanding root crops. Water will remain a production constraint in the province for some time to come, however, since the expected increase in irrigated area is just 2–3% per annum (Provincial Govern- ment of Ilocos Norte, 1999). Water-sharing is an alternative option, but it will require detailed surveys to determine to what extent water can be shared among land units belonging to the same irrigation system. Even for the base scenario, maximizing income with the consequent high share of nonrice production (Table 1) would not threaten

Table 1. Values of goals with water-sharing.

Goal

Activity Maximize rice Maximize production income

Rice production (t) 726,767 279,897 Nonrice production (t) 1,988 2,953,522 Employment (1,000 labor-d) 12,006 12,357 Total farmers’ income (106 US$) 123 1,125

Regional land-use analysis to support agricultural and environmental . . . 479 self-sufficiency in rice production for the province. The results imply that water is an important constraint to agricultural development and that expansion of irrigated area and water-sharing would result in higher productivity and more income for the province. The current study is part of an ongoing study for Ilocos Norte. The analysis presented here indicates technical opportunities but does not consider socioeconomic limitations such as constraints to adopting “best farmers’ practices” or product or labor market constraints. The next case study, for the northern Atlantic Zone, Costa Rica, presents examples where relatively more socioeconomic considerations are taken into account.

Predictive case study: the northern Atlantic Zone, Costa Rica Site description and problem definition The northern Atlantic Zone (AZ) is in the Caribbean lowland of Costa Rica. The climate is humid tropical with a mean daily temperature of 26 °C, 3,500–5,500 mm annual rainfall, and 85–90% average relative humidity. The total surface is about 447,000 ha, 334,000 ha of which are suitable for agriculture. From these 334,000 ha, some 55,000 ha are protected area for nature conservation. Current land use is natural forest (49%), cattle ranching (38%), banana plantations (10%), and crops (3%). Sub- stantial deforestation has taken place since the late 19th century, with negative environmental effects such as loss of biodiversity, loss of tourist attractions, and increased greenhouse gas emissions. The protected status of some national parks that conserve unique tropical forest habitats is sometimes disputed or violated by farmers and log- gers. In agriculture, concern is raised about the negative effects on human health and the environment of large amounts of biocides used, especially in the cultivation of bananas. Recently, the Costa Rican government has called for the execution of research that explicitly analyzes the trade-offs between economic and environmental goals (SEPSA 1997). Economic goals relevant to the AZ are, among others, the generation of foreign currency and farmers’ income, which can be captured with the term “regional economic surplus.” SOLUS was implemented for exploratory and predictive land-use studies in the northern AZ by the Research Program on Sustainability in Agriculture (REPOSA). First, in exploratory scenarios, outer boundaries to development and trade-offs between various economic and sustainability parameters were calculated (Bouman et al 1999a). Next, predictive scenarios were executed to study the effects of agricultural policies on changes in land use and selection of production technologies (Schipper et al 2000). Based on the problem statement given above, examples of predictive scenarios are presented here for agricultural policies that aim to (1) conserve forested area and (2) reduce biocide use. The effects of the policies are studied by comparing scenario results with model results obtained without such a policy, called the base run.

