The Drying of the Arkavathy River: Understanding Hydrological Change in a Human-Dominated Watershed

The drying of the Arkavathy river: understanding hydrological change in a human-dominated watershed

Gopal Penny

A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy

Engineering - Civil and Environmental Engineering

in the

Graduate Division

of the

University of California, Berkeley

Committee in charge:

Associate Professor Sally E. Thompson, Chair Professor Fotini K. Chow Assistant Professor Iryna Dronova Veena Srinivasan, PhD

Summer 2017 The drying of the Arkavathy river: understanding hydrological change in a human-dominated watershed

Abstract

The drying of the Arkavathy river: understanding hydrological change in a human-dominated watershed by Gopal Penny Doctor of Philosophy in Engineering - Civil and Environmental Engineering University of California, Berkeley Associate Professor Sally E. Thompson, Chair

Human interventions in the hydrologic cycle have intensified to the extent that water resources cannot be managed and understood in isolation from anthropogenic influences. New approaches are needed to understand the effects of humans on hydrology, especially in regions of the world with limited hydrologic records. This dissertation focuses on a case study of the Arkavathy watershed adjacent to Bangalore, India, which has been transformed by rapid urbanization, intensification of agriculture, and over-exploitation of water resources over the last 50 years. During this time, the disappearance of streamflow in the watershed was largely overlooked as Bangalore shifted from Arkavathy-sourced water supply to im- ported water and farmers from surface water to groundwater irrigation. With Bangalore continuing to expand its water footprint and local groundwater resources drying up, moving towards sustainable water resources management in the Arkavathy requires overcoming the general absence of local hydrological records to develop an understanding of the changing hydrology of the watershed. To this end, a multifaceted research approach is developed and applied to the Arkavathy watershed to identify the dominant hydrologic dynamics within the watershed and understand the conditions under which hydrologic change occurred. This research reveals a number of important findings. First, humans are the primary drivers of change in this watershed, as neither precipitation variability nor increases in temperature can explain the observed changes in hydrology. Second, hydrologic change within the watershed is spatially heterogeneous, with drying occurring in the northern part of the watershed and increased surface water availability downstream of Bangalore. Third, streamflow decline in the northern Arkavathy has most likely been caused by extensive groundwater depletion driven by groundwater irrigated agriculture. And finally, management strategies designed to reverse groundwater depletion by constructing check dams within the surface water network are unlikely to succeed on the scales pertinent to watershed management. In addition to understanding water resources within the Arkavathy, this work serves as a foundation for understanding the trajectory of water resources in the region. This research also presents an approach for investigating historical hydrologic change in a poorly monitored watershed, un- 2 derstanding human-water interactions, and supporting long-term predictions for sustainable water management. i

Contents

Contents i

List of Figures iv

List of Tables vi

1 Introduction 1 1.1 The need to study coupled human-water systems ...... 2 1.2 The Arkavathy watershed ...... 2 1.2.1 Regional context ...... 3 1.2.2 Potential drivers of change ...... 5 1.2.3 Climate and geography ...... 6 1.3 Research approach ...... 8 1.3.1 Conceptual framework ...... 8 1.3.2 Summary of chapters ...... 10

2 Multiple hypotheses of change 13 2.1 Introduction ...... 13 2.1.1 Challenges in rapidly growing, data-scarce regions ...... 13 2.1.2 Use-inspired science in data-scarce regions ...... 14 2.2 Drying of TG Halli reservoir ...... 15 2.2.1 Study area and the problem ...... 15 2.2.2 The debate about causes and solutions ...... 18 2.2.3 The multiple hypotheses approach ...... 19 2.3 Methods ...... 21 2.3.1 Data sources and quality assurance ...... 21 2.3.2 Analysis techniques ...... 21 2.4 Results ...... 26 2.4.1 Lack of evidence of climatic drivers ...... 26 2.4.2 Evidence of human drivers ...... 27 2.5 Discussion ...... 29 2.6 Conclusions ...... 30 ii

3 Spatial changes in surface water 32 3.1 Introduction ...... 32 3.2 Methods ...... 35 3.2.1 Study area and remote sensing analysis ...... 35 3.2.2 Statistical model of tank water extent ...... 37 3.2.3 Hydrologic change and land use ...... 40 3.3 Results ...... 41 3.3.1 Accuracy assessment ...... 41 3.3.2 Heterogeneity of long-term hydrologic change ...... 42 3.3.3 Streamﬂow decline and agricultural practices ...... 43 3.4 Discussion ...... 44 3.4.1 Long-term hydrological trends and human drivers of change . . . . . 44 3.4.2 Assessing the classiﬁcation and model uncertainty ...... 46 3.5 Conclusions ...... 47

4 Process-based reconstruction 48 4.1 Introduction ...... 48 4.1.1 Typology of hydrologic reconstruction ...... 49 4.1.2 A process-based reconstruction approach ...... 52 4.2 Methods ...... 53 4.2.1 Study area and runoff generation in semi-arid catchments ...... 53 4.2.2 Field instrumentation and experiments ...... 56 4.2.3 Storm event analysis ...... 58 4.3 Results ...... 59 4.3.1 Overland flow ...... 59 4.3.2 Saturated hydraulic conductivity ...... 60 4.3.3 Storm dynamics and timing ...... 60 4.3.4 Open well survey ...... 61 4.4 Discussion ...... 63 4.4.1 Contemporary streamflow generation ...... 63 4.4.2 Historical streamflow generation ...... 64 4.5 Conclusions ...... 66

5 Check dams and groundwater recharge 68 5.1 Introduction ...... 68 5.2 Methods ...... 69 5.2.1 Study area ...... 69 5.2.2 Field instrumentation and data analysis ...... 70 5.2.3 Thirumagondonahalli watershed simulation ...... 72 5.2.4 Simulation of synthetic check dam networks ...... 73 5.3 Results ...... 75 5.3.1 Observed streamﬂow dynamics ...... 75 iii

5.3.2 Eﬀect of check dams in the Thirumagondonahalli watershed . . . . . 76 5.3.3 Features of synthetic check dam networks ...... 77 5.4 Discussion ...... 78 5.5 Conclusions ...... 80

6 Conclusion 82 6.1 Summary of ﬁndings ...... 82 6.2 Future work ...... 83

A Supporting information for Chapter 3 85 A.1 Remote sensing analysis ...... 85 A.1.1 Remote-sensing images and supplementary data ...... 85 A.1.2 Classiﬁcation method ...... 89 A.1.3 Validation of classiﬁcation method ...... 92 A.2 Statistical model design ...... 95 A.2.1 Dry season analysis ...... 95 A.2.2 Collinearity analysis ...... 95 A.3 Statistical model analyses ...... 96 A.3.1 Precipitation timeseries ...... 96 A.4 Hydrological trends and agricultural land use ...... 103

B Supporting information for Chapter 4 104

C Supporting information for Chapter 5 105 C.1 Check dam water balance ...... 105 C.1.1 Evaporation ...... 105 C.1.2 Recharge ...... 106 C.2 Precipitation variability ...... 106

References 108 iv

List of Figures

1.1 Arkavathy watershed location map ...... 7 1.2 Conceptual framework of sociohydrologic system and research approach . . . . . 9

2.1 Important features of the TG Halli watershed ...... 16 2.2 Changes in hydrology and hydrometeorology from secondary data ...... 17 2.3 Irrigation water supply over time ...... 18 2.4 Change in Eucalyptus area, 1973–2001 ...... 28

3.1 Arkavathy map and Landsat scene boundaries ...... 33 3.2 Aerial photos of a small tank before and after runoﬀ events ...... 35 3.3 Map of major subwatersheds and tank cluster watersheds ...... 39 3.4 Tanks-scale validation of classiﬁcation ...... 41 3.5 Long-term hydrologic trend for each tank cluster ...... 43 3.6 Agricultural land use and hydrologic change ...... 44

4.1 Typology of hydrologic reconstruction ...... 49 4.2 Map of study watersheds and instrumentation within the TG Halli watershed . 53 4.3 Framework for investigating streamflow generation mechanisms ...... 55 4.4 Deuterium concentrations throughout two storm events ...... 59 4.5 Saturated hydraulic conductivity and precipitation rates ...... 60 4.6 Probability of runoff based on storm size and soil moisture ...... 61 4.7 Hydrologic variables for a large runoff event on 07 October 2014 ...... 62 4.8 Time lag from peak inflow to peak soil moisture ...... 62 4.9 Elevation of wells compared with elevation of the nearest stream channel . . . . 63 4.10 Depth to the bottom of dry wells in 2014 and water table depths c. 1970 . . . . 65

5.1 Locations of check dams in study watersheds ...... 70 5.2 Photos of a check dam under empty and near-full conditions ...... 71 5.3 Synthetic check dam configurations ...... 74 5.4 Daily timeseries of precipitation and water storage in a check dam and tank . . 75 5.5 Check dam storage status during runoff events ...... 76 5.6 Partitioning of precipitation and runoff into components of the water balance . . 77 v

5.7 Simulation results including check dam recharge, Gini coeﬃcient, and low-ﬂow years ...... 79

A.1 Timeline of classified Landsat scenes ...... 87 A.2 Treatment of missing pixels due to Landsat 7 SLC error ...... 88 A.3 Treatment of clouds and cloud shadows ...... 89 A.4 Flowchart of classification algorithm ...... 90 A.5 Dynamics of water extent (with Landsat classification for select years) and precipitation for example tank (A) ...... 92 A.6 Dynamics of water extent (with Landsat classification for select years) and precipitation for example tank (B) ...... 92 A.7 Pixel-level producer and consumer accuracy tables ...... 93 A.8 Additional tank-scale validation of classification ...... 94 A.9 Timeseries of water extent in reservoirs ...... 94 A.10 Dry-season analysis of water extent ...... 95 A.11 Annual precipitation in the Arkavathy watershed ...... 96 A.12 Timeseries of total precipitation metric for each tank cluster ...... 97 A.13 Quantile-quantile plot of residuals from regression ...... 98 A.14 Map of subwatershed names and tank cluster IDs ...... 99 A.15 Variability in tank water extent due to precipitation and hydrologic change . . . 102 A.16 Hydrologic change and time-averaged agriculture land use fraction ...... 103

B.1 Locations of ﬁeld measurements, open wells, and historical water table depths . 104

C.1 Variability of precipitation from two rain gauges ...... 107 vi

List of Tables

2.1 Parameters descriptions and data sources ...... 21 2.2 Number and type of stream encroachments ...... 26

A.1 Dates and identiﬁers of Landsat scenes ...... 85 A.2 Spatial and hydrologic data sources ...... 87 A.3 Regression results, including conﬁdence intervals around B1,j ...... 100 vii

Acknowledgments

My dissertation was made possible with support from multiple agencies and numerous in- dividuals. My research was primarily funded by the National Science Foundation (NSF) Graduate Research Fellowship Program under Grant No. DGE 1106400, and field research was funded by the NSF GROW program, USAID RI Fellowship, and NSF CNIC 1427420. This work was enabled through a collaboration with Ashoka Trust for Research in Ecology and the Environment (ATREE) in Bangalore, whose project “Adapting to Climate Change in Urbanizing Watersheds (ACCUWa) in India” was funded by the International Development Research Centre (IDRC), Canada, Grant No. 107086-001. The research contained in this thesis was supervised by Prof. Sally Thompson, and I cannot imagine a more insightful or supportive mentor. Sincere thanks to Veena Srinivasan whose mentorship was essential in shaping this research and my beliefs about the role of hydrology in society; to Sharad Lele for his guidance and support; and to everyone at ATREE who welcomed me into their community and supported this research, including Bejoy Thomas, Priyanka Jamwal, H. Usha, Kiruba Jeremiah, M. Mariappan, Apoorva R., Karthik Madyastha, G. Manjunath, C. Varan, and the EcoInformatics Lab. Thanks also to my committee members and co-authors Iryna Dronova, Tina K. Chow, Joshua Peschel, and Sierra Young; and to fellow members of the Thompson lab family, with specific thanks to David Dralle, Gabrielle Boisramé,and Xue Feng. Finally, I especially want to acknowledge my parents, Michael and Heather, my brother, Toshy, and my friends for their kind and steady support. 1

Chapter 1

Introduction

As societies continue to develop and increase demand for natural resources, water will frequently be the limiting factor on food supply (Hanjra and Qureshi, 2010; Jury and Vaux, 2007; Rockströmet al., 2009) and water scarcity the underlying source of conflict in some regions (Gleick, 2014, 1993). The consequences of water scarcity will be greatest in developing countries, where economic, institutional and demographic challenges often coexist with physical water scarcity (Fader et al., 2016; Molden, 2007; Vörösmartyet al., 2013). Low and middle income countries are hampered by limited resources to build and maintain water infrastructure, collect environmental data, and study hydrological systems (Houghton-Carr and Fry, 2006; Sene and Farquharson, 1998; Srinivasan et al., 2012a, 2017). Although these concerns can be partially alleviated by successful political negotiations over transbound- ary water sources (Müller et al., 2017) as well as transnational exchange of water resources through virtual water trading (Konar et al., 2011), these tools are unlikely to resolve water scarcity issues (Carr and D’Odorico, 2017; Suweis et al., 2015). There remains ample concern over sustainable provisioning of water resources for humans and nature in many regions of the world, especially given that population growth and climate change will exacerbate weather extremes, including drought (Fader et al., 2016) and adequate provisioning of water resources must be addressed at regional to local scales (Steffen et al., 2015). A common strategy to improve water security is to implement Integrated Water Resources Management (IWRM), which applies a holistic and flexible approach to water management (Biswas, 2004; GWP, 2000). For instance as part of the Sustainable Development Goals for Water and Sanitation, the United Nations has proposed applying Integrated Water Resources Management (IWRM) at all basin scales (Target 6.5, United Nations, 2016). Although IWRM and implementation of adaptive water governance must be pursued, successful IWRM requires an understanding of the quantity of water resources available which is complicated by the extensive anthropogenic modification of the environment in the Anthropocene (Crutzen, 2002; Steffen et al., 2007; Zalasiewicz and Williams, 2010). CHAPTER 1. INTRODUCTION 2

1.1 The need to study coupled human-water systems

Land and water management are transforming the environmental systems on which societies depend (Vörösmartyet al., 2004; Vörösmarty, 2000). Many of the world’s river basins are impaired in one way or another (Molden, 2007; Vörösmartyet al., 2010), and excessive groundwater depletion is common in many regions (Wada et al., 2012). These changes challenge the traditional models used by hydrologists to predict the behavior of hydrologic systems (Milly et al., 2008). Understanding and predicting the trajectory of water resources will require new approaches to understand the interactions between humans and water (Mon- tanari et al., 2013; Sivapalan et al., 2012; Srinivasan et al., 2016; Wagener et al., 2010). Multiple approaches have been suggested to address these challenges. One approach involves co-evolutionary modeling, where interactions between societal and environmental systems are incorporated into a dynamical system consisting of multiple differential equa- tions that describe relationships among state variables in the system (Di Baldassarre et al., 2013a,b; Elshafei et al., 2014, 2015). Although these models can potentially identify emergent dynamics (Di Baldassarre et al., 2013a), they are unlikely to capture dramatic shifts in system behavior initiated by exogenous drivers (e.g., technology), regime changes (Beven, 2016; Falkenmark et al., 2007), or system collapse (Holling, 2001). Other avenues have been suggested, such as comparative hydrology (inferring properties by considering similarities and differences among a large number of sites) and hydrologic reconstruction (reproducing historical processes or conditions) (Falkenmark et al., 1989; Gaálet al., 2012; Thompson et al., 2011). These could serve to develop a foundation for sociohydrology and inform co- evolutionary modeling (Sivapalan et al., 2012; Thompson et al., 2013b), and could proceed in parallel to ongoing work in predictions in ungauged basins (PUB, Blöschl, 2016; Siva- palan et al., 2003). Regardless of methodology, prioritizing hydrologic research in regions with pressing issues of water scarcity and water policy would serve the broader, long-term goals of prediction in hydrology as well as immediate and pragmatic purposes of sustainable water management and protection of ecosystem services (Falkenmark et al., 2009; Thompson et al., 2013b). These ideas constitute the motivation for this dissertation, which focuses on the human-dominated and water-stressed Arkavathy watershed, where research to address urgent water management can also yield new insights into human-water interactions and support better long-term predictions in hydrology (Falkenmark et al., 2007).

1.2 The Arkavathy watershed

The Arkavathy watershed in Karnataka, India, is a valuable case study given the intercon- nectedness of hydrologic and social systems in the watershed and extent of their historical interactions. The Arkavathy provided Bangalore, which is adjacent to and partially overlaps the Arkavathy, with all of its water through 1970, and supplied a sustainable network of surface water irrigation that existed for centuries to support agriculture in the remainder of the watershed. Over the last 50 years, water reservoirs in the Arkavathy have mostly CHAPTER 1. INTRODUCTION 3

gone dry, including two major reservoirs that supplied Bangalore, many surface irrigation reservoirs distributed throughout the watershed, and groundwater storage, as the water table has fallen dramatically (Lele et al., 2013). These changes have already resulted in social consequences. Some farmers have deserted their agricultural land (Lele and Sowmyashree, 2016) and those that remain must decide among drilling for deep groundwater, subsisting on rainfed agriculture, or letting their land go fallow. Bangalore, which now imports all of its water from the Cauvery river (Suresh, 2001), faces the challenge of supplying water to its rapidly growing population under tenuous circumstances with potential for a water crisis (Balasubramanian, 2016). There has been considerable uncertainty regarding the drying of water resources within the Arkavathy, and water management has proceeded largely in the absence of detailed understanding regarding the hydrologic behavior of the watershed (ISRO and IN-RIMT , 2000). The challenge of understanding hydrologic functioning is hampered by insuﬃcient hydrologic records due to poor environmental monitoring, a problem that is widespread in the region (Batchelor et al., 2003). There is an immediate need to understand why the watershed dried and inform watershed management, as well as a need for methodological advancement to conduct hydrologic studies in regions that lack adequate hydrologic records, both of which are the subject of this dissertation. The selection of the Arkavathy watershed as a study site was supported by the opportunity to collaborate with Ashoka Trust for Research in Ecology and the Environment (ATREE), a non-proﬁt research institution in Bangalore. ATREE implemented a 5-year research project to understand water availability, quality, and provisioning within the Arka- vathy watershed, considering the environmental and physical processes as well as water treatment technologies, social institutions, and governance (Lele et al., 2013). The ongoing local presence, expertise, and data collection within ATREE enabled the research in this dissertation.

1.2.1 Regional context The drying trend in the Arkavathy is, unfortunately, emblematic of the broader trends in water supply and management in south India. India’s population is expected to grow from 1.3 billion today to 1.6 billion in 2050 (United Nations, 2013), likely aggravating water scarcity challenges that much of the country already faces (Jury and Vaux, 2007). Unsustainable groundwater depletion is the norm in many of the country’s aquifers (Fishman et al., 2011; Reddy, 2005; Rodell et al., 2009; Tiwari et al., 2009) as agricultural practices have largely shifted from surface water irrigation to groundwater irrigation (Shah et al., 2003; Wada et al., 2012). Water demands on the country’s largest rivers have frequently resulted in water disputes between states, and most of the states in south India have recently been involved in water disputes of some sort (Briscoe and Malik, 2006; Richards and Singh, 2002; Thatheyus et al., 2013). The Cauvery river (to which the Arkavathy is a tributary) originates in the state of Karnataka and ﬂows into Tamil Nadu and has been a source of contention since the late 1800s (Anand, 2004). The dispute continues today with statewide protests frequently CHAPTER 1. INTRODUCTION 4

occurring in drought years (Gleick, 2011). While political battles continue, Karnataka is considering augmenting streamflow in the Cauvery by constructing infrastructure for an inter-basin water transfer from west-flowing rivers on the opposite side of the Western-Ghats to the Cauvery which flows east (Belgaumkar, 2013). This continues a trend of supply-side solutions to water challenges (Perrin et al., 2012) in which management efforts frequently fail to account for the anthropogenic drivers of water scarcity and their interaction with watershed hydrology processes (Bouwer et al., 2006). The historical importance of the Arkavathy for Bangalore’s water supply and its similarities to many other water systems in India makes it an attractive case study. Many of the features found in the Arkavathy are common throughout the region. For instance, the watershed contains a mix of rainfed and irrigated agriculture. For centuries, irrigated agriculture in south India was supported by a system of man-made lakes known as tanks, in which a long, earthen dam was constructed across the stream channel and shallow valley floor. These tanks systems have supported irrigated agriculture in south India since at least 200 A.D. (Ludden, 1979). Seasonal streams would fill the tanks during the monsoon season, and overflow downstream as part of a “cascading tank system” (Shah, 2003a). Until groundwater displaced surface water as the major source of irrigation in the 20th century, tanks were actively controlled to release water to farmers (Van Meter et al., 2014). Today, tanks are mostly unmanaged groundwater recharge basins, and tanks in the Arkavathy rarely overflow (Lele et al., 2013). Nevertheless, they capture and remove considerable amounts of water from the river network via evaporation and recharge (Molden and Sakthivadivel, 1999). In the second half of the 20th century, groundwater irrigation began supplanting tank irrigation and by the end of the century became widespread throughout India. Borewell technology was introduced in the 1960s and rapidly adopted. Approximately 60% of irrigated plots in India now utilize groundwater (Shah et al., 2003). Groundwater pumping has caused especially large reductions in aquifer storage in parts of northern India where the severity of groundwater depletion has raised alarm about potential water crises (Rodell et al., 2009). While the loss of groundwater in south India is not as severe, most irrigated agriculture in the Arkavathy watershed is groundwater sourced and considerable groundwater depletion has been observed (ISRO and IN-RIMT , 2000; Lele et al., 2013). As groundwater depletion became a concern in India, watershed development and managed aquifer recharge to reverse the depletion emerged as a common strategy and is still being pursued today. Many existing irrigation tanks were converted to groundwater recharge basins. Watershed development initiatives (GOI , 1994, 2002) promoted the construction of check dams, or small in-stream obstructions to flow, which are designed to capture a portion of streamflow and retain it for aquifer recharge (Gale et al., 2006). Another adaptive response to groundwater depletion favored by many individual farmers involved the conversion of traditional cropland to Eucalyptus plantations, which require little maintenance and no irrigation. Eucalyptus plantations were introduced to north India in the 1960s and have been widely adopted throughout India since (Puri and Nair, 2004), including in the Arkavathy. Eucalyptus plantations do not require irrigation, but have the potential to reduce subsurface water storage (and therefore streamflow) through high evapotranspiration CHAPTER 1. INTRODUCTION 5

rates (Calder et al., 1997). Research characterizing anthropogenic change in Indian water systems has typically focused on either local-scale management practices (Gale et al., 2006) or large-scale climatic drivers and total basin streamﬂow (Gosain et al., 2006a). More research is needed to connect small-scale processes (e.g., runoﬀ generation or check dam construction) with their large- scale consequences (e.g., regional surface water depletion). Because of the common features described above, lessons learned from the Arkavathy case study should be informative for regional water management.

1.2.2 Potential drivers of change One of the immediate challenges faced by water managers is the lack of an evidence basis that can be used to interpret the observed changes in flow in the Arkavathy river. Like many other watersheds in south India, social and environmental changes in the Arkavathy watershed have outpaced scientific understanding of the physical behavior of the system. A lack of consensus regarding the cause of the drying has led to uncoordinated and potentially confounding management actions by different water agencies (e.g., see Chapter 2, Section 2.1). To consider the potential drivers of the drying of the river, a surface water balance expression can be useful, with the control volume encompassing the spatial extent of the watershed and extending vertically from the bottom of the vadose zone to the top of the canopy: dS = P + A − ET − L − (Q + Q ) (1.1) dt b f In this equation, S is the volume of water stored within the catchment, P is precipitation, A represents anthropogenic imports, either via urban water imports from the Cauvery or agricultural water imports (including pumping from deep aquifers outside the surface water control volume) and exports, largely for water supply to Bangalore (which decreased over time as the city has increasingly relied on Cauvery water supply), ET is evapotranspiration and L leakage from the surface water catchment to deep aquifers. The term Q is streamflow, partitioned into a quickflow component Qf and a baseflow component Qb. Quickflow is often assumed to be generated by surface processes (e.g. infiltration excess runoff), and baseflow is assumed to be generated by soil or groundwater stores feeding the channel. Previous work has suggested that water managers, residents in the watershed, and other stakeholders collectively hold several hypotheses for the drying of the river (Lele et al., 2013), attributing the hydrological change to (H1) changes in precipitation, (H2) global warming, (H3) groundwater pumping, (H4) changes in land use, and (H5) obstructions to flow. Each of these hypotheses is accompanied by a plausible explanation. The precipitation hypothesis (H1) suggests two potential mechanisms for reducing streamflow: (a) reduced precipitation volumes P could simply reduce the total volume of water available for flow annually (Q), or (b) reduced rainfall intensities or increased inter-storm arrival times could increase infiltration (along with either L or ET ) at the expense of surface CHAPTER 1. INTRODUCTION 6

runoff (Qf ). Multiple studies about the susceptibility of south India to water stress indicate the magnitude of challenges that India faces from a water scarcity perspective (O’Brien et al., 2004; Palmer et al., 2008). However, climate models do not predict major changes in precipitation in south India in coming decades (Gosain et al., 2011; Tripathi et al., 2006). Climate change (H2) has been observed in this region as an increasing trend in average annual temperature, which could promote an increase in potential (and thus actual) evapotranspiration (ET ). Increasing temperature could therefore decrease soil water storage and reduce baseflow (Qb). A reduction in the height of the shallow water table due to groundwater extraction (H3) could reduce baseflow generation by increasing the leakage term. Although the interactions between a stream and groundwater are complex and dependent on local hydrogeology (Brunke and Gonser, 1997; Sophocleous, 2002), repeated studies demonstrate that groundwater depletion can decrease streamflow (Hughes et al., 2012; Jenkins, 1968; Kinal and Stoneman, 2012; Theis, 1940). The hypothesis regarding changes in land use (H4) relates to the increase in Eucalyptus plantations, which have the capacity to remove considerable amounts of water from the soil as ET , reducing catchment storage and baseflow (Qb). Some Eucalyptus plantations in south India were found to transpire at rates close to potential evapotranspiration (Kallarackal and Somen, 1997), tapping into subsurface water storage and transpiring more water than the total annual precipitation in some cases (Calder et al., 1997). Numerous studies indicate that Eucalyptus afforestation can reduce streamflow (Bosch and Hewlett, 1982; Langford, 1976). The hypothesis regarding obstructions to flow (H5) suggests that overland flow is impounded either on the hillslope or within the stream by check dams or other obstructions. The water then evaporates or infiltrates and drains to the water table instead of flowing downstream, leading to short term increases in S at the expense of total flow Q, and long term increases in leakage (L) or evapotranspiration (ET ) at the expense of streamflow (Q). Check dams are useful for increasing groundwater (Gale et al., 2006), but potentially at the expense of surface water runoff (Sakthivadivel, 2007). The use of tanks and check dams to recharge groundwater remains an active topic of research (Glendenning et al., 2012; Haddad, 2015).

1.2.3 Climate and geography The Arkavathy watershed is adjacent to Bangalore, India (13.0◦N, 77.6◦E, population: ˜10 million, Figure 1.1) with a semi-arid climate and average annual precipitation of 820 mm. Precipitation falls mostly during the southeast monsoon (June-September) and northeast monsoon (October-December) seasons. Situated at the southern end of the Deccan Plateau, most of the watershed is a pediplain in which mountains have been eroded to the extent that the landscape is mostly ﬂat with a deep weathered soil layer below (ISRO and IN- RIMT , 2000). Remnants of old mountains are visible as denudational hills jutting out of the ﬂat landscape. In most cases gneiss or granite bedrock lies below the soil layer. Soils are CHAPTER 1. INTRODUCTION 7

10 km05

1 India

Bangalore Rivers Tanks Reservoirs: 1 Hesaraghatta 2 TG Halli

Figure 1.1: Location map. (left, top) India. (left, bottom) Karnataka. (right) Arkavathy watershed, including the two historical drinking water reservoirs, TG Halli and Hesaraghatta (both reservoirs are now dry). The TG Halli watershed, where most of the research was conducted, comprises the upper part of the Arkavathy, north of the TG Halli reservoir. generally loamy with varying clay content. In some areas, an argillic (clay-rich) horizon lies about a half meter below the soil surface (ISRO and IN-RIMT , 2000). Over eighty-five percent of the watershed is flat, comprised of gentle slopes (less than 3% grade). Less than three percent of the watershed is strongly sloped (greater than 10% grade) (ISRO and IN-RIMT , 2000). The river network is punctuated by hundreds of tanks, and first- and second-order streams are further fragmented by check dams. Land use in the watershed is mostly agricultural, with some forested and protected areas in the south and rapidly expanding urban areas around Bangalore. Smaller cities have also expanded over the last few decades, in particular those connected by highway to Bangalore. Outside the urban areas, the watershed is predominantly under agricultural cultivation with a mix of crops including rice, corn, grapes, potatoes, and vegetables. In recent decades some farm owners have switched to Eucalyptus plantations. Farmland is a mix of rainfed and groundwater- irrigated agriculture, and in recent decades farmers have increased double and triple cropping CHAPTER 1. INTRODUCTION 8 by increasing irrigation. The aquifer water table is now hundreds of meters below the land surface, but anecdotal observations of the water table prior to the expansion of groundwater irrigation suggest that the water table used to be very close to the land surface (Lele et al., 2013).

1.3 Research approach

1.3.1 Conceptual framework The Arkavathy can be conceptualized as a social-ecological system (SES, Figure 1.2, top, after Bennett et al., 2009) in which humans modify the environment to benefit from ecosystem services (e.g., crop production). The modified system produces new environmental outcomes that benefit society, mediated by the physical processes of the hydrologic system. Although such a system can be managed sustainably (Ostrom, 2009), in the absence of constraints on the system, land and water management intensify to maximize ecosystem services (Folke, 2006; Folke et al., 2010). When the system is most integrated and providing the most benefits, it is also in a state of low resilience (Holling, 2001) and eventually the environmental system likely becomes strained and produces undesirable outcomes and social consequences. Social adaptation yields new human decisions and updated environmental management (Folke et al., 2005; Walker et al., 2004). Although the evolution of the Arkavathy system is ongoing, considerable hydrologic change and social consequences occurred prior to this study and in the absence of rigor- ous hydrologic observation. The lack of empirical observations over the period of change calls for an innovative approach to understand this system. The research framework shown in Figure 1.2 (bottom, in blue) describes an approach for investigating hydrologic dynamics and change in poorly-monitored sociohydrologic systems. Although the four stages are not mutually exclusive and can largely be studied in parallel, they represent distinct features of the research plan and each stage of analysis corresponds to a different chapter in this dissertation. Problems in social-ecological systems are likely to be noticed after social consequences have occurred and from a research perspective, the identification of the problem is likely to occur during an adaption phase, especially in poorly monitored systems. Initial steps include describing its relevant history and defining research objectives, and likely involve some engagement with stakeholders and managers of the system (Lele et al., 2013). The multiple hypotheses stage of analysis involves the development and initial evaluation of multiple hypotheses offering potential explanations of pertinent dynamics of the system. The method of multiple working hypotheses was initially developed by Chamberlin (1965b) as a way to circumvent researcher confirmation bias and promotes thorough investigation in developing theories about a given system. In this framework, all plausible hypotheses are considered and tested independently, allowing for the possibility that more than one hypothesis could provide a valid explanation of the observed system dynamics. In terms CHAPTER 1. INTRODUCTION 9

of investigating social-ecological systems, stakeholders are likely to be helpful in identifying plausible hypotheses and evaluation of stakeholder hypotheses may be essential for ongoing credibility of researchers and engagement with stakeholders (Lele et al., 2013). This stage of the analysis focuses on an initial evaluation of the hypotheses using readily accessible data with the objective of ruling out hypotheses which are contradicted or unsupported by evidence. The development and initial evaluation of hypotheses serves as a way to constrain the scope of subsequent research. The refinement stage of analysis seeks to further constrain and clarify the social-ecological system in question, ideally by further rejecting or supporting the remaining plausible hypotheses. To advance beyond the initial hypothesis testing, this stage will likely require generating data sets or combining historical data sets in new ways, for example through remote sensing, surveys, observations, or other documentary sources. This refinement stage is intended to help identify which data that will be needed to understand the processes controlling system dynamics, as well as define boundaries or identify spatial or temporal features of the system in question. The above analyses help constrain and clarify the problem but are unlikely to clearly attribute the drivers of change. An understanding of the physical processes that relate environmental modification to environmental outcomes is essential to causally attribute observed dynamics to appropriate drivers. One way to produce such an understanding of historical

Human Social benefits, decisions consequences

Environmental Environmental modification outcomes

Multiple hypotheses Refinement Physical processes System modeling

Research approach Figure 1.2: A social-ecological system (top, in green) and methodological approach to understand system behavior (bottom, in blue), including four stages of analysis. CHAPTER 1. INTRODUCTION 10 processes is to conduct a detailed investigation of contemporary processes, followed by a search for additional pieces of evidence that allow for reconstruction of historical hydrologic processes, such as sedimentary evidence (Benito et al., 2015), chemical profiles (Allison, 1988), or other auxiliary sources (Brázdilet al., 2006). The focus of the system modeling stage is potentially twofold, including evaluation of hypotheses that was not feasible in earlier stages, as well as exploration and prediction of future scenarios. The previous stages of analysis are important for correctly specifying model dynamics and model parameters, while protecting against equifinality (Beven, 2002; Savenije, 2001) and model structural errors (Beven, 2006). System modeling can take a variety of forms designed to match the research or management question of interest, including prediction of future behavior, exploration of alternative scenarios under different drivers, or simulation of long-term evolutionary dynamics. This research framework was applied in studying hydrologic change in the Arkavathy watershed, with the four stages of analysis corresponding to Chapters 2–5, which seek to (i) identify and evaluate multiple hypotheses on an initial basis (Chapter 2); (ii) constrain and clarify the problem using a remote sensing analysis of tanks (Chapter 3); (iii) investigate the physical processes controlling hydrologic system behavior (Chapter 4); and (v) simulate relevant processes under multiple scenarios of check dam construction (Chapter 5). The methods incorporate best practices from PUB (Predictions in Ungauged Basins) for investigating ungauged basins (Blöschlet al., 2013) and build on emerging ideas from the sociohydrology literature (Sivapalan et al., 2012). The concept of hydrologic reconstruction (Thompson et al., 2013b) is developed in considerable detail throughout the dissertation, and especially in Chapter 4. Chapter 2 has been published in Hydrology and Earth System Sciences (HESS, Srinivasan et al., 2015), and at the time of writing Chapters 3 and 4 are under review in HESS (Penny et al., 2016) and Water Resources Research (Penny et al., 2017), respectively.

