The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of the Food and Agriculture Organization of the United Nations concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. The word “countries” appearing in the text refers to countries, territories and areas without distinction. The designations “developed” and “developing” countries are intended for statistical convenience and do not necessarily express a judgement about the stage reached by a particular country or area in the development process. The opinions expressed in the articles by contributing authors are not necessarily those of FAO.

The EC-FAO Partnership Programme on Information and Analysis for Sustainable Forest Management: Linking National and International Efforts in South Asia and Southeast Asia is designed to enhance country capacities to collect and analyze relevant data, to disseminate up-to- date information on and to make this information more readily available for strategic decision-making. Thirteen countries in South and Southeast Asia (Bangladesh, Bhutan, Cambodia, India, Indonesia, Lao P.D.R., Malaysia, Nepal, Pakistan, the Philippines, Sri Lanka, Thailand and Viet Nam) participate in the Programme. Operating under the guidance of the Asia-Pacific Forestry Commission (APFC) Working Group on Statistics and Information, the initiative is implemented by the Food and Agriculture Organization of the United Nations (FAO) in close partnership with experts from participating countries. It draws on experience gained from similar EC-FAO efforts in Africa, and the Caribbean and Latin America and is funded by the European Commission.

Cover design: Tan Lay Cheng Photo credits: Samsudin Musa

For copies write to: Patrick B. Durst Senior Forestry Officer FAO Regional Office for Asia and the Pacific 39 Phra Atit Road Bangkok 10200 Thailand

Printed and published in Bangkok, Thailand

© FAO 2003 ISBN 974-7946-33-5 2 3 EUROPEAN COMMISSION DIRECTORATE-GENERAL DEVELOPMENT

Information and Analysis for Sustainable Forest Management: Linking National and International Efforts in South and Southeast Asia

EC-FAO PARTNERSHIP PROGRAMME (2000–2002) Tropical Forestry Budget Line B7-6201/1B/98/0531 PROJECT GCP/RAS/173/EC

ASSESSING THE STATUS OF LOGGED-OVER PRODUCTION FORESTS THE DEVELOPMENT OF A RAPID APPRAISAL TECHNIQUE

by Samsudin Musa, Abd. Rahman Kassim, Safiah Muhammad Yusoff & Shamsudin Ibrahim

4 Information and Analysis for Sustainable Forest Management: Linking National and International Efforts in South and Southeast Asia

EC-FAO PARTNERSHIP PROGRAMME (2000–2002) Tropical Forestry Budget Line B7-6201/1B/98/0531 PROJECT GCP/RAS/173/EC

Background Over the last several decades, deforestation and degradation of tropical forests as well as their wasteful utilization have received increased attention. Forest resources contribute significantly towards foreign exchange earnings in many countries and over-exploitation of resources cannot be in their long-term interest. In the early 1980s, there were many predictions that tropical forests – including those of Malaysia – would be completely depleted by 2000 (Samsudin and Kasinathan 1989). Although such predictions were overly pessimistic and tropical forests remain an important land use in the Asia-Pacific region, some countries (e.g. the Philippines and Thailand) have turned from being exporters to net importers of timber (Appanah 2000). The widespread concern regarding forest abuse has triggered numerous forest assessments and inventories. Most countries in the Asia-Pacific region assess the extent of their forest areas on a regular basis. The area of natural forests is usually known, although area assessments may not necessarily be completely accurate. In some countries, data on are also available and post-harvesting inventories determine the needs for silvicultural treatments. However, once forest operators leave forest areas and road conditions deteriorate, regular inventories cease. As a result, knowledge of the status of logged-over forest areas is scanty, and the assumption that previously logged forests

5 will be ready for re-entry is frequently based on inadequate knowledge of forest stand volumes and composition. In fact, there is a widespread concern that many production forests are now degraded and will yield, during the second harvest, substantially lower commercial volumes than during the first harvest. The knowledge gap is a major concern as wood-based industries rely on a continuous flow of raw material. In many countries, all the old-growth (production) forests have been or will soon be exploited and wood supply will have to rely on logged-over or second- growth (or even third cycle) or residual forests1. Assessing the status of logged-over forests in terms of expected volumes, species composition and timber quality thus has high priority. In Peninsular Malaysia, information based on preliminary investigations (Yong 1998) and observations indicate that most residual forest stands have not regenerated according to assumptions and will not be ready for commercial harvesting on a sustainable basis at the end of the cutting cycle as expected. The effects of (poor) harvesting practices and illegal logging on forest conditions remain unclear. Due to the large extent, high variability and inaccessibility of many natural forests in the region, conventional forest inventories are extremely costly. Alternative assessment are needed, which allow for the rapid appraisal of stand conditions. This report provides an account of a methodology that can be used for assessing logged-over forests rapidly at the broad management level. For operational level inventories, the design and intensity of sampling will be different.

Introduction In Malaysia, the forestry sector continues to play an important role in the socio-economic development of the nation. The total export of timber and timber products, including wooden and rattan furniture in 2001 was impressive and was valued at about US$3.75 billion. The country is fortunate in that forest products are still available from old-growth production forests to meet its timber requirements, although in some states (e.g. Johor, Selangor and Negeri Sembilan), timber harvesting has started in the second-growth (logged-over) forests. However, the old-growth production forests are expected to be exploited completely over the next few years and forest production will have to shift to second-growth forests. This will be a challenge for the forestry sector as the structure and composition of second-growth forests are quite different compared to old-growth forests. Research indicates that past logging practices caused serious damage to residual stands and regeneration (e.g. Canonizado 1978; Pinard 1995; Taumas 1999), often exceeding 50 percent2. There is also excessive soil damage and compaction due to the high density of logging roads and skid trails, and heavy blading by bulldozers. As a result, most second-growth forests are anticipated to contain less commercial species, and the size distribution patterns of the trees are expected to be highly variable. Due to high mortality and poor growth rates, productivity under the selective management system, currently being applied in production forests, is far less than the assumed 2.5 m3/ha (Ismail Harun and Appanah 2000). This means that the existing logged-over forest, which comprises more than 80 percent of the production forest in Peninsular Malaysia, needs to be assessed and new management approaches will have to be formulated to enhance the productivity of the stands. The total forest resources in Malaysia cover 18.9 million ha or 59 percent of the total land area. In Peninsular Malaysia, out of a total forested area of 5.55 million ha, Permanent Reserved Forests constitute 4.58 million ha, of which 2.91 million ha are classified as Production Forests while the remaining 1.67 million ha are Protection Forests (Table 1). The area of logged-over forests is about 3.11 million ha or 65 percent of the total inland forest. These forests will be re-logged in the near future to meet the timber demands of the country.

Table 1. Permanent Reserved Forests (PRF) in Peninsular Malaysia (ha) Status PRF Stateland Total

1 The role of plantations in supplying wood is also expected to increase considerably. 2 Tay et al. 2002 report as much as 60 percent damage to the residual stand in Sabah, Malaysia.

6 Production Protection Undisturbed Inland Forest 775 378 735 221 138 481 1 649 080 Undisturbed Peat Swamp Forest 111 494 4 576 61 257 177 327 Sub-total 886 872 739 797 199 738 1 826 407 Logged Over Inland Forest < 1960 154 046 153 283 54 449 361 778 1961-1970 206 633 122 823 45 011 374 467 1971-1980 533 724 190 271 134 796 858 791 1981-1990 734 175 300 624 273 483 1 308 282 1991-1992 75 496 41 814 90 716 208 026 Sub-total 1 704 074 808 815 598 455 3 111 344 Logged Over Peat Swamp Forest < 1980 35 897 2 621 63 544 102 062 1981-1990 1 761 70 60 510 62 341 1991-1992 49 140 - 15 250 64 390 Sub-total 86 798 2 691 139 304 228 793 Mangrove Forest 88 827 80 13 716 102 623 Plantation Forest 89 960 746 - 90 706 Shifting Cultivation Forest 27 865 105 060 18 298 151 223 Degraded Forest 23 586 14 858 283 38 727 Total 2 907 982 1 672 047 969 794 5 549 823 Source: National III, Forestry Department Peninsular Malaysia

Objectives of the study

This study was designed to develop a rapid appraisal technique for assessing the status of logged- over forests at the management level in Peninsular Malaysia. It is based on existing methodologies and develops a pragmatic approach applicable for the prevailing conditions in Peninsular Malaysia.