480 Bouman et al Implementation of SOLUS from regional characterization Based on regional characterization, the study area was divided into subregions to account for spatial variation in (farm-gate) product prices caused by spatial variation in transportation costs. Climatic characteristics and major topographical features are rather homogeneous throughout the area, doing away with the need to include these biophysical characteristics in the zonation. For each subregion, land and labor resources were calculated by map overlaying in GIS. Three main land units were distinguished: fertile well-drained (FW), infertile well-drained (IW), and fertile poorly drained (FP). Agricultural labor availability was derived from combining administrative boundaries with population and labor market data. The optimization model, called REALM (regional economic and agricultural land-use model), selects, per subregion and per land unit, land-use systems by maximizing regional economic surplus. Because of the relatively large size of the northern AZ and the presence of some major export commodities (e.g., bananas), (inter)national price formation for most products is expected to be affected by production in the area. Therefore, product markets were modeled using calculated price elasticities of demand, the share of the supply from the region in total supply (domestic or world market), and price elasticities of the supply of other suppliers. Also, the presence of a labor market was modeled, including competition for labor between subregions, along with possibilities to attract labor from the nonagricultural sector and from outside the area based on a calculated elasticity of national labor supply. Since the elasticities used in the modeling of the product and labor markets reflect aggregate producer (and consumer) behavior, REALM can be used for predictive as well as exploratory land- use studies. Land-use systems were generated for pasture, bananas, black beans, cassava, maize, palm heart, pineapples, plantains, and natural forests for sustainable timber extraction; for two herd systems that could be used to graze the pastures; and for five feed supplements. By varying production targets and production technologies, and by combining these with the three land units, a total of 3,108 different actual and technologically potential land-use systems were generated. Two sustainability indicators were calculated to express the use of biocides: the total quantity of active ingredients in biocides applied (BIOA), and a so-called biocide index (BIOI) that quantifies the environmental hazard of biocides by taking into account the amount of toxic ingredients used, their toxicity level, and their half-life time. Within the optimization model, the values of these sustainability parameters are summed over all selected land-use systems so that aggregate values are obtained per land unit, per subregion, and for the northern AZ as a whole.

Forest conservation Recently, the Costa Rican government has introduced a policy to stimulate landowners to keep part of their property under natural forest. In return for not cutting down trees, a landowner can obtain a subsidy of $43 ha–1 y–1, initially for a period of 5 years. The subsidy was created as a result of the international discussion around glo-

Regional land-use analysis to support agricultural and environmental . . . 481 bal warming since maintaining or creating forest is seen as a means to sequester carbon dioxide. To analyze the effect of a forest subsidization policy, premiums were allocated to the land-use system “natural forest” in the optimization model. Besides government subsidies, such premiums may also express the value of nontimber products and other “social” services (such as tourist attractions, conservation of biodiversity). The economic returns of natural forest by sustainable exploitation of wood are about $16 ha–1 y–1. In the base run, no extra premium was given to natural forest. Next, premiums of $111, $122, and $133 ha–1 y–1 were given. Premiums up to $111 ha–1 y–1 were not sufficient to induce landowners to maintain their natural forests (Table 2). On the other hand, premiums of $122 and $133 ha–1 would lead to forest areas of about 120,000 ha and 200,000 ha (at the expense of decreased pasture area), respectively, compared with the current amount of 84,000 ha. Based on these results, it can be expected that the current level of subsidy offered by the Costa Rican government will not be enough to increase or even maintain the current amount of forested area in the northern AZ. Even though an annual premium of $111 ha–1 raised the annual return of natural forest to $117 ha–1, this was still lower than the shadow price of land in all subregions and for each land unit. In the case of a premium of $122 ha–1, however, returns of natural forest exceeded the shadow prices of the land belonging to the fertile poorly drained and infertile well-drained land units in most subregions. On the other hand, the fertile well-drained land unit had shadow prices between $188 and $204 ha–1 (depending on the subregion) and a premium would have to exceed $172 ha–1 for natural forest to become an economically attractive option.

Biocide taxing Regulation and control of agricultural input use have been identified as an important policy option to reduce certain negative externalities of agricultural production (SEPSA 1997). We expect that taxing an input that is currently not taxed, as is now the case with biocides in Costa Rica, will lead to less use of this input. Two ways of imple-

Table 2. Effect of a premium on the area of natural forest: regional economic surplus and land area under natural forest (per land unit).

Land under forest (000 ha) Economic surplus ($106)FWa FPa IWa Totala (118) (136) (86) (340)

Premium ($ ha–1) 0 276 0 0 0 0 111 276 0 0 0 0 122 276 0 63 56 119 133 278 0 122 77 200

aFW = fertile well-drained, FP = fertile poorly drained, IW = infertile well-drained; total available area given in parentheses.

482 Bouman et al Table 3. Effects of alternative ways of taxing biocide use on regional economic surplus and on regional amount of active ingredients (BIOA) and biocide index (BIOI).