1.3.2 Summary of chapters Chapter 2 overview: Multiple hypotheses of change In this chapter, five hypotheses are developed as plausible explanations for the observed decline in streamflow in the TG Halli watershed, the northern portion of the Arkavathy (Figure 1.1). The following hypotheses are then tested using data from field surveys and reliable secondary sources: (H1) changes in rainfall amount, timing and storm intensity, (H2) rising temperatures, (H3) increased groundwater extraction, (H4) expansion of eucalyptus plantations, and (H5) increased fragmentation of the river channel. The results indicate that anthropogenic drivers of change such as groundwater pumping, expansion of eucalyptus plantations, and to a lesser extent channel fragmentation (H3, H4, H5), are much more likely to have caused the decline in surface flows in the TG Halli catchment than changing climate (H1, H2). The analyses show that direct human interventions play a significant role in altering the hydrology of watersheds. The multiple working CHAPTER 1. INTRODUCTION 11 hypotheses approach presents a systematic way to quantify the relative contributions of anthropogenic drivers to hydrologic change.

Chapter 3 overview: Spatial changes in surface water In this chapter, we use a remote sensing approach to understand hydrologic change by characterizing hydrologic trends in tank water extent (surface area) and comparing these trends with historical land use maps to assess human drivers of change. We classify water extent in nearly 1700 tanks in Landsat images over the course of the study period, 1973–2010. We group tanks into clusters and model water extent of the tank clusters in a multiple regression on simple hydrological covariates (including precipitation) and time. Interannual variability in precipitation accounts for most of the predicted variability in water extent, yet precipitation does not exhibit statistically significant temporal trends in any part of the watershed. After controlling for precipitation variability, we find statistically significant trends in water extent, both positive and negative, in 13 of the clusters. These trends likely reflect a changing hydrologic regime, indicated by a non-stationary relationship between precipitation that occurs in the watershed and the runoff that is generated and fills tanks. Tank water extent increased in a region downstream of Bangalore, likely due to increased urban effluents, and tank water extent declined in the northern portion of the Arkavathy. Comparison of these drying trends with land use indicate that they were most strongly associated with irrigated agriculture, suggesting that groundwater pumping for irrigation could be an important driver of hydrological change in this watershed. This analysis demonstrates that most of the drying in the Arkavathy is located within the TG Halli watershed in the northern portion of the watershed. Furthermore, it suggests a link between groundwater pumping for irrigation and the observed streamflow declines.

Chapter 4 overview: Process-based reconstruction In this chapter, we develop a typology of hydrologic reconstruction consisting of data reconstruction; phenomenological reconstruction; theory-based reconstruction; and process-based reconstruction. The latter seeks to understand the historical processes controlling nonstationary system dynamics. We use a process-based reconstruction to understand current and past mechanisms of streamflow generation in the TG Halli watershed. We employ a range of field instrumentation and experiments to identify contemporary streamflow generation mechanisms in the TG Halli watershed, which we use to constrain our understanding of historical hydrologic processes. Additional sources of evidence indicate the presence and subsequent loss of an historical shallow groundwater table that discharged to the stream in the past. While ongoing land and water management decisions complicate the trajectory of water resources in the TG Halli watershed, multiple sources of independent evidence indicate that groundwater pumping is the most likely driver of streamflow decline in the TG Halli watershed, consistent with the findings in Chapter 3. Furthermore, the findings regarding CHAPTER 1. INTRODUCTION 12 hydrologic processes can be used as a foundation for predictions and exploration alternative management scenarios.

Chapter 5 overview: Check dams and groundwater recharge In this chapter, we explore the effect of check dams on streamflow and groundwater recharge. The policy response to groundwater depletion in India has largely attempted to reduce or reverse the depletion through a watershed development program in which groundwater recharge is encouraged along the river network in the form of check dams. We investigate recharge by check dams within the study catchments in the TG Halli watershed by monitoring dynamics of a single check dam and considering check dam chains and watershed dynamics via hydrologic modeling. We then explore the potential of check dams to facilitate groundwater recharge in the study catchment and in idealized catchments in which we control for multiple configurations of check dam construction. Check dam construction has a positive influence on groundwater recharge and has the potential to increase water availability in upstream areas, potentially increasing the equity of water access among users. However, based on the characteristics of installed check dams in the study watershed, the net effect of check dams is unlikely to dramatically shift the water balance and restore groundwater to historical levels. 13

Chapter 2 A multiple hypothesis approach in a data-scarce region1

2.1 Introduction

Freshwater has been identified as one of the gravest challenges of the twenty-first century (Srinivasan et al., 2012a; Vörösmartyet al., 2010; Wagener et al., 2010). Human demands for water have increased while annual freshwater available globally has remained more or less constant through history. To make sound policy choices, water managers need to know how water availability is changing. They must reconcile the ability to meet the needs of their populations and economies with the potential impacts on the well-being of downstream users, ecosystems and/or future generations. But predicting water availability is particularly challenging in rapidly growing regions, which are undergoing population growth, agricultural intensification and industrialization. Human modifications of land and waterscapes are changing the dynamics of the water cycle at unprecedented rates. Many of these regions also lack the long term hydrologic monitoring records needed to make such analyses possible. As a result, water managers lack the scientific basis to articulate trade-offs. This often leads to policies that address only part of the problem at best, or, at worst have negative or paradoxical outcomes (Sivapalan et al., 2014).

2.1.1 Challenges in rapidly growing, data-scarce regions Making hydrological predictions is a non-trivial problem in any context, but it is confounded by three issues encountered in rapidly changing, data-scarce regions: (i) non-stationarity arising from anthropogenic drivers, (ii) the sparse availability of historical data, and (iii) lack of original, place-based scientiﬁc research leading to oversimpliﬁed assumptions. The prediction challenges arise both from the nature of the system (point i) and researcher

1This chapter was published in 2015 with the title “Why is the Arkavathy River drying? A multiple- hypothesis approach in a data-scarce region” in Hydrology and Earth System Sciences (HESS). CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 14 constraints (points ii and iii), but the net result is that water managers are forced to rely on conceptual models that poorly represent the underlying system. Multiple drivers of change: Traditional water resources management is based on the assumption of stationarity — the idea that natural systems fluctuate within an unchanging envelope of variability (Milly et al., 2008). However, the impact of humans on the water cycle, either directly through modification of landscapes and waterscapes, or indirectly via climate change, has been identified as a defining challenge for hydrology (Thompson et al., 2013a). The potential impacts of climate change on the hydrologic cycle have received enormous attention from researchers and decision makers in recent years (Huntington, 2006; Stocker and Raible, 2005). The role of other direct human interventions like groundwater extraction, small dams and urbanization is also coming under increased scrutiny (Arrigoni et al., 2010; Cai and Zeng, 2013; Hu et al., 2015; Wang and Hejazi, 2011; Zeng and Cai, 2014). Data sparseness: The task of determining cause and effect relationships with respect to hydrologic behaviour is complicated in regions with sparse or recent hydrologic records (Maeda and Torres, 2012). Despite long-standing global calls to improve data sharing and transparency (Arzberger et al., 2004; Bonell et al., 2006; Dunne and Leopold, 1978; Reich- man et al., 2011; Sivapalan et al., 2003), data scarcity remains an impediment to research in many parts of the world. Data scarcity has motivated concerted research efforts such as the Predictions in Ungauged Basins (PUB) effort of IAHS (Sivapalan, 2003; Sivapalan et al., 2003; Wagener et al., 2004). However, these efforts are generally not suitable for predictions in non-stationary, human-impacted basins (Sivapalan et al., 2012; Srinivasan et al., 2012b). In such cases, lack of data confounds both conceptual understanding and building of quantitative models that explain “how the water system works”. Over-simplifying assumptions: Investments in the water sector must be made even in the absence of long-term records. In the absence of reliable data, modellers are then often forced to make many simplifying assumptions. The choices seem too often to be dictated by what can be modelled rather than what matters, leading to so-called modeller myopia (Buytaert, 2015). For instance, Gosain et al. (2006b) predict water availability in space and time in several Indian river basins under climate change, but do not incorporate man-made structures like dams or diversions into their basic model or trend analyses. Mujumdar and Ghosh (2008) modelled flows in the Mahanadi river of eastern India; their model assumed that recent declines in streamflow reflect a “climate signal”, without considering the possible influence of more proximate factors like groundwater pumping. Similarly, numerous water resources modelling projects in India decouple the effects of groundwater depletion from surface water responses, even where groundwater overexploitation is known to be a problem (Garg et al., 2013; Gosain et al., 2011; Kelkar et al., 2008).

2.1.2 Use-inspired science in data-scarce regions The mismatch between the needs of water managers and what off-the-shelf models can generate is not a sufficient reason for inaction or ad-hoc decision making in regions with rapidly increasing water demand. There is an urgent need to formulate new approaches to CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 15 frame and conduct hydrologic investigations in human dominated, data scarce situations. The conventional response would be to initiate primary data collection, and build new site- specific models from scratch. However, hydrologic data collection is expensive and takes many years. In contrast, information is often needed quickly and projects are limited by time and resource constraints. How should hydrologists proceed in these circumstances? First, as Thompson et al. (2013a) suggest, hydrologists should adopt a use-inspired science approach by pursuing scientific understanding while also addressing policy and management goals. This requires identifying the most pressing societal problems and working backwards from them. Second, as Buytaert et al. (2014) suggest, knowledge may be dispersed amongst multiple parties. While researchers and managers may hold some expert knowledge, citizens who have lived through change in the basin may also have useful insights. Third, it should be possible to use this knowledge to identify working hypotheses (Chamberlin, 1965b) that might explain the hydrological phenomenon of interest, and then use the sparse data to accept or reject at least some of them. This approach would then guide the choice of future data collection and sophisticated modelling efforts, targeting the most critical knowledge gaps. We use the above approach to narrow down possible causal mechanisms of hydrologic change in the Arkavathy watershed in South India. Five possible hypotheses that link anthropogenic and climatic changes to the water scarcity in the watershed are outlined and investigated.

2.2 Drying of TG Halli reservoir

2.2.1 Study area and the problem The Arkavathy river is located in Karnataka State in southern India (Figure 2.1). The rivers catchment overlaps with the western portion of the rapidly growing metropolis of Bengaluru (Bangalore). The region is seasonally monsoonal, receiving approximately 830 mm of precipitation annually. The main stem of the Arkavathy river has its headwaters in the Nandi Hills north of Bengaluru and is joined by its ﬁrst major tributary, the Kumudvathy river at Thippagondanahalli (TG Halli) village, where a reservoir was constructed in 1935 to supply water to Bengaluru. This reservoir has a catchment area of approximately 1447 km2. The TG Halli reservoir catchment also contains an older water supply reservoir at Hesarghatta, as well as an estimated 617 small surface storage structures called “tanks”. Tanks are traditional in-stream water harvesting systems that were commonly built in South India and Sri Lanka over the last six centuries to store monsoon runoﬀ for post-monsoon irrigation (Shah, 2003b; Vaidyanathan et al., 2001). The cumulative storage of all these tanks (297 of which are more than 50 Ha in size) and Hesarghatta reservoir is estimated to be 143 Million cubic metres (MCM), i.e., about one and a half times the storage capacity of TG Halli reservoir (ISRO and IN-RIMT , 2000). Most of the TG Halli catchment is underlain by gneissic and granitic aquifers. Highly CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 16

Figure 2.1: The TG Halli catchment with major features. BBMP is the Greater Bangalore Munic- ipal Corporation Boundary. (Data Source: Survey of India toposheets at 1:50,000 scale; ASTER DEM imagery, maps prepared at the ATREE EcoInformatics Lab.) weathered soils extend to about 20m below grade level (BGL), and form a shallow aquifer in which seasonal perched water tables can develop. Between about 20-60 m BGL lies a fractured rock zone with considerable jointing and cracking, acting as a deeper aquifer. Groundwater yields decline beyond 60 m BGL, although fractures continue to be encountered down to 300m. From 1937 up to the 1980s, the TG Halli reservoir was a major source of water for Bengaluru. However, inflow to the reservoir has steadily declined since the early 1980s (Figure 2.2a), and today it supplies only 0-25% of its design capacity. Average inflows into the TG Halli Reservoir have decreased from 385 MLD (ML day-1 or 140,000 ML yr−1) pre- 1975 to about 65 MLD post-2000 (24,000 ML yr−1), a decline of 320 MLD. The cascading irrigation tanks dotting the catchment are also mostly dry, indicating that the loss of surface runoff has occurred throughout the catchment (Lele et al., 2013). CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 17

5 x 10 8 A Median ow Median ow Median ow 6 1938-1975: 1975-2000: 2000-2010: 4 108,000 ML/yr 49,000 ML/yr 25,000 ML/yr

Inflow (ML) 2 0 2000 B

1500

1000

Annual P (mm) 500

0 C 1800 1700 1600 1500 Annual PET (mm) 1400 Months baseflow 0.4 D 4

0.3 3

0.2 2

0.1 1

Baseflow Index 0 0 1940 1950 1960 1970 1980 1990 2000 2010 Years Figure 2.2: Changes in hydrology and hydrometeorology of the Arkavathy basin, 1970–2010. (a) Annual inflows into the TG Halli reservoir. The 1938-1975, 1975-2000, and 2000-2010 median and mean annual inflows illustrate the decline in inflow that has occurred in recent decades. (b) Area-averaged annual rainfall over the 7 Taluk (local government areas) comprising the TG Halli catchment. (c) Potential evapotranspiration as estimated from the Hargreaves equation for the TG Halli catchment. (d) Two estimates of baseflow contribution to the TG Halli inflows: the number of months per year when 100% of flow was derived from baseflow (bars) and the baseflow index computed from daily inflow data (dots)

The drying of ﬂows into the TG Halli Reservoir and tanks in the catchment has clear implications for the 800,000 people that live in the catchment, both in terms of current water availability and because the declining ﬂows may be an indicator of the overall unsustainability of water use in the basin jeopardizing future populations and economic growth. CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 18

Neelamangala Taluk Doddaballapur Taluk Well Census 100% 40% BC1970−1990 2000-2010 A Surface Water B C 1970−1990 1990-20001990−2000 2010-2014 Open Well 2000−2010 80% 2010−2014 Bore Well 30%

60%

20% 40% Frequency % Irrigation Source % 10% 20%

0 0 1975 1980 1985 1990 1995 2000 2005 2010 1975 1980 1985 1990 1995 2000 2005 2010 0 50 100 150 200 250 300 350 400 Years Depth to water (m) Figure 2.3: Water use in the TG Halli Catchment. (a,b) Sources of irrigation water used over time in two Taluks (Local Government areas), indicating the reduction in surface and shallow well use and their replacement by deep bore wells over time. Data Source: Department of Economics and Statistics, Karnataka. (c) Depths at which water was encountered at time of drilling. Data from well census conducted by ATREE ﬁeld hydrology team in summer of 2014 covering 482 boreholes in a 26 sq km area in TG Halli catchment.

2.2.2 The debate about causes and solutions Given the urgency of the problem, several uncoordinated and often contradictory actions have been undertaken. One reason for this is that the causes of the inflow reductions to the TG Halli reservoir remain unclear. In order to formulate hypotheses that could be investigated systematically, we consulted a range of sources to understand the positions and perceptions of different groups: one-on-one meetings with government officials, written policy documents and reports, a comprehensive literature review (Lele et al., 2013) and an expert consultation meeting held at ATREE, Bangalore in November, 2012. Additionally, at the launch of the research project in early 2013, a meeting was convened by the Chief Secretary of the state which included the research team and the heads of all government agencies engaged in water issues. The research team also made several reconnaissance visits, attended over a dozen stakeholder meetings hosted by other groups, and held more than sixty “Water Literacy Meetings” in the TG Halli catchment villages in 2014 and 2015, which were collectively attended by over 500 farmers. Finally, the research team conducted over two dozen focus group discussions in 2013 and 2014, targeting specific stakeholder groups. This initial review identified several policy positions that reflect different perceptions on the drying of the river:

• The Bangalore Water Supply and Sewerage Board (BWSSB), which owns and operates the TG Halli reservoir, commissioned a study (ISRO and IN-RIMT , 2000). This study identiﬁed several possible causes for the decline of inﬂows into the TG Halli — declines in rainfall, groundwater pumping, and obstructions in streams. However, the study did not quantify the relative magnitudes of these factors and did not recommend actions CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 19

to directly address them. BWSSB has taken no speciﬁc actions of its own to address the problem.

• The Cauvery Neeravari Nigam Ltd. (CNNL) was made responsible by the state government for “rejuvenating” the Arkavathy river. CNNL commissioned its own study (CNNL, 2010), which concluded that the primary reason for reduced inﬂow into TG Halli is obstructions in the channels. Accordingly, the agency response has been to bulldoze the obstructions and desilt the channels.

• Meanwhile, local rural development programmes have focused on constructing check dams to recharge the shallow aquifer and ostensibly restore baseﬂow in the stream.

• A number of citizen’s groups and social movements have emerged with the objective of “Rejuvenating the River” (Revive Kumudavathi, 2017) or “Saving Bangalore’s Lakes” (Arkavathi Rejuvenation, 2017). These groups are focused on removal of Eucalyptus trees, desilting lake beds, and diverting treated wastewater into lakes to recharge the shallow aquifer.

• The state Water Resources Development Organization (WRDO) has argued that climate change, via declining rainfall and rising temperatures, is responsible for the drying of the river. This perception was also held by most farmers we interacted with during the Water Literacy Meetings, many of whom favour inter-basin imports from west ﬂowing rivers.

2.2.3 The multiple hypotheses approach By examining the different explanations of the causes of streamflow decline and plausible runoff generation mechanisms, we identified and investigated all plausible hypotheses that could explain the observed changes in the Arkavathy basin: Hypothesis 1: Changes in rainfall: Changes in rainfall as the primary driver of streamflow could induce changes in surface runoff generation. The climate change literature for this part of Karnataka mentions a possible shift in the monsoon, such that the southwest monsoon June-July-August-September (JJAS) season rainfall would probably decline, and post-monsoon October-November-December (OND) rainfall could increase. A change in the seasonality of precipitation could result in a change in rainfall partitioning to runoff, because a greater fraction is partitioned to evaporation and transpiration. Additionally, if both seasonal and annual rainfall patterns are unchanged, a reduction in the mean storm intensity or depth could result in a failure to trigger infiltration-excess or saturation-excess runoff. The previous study commissioned by BWSSB found that rainfall in excess of 20mm day-1 is needed to generate significant inflows into TG Halli reservoir (ISRO and IN-RIMT , 2000). Hypothesis 2: Increasing potential evapotranspiration due to climate change: Increases in potential evapotranspiration could result in an increase in actual evapotranspi- CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 20 ration, reducing recharge and therefore baseflow. Of the major drivers of potential evapotranspiration (temperature, humidity, solar radiation, and wind speed), it is known that temperature has been increasing in south India over the past century, with most stations reporting temperature increases on the order of 1◦Cper100 years (Arora et al., 2005; Hingane et al., 1985). On the other hand, solar radiation trends in the region have been negative, in association with the formation of Atmospheric Brown Clouds (Ramanathan et al., 2005) and wind speed trends in India also appear to also be declining (McVicar et al., 2012). Empir- ical evidence from evaporation gauges throughout India also suggests that pan evaporation has declined over the twentieth century (Chattopadhyay and Hulme, 1997). Nonetheless, temperature-driven increases in evaporative demand could have altered rainfall-runoff partitioning in the catchment and contributed to the reduced streamflow. Hypothesis 3: Declining baseflow due to groundwater overexploitation Pre- vious studies in well-monitored basins have shown that groundwater depletion can reduce baseflow contributions to streamflow, reducing overall flows (Cai and Zeng, 2013; Zeng and Cai, 2014). Our hypothesis is that reduction in groundwater storage induced by pumping lowers the seasonal water table, resulting in the water table intersecting the river channel less frequently and for shorter periods of time, ultimately reducing the baseflow contribution to the Arkavathy. The decline in the Arkavathy river flow has occurred concurrently with an expansion of groundwater extraction in the basin and across Karnataka. Although groundwater monitoring in the region is minimal, irrigation data clearly show a shift from surface to groundwater and and open wells to deep borewells (Figure 2.3a,b) (DES, 2012). Hypothesis 4: Increasing actual evapotranspiration due to expansion of plantations Numerous studies indicate that where a catchment area is converted from rain-fed agriculture to deep-rooted perennial vegetation, it can result in decreases in flow (Brown et al., 2005). Eucalyptus cultivation was actively promoted among farmers by the state government under its farm forestry programme in the 1980s (Shiva et al., 1981). Field surveys within the TG Halli catchment indicate a significant increase in Eucalyptus plantation area in the past 40 years. Several studies have documented that Eucalyptus plantations create unsaturated conditions over a deep root zone, and can thus reduce subsurface contributions to streamflow (Calder et al., 1993; Farley et al., 2005). Hypothesis 5: Million puddle theory The final hypothesis is that the construction of (largely illegal) structures in the channel, along with construction of check dams and unculverted roads has resulted in the channels in the upper catchment becoming disconnected. In other words, a once-connected, flowing river has been replaced by a “million puddles”. A portion of the water in these puddles evaporates or is transpired by riparian vegetation and becomes unavailable. CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 21

Table 2.1: Details of various data sets used.

Parameter Type of data Source of data Precipitation Daily rainfall from four rain Indian Meteorological Depart- gauges (1934–2010) ment Temperature Monthly max, min. and mean Indian Meteorological Depart- temperature from two weather ment stations (1901–2010) Surface flow Daily Inflows (1975–2010) and Bangalore Water Supply and Monthly Inflows (1937–2010) Sewerage Board into TG Halli Reservoir Area under Eucalyptus Topographical sheet Survey of India plantations in 1973–1979 Area under Eucalyptus in Land use map Karnataka State Remote Sens- 2001 ing Applications Centre Groundwater levels Well Census of 472 wells in ATREE hydrology team field two milli-watershed covering 26 survey km2 Groundwater extraction Irrigated Area (1970–2012) KA State Annual Season Crop Report Irrigated Area (1981, 1991, Census of India, Village 2001) Amenities Dataset Channel obstructions Number of check dams and un- Primary Survey by Zoomin culverted roads Tech. Number of check dams in two ATREE hydrology team field milli-watershed covering 26 sq survey km

2.3 Methods

2.3.1 Data sources and quality assurance To test these hypotheses, we collected available secondary data within and around the Arka- vathy basin. Data were quality-checked and triangulated against other sources and supplemented with field surveys when needed (Table 2.1). Monthly inflow data for the TG Halli reservoir were obtained from Bangalore Water Sup- ply and Sewerage Board (BWSSB) for the period 1937 to 2010. Additionally, daily records of inflows from 1970 onwards were obtained from the local BWSSB offices and digitized. The daily and monthly data were cross-validated and any errors were corrected.

2.3.2 Analysis techniques The goal of the analysis was two-fold: (i) to determine whether the perceived changes in hydrological drivers have occurred, and (ii) whether the magnitude of changes in the drivers CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 22 could explain the magnitude of the change in ﬂow in the Arkavathy basin (i.e. consistent with the observed 320 MLD reductions in ﬂow).

Hypothesis 1: Declining Rainfall Data from four long-term rainfall gauges located in Devanahalli, Doddaballapura, Magadi and Nelamangala towns within the TG Halli catchment were available for analysis (see Figure 2.1 for gauge locations). These gauges provide daily rainfall data over 75 years (1934–2010). Although 18 additional rainfall gauges, operated by various government agencies, exist within the catchment, these gauges do not provide continuous data over a suﬃcient time period to allow trend analysis. As a quality control procedure, we performed double mass plots, compared the total number of rainy days between the gauges, and excluded years where the total number of rainy days represented a low outlier (indicating a likelihood of missing data). Outlier years were determined to be those where the number of recorded rain days was less than f25 − 1.5(f75 − f25), where f75 represents the 75th percentile and f25 the 25th percentile of the total number of rain days. Annual rainfall was computed over the water year (June to May). Seasonal rainfall totals were computed in terms of Pre-Monsoon (January-February-March-April-May: JFMAM), Monsoon (June-July-August-September: JJAS) and Post-Monsoon (October-November-December: OND) rainfall totals. To identify changes in rainfall depths at daily timescales, the number of days per year in which rainfall volumes exceeded 10 mm, 25 mm and 50 mm were determined for the 1934–2009 period. Trend detection was undertaken for each of the above datasets in two ways. First, we determined if a trend was present over the full time series. As the data generally did not conform to the assumptions for least-squares regression, we evaluated the trends using a non-parametric Mann-Kendall test. Second, we evaluated whether a change in the mean values of the meteorological parameters had occurred from the pre-1970 and post-1970 period, taking 1970 as a point after which the Arkavathy river ﬂows obviously declined. Where the data were normally distributed we made these comparisons with t-tests, otherwise non-parameteric Mann-Whitney-Wilcoxon tests were used.

Hypothesis 2: Increasing potential evaporation due to climate change In the absence of detailed meteorological data in the Arkavathy watershed, we estimated changes in the mean daily potential evaporation rate as a function of temperature using the modiﬁed 1985 Hargreaves evapotranspiration equation (Hargreaves and Samani, 1985):

0.5 PET = 0.0023 × Ra × (TC + 17.8) × TR , (2.1) −1 where Ra is the extraterrestrial solar radiation (mm day ), TC is the average daily temper- ◦ ature ( C) calculated as (TMax + TMin) /2. TR is the temperature range (TR = TMax −TMin, TMax is the maximum daily temperature and TMin is the minimum daily temperature). All results were averaged to the annual timescale. The resulting PET time-series was analyzed to determine the presence of trends or step-changes in the mean PET. Temperature data from 1901 to 2001 were obtained for one station each in Bangalore Urban and Bangalore CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 23

Rural districts from the Indian Meteorological Department. The data were checked to ensure that there were no missing data and that temperatures were within expected ranges. Ex- traterrestrial solar radiation was computed based on the weather station latitudes (Maurer, 2014) using the method of Spencer (1971), with an accuracy of 0.01% (Duffie and Beckman, 2013). The modified Hargreaves equation relies on the diurnal temperature range to provide a surrogate for solar radiation, and is widely used to estimate potential evaporation when only limited ground data (temperature) are available. The resulting PET estimates are typically within 10% or better of those derived from lysimeter or Penman-Monteith methods, when results are averaged over 5-day or greater time periods (Hargreaves and Allen, 2003). Limi- tations of the method lie in the fact that the relationship between diurnal temperature range and other drivers of potential evaporation (e.g. net radiation and vapour pressure deficit) may not be stationary over long time periods. In South India, such non-stationarity is likely to be associated with so-called “solar dimming” due to increased upper atmospheric pollution (Chattopadhyay and Hulme, 1997). We anticipate that errors due to non-stationarity are likely to lead to an over-estimation of potential evaporation via the Hargreaves equation.

Hypothesis 3: Declining baseflow due to groundwater overexploitation Long-term groundwater level data (> 10 years) existed for only two shallow wells within a 5 km buffer of the TG Halli catchment. These reported stable water levels of 10-30 m BGL. However, in the course of extensive field visits, no water was seen in any other open well in the region. We concluded that the two monitoring wells are not representative of surrounding conditions. There are also no deep borewell piezometers with long-term water level data in the catchment area. To infer potential changes in groundwater levels, we conducted a comprehensive census of borewells in a 26 km2 area in the TG Halli catchment in the summer of 2014. Data for a total of 472 borewells were recorded. For each borewell, the owner was interviewed to obtain details of the year of construction, use, status, depths of yielding fractures and year of failure (if applicable). Together, these data provide an understanding of how groundwater levels have changed in the last four decades. We undertook two different analyses to explore whether changes in groundwater were compatible with the observed changes in surface flow. In one analysis we used a baseflow recession technique to benchmark the changes in mobile subsurface water storage that would be needed to account for the decline in annual flows and then estimated how these changes might manifest as a decline in groundwater levels. If this change in storage greatly exceeds observed well declines in the catchment, then the hypothesis that lower groundwater levels have lead to streamflow reductions could be rejected. In a second analysis, we performed a baseflow separation on the daily runoff data from 1970 onwards to determine how the trends in total streamflow were reflected by changes in quick-flow and baseflow. Recession analysis: We follow Brutsaert and Nieber (1977) in positing a nonlinear relationship between storage (S, [ML]) and discharge (Q, [MLD]) of the form:

S = aQb. (2.2) CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 24

A mass balance during periods of flow recession (i.e. when rainfall P is negligible) would be given by: dS = −ET − Q, (2.3) dt where recharge to groundwater and inter-basin transfers are assumed negligible and ET represents evapotranspiration. If 2.2 and 2.3 are coupled and differentiated, then the following expression is obtained relating flow to its rate of change: dQ abQb−1 = −ET − Q, (2.4) dt Under the assumptions that ET is slow in comparison to flow, so that ET→ 0, 2.3 and 2.4 simplifies to dQ 1 = − Q2−b, (2.5) dt ab Taking the logarithms of the absolute values of this expression, one obtains:

dQ 1 log = log + (2 − b) log (Q) . (2.6) dt ab

That is, a plot of the logarithms of the rate of change of the discharge against the logarithms of the actual discharge at any point in time contains sufficient information (in the form of an intercept and slope) to estimate the parameters of the original storage-discharge expression. To do this, the lower envelope of the expression must be fitted in order to minimize the effects of neglecting evaporation and to focus the analysis on the groundwater response (Brutsaert and Nieber, 1977). No significant changes in the recession behaviour over time were identified from this analysis. This methodology was applied to the monthly flow data from the Arkavarthy at TG Halli, focusing on the seasonal recessions from 1937- 1970 (i.e. prior to the discernible reductions in river flow). There are two major limitations to using monthly data for this analysis. First, the estimation of the rate of change of the flow is coarse. Second, the contribution of rainfall to runoff events is unlikely to be negligible, even during the seasonal recession. However, because the daily flow data were only available for the post-1970 period, the monthly analysis provides the only opportunity to evaluate the storage-discharge relationship when the river was flowing ‘normally’. As outlined in the results, the calibrated model had an exponent b = 1.43, very close to the theoretical value of 1.5, offering some reassurance that the results are reasonable. Using the parameterized storage-discharge equation, we estimated the mobile storage averaged over the catchment area, needed to produce the mean of the peak monthly flows for all years prior to 1970. The resulting storage volume can be normalized by the mobile porosity of the aquifer sediments to generate an estimate of the drop in the surface water table depth required to explain the ‘missing’ flow volume after 1970. Baseflow trends: Using the daily data from 1970–2010, we undertook a baseflow separation using a digital filter (Nathan and McMahon, 1990) and computed the annual baseflow CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 25 and the baseflow index for each water year. Again, we analyzed trends in these indices using the methods described previously. Additionally, we analyzed baseflow trends in both the monthly data from 1937–2008, and in the daily data available for 1970 onward. For the monthly data, we defined a baseflow month as a month when there was streamflow. This definition implies that 100% of the flow in these months is from baseflow, a much higher standard than the baseflow index which indicates the proportion of baseflow that occurred through time.

Hypothesis 4: Increasing actual evapotranspiration due to expansion of plantations We calculated the change in Eucalyptus plantation area from 1973 to 2001 by comparing the mapped land uses in both years. We used two sources: a land use map provided by Karnataka State Remote Sensing Application Centre (KSRSAC) and Survey of India topographic sheets. The KSRSAC land use map was derived from Indian Remote Sensing (IRS) LISS-3 merged with PAN satellite imagery with an effective 6m resolution. The map reported the area under Eucalyptus plantations in 2001. For other years, no such maps were readily available. So we digitized 1:50,000 scale topographic maps prepared by the Survey of India during the 1970s (1973–1979), which show Eucalyptus plantations on public lands. We made three assumptions about water use by Eucalyptus plantations (which are typically unirrigated). First, the plantations could not themselves have led to groundwater mining (as has been claimed in other parts of Karnataka (Calder et al., 1993)), because shallow groundwater in the region had largely disappeared by time Eucalyptus plantations were promoted under the Social Forestry Program in the early 1980s. Second, we assumed that Eucalyptus transpires at a rate of 830 mm yr−1 (the annual average rainfall). In effect the trees were perfectly efficient in utilization of rainwater, given that potential evaporation of 1650 mm yr−1 greatly exceeds annual rainfall and that many plantations implement practices to limit surface runoff. Third, we assumed the plantations displaced rainfed coarse cereal crops such as maize or millet, which have a seasonal ET of about 290 mm yr−1 for a single crop and 540 mm yr−1 for a double crop (Allen et al., 1998).

Hypothesis 5: Million puddle theory Data on the number of channel obstructions in the TG Halli catchment were available in a report commissioned by Cauvery Neeravari Nigam Limited (CNNL) (CNNL, 2010). A total of 344 obstructive structures were recorded including roads, bridges and unculverted roads) of which 277 were small check dams (Table 2.2). The density of check dams estimated from the report is 0.2 km−2 of watershed. To validate the CNNL data, we conducted a comprehensive survey of all stream obstructions in two milli-watersheds covering a 26 km2 area within in TG Halli catchment. Over 40 check dams were found in the 26 km2 area, indicating a check-dam density of 1.35 km−2. Even after discounting 20% that were leaky or silted, it appears that the CNNL data are an underestimate of the number of check dams. We therefore assumed the higher density of 1.35 km−2 for our analysis. CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 26

Table 2.2: Number and type of stream encroachments in each section TG Halli catchment Source: Zoomin Tech Report to CNNL, 2011.

Type Hesarghatta Kumudvathy Arkavathy Check dam 70 65 142 Bridge 4 23 31 Road 0 2 7

The volumes of typical obstructions were estimated based on stream profiles made using a Dumpy-Level instrument on seven check-dams. Interpolation of the stream profiles allowed us to estimate the maximum storage volumes as ranging between 100 and 1500 m3 with an average of 325 m3. We multiplied this storage by the basin area and density of obstructing structures to obtain a peak storage volume for the whole basin. We then plotted the cumulative density function of the daily inflow events into the TG Halli for 15 years from 1976 to 1990 (the period before check-dams and unculverted roads were constructed for which we had daily inflow records). We took all flow events less than or equal to the peak storage volume and assumed that the entire flow would be impounded. For events that generated inflows greater than the peak storage, the volume impounded was capped by the peak storage the catchment; anything higher would have overflowed. The volumes impounded were summed to estimate the total loss downstream. This calculation is likely to overestimate the fraction of daily runoff that is impounded behind check dams and unculverted roads, since the structures are unlikely to be empty at the beginning of every rain event.