Literature review

Forest inventory Forest inventories describe the quantity and quality of trees and other organisms of the forest and the characteristics of the land on which the forest grows. A forest assessment assigns values to the resource. The main objectives of a forest inventory are to obtain estimates of the timber volume to prepare an area for timber sale and to prepare operational plans for logging based on the quantity and location of the timber. Data recorded during inventories (e.g. timber volumes, growth rates, size distribution pattern, species composition, stand conditions and location of stands) are also important for sound forest management (Taylor et al. 1989a). Sampling methods

Cost and reliability are important considerations when designing an inventory. Forest inventories can involve large areas and a 100-percent inventory is often not possible and for most purposes unnecessary (it might be a specific research requirement). Complete inventories are expensive, tedious and as a result non-sampling errors (incorrect recording of tree diameters, heights and quality) tend to increase (Bell 1997). For this reason, sampling has been introduced, which if properly done, provides reasonable estimates of the true population. It is crucial to select a sampling method or combination of sampling methods that allows for the most efficient collection of data. There are numerous different sampling designs and no particular design meets the needs of all inventories. Each forest area varies and in meeting the objectives of an inventory a design must be selected that provides best estimates at a reasonable cost. Seven basic factors influence the choice of the design (Taylor et al. 1989b):  Information required and its desired precision;

7  Composition of the forest and its variability;  Topography and accessibility;  Human-resource availability and level of skills;  Availability of time and funds;  Availability of maps, aerial photographs and other relevant information; and  Designers’ knowledge of statistics and sampling theory.

The basic inventory designs include the following categories: A. Probability sampling  Simple random sampling;  Stratified random sampling;  Multi-stage sampling;  Multi-phase sampling; and  Sampling with varying probabilities. B. Non-random sampling  Selective sampling  Systematic sampling

In probability sampling, the probability of selecting any sampling unit is known prior to actual chance of selection. In non-random sampling, the units that constitute the sample are not chosen by probability laws, but by personal judgment or systematically. Random sampling requires an equal chance of selecting all possible combinations of the n sampling units from the total population. In systematic sampling, the sampling units are spaced at fixed intervals throughout the population. This approach has many advantages, which explains its regular use. It is able to provide good estimates of the population means by spreading the sample over the entire population. Inventories based on systematic sampling are usually faster and less expensive to execute since the choice of sample locations is mechanical and uniform. Travel between sample units is easier and shorter as fixed directional bearings are followed.

Stratification We often have some knowledge of a population, which can be used to increase the precision or usefulness of our sample. Stratified random sampling is a method that takes advantage of available information about the population. In this method, the units of the population are grouped together on the basis of similarity of some characteristics. Each group or stratum is then sampled and the group estimates are combined to provide population estimates. Stratification is achieved by sub-dividing the forest area into strata on the basis of some features such as topography, forest types, density classes, volume, age and site. Stratified random sampling has two main advantages over unrestricted random sampling:  Separate estimates of the means and standard errors are made for each stratum.  For a given sampling intensity, stratification often gives a more precise estimate. However, this assumes that the homogeneity of the stratum is greater than the homogeneity of the whole population.

As each stratum is more uniform it should give less variation within forest types than between types. This in turn will give more precise estimates of parameters (e.g. basal area and number of stems). Forest stratification should be considered to increase the cost-effectiveness of an inventory. In addition, in some cases it can be used to reduce the total number of plots. Normally, forest stratification is based on remotely sensed data such as aerial photographs and satellite images. However, the ability to stratify tropical forests with reliable precision has been

8 questioned. Hence, forests are classified usually according to available historical information. Logged areas may be classified according to years since logging (e.g. 1-10 years, 11-20 years, 20- 30 years after logging). Such an approach has been adopted for national forest inventories in Malaysia. Recent technological advances in remote sensing (i.e. high-resolution images) have enhanced the capability of to stratify and classify forests more accurately. Nezry et al. (2000) reported obtaining fairly reliable estimates of standing tree volume in Sarawak, Malaysia, based on Synthetic Aperture Radar and optical synergy and forestry knowledge (forest structure models). New approaches in data analysis have also assisted greatly in the stratification of tropical forests. One such development is the Canopy Density Mapping and Monitoring Model (in short the FCD model), which was developed by Rikimaru with assistance from the International Tropical Timber Organization (ITTO). Rikimaru (2000) developed a modelling software that utilizes Landsat imagery and assesses the forest status based on canopy density through several pertinent indices such as advanced vegetation index, bare soil index, shadow index and thermal index (Rikimaru 2002). Studies in Indonesia (Urquizo 1998) and Malaysia (Alias et al. 2001) have also reported positive results in forest stratification.

Efficiency in sampling design The most efficient sampling design is one that, for a specific cost gives the smallest error for the parameter to be estimated, or for an accepted error is the least cost (Taylor et al. 1989b). Defining efficiency is straightforward, but determination of the most efficient design is complicated for several reasons:  It is impossible to consider all possible sampling designs when choosing a design. Some characteristics of the contemplated design are taken for granted. In field sampling, the unit area at each stage, the number of stages and stratification are fixed prior to optimization. This is based on prior experience and leads to only partial optimization.  Complete estimation of the error should take into account not only the sampling error, but also biases and measurement errors. The latter are difficult to determine. In most cases, efficiency studies deal only with sampling errors and are valid only if the measurement errors are reduced to the minimum.  A sampling design is optimal for a given parameter such as gross volume of large trees over 60 cm diameter at breast height (dbh). But it may not be optimal for considerably smaller trees. Consequently, it is crucial to select the most important inventory parameters for consideration in the design stage. Sometimes determining the most important parameter is not easy as consideration needs to be given to others as well.

9 Plot design The kind of sampling units, their size and shape, the number of units, their distribution, their measurement and their analysis are important considerations in the overall inventory methodology. It is common to used fixed area plots (e.g. strip plots, rectangular, square, L-shaped and circular plots of various sizes). Variable plots, such as point sampling, are also used. Some approaches rely on a combination of fixed and variable plots.

Fixed plots Fixed plots have a fixed size with a certain shape. Unbiased estimates of stand parameters can be obtained from any plot size or shape, although the precision and survey cost can vary significantly. It is usually more efficient to use small sampling units as they exhibit higher variability. However, in very variable forests, small sampling units may result in a high coefficient of variation (CV). Plots can assume any shape. Theoretically, a rectangular plot with long at right angles to the contours is most efficient as the plots tend to cross the range of stand conditions (Taylor et al. 1989b). Plot shapes do not affect the size of the sampling error, but the characteristics of the shape influence the ease of establishment and the length of the plot perimeter. Circular plots have the advantage that a single dimension, the plot radius, can be used to define the perimeter. Their biggest advantage is that they have the smallest area to perimeter ratio of any shape, which reduces the number of borderline trees. Fewer borderline trees reduce the number of measurement errors. In many inventory practices employing fixed plots, sub-plots of varying sizes are also established to enumerate trees within specified size classes on each sub-plot. The intent is to reduce the plot area for small trees and increase it for the larger trees to tally approximately an equal number of trees in several sized classes. When sub-plots are used, the sampling intensities of the sub-plots are always less than those of the main primary plot.

Variable plots Variable plot cruising is one of the latest methods to have been developed but it is still not practised widely in the Asia-Pacific region. The method, also called point sampling or plotless cruising, has been used extensively in temperate countries and has also been thoroughly tested in a wide variety of tropical situations. Many features of this method are similar to the fixed plots (e.g. determination of plot locations and numbers, measurements of tree diameter, defects and tree quality). However, its application has many differences, and thus comparisons should be limited to the relative costs of obtaining the same pre-determined sampling error, or relative sampling error at a fixed cost. (Samsudin and Kasinathan 1998) Variable radius, instead of fixed radius plots, are used in the variable plot method. It is actually a multi-plot with each tree having its own plot size depending on its diameter. It does not require measurement of plot diameters or tree diameters to compare the basal area per hectare. A tree that has a diameter large enough to be within the fixed critical angle of the angle gauge (e.g. relaskop or ) is tallied. Trees too small or too distant are ignored. In fixed plot sampling, the probability of tree selection is proportional to stem frequency; in point sampling it is proportional to the stem basal area.