Tax rate

Base Flat tax Tax A Tax B (%)

Type of biocide Slightly toxic 0 100 20 10 Medium toxic 0 100 50 30 Very toxic 0 100 200 150

Results (value) (% change) Economic surplus ($ × 106) 267.6 –18.7 –4.3 –2.2 BIOA (kg × 106) 1.9 –13.1 –3.9 –3.8 BIOI (106) 84.1 –4.0 –81.9 –81.9 menting a tax on biocides were studied: a flat tax and a progressive tax. With the flat tax, all biocides were taxed equally, whereas the level of progressive tax was related to the environmental damage caused by a specific biocide as expressed by its biocide index (BIOI). Taxing all biocides at a uniform rate of 100% led to a reduced use of biocides in terms of BIOA of 13% relative to the base scenario, while the BIOI decreased by only 4% (Table 3). However, the total economic surplus decreased by nearly 19%. Thus, a relatively modest environmental gain was obtained at high economic costs. In contrast, a progressive tax regime where different tax rates were applied to three categories of biocides depending on their degree of toxicity (i.e., slightly, medium, and very toxic) resulted in a much larger reduction in the BIOI, while at the same time preserv- ing more of the economic surplus. For example, applying taxes of 20%, 50%, and 200% (Tax A) to the categories of slightly, medium, and very toxic biocides, respectively, led to a reduction in the economic surplus of 4%, while reducing the BIOI by more than 80%. When tax rates were reduced to 10%, 30%, and 150% (Tax B), respectively, for the three categories of biocides, economic surplus decreased by just 2% with the same environmental improvement. Thus, there appears to be considerable scope for tax policies to induce the adoption of less biocide-intensive land-use systems while maintaining aggregate income.

Land-use analysis in rainfed rice ecosystems Rainfed rice areas are mostly harsh environments, characterized by spatial variability in environmental conditions (soil, topography, weather) and temporal variability in weather, especially rainfall amount and distribution. Yields are therefore relatively low and unstable. The dominant constraints to production are abiotic stresses—among which the lack of water is commonly considered as the most severe—and uncertain

Regional land-use analysis to support agricultural and environmental . . . 483 returns to purchased inputs due to unstable yields (Wade 1998). In such environments, farmers’ behavior and use of land are guided very much by their perception of risk (Roetter and Van Keulen 1997, Singh HN et al, this volume). This is especially important in subsistence agriculture, where rural livelihood often depends on the season’s harvest. Any land-use analysis should therefore explicitly recognize and address these critical biophysical and human characteristics of rainfed lowlands. So far, neither the temporal variability of weather and economic parameters nor the decision behavior of farmers under risk has been taken into account in SOLUS or LUPAS.

Spatial and temporal variability In SOLUS and LUPAS, physical production of a land-use system is a key technical coefficient in the optimization model. Production is calculated or predetermined in the target-oriented approach from, among others, biophysical properties of the various land units identified. In rainfed environments, where water availability largely determines production, the amount and distribution of rainfall, the terrain (slope, position within the landscape), and hydrological soil properties are key properties by which to distinguish land units. Novel ways for mapping and delineating such properties and their spatial variability are presented by Oberthür et al (this volume, 1999). Other approaches combine biophysical land characteristics with hydrological modeling to generate yield surfaces. In SOLUS and LUPAS, land units are considered homogeneous in biophysical and socioeconomic conditions and technical coefficients are determined for “average” conditions. Simulation models can help translate variability (or uncertainty) in biophysical parameters into variability in crop yield. For instance, Bouman (1994) used Monte Carlo techniques and an ecophysiological rice growth model to generate probability distributions of rice yield from variability in soil properties and management parameters. The main challenge is to translate the resulting maps into several manageable land units that can be handled in the optimization model, while retaining information on parameter variability. Simulation models are also suitable for calculating temporal variability in yield caused by variation in weather (Hammer and Muchow 1991). GIS linking soil and climatic data surfaces with yield probability distributions generated by crop simulation models in combination with agroeconomic data has been applied to calculate spatio-temporal variability of yield, production, and economic risk for well-defined production systems (Roetter and Dreiser 1994). Besides simulation modeling and GIS techniques, expert knowledge, field inquiries, and experimental data can all help quantify variability in yield (or other technical coefficients) of land-use systems. Once the variability is quantified, several methods can be used to address such variability in optimization models (Hazell and Norton 1986). A main research question, however, is how to handle the spatial representation of variable model input and output parameters.