2.4 Results

Results are presented separately for each of the hypotheses, grouped into climatic and human drivers.

2.4.1 Lack of evidence of climatic drivers

Hypothesis 1: Declining Rainfall Annual rainfall trends: Figure 2.2b shows the area-averaged monthly and annual rainfall over the basin for the years 1934–2010. With an average of 830 mm yr−1 and standard deviation of 210 mm yr−1, the monthly rainfall time series does not show any trend, and no statistically signiﬁcant trend emerges in the annual rainfall. Similarly, no signiﬁcant changes are visible in the pre- and post-1970 in mean annual and monthly rainfall totals. The data do exhibit high decadal variability in rainfall, and it is clear that the 1970–1980 period was exceptionally wet. However, there is no evidence that total rainfall volumes have changed in the region. CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 27

Seasonal rainfall trends: With the exception of Devanahally, we did not identify any statistically significant shifts in the timing of the rainfall over the last 80 years. The observed trend in Devanahally was for an increase in JJAS rainfall, contrary to the predictions of climate models. Moreover, more rain in JJAS cannot explain the decline in flow production at other times of the year. Change in rainfall intensity: No statistically significant trends in daily rainfall volumes exceeding threshold values of 10 mm, 25 mm or 50 mm could be identified at the 95th percentile level at any of the four gauges. Although we cannot exclude the possibility of changes in sub-daily rainfall intensities, analysis of rainfall data in the TG Halli catchment area shows no meaningful historical trends in precipitation volumes, timing or storm characteristics. We find no evidence that rainfall-driven changes could be responsible for the change in flow in the TG Halli catchment.

Hypothesis 2: Increasing ET due to increase in temperature The rise in temperature of about 0.6–1 ◦C per 100 years was within the range predicted by other studies (Arora et al., 2005; Kothawale and Rupa Kumar, 2005). The estimated PET from the Hargreaves equation averaged to the annual scale is shown in Figure 2.2c. As indicated in the figure, there is no statistically significant trend in PET within the basin. We conclude that there is no evidence to support the hypothesis that increasing temperature is increasing potential evaporation and leading to a decline in streamflow.

2.4.2 Evidence of human drivers

Hypothesis 3: Declining baseflow due to groundwater overexploitation From the recession analysis, the fitted storage discharge relationship for the pre-1970 period was: S = 595Q0.57 (2.7) Where S and Q are given in units of ML per month, consistent with the monthly time- step. The slope of the lower envelope was 1.43, very close to the 1.5 slope predicted by the nonlinear Dupuit-Boussinesq theory and found by Brutsaert and Nieber (1977) in their original analysis. We estimated S for the mean of the peak monthly flows from the years prior to 1970 (65,000 ML) using Eq. 2.7, and normalized this total stored volume by the catchment area. This leads to a prediction that on average, mobile storage would need to decline by 0.24 m across the catchment to reduce the peak monthly flow rates to zero. We can then use porosity estimates of 20% for the unconfined sediments, and 1% for the fractured rock to estimate the order of magnitude of the groundwater declines that could effectively remove 0.24 m of mobile water from being in connection with the surface channels. This works out to a decline of approximately 1.25 m in the surficial aquifer, or a decline of approximately 25 m in the fractured rock aquifer. (Figure 2.3). As can be seen from Figure 2.2d, baseflow started declining after the early 1980s but after 1992, there was not a single month when there was baseflow into the reservoir. The CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 28

Figure 2.4: Change in Eucalyptus Area in Arkavathy basin between 1973 and 2001. baseflow index, which was computed from daily data and indicates the share of baseflow in the total annual inflow, also declined consistently from 1970–2010. The loss of below-ground storage observed in the Arkavarthy basin is of the correct order of magnitude to explain the contemporary absence of surface flow, and the hypothesis that loss of groundwater storage in the surface aquifer should be retained for further investigation.

Hypothesis 4: Increasing actual evapotranspiration due to expansion of plantations The area under Eucalyptus plantations in 1973, as indicated by Survey of India toposheets was only 11 km2, all of it within the boundaries of state Reserve Forests. By 2001, the area under Eucalyptus plantations had increased to 104 km2 (Figure 2.4). Conversion of 93 km2 of rainfed crops to Eucalyptus plantations would thus translate to a loss of runoff of 75–135 MLD by the year 2001. This figure is significant compared to the observed runoff decline of about 320 MLD, suggesting that expansion of Eucalyptus could be a significant contributor.

Hypothesis 5: Million Puddle Theory Based on the assumptions about check-dam and encroachment water storage, the total loss in runoﬀ at the basin scale attributable to channel encroachment is on the order of 18–54 CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 29

MLD. While there are substantial uncertainties associated with this number (for example, other sources of surface water storage such as within-farm impoundments or impoundments in housing plots are ignored leading to an underestimate of the total volume, while the assumption that all storages empty prior to each rainfall event undoubtedly represents an overestimate), the order of magnitudes unequivocally indicate that while the million puddle theory could have contributed to a fraction of the loss in runoﬀ, it cannot account for the entire loss of inﬂow into the TG Halli.

2.5 Discussion

Our analysis indicates that rainfall changes or temperature increases are cannot account for any significant fraction of the decline in inflows into TG Halli reservoir. The causes found to be plausible are groundwater extraction, expansion of Eucalyptus plantations and to some extent increased obstructions in the stream course. Importantly, many policy approaches currently under consideration do not reflect the major underlying causes of the drying of the Arkavathy river, and in some cases (check dam construction) are clearly counter-productive. In the future, climate change could play a critical role in exacerbating water stress, but climate stressors will only add to existing local stresses. Although the hypotheses have been framed as independent, the mechanisms undoubtedly interact with each other, so that their inter-relations should be considered in formulating a conceptual model of the catchment, and in attributing the effects of each mechanism in terms of the change in river flow. For example, check dams not only impound flow, but also locally elevate recharge. Check dams may thus facilitate high levels of groundwater extraction locally. Spatial heterogeneity in water table levels and Eucalyptus root zone access to saturated conditions may vary throughout the catchment, meaning that the assumption that Eucalyptus plantations do not contribute to groundwater mining and reduced baseflow should be relaxed in future studies. The analysis presented here is preliminary. Further work is needed to understand the hydrological processes in the catchment, including the contemporary and historical flow generation pathways and their changes. There are, however, suggestive clues of timing that suggest a potential working hypothesis for the flow generation mechanisms. Expansion of electricity and installation of wells began to increase in the late 1960s, although this period also coincided with a period of relatively high rainfall and streamflow in the 1970s. Flow declines began to emerge in the early 1980s, with baseflow indices and numbers of “baseflow months” plummeting in the early 1990s, approximately at the same time that open wells went dry and deeper borewells become more prevalent (Figure 2.3a,b). During this period, the baseflow index declined, suggesting that less and less of the streamflow entering the TG Halli Reservoir was associated with groundwater inputs. These trends are highly reminiscent of those projected by models of the Republican river basin (Zeng and Cai, 2014) as a function of increasing groundwater extraction, with reduced baseflow and an increasingly erratic quickflow response. CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 30

Inflows continued to decline after 1992, suggesting that additional mechanisms beyond the decline in baseflow must be considered. Possible additional mechanisms include the conversion of the Arkavathy river into a “losing” river, which provides a source of recharge to the local aquifers, the continued expansion of Eucalyptus plantations and increasing implementation of management techniques that prevent surface runoff from leaving farm fields, and increasing obstruction of the stream channels. Based on these observations, further research targeting runoff generation mechanisms, establishing the pathways for surface water–groundwater connections, evaluating the effect of land use on water balance and estimating groundwater extraction rates has now been initiated in the catchment (See www.atree.org/accuwa). Finally, from a policy perspective, the fuzzy perception of the causes of streamflow decline and the lack of coordination between agencies have resulted in contradictory policies. The range of policy responses observed reflect both different stakeholder interests and different explanations of the hydrologic causes of the declining river flow. For instance, even as CNNL is removing encroachments and blockages under the Rejuvenation of Arkavathy River Program, new check dams continue to be authorized the under the Mahatma Gandhi National Rural Employment Guarantee Scheme (MGNREGA). Interestingly, the case study illustrates that the reluctance to acknowledge human feedback is not limited to hydrologists. Even farmers living in the catchment often do not fully acknowledge their role in altering the hydrology. The actors involved have not made any substantive effort to scientifically validate and reconcile these views, resulting in significant wasted investment.

2.6 Conclusions

The TG Halli catchment case study shows that humans can play a significant role in altering the hydrology of watersheds (Grafton et al., 2013; Wang and Cai, 2009) in a variety of ways. Indeed, the results presented in this paper suggest that proximate drivers like groundwater pumping and land use change, rather than just climate change, are the most likely causes of the drying of the Arkavathy river. This work strengthens the case for use-inspired sociohydrology as a science of water and society that explicitly includes human feedback on hydrologic processes (Sivapalan et al., 2012). In particular, the paper makes three contributions to this nascent field. First, the study highlights the importance of accounting for multiple anthropogenic drivers of change. There has been a tendency within the hydrology community to understate the role of humans in altering hydrology beyond large structures like dams, or more recently climate change. The dominant conceptualisation remains that of the hydrologic system as being separate from society. This case study shows why attention to direct and dispersed human modifications of this system is needed. Second, the study offers guidance on how human feedback ought to be addressed in a region where data are scarce and unreliable. By adopting a multiple hypothesis approach, we illustrate how even limited data sources can be marshalled to eliminate some of them and identify critical knowledge gaps. This approach can inform primary data collection efforts and lead to the development CHAPTER 2. MULTIPLE HYPOTHESES OF CHANGE 31 of better models of the catchment. Third, the hypotheses themselves are derived not just from the academic literature but also from perceptions of all stakeholders in the debate. This ensures the legitimacy and usefulness of the research. 32

Chapter 3 Spatial characterization of long-term hydrological change1

3.1 Introduction

Human water consumption is straining water resources worldwide (Gleick, 2014; Lall et al., 2008; Vogel et al., 2015; Wada et al., 2012), with developing nations particularly vulnera- ble to water scarcity (Vörösmartyet al., 2010). The causes of water scarcity are complex (Srinivasan et al., 2012a) and in south India have been associated with urbanization (Srini- vasan et al., 2013), groundwater depletion (Reddy, 2005), degradation of rainwater harvesting structures (Gunnell and Krishnamurthy, 2003), and interstate water disputes (Anand, 2004). Effective management of water resources in south India requires an understanding that relates changes in hydrology to the evolving human drivers of such change. Such human interventions in the water cycle often occur due to decisions made at local scales, and therefore exhibit considerable spatial heterogeneity when considered at larger scales. This is problem- atic in this region because most research linking human drivers to hydrological responses focuses on either the local scale (Perrin et al., 2012; Van Meter et al., 2016b), or regional to national scales (Devineni et al., 2013; Gosain et al., 2011; Tiwari et al., 2009). There is little research that addresses the emergent effects and heterogeneity of human-driven hydrological change across the watershed scales at which management decisions must typically be made. The gap in scientific understanding at management-relevant scales is strongly associated with lack of data resolution at these scales, and forces water managers to make decisions without sufficient information about cause and effect within watersheds (Batchelor et al., 2003; Glendenning et al., 2012; Lele et al., 2013; Srinivasan et al., 2015). The data scarcity that challenges understanding of human-driven hydrological change in south India is a common challenge in hydrology and has been extensively explored through “predictions in ungauged basins” (PUB) (Bonell et al., 2006; Hrachowitz et al., 2013) over the

1This chapter was submitted to Hydrology and Earth System Sciences (HESS) in 2016 with the title “Spatial characterization of long-term hydrological change in the Arkavathy watershed adjacent to Bangalore, India.” At time of filing this thesis, the article is under review and available through HESSD. CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 33 past two decades. The methodologies de- India veloped through the PUB initiative focused Karnataka Arkavathy strongly on near-“natural” basins, where WRS-1 proxies for flow behavior (whether climatic, WRS-2 geographic or geomorphic) could be used to form a space in which to extrapolate flows observed in gauged basins to those in the un- (a) gauged site (Blöschlet al., 2013). Extending these techniques to heavily managed catch- 0 20 km10 ments presents numerous challenges, includ- (b) ing the identification of suitable proxies to define the effects of human intervention and non-stationarity of the water cycle (Thomp- son et al., 2013b). Given the complexity 1 of these managed systems, hydrological reconstruction to infer or reproduce the history of hydrological change can help identify the predominant processes that relate 2 human water use and management with the hydrological response. Here we present such a hydrological re- 3 construction covering four decades of extensive hydrological change in the Arkavathy 4 watershed near Bangalore, India (Fig. 3.1). Concern about water scarcity in the Arka- vathy watershed has grown with the loss of historical monsoon-season river flow and reduced inflows to the TG Halli reservoir, which was the primary water supply reservoir for Bangalore between the 1930s and Rivers 1970s. These inflows have declined by nearly Tanks 5 Reservoirs: 80% since the late 1970s, a time period that 1 Hesaraghatta 2 TG Halli also included groundwater depletion and loss 3 Manchanabele 4 Byramangala Boundaries: of storage in surface reservoirs. Analysis by 5 Harobele Arkavathy Srinivasan et al. (2015) showed that neither Stream gauge Bangalore trends in precipitation nor evaporative demand could explain the observed changes Figure 3.1: Site map. (a) Location of the Arka- in river flow. Instead, reductions in river vathy watershed in Karnataka, India, and scene channel flow were probably caused by hu- boundaries for Landsat 1–3 (WRS-1) and 4–8 man drivers of change such as expansion (WRS-2). (b) Map of the watershed including tanks, reservoirs, gauge locations, river network, of Eucalyptus plantations, groundwater de- and Bangalore. pletion associated with irrigated agriculture, CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 34 and the construction of in-stream check dams (Srinivasan et al., 2015). Groundwater irrigation grew in popularity in India in the 1960s (Briscoe and Malik, 2006), supplanting tank irrigation in south India in the following decades with the widespread adoption of borewells for groundwater pumping (Janakarajan, 1993a). Groundwater is now the dominant source of irrigation water in the Arkavathy watershed (Lele et al., 2013; Srini- vasan et al., 2015). The availability of year-round reliable water supplies led to increases in the extent and intensity of agricultural production, and thus further demand for water. Replacement of traditional crops with Eucalyptus plantations, and population growth and urbanization around the periphery of Bangalore, the road network, and other urban hubs have also likely increased water demand. As villages and farmers became more reliant on groundwater, they attempted to augment groundwater recharge by constructing hundreds, if not thousands, of in-stream check dams which impound a portion of streamflow which is then removed from the channel via groundwater recharge or evaporation (Srinivasan et al., 2015). These decentralized land and water management decisions are spatially heterogeneous and characterizing their effects on surface water is hindered by the lack of hydrological records in the Arkavathy. However, spatially explicit characterization of variations in these drivers and hydrological change across the watershed could offer a basis for drawing conclusions about the likely causes of change, thus assisting in the development of management approaches. To date, such analysis has been limited to anecdotal stakeholder accounts (Lele et al., 2013). Our reconstruction relies on developing a history of change in post-monsoon storage in widely distributed surface rainwater harvesting structures known as tanks (Vaidyanathan et al., 2001; Van Meter et al., 2014). Agriculture in south India was historically sustained by a series of reservoirs known collectively as the “cascading irrigation tank system”. Nearly 1700 tanks have been constructed in the Arkavathy watershed. Tanks typically consist of a long, shallow dam bund constructed across a river to harvest surface runoff during the monsoon and supply irrigation water during the dry season. The bund impedes streamflow until the tank fills, overflows, and “cascades” into downstream tanks. Although the dam bunds remain in place, village-level water managers report that the tanks rarely fill up and overflow in large portions of the Arkavathy (ATREE et al., 2015), similar to other watersheds in south India (Gunnell and Krishnamurthy, 2003; Janakarajan, 1993b; Kumar et al., 2016). This decline of tank water is a cause of concern in the Arkavathy and much of the region, and multiple efforts have been initiated to rejuvenate tanks, often without clear understanding of the drivers of degradation of the system (Kumar et al., 2016; Srinivasan et al., 2015). An illustrative example of one of the tanks in the watershed is shown in Fig. 3.2 for two conditions - one prior to a runoff event, and one following a runoff event in August 2014. This tank, like all tanks in the watershed, is directly connected to surface flow in the river channel network. Consequently changes in the water surface area within tanks (tank water extent) - such as the changes occurring between the two images shown in Figure 3.2 - provide a proxy for surface flow generation over the upstream catchment area. In situ measurements of tank water storage have been successfully used to calibrate and validate hydrological models in Andhra Pradesh (Perrin et al., 2012) and Tamil Nadu (Van Meter et al., 2016b). Other studies in south India (Mialhe et al., 2008), the USA (Halabisky CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 35

Tank River bund channel

July 24, 2014 August 26, 2014

Figure 3.2: Aerial photos of a small tank containing turbid water in the Arkavathy watershed before and after runoﬀ events in August 2014. The tank receives water from the channel and directly from adjacent agricultural plots, and water extent increases with storage.

et al., 2016), Africa (Gardelle et al., 2010; Liebe et al., 2005, 2009; Meigh, 1995; Sawunyama et al., 2006) and South America (Rodrigues et al., 2012) also use surface water bodies as aggregators of streamflow. Hydrological changes in the Arkavathy watershed should be apparent in historical satellite imagery, as the period of reported hydrological change in the Arkavathy (from the late 1970s onward) coincides with the initial image collection by Landsat satellites in 1972. We develop an automated approach for estimating tank water extent in the Arkavathy watershed using Landsat imagery and apply this approach to reconstruct a timeseries of water extent in tanks from 1973 to 2010. We then undertake a statistical analysis that identifies temporal trends in water extent while controlling for variability in precipitation over the study period. We interpret long-term trends in tank water extent that remain after controlling for precipitation variations as an indication of spatially-variable hydrologic nonstationarity. Specifically, we hypothesize that declines in tank water extent, estimated independently of precipitation variations, derive from human activity that depletes local groundwater resources, such as groundwater irrigation or groundwater mining by Eucalyptus plantations, or that impedes flow generation in the channel network, such as check-dam construction. To explore this hypothesis, we compare the non-precipitation-related temporal trends of tank water extent against land use profiles developed by Lele and Sowmyashree (2016). Each of these analyses: the remote sensing, tank water extent modeling, and land use - water extent trend comparison is outlined in the methods section below.

3.2 Methods

3.2.1 Study area and remote sensing analysis The Arkavathy watershed spans 4,253 km2 on the western edge of the city of Bangalore in Karnataka, south India (Fig. 3.1). It has a monsoonal climate, with the rainy season lasting from June to November, relatively stable daily maximum temperature of 27◦C, and mean CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 36 annual rainfall of 820 mm. The monsoon season includes the southwest monsoon from June- September and the northeast monsoon from October-November (we exclude December, the official end of the northeast monsoon, which brings less than 5% of annual rainfall). We therefore refer to several discrete periods of time within the year as the pre-monsoon period, taken as April-May, the wet season or monsoon season, between June and November, the post-monsoon period, taken as December-February, and the dry season, December-May. We also refer to the monsoon year, analogous to the usual concept of the water year, spanning the period from April to March of the following year. Temperature peaks near the end of the dry season in April around 34◦C, before pre-monsoon rainfall arrives sporadically in April and May. The river is gauged at TG Halli reservoir (Location 2, Fig. 3.1b) and upstream of Harobele reservoir (Location 5, Fig. 3.1b). The watershed contains a mix of urban, natural and agricultural land uses. Agricultural land can be divided into rainfed grain crops, irrigated vegetable crops, Eucalyptus plantations, and other irrigated tree plantations (e.g., areca nut). Most present-day irrigation water in the Arkavathy is sourced from a deep, fractured rock aquifer. Irrigation from tanks is now significant in only a few locations, mostly located downstream of Bangalore. The city of Bangalore imports water from the regional Cauvery river and returns some urban wastewater to the Arkavathy system. Although many tanks are no longer in use, the tank structures remain intact and continue to capture inflow. The aim of the remote sensing analysis was to generate a timeseries of the surface area of water stored in each tank (referred to from now on as the ‘tank water extent’) in the Arkavathy watershed. There is minimal rainfall or flow outside the monsoon period, and analysis of tank areas within the monsoon period is inhibited by extensive cloud cover. The analysis therefore focuses on post monsoon images from the months of December and January. A detailed description of the remote sensing methods employed is provided in the Ap- pendix A, Section A.1, and the main steps are summarized below. Landsat imagery was used for analyses. Sixteen (16) images taken in December or January between 1973 and 2010 were classified to provide information about end-of-monsoon tank water extent. An additional 32 images were also classified to assist in validation, and to provide information about tank water extent variations during the dry season (see Appendix A Fig. A.1 and Table A.1 for imagery dates). A range of pre-processing and quality assurance/quality control procedures were performed on the imagery, including converting all Landsat imagery to top-of-atmosphere reflectance (Chander et al., 2009), corrections for scan-line error in Landsat 7 ETM+ images (Catts et al., 1985; Chen et al., 2011; Scaramuzza et al., 2005), and masking of cloud shadows (Craven et al., 2002; Irish, 2000; Zhu and Woodcock, 2012). The location of tanks within the resulting images was determined using a shapefile of tank boundaries obtained from the Karnataka State Remote Sensing Application Centre (KSRSAC, karnataka.gov.in/ksrsac), supplemented by 1970s topographic maps (surveyofindia.gov.in) for the beginning of the study period (see Table A.2 for all data sources used in this study). Within each identified tank boundary, a two-stage approach for estimating water extent was followed (see CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 37

Section A.1). First, pixels having definitive spectral properties of water were identified and classified as “apparent” water pixels. Second, spectral unmixing was used to estimate the water fraction in all pixels within 60 m of any apparent water pixels. Tanks were excluded from the analysis if their boundaries intersected with a cloud or cloud shadow, if ≥25% of their area overlapped with missing pixels due to the SLC error in Landsat 7, or if ≥25% of their area overlapped with the edge of the scene.

Validation of classification method Classification results were validated against a 5 m resolution LISS IV satellite image from 26 February 2014 using a classified Landsat image from 27 February 2014. The classifications were compared at the pixel scale and tank scale. Pixel level validation details are described in the Appendix A, Section A.4. Tank scale Landsat results were compared directly to the results from the reference (LISS) classification, ignoring tanks in which there were obvious differences due to the incongruous image capture dates (e.g., cloud cover). We also used Digital Globe imagery available from Google Earth (Google Earth, 2016) to assess the validity of the classification in normal (680–955 mm) versus wet (>955 mm) precipitation years during the study period. Given the limited availability of these images, we were unable to find a dry-year image (<680 mm) within the study period that was suitable for comparison with a mostly cloud-free Landsat image. We manually delineated 18 tanks in the normal year (2009) and 34 tanks in wet years (2004 and 2005), and compared the manual delineation with classification of Landsat images from the same time period using a linear regression.

3.2.2 Statistical model of tank water extent The aim of the statistical model is to identify changes in tank water extent that could be attributed to changes in streamﬂow production in the Arkavathy watershed. To achieve this, the model should control for drivers of water extent variability other than streamﬂow. Bathymetric surveys in the Arkavathy watershed indicate that tank surface area is a function of tank volumetric storage (Young et al., 2017). Thus, a volumetric water balance for a tank can be used to consider the drivers of water extent variability, as follows:

t t t t X2 X2 X2 X2 S(t2) = S(t1)+ (P −Drainage−ET )Atank + Qin − Qout − W ithdrawals, (3.1)

t1 t1 t1 t1 where S indicates tank storage at time t2 when the Landsat image was taken, S(t1) is the storage in the tank at some prior time t1, P is the daily precipitation depth over the tank area, Drainage the drainage from the tank floor, ET evaporation from the tank surface area, Atank is the tank surface area, Qin the streamflow entering the tank, Qout the overflows leaving the tank, W ithdrawals any anthropogenic withdrawal from the tank itself, and sums are taken from t1 to t2. The statistical model we developed addressed these sources of variation by (i) approximating S(t1) with zero, (ii) directly accounting for variations in P (and thus CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 38

their contribution to variations in Qin), (iii) neglecting variations in Qout, for two reasons - firstly, because watershed managers report that tanks rarely overflow, so Qout can reasonably be approximated as ≈ 0, and secondly because any overflow that does occur implies that S is equal to its maximum Smax, so that variations in overflow cannot contribute to changes in observed S, (iv) treating the sum of Drainage, ET and W ithdrawal fluxes as a stationary cumulative loss term, and (v) accounting for time trends in tank water extent that remain having accounted for (i)-(iv). Such time trends, if present, would indicate the presence of non-stationarity in tank water extents that could not be explained by precipitation. We confirmed that (i) is reasonable by analyzing carry-over storage across the dry season using 2014 imagery (selected because of high image availability). Carry-over storage from 2013 monsoon to the start of the 2014 monsoon was ≤25% or approximately ≤12.5% of post monsoon storage for more than 50% of tank clusters, and ≤50% or approximately ≤35% of storage for more than 75% of clusters (storage to volume conversions are based on bathymetric data reported in Young et al., 2017). Tank clusters with the highest carryover storage (as inferred from water extent) were found in urban subwatersheds or hilly sub watersheds in the southern part of the Arkavathy watershed (see Fig. A.10). These results suggest that carry-over storage is minimal in most parts of the watershed and that neglecting its effect on tank water extent variability is reasonable. Variations in P (ii) were accounted for using daily rainfall data from 62 gauges operated by the Karnataka State Natural Disaster Monitoring Centre (KSNDMC). Precipitation trends were analyzed using Mann-Kendall non-parameteric tests. Exploratory analysis at the whole-basin scale indicated that tank water extents were most related to precipitation totals from September 1 to to the date of Landsat image acquisition. Contemporary observations in the Arkavathy watershed suggest that only the largest or most intense storms generate runoff. The average depth of large storms (>10 mm/day) from September 1 to the date of the Landsat image was used as a metric of extreme rainfall occurrence to account for these observations. Finally, we confirmed that treating the sum of Drainage, ET and W ithdrawal fluxes as a lumped linear loss term was reasonable by focusing on the early dry season (December 1 onward). Rainfall is negligible in this period. Previous analysis of monitored locations shows that since the early 1970s, no streamflow occurred in the Arkavathy watershed other than in months when rainfall occurred (Srinivasan et al., 2015). Changes in tank water extent from December 1 into the early dry season are therefore dominated by loss terms. We confirmed that these losses were stationary in 6 of the 8 watersheds analyzed, and that they were well described by a linear model, based on linear regression and non-parameteric Mann–Kendall tests using classified tank water extents obtained from 27 dry season Landsat images (see Fig. A.10). All analyses proceeded by considering two spatial scales: 8 subwatersheds, which represent regions of relatively homogeneous climatic forcing, and 42 smaller hydrologically- connected subwatershed units, each containing at least 15 tanks having non-zero water extent in at least 4 post-monsoon images, which are referred to as “tank clusters” (Fig. 3.3). Aggre- CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 39 gated tank water extents for each cluster form the basis for statistical analysis. Ag- Subwatershed gregating data in this way overcomes some Hesaraghatta of the challenges associated with a relatively Kumudavathy short record and frequently dry tanks, while offering enough spatial resolution to identify TG Halli East variability in trends across the Arkavathy Vrishabhavati watershed. The analysis excluded reservoirs, Manchanabele because the water extent in a reservoir is also influenced by active management and wa- Suvarnamukhi ter transfers. Analysis also excluded tanks Kanakapura that were constructed during the study pe- Harobele riod from consideration prior to their construction. These model features (i) - (v) were incor- Figure 3.3: Subwatersheds and tank cluster watersheds. Each tank cluster contained at least 15 porated into a multivariate regression with tanks. interactions between continuous covariates and categorical variables (e.g., see Cohen et al., 2003; Jaccard et al., 1990). The covariates used were cumulative monsoon season rainfall (from September 1 onward), denoted Ptotal; average depth of large storms during the monsoon season (from September 1 onward), denoted Pextreme; time delay between the end of the monsoon (December 1) and the date of Landsat image acquisition, denoted DSD; and the year in which the observation was made, denoted Y ear. The Ptotal, Pextreme, and DSD covariates were modeled as fixed effects which interact with the subwatersheds. In other words, the response of the tank water extent to these variables was allowed to vary for each subwatershed, but was assumed to be consistent for the tank clusters within the subwatershed. The year effect was estimated separately for each tank cluster. The model can be written as follows:

Acluster,ij = C0 + C1,kPtotal,ij + C2,kPextreme,ij + C3,kDSDi + B1,jY eari + eij (3.2) The subscripts refer to the Landsat scene (i), tank clusters (j), and subwatersheds (k). Other than the intercept (C0), the fixed effects differ for each subwatershed (C1,k, C2,k, and C3,k) or tank cluster (B1,j). The errors for each observation are included as eij. The model predicts the tank water extent per cluster (Acluster,ij), normalized by its maximum. Tank clusters were only analyzed if ≤ 30% of the total cluster tank area was missing (due to tanks being omitted for QA/QC purposes in classification, or not having been constructed by the date of analysis). All covariates were centered by subtracting the mean before being input into the model. We confirmed that collinearity between covariates was minimal and did not impact interpretation of confidence intervals or model output using Generalized Variance Inflation Factors (Fox, 2008; Fox and Monette, 1992) (see Appendix A, Section A.2 for details). The model performance was assessed using multiple R2 statistics and significance of all effects. CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 40

The primary result of interest is the Y ear effect on tank water extent for each cluster, B1,j. This effect represents a temporal trend in total tank water storage over time (as a percent change over time), after controlling for a stationary relationship between tank water storage and the covariates (Ptotal, Pextreme, DSD). In the 6 watersheds where dry season losses were stationary, we attribute this change to changing inflows, as all other sources of non-stationarity are controlled for. In the two subwatersheds where a change in the effect of dry season water loss on tank storage was detected, B1,j captures the combined effect of hydrological change and non-stationarity in dry-season tank water losses. Because the value of B1,j is the key result of interest, additional analyses were performed to confirm its importance. Specifically the model was refit while omitting the Y ear effect 2 B1,j. The performance of the two models (with and without B1,j) was compared via R metrics. The significance of deviations between the two model predictions was tested using an F-test (H0 : B1,j = 0, HA : B1,j 6= 0, for at least one value of j).

3.2.3 Hydrologic change and land use We used four land use maps of the TG Halli watershed, encompassing the three subwatersheds (containing a total of 17 tank clusters) upstream of the TG Halli reservoir (TG Halli East, Kumudavathy, and Hesaraghatta), developed for 1973-74, 1991-92, 2001-02, and 2013- 14 (Lele and Sowmyashree, 2016). The maps differentiate agricultural land use classes into rainfed crops, irrigated agriculture, and Eucalyptus plantations. Irrigated agriculture in this region is supplied almost exclusively by groundwater, allowing us to test whether groundwater irrigated agriculture, increased water utilization by Eucalyptus plantations (Srinivasan et al., 2015), both, or neither, are associated with the identified streamflow trend. In the early 1970s, rainfed agriculture was the primary land use in the TG Halli watershed. Over the study period, many farmers adopted groundwater irrigation and others converted their fields to Eucalyptus plantations, which have the potential to mine shallow groundwater or to significantly reduce deep recharge. These land use changes have the potential to reduce surface water flows by depleting subsurface water availability and baseflow over time, likely resulting in a non-stationary streamflow response. This non-stationarity, in conjunction with the relatively sparse availability of land cover data over time, complicated a direct analysis of land use against tank water level. Instead, a space-for-time approach was used to compare the differences in time-averaged land use across each tank cluster to the differences in the Y ear effect B1,j found for each cluster. We therefore calculate the time-average land use fraction corresponding to irrigated crops (Airrigated,avg) and Eucalyptus plantations (AEucs,avg) for each of the 17 tank cluster watersheds and regress (B1,j) against these these land fractions:

B1,j = CEucsAEucs,j + CirrigatedAirrigated,j (3.3)

The coeﬃcients, CEucs and Cirrigated, correspond to the sensitivity of hydrological change to time average Eucalyptus land cover and irrigated agriculture land cover, across all 17 tank clusters. This analysis is not designed to directly infer causation, but rather to understand associations between streamﬂow decline and agricultural practices. CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 41

3.3 Results

3.3.1 Accuracy assessment The Landsat classification yielded timeseries of water extent in each of the tanks throughout the watershed (e.g., see Figs. A.5 & A.6). The Landsat classification agreed well with the reference LISS classification at the tank scale, and accuracy improved with increasing tank size. A regression of Landsat extent versus reference extent (Figure 3.4) for tanks less than 25 hectares (278 pixels) had a slope of 0.98 and coefficient of determination (R2) of 0.95. When all tanks and reservoirs were included, the regression line had a slope of 1.02 and coefficient of determination of 0.99. Over 99% of dry tanks were correctly classified as dry, but error was considerably large for small tanks with non-zero water extent less than 2.5 ha (28 pixels), due to false positives in the reference classification as well as errors the Landsat classification. For tanks between 2.5 and 10 ha the classification performed considerably better. The mean absolute error increased as the extent of the water body increased, but mean percent error decreased with water body size. Pixel scale accuracy assessment (see

(a) (b)

(c)

Figure 3.4: Comparison of Landsat and reference (LISS) classification from February 2014 images. (a) Water extent in tanks less than 25 ha. (b) Water extent in all tanks and reservoirs. (c) Error in the Landsat classification for tanks and reservoirs. Relative error decreases with increasing tank size. Only three of the five reservoirs are included because the LISS image excluded the Harobele reservoir and there was considerable change in an algae bloom in the Byramangala reservoir in the time between the acquisition of the LISS and Landsat images. CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 42

Appendix A) indicated that classification at pixel scales was accurate for completely wet or completely dry pixels (producer accuracies of 84% and 99% respectively), and lower for mixed pixels (producer accuracy 41–82%), see Fig. A.7. Comparison of our automated Landsat classification similarly compared well with the Google Earth manual delineation of tanks in both normal years (R2 = 0.97) and wet years (R2 = 0.97) (see Fig. A.8). Although the time-variation in most tanks has not been reported via in situ measurements, trends in water storage over time are widely known for major reservoirs. The TG Halli and Hesaraghatta reservoirs declined from a peak storage in the 1970s to much lower contemporary storage. Large increases in water extent were observed in Manchanabele reservoir, which was constructed in 1993, and Harobele reservoir which was constructed in 2004. These anecdotal trends corroborate our findings for these specific structures (Figure A.9).