Inventory levels Inventories undertaken at various levels involve different intensities and cover differing spatial scales. Stand inventories provide information on a stand-to-stand basis for detailed management planning. This level of inventory is called the management level inventory and is usually undertaken for planning purposes and the development of mid-term management plans. More intense sampling is the operational-level inventory and it is called the district inventory. Sampling intensities are higher and normally cover 10 percent of an area for such purposes as pre-felling

10 inventories. The information is used for calculating allowable cuts for sustained yield management. The land units include several stands and they may range from several hundred (such as in Malaysia) to several thousand hectares. Sampling errors tolerated at this level normally would be less than 20 percent. National inventories are also conducted to provide baseline information for the whole country for policy and broad planning purposes. In Malaysia, three national inventories have been undertaken at 10-year intervals. Currently, the Fourth National Inventory is being implemented. Systematic sampling and cluster sampling are the two most commonly used methods. In some countries, existing information from management inventories is compiled to obtain national estimates. The upscaling of management inventory data to national data is problematic, as they may yield results that are 20 to 30 percent above the results from adding up these lower level inventories (Pelz 1993).

Study area The study was carried out in Hutan Simpan Tekam or Tekam Forest Reserve, which is situated in the district of Jerantut, Pahang, Malaysia. Tekam Forest Reserve is also part of the main range called Banjaran Titiwangsa that stretches in the centre of Peninsular Malaysia. The study area, which covers an area of 10 x 10 km lies between 102o 32’ 24” E to 102o 37’ 48” E and 03o 57’ 36” N to 04o 03’ 00” N (Figure 1). It is accessible from a logging road and adjacent oil palm plantation roads. The forest type of the area is defined as hill dipterocarp, which is common at elevations between 300 to 750 m above sea level and forms the bulk of the production forests in Peninsular Malaysia. Hill dipterocarp forests are less commercially productive than lowland forests, but are still rich in floristic composition. Many lowland forest species are also found here but less frequently, while many species that occur here are not found in the lowland dipterocarp forests. The topography of the area is undulating with steep and rugged slopes exceeding 45º. The elevation ranges from 60 to 800 m. The area was selected because it is adjacent to the Forest Research Institute Malaysia (FRIM) research station. It is accessible, as harvesting activities are still ongoing in the north of the study area. The earliest logging occurred in 1976 and the most recent in 1986. The total area covered by the study is about 11 800 ha. There are 27 full compartments within the study area covering about 8 015 ha. The remaining 3 790 ha consist of partial compartments (Table 2).

Sampling design In the most direct approach, plots are placed randomly or systematically over the entire area. However, such an approach may not be efficient in stands that are not uniform and where variation is high. In this case, stratified sampling is more efficient especially when the area has been logged at different times (temporal differences) and different intensities. For example, cutting intensities may vary among compartments. Also, there is a tendency to remove more trees closer to the road and to avoid removing trees in difficult terrain. Thus it is better to calculate separate estimates for each stratum. Overall estimates can still be made for the whole stand. Stratified estimates of the population mean and total produce smaller variance than non-stratified estimates. The practical implication is that stratified sampling produces more precise estimates (i.e. the standard error is smaller). Stratified sampling may also allow a reduction of the total sampling units. However, the advantage of this approach is realized only if the stratification is done properly. For surveying logged-over forests, the ability to stratify is very useful. In this study, the stratification is carried out with the help of satellite data and the strata are classified based on tree densities. For the development of a rapid appraisal technique the stratified sampling design is the most practical. It minimizes costs by localizing inventory samples. The stratification of the forest area is undertaken based on recent satellite images. The timing of the last logging entry could also be used as a stratification criterion. This approach has been adopted for logged-over forests by the Forestry Department Peninsular Malaysia in its national forest inventory. In this study, the

11 condition of the forest was highly variable and appeared to depend more on the quality of the logging operations than on the timing of the last logging entry.

LOCATION OF THE STUDY AREA TEKAM FR, JERANTUT, PAHANG 200000 300000 400000 500000 600000 700000

Scale 1:300000 N 0 50 100 150 Kilometer s

W E 700000 S 700000 600000 600000 500000 500000 400000 400000 300000

STUDY AREA : TEKAM FR, 300000 JERANTUT, PAHANG

11 0 10 9 10 8 11 3 10 7 11 1 11 2

10 6 11 4 10 5 10 4 43 11 5

10 2 200000 42 56 10 3 2 72 11 6 11 7 1 3 41 44 40 57 70 10 1 11 8 4 16 58 71 73 10 0 12 0 6 5 54 55 11 9 17 45 12 1 12 4 15 69 74 75 8 39 53 12 2 12 5 7 46 59 99 12 3 12 6 14 18 76 77 12 8 12 7 200000 9 51 67 68 98 13 19 47 52 13 4 12 9 10 26 27 60 11 12 25 66 78 13 5 13 0 79 97 21 20 61 13 3 24 37 48 50 23 65 13 2 13 1 28 64 22 29 62 83 82 36 49 81 13 8 63 84 30 31 85 93 86 87 32 92 33 88 91

34 89 35 90

200000 300000 400000 500000 600000 700000

Figure 1. Location of study area

Accessibility within the logged-over forest is problematic. Therefore, a cluster sampling approach was adopted to reduce travelling time. In such an approach several plots are located close to each other to form a sampling unit. Each cluster forms a group of secondary plots at each location. The unit of observation is not the individual secondary plots but the entire cluster. In cluster sampling, the first stage is a selection of primary points rather than finite sampling unit areas. The second stage is a cluster of sample plots centered on the primary point, laid out in a pre-determined format.

12 Table 2. Compartments within the study area showing years after logging and extent Compartment Year since logging (years) Size (ha) 75 22 325.077 77 22 230.870 79 25 433.444 81 25 316.174 82 25 209.482 97 25 290.275 98 25 224.452 99 18 217.427 100 18 226.375 118 17 270.231 119 17 409.571 120 15 407.207 121 17 349.329 122 18 192.442 123 18 402.613 124 15 240.829 125 15 228.879 126 15 243.230 127 15 310.859 128 18 315.257 129 18 231.012 130 25 348.614 131 25 367.580 132 25 291.443 133 18 174.723 134 18 385.079 135 19 372.667 Portions of other compartments 3 790 Total 11 805

To overcome bias, plots are established at the intersection of 1-km grids. For a rapid appraisal the plots are located in the vicinity of accessible roads. Each cluster is considered a sampling unit and consists of five plots. The centre plot of the cluster is located at the grid intersection while the other four plots are located at a distance of 100 m to the north, east, south and west (Figure 2). Thus the effective size of the sampling unit is 100 x 100 m or 1 ha. The number of plots depends on the desired level of accuracy. It is also influenced by the availability of funds, which in turn depends to some degree on the estimated value of the forest stand. Usually, more valuable stands are sampled more intensively. In most forest inventory operations the probability level is accepted at 95 percent. The accepted standard error (SE%) for volume estimates of production forests can vary from 10 to 20 percent depending on the forest types and stand conditions. For Peninsular Malaysia, the SE% for national and management unit inventories is 15 percent for areas logged more then 20 years ago. For areas logged less then 20 years ago, the SE% is 20 percent. Based on the formula below, the number of samples for various strata as adopted by the Forestry Department Peninsular Malaysia is shown in Table 3. 2 2 2 ni = t * (CVi% )/(SEi% )

ni = total number of plots for stratum i t = value on a confidence (probability) level of 95% ≈ 2 CVi% = coefficient of variation of stratum i SEi% = standard error