Risk Variability in yield turns into risk when it affects farmers’ livelihoods and influences their decisions on land use. Besides the biophysical variability (or, derived from it, the economic variability in financial returns), farmers’ behavior toward risk is impor-

484 Bouman et al tant in risk analysis (Roetter and Van Keulen 1997). Several techniques handle risk in optimization modeling, most of them based on expressions of variability and farmers’ risk perceptions (Hazell and Norton 1986, Selvarajan et al 1997). In the simplest techniques, mean yields in an objective function are modified by an expression of variability (such as standard deviation) times a farmer’s risk aversion factor, which should be derived from interviews. Another approach is to quantify farmers’ utility functions—that include their risk perception—and to make these the objective function of the optimization model (Kruseman et al 1995). In regional land-use studies, farmers could be categorized according to their utility functions. Farm categories could then be optimized individually or together in an iterative manner taking into account feedback at higher levels of spatial aggregation (e.g., Roebeling et al 2000).

Conclusions and recommendations Characterization of rainfed environments is not a goal in itself, but depends on the type of information to be generated. It needs to be based on a sound understanding of the prevailing biophysical and socioeconomic processes, be it at the field, farm, or regional level. Both exploratory and predictive land-use studies have in common that they synthesize fragmented agricultural knowledge and integrate data on resources over time and space. In rainfed rice areas, high temporal and spatial variability of production resources complicates the analysis. Farmers’ diverse responses to climatic and economic risks must be taken into account, which eventually demands stronger links between on-farm research and operational research for meaningful policy formulation and implementation.

References Bouman BAM. 1994. A framework to deal with uncertainty in soil and management parameters in crop yield simulation: a case study for rice. Agric. Syst. 46:1-17. Bouman BAM, Jansen HGP, Schipper RA, Nieuwenhuyse A, Hengsdijk H, Bouma J. 1999a. A framework for integrated biophysical and economic land use analysis at different scales. Agric. Ecosyst. Environ. 75:55-73. Bouman BAM, Nieuwenhuyse A, Ibrahim M. 1999b. Pasture degradation and restoration by legumes in humid tropical Costa Rica. Trop. Grassl. 33:98-110. Bouman BAM, Jansen HGP, Schipper RA, Bouma J, Kuyvenhoven A, Van Ittersum MK. 2000. A toolbox for land use analysis. In: Bouman BAM, Jansen HGP, Schipper RA, Hengsdijk H, Nieuwenhuyse A, editors. Tools for land use analysis on different scales, with case studies for Costa Rica. Dordrecht (Netherlands): Kluwer Academic Publishers. p 213- 232. Crissman CC, Antle JM, Capaldo SM. 1997. Economic, environmental, and health trade-offs in agriculture: pesticides and the sustainability of Andean potato production. Dordrecht (Netherlands): Kluwer Academic Publishers.