3.3.2 Heterogeneity of long-term hydrologic change Trend analysis of the 62 rain gauges in the watershed showed that there were no statistically significant trends in rainfall at whole watershed (see Fig. A.11), subwatershed (not shown), or tank cluster scales (see Fig. A.12). Precipitation has thus been stationary (with considerable inter-annual variability) during the period of analysis, and any identified trends in tank water extent over time can exclude consideration of precipitation change as a driver. The multivariate analysis explained nearly 70% of the variation in tank cluster water extent (R2 = 0.68). Model residuals were normally distributed (Figure A.13). The effects of both precipitation covariates (Ptotal and Pextreme) were significant (the 95% confidence interval of the slopes excluded zero) in nearly all subwatersheds, and the effect of dry-season water loss was significant in the two subwatersheds that flow into TG Halli reservoir. The model suggested that inter-annual variability in precipitation (Ptotal and Pextreme) explained 63% of the total variability in tank water extent through the period of record, while the date of observation explained 10% of the variability. The multivariate analysis also identified significant Y ear effects B1,j (Table A.3, Fig. S14) in 13 tank clusters. B1,j varied in its sign and statistical significance among tank clusters, and explained 27% of the total variation in tank water extent (see Fig. A.15 for a comparison of the effects of precipitation and B1,j across each cluster.). In the two subwatersheds flowing directly into the TG Halli reservoir, B1,j captured the combined effect of non-stationarity in streamflow generation and non-stationarity in dry-season tank water losses (lower tank losses increase B1,j). If the sign of B1,j is negative in these tanks, it implies that the effect of non-stationarity in streamflow generation must both be negative and exceed the effects of reduced tank water losses. We converted the units of B1,j to an areal rate of change over time per 10 km2 of catchment area (Figure 3.5). In the three subwatersheds upstream of TG Halli reservoir, most tank clusters exhibit negative Bi,j values, implying reductions in streamflow generation. Tanks within Bangalore generally exhibited negative Y ear effects, and tanks at the city periphery and immediately downstream of the city had positive effects. Other regions of the watershed exhibited mixed values of Bi,j, but none were statistically significant at the 95% confidence level. CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 43

We confirmed that the Y ear effect B1,j was important for understanding the variations in tank water extent. Omitting the Y ear effect from the model lowered the R2 from 0.68 to 0.58. Furthermore, the model predictions with and without the Y ear effect were significantly different according to the F-test (p < 3.1x10−11). These results allow us to reject the null hypothesis that B1,j = 0 and Y ear effects could be ignored. Overall, the results indicate that while interannual variations in rainfall totals and extremes explain the majority of interannual variation in tank water level, a trend in tank water level is present in several regions of the Arkavathy watershed that is independent of rainfall variability. This trend cannot be explained by trends in rainfall (which were negligible), by trends in dry season tank water loss rates (when present, had the opposite sign to the identified trend in water level), or by changes in outflows (which were constrained to occur when tank storage is at its peak). The results suggest that changes in 4 streamflow production independent of rainfall are occurring in discrete locations in the 0 Arkavathy watershed, and that the sign of these changes varies through space. −4

−8 3.3.3 Streamflow decline Temporal trend (ha decade−1 10 km−2) and agricultural practices The regression of the Y ear effect B on ir- Figure 3.5: Values of B , the Y ear effect on clus- 1,j i,j rigated agriculture and Eucalyptus land use ter water extent, 1973–2010, given as change in water surface area (ha) per decade per 10 km2 of areas explained most of the differences in 2 watershed area. White space indicates subwater- B1,j between tank clusters (R = 0.68). The shed boundaries, and black lines indicate statisti- relationship between irrigated crops and B1,j cal significance of the cluster trend. The sign of was statistically significant (95% confidence Bi,j offers insight into likely trends in runoff ra- intervals of Cirrigated excluded zero), and the tio (streamflow generated within each tank cluster relationship with Eucalyptus plantations was per unit incident rainfall). not statistically significant (Fig. 3.6). CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 44

0.20 ● Eucalyptus 0.15 ● ● Irrigated crop 0.10 ● ● 0.05 ● 0.00 ● Land use fraction

1970 1980 1990 2000 2010

Eucalyptus ● Irrigated crop ●

−1.0−0.50.0 Model coefficients

Figure 3.6: Agricultural land use and hydrological change. (Top) Land use fraction of Eucalyp- tus plantations and irrigated crops in four land use maps. (Bottom) Model coeﬃcients (CEucs, Cirrigated)relating hydrological change to Eucalyptus and irrigated crops based from the multivariate linear regression. Horizontal lines indicate 95% conﬁdence intervals.

3.4 Discussion

3.4.1 Long-term hydrological trends and human drivers of change Tank water extent at the end of the monsoon season can be primarily attributed to the storage of monsoon season streamflow: tanks in the Arkavathy watershed rarely overflow, there is little carry-over storage year to year, and loss processes do not extensively deplete the tanks from the end of the monsoon period to the time when tank water extents were observed by Landsat. Thus, storage of water in tanks offers an integrated measure of tank inflows during the previous wet season. Statistical analysis of the tank water extents suggests that while inter-annual variability in tank water extent is largely explained by precipitation, in many parts of the watershed, this variability is superimposed on a longer-term trend in tank water extent that is independent of precipitation, representing a non-stationarity in inflows. Analysis of rain gauges indicated that precipitation has been stationary within the watershed during the study period. Non-stationarity in inflows, coupled with stationarity in precipitation, indicate changes in flow production per unit precipitation (runoff ratio) - commonly an indicator of changing hydrological processes in a watershed (Hughes et al., 2012). Historical land use maps for the TG Halli watershed indicate that there is an association between the inferred streamflow generation trends - particularly streamflow declines - and human drivers of change. We hypothesized that the inferred decline in streamflow would correspond with agricultural practices associated with groundwater depletion. Although little data exist to describe historical declines of the water table, contemporary farmers typically have to drill new borewells to depths exceeding 100 m to reach any groundwater. CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 45

If a loss of baseflow due to groundwater depletion and the disconnection of the water table from the stream channel is a primary driver of streamflow decline, we would expect the negative trends in streamflow to correspond with irrigated agriculture, which is supplied almost entirely by groundwater in the TG Halli watershed. In the linear model relating the Y ear effect B1,j to land use in the TG Halli watershed (Equation 3.2, Section 3.1), the time-averaged irrigated crop land use area is a clearer and stronger predictor of declines in tank water extent than Eucalyptus land use (Fig. 3.6). Moreover, other exploratory analyses showed that irrigated crop land-use is more correlated 2 2 with B1,j (R = 0.68, see Fig. A.16) than rainfed crops (R = 0.5) and all other land-use types (R2 ¡ 0.38). Areas retaining mostly rainfed crops exhibit lower absolute values of B1,j, and lower values of B1,j are associated with areas with higher conversion of rainfed crops to irrigated crops. The finding that Eucalyptus plantations do not play a major role in streamflow decline is consistent with field experiments, which show that that Eucalyptus plantations tend to reduce soil infiltration capacity and therefore would increase infiltration excess runoff (Penny et al., 2017). There could be some relationship between Eucalyptus plantations and non-stationary hydrologic processes, but if so it is secondary to that of irrigated crops. Areas with a high fraction of irrigated agriculture are also likely to contain relatively higher densities of check dams than other land use types, given the desire to recharge diminished groundwater resources. In the absence of datasets describing the spatial distribution and hydrological properties of check dams (or a viable way to develop such a dataset), this analysis is unable to separate the effect of loss of baseflow due to groundwater pumping from the in-stream losses due to check dams. Both processes likely play a role in observed hydrological changes. Recession analyses indicate that the loss of the shallow water table could plausibly explain the observed magnitude of streamflow declines (Srinivasan et al., 2015), and check dams exacerbate the loss of streamflow by converting water in the stream channel to groundwater recharge. The most negative values of B1,j and thus the largest inferred reductions in streamflow production occurred in the northernmost regions of the Arkavathy where elevation is higher than other areas of the watershed. Although it may appear that the pattern of decline could be related to upstream-downstream processes and the presence or absence of irrigation return flows (Van Meter et al., 2016b, e.g., see), we are doubtful that this effect is important in the Arkavathy today. Indirect evidence (e.g., surveys) indicates that the water table is hundreds of meters below the surface in northern parts of the Arkavathy watershed (Srinivasan et al., 2015). Furthermore, the relief in the watershed is ≈ 100 m over a distance of 50 km in the TG Halli watershed, meaning that system-wide return flows connecting upstream to downstream are unlikely. Urbanization could result in increased streamflow being routed to downstream tanks, due to increases in impervious surfaces, the fallowing of agricultural land in anticipation of urbanization, and reduced consumptive water use. Increased urban water use produces increased urban effluent, which is discharged to the surface channel network where it can contribute to increases in tank water storage downstream. The observed positive Y ear CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 46 trends downstream and on the periphery of the Bangalore urban area are consistent with the substantial increases in Bangalore’s imports from the Cauvery river: from 185 million liters per day (MLD) in 1974 to 1350 MLD currently (BWSSB, 2017). Additionally, as the city has grown, groundwater pumping for urban areas has increased to an estimated 600 MLD (Lele et al., 2013). About 40% of Bangalore’s sewage of 1400 MLD flows to Byramangala reservoir (Jamwal et al., 2015). This has contributed to additional inflows to Byramangala reservoir and more irrigated agriculture directly downstream of the reservoir. Tanks within urban areas can also exhibit drying trends. For instance, tanks may be encroached upon as residential areas expand. Additional urban wastewater inflow can lead to expansion of algae blooms covering the tank water surface, which can appear as a “drying” of the tank in this analysis.

3.4.2 Assessing the classification and model uncertainty The classification of small tanks in the Arkavathy watershed poses challenges associated with harmonization of different Landsat sensors and the variability in the spectral properties of “wet” tanks due to variations in water quality and vegetation extent. The classification tends to overestimate the amount of water in dry pixels and underestimate the amount of water in wet and mixed pixels. Because our classification scheme is designed to avoid bias between images taken with different Landsat sensors, we likely sacrifice some precision with sensors from Landsat missions 5–8. Because these mixed pixels lie at the boundary of the wetted tank area, classification error would be sensitive to geo-registration error in one or both of the Landsat and LISS images. Error could also arise from our specification that water pixels must lie within 60 m of clearly identifiable water bodies, or the assumptions made during spectral unmixing. Although the classification scheme accounted for only two classes, the spectral properties of the land class varied among dry soil, wet soil, sparse vegetation, and irrigated agriculture. Classification of water was complicated by vegetation in tanks, varying degrees of turbidity, and algae blooms in tanks with considerable wastewater inflow. Errors at the pixel and tank scales are likely unavoidable given the spectral heterogeneity of both land and water pixels. In particular, tanks containing water of variable turbidity, excessive vegetation, or algae blooms are prone to classification errors. Because pixel-scale errors are unbiased, accuracy at the tank scale improves as tank size increases. Error is further mitigated by grouping tanks into clusters in the statistical model. The uncertainty of the classification (R2 = 0.99 when all water bodies are included) is small compared with the uncertainty of the statistical model (R2 = 0.68). Although the results of our statistical model imply a non-trivial amount of unexplained variation, Gardelle et al. (2010) reported similar performance (R2 = 0.78) for a model relating precipitation and water extent in a single lake, and noted that the correlation was valid only for a nine-year subset of the five-decade study period. The sources of uncertainty include the complex hydrological processes that relate precipitation, streamflow, and tank water storage, as well as the nonlinear and heterogeneous relationship between water extent and water storage, CHAPTER 3. SPATIAL CHANGES IN SURFACE WATER 47 the neglect of pre-monsoon tank water extent in the model, and the non-stationary behavior of dry-season losses in the two northernmost watersheds. Given this uncertainty, results of our analysis are reasonable given the simplicity of the model and the complexity and heterogeneity of the watershed hydrological response.

3.5 Conclusions

The Arkavathy watershed embodies many of the water security challenges confronting southern India. With data limitations hampering the characterization of changing water supplies in the watershed, remote sensing tools provide insights into the history and spatial pattern of change in water availability and hydrological function. We were able to take advantage of a pre-existing “sensing network” provided by the irrigation tank system throughout the Arkavathy watershed. The high number of tanks in this watershed allowed for a comparison of hydrological change with land use at spatial scales appropriate for a first-order analysis. The analysis reveals that while tank water extent inter-annual variations are dominated by inter-annual variation in precipitation, an independent time trend in tank water extent occurs for a subset of the watershed. This trend is not spatially homogeneous, but varies in its magnitude and sign among different regions of the watershed. These differences appear to be associated with differing patterns of land use across the watershed. A comparison of the hydrological trends with agricultural practices within the TG Halli watershed showed that declines in tank water extent over time, controlling for precipitation, are more closely associated with groundwater irrigated agriculture than other kinds of land use, including Eucalyptus plantations. This association is consistent with hypothesized effects of groundwater depletion on streamflow generation in the Arkavathy, and with the potential influence of check-dams in fragmenting the surface flow network (Srinivasan et al., 2015). Further investigation could attempt to attribute the cause of the inferred streamflow decline, either via a more sophisticated statistical analysis considering the many potential drivers of change or via a mechanistic model of catchment hydrological functioning. Ideally such analysis would also separate the relative effects of loss of baseflow due to groundwater pumping and conversion of surface flows to groundwater recharge via check dams. Surface networks of rainwater harvesting structures are employed in seasonal climates worldwide, whether in cascading tank systems in southern India and Sri Lanka, or hillslope farm dams in Australia (Callow and Smettem, 2009; Roohi and Webb, 2012), North-East Brazil (de Araújoand Medeiros, 2013; de Toledo et al., 2014; Lima Neto et al., 2011; Malveira et al., 2012), South Africa (Hughes and Mantel, 2010), the US Great Plains (Womack and Others, 2012) and China (Xiankun, 2014; Xu et al., 2013). Capitalizing on these networks as proxy indicators of rainfall and streamflow variation, as in the Arkavathy, could prove a valuable approach to circumventing problems of data scarcity and characterizing changing hydrological conditions. 48

Chapter 4 A process-based hydrologic reconstruction1

4.1 Introduction

The extent of human intervention in the hydrologic cycle is unprecedented (Vörösmartyet al., 2004, 2013), undermining traditional assumptions of stationarity (Ehret et al., 2014; Milly et al., 2008; Peel and Blöschl, 2011) and forcing water scientists to make predictions and design sustainable management strategies in a nonstationary water cycle that is continually evolving in response to human drivers (Montanari et al., 2013; Pande and Sivapalan, 2017; Savenije et al., 2014; Sivapalan et al., 2012). These challenges are amplified by a relative lack of observational data that describes the nonstationary behavior of human-influenced water systems on decadal timescales or longer (Thompson et al., 2013b). To adequately understand and make predictions about these systems, new empirical insights are needed to develop historical baselines (Fox et al., 2015; Macdonald and Black, 2010), identify and characterize nonstationarity (Barriendos et al., 2003; Rossi et al., 2009; Saghafian et al., 2008), describe emergent system properties (Dafforn et al., 2016; Kim et al., 2011; Konar et al., 2016; Ruddell and Kumar, 2009), explain human-water interactions (Steffens and Franz, 2012), and ultimately inform decision-making approaches (Gober et al., 2017). Given the frequent mismatch between data availability and data requirements, there is scope to improve understanding of sociohydrologic systems through sociohydrologic reconstruction efforts that extend traditional social and hydrologic data beyond the instrumented record (Thompson et al., 2013b). Reconstruction of the hydrologic components of these systems has strong analogies to the ongoing efforts in “Predictions in Ungauged Basins” (PUB, Blöschl et al., 2013; Hrachowitz et al., 2013), especially as this initiative increasingly addresses water management and nonstationarity as critical elements of PUB (Blöschl, 2016).

1This chapter was submitted to Water Resources Research (WRR) in 2017 with the title “A process-based hydrologic reconstruction to understand streamﬂow decline in a human-dominated semiarid catchment.” At time of ﬁling this thesis, the article is under review. CHAPTER 4. PROCESS-BASED RECONSTRUCTION 49

4.1.1 Typology of hydrologic reconstruction Reconstruction of historical conditions using proxy data or models is standard in paleocli- matology and related fields (Cronin, 2009, e.g.), and similar approaches have been applied in palaeoflood hydrology (Baker, 2006, 2008; Benito and Thorndycraft, 2005) and historical hydrology (Brázdilet al., 2006). These disciplines naturally produce studies that are relevant to sociohydrology (e.g., see Machado et al., 2015), but the methods used to reconstruct deep-time changes in natural processes are not necessarily suitable for sociohydrologic reconstruction, where nonstationarity and a focus on human-water interactions could challenge the identification of process models with which to understand historical system behavior. Consequently, while hydrologic reconstruction of human impacted systems could proceed analogously to paeleoscience reconstructions with a focus on either data reconstruction via proxies or theoretical reconstruction via application of physical models, at least two other forms of reconstruction can be envisaged for sociohydrologic systems: a phenomenological approach that aims to build associations between historical hydrologic and social phenomena, and a process-based approach that attempts to understand such phenomenological links through mechanistic reasoning. A typology of reconstruction, illustrated in Figure 4.1, contains four categories of reconstruction, each with the capacity to improve our understanding of sociohydrologic systems. The categories have strong analogies to scientific models of data (descriptive), models of phenomena (associative), models of theory (predictive, developed through hypothesis testing), and robust-process explanations (mechanistic) (Frigg and Hartmann, 2017; Kleinhans et al., 2005). To date, these different types of reconstruction have been exploited to differing extents in hydrologic and sociohydrologic research. Data reconstruction aims to generate quantitative historical time series describing events or processes of interest. For example, environmental and documentary evidence has been used to reconstruct historical climates in Europe (Brázdilet al., 2005), and a broad range

Phenomenological Theory-based Data reconstruction

Environmental Environmental and human Processes and societal Drivers outcomes

Process-based Phenomenological

Figure 4.1: A simple sociohydrologic system and typology of hydrologic reconstruction. Data reconstruction involves generating timeseries or other information about historical events. Phenomeno- logical reconstruction associates historical outcomes or nonstationarity with drivers of change. Theory-based reconstruction applies knowledge about process to explore system dynamics, including drivers and outcomes (outward arrows). Process-based reconstruction seeks to identify the mechanism relating drivers and outcomes (inward arrows) through experiments, theory, and additional observations. CHAPTER 4. PROCESS-BASED RECONSTRUCTION 50 of techniques including analysis of epigraphic marks, flood observations, archival descriptions and paleoflood deposits have been used to reconstruct flood frequencies prior to the instrumented record (Benito et al., 2015). Geophysical proxies (tree rings, pollen records, sedimentary deposits), historical archives (tax records, written records), historical maps (Zlinszky and Timár, 2013), or combinations of these (Büntgenet al., 2011) have all been employed to conduct data reconstructions of physical systems. Social analogues to such data include historical economic statistics (Mitchell, 1971), historical population reconstructions (Russell, 1948), and, for example, reconstructions of diet from isotopic records (Müldnerand Richards, 2007). Data reconstruction is instrumental in quantifying historical environmental conditions and extreme events, and providing perspectives on variability in social and hydrological conditions on scales of decades to millennia. However, it does not necessarily offer direct insights into drivers of hydrologic change or the mechanisms by which changes are produced, the latter being of particular interest for sociohydrologic research. Phenomenological reconstruction aims to produce an explanatory narrative about historical change by making associations between observable events (drivers and outcomes) based on their concurrence or sequence. This strategy is similar to actual-sequence explanations (Kleinhans et al., 2005). It has been widely employed in historical analyses of the effects of climate change on society, for example by identifying associations in time between periods of rapid climate change and societal collapse (Tol and Wagner, 2010; Weiss and Bradley, 2001; Zhang et al., 2007). In hydrologic applications, Benito et al. (2010) identified changes in flood frequency in the Guadalent´ınriver in Spain and in parallel documented a historical sequence of human-driven events such as land use change that could explain the observed nonstationarity in flood frequency. Foulds et al. (2014) used geophysical evidence to reconstruct two centuries of flooding and climate oscillations in the Cambrian Mountains in Wales, and connected changes in flood frequency to seasonal climate variations. Hutton et al. (2017a,b) developed characteristics of multiple large-scale infrastructure and water management projects in California, and associate historical changes (1920–present) in seasonal inflow to the San Fransciso Bay Delta with these projects. Phenomenological reconstructions offer additional insight into potential associations between social and hydrologic outcomes, but have been criticized for neglecting complexity and offering at best weak evidence of causality (Salehyan, 2008; Zhang et al., 2011). One avenue towards causal reasoning is deductive “theory-based” reconstruction. This form of reconstruction attempts to test hypotheses against historical observations, or to use hypotheses to reconstruct feasible narratives of past behavior. Typically, model simulations are used to achieve this reconstruction. In climate reconstruction, this approach has found wide application with the use of GCMs to make predictions about historical behavior of the climate system that are then tested against climate proxy datasets (Chandler et al., 1992; Joussaume and Taylor, 1995; Smerdon et al., 2015). In sociohydrology, models have been hypothesized to represent coupled human-water systems, often including phenomenological descriptions of social response such as the ‘memory’ of past flood events that inhibits de- velopments on floodplains (Viglione et al., 2014), or the ‘sensitivity’ to environmental harm that increases with the duration of such harm and alters the value accorded to environmen- CHAPTER 4. PROCESS-BASED RECONSTRUCTION 51 tal protection (Elshafei et al., 2014). The simulated model output can then be compared to observations or reconstructed timeseries as a test of the hypothesized system dynamics (Di Baldassarre et al., 2009, 2013a,b; Elshafei et al., 2015; Kuil et al., 2016). Despite the appeal of theory-based reconstruction, human-induced nonstationarity in catchment conditions and parameters is likely to introduce uncertainty and, potentially, parameter and model structural errors (Beven, 2016). There are clear situations where climate change induces hydrologic changes (Spencer et al., 2009) such as permafrost melting in the Arctic, which is associated with deeper active soil layers, increased groundwater-river connectivity, and prolonged discharge into rivers (Walvoord and Striegl, 2007; Yang et al., 2002). Land use change, such as urbanization, is notorious for shifting runoff generation and routing processes relative to natural systems (Carlson and Arthur, 2000; Rose and Peters, 2001; Van Meter et al., 2016a). Such changes may pose challenges to theory-based reconstruction or confound attribution of cause and effect. They suggest that a fourth mode of reconstruction, process-based reconstruction, may be valuable in assigning cause and effect by using contemporary and historical observations to understand the mechanistic connections between drivers and outcomes (rather than prescribing these mechanisms, as in a theory-based reconstruction). This approach has analogies to the development of robust-process explanatory narratives (Kleinhans et al., 2005), and has the advantage of helping constrain modeling and interpretation of phenomena, reducing the potential for interpretation pitfalls such as equifinality (Beven, 2006) or reverse causation bias (Maclure and Schneeweiss, 2001). Of course, incomplete data availability and the inability to conduct experimental manipulations or observations of historical processes challenges process-based reconstruction. From the perspective of social science research, process-based reconstructions are likely to be informed by qualitative understanding of behaviors and decision-making (as could be revealed from written or oral histories or interviews) in addition to data analysis or modeling. One potential avenue to reconstruct historical hydrologic processes is to apply the method of multiple hypotheses (Chamberlin, 1965a) by identifying several plausible hypotheses describing the processes through which causation occurs and seeking evidence to falsify each one — in this case, ruling out specific historical mechanisms. Clearly there are overlaps, synergies and interdependencies between these different types of reconstruction. For example, hydrological process understanding at large scales is often necessarily phenomenological in nature. In many cases the most robust approaches to understanding historical dynamics will draw on multiple reconstruction approaches. Machado et al. (2015) reconstructed flood frequencies in the Tagus river in Spain (data reconstruction), noted that drastically reduced flooding in the mid-20th century was preceded by construction of large reservoirs on the river (phenomenological reconstruction), and accurately simulated historical behavior using a model that incorporated these dynamics (theory-based reconstruction). Bürger et al. (2006) used documentary evidence to reconstruct characteristics of a large flood event in the early 19th century in the Neckar river in Germany (data reconstruction), then modeled the climatic and hydrologic processes that could have produced such an extreme event (theory-based reconstruction). CHAPTER 4. PROCESS-BASED RECONSTRUCTION 52

4.1.2 A process-based reconstruction approach Here, we undertake a process-based reconstruction of the strongly nonstationary and human- impacted hydrology of the TG Halli watershed, a subwatershed in the Arkavathy basin, located near the city of Bangalore in Karnataka, India. The TG Halli watershed terminates in a water supply reservoir, TG Halli reservoir. Inflows to this reservoir declined dramatically since the 1970s, reducing water supply to Bangalore. These declines were coincident with a suite of other hydrologic changes in the catchment, including reduced inflow to the widespread ”cascading tank” (Meter et al., 2014) rainwater harvesting system in the catchment (Penny et al., 2016), dramatic declines in groundwater level (Lele et al., 2013; Srini- vasan et al., 2015), and abandonment of shallow surface wells and bores. The declines were also coincident with considerable changes in land and water management including expansion of irrigated agriculture and Eucalyptus plantations, rapid urbanization, and widespread watershed management efforts (Lele et al., 2013; Penny et al., 2016). The declines in flow cannot be attributed to climatic changes (Srinivasan et al., 2015), suggesting that it is the changes in land and water management that produced the changes in hydrology. The TG Halli watershed is a candidate for hydrologic reconstruction for several reasons: (i) data are sparse, suggesting data reconstruction could be valuable, (ii) the associations between hydrologic change and human-induced land and water management changes remain unclear, and (iii) the hydrologic processes in the watershed remain obscure. These factors impede hydrologic predictions and hamper evaluation of management alternatives. Management responses to water scarcity from local and state agencies have been largely uncoordinated and at times contradictory (Srinivasan et al., 2015), despite the many symp- toms of water crisis in the basin, including escalating costs and reduced predictability of groundwater supplies, rapid land conversion, and near complete loss of surface water resources. Reconstructing hydrologic change in this system has the potential to reveal the historical and complex set of relationships linking hydrology (at point, field, subcatchment scales) with local and regional agents (farmers, water management agencies, state policy). Reconstruction would, therefore, contribute new insights to the study of sociohydrology as well as provide understanding and information pertinent to local and collective actors in the system, who could adjust their actions and management responses accordingly (Thompson et al., 2013b). Hydrologic reconstruction can also provide constraints on a more detailed sociohydrologic exploration of linked changed between anthropogenic interventions and hydrologic responses over time (see Srinivasan et al., for such an analysis on the TG Halli watershed). Efforts in data reconstruction and phenomenological reconstruction in the TG Halli watershed have produced valuable insights regarding potential drivers and hydrologic outcomes of human activity within the watershed (Penny et al., 2016; Srinivasan et al., 2015), but there remains a need to demonstrate the causal processes linking human drivers to the observed drying of the river. Here, we focus on a process reconstruction with the objective of understanding contemporary and historical runoff generation mechanisms in the watershed. We adopt a threefold strategy, beginning with experimental and observational research aimed CHAPTER 4. PROCESS-BASED RECONSTRUCTION 53 at elucidating contemporary runoff generation mechanisms. These findings are then used to constrain a perceptual model of past system behavior and generate a range of process-relevant hypotheses. Finally, these hypotheses are tested against historical information about system function derived from a broad range of evidence, including historical flow records, farmer surveys (Srinivasan et al., 2015), and the features of historical infrastructure designed and constructed prior to the observed hydrologic changes.

4.2 Methods

4.2.1 Study area and runoﬀ generation in semi-arid catchments The TG Halli watershed is located in Karnataka, India, spanning 1,447 km2 to the northwest of Bangalore (Figure 4.2). The watershed contains two main river channels, which drain into the TG Halli reservoir at the watershed outlet. The climate is tropical semi-arid and the watershed receives approximately 700 mm of annual rainfall, which arrives mostly during monsoon season between June and November. Temperatures are fairly consistent throughout the year around 25 ◦C, peaking prior to the monsoon season in June. The watershed topography is mostly ﬂat or very gently sloping, with 86% percent of the watershed consisting of slopes less than 3% (ISRO and IN-RIMT , 2000). The eastern portion of the watershed contains the western periphery of the Bangalore urban area, which was historically supplied with drinking water by TG Halli reservoir. Outside

(a) (c) (d) 0 5 10 km

India 0 1 2 km

(e)

(b)

Tank water level Weather station TG Halli Soil moisture Reservoir Storm isotopes

Figure 4.2: TG Halli location and instrumentation of the three study watersheds. (a) India. (b) Karnataka. (c) TG Halli watershed with tanks and major stream channels. (d) Thirumagondona- halli study watershed with Hadonahalli weather station and soil moisture sites. (e) Doddatumkur (large tank) and SM Golahalli (small tank) study watersheds with SM Golahalli weather station. Storm isotopes were taken near Ekasipura village within the Doddatumkur watershed. CHAPTER 4. PROCESS-BASED RECONSTRUCTION 54

Bangalore, the landscape is dominated by agriculture, both rainfed and irrigated. Urban- ization has caused the urban area to expand into agricultural land at a rate of 2.3 km2 per year over the study period, and 3.8 km2 per year since 2002 (Lele and Sowmyashree, 2016). The river network in the TG Halli watershed is fragmented by a series of surface water harvesting structures known as tanks (Vaidyanathan et al., 2001), which consist of a shallow bund wall built across natural river valleys (Van Meter et al., 2016b). Most of the tanks were constructed centuries ago and formed a “cascading tank system,” in which tanks would fill and overflow downstream, forming a connected network. As surface water flows in the basin have declined, tanks in the watershed now rarely overflow. Irrigation, historically supplied by stored water in the tanks, is now primarily sourced from groundwater (Lele et al., 2013). Extensive groundwater pumping has resulted in water table depletion on the order of 100 m (Srinivasan et al., 2015). Numerous check dams have been constructed in headwater channels to increase groundwater recharge. The check dams function like weirs, with the impounded water being lost to evaporation or recharging local groundwater. Farmers have adapted to reduced groundwater availability by converting their fields to Eucalyptus plantations, which require no irrigation and little maintenance. Other farmers have abandoned their crops and left their land fallow. Field studies focused on three subwatersheds of the TG Halli for study, each defined by its receiving tank: SM Golahalli, Doddatumkur, and Thirumagondonahalli (Figure 4.2d,e). The land use characteristics of these study watersheds are representative of much of the TG Halli watershed, with a mix of rainfed crops, groundwater irrigated crops, Eucalyptus plantations, perennial irrigated plantations, and fallow land. We selected study watersheds with a contrast of groundwater irrigation and Eucalyptus plantations because these two land use categories have been suggested as potential drivers of streamflow decline (Srinivasan et al., 2015). The immediate upstream watersheds of SM Gollahalli and Doddatumkur tanks contained higher percentages of groundwater irrigated agriculture, and the Thirumagondonahalli watershed contained greater coverage of Eucalyptus plantations.

Hypothesized runoff generation processes Field work and data analyses in the three subwatersheds aimed to identify runoff generation processes that could be generating contemporary streamflow, and to find evidence relating to historical runoff generation mechanisms. Since rainfall and temperature appeared to be largely stationary in the basin, we hypothesized that a change in runoff generation mechanisms operating in the basin may be responsible for the large observed reductions in surface water flow. In semi-arid subtropical watersheds such as TG Halli, streamflow generation is often associated with overland flow (Yair and Lavee, 1985), most typically due to infiltration excess (Hortonian) runoff. However, subsurface stormflow and saturation excess overland flow have also been observed in semi-arid watersheds (Beven, 2002). The runoff generation mechanisms in the TG Halli watershed have not been determined, although the presence of a shallow argillic layer in the soil column as well as the observation that relatively high inflow CHAPTER 4. PROCESS-BASED RECONSTRUCTION 55 to TG Halli occurred late in monsoon season led previous researchers to suggest saturation excess runoff as a potential streamflow generation mechanism (ISRO and IN-RIMT , 2000). Given the ambiguity regarding contemporary and historical runoff generation mechanisms in the basin, field work attempted to demonstrate the occurrence or absence of (i) infiltration excess runoff, (ii) saturation excess runoff, and (iii) groundwater discharge to the stream channel under contemporary conditions (Figure 4.3). Infiltration excess runoff would be supported by the presence of overland flow, by channel flows consist primarily of “new” water, and by soil conductivities that were regularly lower than rainfall intensities (Horton, 1933). Saturation excess runoff would be supported by the presence of overland flow consisting of a mix of old and new water, by saturation of surficial soils above an impermeable layer prior to runoff, and by soil hydraulic conductivities exceeding rainfall intensities (Dunne and Black, 1970; Hewlett and Hibbert, 1967). Groundwater discharge would be associated with a shallow or perched groundwater table and would result in a discharge of old water to the stream (Whipkey, 1965).

Visual observations, Soil moisture, Soil moisture, flow traps tank inflows well survey

Overland flow Shallow soils Perched / shallow occurs? saturated? water table?

Infiltration Saturation Groundwater Excess Excess discharge

Runoff contains old K > Intensity? sat water?

Rainfall & infiltrometer Isotope tracers measurements

Figure 4.3: Three canonical runoff generation mechanisms are hypothesized as alternative streamflow generation pathways in the TG Halli catchment: infiltration excess, saturation excess, and subsurface stormflow and groundwater discharge to the channel. This figure shows how different pieces of evidence would support the occurrence of these mechanisms, and relates potential findings to research tasks undertaken in the study watersheds. CHAPTER 4. PROCESS-BASED RECONSTRUCTION 56

4.2.2 Field instrumentation and experiments Water level and tank inflow Streamflow estimates were generated from measurements of tank water level and tank bathymetry. Each of the three study tanks (SM Golahalli, Doddatumkur, and Thirumagon- donahalli) was instrumented with an Odyssey Capacitative Water Level Logger (Dataflow Systems Inc, 2017), which was manually calibrated to water levels in the tank. Unmanned vehicles were used to map the bathymetry of the tanks, merging maps generated from aerial vehicles for dry areas and surface vehicles for wetted parts of the tanks (Young et al., 2017). The combined bathymetric surveys and water level measurements were used to calculate tank water storage from the water level timeseries. Streamflow records were then generated for each of the watersheds for the 2014–2016 monsoon seasons. The capacitance sensors were sensitive to temperature, leading to daily fluctuations in the apparent water level and overestimates of the water level during inflow events as cold water mixed into the tank. As temperature levels equilibrated after the storm, the tank water levels appeared to drop. Because these temperature effects were challenging to separate from any actual changes in tank water storage, sub-daily analyses relied primarily on estimates of the timing of peak flow, used in Section 4.2.3, which was unaffected by the temperature fluctuations.