13 N

Grid

Sample plots Sampling unit

100 m

Grid 100 m Grid

100 m 100 m

Grid

Plot design

Quarter circle

Variable 20m plots

Fixed plots 2m 

5m

Sampling unit plot centre

Figure 2. Layout of the sampling unit and plot layout

14 Table 3. Number of sample units per stratum at state and national levels

Forest type Stratum Statistics No. of units per state Name No. CV% SE% Dipterocarp Virgin forest – good to superior 11 30 15 16 production Virgin forest – poor to moderate 12 45 15 36 forest Logged-over 1-10 years ago 20 50 20 25 Logged-over 11-20 years ago 21 45 20 20 Logged-over 21-30 years ago 22 40 15 28 Logged-over 31-40 years ago 23 35 15 22 Logged-over 41+ years ago 24 35 15 22

Estimating the sample size required for estimating population based on a two-stage sampling design requires reliable estimates for both the primary sampling units (i.e. the forest strata) and the secondary units (i.e. plots within each stratum). In most situations, this information is unavailable before the inventory (Shiver and Borders 1996). It is possible to use the information from the previous national forest inventories as a guide but it may not be accurate because of the changes and variability between different logged-over forest stands. However, in general it is the variation between the forest strata (primary sampling units) that is much greater than the variation between plots of each stratum (secondary sampling units). Consequently, as the number of plots is limited, the aim should be to distribute the plots to all the strata proportionally. The management unit for Peninsular Malaysia is the state. Following the stratification by years since logging, the study area has two strata, namely logged-over (11 to 10 years) and logged-over (21 to 30 years). The total number of sampling units required based on the CV for these classes obtained from the Third National Inventory undertaken in 1991/92 amounts to 48. This information could be used as a guide but would not be very accurate. However, the study area is much smaller and for a rapid appraisal the number of plots for each of the classes identified should be fewer. The calculation of the CV is based on the preliminary inventory samples and the number of plots is determined for each class. This is undertaken after the collection of data in the field. At the same time the stratification is not based on years after logging but on tree density of the residual stands (see “classification”, below).

Plot layout

Each sampling unit consists of a cluster of five plots. The plot design employs a combination of variable plots (point sampling) and fixed plots (Figure 2). Trees with a dbh equal or greater than 15 cm are enumerated in the variable plots while the smaller trees (less than 15 cm dbh) are enumerated in fixed circular plots. The variable plot samples, proportional to the size of tree and therefore the same sampling intensity, are allowed for small and larger trees. In other words, the larger trees that contribute significantly to the standing volume are given better chances of being tallied compared to fixed plots (with samples being proportional to the frequency of stems within the plot). This allows for better estimates of basal area and volume. In addition, the point sampling does not require the demarcation of plot boundaries. This reduces costs and also avoids sampling errors normally associated with poor boundary demarcation in fixed plot methods. However, for the smaller trees, the point sampling method is not efficient and thus a fixed plot design is adopted to obtain reliable estimates. Details of the plot design and parameters measured are given in Table 4. The inventory also collects data on the presence of bamboo, rattan, palms and other non-timber forest products (NTFPs). Such information is useful and can indicate the degree of disturbance, as usually the abundance of bamboo and rattan is associated with large canopy openings due to logging.

15 Table 4. Plot design and parameters measured Plot shape/size Tree sizes sampled Parameters measured Variable plot All trees ≥ 15 cm dbh Species, dbh, crown condition, illumination class, wood quality and weed infection Quarter circle Estimate of non-timber Estimated extent and species product distribution e.g. rattan, bamboo, palms

20 m

20 m Fixed circular All saplings ≥ 5 cm and Species, dbh < 15 cm dbh

5 m

Fixed circular All seedlings < 5 cm Species, counts dbh and >1.5 m height.

2 m

Tally sheet The two tally sheets used for collecting data in the field are shown in Annex 1. The first sheet is for general information such as identification, location, elevation and NTFPs. Information on location includes a global positioning system (GPS) reading and forest strata information according to tree densities. Information on NTFPs is limited to rattan, bamboo, palms and others. For trees above 15 cm dbh, in addition to diameter, information is also collected on stem quality, crown quality, illumination and other species (e.g. climbers and lianas). The information is important because in addition to standard stock and stand information on the trees in the residual stand, forest managers are also interested in the general condition of those trees. For example, many large trees may have been left behind in the residual stand. This leads to a high volume and basal area. However, most of these trees may be damaged and decaying and from a commercial point of view they are not valuable. Details of the measurements and codes used are given in Annex 2. The second tally sheet is for recording data for the two fixed plots. In the first 5-m radius fixed plot, data on saplings with a dbh between 5 and 15 cm are collected. This will record the number of saplings as well as basal area and volume. However, some forest managers may not be interested in the basal area or volume of small trees. In this case, information of the dbh can be omitted and only trees need to be counted. The next sub-plot is the 2-m radius fixed plot that is used to record seedling numbers. The information is useful for estimating the regeneration potential of the stand. Details of the measurements and codes used are given in Annex 2.

16 Forest stratification Acquisition of spatial data Spatial data collection starts with compiling satellite images, maps and ancillary data (e.g. compartment records). Topographic maps are used and forest maps are required to obtain information on land use, contours, rivers, roads, compartment boundaries and years of logging. A 1998 satellite image of the study area (Figure 3) was purchased from the Malaysian Centre of Remote Sensing (MACRES). Harvesting records were obtained from the Forestry Department.

Figure 3. Raw image of Landsat TM band 543 for Tekam Forest Reserve

Image processing Raw satellite images need to undergo all related pre-processing operations such as geocoding, filtering, masking and unsupervised classification. The quality of this pre-processing contributes substantially to the accuracy of the final thematic products. In this study, two image-processing techniques were adopted, i.e. the conventional technique using ERDAS software and the FCD model.

17 Using ERDAS software The ERDAS remote sensing software was used for processing the Landsat TM image of the study area. First, geometric correction was conducted based on the transformation derived from a set of ground control points (GCP) from the topographic map. This ensured that the image’s location was positioned to its exact and true location. In the next step, the image was filtered to minimize speckle and noise. During unsupervised classification the program classifies the image into several assigned classes, in this case 10 classes (Figure 4). When the topographic map was overlaid onto the image with the 10 classes, features like forest, agriculture, roads and open areas could be determined easily. These features were divided into two categories, i.e. forest and non- forest. Five classes were classified as forest, four as non-forest areas and one as mountain ridge. The classes under the forest category needed to be verified further during ground truthing. Ground-truthed data were used to define regions of interest (ROI) in the classified image. This information was used to reclassify the image during supervised classification. The supervised classification image produced a clearer and better picture of the forest categories (Figure 5).

Figure 4. Image of unsupervised classification using ERDAS

Using the FCD Model Unlike the conventional remote sensing method that assesses the forest status based on qualitative data analysis derived from “training areas”, the FCD model is based on forest conditions, which is a quantitative analysis. FCD utilizes forest canopy density as an essential parameter for characterization of forest conditions. The degree of forest density is expressed in percentages. It also indicates the degree of degradation and hence, prioritizes sites in need of rehabilitation. The principal features of the FCD include 1) rapid stratification of forests into canopy density categories (i.e. 0 to 100 percent); 2) production of tables showing the number of hectares in each category; and 3) a printout of coloured maps that clearly illustrates forest conditions (Rikimaru et al. 1999).

18 Figure 5. Supervised classification image using ERDAS

The source of remote sensing data for the FCD-Mapper is Landsat TM. The FCD-Mapper is a semi-expert system for analysing satellite imagery and is compatible with Microsoft©software.. In this study, the same Landsat TM image that was processed by ERDAS was used. The image was first processed for noise reduction because clouds or cloud shadows or water areas can influence the statistical treatment and analysis of the data adversely. This was followed by range normalization of the TM data for each band. The FCD model combines data from four indices: 1. Advanced Vegetation Index (AVI) 2. Bare Soil Index (BI) 3. Shadow Index or Scaled Shadow Index (SI, SSI) 4. Thermal Index (TI) The four index values were expressed in percentages for each pixel. Using the above four indices the forest canopy density was determined. Similar to the image processed by ERDAS, unsupervised classification was carried out and 10 classes were assigned (Figure 6). However, ground truthing could identify only four different classes. The results of supervised classification are shown in Figure 7. The flowchart of the procedures for the FCD model is illustrated in Figure 8.