Regional land-use analysis to support agricultural and environmental . . . 485 Francisco SR, Navarrete Jr. RL, Dimasuay LB, Lansigan FP, Pascual CM, Lucas MP, Orno JL, Roetter R, Hoanh CT, Laborte AG, Marcos TF, Acosta CG, Agustin EO, Pascua ME, Pascua Jr. SR, Utrera RT, Cruz RT, Obien SR. 1998. A systems approach to agricultural land use planning at the provincial and municipal levels of Ilocos Norte Province, Phil- ippines. In: Roetter R, Hoanh CT, Luat NV, Van Ittersum MK, van Laar HH, editors. Exchange of methodologies in land use planning. SysNet Research Paper Series 1. Los Baños (Philippines): International Rice Research Institute in collaboration with NARS of India, Malaysia, Philippines, Vietnam and Wageningen URC. p 103-109. Hammer GL, Muchow RC. 1991. Quantifying climatic risk to sorghum in Australia’s semiarid tropics and subtropics: model development and simulation. In: Muchow RC, Bellamy JA, editors. Climatic risk in agricultural production: models and management for the semiarid tropics and subtropics. Wallingford (UK): CAB International. p 205-232. Hazell PBR, Norton RD. 1986. Mathematical programming for economic analysis in agriculture. New York (USA): Macmillan Publishing Company. 400 p. Hengsdijk H, Bouman BAM, Nieuwenhuyse A, Jansen HGP. 1999. Integrating systems-analytical and expert knowledge to quantify land use systems: a case study for the Northern Atlantic Zone of Costa Rica. Agric. Syst. 61:109-121. Hoanh CT, Roetter R, Jansen DM, Aggarwal PK, Lansigan FP, Lai NX, Bakar IA, Tawang A. 1998. Generalizing SysNet methodologies for land use planning at the sub-national level. In: Roetter R, Hoanh CT, Luat NV, Van Ittersum MK, van Laar HH, editors. Exchange of methodologies in land use planning. SysNet Research Paper Series 1. Los Baños (Philippines): International Rice Research Institute in collaboration with NARS of In- dia, Malaysia, Philippines, Vietnam and Wageningen URC. p 41-56. Jansen DM, Schipper RA. 1995. A static descriptive approach to quantify land use systems. Neth. J. Agric. Sci. 43:31-46. Kruseman G, Ruben R, Hengsdijk H, van Ittersum MK. 1995. Farm household modeling for estimating the effectiveness of price instruments in land use policy. Neth. J. Agric. Sci. 43:111-123. Kuyvenhoven A, Ruben R, Kruseman G. 1995. Options for sustainable agricultural systems and policy instruments to reach them. In: Bouma J, Kuyvehoven A, Bouman BAM, Luyten JC, Zandstra HG, editors. Eco-regional approaches for sustainable land use and food production. Dordrecht (Netherlands): Kluwer Academic Publishers. p 187-212. Laborte AG, Roetter R, Hoanh CT. 1999. SysNet tools: the multiple goal linear programming (MGLP) model and MapLink. IRRI Technical Bulletin No. 6. Los Baños (Philippines): International Rice Research Institute. 31 p. Lansigan FP, Orno JL, Pascual CM, Navarrete Jr. NR, Francisco SR, Laborte AG, Hoanh CT, Roetter R. 1998. Systems methodology development for exploratory agricultural land use planning for Ilocos Norte, Philippines. In: Roetter R, Hoanh CT, Luat NV, Van Ittersum MK, van Laar HH, editors. Exchange of methodologies in land use planning. SysNet Research Paper Series 1. Los Baños (Philippines): International Rice Research Institute in collaboration with NARS of India, Malaysia, Philippines, Vietnam and Wageningen URC. p 73-80. Lucas MP, Pandey S, Villano RA, Culanay DR, Obien SR. 1999. Characterization and economic analysis of intensive cropping systems in rainfed lowlands of Ilocos Norte, Phil- ippines. Exp. Agric. 35:1-14. Oberthür T, Goovaerts P, Dobermann A. 1999. Mapping soil texture classes using field texturing, particle size distribution and local knowledge by both conventional and geostatistical methods. Eur. J. Soil Sci. 50:457-479.