Weather stations Precipitation measurements were obtained from tipping bucket rain gauges installed near the SM Golahalli tank and the Hadonahalli village (see Figure 4.2) from April 2014 through November 2016. Data from the rain gauges were compared to 15-minute precipitation data from 2011 through 2016 from the Karnataka State Natural Disaster Management Center (KSNDMC) as a quality assurance and control procedure. Rain events reported in the tipping buckets were flagged if they exceeded daily precipitation at the nearest KSNDMC rain gauge by more than 20 mm, or if recorded rainfall intensities exceeded 15 cm per hour. The flagged data were manually removed from the record if rainfall was inconsistent with streamflow or soil moisture records (e.g., when heavy rainfall was not accompanied by any response in streamflow or soil moisture). Overall, 17 events were removed from the Hadonahalli precipitation record and 5 events were removed from the SM Golahalli record. Many of these events occurred in the dry season, likely indicating a sensor malfunction.

Soil moisture Soil moisture sensors were installed at four agriculture sites with different cropping and irrigation technologies, including a grape site with drip irrigation, a cabbage site with flood irrigation, a rainfed site, and a Eucalyptus plantation. At the rainfed, grape, and cabbage sites, four Decagon EC-5 Soil Moisture sensors (Decagon Devices, 2016) were installed, two each at depths of 30 cm and 1 m. The Eucalyptus site was instrumented with Decagon EC-5 CHAPTER 4. PROCESS-BASED RECONSTRUCTION 57 sensors at depths of 30 cm, 1 m, 2 m, and 3 m, and Campbell Scientific CS650 Water Content Reflectometers (Campbell Scientific, 2014) at depths of 1 m, 3 m, and 3.7 m. Soil moisture sensors at the grape, rainfed and Eucalyptus sites reported from the beginning of the 2014 monsoon season through 2016. The cabbage site sensor failed multiple times resulting in large data gaps from October 2014 through March 2015 and ceased collecting useful data in July 2016. The Eucalyptus site sensors were potentially influenced by ponded water in a nearby dug pit. The longest, most consistent soil moisture records were obtained at the rainfed and grape sites, which formed the primary basis for analysis here. Soil moisture records were manually checked for quality control, to ensure self-consistency and coherency across sensors. Our analyses rely on relative soil moisture, S, which we calculated as V W C/n, where n is the porosity. Local soil porosity at each sensor was estimated as the maximum recorded VWC during the study period. At sensors where the recorded VWC never exceeded 0.3, we set porosity to 0.3.

Hydraulic conductivity Hydrologic conductivity at the land surface was measured using a Soil Moisture Equipment (SME) Tension Infiltrometer (Soilmoisture Equipment Corp., 2008) and CSIRO Disk Per- meameter (CSIRO, 1988). At each SME measurement point, the infiltrometer was used to measure infiltration rates at an unsaturated pressure head (-7 to -12 cm) and near-saturated pressure head (-1 to -3.5 cm). Parameters of soil hydraulic conductivity (Ksat and α) were calculated using a nonlinear model based on the method developed by Gardner (1958) and Wooding (1968) (Logsdon and Jaynes, 1993). At each CSIRO site, the permeameter was used to measure infiltration rate at a near-saturated pressure head (-1 to -2 cm) and Ksat was estimated by applying the same method by including α taken from a tension infiltrometer measurement at the same site. Sites were selected to cover the different types of land use in the subwatersheds, specifically crops, Eucalyptus plantations, and beds of dry tanks (for locations, see Supporting Information Figure S1). At 25 of the locations, we made at least 2 replicate measurements, aiming to capture within-site variability. Overall, 83 measurements were made at 35 locations (19 crop, 10 Eucalyptus, 6 tank).

Stable isotope tracer study Stable isotopes of water were analyzed from samples of rainfall, soil water, and runoff collected near Ekasipura village (see Figure 4.2e) over the course of two storms, on 28 September 2014 and 30 September 2014. Each storm lasted less than an hour and generated runoff in the local stream channel. Soil water samples were made from an agricultural field growing corn, and runoff samples collected upstream of a check dam entering the nearest main channel. Precipitation samples were made continuously during the storm using using a funnel and collector, and stored in a sealed flask to prevent evaporation. Soil water was collected using CHAPTER 4. PROCESS-BASED RECONSTRUCTION 58 a suction lysimeter, installed at a depth of 30–35 cm. The lysimeter was installed ahead of time and left under suction to collect water prior to the storm. The first soil water sample taken reflected soil water conditions prior to the storm. The lysimeter was fully emptied after each sample was taken, so that each successive sample represented new water in the lysimeter. Runoff samples were collected manually. Deuterium and δ18O concentrations were obtained from the water samples by the stable isotope laboratory at the University of California, Berkeley.

Overland flow traps To detect the presence of overland flow, two flow traps were installed September 19–25, 2014, in agricultural fields containing corn, one in SM Golahalli and another near Ekasipura village. Another flow trap was installed October 24–31 in a Eucalyptus plantation near the Hadonahalli weather station (see Figure 4.2d,e). Flow traps consisted of a pan collector (with entry facing uphill) that drained into a storage container, and were constructed using locally available household items. Overland flow from the upstream fields collected in the flow trap and remained stored until removal.

Open well survey Long-term groundwater level observations in the TG Halli catchment were either unavailable, or were located in unrepresentative areas. To assess shallow groundwater levels in the study watersheds we surveyed 99 shallow open wells during the 2014 monsoon season (for locations, see Supporting Information Figure S1). The time period of the survey was coincident with storms that produced significant local runoff. Our expectation was that if this runoff were associated with shallow groundwater tables, this would be reflected in the water levels in these wells. Presence or absence of water in each well were noted, and the depth of the well bottom relative to the local land surface was measured, and converted to a common datum using a local digital elevation model DEM and GPS coordinates of each well. These well base depths were compared to stream-bed elevations determined from the same DEM.

4.2.3 Storm event analysis Over the course of the three-year study period, we measured precipitation greater than 1 mm on 174 days at the Hadonahalli station and 107 days at the SM Golahalli station. Daily precipitation was measured for each 24 hour period from 8:30 AM to 8 AM the following day. Because the bulk of precipitation during monsoon season occurred in the late after- noon through the early hours of the morning, we assumed that each 24-hour period with precipitation contained a single storm event. Each day with precipitation greater than 1 mm was treated as a storm event and associated with a suite of hydrologic metrics, including cumulative tank inﬂow, timing of peak tank inﬂow, peak soil moisture, and antecedent soil moisture. CHAPTER 4. PROCESS-BASED RECONSTRUCTION 59

The cumulative tank inflow for any given storm was taken as the tank storage volume at the end of the 24 hour period minus the tank storage volume at the beginning of the 24 hour period. We identified runoff events as those events when there was an increase in instantaneous water level of >5 mm over the 24-hour period and an increase in the average water level of >0 mm from previous day. Both metrics were needed to ensure runoff was attributed to the appropriate day (instantaneous metric) and that the runoff metric was robust to instantaneous conditions and variability (average metric), given the sensitivity of the sensors to temperature. Soil moisture measurements were converted to relative soil moisture, and we noted the peak soil moisture associated with each storm. The peak soil moisture at each depth was set to the maximum soil moisture within the 24-hour storm period across all sensors at that depth in the grape and rainfed monitoring sites. For several subsets of storm events (e.g., events characterized by a given rainfall volume, or by a rainfall volume and peak soil moisture metric), we also calculated “runoff probability” as the number of runoff events divided by the number of precipitation events.

4.3 Results

4.3.1 Overland flow The within-storm isotope analyses suggested that streamflow water was comprised of rainwater, with no evidence of mixing between rainwater and pre-event soil water in the measured runoff. In both storms A (in red, Figure 4.4) and B (in blue), the relative concentration of deuterium (δD) in runoff tracks precipitation, while the concentration in soil water remains relatively constant and distinct from the values in the rain and runoff. In dual isotope space (not shown), runoff samples fell along the local meteoric water line. The first runoff sample in the second storm was more enriched in O18 than rainfall, suggesting the possibility of mixing with evaporatively enriched water at the ini- Figure 4.4: Within-storm deuterium isotopes for tiation of the storm, potentially from small two one-hour storm events, A and B, including amounts of surface ponding carried over be- precipitation (P), soil moisture (S), and runoff tween the storms. The bulk of the runoff (Q) isotopic signatures. The signature of runoff however, matched the profile of new water, tracked precipitation and did not appear to con- as opposed to old water stored in the soil. tain a soil water component. CHAPTER 4. PROCESS-BASED RECONSTRUCTION 60

There were 5 storms at the SM Golahalli rain gauge during the period when flow traps were deployed, including 4 which generated runoff into the tank. Overland flow was produced and stored in flow traps during 3 of these storms at the SM Golahalli tank, and 4 in the Ekasipura location. There were 3 storms in the Hadonahalli watershed during the period of flow trap deployment, 1 of which produced runoff into the Thirumagondonahalli tank and 2 of which filled the flow trap (either the runoff was localized away from the tank or not sufficient to qualify as tank inflow).

4.3.2 Saturated hydraulic conductivity

Saturated hydraulic conductivity (Ksat) was highest at cropped sites with mean (plus or 1.00 minus standard deviation) values of 3.85 ± 0.75 3.49 cm/hr, followed by Eucalyptus plan- 0.50 tations (1.35 ± 0.93 cm/hr) and tank beds 0.25 (0.84±0.81 cm/hr). Precipitation rates were Prob exceed 0.00 comparable with saturated hydraulic con- 0 5 10 15 ductivity, and in large storms precipitation Rain rate, cm/hour rates were greater than Ksat at both cropped and Eucalyptus sites (Figure 4.5). Under a Crops simpliﬁed scenario in which the landscape is

covered by 34% Eucalyptus and 66% crops Eucs ● (the relative proportions in the TG Halli wa- Land cover tershed), and the assumption that infiltra- 0 5 10 15 tion capacity is equal to saturated hydraulic Ksat, cm/hour conductivity (the limiting case), much of the 1.00 landscape would be producing infiltration 0.75 excess runoff in heavy storms (Figure 4.5c). 0.50

4.3.3 Storm 0.25

dynamics and timing Area runoff fraction 0.00 Over the three year study period, most mon- 0 5 10 15 soon season storms at Hadonahalli (61% of Rain rate, cm/hour 174 storms) were smaller than 10 mm, and Figure 4.5: Saturated hydraulic conductivity considerable rainfall (15%) arrived in just 4 (Ksat) and runoff potential. (a) Volume-based large storms (>45 mm). These 4 storms gen- probability of exceedance of precipitation, (b) erated 38% of the total runoff into the Thiru- boxplots Ksat for crops and Eucalyptus land use. magondonahalli tank. (c) Fractional area generating infiltration excess In comparing the storms that produced runoff, assuming the limiting case that infiltration runoff to those that did not, we were unable capacity is equal to Ksat. to detect a clear threshold in precipitation CHAPTER 4. PROCESS-BASED RECONSTRUCTION 61

1.00 Depth ● 30 cm ● 100 cm 0.75 ● ●● Number of events ● ● 5 0.50 ● 20 ● 40

Runoff probability 0.25 Relative soil moisture 0.2−0.8 ● 0.8−1.0 0.00 ●● 0 10 20 30 40 50 Group mean precip, mm

Figure 4.6: Runoff probability for groups of storms binned by precipitation and then relative soil moisture. Within each precipitation group, storms with higher peak soil moisture (filled-in circles) are not more likely to generate runoff than storms with lower peak soil moisture (open circles). rate, volume or antecedent soil moisture that triggered runoff occurrence. We also compared the runoff probability across groups of storms with common peak soil moisture and storm volume (Figure 4.6). The probability of runoff increased for increasing event rainfall, but did not differ for soils near saturation relative to other soil conditions. The difficulty in clearly associating runoff generation with soil and storm characteristics likely reflects the hydrologic heterogeneity that was present even in the small subwatersheds studied here. Rainfall intensity and volume, antecedent soil moisture, and check-dam storage levels likely varied between storms and through space, complicating the observed relationship between discharge into the tank and the point measurements in the catchment. Much of the runoff occurred during a few large storms sharing similar characteristics, which were illustrated by the dynamics in a representative storm occurring on 07 October 2014 (Figure 4.7). In this event, strong rainfall was followed by considerable inflow to the tank. Peak runoff occurred just before midnight, after which the soil profile continued to saturate over the next 1–3 hours with the upper sensor (30 cm) peaking in soil moisture before the lower sensor (1 m). In nearly all storms, the peak in soil moisture occurred after the peak in runoff, regardless of soil moisture site or total storm runoff volume (Figure 4.8).

4.3.4 Open well survey Ninety-nine wells were surveyed during the 2014 monsoon season. Seventy of these wells were completely dry, containing no water. In the remaining 29 wells, visual evidence of surface channels directing flow from adjacent fields into the well could be seen. These wells were managed to store runoff from overland flow. No similar channels were visible at the dry wells. Runoff occurred and filled nearby nearest tanks both before and after the well survey. CHAPTER 4. PROCESS-BASED RECONSTRUCTION 62

10 5

Rain, mm 0 15:00 18:00 21:00 00:00 03:00 06:00 09:00

0.2 0.1 Inflow,

cumecs 0.0 15:00 18:00 21:00 00:00 03:00 06:00 09:00 0 1 50 S 100 Rainfed 0.2 depth, cm 15:00 18:00 21:00 00:00 03:00 06:00 09:00 0 1 50 S 100 Grapes 0.2 depth, cm 15:00 18:00 21:00 00:00 03:00 06:00 09:00 Figure 4.7: Within-storm dynamics for a storm event on 07 October 2014, including 30-minute precipitation at Hadonahalli, inflow to Thirumagondonhalli tank in m3 s-1, and relative soil moisture profiles (S) at the rainfed and grape sites. Soil moisture was measured at 30 cm and 1 m (dashed lines) at each site and interpolated to other depths. Peak inflow occurred before peak soil moisture, which occurred first at the upper soil moisture sensor before the wetting propagated downward.

100

Site & depth Rainfed, 30cm Rainfed, 100cm Grapes, 30cm 1 Grapes, 100cm Event inflow, ML inflow, Event

−10 0 10 20

10 Count 0 −10 0 10 20 Time lag (peak Q to peak S), hours Figure 4.8: Time lag of peak inflow (Q) to peak soil moisture (S) at the grape and rainfed sites. In nearly all storms, peak soil moisture occurred after peak inflow, regardless of the total inflow. CHAPTER 4. PROCESS-BASED RECONSTRUCTION 63

● 5 5 ● ● ● ● ●● ● ● ● ●● ● ● ● ● 0 0 ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●

Relative elevation, m elevation, Relative −5 −5 ● ● ● ● ●

● ●

15 0 0 200 400 Count Distance to stream, m

Figure 4.9: Elevation of the bottom of dry open wells relative to the nearest stream channel, with the approximate elevation of the stream bed marked by the blue dashed line. In general, the relative elevation of the bottom of the well increases moving further away from the stream.

The elevation of the bottom of open wells was, in general, comparable to the elevation of the bed of the nearest stream channel. Relative to the stream, the elevation of well bottoms tended to increase moving further away from the stream (Figure 4.9).

4.4 Discussion

4.4.1 Contemporary streamflow generation Multiple strands of evidence collected in the subwatersheds point to the likelihood that contemporary runoff generation in the studied subwatersheds is associated with infiltration excess overland flow. First, there was clear evidence that overland flow occurred during runoff-producing storms. The flow trap observations detected such flow during during all runoff-producing storms in nearly all of the flow traps. The isotope analysis suggested that runoff was com- posed of new water, rather than containing signals of old water consisting of soil or groundwater. We visually observed overland flows during rainfall events in the study watersheds and saw clear evidence of erosion as flowlines, a precursor to rill formation (Merritt, 1984). The sites where we made these observations were topographically flat, distributed throughout the watershed, and suggest that overland flow generation during storms is widely distributed, rather than being confined to particular topographic positions or land cover types. Simul- taneously, we did not observe any evidence of a shallow water table, and noted a downward propagation of wetting fronts into a drier subsoil during storm events. These observations are inconsistent with contemporary flow production from a shallow or perched water table, CHAPTER 4. PROCESS-BASED RECONSTRUCTION 64 and suggest that overland flow is the primary watershed-scale mode of runoff response. Second, a number of strands of evidence suggest that infiltration excess, rather than saturation excess, is likely the mechanism generating the observed overland flow. The observed saturated hydraulic conductivities in soils were relatively low, and observed rainfall intensities exceeded these conductivities for 25% or more of the storms that occurred. The observation of a downward propagating wetting front throughout the storm, and indeed increasing soil saturation well after surface runoff peaks had been generated is also consistent with the occurrence of infiltration excess runoff. Typically at least a fraction of the observed soil columns remained unsaturated during runoff producing events, again consistent with infiltration excess, but not saturation excess, streamflow generation mechanisms. The insensitivity of runoff probability to soil moisture is also consistent with this mechanism. The only finding we made that was inconsistent (or at least inconclusive) with respect to infiltration excess runoff generation was the absence of a clear threshold in storm intensity around which runoff did or did not occur. We hypothesize that this is likely to be associated with heterogeneity in rainfall intensity across the studied subwatersheds, and potentially also to heterogeneity in water storage in check dams, which fragment the channel network and may obscure the relationship between generation of runoff at field scales and propagation of runoff to the tank at the watershed outlet.

4.4.2 Historical streamflow generation Findings from this study of contemporary streamflow generation constrain our perceptual model of how a change in streamflow generation could have occurred with respect to infiltration excess runoff, saturation excess runoff, and groundwater discharge. Previous analyses suggest that rainfall and its characteristics were near-stationary in the region over the 20th Century (Srinivasan et al., 2015), suggesting that changes in rainfall intensity are unlikely to have occurred. While extensive land use change has occurred, it has primarily consisted of transitions from traditional cropping to Eucalyptus plantations (Lele and Sowmyashree, 2016). Because Eucalyptus plantations exhibit lower saturated hydraulic conductivity than traditional cropland (see Figure 4.5), these changes would be inconsistent with a reduction in runoff generation via an infiltration excess mechanism. Therefore, we conclude that historically higher streamflows must have been associated with an additional runoff generation mechanism, beyond those occurring today. Because the occurrence of saturation excess mechanisms are primarily dictated by geology and landform, which have not been substantially altered in the basin, it is challenging to conceive of a situation where saturation excess runoff occurred in the past but not the present. Conversely, several strands of evidence suggest that the change in surface runoff could be associated with the decline of an historically present seasonal shallow groundwater table. The open wells surveyed in this study are large (approximately 5–10 m diameter), manually constructed and widely installed structures, and historically were used for agricultural irrigation. Presumably, such extensive construction and use of shallow surface wells would only be consistent with a hydrologic regime in which those wells regularly intersected the groundwater table. Moreover, a sample of borewell drill CHAPTER 4. PROCESS-BASED RECONSTRUCTION 65 logs within the TG Halli watershed obtained from the 1970s (for locations, see Appendix Figure B.1) indicate that the depth to the water table encountered during this time period was comparable to the depths of the shallow wells, further supporting the hypothesis that the shallow well depth is a reasonable proxy for past groundwater depths (Figure 4.10). By comparing the contemporary well depths and contemporary channel elevations, it is clear that a groundwater table that was shallow enough to have been accessed by the wells would also have been shallow enough to intersect the contemporary stream channels (Figure 4.9), and moreover, that a head gradient would have existed between the wells and stream channels, supporting groundwater discharge to the channel. There are other circumstantial pieces of evidence supporting the notion that a historical shallow groundwater table could have existed and sustained surface flows during the monsoon season. Long-term monthly inflow data to the TG Halli reservoir indicate the historical presence and subsequent loss of protracted baseflow, in some cases lasting for 1–2 months after the cessation of monsoon season rainfall (Srinivasan et al., 2015). A baseflow analysis suggested that the known decline in the water table could represent a large-enough change in catchment storage to explain the ‘missing’ water in contemporary river discharges (Srinivasan et al., 2015). Phenomenologically, the extent of groundwater irrigated agriculture in subcatchments across the TG Halli watershed is correlated to the magnitude of surface discharge decline since the 1970s (Penny et al., 2016). The evidence presented here relating the loss of shallow groundwater to the loss of streamflow is necessarily circumstantial, given our inability to directly observe historical runoff generation processes. Nonetheless, the fact that diverse strands of evidence are consistent lends confidence to an interpretation that reductions in surface flow in the TG Halli watershed can be attributed, at least in part, to a decline in shallow groundwater levels. In the absence of a shallow water table to route infiltrated water to streams and sustain baseflow, only

●

Depth, m ●

● 15

●

2014 bottom 1970s water of dry wells table

Figure 4.10: Boxplots of the depth to the bottom of dry wells in 2014 and depth to the water table taken from borewell drill logs in the 1970s. The distribution of water table depths in the 1970s is similar to the distribution of well-bottom depths surveyed in 2014. CHAPTER 4. PROCESS-BASED RECONSTRUCTION 66 episodic runoff production via infiltration excess mechanisms to generates streamflow today. Similar changes in runoff dynamics have been observed in other systems. For example, in catchments in Western Australia where drought caused groundwater levels to drop >5 m below the depths of the stream elevation, runoff ratios never exceeded 0.03, while in catchments where groundwater remains within 5 m of the stream elevation, runoff ratios were tightly coupled to groundwater depth, exceeding 0.2 in some cases (Hughes et al., 2012; Ki- nal and Stoneman, 2012). Drops in runoff ratio were noted in 46% of catchments in Eastern Australia following the 10 year Millennium Drought, which also lead to extensive declines in groundwater levels (Saft et al., 2015, 2016), and following groundwater disconnection induced by pumping for agricultural irrigation in the High Plains aquifer in the USA (Kustu et al., 2010). In larger river systems, prolonged declines in groundwater levels are associated with switches from gaining to losing streamwater-groundwater interactions (Brunner et al., 2009), further altering runoff ratios as streamflow is progressively lost to unsaturated soils beneath the streambed. Regardless of the relative importance of runoff production versus losing conditions in the channel for the TG Halli watershed, large declines in groundwater level have an established ability to change the hydrologic regime and functioning of basins at multiple scales (Petrone et al., 2010). Such changes in hydrologic regime underline the strong motivation for understanding process change when undertaking hydrologic reconstruction of nonstationary systems. The simple runoff parameterizations (e.g., via a curve number or solely as infiltration excess runoff) popular in common modeling approaches (e.g. in Indian water resources management, the SWAT model) undermine their potential to explore the history of flow production in managed basins (Neitsch et al., 2011). Furthermore, such approaches are unlikely to be successful in basins where hydrologic regime change has occurred and emphasize the importance of process based reconstruction as a precursor to theoretical or model based approaches. Land and water management changes in the TG Halli catchment are not confined to declining groundwater. Ongoing fragmentation of the flow network by installation of irrigation bunds on farm fields, and installation of check dams in headwater channels are likely to have contributed to the observed surface water declines. Fortunately, the timing of these events (loss of shallow groundwater, installation of check dams, installation of irrigation bunding) is separable. Quantitative attribution of the impact of each of these changes and management efforts on the surface flow behavior remains the subject of ongoing work (see Srinivasan et al.).

4.5 Conclusions

With surface flows into TG Halli reservoir currently at <25% of historical levels, groundwater depths continuing to drop more than 100 m below the land surface, and ongoing land conversion in the TG Halli catchment, the present water management situation has many characteristics of an emerging water crisis (Srinivasan et al., 2012a). Abandonment and CHAPTER 4. PROCESS-BASED RECONSTRUCTION 67 fallowing of crop fields by some farmers (Lele and Sowmyashree, 2016) suggests that collapse and reorganization of the social-ecological system is already occurring at smaller scales and that the system may be in a state of low resilience (e.g., see Holling, 2001). In the absence of reconstruction efforts to understand the genesis of this emerging crisis, policy makers and managers have been left without a consensus evidence base from which to assess the drivers of hydrologic change, and without a reliable platform with which to forecast and develop future scenarios against which to evaluate potential management options. In the case of TG Halli catchment, the data sparse environment and limited historical records have prompted a wide array of reconstruction activities to attempt to generate such an evidence base and foundation for forecasting: data reconstruction (Srinivasan et al., 2015), phenomenological reconstruction (Penny et al., 2016), and — as presented here — process-based reconstruction. These efforts offer scope to test and evaluate potential water resources modeling efforts and strengthen their ability to represent contemporary hydrologic processes, historical change and thus future potential changes. As shown here, there is strong albeit circumstantial evidence of a hydrologic regime shift from a connected groundwater system sustaining relatively high runoff to a disconnected system in which infiltration excess runoff is the remaining form of flow production. Any future modeling and policy effort should be cognizant of this major source of nonstationarity in the functioning of the catchment. Process-reconstruction is clearly a challenging and potentially nebulous task for hydrologists to undertake. Yet given the importance of nonstationary, human influenced water systems for social, economic and environmental well being, and the potential for these systems to undergo significant shifts in hydrologic regime as catchment states change under human influence, it is likely to be essential for robust modeling and prediction efforts in water resources management. Historical hydrologic reconstruction has already demonstrated the critical importance of leveraging multiple evidence sources in order to obtain insight into the past (Benito et al., 2015; Brázdilet al., 2005; Büntgenet al., 2011). If these diverse sources can be synthesized within a hypothesis testing framework, then there is the potential to understand historical catchment functioning, even after extensive human modification and in the absence of direct historical measurement and observations. 68

Chapter 5

The inﬂuence of check dams on watershed-scale groundwater recharge

5.1 Introduction

Groundwater plays a critical role in sustaining global agricultural production (Siebert et al., 2010), such that overuse of groundwater resources has resulted in extensive depletion of groundwater in many regions of the world (Gleeson et al., 2012a; Wada et al., 2010, 2012). The consequences of groundwater depletion vary in severity and have included ecological degradation (Sophocleous, 1997), energy and economic costs for increased pumping (Foster et al., 2004; Konikow and Kendy, 2005; Narayanamoorthy, 2014), loss of livelihood and migration (Fishman et al., 2013; Moench, 2002), and collapse of agricultural systems (Roy and Shah, 2002). These repercussions of groundwater depletion often disproportionately affect poor farmers (Hoogesteger and Wester, 2015; Mukherji and Shah, 2005). Given the nature of groundwater as a common pool, open-access resource (Ostrom et al., 1994), groundwater depletion is unlikely to subside without external influence on farmers (Burness and Brill, 2001; Gardner et al., 1990). Conservation mandates to reduce groundwater depletion are promising (Gleeson et al., 2012b; Sophocleous, 2010), though mostly have occurred in developed countries (Aeschbach-Hertig and Gleeson, 2012). Taxes on groundwater use also have potential (Burness and Brill, 2001), but are unlikely to work in developing countries, which often have the opposite problem in which groundwater is subsidized by free or low-cost electricity (Scott, 2011; Shah et al., 2003). The groundwater crisis is particularly concerning in India, where groundwater depletion rates are among the highest in the world (Aeschbach-Hertig and Gleeson, 2012) and over half a billion people are dependent on groundwater for their livelihoods (Shah et al., 2003). Effective management strategies have been elusive (Shah et al., 2008, 2012), but decentralized rainwater harvesting systems have been promoted as a supply-side solution in recent decades (Agoramoorthy, 2008; Shah et al., 2008), largely through watershed development initiatives (GOI , 1994, 2002). This approach typically includes local capture of surface runoff via land CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 69 management practices (e.g., field leveling and small farm bunds) or check dams, which are small dams constructed within the river channel to capture streamflow in the channel (any water that exceeds the storage capacity of the dam overflows and continues downstream). In both cases the water captured by the landscape or check dams is partitioned to evaporation and groundwater recharge. Check dams, in particular, have been promoted as a viable way to reverse groundwater depletion and alleviate concerns over water scarcity (Agoramoorthy and Hsu, 2008) (check dams have also been promoted for soil and water conservation in other countries in Asia (Xu et al., 2013) and eastern Africa (Nyssen et al., 2010; Rockström, 2000)). The effect of check dams on the water balance is not entirely clear, with some studies reporting beneficial groundwater augmentation (Renganayaki and Elango, 2013) and others studies reporting minimal change in terms of water availability (Dillon et al., 2009; Kerr et al., 2002). The effects of multiple check dams on the water balance at watershed and basin scales has received little attention. We focus on the role of check dams in mediating the water balance at catchment scales in the TG Halli watershed in southern India, where agriculture is largely reliant on groundwater irrigation. As a mitigation strategy to counteract extensive groundwater depletion, numerous check dams have been constructed in the watershed. To this point, it does not appear check dams have been able reduce groundwater depletion, even as efforts to “Rejuvenate the river” (Revive Kumudavathi, 2017) have claimed success in isolated cases (Arpita, 2017). Sustainability in groundwater management has often focused on ensuring that groundwater extraction does not exceed groundwater recharge, the concept of “safe yield” (Aeschbach- Hertig and Gleeson, 2012; Sophocleous, 1997). In this framework, an increase in groundwater recharge would mean that a larger amount of water can be withdrawn while maintaining the sustainability of the system. However, this conceptualization fails to acknowledge surface water–groundwater interactions or the beneficial uses of downstream discharge. For this reason, concerns have been raised that rainwater harvesting in upstream regions reduce water availability in downstream areas (Bouma et al., 2011; Kumar and Ghosh, 2006). The objective of this study is to consider the role of check dams in two research watershed located in the TG Halli watershed. We do so in terms of the effect on the water balance, increases in groundwater recharge throughout the watershed, and the effect on downstream discharge.

5.2 Methods

5.2.1 Study area The TG Halli watershed is located in Karnataka, India (Figure 5.1), just outside the rapidly growing city of Bangalore. Two study catchments were selected within the watershed, each having representative land use for the region with diﬀerent mixes of Eucalyptus plantations and irrigated agriculture. Given the considerable groundwater depletion in this area, farm- CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 70

(a) (c) (d) 0 5 10 km

India 0 1 2 km (e)

(b)

Tank water level Weather station TG Halli Check dam Reservoir Monitored check dam

Figure 5.1: Location map. (a) India. (b) Karnataka. (c) TG Halli watershed, including tanks and major river channels. (d) Thirumagondonahalli tank and study catchment, including the Hadonahalli weather station and check dams. (e) Doddatumkur (large tank) and SM Golahalli (small tank) study catchments, including the SM Golahalli weather station. Check dam locations were identiﬁed from a stream survey of the study catchments. ers have had to cope with decreasing availability of groundwater for irrigation. Hundreds of man-made lakes, or tanks, have existed in the watershed for centuries as surface water irrigation reservoirs. However, the tanks could only provide irrigation supply when they contained water, and were controlled by tank operators who opened and closed sluice gates to provision water among farmers directly downstream in the tank command area. Ground- water, on the other hand, was not constrained to farmers living near the stream, and gave farmers control over the timing and quantity of irrigation supply. When groundwater became increasingly accessible through new technology in the latter half of the 20th century, groundwater irrigation supplanted tank irrigation throughout the TG Halli watershed. As a response to groundwater depletion, numerous check dams have been constructed in the watershed in recent decades. We surveyed stream channels in each of these catchments and georeferenced 37 check dams within the channels (see Figure 5.1d,e, and Figure 5.2 for an example). The eﬀect of these check dams on the catchment water balance was analyzed and modeled as described below.

5.2.2 Field instrumentation and data analysis Weather stations were installed in each of the Thirumagondonahalli and SM Golahalli tank watersheds (Figure 5.1d,e). These Aeron Systems Wireless Weather Stations (Aeron Sys- CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 71

Figure 5.2: A check dam in the study area under dry conditions and after filling with streamflow. tems, 2017) included a tipping bucket rain gauge for precipitation, air temperature and humidity sensors, a silicon photodiode for incoming solar radiation, and a cup anemometer for wind speed. Additionally, 15-minute precipitation data was obtained from Karnataka State Natural Disaster Management Centre (KSNDMC) for two locations just outside the study watersheds. Streamflow was measured using water level sensors and bathymetry surveys of tanks and check dams. Odyssey Capacitative Water Level Loggers (Dataflow Systems Inc, 2017) were installed in three tanks and 12 check dams beginning in June 2014, and calibrated to water level in tanks and check dams in each catchment. In tanks, the water level sensors collected nearly continuous data through 2016. Many of the check dam sensors were impaired or taken by wildlife or humans, but valuable data was collected from two check dams in the Doddatumkur watershed for the 2014 monsoon season and nearly continuous data was collected at one of the check dams in the Thirumagondonahalli watershed from June 2014 through December 2016 (see Figure 5.1 for these three “monitored” check dams). The water levels sensors were calibrated to water level in each of the check dams and tanks. Bathymetric surveys were completed in the tanks by mapping dry areas using aerial photography and wet areas using sonar (Young et al., 2017). Check dam bathymetry was mapped at the three monitored check dams using water tube level surveys (Critchley et al., 1991). Using stage-volume relationships from the bathymetric surveys, timeseries of water level from the capacitance sensors were converted to timeseries of water storage, and inflow was calculated as the change in water storage over time. Streams were delineated using a local digital elevation model (DEM), which accurately mapped large rivers through third order streams, but poorly mapped first and second order streams due to the flat topography. Smaller streams in the study watersheds were mapped by walking each of the stream channels, during which the height and width of all check dams were also measured. Because the DEM was unable to accurately delineate first and second watersheds, we manually delineated these watersheds in QGIS (QGIS Development Team, CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 72

2016) by setting ridges equidistant from parallel streams and using the DEM as a guide where the landscape was not transected by streams. The water level sensors were sensitive to temperature, leading to artificial daily oscillations in the apparent water level of the tank, as well as overshooting tank water level during storms as cold water entered into the tank or check dam. We avoided analyzing sub-daily streamflow timeseries, instead focusing on daily timescales where the oscillations were not apparent. Observations of precipitation and streamflow were aggregated to daily timescales, with each day spanning a 24-hour period beginning at 8:30 am, the daily time period used by the Karnataka State Natural Disaster Management Centre (KSNDMC). We considered any daily time period with rainfall as a “storm event” and any daily time period with runoff as a “runoff event”. Streamflow typically ceased shortly after rainfall abated, lending confidence to the assumption that each runoff event was contained within the 24-hour period. Daily evaporation rates for the three monitored check dams were calculated using a modified Penman equation developed by Shuttleworth (2007) for open water bodies. Recharge rates were then calculated as the difference between the daily change in check dam storage and the evaporative losses. See the Appendix C for detailed information about the calculation of evaporation and recharge rates.