Ground truthing Ground truthing was carried out to verify the compartment boundaries, forest classes and to note any special feature that could not be detected in the image. For each forest class, several ground truthing plots were established and information such as stem frequency, sizes and structure was collected. Basically, ground truthing provides details of the forest condition including canopy layers, dominant tree species, elevation, location, understorey vegetation and tree diameters. The data were used to define the classified image and related maps were produced.

19 Figure 6. Unsupervised classification image using FCD

20 Figure 7. Supervised classification using FCD

21 LANDSAT TM data

Noise reduction process scan line noise, atmospheric noise, cloud area, cloud shadow area, water area, etc.

Range Normalization of TM data for each bands

Advanced vegetation index Bare soil index Shadow index Thermal index

Vegetation/bare soil Black soil detection synthesis model

Advanced shadow index Spatial process

Shadow Vegetation density % Scaled shadow index percentage for forest

Integration model

Forest canopy density map

Figure 8. Flow chart of FCD Mapping Model (Rikimaru, 2002)

22 Results and discussion Image analysis results Processed by ERDAS software The supervised classification image (Figure 5) produced five forest classes that, when verified on the ground, gave the following results (Table 5).

Table 5. Verification of the forest classes in the ERDAS supervised image

Class Area (ha) Verification 1 2 463 Ridge areas with a canopy structure consisting of two layers. Trees consist mainly of medium-sized commercial trees and many seedlings. Some open areas are colonized, mainly by ferns. 2 4 955 Lowland areas with three distinct canopy layers. These regenerating forests consist of mainly less valuable non-dipterocarp trees, which are small with dbh < 60 cm, although occasionally there are some bigger trees. 3 3 678 Forest areas, which are more open (probably due to heavy logging). They have a one-layered canopy structure with very few climax species. They are mostly colonized by pioneers and other light-demanding species. Bamboos and palms are abundant in some areas. 4 539 These are areas that are more representative of the original forests in terms of their canopy structure. They have three canopy layers and many large trees. However, non-dipterocarps dominate. Occasional dipterocarps may be protected and other trees that were not removed during logging. 5 239 These areas are also representative of the original forests both in canopy structure and species composition. They may consist of fully regenerated forests or areas that were not disturbed by logging activities. The canopy has three layers and valuable commercial species of dipterocarps and non- dipterocarps abound.

Another feature that is clear from the image is the elevation of the study area (Figure 9). When overlaid onto the contour map, the classified image clearly showed the differences in elevation. This was also verified by the ground truthing.

The map depicting the year of the last logging entry was also overlaid onto the classification image (Figure 10). This procedure intends to relate the forest condition with the year the forest was logged. The compartment number, size and year since logging are shown in Table 2. Attempts were made to obtain information on the logging intensities and volume of timber removed from the areas. Visual comparison of forest conditions indicated that years since logging may be a poor parameter for stratification of the forests because of differences among logging operations. The condition of the residual stand depends to a considerable extent on the logging practices employed by different contractors. In extreme cases of poor logging the area was badly logged and the residual forest was not able to regenerate according to the prescribed cutting cycle.

Processed by the FCD model Supervised classification reduced the 10 classes assigned by the unsupervised classification to four classes (Table 6). Forest density class 2 is most extensive covering 46 percent of the area. The second most extensive class is forest density class 1, which covers 34 percent of the area. Based on the canopy densities, it can be assumed that these two classes have fewer big trees and more saplings and seedlings. This indicates a stand that has been logged more recently and requires a longer time to recover before harvesting can be carried out again. The supervised classification also shows that the area with more mature trees and higher canopy densities covers only about 20 percent of the total area.

23 512000 514000 516000 518000 520000 522000 524000 450000 N 450000 ELEVATION MAP

448000 TEKAM FR, JERANTUT, PAHANG 448000 Scale 1 : 70,000 012Km 446000 446000

LEGEND Cliff

444000 2 layers, many seedlings, medi um comm. trees

444000 3 l ayers, smal l, less valuable trees 1 l ayer, small non-comm. t rees 3 l ayers, many seedl ings, 442000 large non-comm. trees 3 l ayers, densed, 442000 large comm. trees Non-Forest Contour 440000 440000

512000 514000 516000 518000 520000 522000 524000

Figure 9. Elevation overlaid onto classification image

Table 6. FCD results based on the supervised image FCD classes Class Colour Coverage area (% crown) (ha) % >70 4 Green 472 4 50-70 3 Yellow 1 889 16 30-50 2 Brown 5 431 46 <30 1 Blue 4 014 34 Total 11 806 100

24 514000 516000 518000 520000 522000 524000 N

101 118 1984 1984 448000 120 YEAR OF LOGGI NG, 448000 100 1986 TEKAM FR, 119 121 JERANTUT, PAHANG 1984 1984 124 1986 Scale 1 : 70,000 75 446000 1979 122 125 126 012Km 446000 99 1983 1986 1983 123 1986 1983 128 LEGEND 77 1983 127 Cliff 98 1986 2 layers, many seedlings, 134 444000 medium comm. trees 1976 1983 129 3 layers, small, less val uable trees 444000 1983 1 layer, 78 small non-comm. trees 97 135 130 3 layers, many seedlings, 1976 1982 1976 large non-comm. trees 79 133 3 layers, densed,

1981 442000 large comm. trees 1983 Non-Forest

442000 132 131 Road 1976 1976 Trail 82 127 Compartment No. 1976 1986 Year of Logging 81 440000

440000 1976 138 1138975 514000 516000 518000 520000 522000 524000 Figure 10. Year of logging overlaid to supervised classification image.

Field data collection Data collection in the field took place between late December 2001 and February 2002. The planned activities were slowed down by unusually heavy rains. A reconnaissance inventory consisting of a small numbers of plots was conducted in each forest stratum. Plots were located at grid intersections close to roads. For the point sampling of the large trees, a wedge prism of basal area factor (BAF) 4 was used. The choice of BAF usually depends on the density and size of trees within the stand. Denser stands and larger trees require a higher BAF factor. Wide-angle could also be used, but they are more expensive and more cumbersome to handle especially in moist conditions. However, they are easier to apply on steep terrain, as relascopes automatically compensate for elevation. The wedge prism was considered adequate for the purpose of the study. It is usually easier to perform the survey with the same BAF for the whole inventory, although changing the BAF for different forest strata does not affect the calculations. In fact, if the crew is not constrained in the field, it is recommended to use a lower BAF of 2 or 3 for more open forest strata with a larger proportion of small trees. Two field crews of four persons each were deployed in the field. Each crew consisted of a field leader who also recorded information. One person equipped with a compass and a GPS led the team to the sampling lines and located plots (This person may also take the prism readings for the point sampling.). The other two persons measured trees, established plots and sometimes took prism readings. The standard equipment used included:

25  Prisms  Compass  GPS (Garmin)  Diameter tapes  Clinometers  Flagging tape  50 m distance tapes  Laser impulse – for measurements of distance, height and bearings Sampling points were determined on the map and located on the ground with the help of topography maps and the GPS. Points were demarcated clearly on the ground with PVC pipes. During the actual inventory, some points may be revisited for auditing or for continuous and periodic measurements.

Data input Data variables Two sets of data were entered into a spreadsheet program (Microsoft© Excel 2000). The variables included in the first data set were: (1) Compartment number (2) Forest density class (3) Sample number (4) Elevation (5) Type of NTFP and its abundance (6) Tree species code (7) Dbh for trees above 15 cm in diameter (8) Timber quality (9) Crown form (10) Crown illumination (11) Competing vegetation

The second data set consisted of 1) to 4) in the first set plus: (1) Tree species code for trees (2) Diameter of trees between 5-15 cm dbh (3) Number of trees between 1.5 m height and 5 cm dbh

Data verification The data in the spreadsheets were exported to S-Plus for Windows Version 6 Release 2 for data verification. Summary statistics (e.g. mean, median total sum) were produced for each variable as well as quartiles for continuous variables, and counts for categorical variables. If a variable exceeded logical consistencies, the raw data were checked, and if necessary ground checks were made to verify data accuracy.