486 Bouman et al Pinstrup-Andersen P, Pandya-Lorch R. 1995. Prospects for world food security and distribution. In: Eco-regional approaches for sustainable land use and food production. Bouma JA, Kuyvenhoven A, Bouman BAM, Luyten JC, Zandstra HG, editors. Dordrecht (Neth- erlands): Kluwer Academic Publishers. p 89-112. Rabbinge R, Van Latesteijn H. 1992. Long term options for land use in the European Commu- nity. Agric. Syst. 40:195-210. Roebeling PC, Jansen HGP, Schipper RA, Saenz F, Castro E, Ruben R, Hengsdijk H, Bouman BAM. 2000. Farm modeling for policy analysis at the farm and regional level. In: Bouman BAM, Jansen HGP, Schipper RA, Hengsdijk H, Nieuwenhuyse A, editors. Tools for land use analysis at different scales, with case studies for Costa Rica. Dordrecht (Neth- erlands): Kluwer Academic Publishers. p 171-198. Roetter R, Dreiser C. 1994. Extrapolation of maize fertiliser trial results by using crop-growth simulation: results for Murang’a District, Kenya. In: Fresco LO, Stroosnijder L, Bouma J, van Keulen H, editors. The future of the land: mobilising and integrating knowledge for land use options. UK: John Wiley and Sons. p 249-260. Roetter R, van Keulen H. 1997. Variations in yield response to fertilizer application in the tropics. II. Risks and opportunities for small-holders cultivating maize on Kenya’s ar- able land. Agric. Syst. 53:69-95. Roetter R, Hoanh CT, Luat NV, Van Ittersum MK, van Laar HH, editors. 1998. Exchange of methodologies in land use planning. SysNet Research Paper Series 1. Los Baños (Phil- ippines): International Rice Research Institute in collaboration with NARS of India, Malaysia, Philippines, Vietnam and Wageningen URC. 154 p. Roetter R, Laborte AG, Hoanh CT, van Oort P, Cabrera JMCA, Lucas M, Francisco S, van Keulen H. 1999. Resource use analysis at regional scale: explorations for rice systems in Southeast Asia. In: CD-Rom Proceedings SAAD3, 8-10 Nov. 1999, Lima, Peru. Dordrecht (Netherlands): Kluwer Academic Publishers. Schipper RA, Bouman BAM, Jansen HGP, Hengsdijk H, Nieuwenhuyse A. 2000. Integrated biophysical and socio-economic analysis of regional land use. In: Bouman BAM, Jansen HGP, Schipper RA, Hengsdijk H, Nieuwenhuyse A, editors. Tools for land use analysis on different scales, with case studies for Costa Rica. Dordrecht (Netherlands): Kluwer Academic Publishers. p 115-144. Selvarajan S, Aggarwal PK, Pandey S, Lansigan FP, Bandyopadyay SK. 1997. Systems approach for analyzing trade-offs between income, risk and water use in rice-wheat production in northern India. Field Crops Res. 51:147-161. SEPSA (Secretaría Ejecutiva de Planificación Sectoral Agropecuaria). 1997. Políticas del sector agropecuario (revisión y ajuste). Ministry of Agriculture and Livestock (MAG), San José, Costa Rica. Timmer P, Falcon W, Pearson S. 1983. Food policy analysis. Baltimore, Md. (USA): Johns Hopkins University Press. Tripathi BP, Ladha JK, Timsina J, Pascua SR. 1997. Nitrogen dynamics and balance in intensified rainfed lowland rice-based cropping systems. Soil Sci. Soc. Am. J. 61:812-821. Van Ittersum MK, Rabbinge R. 1997. Concepts in production ecology for analysis and quantification of agricultural input-output combinations. Field Crops Res. 52:197-208. Veldkamp A, Fresco LO. 1997. Exploring land use scenarios, an alternative approach based on actual land use. Agric. Syst. 55:1-17.

Regional land-use analysis to support agricultural and environmental . . . 487 Wade L. 1998. Nutrient research on rainfed lowland rice in relation to 1995 review. In: Ladha JK, Wade L, Dobermann A, Reichardt W, Kirk GJD, Piggin C, editors. Rainfed lowland rice: advances in nutrient management research. Manila (Philippines): International Rice Research Institute. p 29-40.

Notes Authors’ addresses: B.A.M. Bouman, R. Roetter, A.G. Laborte, International Rice Research Institute, DAPO Box 7777, Metro Manila, Philippines; R.A. Schipper, Department of Development Economics, Wageningen University, Netherlands. Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char- acterizing and understanding rainfed environments. Proceedings of the International Workshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute. 488 p.

488 Bouman et al