5.2.3 Thirumagondonahalli watershed simulation We simulated hydrology of the Thirumagondonahalli watershed for the entire three-year study period using a lumped runoff model that was input into a check dam network model. During this period, streamflow was primarily driven by infiltration excess runoff (Hortonian overland flow), which occurs whenever precipitation rates exceed hydraulic conductivity. For simplicity, we assumed that infiltration capacity could be represented by hydraulic conductivity (Ksat, the limiting case) and that any precipitation in excess of Ksat would runoff. We simulated runoff over the study period using 30-minute precipitation data from the Hadonahalli weather station, and aggregated runoff to daily timescales. Runoff immediately flowed into the the nearest channel where it was then routed through the check dam network. The water storage in each check dam was updated as follows:

St = St−1 + Qin,t − Qout,t − Et − Lt (5.1)

The updated storage for the current day (St) was calculated as the storage from the previous day (St−1) plus any inflow (Qin,t) and minus any losses. Overflow (Qout,t) was calculated as any water in excess of the storage capacity (i.e., max(0,St−1 + Qin,t − Scapacity)). Any overflow was then routed to the downstream receiving water body. Daily vertical rates (mm/day) of evaporation and recharge were taken as constant values using the mean estimated fluxes as calculated above (Section 5.2.2). The volumetric evaporation (Et) and recharge (Lt) were calculated by multiplying the vertical rates by the water surface area of each check dam. The storage-area relationship was represented by a simple geometry in which the check dam was constructed in a rectangular channel with constant bed slope. CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 73

In check dams where we mapped bathymetry, we calibrated this relationship using the bed slope, which accurately reproduced the measured storage-area relationship (R2>0.95). For check dams in which the bathymetry was not measured, we assumed a rectangular channel using the measured width of the check dam and a bed slope of 2% (typical of hillslopes in the watershed). On daily timescales, inflow and outflow were effectively instantaneous processes and no sub-daily routing was deemed necessary. We chose to use a lumped runoff model and ignore spatial heterogeneity for two primary reasons. First, spatial heterogeneity of hydraulic conductivity would be most important in small storms, but these storms are most likely to be spatially heterogeneous themselves (see Appendix, Figure C.1). Because the rain gauge was unable to capture spatial variability in rainfall, capturing spatial variation in runoff was unfeasible. Second, the objective of the model was not to perfectly predict runoff, but rather to examine the effect of check dams, and the assumption of spatial stationarity of runoff throughout the watershed does not affect the role that check dams play in partitioning streamflow. Saturated hydraulic conductivity in the model was calibrated so that the total simulated inflow to the Thirumagondonahalli tank matched the observed inflow. This produced a sensible hydrograph and saturated hydraulic conductivity (Ksat) of 23 mm/hour, similar to the measured hydraulic conductivities within the catchment (see Chapter 4). Check dam capacities were calculated from bathymetry for check dams in which the bathymetry was measured. For other check dams, the capacity was estimated using the simple geometry described above (a rectangular channel with constant beds slope of 2%) along with the height and width of the check dam. Check dam recharge and evaporation were modeled as constant vertical fluxes occurring over the wetted area of the check dam, and the wetted area was calculated on a daily basis from the storage-area relationship. To test how sensitive the water balance was to varying check dam storage capacity, we reran the simulation using check dam capacities that were 5 times and 20 times greater than the original capacities. In these simulations, we maintained the same check dam network and channel geometry, but increased the height of the dam.

5.2.4 Simulation of synthetic check dam networks To understand the effect of check dams more generally, we identified a variety of configurations in which check dams could be constructed on first and second order streams, and simulated each of these configurations under a range of check dam storage capacities. In each simulation, the total check dam capacity was determined as a fraction of mean annual runoff, and the storage capacity was divided among the check dams as determined by each configuration (Figure 5.3). In the “even” configuration, 50% of the storage was distributed equally among the first order streams, with 25% of the storage on each of the second order streams. In the “first-order” configuration, 90% of the storage was distributed equally among the six first-order streams with 5% of storage on each of the two second order streams. In the “second-order” configuration, the second order streams each contained 45% of the storage, the remainder distributed equally among first-order streams. In the “opposite” configura- CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 74

Even 1st order 2nd order Opposite Lopsided

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Figure 5.3: Check dam configurations for the synthetic stream network, with circle size corresponding to total check dam storage capacity on each reach (see text for specifics). The arrow indicates direction of flow and discharge into the tank.

tion, first-order streams in one of the subwatersheds contained 45% of the storage with the opposite second order stream containing 45%. In the lopsided configuration, 90% of the storage was contained within one second-order watershed, half of it being distributed equally among the first order streams and the other half on second order streams, and the other second order watershed contained 10% of the total storage. The immediate catchment areas of all stream reaches were assumed to be equal (Kirchner, 1993). We simulated each configuration under each scenario of storage capacity using 50 years of hourly precipitation created with a random weather generator (Fatichi et al., 2011; Ivanov et al., 2007), which was parameterized with climate observations from the study watershed. We calibrated saturated hydraulic conductivity so that the runoff ratio over the entire 50-year period matched the runoff ratio from the three-year observational period. This calibration resulted in a Ksat of 17 mm/hour, a sensible value given that this synthetic model was run on hourly timescales, instead of the 30-minute timescale in Section 5.2.3. Although the observed check dam capacity in the study watersheds was quite low compared with annual runoff, other studies have reported high storage capacities from watershed development programs in other states within India. We allowed total check dams storage capacity to vary from 0.1% to 200% of mean annual runoff to explore the dynamics of a wide range of plausible systems. For each of these configurations and storage fractions, we calculated total groundwater recharge from check dams as a fraction of total runoff, as a way of estimating the effect of check dams on streamflow. To capture the variation in benefits of groundwater recharge dictated by the check dam network, we also calculated the Gini coefficient (Sen, 1973; Yitzhaki, 1998) of recharge among the nine subwatersheds. The Gini coefficient is a measure of inequality, equal to 0 in the case of perfect equality and with an upper limit of 1 in the case of perfect inequality, where all wealth is concentrated among a small fraction of the population. The Gini coefficient used here indicates the inequality of groundwater recharge from check dams in the synthetic watershed. Lastly, as a way to estimate the risk to downstream systems, we calculated the fraction of years in which the total streamflow in the third order streams was less than 20% of mean annual streamflow under natural conditions (i.e., no check dams). A higher percentage of CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 75

125 100 75 50

Rain, mm 25 0

3 2014−07 2015−01 2015−07 2016−01 2016−07 2017−01

0 2014−07 2015−01 2015−07 2016−01 2016−07 2017−01 Check dam storage, m

150

100

Tank storage, ML storage, Tank 0 2014−07 2015−01 2015−07 2016−01 2016−07 2017−01

Figure 5.4: Daily timeseries of (a) precipitation at Hadonahalli weather station, (b) water storage in the monitored check dam in the Thirumagondonahalli watershed, with the capacity of the check dam represented by the dashed line (small overshoots are expected due to cold water inﬂow in storms), and (c) water storage in Thirumagondonahalli tank.

low-ﬂow years would indicate that water-dependent social or ecological systems would be less resilient to variability in streamﬂow (Brown and Lall, 2006).

5.3 Results

5.3.1 Observed streamflow dynamics Streamflow exhibited sharp rising and falling limbs at the onset and abatement of precipitation, and even in the bigger storms runoff into the tank generally ceased within hours of the end of the storm. The instrumented check dam in the Thirumagondonahalli watershed regularly filled even in some smaller storms. On the other hand, the tank did not fill or overflow during the study period, and most of the inflow to the tank was contained within a few large storms (Figure 5.4). Because the water level sensors were sensitive to variables other than water level (e.g., temperature, electrical conductivity) it was difficult to know precisely when the check dam was overflowing. All of the large inflow events into the Thirumagondonahalli tank occurred when the check dam was at or near capacity, suggesting that the check dam overflowed in CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 76

125 Check dam ● status ● 100 ● Near full Not full 75

50 ●

Tank inflow, ML inflow, Tank ● 25 ● ● ● ● ●● ●● 0 ●●●●●●●● ●● 0 10 20 30 Peak check dam storage, m3

Figure 5.5: Tank inflow versus peak check dam storage for all runoff events into the tank. Tank inflows only occurred when the check dam was full and, presumably, overflowing. these events (Figure 5.5). The partitioning of water captured by check dams favored groundwater recharge (82–86% of captured inflow) over evaporation (14–18%). In this sense the check dams were “efficient” at recharging groundwater.

5.3.2 Effect of check dams in the Thirumagondonahalli watershed The Thirumagondonahalli watershed model produced representative streamflow in the largest storms. Over the three-year study period, 50% of observed inflow was accounted for by the largest 7 inflow events, and 50% of the simulated inflow was accounted for by the 5 largest inflow events. On the other hand, the model reproduced the behavior of small storms poorly. The 130 smallest observed inflow events comprised 10% of observed discharge, compared with the 8 smallest simulated inflow events making up 10% of the simulated discharge. The reasons for this discrepancy most likely have to do with the spatial and temporal variability of precipitation. The rain gauge is unlikely to observe all storms that generate runoff in the catchment, especially small storms that are highly variable in space. Furthermore, short and intense storms that produced observed runoff would be unlikely to produce runoff in the simulation, where rainfall was aggregated to 30 minute time periods. Fortunately, the inability to reproduce these small storms is unlikely to affect the partitioning of the water balance in the simulation, because recharge and evaporative rates are based on the water surface area of the check dam, which is proportional to the square root of storage. The difference between a small and a very small inflow is unlikely to have a significant effect on CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 77

Capacity x1

Percent of P Percent of Q

Capacity x5 Lcheck

Percent of P Echeck

Percent of Q Qtank P−Q Capacity x20

Percent of P Percent of Q

0 25 50 75 100

Figure 5.6: Partitioning of precipitation and runoff into various components of the water balance. The green (P −Q) portion represents water that infiltrates or is otherwise captured on the landscape (94.3% of total precipitation) and the remaining portion of precipitation represents net runoff, Q (5.7% of precipitation, which is unchanged in each of the scenarios). Check dam evaporation Echeck and recharge Lcheck comprise a very small portion of the water balance under current conditions (0.16% of total precipitation), and even if the capacities of existing check dams were increased by a factor of 5 or 20, the net capture of check dams would be less than 1% of the total water balance, although they would increase to approximately 15% of runoff in the latter case.

the check dam surface area, or therefore on loss rates. The effect of check dams can be understood by considering the total recharge and evaporation from check dams as a fraction of precipitation. In other words, we partition precipitation into landcape capture (P − Q) and runoff, which can be further partitioned as Q = Qtank + Echeck + Lcheck, where Q is total runoff into the channel, Qtank the inflow into the tank, Echeck the cumulative evaporation from check dams, and Lcheck the cumulative recharge, or leakage, from check dams (no check dam storage term is needed because the change in storage over long timescales is negligible). From observations, Q is approximately equal to 5.7% of precipitation. Nearly all runoff that flowed into check dams continued downstream as overflow, and check dam recharge and check dam evaporation comprised a very small portion of the water balance (less than 5% of total streamflow and less than 1% of total rainfall). After increasing the storage by a factor of 20, check dams still did not significantly alter the water balance of the catchment, although they did capture a larger fraction of runoff (Figure 5.6).

5.3.3 Features of synthetic check dam networks The statistical distribution of precipitation from the random weather generator was similar to the observed precipitation, with a lower mean value (plus or minus standard deviation) CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 78 of 746 ± 126 mm and shorter upper tail, meaning it underestimated extreme events. The mean runoff ratio was 5.5%, and runoff in the watershed (i.e., any flow that made it to a channel) was 43 ± 27 mm, showing a wide range of variability. On average, when the total check dam capacity was less then 30% of mean annual flow, the check dams captured more runoff than their storage capacity. In this case, an increase to check dam capacity resulted in a greater increase in check dam recharge (Figure 5.7, left). The marginal benefit of additional check dam capacity started decreasing with greater storage capacities, and tapered off significantly for total capacities greater than 80% of mean annual runoff. The lopsided configuration tapered off earlier, due to storage being concentrated on one side of the watershed. The Gini coefficient consistently decreased (increasing equality) with increasing check dam capacity, although the rate of change diminished as check dam capacity increased (Fig- ure 5.7, center). Among the configurations, the coefficient was highest when check dams were concentrated on second order streams, where little benefit was given to the six first- order streams. The configurations in which the Gini coefficient was lowest were the even configuration and 1st order configuration, when storage was distributed among the 1st order watersheds (and even in the 1st order configuration, there was a small amount of storage on the second order watersheds). Increasing storage on first-order watersheds unambiguously increase the equality of groundwater recharge in this synthetic study. On the other hand, increasing check dams storage reduced the amount of discharge to the tank at the outlet of the synthetic watershed. In the natural (no check dams) case, less than 10% of the years were marked as “low-flow” (Figure 5.7, right). As total check dam capacity increased, the fraction of low-flow years increased rapidly in the even and 1st order configurations, although the rate of change tapered off for the 1st order configuration after check dam capacity exceeded 0.8 mean annual flow. The lopsided configuration exhibited the least sensitivity to increasing check dam capacity because one side of the watershed was mostly free of check dams.

5.4 Discussion

Field surveys and monitoring combined with watershed modeling demonstrate that check dams in the TG Halli watershed capture a small percentage of streamflow and have little influence in mediating the water balance. Although these check dams are efficient, both in terms of promoting recharge (opposed to evaporation) and capturing flow (relative to their capacity), they are unlikely to reverse the trend of groundwater decline in the watershed. One of the goals of check dams is to increase agricultural production of the watershed, thereby increasing food supply and improving farmer livelihoods. To understand the relative effect of check dams on agricultural production, we assume that crop production is proportional to evapotranspiration, using the following argument. Bouma et al. (2011) calculated the value of water based on the revenue of crops consuming that water. This connection between crop revenue and water value implies that the total value of crops produced will CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 79

0.8 0.6 0.75 0.6 Even 0.4 0.50 1st order 0.4 2nd order 0.2 Gini coefficient Low flow years flow Low Opposite 0.25 Check dam recharge 0.2 Lopsided 0.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 Check dam capacity Check dam capacity Check dam capacity

Figure 5.7: Simulation results of synthetic check dam networks versus total check dam storage capacity as a fraction of mean annual runoff (Q¯). (a) Check dam recharge as a fraction of Q¯, with dashed line of equality. (b) Gini coefficient, showing greater equality among subcatchments with increasing check dam storage. (c) Fraction of years in which flow does not exceed a low-flow threshold of 0.2Q¯.

increase with the amount of water available. This makes sense when considering that some farms in the TG Halli watershed have been left fallow, but could be cultivated if there were additional water available. Because extensive ground water pumping is likely to recover any groundwater recharge and use that water to supply crops, any recharge would be converted to ET . To understand the relative increase in agricultural production due to check dams, therefore, we need to understand the relative increase in ET . With this objective, we consider the following water balance for the catchment of a tank, over long timescales (> 1 year) and including the aquifer in the storage term: ∆S = P − ET − Q (5.2) ∆t From our observations, we can assume that Q = 0.057P , and that it is entirely captured by the tank at the outlet of the catchment. Inﬂow to the tank can be partition into groundwater recharge (and therefore incorporated into the ∆S/∆t term) and Etank, which is 30% of Q or 1.8% of P , and can be incorporated into the P term. Furthermore, groundwater depletion entails that ∆S/∆t is negative, although the trend of borewells going dry and farmers abandoning their ﬁelds suggests that ∆S/∆t may be approaching zero. This yields a new water balance: ∆S = 0.98P − ET , and noting that ∆S/∆t ≤ 0, (5.3) ∆t ET ≥ 0.98P (5.4)

Summarizing the above arguments, total ET is, roughly, greater than or equal to precipitation, and a change in recharge would produce an equivalent change in evapotranspiration. This implies that a change in recharge equivalent to 1% of precipitation would result in a relative change in evapotranspiration ≤1% and therefore, a relative change in agricultural CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 80 production of ≤1%. Given the limited storage capacities of check dams in the study catchments (˜1% of precipitation), they appear to have a negligible effect on catchment groundwater recharge and total agricultural production. From the perspective of an individual farmers, check dams can make sense to replenish local groundwater supply. Unfortunately, those benefits are unlikely to be shared widely among other farmers or communities not in the direct vicinity of the check dam, contrary to findings from other regions (Renganayaki and Elango, 2013). When the total storage capacity of check dams is small relative to annual precipitation, as is observed in the TG Halli watershed, the synthetic analysis found that the configuration of check dams did not play a significant role in controlling check dam capture and recharge. However, with greater storage capacity of check dams, the configuration played an increasingly important role in mediating the catchment water balance and redistributing groundwater availability. Increasing the storage capacity of check dams had a clear effect on the equality of groundwater recharge, as indicated by the Gini coefficient. Conversely, the cases in which equality of recharge was greatest corresponded to cases with a high percentage of low-flow years. We can define the resilience of a system as being proportional to overall water availability (analogous to Brown and Lall, 2006) and dependent on some minimum discharge threshold. Under this definition, resilience of downstream systems (ecological or agricultural) decreases with increasing check dam capacity. In other words, there was an inverse relationship between groundwater recharge equality and downstream discharge resilience. Although the quantity and total capacity of check dams in the TG Halli watershed is small enough that check dams are unlikely to affect resilience, these characteristics should be considered if total check dam capacity were to approach much higher values. The cumulative quantity of upstream storage capacity in other watersheds has been reported as being similarly low (1-2% of streamflow, Dillon et al., 2009) as well as considerably higher and comparable to total streamflow (Bouma et al., 2011). It is possible that check dams in different hydrogeology or climate would produce different dynamics, and this should be taken into account when considering check dams in other regions.

5.5 Conclusions

Check dams have been promoted in the TG Halli watershed, and indeed in much of India, as a way to promote groundwater recharge and sustainability. Based on an approach that included field observations and watershed simulations, we found that check dams do not play a significant role in water balance partitioning in the TG Halli watershed. Exploratory modeling of synthetic check dam networks found that the total storage capacity and configuration of check dams both matter in terms of water balance partitioning, groundwater recharge equity, and downstream resilience to check dam construction. These findings suggest that check dams would only play a major role in the water balance of the TG Halli if they are constructed in higher numbers with large storage capacities. CHAPTER 5. CHECK DAMS AND GROUNDWATER RECHARGE 81

Because the total quantity of check dams within the TG Halli is currently low and the check dams only capture a small fraction of streamflow, the marginal benefit (increase in groundwater recharge) of new check dams is relatively high, but the net impact of new check dams at the watershed scale is still likely to be negligible. The notion that check dams yield significant benefits beyond farmers in the vicinity of the check dam may be overstated, particularly in the TG Halli watershed. Although check dams have not yielded meaningful benefits in terms of catchment-scale mitigation of groundwater depletion, they may make sense to individual farmers and villages who wish to increase their groundwater recharge, as the efficiency of check dams is high. Other rainwater harvesting approaches, such as landscape capture, could have a greater impact than check dams in mediating the water balance. More research on these practices would be valuable in understanding the cumulative impact of rainwater harvesting initiatives on the water balance and agricultural productivity. Lastly, increasing groundwater recharge can increase the equality access to groundwater supply among farmers, but with the consequence of negatively affecting the resilience of downstream systems. As check dams and other rainwater harvesting strategies are promoted in this region, the tradeoffs between equity and resilience should be considered in more detail. 82

Chapter 6

Conclusion

6.1 Summary of ﬁndings

This dissertation focused on a heavily-modified and poorly-monitored watershed, where understanding of change was limited to hypotheses dispersed among various stakeholders (farmers, tank managers, water managers) and not necessarily grounded in empirical data. These perspectives were useful in shaping a research agenda, but were not necessarily indicative of the processes driving hydrologic change in the watershed. We developed a research framework for understanding historical hydrologic change in data-sparse regions (Figure 1.2), a critical challenge for the field of hydrology (Sivapalan et al., 2012, 2014; Thompson et al., 2013b). We applied this framework to understand spatial heterogeneity of hydrologic change, associated this change most strongly with groundwater irrigated agriculture, identified the physical mechanisms of streamflow generation and decline, and explored the potential of check dams as a mitigation strategy. Chapter 2 required compiling information from stakeholders and agencies, and applied the method of multiple working hypotheses to test the viability of each of five hypotheses regarding the nature of change in the TG Halli watershed. No evidence was found in support of the hypotheses that climatic drivers were forcing the hydrologic change, indicating that anthropogenic activities were driving nonstationarity in the watershed. This initial analysis found that groundwater depletion, expansion of Eucalyptus plantations, and to a lesser degree check dam construction could be driving hydrologic change. To understand heterogeneity of hydrologic change, Chapter 3 focused on reconstructing hydrologic change of tank water extent using remote sensing of Landsat imagery. By modeling water extent in a multiple regression and with precipitation and time as predictor variables, we were able to isolate temporal trends in hydrology from natural variability of precipitation. The temporal trends indicated that drying was strongest in the TG Halli watershed, while streamflow increased downstream of Bangalore. Using land use maps developed for the TG Halli watershed, we found that streamflow decline was most closely associated with irrigated agriculture, which is almost entirely sourced by groundwater. Eucalyptus CHAPTER 6. CONCLUSION 83 land use was somewhat predictive of depletion, but the relationship was not statistically significant when accounting for irrigated crops and any effect of Eucalyptus plantations was secondary to that of direct groundwater abstraction. This remote sensing approach was useful both in terms of identifying a fingerprint of human-induced hydrologic change as well as developing a dataset that could be used to associate change with social and geophysical properties of the system. To develop a mechanistic basis for hydrologic change, Chapter 4 focused on understand contemporary and historical hydrologic processes controlling streamflow generation in two study catchments within the TG Halli watershed. Neither saturation excess runoff nor groundwater discharge were supported by the data. Stable isotope tracers, measurements of soil properties, and analysis of rainfall-soil-streamflow dynamics indicated a prevalence of infiltration excess runoff in the study catchments. Additional evidence indicated the historical presence and subsequent loss of a shallow groundwater, and the elevation of infrastructure indicated that the water table would have been above the elevation of nearby streams. This process-based hydrologic reconstruction supported the hypotheses that a decline in groundwater would result in a reduction of streamflow. To understand a commonly referenced strategy to mitigate or reverse groundwater depletion, Chapter 5 focused on the role of check dams in mediating the water balance and promoting groundwater recharge. Data analysis found that major tank inflows occur when check dams overflow, but watershed simulations indicated this coincidence was a result of large storms generating considerable runoff throughout the watershed. Although check dams were efficient in terms of partitioning water between recharge and evaporation, the total capacity of check dams was small and the net capture by check dams less that 1% of total rainfall. A synthetic analysis found that the configuration of check dams on the network was important, and that check dams could have a considerable effect on downstream flows if constructed in large enough quantities. Check dams could be useful to individual farmers or villages, but as a watershed management strategy have not had much of an effect in the TG Halli watershed.

6.2 Future work

Groundwater depletion appears to be the primary driver of streamflow decline in the Arka- vathy watershed. Considerable groundwater depletion was driven by groundwater irrigated agriculture, but Eucalyptus plantations likely played an ancillary role in reducing groundwater recharge given their effect on hydraulic conductivity and potential for increasing transpiration. Given that baseflow had ceased by the mid-1990s, any subsequent declines would have to be explained by another mechanism. On-site rainwater harvesting strategies, including leveling of plots and bunding along the edges, are likely to reduce infiltration excess overland flow and could have played a role in reducing streamflow. These interventions, in addition to check dams, should be investigated in more detail when analyzing the water balance of groundwater-depleted areas. CHAPTER 6. CONCLUSION 84

It appears unlikely that the Arkavathy will return to its natural (or even preindustrial) hydrologic conditions. Nevertheless, a sustainable management approach should be eventually attainable, if collective action by farmers or effective policy levers can curtail over-abstraction groundwater of irrigation. Education programs have been valuable in shifting public perception about water, albeit in other areas of the water industry (e.g., water recycling, Ormerod and Scott, 2012), and ATREE has worked diligently in building trust with farmers and educating informing them of research findings. Electricity for groundwater pumping is subsidized entirely by the state of Karnataka, such that the cost of pumping groundwater is an externality to farmers, without even considering the externality of groundwater and surface water depletion. Subsidies for electricity could be curtailed or reduced, which would impose monetary consequences to groundwater abstraction and likely reduce depletion. Although such a policy may be politically unfeasible and would be difficult to implement (Shah et al., 2008), it may be possible with sufficient public education regarding the effects of groundwater depletion. Such a policy could be combined with a regulatory approach to conserve agriculture water, monitored using satellite-based estimates of evapotranspiration (Fisher et al., 2017; Semmens et al., 2015). Unfortunately, these solutions do not appear to be currently viable, and creative ideas and public outreach will most likely be required to ultimately achieve sustainable management within the watershed. More broadly, the trajectory of this system was dictated by social institutions and mediated by the hydrogeology. Future work should focus on understanding the characteristics of the human and hydrologic systems that produced such a scenario. In other words, what economic, political, and technological characteristics results in the types of decisions that were made? What geographic and geologic characteristics caused the system to develop in the way it did? And furthermore, are there lessons from the Arkavathy that can be generalized to more broadly indicate the behavior of sociohydrologic systems? Obviously, generalizations must be made cautiously, but nevertheless could be useful in identifying problems before they become severe, as happened in the Arkavathy watershed. The story of the Arkavathy watershed is not finished, and humans will continue to be the driving force shaping the hydrologic behavior of the watershed. As Bangalore continues to grow, it will drive demand for more food and more water. If the system continues without adequate planning, water resources in the Arkavathy will be further strained and transformed. Unfortunately, such a scenario would likely result in further degradation of the hydrologic system and greater inequality with the greatest consequences affecting the poorest stakeholders. Although there is no immediate fix to the challenge of water scarcity in the Arkavathy watershed, strategic policy interventions should seek to incorporate externalities into the cost of groundwater abstraction and adopt policies that promote sustainable agriculture. Even though the system is unlikely to be restored to its natural conditions, care should be taken to ensure the sustainability of this social-ecological system so that it may continue to serve the people who live in the watershed. 85

Appendix A

Supporting information for Chapter 3

A.1 Remote sensing analysis

A.1.1 Remote-sensing images and supplementary data Tracking water storage in the tanks at monthly or higher temporal resolution would be desirable, but is precluded because remotely sensed images from the monsoon season often contain large areas of cloud cover. This analysis therefore focuses on post monsoon images from the months of December and January. We selected 48 Landsat images for classiﬁcation, including 16 acceptable post-monsoon images from 1973 to 2010 for analyzing long-term variability in tank water extent, as described in Section 2.4 (see Fig. A.1 and Table A.1 for details).

Table A.1: Landsat scenes classiﬁed. The path and row numbers refer to WRS-1 for Landsat 1–3 and WRS-2 for Landsat 5–8. “Use” column indicates whether the scene was used for calculation of long-term trends (LTT), dry-season analysis (DSA), or for accuracy assessment (ACC). The ACC images from 2004 through 2009 were used in conjunction with Google Earth images, while the 2014 ACC image was compared with the LISS IV image.

Date Mission Path Row Cloud Free Source Use 1973 January 22 Landsat 1 154 051 Yes USGS LTT,DSA 1973 February 27 Landsat 1 154 051 Yes USGS DSA 1976 January 17 Landsat 2 155 051 Yes USGS LTT 1976 December 05 Landsat 2 154 051 USGS DSA 1976 December 24 Landsat 2 155 051 Yes USGS DSA 1977 January 10 Landsat 2 154 051 Yes USGS LTT,DSA 1977 January 11 Landsat 2 155 051 Yes USGS DSA 1977 January 28 Landsat 2 154 051 Yes USGS DSA 1986 December 15 Landsat 5 144 051 Yes NRSC LTT 1990 January 24 Landsat 5 144 051 Yes NRSC LTT 1990 December 26 Landsat 5 144 051 NRSC LTT Continued on next page APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 86

Date Mission Path Row Cloud Free Source Use 1992 December 31 Landsat 5 144 051 Yes NRSC LTT 1995 January 22 Landsat 5 144 051 Yes NRSC LTT 1999 February 02 Landsat 5 144 051 Yes USGS DSA 1999 February 18 Landsat 5 144 051 Yes USGS DSA 1999 April 07 Landsat 5 144 051 Yes USGS DSA 1999 November 09 Landsat 7 144 051 Yes USGS DSA 2000 March 16 Landsat 7 144 051 Yes USGS DSA 2000 May 03 Landsat 7 144 051 USGS DSA 2001 January 14 Landsat 7 144 051 Yes USGS LTT 2001 October 29 Landsat 7 144 051 USGS DSA 2002 February 18 Landsat 7 144 051 Yes USGS DSA 2002 April 07 Landsat 7 144 051 USGS DSA 2002 December 03 Landsat 7 144 051 Yes USGS LTT,DSA 2003 March 09 Landsat 7 144 051 Yes USGS DSA 2003 December 22 Landsat 7 144 051 Yes USGS LTT 2004 December 08 Landsat 7 144 051 Yes USGS ACC 2005 February 10 Landsat 7 144 051 Yes USGS ACC 2005 December 27 Landsat 7 144 051 USGS LTT,DSA,ACC 2006 January 12 Landsat 7 144 051 Yes USGS DSA 2006 January 28 Landsat 7 144 051 USGS DSA 2006 March 01 Landsat 7 144 051 USGS DSA 2006 December 30 Landsat 7 144 051 Yes USGS LTT 2008 January 02 Landsat 7 144 051 Yes USGS LTT,DSA 2008 January 18 Landsat 7 144 051 Yes USGS DSA 2008 February 19 Landsat 7 144 051 Yes USGS DSA 2008 March 06 Landsat 7 144 051 Yes USGS DSA 2009 January 20 Landsat 7 144 051 Yes USGS LTT 2009 November 28 Landsat 5 144 051 Yes USGS ACC 2010 January 07 Landsat 7 144 051 USGS LTT 2013 November 07 Landsat 8 144 051 USGS DSA 2014 January 10 Landsat 8 144 051 USGS DSA 2014 January 26 Landsat 8 144 051 Yes USGS DSA 2014 February 11 Landsat 8 144 051 Yes USGS DSA 2014 February 27 Landsat 8 144 051 USGS DSA,ACC 2014 March 15 Landsat 8 144 051 Yes USGS DSA 2014 March 31 Landsat 8 144 051 Yes USGS DSA 2014 April 16 Landsat 8 144 051 USGS DSA

The 2014 Landsat imagery was used for remote-sensing validation and dry-season analysis, but was not included in the 1973–2010 study period. Most images were downloaded from Earth Explorer (earthexplorer.usgs.gov), except for ﬁve images from 1986 through 1993, which were purchased from the National Remote Sensing Centre (NRSC, nrsc.gov.in). An APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 87

Scene use Long-term trend Dry-season analysis Accuracy assessment

Figure A.1: Landsat scenes classified in this study (N = 48), with the year corresponding to the date on January 1. Decades are separated by dashed vertical lines using ”monsoon year” (e.g., the January 1990 image is grouped with the 1980s because it corresponds to the 1989 monsoon year). image from the Land Imagery Scan Sensor (LISS-IV) were also purchased from NRSC and used for accuracy assessment. A shapefile of tank boundaries was obtained from the Kar- nataka State Remote Sensing Application Centre (KSRSAC, karnataka.gov.in/ksrsac) to aid in classification of water bodies. Topographic maps completed in the 1970s by the Survey of India (surveyofindia.gov.in) were manually georeferenced and used to verify tank boundaries at the beginning of the study period. Other supplementary datasets were obtained from NASA Reverb (reverb.echo.nasa.gov) and Karnataka State Natural Disaster Monitor- ing Centre (KSNDMC) as listed in Table A.2.

Table A.2: Data sources used in this paper.

Dataset Date Resolution Source Landsat images 1973–2010 & 2014 30 m USGS & NRSC LISS IV image 2014 5 m NRSC Land use map 2001 30 m KSRSAC Tank boundaries - - KSRSAC Topographic maps 1970s - Survey of India Aster DEM - 30 m NASA Reverb Daily Precipitation (62 stations) 1972–2010 0.69 10km2 KSNDMC

NRSC images were manually georeferenced using reference points from the higher-resolution LISS image, with root mean squared error (RMSE) less than 0.5 pixels in all images. All APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 88

Landsat images were cropped to the extent of the Arkavathy watershed and converted to top-of-atmosphere (TOA) reflectance (Chander et al., 2009), which was used for training and classification of all images. Landsat 7 ETM+ scenes acquired after May 31, 2003 contained gaps due to a failure of the Scan Line Corrector (SLC) (Scaramuzza et al., 2005). Although gap-filling techniques for the SLC error generally use successive images to fill missing pixels (e.g., Chen et al., 2011), we used a single-image gap-filling approach because of the inherent temporal variability of tank water extent. We used pixels along the edge of the gap to fill missing pixels similar to Catts et al. (1985) but instead of interpolation, which would cause spectral homogenization in missing pixels, we repeated edge pixels towards the center of the gaps using using successive grayscale dilation (see Fig. A.2).

Figure A.2: Left: Landsat false color composite (FCC) of a tank on 17 December 2005, with missing pixels visible as black diagonal bands. Middle: FCC after missing pixels were ﬁlled using successive grayscale dilation. Right: Classiﬁcation of water in the image shown in blue.

We used cloud-free images where possible, but in some years the only viable post-monsoon image contained some cloud cover. Cloud shadows were particularly troublesome because the spectral reflectance of land in a cloud shadow was often similar to that of water. We applied the fmask algorithm (Zhu and Woodcock, 2012) to identify clouds and cloud shadows, making minor modifications to improve the method for the Bangalore region as follows: (i) we included the filters from the automatic cloud cover assessment algorithm (ACCA, Irish, 2000) when determining the potential cloud pixels, which reduced false positives for clouds in urban areas, and (ii) we removed clouds whose height (determined with fmask) was an outlier. This approach was possible because the topography was relatively flat and the selected images contained only cumulus clouds which exhibit relatively consistent base height at the lifting condensation level (Craven et al., 2002). Outliers were determined as clouds with a height less than H25 − 1.5(H75 − H25) or greater than H75 + 1.5(H75 − H25) where H25 and H75 are the first and third quartiles of cloud height and H75 − H25 is the interquartile range. This procedure helped prevent erroneous classification of cold, white land pixels as clouds and limited the potential for erroneous classification of water bodies as shadows (see Fig. A.3). APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 89

Figure A.3: Left: Landsat false color composite on 27 December 2005 showing clouds and shadows near multiple tanks, with missing pixels visible as black diagonal bands. Middle: FCC after missing pixels were filled using successive grayscale dilation. Tank boundaries are shown in light gray. Right: Classification of clouds (yellow), cloud shadows (gray), water in tanks (blue), and areas classified as water but removed from the analysis due to clouds or cloud shadows (red outline).