Data analysis Analysis software S-Plus for Windows Version 6 Release 2 was used to analyse the data. A program code based on S- language was written to summarize the data by each forest stratum classified using the FDC Model.

26 Analysis approach The study site was divided into four strata: (1) > 70 percent; (2) 50-70 percent; (3) 30-50 percent; and (4) < 30 percent of canopy density cover. Tree species were grouped into dipterocarps, non- dipterocarps and all species combined. Four data analysis program codes (FAO 1 to 4) were written:

FAO-1 A data summary of trees greater than 15 cm dbh for each stratum and species group. The stand parameters analysed were: a. average trees per hectare b. average basal area per hectare c. average volume per hectare d. CV for basal area per hectare e. CV for volume per hectare f. point sample estimators

Stand and stock tables were generated for each sampling point. A calculation of the stand and stock tables is given below: Analysis steps: a. Calculate basal area per tree (bat) and volume per trees (volt).

2 batj = ! * dbhi / 40000 [1] voltj = batj * FQ * Log * 5 [2]

where FQ: form quotient of 0.65 Log: number of 5 m logs j is the individual tree record

The number of 5 m logs differs by size class (Table 7).

Table 7. Tree dbh and its equivalent 5 m logs Tree dbh Number of 5 m logs 15< 30 cm 1 30< 60 cm 2 60< 75 cm 3 > 75 cm 4

For example, for tree no. 4 (see Table 6) with a dbh of 30.8 cm, the basal area and volume are: 2 2 bat4 = ! * 30.8 / 40000 = 0.07 m 3 volt4 = 0.07 * 0.65 * 2 * 5 = 0.48 m

b. Calculate the number of trees per hectare (tph), basal area per hectare (bah) and volume per hectare (volh) for individual tree data

tphj = BAF/batj [4a] bahj = batj * tphj [4b] volhj = voltj * tphj [4c] where BAF is 4.

27 For example, for sample tree no. 1 (Table 8) with a dbh of 20 cm, the number of trees, basal area and volume per hectare, tph1 = 4/0.03 = 130 trees 2 bah1 = 0.03 * 130 = 4 m /ha 3 volh1= 0.10 * 130 = 13 m /ha

Table 8. Example of stand and stock table for point sample 1 (1) (2) (3) (4) (5) (6) (7) Sample dbh bat volt tph bah volh tree (m3/tree) (m2/tree) BAF/(3) (m2/ha) (m3/ha) (trees/ha) (3) x (5) (4) x (5) 1 19.80 0.03 0.10 130 4.00 13.00 2 130.60 1.34 17.41 3 4.00 52.00 3 59.90 0.28 1.83 14 4.00 26.00 4 30.80 0.07 0.48 54 4.00 26.00 5 56.60 0.25 1.64 16 4.00 26.00 6 66.30 0.35 3.37 12 4.00 39.00 7 41.50 0.14 0.88 30 4.00 26.00 c. Summarize data by sampling points The summary of sampling point data for each parameter is constructed by summation of the individual tree data (Table 7).

tphi = " tphj [5a] bahi = " bahj [5b] volhi= " volhj [5c]

where i is the sampling point tph, bah, volh and j as explained above

For example, for point sample 1 (Table 2), the number of trees, basal area and volume per hectare are: tph1 = 130 +3 +…..+ 30 = 258 trees/ha 2 bah1 = 4 + 4 +……. + 4 = 28 m /ha 3 volh1= 13 + 52 + … +26 = 208 m /ha d. Summarize the data by forest density class and calculate the CV The calculation of tph, bah and volh is done by dividing the sum of parameters of a sampling point within each forest canopy density class by the number of sampling points.

tphh = " tphi/n [6a] bahh = " bahi/n [6b] volhh = " volhi/n [6c]

where k is the forest canopy density class. tph, bah volh and j as explained above n is the number of sampling points in each forest density class For example, for Forest Canopy Density Class 1 (Table 9) for trees of 15 cm dbh and above, the number of trees, basal area and volume is:

28 tph1 = (258 + 469 +…..+ 161)/5 = 195.8 trees/ha 2 bah1 = (28 + 40 +……. + 20)/5 = 22.4 m /ha 3 volh1= (208 + 325 +……+ 130)/5 = 174.2 m /ha

Table 9. Example of stand and stock table summary for Forest Canopy Density Class 1 Sampling tph bah volh point (trees/ha) (m2/ha) (m3/ha) 1 258 28 208 2 469 40 325 3 28 12 130 4631278 5 161 20 130

The CV was calculated as the ratio of the standard deviation (Sy) of the sampling units to the mean within each forest canopy density class:

CV = Sy/Ybar where CV is the coefficient of variation Sy is the standard deviation of the sampling unit to the mean Ybar is the mean of Y, and Y can be the number of trees, basal area or volume per hectare For example, if the mean of stratum A is 25.5 and the standard deviation is 10, the CV is: (10/25) * 100 = 40%.

e. Point sampling estimators Besides average values, other important point sample estimators are variances and standard errors, which can be used to determine the tract total of variables of interest. In general, the following equation is used to calculate the average value, variances and standard error of a variable (e.g. basal area or volume per hectare):

Yi = "BAF * Yij / batij [7]

where: Yi is the per hectare estimate of Y (or characteristic of interest) at point i (i=1,2….,n) BAF is the Basal Area Factor batij is the basal area of tree j on point i mi the number of sample trees in point i Yij the volume or basal area for tree j on point i

The variance of Y among points is 2 2 2 S y = ("Yi - ("Yi) /n)/(n-1) [8] The variance of Ybar is 2 2 S ybar = S y/n [9] The standard error of Ybar is

Sybar =Sy/√n [10] Estimates of the stand parameters of the tract total can be obtained simply by:

Thaty = Ybar * A [11] SThat = A * Sybar [12]

29 Where A is the tract area in hectares Thaty is the tract total value of Y Ybar is the per hectare average value of Y SThat is the tract total standard error Sybar is the per hectare standard error of Ybar

For example, for Forest Canopy Density Class 1, >15 cm dbh and above (Table 9), If Yi = volume per hectare at point i (i.e., Y1 = 208, Y2 =325, and so on), therefore "Yi = 871 and "Yi2=188 773 Using the previous equation, the average volume per hectare is calculated as Ybar =(1/n) * "Yi = 871/5 = 174.2 m3 The variance among points is calculated as 2 2 2 2 3 S y = ("Yi - ("Yi) /n)/(n-1) = (188 773 - (871) /5)/(5-1) = 9 261.2 m The variance of the mean is calculated as 2 2 3 S ybar = S y/n = 9 261.2/5 = 1 852.2 m 3 3 The standard error of the mean (Sybar ) per hectare is √1852.2 m = 43.04 m An approximate 95 percent confidence interval for the mean volume per hectare is calculated as 174.2 ± 2(43.04) or LCL = 88.1 m3 UCL = 260.3 m3 Given a tract area of 50 ha, the total tract stand parameters can be calculated,

Thaty = Ybar * A 3 Thaty = 174.2 * 50 = 8 710 m The standard error of the tract total is calculated as 3 SThat = A * Sy = 50 * 43.0 =2151.9 m An approximate 95 percent confidence interval for the mean volume per hectare is calculated as 8 710 ± 2(2 151.9) or LCL = 13 013.8 m3 and UCL = 4 406.2 m3

FAO-2 A data summary of trees between 5 to 15 cm dbh. The stand parameters analysed were: a. trees per hectare b. basal area per hectare c. CV for basal area per hectare

The method of calculation for the number of trees, basal area and the CV is similar to FAO-1.