A.1.2 Classification method The tank water classification method relied on separating pixels containing water from pixels containing land in a spatial region defined by the mapped tank boundaries. Water stored in tanks in the Arkavathy watershed varied from clear (with low reflectance in all Landsat reflectance bands) to turbid (more reflective in the visible (Moore, 1980) and NIR bands (Whitlock et al., 1981)). Turbid water exhibited its highest reflectance in the red band due to the red soils in the Arkavathy watershed (Novo et al., 1989). Land cover surrounding wetted areas of tanks included vegetation, bare soil, and built-up urban land. We grouped these classes into a single land class, which was characterized by high reflectance in the NIR band and lower reflectance in visible bands (McFeeters, 1996). These characteristics primarily distinguish land from water in the Arkavathy, which has low reflectance in the NIR band and either low reflectance in the green band (clear water) or high reflectance in the red band (turbid water). We developed an automated classification algorithm that distinguished areas of clear water and turbid water from land in each pixel, allowing rapid and consistent classification approach across images and Landsat sensors. We used a two-stage approach for estimating water extent in tanks. First, pixels having definitive spectral properties of water were identified and classified as “apparent” water pixels. Second, spectral unmixing was used to estimate the water fraction in all pixels within 60 m of any apparent water pixels. A conceptual representation of this algorithm is provided in Figure A.4, and the steps described below are cross referenced to the numbered panels in the figure. The only user input to the classification algorithm for each scene was to select a reservoir containing clear water with which to train the image (Fig. A.4, step 1). The Normalized Difference Water Index by McFeeters (1996), NDWI = (green - NIR) / (green + NIR), APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 90

1: Select training reservoir. Calculate NDWI

2: Otsu method for NDWI 3a: Water pixels -- NDWI threshold at training threshold for clear water reservoir. Calculate spectral means for land 3b: Identify apparent clear & clear water water pixels. Dilate two pixels

NDWI 3c: Spectral unmixing for clear water fraction with 4a: Land pixels -- NIR & red green, NIR bands thresholds for turbid water x NIR

Green

Red NIR 1 4b: Identify apparent turbid 0 x water pixels. Dilate two pixels

4c: Spectral unmixing for turbid water fraction with red, NIR bands

5: Water fraction for each 6: Tank water extent is pixel is max of clear fraction sum of water fraction and turbid fraction of pixels in tank

7: Set flags for clouds/shadows, MSS NA, SLC missing pixels. Remove corresponding tanks

Figure A.4: Flowchart of classification method. In Steps 3 and 4, clear water fraction and turbid water fraction are each calculated for all pixels in the image before they are combined into water fraction in Step 5. Color images are from Landsat, with red, green, and blue in the image corresponding to NIR, Red, and Green bands from Landsat TM. APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 91 reveals a clear distinction between land and water pixels. In each image, we divided pixels within the training reservoir (or a rectangular window of pixels around the training reservoir if the reservoir was mostly full) into water or land classes using Otsu’s method (Otsu, 1979), which clusters grayscale pixels into two classes by minimizing the within-class variance. The water and land pixels at the training reservoir were used to calculate the spectral means of land pixels and clear water pixels (step 2). The minimum NDWI of water pixels at the training reservoir (step 3a) was used as a threshold to create a mask of apparent clear water for the entire scene (step 3b) which was then dilated using a 5x5 square kernel (a 3x3 kernel for MSS scenes). All pixels within the dilated mask were transformed to a single component,x ˆ, parallel to the transect between the spectral means of clear water and land in the 2-dimensional space of NIR and green reflectance (step 3c). Pixels falling between thex ˆ means of clear water and land were assigned a clear water fraction. Clear water fraction was set to 1 in pixels at or below the clear waterx ˆ mean, and linearly decreased to 0 for pixels at or above the landx ˆ mean. A similar procedure of masking, dilating, and unmixing was performed for turbid water, with minor changes. The criteria for apparent turbid water pixels were determined from land pixels near the training reservoir as the 98th percentile of red reflectance and the 98th percentile of NDWI (step 4a), provided that red reflectance was greater than NIR reflectance. Pixels meeting these criteria were included in the turbid water mask and dilated to include the surrounding area (step 4b). Spectral unmixing was conducted similarly to clear water, except the component for unmixing,y ˆ, was taken along the transect between the spectral means of turbid water and land in the NIR-red space (step 4c). Finally, the water area in each pixel was taken as the higher value of clear water area and turbid water area (step 5). Tank water extent was calculated as the sum of water area of all pixels within two pixels of the mapped tank boundary (step 6). Examples of raw and classified images for a small tank (≈ 25 ha) and a large tank (≈ 160 ha), resulting timeseries of tank water extents, and the relationship between tank water extents and the preceding monsoon rainfall characteristics are shown in Figs. A.5 and A.6. We did not estimate the area of water in any tank that was flagged for the following quality concern criteria: (i) spatial overlap or adjacency of dry tank boundary or wetted tank area with clouds or cloud shadows, (ii) spatial overlap of greater than 25% of dry or wet tank area with missing pixels due to the SLC error in Landsat 7 images, or (iii) greater than 25% spatial overlap of dry or wet tank area with the edge of the scene from MSS images (step 7). In each of these cases, the tank area was recorded as “NA”. Remote sensing and spatial processing were scripted in R (R Core Team, 2016) using the raster (Hijmans, 2015), rgeos (Bivand and Rundel, 2016), sp (Pebesma and Bivand, 2005), and rgdal (Bivand et al., 2016) packages, as well as ggplot (Wickham, 2009) for plotting. Watershed delineation and extraction of the cascading tank network were completed in GRASS GIS (GRASS Development Team, 2016). APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 92

25 Image 20 shown 15 10 Yes 5 No

Water extent, ha extent, Water 0 1980 1990 2000 2010 Monsoon Year 1972 1975 1994 2008 25 25 20 20 15 15 10 10 5 5

Water extent, ha extent, Water 0 ha extent, Water 0 200 300 400 500 600 20 25 30 35 40 Ptotal (mm) Pextreme (mm)

Figure A.5: Left, top: Timeseries of post-monsoon tank water extent with selected Landsat images. Left, bottom: post-monsoon water extent versus monsoon season precipitation (Ptotal and Pextreme). Right: Landsat images (NIR-red-green mapped to red-green-blue) and corresponding classiﬁed water fraction.

160 Image 120 shown 80 Yes 40 No Water extent, ha extent, Water 1980 1990 2000 2010 Monsoon Year 1972 1986 2002 2007 160 160 120 120 80 80 40 40 Water extent, ha extent, Water ha extent, Water 200 400 600 20 30 40 50 60 Ptotal (mm) Pextreme (mm)

Figure A.6: Left, top: Timeseries of post-monsson tank water extent with selected Landsat images. Left, bottom: post-monsoon water extent versus monsoon season precipitation (Ptotal and Pextreme). Right: Landsat images (NIR-red-green mapped to red-green-blue) and corresponding classiﬁed water fraction.

A.1.3 Validation of classification method To validate the classification results, we used a 5 m resolution LISS IV satellite image from 26 February 2014 to compare with a classified Landsat image from 27 February 2014. The LISS IV image was classified in ENVI software (Harris Geospatial Solutions Inc.) using support vector machine (SVM) classification with four land classes and four water classes. After classification, the water classes were merged into a single water class and resampled to the resolution of Landsat so that the resulting grayscale classification contained a water fraction in the range [0,1] for each pixel. At the pixel level, a traditional confusion matrix is inappropriate for continuous classification data (Congalton and Green, 2009). Thus, we evaluated the error (Landsat water fraction APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 93

minus reference water fraction) in all Error (Landsat - LISS water fraction) pixels within tanks by binning the pixel a Underclassified Correct Overclassified error into categories representing under- Bins (-0.6,-0.2][-1,-0.6](-0.2,0.2] (0.2,0.6] (0.6,1] Total classified (-1 to -0.2), correct (-0.2 to [0,0.01) 0 0 99.5% 0.5 0.1 444793 [0.01,0.2) 0 0 81.9% 17.8 0.3 5133 0.2) and over-classified (0.2 to 1). We [0.2,0.4) 0 42.9 40.8% 16.3 0 2597 further separated pixels into groups by [0.4,0.6) 0 33.5 57.2% 9.3 0 1982 binning the producer (reference) wa- [0.6,0.8) 18.2 14.7 64.2% 2.9 0 1747 ter fraction and user (Landsat) wa- [0.8,0.99) 11.4 27.9 60.7% 0 0 2258 LISS water fraction LISS water ter fraction. We calculated producer’s [0.99,1] 5.7 10.3 84% 0 0 8681 and user’s accuracy for each water frac- Total 1066 3561 458586 3725 253 467191 tion bin to form both a producer er- Bins (-0.6,-0.2][-1,-0.6](-0.2,0.2] (0.2,0.6] (0.6,1] Total ror matrix and consumer error matrix [0,0.01) 0.2 0.4 99.4% 0 0 443774 (Fig A.7). Producer accuracy was 84% [0.01,0.2) 0.9 2.4 96.7% 0 0 5196 [0.2,0.4) 1 5.7 33.8% 59.5 0 3752 for wet pixels and 99% for dry pix- [0.4,0.6) 0 15.9 46.4% 37.8 0 2767 els, and because of the high number [0.6,0.8) 0 37.6 49.1% 10.6 2.7 3244 of dry pixels the overall accuracy was [0.8,0.99) 0 0 95.8% 1.4 2.9 5750 Landsat water fraction Landsat water 98%. Pixels containing a mix of wa- [0.99,1] 0 0 99.1% 0.9 0 2708 ter and land (20–80% water) had lower Total 1066 3561 458586 3725 253 467191 b 5000 producer accuracy (41–82%). Overall, N = 1066 N = 3561 N = 3725 N = 253 the classification errors were unbiased 4000

and the histogram of classiﬁcation er- 3000 rors (excluding pixels with zero error) 2000

was approximately normally distributed Count (# pixels) (Figure A.7b). 1000 We also used Digital Globe imagery 0 −1.0−0.6−0.2 0.2 0.6 1.0 available from Google Earth (Google Error (Landsat - LISS water fraction) Earth, 2016) to assess the validity of the classification in normal (680–955 mm) Figure A.7: (a) Pixel-level producer and consumer ac- and wet (>955 mm) precipitation years curacy tables, given by percent of pixels in each error during the study period. Given the lim- bin. Pixels are grouped into rows by the producer or consumer water fraction and then binned into columns ited availability of these images, we were by the error (Landsat - LISS water fraction). (b) His- unable to find a dry-year image (<680 togram of non-zero classification errors (excluding pix- mm) within the study period that was els with error = 0). suitable for comparison with a mostly cloud-free Landsat image. We manually delineated 18 tanks in the normal year (2009) and 34 tanks in wet years (2004 and 2005), and compared the manual delineation with classification of Landsat images from the same time period using a linear regression (Fig. A.8). The long-term trajectory of reservoirs in the Arkavathy watershed is widely known. TG Halli and Hesaraghatta reservoirs decreased over time. Manchanabele was constructed in 1993, and Harobele in 2004. These general trends are reproduced by the classification (Fig- ure A.9). APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 94

80 ●

Monsoon 60 group ● Normal ● Wet

40 Landsat date ● ● 2004−12−08 ● 2005−02−10 ● ● ● ● 2005−12−27 ● Landsat water extent, ha extent, Landsat water 20 ● ● ● 2009−11−28

●●● ●●● ●● ●●● 0 ●● ●

0 20 40 60 80 Google Earth water extent, ha

Figure A.8: Comparison of automatically classiﬁed tanks using Landsat and manually delineated tanks using Google Earth. The four Landsat images correspond to Digital Globe imagery collected on 2004-12-04, 2005-02-09, 2005-12-07 (wet year), and 2009-12-07 (normal year), respectively.

600

TG Halli 400 Hesaraghatta Manchanabele Harobele 200 Reservoir extent (ha) Reservoir extent

0 1970 1980 1990 2000 2010 Year

Figure A.9: Water extent in reservoirs from the automated classiﬁcation, with best ﬁt trend lines. The trends reproduce characteristic behavior of the reservoirs. APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 95

A.2 Statistical model design

A.2.1 Dry season analysis

a b

Figure A.10: Dry season analysis. (a) Post-monsoon drying of all tanks in each subwatershed, relative to the water extent at start of the dry season. (b) Histogram of the remaining water at the end of dry season as a fraction of the start of the dry season for all tank clusters. (c) Confidence intervals on the Mann–Kendall test statistic (tau) for a trend in the rate of tank water loss in dry season. Most subwatersheds do not exhibit a statistically significant trend in the rate of dry-season water loss (the confidence intervals include zero), but the Hesaraghatta and TG Halli subwatersheds exhibit a significant decreasing trend, meaning that tanks dry at a slower rate now than in the past.

A.2.2 Collinearity analysis To check that the estimates of the model effects were not substantially affected by correlation 1/(2df) among the covariates, we calculated the Generalized Variance Inflation Factor√ (GV IF ) for each of the covariates (Fox and Monette, 1992). This factor is analogous to VIF , which is the effect of collinearity on the confidence intervals of each covariate coefficient — it has a lower limit of 1 (no effect), and values less than 2 (a doubling of the confidence intervals) give reasonable assurance that multicollinearity does not greatly affect the confidence intervals 1/(2df) (Fox, 2008). This factor for the time (Y eari) predictor was calculated as GV IF = 1.01, indicating that collinearity has a negligible effect on the estimation of B1,j. Although multicollinearity among other variables was not a concern (we were most interested in confidence APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 96 intervals around the non-precipitation-related time trend), the calculated index was nevertheless reasonable (GV IF 1/(2df) < 1.72) for all other variables. Precipitation trends were also computed independently for each watershed and tank cluster for the period of analysis, and their significance assessed using a non-parametric Mann-Kendall test.

A.3 Statistical model analyses

A.3.1 Precipitation timeseries

1200 ● ● ●

1000 ●

● ● ● ● In analysis 800 ● Other years ● ●

Annual precipitation, mm Annual ● 600 ● ● ● ● ● 1970 1980 1990 2000 2010 Year

Figure A.11: Annual precipitation in the Arkavathy watershed over the course of the study period, as an average of annual precipitation from the 62 rain gauges. Mean annual precipitation is 820 mm, with a standard deviation of 180 mm. There is no statistically significant trend in precipitation when considering precipitation from all years, nor is there is a statistically significant trend when considering only the years from the the analysis (in both cases, the 95% confidence interval of the trend includes zero). APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 97

Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 700 p=0.39 p=0.44 p=0.39 p=0.34 p=0.44 p=0.39 ● ● 600 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 500 ● ● ● ● ● ● ● ● ● ● ● ● ● ● 400 ● ● ● ● ●● ● ●● ●● ● ● ● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● 300 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 200 ● ● ● ●● ● ● ● ●

Cluster 7 Cluster 8 Cluster 9 Cluster 10 Cluster 11 Cluster 12 700 p=0.44 p=0.56 p=0.50 p=0.50 p=0.44 p=0.50 600 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 500 ● ● ● ● ●

400 ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 300 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 200 ● ● ● ● ●

Cluster 13 Cluster 14 Cluster 15 Cluster 16 Cluster 17 Cluster 18

700 ● p=0.44 p=0.34 p=0.34 p=0.39 p=0.44 ● p=0.56 600 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 500 ● ● ● ● ● ● ● ● ● ● ● ●

400 ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 300 ● ● ● ● ● ● ● ● ● ● ● ● ● ● 200 ● ● ● ● ● ● ● ● ● ● ● ●

Cluster 19 Cluster 20 Cluster 21 Cluster 22 Cluster 23 Cluster 24

700 ● ● p=0.50 ● p=0.39 ● p=0.50 ● p=0.50 p=0.50 p=0.30 600 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 500 ● ● ● ● ● ● ● ● ● ●

total,ij 400 ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● P ● ● ● ● ● ● ● ● ● ● ●● 300 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 200 ● ● ● ● ● ●

Cluster 25 Cluster 26 Cluster 27 Cluster 28 Cluster 29 Cluster 30

700 ● ● ● p=0.26 p=0.44 p=0.34 ● p=0.26 ● p=0.44 p=0.39 600 ● ● ● ● ● ● ● ● ● 500 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

400 ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● 300 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 200 ● ● ● ● ● ●

Cluster 31 Cluster 32 Cluster 33 Cluster 34 Cluster 35 Cluster 36 700 p=0.34 ● p=0.30 ● p=0.26 ● p=0.39 ● p=0.26 ● p=0.26 ● 600 ● ● ● ● ● ● ● ● ● ● ● ● 500 ● ● ● ● ● ● ● ● ● ● ● ●

400 ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● 300 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 200 ● ● ● ● ● ●

Cluster 37 Cluster 38 Cluster 39 Cluster 40 Cluster 41 Cluster 42

700 ● ● p=0.30 p=0.30 p=0.34 ● p=0.26 ● p=0.30 ● p=0.39 ● 600 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 500 ● ● ● ● ● ● ● 400 ● ● ●● ●● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ●● ● ● 300 ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 200 ● ● ● ● ● ● 1980 2010 1980 2010 1980 2010 1980 2010 1980 2010 1980 2010 Year

Figure A.12: Total precipitation metric (Ptotal,ij) as calculated for each tank cluster and year over the study period. None of precipitation timeseries exhibited a statistically signiﬁcant trend over time, as shown by the Mann-Kendall p-values in each plot (>0.05 in call cases). The extreme precipitation metric (Pextreme,ij) similarly did not exhibit any statistically signiﬁcant trends in any of the clusters. APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 98

Figure A.13: Quantile-quantile plot of residuals from multiple regression, with residuals normalized by mean and standard deviation and plotted against a theoretical normal distribution. APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 99

Figure A.14: Subwatershed names and cluster IDs used in the multiple regression. These identiﬁers are needed to associate the results in Table A.3 with their spatial locations, shown in this ﬁgure. The Manchanabele and Harobele subwatersheds here are named for reservoirs within the watershed, which are not located at the subwatershed outlet. APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 100

Table A.3: Results from multiple regression for all variables. The effects that apply at the subwatershed level are reported as directly output from the model. The temporal trend of each cluster is reported as percent change in water extent per year, relative to the max water extent recorded for the cluster (last column). To convert to ha per decade per 10 km2 of catchment (as in Figure A.16), the coefficient estimate was multiplied by the max cluster extent and divided by the watershed area. The SS column indicates whether or not the predictor coefficient is statistically significant.

Watershed Area (km2) Coeﬀ Estimate 95% Conf. Interval Max extent (ha) Hesaraghatta 601 C1,k 0.0030 [ 0.0011, 0.0048] - Kumudavathy 441 C1,k 0.0048 [ 0.0031, 0.0065] - TG Halli East 406 C1,k 0.0053 [ 0.0032, 0.0073] - Vrishabhavati 558 C1,k 0.0026 [ 0.0013, 0.0040] - Manchanabele 519 C1,k 0.0052 [ 0.0031, 0.0074] - Suvarnamukhi 315 C1,k 0.0029 [ 0.0011, 0.0048] - Kanakapura 466 C1,k 0.0027 [ 0.0007, 0.0048] - Harobele 855 C1,k 0.0037 [ 0.0018, 0.0057] - Hesaraghatta 601 C2,k 0.0824 [ 0.0192, 0.1456] - Kumudavathy 441 C2,k 0.0635 [ 0.0065, 0.1206] - TG Halli East 406 C2,k 0.0563 [-0.0117, 0.1243] - Vrishabhavati 558 C2,k 0.0960 [ 0.0509, 0.1412] - Manchanabele 519 C2,k 0.0345 [-0.0293, 0.0983] - Suvarnamukhi 315 C2,k 0.0855 [ 0.0277, 0.1433] - Kanakapura 466 C2,k 0.0728 [ 0.0025, 0.1431] - Harobele 855 C2,k 0.0840 [ 0.0295, 0.1385] - Hesaraghatta 601 C3,k -0.0062 [-0.0145, 0.0021] - Kumudavathy 441 C3,k -0.0187 [-0.0276,-0.0097] - TG Halli East 406 C3,k -0.0169 [-0.0277,-0.0062] - Vrishabhavati 558 C3,k -0.0042 [-0.0119, 0.0034] - Manchanabele 519 C3,k 0.0046 [-0.0078, 0.0170] - Suvarnamukhi 315 C3,k -0.0023 [-0.0127, 0.0080] - Kanakapura 466 C3,k 0.0081 [-0.0033, 0.0196] - Harobele 855 C3,k -0.0021 [-0.0122, 0.0080] - Cluster 1 78 B1,j -1.5124 [-2.3287,-0.6962] 441.25 Cluster 2 160 B1,j -1.0844 [-1.8635,-0.3053] 849.61 Cluster 3 34 B1,j -1.3939 [-2.1911,-0.5966] 145.72 Cluster 4 56 B1,j -1.4703 [-2.3559,-0.5848] 230.09 Cluster 5 178 B1,j -0.6510 [-1.3507, 0.0486] 829.00 Cluster 6 31 B1,j -1.2561 [-1.9116,-0.6006] 122.81 Cluster 7 64 B1,j -0.9712 [-1.6454,-0.2970] 126.88 Cluster 8 97 B1,j -1.0577 [-1.7169,-0.3985] 558.47 Cluster 9 80 B1,j -0.5055 [-1.1515, 0.1404] 297.76 Cluster 10 87 B1,j -0.2324 [-0.8429, 0.3781] 241.72 Cluster 11 42 B1,j -0.6764 [-1.3379,-0.0150] 96.40 Cluster 12 70 B1,j 0.3645 [-0.2569, 0.9860] 105.45 Continued on next page APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 101

Watershed Area (km2) Coeﬀ Estimate 95% Conf. Interval Max extent (ha) Cluster 13 65 B1,j 0.1950 [-0.3439, 0.7340] 80.60 Cluster 14 92 B1,j -0.9556 [-1.5062,-0.4049] 234.53 Cluster 15 143 B1,j -0.1969 [-0.7689, 0.3751] 349.95 Cluster 16 100 B1,j -0.0431 [-0.6123, 0.5260] 102.28 Cluster 17 70 B1,j 0.3364 [-0.1944, 0.8672] 156.83 Cluster 18 95 B1,j -0.1915 [-0.5556, 0.1726] 95.42 Cluster 19 73 B1,j 0.7183 [ 0.0552, 1.3815] 183.03 Cluster 20 78 B1,j 0.6817 [ 0.1575, 1.2059] 150.55 Cluster 21 49 B1,j 0.1623 [-0.6215, 0.9461] 84.40 Cluster 22 66 B1,j 0.0728 [-0.5888, 0.7343] 203.93 Cluster 23 88 B1,j 0.9075 [ 0.2841, 1.5309] 433.71 Cluster 24 46 B1,j 0.2055 [-0.3247, 0.7356] 73.66 Cluster 25 64 B1,j -0.2466 [-0.8272, 0.3339] 128.42 Cluster 26 222 B1,j 0.1375 [-0.4092, 0.6842] 90.15 Cluster 27 159 B1,j -0.0646 [-0.6181, 0.4889] 154.36 Cluster 28 138 B1,j -0.5417 [-1.5775, 0.4940] 151.75 Cluster 29 73 B1,j 1.1434 [ 0.5227, 1.7641] 138.12 Cluster 30 85 B1,j 0.2577 [-0.3536, 0.8689] 182.95 Cluster 31 49 B1,j 1.0800 [-0.1694, 2.3295] 138.25 Cluster 32 74 B1,j -0.0363 [-0.7226, 0.6499] 230.90 Cluster 33 34 B1,j 0.1186 [-0.4931, 0.7303] 44.89 Cluster 34 132 B1,j 0.7047 [-0.2094, 1.6188] 120.77 Cluster 35 54 B1,j -0.3459 [-1.0664, 0.3745] 113.29 Cluster 36 179 B1,j 0.0850 [-1.0226, 1.1927] 134.09 Cluster 37 101 B1,j -0.2756 [-1.0543, 0.5031] 243.98 Cluster 38 149 B1,j -0.0055 [-0.6497, 0.6387] 153.70 Cluster 39 182 B1,j 0.1862 [-0.4335, 0.8060] 121.96 Cluster 40 92 B1,j -0.7117 [-1.6038, 0.1805] 227.21 Cluster 41 202 B1,j -0.9949 [-2.7348, 0.7449] 118.67 Cluster 42 231 B1,j 0.2872 [-0.3407, 0.9150] 102.30 APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 102

30 ● )

25 ) extreme,ij ● P

( ● ● ● ●● ●

⋅ sd ● ● ● ● ● ● ● 2 ,k ● ● ●● 20 ● ● ● + C ● ● ● ● ● ● ) ● ● ● ● ● ● ● ● ● ● total,ij ● P % change cluster area ( ( 15 ● ⋅ sd 1 ,k C

●

10 ●

−50−25 0 25 B 1,j ⋅timeperiod (% change cluster area)

Figure A.15: Variability in tank water extent due to precipitation is comparable with changes due to the temporal trend B1,j over the time period of the study, 38 years. In other words, normal variability in precipitation equivalent to the standard deviation (180 mm for annual precipitation) would lead to changes in cluster water extent of about 15–25% of the maximum cluster water extent. Mean annual precipitation regularly varies between 680 and 960 mm, the lower and upper quartiles, respectively. By comparison, the eﬀect of long-term hydrologic changes led to changes of >25% of the maximum cluster water extent in some clusters, although the trend is not signiﬁcant in all of these clusters. APPENDIX A. SUPPORTING INFORMATION FOR CHAPTER 3 103

A.4 Hydrological trends and agricultural land use

● ● ● ● ● ● 0 ● 0 ● ● ● ● ●

● ● −2 ● ● −2 ● ● ● ● ● ● 1 ,j 1 ,j B −4 B −4

● ●

● ● ● ● −6 ● ● −6 ● ●

−8 −8 ● ● 0.050 0.075 0.100 0.03 0.06 0.09 0.12 A irrigated,j A Eucs,j

Figure A.16: Temporal trend (B1,j) versus time-averaged agricultural land use fraction of irrigated 2 2 crops (Airrigated,j,R = 0.66) and Eucalyptus plantations (AEucs,j,R = 0.38) for the 17 tank clusters within the TG Halli watershed (Clusters 1–17 in Fig. A.14). Both categories of land use are negatively correlated with the the temporal trend parameter, B1,j. 104

Appendix B

Supporting information for Chapter 4

This supporting information provides additional details regarding certain portions of the ﬁeld experimentation and observations presented in the main article.

(a) (b) 10 km05

0 1 2 km

(c)

Figure B.1: Location maps. (a) Locations of saturated hydraulic conductivity (Ksat) measurements in green. (b) Locations of open wells from the well survey, including dry wells in red and managed wells containing water in gray. (c) Locations of borewell drill logs from the 1970s, referenced in the discussion section in the main text. 105

Appendix C

Supporting information for Chapter 5

C.1 Check dam water balance

The water balance for a check dam is as follows: ∆S = Q − Q − E − L (C.1) ∆t in out check check

where, ∆S is the change in storage of the check dam over time period ∆t, Qin is the inflow generated from catchment, Qout any overflow, Echeck is the loss to evaporation, and Lcheck is the leakage or groundwater recharge from the check dam, as water percolates through the bottom or side walls of the dam. During rain events, the rates of inflow and and change in storage are much greater than Echeck or Lcheck, and net inflow can be calculated as the change in storage (Qin − Qout = ∆S/∆t). The inflow Qin can be calculated if check dam storage is less than capacity, but in any event where storage approaches capacity, only net inflow (Qin − Qout) can be calculated.

C.1.1 Evaporation The open water evaporation from the check dams was estimated using the Shuttleworth equation (Equation C.2, Shuttleworth, 2007), a modiﬁed form of Penman equation, in metric units. Evaporation was calculated for 30 minute time periods and aggregated to daily evaporative loss. The weather data were obtained from weather stations in SM Golahalli and Thirumagondonahalli study catchments.

∆A0 + γ[6.43(1 + 0.536U )D] E = m (mm day-1) (C.2) λ(∆ + γ)

In the above equation, A0 is the measured or estimated energy available for evaporation from the free water surface (MJ m-2 day-1), ∆ is the rate of change in saturated vapour ◦ -1 ◦ -1 pressure (kPa C ), γ is the Psychrometric constant (kPa C ), Um is the wind speed APPENDIX C. SUPPORTING INFORMATION FOR CHAPTER 5 106

(ms-1), D is the vapour pressure deﬁcit (kPa), and λ is the latent heat of vaporisation of water (MJ kg-1). The Equation (C.2) implicitly includes division of numerator by density of water (1000 kg m-3) to obtain evaporation in mm per day.

C.1.2 Recharge We estimated check dam recharge by calculated the percolation rates of three check dams through the monsoon of 2014 (July–December). On dry days when there was no inﬂow or overﬂow events, percolation Lcheck from check dam was calculated a using simple water balance approach (Glendenning and Vervoort, 2010; Sharda et al., 2006):

For dry days, L = −(E + ∆S) (C.3)

We defined a dry day as one that satisfies all the following conditions: (1) the change in storage is negative (∆S ¡ 0), (2) the measured storage in the check dam was less than its storage capacity (this issue occasionally occurred after storms, likely due to temperature sensitivity of the water level sensors), and (3) the absolute value of change in storage is greater than evaporation (abs(∆S) ¿ E). Check dam evaporation from the Eq. C.2 using weather data from the nearest weather station was aggregated to daily timescales, and the change in storage storage (∆S) was calculated as the difference in volume of water stored in a check dam at 23:30 hours over consecutive days. We calculated the volumetric percolation for dry days using Eq. C.3 and then divided the volumetric daily percolation by daily average water spread area to approximate a vertical percolation rate. Percolation rates averaged 0.02 m per day for all three check dams. The rates were within the range of what was obtained by measuring percolation directly using a infiltrometer in different locations in the study catchments (see Chaper 4). The percolation rates were fairly consistent throughout the monsoon season. The water balance of check dams can be calculated from the estimates of evaporation and recharge. Check dams were largely efficient in terms of effectively partitioning a majority of captured inflow to groundwater recharge. The percentage of captured water that was evaporated was fairly small (14–18%) and the percentage recharged quite high (82–86%).

C.2 Precipitation variability

To estimate the eﬀect of spatial variability of precipitation, we compared storm event precipitation from the Hadonahalli rain gauge with data from a nearby rain gauge at Tubagere (2.7 km away and approximately due east) operated the Karnataka State Natural Disaster Management Cetnre (KSNDMC). Total rainfall in large events was fairly consistent across these two gauges, but precipitation was spatially variable with higher percent variation for smaller storms (Figure C.1). APPENDIX C. SUPPORTING INFORMATION FOR CHAPTER 5 107

1.00 ● ●

●

0.75

0.50 Percent difference Percent

0.25

0.00

1 10 100 Precipitation, mm

Figure C.1: Precipitation comparison between the Hadonahalli rain gauge and data from KSNDMC rain gauge at Tubagere, which is 2.7 km away and approximately due east. The horizontal axis is the maximum event precipitation between the two gauges, and the vertical axis the percent diﬀerence relative to the maximum. The boxplots represent the distribution of percent diﬀerence over bins of equal size. Each point represents a single storm event.

The effect of precipitation variability likely influenced measurements of some of the largest storms. In one storm on 24 August 2014, inflow to the tank was about 125 ML, even though both rain gauges measured about 25 mm of rain which is equivalent to only 75 ML if average over the catchment. The largest precipitation events on 06 September 2015 contained two precipitation peaks, the first containing 100 mm of rain in the second containing 25 mm of rain. After the first peak there was a large amount of inflow into Thirumagondonahalli tank, but the second rainfall peak brought no inflow to the tank. 108

References

Aeron Systems (2017), Wireless Weather Station with New Wireless Data Logger, Aeron Systems Pvt. Ltd., Pune, India. Aeschbach-Hertig, W., and T. Gleeson (2012), Regional strategies for the accelerat- ing global problem of groundwater depletion, Nature Geoscience, 5 (12), 853–861, doi: 10.1038/ngeo1617. Agoramoorthy, G. (2008), Can India meet the increasing food demand by 2020?, Futures, 40 (5), 503–506, doi:10.1016/j.futures.2007.10.008. Agoramoorthy, G., and M. J. Hsu (2008), Small size, Big Potential: Check Dams For Sus- tainable Development, Environment: Science and Policy for Sustainable Development, 50 (4), 22–35, doi:10.3200/ENVT.50.4.22-35. Allen, R. G., L. S. Pereira, D. Raes, M. Smith, and Others (1998), Crop evapotranspiration - Guidelines for computing crop water requirements - FAO Irrigation and drainage paper 56, FAO, Rome, 300. Allison, G. (1988), A review of some of the physical, chemical and isotopic techniques available for estimating groundwater recharge, Estimation of natural groundwater recharge, pp. 49–72. Anand, P. B. (2004), Water and Identity: An analysis of the Cauvery River water dispute., University of Bradford, pp. 1–41. Arkavathi Rejuvenation (2017), Arkavathi Rejuvenation Group Page, https://www. facebook.com/arkavathi.rejuvenation. Arora, M., N. K. Goel, and P. Singh (2005), Evaluation of temperature trends over India, Hydrological sciences journal, 50 (1). Arpita, R. (2017), Tanks, borewells in Kumudavathi basin recharged now, The Times of India, pp. Nelamangala, June 19, 2017. Arrigoni, A. S., M. C. Greenwood, and J. N. Moore (2010), Relative impact of anthropogenic modiﬁcations versus climate change on the natural ﬂow regimes of rivers in the Northern Rocky Mountains, United States, Water Resources Research, 46 (12). REFERENCES 109

Arzberger, P., P. Schroeder, A. Beaulieu, G. Bowker, K. Casey, L. Laaksonen, D. Moorman, P. Uhlir, P. Wouters, and Others (2004), An international framework to promote access to data, Science, pp. 1777–1778.

ATREE, V. Srinivasan, and S. Lele (2015), Forum with traditional watermen (Neerghantis) in the upper Arkavathy sub-basin.

Baker, V. R. (2006), Palaeoﬂood hydrology in a global context, CATENA, 66 (1-2), 161–168, doi:10.1016/j.catena.2005.11.016.

Baker, V. R. (2008), Paleoﬂood hydrology: Origin, progress, prospects, Geomorphology, 101 (1-2), 1–13, doi:10.1016/j.geomorph.2008.05.016.