FAO-3 A data summary of trees between 1.5 m height to 5 cm dbh. The stand parameter analysed was: a. Number of trees per hectare

FAO-4 A data summary of NTFPs. The stand parameter analysed was: a. Percentage of area occupied by the vegetation

30 Data analysis The analysis of the plots placed in the different strata revealed the distribution of trees (Table 10). It must be stressed that this is not a complete ground truthing exercise and thus the number of plots for each stratum is not sufficient. A more complete ground truthing exercise requires about 10 plots per stratum. At this stage, the results are only of an indicative nature. Consequently, the results did not yield a consistent trend for the different FCD classes. However, some patterns emerged. FCD class 1 (blue stratum) for trees 15 cm dbh and above, although the total number of trees (360 trees/ha) is much higher compared to the other classes (class 4 with lowest average trees per hectare), the volume per hectare (174 m3/ha) is lower than that of class 4 (green stratum), which had 224 m3/ha with only 253 trees/ha (Table 10). This could be attributed to class 1 having more small trees compared to class 4, which had larger trees. It could also mean that the logging intensity in classes 1 and 2 could have been higher. The FCD class maps indicate that most of classes 1 and 2 are located closer to the roads compared to class 4. At the same time, it could also mean that classes 1 and 2 have a much lower number of large trees. If records on the number of trees and logging intensity were available the actual situation could be verified based on the number of trees removed. However, based on experience and field inspection, this assumption is probably correct. Logging records that were provided contained information of logging year, total volume removed and cutting regime. Information on the stocking of the original stand conditions and logging intensity was not available. However, the trend is less clear in a comparison of classes 2 and 3.

Table 10. Summary of point sample estimators by minimum dbh class and FCD class Dbh class FCD No avg avg bah avg volh Variance Standard error QD (cm) class pt tph bah volh bah volh 15++ 1 5 360.8 28.0 174.2 17.6 635.4 4.2 25.2 31.4 15++ 2 6 228.5 23.3 160.3 1.5 289.2 1.2 17.0 36.1 15++ 3 7 341.9 26.9 163.4 8.2 835.8 2.9 28.9 31.6 15++ 4 8 253.1 28.5 224.3 27.1 1 785.1 5.2 42.3 37.9 30++ 1 5 110.8 18.4 143.0 4.2 422.5 2.0 20.6 46.0 30++ 2 6 113.5 18.7 145.2 2.8 320.2 1.7 17.9 45.8 30++ 3 7 104.9 17.7 133.7 20.5 1 204.8 4.5 34.7 46.4 30++ 4 8 128.0 24.0 209.6 17.7 1 553.8 4.2 39.4 48.9

45++ 1 5 46.6 12.8 106.6 7.0 463.1 2.7 21.5 59.2 45++ 2 6 45.0 12.0 101.8 6.4 523.0 2.5 22.9 58.3 45++ 3 5 73.5 17.6 140.4 20.2 1 122.2 4.5 33.5 55.2 45++ 4 8 47.2 16.0 157.6 11.4 1 131.3 3.4 33.6 65.7 60++ 1 4 16.1 7.0 74.8 1.0 151.4 1.0 12.3 74.5 60++ 2 5 14.6 6.4 70.2 2.6 432.6 1.6 20.8 74.8 60++ 3 5 17.1 6.4 67.6 1.0 142.0 1.0 11.9 68.9 60++ 4 7 22.2 11.4 135.6 7.2 892.1 2.7 29.9 81.0 Note: No. pt: Number of point samples in each FCD class avg: average tph: average trees per hectare (trees/ha) bah: average basal area per hectare (m2/ha) volh: average volume per hectare (m3/ha) QD: Quadratic mean dbh (cm) Class 1: blue stratum on the FCD map Class 2: brown stratum on the FCD map Class 3: yellow stratum on the FCD map Class 4: green stratum on the FCD map

31 Class 3 had a significantly higher concentration of trees greater than 45 cm dbh compared to the other classes. However, it did not contain as many large trees. This could be seen from the quadratic mean. Class 4 had more trees of larger sizes (>60 cm), where the quadratic mean was higher and the basal area and volume of trees was also proportionally higher. Class 3 was comparatively more uniform than class 4, which was considered least disturbed or best regenerated. The most uniform was class 2. This is reflected by the low variance and standard error for this class, which is considerable compared to the other three classes. Figures 11, 12 and 13 provide a graphical presentation of tree distribution by FCD classes expressed in trees/ha, basal area/ha and volume/ha respectively.

300 FCD 1 FCD 2 FCD 3 FCD 4

200 trees per hectare

100

0 15-30 30-45 45-60 60++ dbh class (cm)

Figure 11. Distribution of number of trees per hectare by dbh class and Forest Canopy Density (FCD) class

FCD 1 11 FCD 2 FCD 3 FCD 4

9 basal basal area per hectare

7

5 15-30 30-45 45-60 60++ dbh class (cm )

Figure 12. Distribution of basal area per hectare by dbh class and Forest Canopy Density (FCD) class

32 120 FCD 1 FCD 2 FCD 3 FCD 4

80 volume per hectare per volume

40

0 15-30 30-45 45-60 60++ dbh class (cm)

Figure 13. Distribution of volume per hectare by dbh class and Forest Canopy Density FCD) class

Calculation of sample size The calculation of the sample size depends on the CV and allowable error. In theory, point sampling occurs on an infinite population. Since a point has no area, there can be an infinite number of points even in the smallest area. Based on preliminary sampling, sample size can be estimated as: n = 4(CV)2/(AE%)2 [13] where: n is the sample size of characteristics of interest 4 is the approximate Z2 value for 95 percent confidence CV is the coefficient of variation in percent AE% is the allowable error in percent

For example, for FCD Class 1 with a CV of basal area per hectare of 34 percent (Table 11) and with and allowable error of 10 percent: n = 4(33.5)2/102 = 44.89 sample points

In other words, a minimum of 45 sample points is required to estimate the basal area of trees greater than 15 cm dbh with 95 percent confidence and a 10 percent allowable error.

Table 11 shows the number of samples required for the study area based on the CV for the different FCD classes at different accuracy levels for trees with a dbh above 15 cm for different allowable errors. A higher allowable error reduces the number of samples required. Forest managers will have to decide on the desirable level of accuracy to determine the number of sampling plots. For a rapid assessment of forest conditions, an allowable error of 15 percent is recommended. At this level, the total number of samples required for all the FCD classes within the study area calculated based on volume/ha amounts to 121 plots. The variation is higher for the higher density forest classes. However, the number of samples required for FCD Class 2 is very small (12 samples) because of the significantly lower CV for that class. The CV for each FCD class will be different if calculations are based on different dbh classes and subsequently, the number of samples required will also differ. Thus if managers are only interested in the large trees (e.g. trees with dbh >45 cm), the number of samples required should be calculated based on the CV of large trees.

33 Table 11. Number of samples for trees > 15cm dbh FCD class CV No. of samples at No. of samples at 15% No. of samples at 20% 10 % A.E A.E A.E bah volh bah volh bah volh bah volh 1 33.5 32.4 45 42 20 19 12 11 2 12.9 26.0 7 27 3 12 27 3 28.1 46.8 32 88 14 39 822 4 51.7 53.3 107 114 48 51 27 29 Total no. of samples 191 271 85 121 49 69 Note: bah: average basal area per hectare (m2/ha) volh: average volume per hectare (m3/ha) CV: Coefficient of variation (%) A.E: Allowable error

Once the number of samples required for each FCD class has been determined, the sampling points can be systematically distributed across the study area. A system of computer-generated grids is on the map based on the existing RSO grids to avoid bias (Figure 14). Sampling points need to be established at the grid intersections. The sampling points required are then distributed randomly for each FCD class (Figure 14). For example, for FCD Class 1 the 19 samples required are distributed randomly within the 200-m grid intersections for that class. For practical reasons, the distribution of the sampling points could also be based on their distance from existing roads. In this approach, preference is given to samples that are located closer to roads. Although this may generate a bias, it is more cost effective, as accessibility can be a major constraint in making an inventory of logged-over forests.

Figure 14. Sample points on FCD classification image

34 Conclusion This report describes a rapid appraisal technique for logged-over forests. A two-stage sampling design is recommended with each sampling unit consisting of a cluster of five sample plots. The sampling design also combines variable sampling plots for trees with diameters above 15 cm with circular fixed plots for smaller trees. For the variable (point) sampling, a prism with a BAF of 4 is recommended based on the size and distribution of trees within the stand. The main reason for adopting a two-stage sampling design is that remote sensing data coupled with the use of the Forest Canopy Density Model can be used for stratifying the forest. The rapid inventory method proposed enables the forest manager to classify the forest by canopy density classes and subsequently to obtain stock and stand reports for each class. The method is practical and can be implemented easily by the field crew with minimal training and supervision. Although detailed inventory data were not available, the FCD classification combined with ground truthing indicates that the forests in the study area consist mainly of young regenerating trees. The year since the last logging does not reflect adequately the readiness of the area to be re- logged. Many areas will not regenerate sufficiently within the cutting cycle of 30 years without intensive silvicultural treatments to enhance the productivity and growth environment.

Acknowledgements The authors are very grateful to the EC-FAO Partnership Programme for providing the opportunity to conduct the work. The need to develop an inventory methodology for the logged- over forest is indeed timely as these forests are expected to provide the much needed supply of raw materials for the forest industries sector in Peninsular Malaysia. However, their status in terms of timber yield and species composition is uncertain. A special note of thanks is extended to Thomas Enters for his patience and guidance in the preparation of this document. The authors would also like to acknowledge the continuous support of the Forestry Department, Peninsular Malaysia, in forestry research, which also made the field data collection possible. We are also very thankful to the Director-General of the Forest Research Institute Malaysia for his support rendered in the implementation and completion of this document.

35 References Alias, M.S., Zulkifli, M., Mohd Puat, D. & Ibrahim, S. 2001. Canopy density mapping – a new for forest management. Paper presented at the 13th Malaysian Forestry Conference. 20-25 August 2001. Johor Bahru, Malaysia. Appanah, S. 2000. Trends and issues in tropical forest management: setting the agenda for Malaysia. In I. Shamsudin, Nur Hajar Zamah Shari & Khoo Kean, eds. Tropical forest harvesting: new technologies examined. Kepong, Malaysia, Forest Research Institute. pp. 1-20. Bell, J.F., 1997. Log scaling and timber cruising. Corvallis, USA, Oregon State University. Canonizado, J.A. 1978. Simulation of selective forest management regimes. Malaysian , 41: 128-142. Ismail, H. & Appanah, S. 2000. Forest certification in Malaysia. Major constraints identified. In I. Shamsudin, Nur Hajar Zamah Shari & Khoo Kean, eds. Tropical forest harvesting: new technologies examined. Kepong, Malaysia, Forest Research Institute. pp. 204-213. Nezry, E., Yakam-Simen, F., Romeijn, P., Supit, I. & Demargne, L. 2000. Advanced remote sensing technique for forestry applications: a case study in Sarawak, Malaysia. Invited paper at Multi-Conference on Systemics, Cybernetics and Informatics, 23-26 July, 2000. Orlando, USA. Pinard, M.A. 1995. Carbon retention by reduced impact logging. Ph.D. thesis, University of Florida, Gainesville, USA Pelz, D.R. 1993. Concepts of forest inventories. Proceedings of the IUFRO Conference 1993, Advancement in Forest Inventory and Forest Management Sciences, Seoul, South Korea. pp. 11-21. Rikimaru, A. 2002. Concept of FCD mapping model and semi-expert system. Paper presented at the International Workshop on Tropical Forest Cover Assessment and Conservation Issues in Southeast Asia, 12–14 February 2002. Dehradun, India, IIRS. Rikimaru, A., Miyatake, S. & Dugan, P. 1999. Sky is the limit for forest management tool. Tropical Forest Update, 9(3): 6-8. Samsudin Musa & Kasinathan, K. 1989. Forest inventory manual volume II – current forest inventory practices in ASEAN. Kuala Lumpur, Malaysia, ASEAN Institute of Forest Management. Shiver B.D. & Borders, B.E. 1996. Sampling techniques for forest resource inventory. New York. Wiley. Taumas, R. 2000. Implementation of reduced impact logging (RIL) in Sabah: the Innoprose Corporation Sdn. Bhd. experience. In I. Shamsudin, Nur Hajar Zamah Shari & Khoo Kean, eds. Tropical forest harvesting: new technologies examined. Kepong, Malaysia, Forest Research Institute. pp. 33-59. Taylor, D. L., Samsudin Musa & Kasinathan, K. 1989a. Forest inventory manual volume IV – recommended forest inventory practice in ASEAN. Kuala Lumpur, Malaysia, ASEAN Institute of Forest Management. Taylor, D. L., Samsudin Musa & Kasinathan, K. 1989b. Forest inventory manual volume III – planning and implementing forest inventory. Kuala Lumpur, Malaysia, ASEAN Institute of Forest Management. Tay, J., Healey, J. & Price, C. 2002. Financial assessment of reduced impact logging techniques in Sabah, Malaysia. In T. Enters, P.B. Durst, G.B. Applegate, C.S. Peter Kho & Man, G., eds. Applying reduced impact logging to advance sustainable forest management. RAP Publication 2002/14. Bangkok, Food and Agriculture Organization of the United Nations. pp. 125-140. Urquinzo, M.C.M., Hussin, Y.A. & Weir, M.J.C. 1998. Forest canopy density of logged-over tropical rain forest using satellite images in Bukit Soeharto National Park, East Kalimantan, Indonesia. Proceedings of the FIMP-INTAG International Conference, 26-29 October 1998. Jakarta, Indonesia Yong, T.K. 1998. Forest growth and yield studies in Peninsular Malaysia. The Malaysian Forester, Vol. 61, No. 3: 131-176.

36 Annex 1: Second growth forest: Tally sheet used for Tekam Forest Reserve, Pahang

Date: Compartment: Forest strata:

Line No.: Plot No.: Elevation:

GPS: NTFP’s:

Inventory of trees dbh > 15 cm Point Sampling Plot No. Species name Code Dbh Tree quality Crown quality Illumination Weeds Remarks 1. 2. 3. 4. 5. 6. 7. 8. 9.

Plot information: Sub-plot 1. 5 m Radius Measure saplings 5–15 cm dbh

Name Code Dbh Name Code Dbh

Sub-plot 2 Radius 2 m Measure seedlings < 5cm dbh & height > 1.5 m

Species Code Counts Species Code Counts

37 Annex 2 Clarifications for completing tally sheet

Main plot – Variable circular plot

Column Measurements to record Date Date of enumerations Compartment Compartment number where plot is located Forest strata Forest stratum under FDC Classification Map Line no. Sampling line number Elevation Elevation as measured by appropriate equipment – GPS GPS: GPS coordinates of the plot centre NTFP Record the presence of NTFPs such as rattan, bamboo, palms and medicinal plants. Code 1 if presence is less 25% of plot area Code 2 if presence is greater then 25% of the plot area Species Local or scientific names Code Species code Dbh Record dbh to 1 decimal place for trees greater then 15 cm dbh Tree quality Code 1 – No defects; straight clear bole Code 2 – Minor defects, but can have 1 m log now or in future Code 3 – Defects present such as crooked bole, strong branching, leaning and will not have any log-sized timber Code 4 – Defects such as decay present and will not yield log-sized timber Crown quality Code 1 – Full and completely rounded Code 2 – Full but not completely rounded Code 3 – Not full but more then half full Code 4 – Less then half full Code 5 – Damaged crown and will potentially die Code 6 – Crown completely damaged but presence of re-growth Illumination Code 1 – Emergent trees with complete illumination on all sides Code 2 – Dominant trees receiving full light on top of crown Code 3 – Co-dominant crowns receiving light on part of crown Code 4 – Understorey trees with only part of the crown receiving some direct light Code 5 – Understorey trees not receiving any direct light Weeds Code 1 – No weeds or climbers present Code 2 – Presence of weeds such as climbers but not harmful to tree growth Code 3 – Weeds present but have been effectively treated (cut) Code 4 – Weeds present and have been cut but are re-growing Code 5 – Weeds present and are harmful for tree growth Remarks Record anything of silvicultural importance

Sub-plot 1 (5 m radius circular fixed plot) Column Measurements to record Name Local or scientific name of species Code Record species code Dbh Record diameter for all trees 5–15 cm dbh

Sub-plot 2 (2 m radius circular fixed plot) Column Measurements to record Species Local or scientific name of species Code Record species code Counts Record the number of all seedlings less than 5 cm dbh and greater than 1.5 m

38