Balasubramanian, V. (2016), Water and the future of Bangalore metropolitan region, in Public Policy on Water supply for the Bangalore Metrolpolitan Region, Bangalore.

Barriendos, M., D. Coeur, M. Lang, M. C. Llasat, R. Naulet, F. Lemaitre, and a. Bar- rera (2003), Stationarity analysis of historical ﬂood series in France and Spain (14th20th centuries), Natural Hazards and Earth System Science, 3 (6), 583–592, doi:10.5194/ nhess-3-583-2003.

Batchelor, C., M. Rama Mohan Rao, and S. Manohar Rao (2003), Watershed development: A solution to water shortages in semi-arid India or part of the problem, Land Use and Water Resources Research, pp. 1–10.

Belgaumkar, G. D. (2013), Netravati report to suggest steps to soften blow to ghats.

Benito, G., and V. R. Thorndycraft (2005), Palaeoﬂood hydrology and its role in applied hydrological sciences, in Journal of Hydrology, vol. 313, pp. 3–15, doi:10.1016/j.jhydrol. 2005.02.002.

Benito, G., M. Rico, Y. Sánchez-Moya, A. Sopeña,V. R. Thorndycraft, and M. Barriendos (2010), The impact of late Holocene climatic variability and land use change on the flood hydrology of the Guadalent´ınRiver, southeast Spain, Global and Planetary Change, 70 (1- 4), 53–63, doi:10.1016/j.gloplacha.2009.11.007.

Benito, G., R. Br´azdil, J. Herget, and M. J. Machado (2015), Quantitative historical hydrology in Europe, Hydrology and Earth System Sciences, 19 (8), 3517–3539, doi: 10.5194/hess-19-3517-2015.

Bennett, E. M., G. D. Peterson, and L. J. Gordon (2009), Understanding relationships among multiple ecosystem services., Ecology letters, 12 (12), 1394–404, doi:10.1111/j.1461-0248. 2009.01387.x.

Beven, K. (2002), Runoﬀ generation in semi-arid areas, Dryland Rivers: Hydrology and geomorphology of semi-arid channels, pp. 57–105. REFERENCES 110

Beven, K. (2006), A manifesto for the equiﬁnality thesis, Journal of Hydrology, 320 (1-2), 18–36, doi:10.1016/j.jhydrol.2005.07.007.

Beven, K. (2016), Facets of uncertainty: epistemic uncertainty, non-stationarity, likelihood, hypothesis testing, and communication, Hydrological Sciences Journal, 61 (9), 1652–1665.

Biswas, A. K. (2004), Integrated Water Resources Management: A Reassessment, Water International, 29 (2), 248–256, doi:10.1080/02508060408691775.

Bivand, R., and C. Rundel (2016), rgeos: Interface to Geometry Engine - Open Source (GEOS).

Bivand, R., T. Keitt, and B. Rowlingson (2016), rgdal: Bindings for the Geospatial Data Abstraction Library.

Bl¨oschl, G. (2016), Predictions in ungauged basins–where do we stand?, Proceedings of the International Association of Hydrological Sciences, 373, 57–60.

Bl¨oschl, G., M. Sivapalan, T. Wagener, A. Viglione, and H. Savenije (2013), Runoﬀ Pre- diction in Ungauged Basins: Synthesis Across Processes, Places and Scales, Cambridge University Press.

Bonell, M., J. J. McDonnell, F. N. Scatena, J. Seibert, S. Uhlenbrook, and H. A. J. van Lanen (2006), HELPing FRIENDs in PUBs: charting a course for synergies within international water research programmes in gauged and ungauged basins, Hydrological Processes, 20 (8), 1867–1874.

Bosch, J., and J. Hewlett (1982), A review of catchment experiments to determine the eﬀect of vegetation changes on water yield and evapotranspiration, Journal of hydrology, 55.

Bouma, J. A., T. W. Biggs, and L. M. Bouwer (2011), The downstream externalities of harvesting rainwater in semi-arid watersheds: An Indian case study, Agricultural Water Management, 98 (7), 1162–1170, doi:10.1016/j.agwat.2011.02.010.

Bouwer, L., J. Aerts, P. Droogers, and A. Dolman (2006), Detecting the long-term impacts from climate variability and increasing water consumption on runoﬀ in the Krishna river basin (India), Hydrology and Earth System Sciences, (2003), 703–713.

Br´azdil,R., C. Pﬁster, H. Wanner, H. Von Storch, and J. Luterbacher (2005), Historical climatology in Europe - The state of the art, Climatic Change, 70 (3), 363–430, doi:10. 1007/s10584-005-5924-1.

Br´azdil,R., Z. W. Kundzewicz, and G. Benito (2006), Historical hydrology for studying ﬂood risk in Europe, Hydrological Sciences Journal, 51 (October 2006), 739–764, doi:10. 1623/hysj.51.5.739. REFERENCES 111

Briscoe, J., and R. Malik (2006), India’s water economy: bracing for a turbulent future, Tech. Rep. 34750, World Bank. Brown, A. E., L. Zhang, T. A. McMahon, A. W. Western, and R. A. Vertessy (2005), A review of paired catchment studies for determining changes in water yield resulting from alterations in vegetation, Journal of Hydrology, 310 (1-4), 28–61, doi:10.1016/j.jhydrol. 2004.12.010. Brown, C., and U. Lall (2006), Water and economic development: The role of variability and a framework for resilience, Natural Resources Forum, 30 (4), 306–317, doi:10.1111/j. 1477-8947.2006.00118.x. Brunke, M., and T. Gonser (1997), The ecological significance of exchange processes between rivers and groundwater, Freshwater biology, pp. 1–33. Brunner, P., C. T. Simmons, and P. G. Cook (2009), Spatial and temporal aspects of the transition from connection to disconnection between rivers, lakes and groundwater, Journal of Hydrology, 376 (12), 159–169, doi:10.1016/j.jhydrol.2009.07.023. Brutsaert, W., and J. L. Nieber (1977), Regionalized drought flow hydrographs from a mature glaciated plateau, doi:10.1029/WR013i003p00637. Büntgen, U., R. Brázdil, K. U. Heussner, J. Hofmann, R. Kontic, T. Kyncl, C. Pfister, K. Chromá,and W. Tegel (2011), Combined dendro-documentary evidence of Central European hydroclimatic springtime extremes over the last millennium, Quaternary Science Reviews, 30 (27-28), 3947–3959, doi:10.1016/j.quascirev.2011.10.010. Bürger,K., P. Dostal, J. Seidel, F. Imbery, M. Barriendos, H. Mayer, and R. Glaser (2006), Hydrometeorological reconstruction of the 1824 flood event in the Neckar River basin (southwest Germany), Hydrological Sciences Journal, 51 (October 2006), 864–877, doi: 10.1623/hysj.51.5.864. Burness, H. S., and T. C. Brill (2001), The role for policy in common pool groundwater use, Resource and Energy Economics, 23 (1), 19–40, doi:10.1016/S0928-7655(00)00029-4. Buytaert, W. (2015), Interactive comment on Why is the Arkavathy River drying? A multiple hypothesis approach in a data scarce region, HESS Interactive Comment, 12 (C65-67). Buytaert, W., Z. Zulkafli, S. Grainger, L. Acosta, J. Bastiaensen, B. De Bièvre,J. Bhusal, T. Chanie, J. Clark, A. Dewulf, M. Foggin, D. M. Hannah, C. Hergarten, A. Isaeva, T. Karpouzoglou, B. Pandeya, D. Paudel, K. Sharma, T. Steenhuis, S. Tilahun, M. Zhu- manova, and G. Van Hecken (2014), Citizen science in hydrology and water resources: opportunities for knowledge generation, ecosystem service management, and sustainable development, Name: Frontiers in Earth Science, 2, 26, doi:10.3389/feart.2014.00026. BWSSB (2017), About BWSSB (Bangalore Water Supply and Sewerage Board). REFERENCES 112

Cai, X., and R. Zeng (2013), Assessing the Stream Flow Effects of Groundwater Pumping and Return Flow from Irrigation, chap. 40, pp. 416–425, doi:10.1061/9780784412947.040. Calder, I., P. Rosier, K. Prasanna, and S. Parameswarappa (1997), Eucalyptus water use greater than rainfall input-possible explanation from southern India, Hydrology and Earth System Sciences, pp. 249–256. Calder, I. R., R. L. Hall, and K. T. Prasanna (1993), Hydrological impact of Eucalyptus plantation in India, Journal of Hydrology, 150 (2), 635–648, doi:10.1016/0022-1694(93) 90129-W. Callow, J. N., and K. R. J. Smettem (2009), The effect of farm dams and constructed banks on hydrologic connectivity and runoff estimation in agricultural landscapes, Environmental Modelling and Software, 24, 959–968, doi:10.1016/j.envsoft.2009.02.003. Campbell Scientific, I. (2014), CS650 and CS655 Water Content Reflectometers, Campbell Scientific, Inc. Carlson, T. N., and S. T. Arthur (2000), The impact of land useland cover changes due to urbanization on surface microclimate and hydrology: a satellite perspective, Global and planetary change, 25 (1), 49–65. Carr, J. A., and P. D’Odorico (2017), The water–food nexus and virtual water trade, Water Diplomacy in Action: Contingent Approaches to Managing Complex Water Problems, 1, 95. Catts, G., S. Khorram, A. Knight, and S. DeGloria (1985), Remote sensing of tidal chlorophyll-a variations in estuaries, International Journal of Remote Sensing, 6 (11), 1685–1706, doi:10.1080/01431168508948318. Chamberlin, T. (1965a), The method of multiple working hypotheses, Science, 103 (3), 349– 354, doi:10.1086/629752. Chamberlin, T. C. (1965b), The method of multiple working hypotheses, Science, 15 (3671), 754–759, doi:10.1126/science.148.3671.754. Chander, G., B. L. Markham, and D. L. Helder (2009), Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors, Remote Sensing of Environment, 113 (5), 893–903, doi:10.1016/j.rse.2009.01.007. Chandler, M. A., D. Rind, and R. Ruedy (1992), Pangaean climate during the Early Juras- sic: GCM simulations and the sedimentary record of paleoclimate, Geological Society of America Bulletin, 104 (5), 543–559. Chattopadhyay, N., and M. Hulme (1997), Evaporation and potential evapotranspiration in India under conditions of recent and future climate change, Agricultural and Forest Meteorology, 87 (1), 55–73. REFERENCES 113

Chen, J., X. Zhu, J. E. Vogelmann, F. Gao, and S. Jin (2011), A simple and effective method for filling gaps in Landsat ETM+ SLC-off images, Remote Sensing of Environment, 115 (4), 1053–1064, doi:10.1016/j.rse.2010.12.010. CNNL (2010), Study of Obstructions to flow of natural water in the course of the Arkavathy River and its tributaries from the origin upto the Thippagondanahalli reservoir, Bangalore. Cohen, J., P. Cohen, S. G. West, and L. Aiken (2003), Applied Multiple Regression / Cor- relation Analysis for the Behavioral Sciences, vol. Third Edit, 703 S. pp., Routledge, doi:10.2307/2064799. Congalton, R. G., and K. Green (2009), Assessing the Accuracy of Remotely Sensed Data: Principles and Practices, vol. 2, CRC/Taylor & Francis Group, LLC, Boca Raton, London, New York, doi:10.1111/j.1477-9730.2010.00574 2.x. Craven, J. P., R. E. Jewell, and H. E. Brooks (2002), Comparison between observed convective cloud-base heights and lifting condensation level for two different lifted parcels, Weather and Forecasting, 17 (4), 885–890, doi:10.1175/1520-0434(2002)017h0885: CBOCCBi2.0.CO;2. Critchley, W., K. Siegert, C. Chapman, and M. Finkel (1991), Water harvesting: a manual for the design and construction of water harvesting schemes for plant production, Tech. rep., FAO. Cronin, T. M. (2009), Paleoclimates: understanding climate change past and present, Columbia University Press. Crutzen, P. J. (2002), The Anthropocene, in Earth System Science in the Anthropocene, vol. 12, pp. 13–18, Springer-Verlag, Berlin/Heidelberg, doi:10.1007/3-540-26590-2 3. CSIRO (1988), CSIRO Disc Permeameter Instruction Manual. Dafforn, K. A., E. L. Johnston, A. Ferguson, C. L. Humphrey, W. Monk, S. J. Nichols, S. L. Simpson, M. G. Tulbure, and D. J. Baird (2016), Big data opportunities and challenges for assessing multiple stressors across scales in aquatic ecosystems, Marine and Freshwater Research, 67 (4), 393–413. Dataflow Systems Inc (2017), Odyssey Capacitive Water Level Logger, Dataflow Systems Inc. de Araújo,J. C., and P. H. A. Medeiros (2013), Impact of dense reservoir networks on water resources in semiarid environments, Australian Journal of Water Resources, 17 (1), 87–100. de Toledo, C. E., J. C. de Araújo,and C. L. de Almeida (2014), The use of remote-sensing techniques to monitor dense reservoir networks in the Brazilian semiarid region, Interna- tional Journal of Remote Sensing, 35 (10), 3683–3699. REFERENCES 114

Decagon Devices, I. (2016), EC-5 Soil Moisture Sensor Operators Manual, Decagon Devices, Inc. DES (2012), Annual season crop report for 1970-2012, Government of Karnataka, Bangalore, India. Devineni, N., S. Perveen, and U. Lall (2013), Assessing chronic and climate-induced water risk through spatially distributed cumulative deficit measures: A new picture of water sustainability in India, Water Resources Research, 49 (4), 2135–2145, doi:10.1002/wrcr. 20184. Di Baldassarre, G., A. Castellarin, and A. Brath (2009), Analysis of the effects of levee heightening on flood propagation: example of the River Po, Italy, Hydrological sciences journal, 54 (6), 1007–1017. Di Baldassarre, G., A. Viglione, G. Carr, L. Kuil, J. L. Salinas, and G. Blöschl (2013a), Socio-hydrology: conceptualising human-flood interactions, Hydrology and Earth System Sciences, 17 (8), 3295. Di Baldassarre, G., M. Kooy, J. S. Kemerink, and L. Brandimarte (2013b), Towards understanding the dynamic behaviour of floodplains as human-water systems, Hydrology and Earth System Sciences, 17 (8), 3235. Dillon, P., I. Gale, S. Contreras, P. Pavelic, R. Evans, and J. Ward (2009), Managing aquifer recharge and discharge to sustain irrigation livelihoods under water scarcity and climate change, IAHS-AISH Publication, 330 (September), 1–12. Duffie, J. A., and W. A. Beckman (2013), Solar Engineering of Thermal Processes, 3 ed., John Wiley \& Sons, Hoboken, New Jersey. Dunne, T., and R. D. Black (1970), Partial area contributions to storm runoff in a small New England watershed, Water resources research, 6 (5), 1296–1311. Dunne, T., and L. B. Leopold (1978), Water in environmental planning, W. H. Freeman and Co., San Francisco, CA. Ehret, U., H. V. Gupta, M. Sivapalan, S. V. Weijs, S. J. Schymanski, G. Blöschl, A. N. Gelfan, C. Harman, A. Kleidon, T. A. Bogaard, D. Wang, T. Wagener, U. Scherer, E. Zehe, M. F. P. Bierkens, G. Di Baldassarre, J. Parajka, L. P. H. Van Beek, A. Van Griensven, M. C. Westhoff, and H. C. Winsemius (2014), Advancing catchment hydrology to deal with predictions under change, Hydrology and Earth System Sciences, 18 (2), 649–671, doi:10.5194/hess-18-649-2014. Elshafei, Y., M. Sivapalan, M. Tonts, and M. R. Hipsey (2014), A prototype framework for models of socio-hydrology: Identification of key feedback loops and param- eterisation approach, Hydrology and Earth System Sciences, 18 (6), 2141–2166, doi: 10.5194/hess-18-2141-2014. REFERENCES 115

Elshafei, Y., J. Z. Coletti, M. Sivapalan, and M. R. Hipsey (2015), A model of the socio-hydrologic dynamics in a semiarid catchment: Isolating feedbacks in the coupled human-hydrology system, Water Resources Research, 51 (8), 6442–6471, doi:10.1002/ 2015WR017048.

Fader, M., M. C. Rulli, J. Carr, J. Dell’Angelo, P. D’Odorico, J. A. Gephart, M. Kummu, N. Magliocca, M. Porkka, C. Prell, M. J. Puma, Z. Ratajczak, D. A. Seekell, S. Suweis, and A. Tavoni (2016), Past and present biophysical redundancy of countries as a buﬀer to changes in food supply, Environmental Research Letters, 11 (5), 055,008, doi:10.1088/ 1748-9326/11/5/055008.

Falkenmark, M., T. Chapman, and Others (1989), Comparative hydrology: An ecological approach to land and water resources, The Unesco Press.

Falkenmark, M., C. M. Finlayson, and L. J. Gordon (2007), Agriculture, water, and ecosystems: avoiding the costs of going too far, Water for Food, Water for Life - A Comprehensive Assessment of Water Management in Agriculture, pp. 233–277, doi: 10.4324/9781849773799.

Falkenmark, M., J. Rockstr¨om,and L. Karlberg (2009), Present and future water requirements for feeding humanity, Food Security, 1 (1), 59–69, doi:10.1007/s12571-008-0003-x.

Farley, K. A., E. G. Jobbagy, and R. B. Jackson (2005), Eﬀects of aﬀorestation on water yield: a global synthesis with implications for policy, Global Change Biology, 11 (10), 1565–1576, doi:10.1111/j.1365-2486.2005.01011.x.

Fatichi, S., V. Y. Ivanov, and E. Caporali (2011), Simulation of future climate scenarios with a weather generator, Advances in Water Resources, 34 (4), 448–467, doi:10.1016/j. advwatres.2010.12.013.

Fisher, J. B., F. Melton, E. Middleton, C. Hain, M. Anderson, R. Allen, M. F. McCabe, S. Hook, D. Baldocchi, P. A. Townsend, A. Kilic, K. Tu, D. D. Miralles, J. Perret, J. P. Lagouarde, D. Waliser, A. J. Purdy, A. French, D. Schimel, J. S. Famiglietti, G. Stephens, and E. F. Wood (2017), The future of evapotranspiration: Global requirements for ecosystem functioning, carbon and climate feedbacks, agricultural management, and water resources, Water Resources Research, 53 (4), 2618–2626, doi:10.1002/2016WR020175.

Fishman, R., M. Jain, and A. Kishore (2013), Patterns of migration, water scarcity and caste in rural northern Gujarat, Tech. Rep. February, International Growth Centre, London.

Fishman, R. M., T. Siegfried, P. Raj, V. Modi, and U. Lall (2011), Over-extraction from shallow bedrock versus deep alluvial aquifers: Reliability versus sustainability consider- ations for India’s groundwater irrigation, Water Resources Research, 47 (12), W00L05, doi:10.1029/2011WR010617. REFERENCES 116

Folke, C. (2006), Resilience: The emergence of a perspective for social-ecological systems analyses, Global Environmental Change, 16 (3), 253–267, doi:10.1016/j.gloenvcha.2006.04. 002. Folke, C., T. Hahn, P. Olsson, and J. Norberg (2005), Adaptive Governance of Social- Ecological Systems, Annu. Rev. Environ. Resourc., 30 (1), 441–473, doi:10.1146/annurev. energy.30.050504.144511. Folke, C., S. R. Carpenter, B. Walker, M. Scheffer, T. Chapin, and J. Rockström(2010), Resilience thinking: Integrating resilience, adaptability and transformability, Ecology and Society, 15 (4), doi:10.1038/nnano.2011.191. Foster, S., H. Garduno, R. Evans, D. Olson, Y. Tian, W. Zhang, and Z. Han (2004), Qua- ternary aquifer of the North China Plain - Assessing and achieving groundwater resource sustainability, Hydrogeology Journal, 12 (1), 81–93, doi:10.1007/s10040-003-0300-6. Foulds, S. A., H. M. Griffiths, M. G. Macklin, and P. A. Brewer (2014), Geomorphological records of extreme floods and their relationship to decadal-scale climate change, Geomor- phology, 216, 193–207, doi:10.1016/j.geomorph.2014.04.003. Fox, J. (2008), Applied regression analysis and generalized linear models, 2nd editio ed., Sage Publications, Inc. Fox, J., and G. Monette (1992), Generalized collinearity diagnostics, Journal of the American Statistical Association, 87 (417), 178–183. Fox, P., P. H. Hutton, D. J. Howes, A. J. Draper, and L. Sears (2015), Reconstructing the natural hydrology of the San Francisco Bay-Delta watershed, Hydrology and Earth System Sciences, 19 (10), 4257–4274, doi:10.5194/hess-19-4257-2015. Frigg, R., and S. Hartmann (2017), Models in Science, The Stanford Encyclopedia of Philos- ophy (Spring 2017 Edition). Gaál,L., J. Szolgay, S. Kohnová, J. Parajka, R. Merz, A. Viglione, and G. Blöschl (2012), Flood timescales: Understanding the interplay of climate and catchment processes through comparative hydrology, Water Resources Research, 48 (4), doi:10.1029/2011WR011509. Gale, I., D. Macdonald, R. Calow, I. Neumann, M. Moench, H. Kulkarni, S. Mudrakartha, and K. Palanisami (2006), Managed Aquifer Recharge: an assessment of its role and effec- tiveness in watershed management, Tech. rep., British Geological Survey Commissioned Report, Keyworth, Nottingham. Gardelle, J., P. Hiernaux, L. Kergoat, and M. Grippa (2010), Less rain, more water in ponds: a remote sensing study of the dynamics of surface waters from 1950 to present in pastoral Sahel (Gourma region, Mali), Hydrology and Earth System Sciences, 14 (2), 309–324, doi:10.5194/hess-14-309-2010. REFERENCES 117

Gardner, R., E. Ostrom, and J. Walker (1990), The Nature of Common-Pool Resource Problems, Rationality and Society, 2 (3), 335–358, doi:10.1177/1043463190002003005. Gardner, W. R. (1958), Some steady-state solutions of the unsaturated moisture flow equation with application to evaporation from a water table., Soil Science, 85 (4), 228–232, doi:10.1097/00010694-195804000-00006. Garg, K. K., S. P. Wani, J. Barron, L. Karlberg, and J. Rockström(2013), Up-scaling potential impacts on water flows from agricultural water interventions: opportunities and trade-offs in the Osman Sagar catchment, Musi sub-basin, India, Hydrological Processes, 27 (26), 3905–3921. Gleeson, T., Y. Wada, M. F. P. Bierkens, and L. P. H. van Beek (2012a), Water balance of global aquifers revealed by groundwater footprint, Nature, 488 (7410), 197–200, doi: 10.1038/nature11295. Gleeson, T., W. M. Alley, D. M. Allen, M. A. Sophocleous, Y. Zhou, M. Taniguchi, and J. Vandersteen (2012b), Towards sustainable groundwater use: Setting long-term goals, backcasting, and managing adaptively, Ground Water, 50 (1), 19–26, doi:10.1111/j. 1745-6584.2011.00825.x. Gleick, P. (2014), The World’s Water Volume 8, vol. 8, Island Press/Center for Resource Economics, Washington, DC. Gleick, P. H. (1993), Water and conflict: Fresh water resources and international security, International Security, 18, 79–112. Gleick, P. H. (2011), The World’s Water Volume 7: The Biennial Report on Freshwater Resources, vol. 7, Island Press, doi:10.1007/978-1-59726-228-6. Glendenning, C., and R. Vervoort (2010), Hydrological impacts of rainwater harvesting (RWH) in a case study catchment: The Arvari River, Rajasthan, India. Part 1: Field- scale impacts, Agricultural Water Management, 98 (2), 331–342, doi:10.1016/j.agwat.2010. 09.003. Glendenning, C., F. van Ogtrop, a.K. Mishra, and R. Vervoort (2012), Balancing watershed and local scale impacts of rain water harvesting in IndiaA review, Agricultural Water Management, 107, 1–13, doi:10.1016/j.agwat.2012.01.011. Gober, P., D. D. White, R. Quay, D. A. Sampson, and C. W. Kirkwood (2017), Socio- hydrology modelling for an uncertain future, with examples from the USA and Canada, Geological Society, London, Special Publications, 408 (1), 183–199. GOI (1994), Drought-Prone Areas Programme and Desert Development Programme: C. H. Hanumantha Rao Committee Report, Tech. rep., Ministry of Rural Development, Gov- ernment of India, Delhi, 1994. REFERENCES 118

GOI (2002), Guidelines for Hariyali, Tech. rep., Department of Land Resources, Ministry of Rural Development, Government of India, New Delhi, India.

Google Earth (2016), version 7.1.7.2606. Karnataka, India, 2004-2009, approx. 13◦00’N 77◦30’E. DigitalGlobe 2016, [accessed March, 2017].

Gosain, A., S. Rao, and D. Basuray (2006a), Climate change impact assessment on hydrology of Indian river basins, Current science, 90 (3).

Gosain, A. K., S. Rao, and D. Basuray (2006b), Climate change impact assessment on hydrology of Indian river basins, Current Science, 90 (3), 346–353.

Gosain, A. K., S. Rao, and A. Arora (2011), Climate change impact assessment of water resources of India, Current Science(Bangalore), 101 (3), 356–371.

Grafton, R. Q., J. Pittock, R. Davis, J. Williams, G. Fu, M. Warburton, B. Udall, R. McKen- zie, X. Yu, N. Che, D. Connell, Q. Jiang, T. Kompas, A. Lynch, R. Norris, H. Possing- ham, and J. Quiggin (2013), Global insights into water resources, climate change and governance, Nature Climate Change, 3 (4), 315–321, doi:10.1038/nclimate1746.

GRASS Development Team (2016), Geographic Resources Analysis Support System (GRASS GIS) Software, Version 7.0, Open Source Geospatial Foundation.

Gunnell, Y., and A. Krishnamurthy (2003), Past and Present Status of Runoﬀ Harvesting Systems in Dryland Peninsular India: A Critical Review, Ambio, 32 (4), 320–323, doi: 10.1579/0044-7447-32.4.320.

GWP (2000), Integrated Water Resources Management, Tech. rep., Global Water Partner- ship, Stockholm, doi:10.2166/wst.2010.262.

Haddad, M. (2015), Social, religious, and cultural inﬂuences on the sustainability of water and its use, in Sustainable Water Use and Management: Examples of New Approaches and Perspectives, edited by W. Leal Filho and V. S¨umer,Green Energy and Technology, p. 406, Springer International Publishing, Cham, doi:10.1007/978-3-319-12394-3.

Halabisky, M., L. M. Moskal, A. Gillespie, and M. Hannam (2016), Reconstructing semi- arid wetland surface water dynamics through spectral mixture analysis of a time series of Landsat satellite images (1984-2011), Remote Sensing of Environment, 177, 171–183, doi:10.1016/j.rse.2016.02.040.

Hanjra, M. A., and M. E. Qureshi (2010), Global water crisis and future food security in an era of climate change, Food Policy, 35 (5), 365–377, doi:10.1016/j.foodpol.2010.05.006.

Hargreaves, G. H., and R. G. Allen (2003), History and evaluation of Hargreaves evapotranspiration equation, Journal of Irrigation and Drainage Engineering, 129 (1), 53–63, doi:10.1061/(ASCE)0733-9437(2003)129:1(53). REFERENCES 119

Hargreaves, G. H., and Z. Samani (1985), Reference Crop Evapotranspiration from Ambient Air Temperature, in American Society of Agricultural Engineering Meeting, Chicago, 1985, vol. 24, doi:10.13031/2013.26773.

Hewlett, J. D., and A. R. Hibbert (1967), Factors aﬀecting the response of small watersheds to precipitation in humid areas, Forest hydrology, 1, 275–290.

Hijmans, R. J. (2015), raster: Geographic Data Analysis and Modeling.

Hingane, L. S., K. Rupa Kumar, and B. V. Ramana Murty (1985), Long-term trends of surface air temperature in india, Journal of Climatology, 5 (5), 521–528, doi:10.1002/joc. 3370050505.

Holling, C. S. (2001), Understanding the complexity of economic, ecological, and social systems, doi:10.1007/s10021-001-0101-5.

Hoogesteger, J., and P. Wester (2015), Intensive groundwater use and (in)equity: Processes and governance challenges, doi:10.1016/j.envsci.2015.04.004.

Horton, R. E. (1933), The role of inﬁltration in the hydrologic cycle, Eos, Transactions American Geophysical Union, 14 (1), 446–460.

Houghton-Carr, H., and M. Fry (2006), The decline of hydrological data collection for development of integrated water resource management tools in Southern Africa, in Climate Variability and ChangeHydrological Impacts, November, pp. 51–55, IAHS Publ, Cuba.

Hrachowitz, M., H. H. G. Savenije, G. Bl¨oschl, J. J. McDonnell, M. Sivapalan, J. W. Pomeroy, B. Arheimer, T. Blume, M. P. Clark, U. Ehret, and Others (2013), A decade of Predictions in Ungauged Basins (PUB)a review, Hydrological sciences journal, 58 (6), 1198–1255.

Hu, Z., L. Wang, Z. Wang, Y. Hong, and H. Zheng (2015), Quantitative assessment of climate and human impacts on surface water resources in a typical semi-arid watershed in the middle reaches of the Yellow River from 1985 to 2006, International Journal of Climatology, 35 (1), 97–113.

Hughes, D. A., and S. K. Mantel (2010), Estimating the uncertainty in simulating the impacts of small farm dams on streamﬂow regimes in South Africa, Hydrological Sciences Journal– Journal des Sciences Hydrologiques, 55 (4), 578–592.

Hughes, J. D., K. C. Petrone, and R. P. Silberstein (2012), Drought, groundwater storage and stream ﬂow decline in southwestern Australia, Geophysical Research Letters, 39 (3).

Huntington, T. G. (2006), Evidence for intensiﬁcation of the global water cycle: review and synthesis, Journal of Hydrology, 319 (1), 83–95. REFERENCES 120

Hutton, P. H., J. S. Rath, and S. B. Roy (2017a), Freshwater ﬂow to the San Francisco Bay-Delta estuary over nine decades (Part 1): Trend evaluation, Hydrological Processes, In press.

Hutton, P. H., J. S. Rath, and S. B. Roy (2017b), Freshwater ﬂow to the San Francisco Bay-Delta estuary over nine decades (part 2): Change attribution, Hydrological Processes, In press.

Irish, R. R. (2000), Landsat 7 automatic cloud cover assessment, AeroSense 2000, 4049, 348–355, doi:10.1117/12.410358.

ISRO, and IN-RIMT (2000), Tippagondanahalli Reservoir (TGR): A remote sensing based evaluation., Technical report for Bangalore Metropolitan Region Development Authority, Bengaluru.

Ivanov, V. Y., R. L. Bras, and D. C. Curtis (2007), A weather generator for hydrological, ecological, and agricultural applications, Water Resources Research, 43 (10), doi:10.1029/ 2006WR005364.

Jaccard, J., C. K. Wan, and R. Turrisi (1990), The Detection and Interpretation of Interac- tion Eﬀects Between Continuous Variables in Multiple Regression, Multivariate Behavioral Research, 25 (4), 467–478, doi:10.1207/s15327906mbr2504 4.

Jamwal, P., T. Zuhail, P. R. Urs, and V. Srinivasan (2015), Contribution of sewage treatment to pollution abatement of urban streams, Current Science, 108, No 4,, 677–685.

Janakarajan, S. (1993a), In Search of Tanks: Some Hidden Facts, Economic and Political Weekly, 28 (26), A53–A60.

Janakarajan, S. (1993b), Economic and Social Implications of Groundwater Irrigation: Some Evidence from South India, Indian journal of agricultural economics, 48 (1), 65–75.

Jenkins, C. (1968), Techniques for computing rate and volume of stream depletion by Wellsa, Groundwater.

Joussaume, S., and K. E. Taylor (1995), Status of the paleoclimate modeling intercomparison project (PMIP), World Meteorological Organization-Publications-WMO TD, pp. 425–430.

Jury, W. A., and H. J. J. Vaux (2007), The emerging global water crisis: managing scarcity and conﬂict between water users, Advances in agronomy, 95 (07), doi: 10.1016/S0065-2113(07)95001-4.

Kallarackal, J., and C. K. Somen (1997), Water use by Eucalyptus tereticornis stands of diﬀering density in southern India., Tree physiology, 17, 195–203. REFERENCES 121

Kelkar, U., K. K. Narula, V. P. Sharma, and U. Chandna (2008), Vulnerability and adaptation to climate variability and water stress in Uttarakhand State, India, Global Environ- mental Change, 18 (4), 564–574.

Kerr, J., G. Pangare, and V. L. Pangare (2002), Watershed Development Projects In India: An Evaluation, Tech. rep., International Food Policy Research Institute, Washington D.C.

Kim, J., J. Yun, H. Kwon, and J. Chun (2011), First step toward bridging ecological and socio-economic systems: Linking thermodynamics, complexity, and sustainability, in Pro- ceedings of the 2011 TERRECO Science Conference, Karlsruhe Institute of Technology, Garmisch-Partenkirchen, Germany.

Kinal, J., and G. L. Stoneman (2012), Disconnection of groundwater from surface water causes a fundamental change in hydrology in a forested catchment in south-western Aus- tralia, Journal of Hydrology, 472, 14–24.

Kirchner, J. W. (1993), Stastical inevitability of Horton’s law and the apparent randomness of stream channel networks., Geology, 21 (3), 591–594, doi:10.1130/0091-7613(1993)021h0591.

Kleinhans, M. G., C. J. J. Buskes, and H. W. de Regt (2005), Terra Incognita: Explanation and Reduction in Earth Science, International Studies in the Philosophy of Science, 19, No. 3 (October 2005), 289–317, doi:10.1080/02698590500462356.

Konar, M., C. Dalin, S. Suweis, N. Hanasaki, A. Rinaldo, and I. Rodriguez-Iturbe (2011), Water for food: The global virtual water trade network, Water Resources Research, 47 (5), W05,520, doi:10.1029/2010WR010307.

Konar, M., T. P. Evans, M. Levy, C. A. Scott, T. J. Troy, C. J. V¨or¨osmarty, and M. Siva- palan (2016), Water resources sustainability in a globalizing world: who uses the water?, Hydrological Processes, 30 (18), 3330–3336.

Konikow, L. F., and E. Kendy (2005), Groundwater depletion: A global problem, Hydroge- ology Journal, 13 (1), 317–320, doi:10.1007/s10040-004-0411-8.

Kothawale, D. R., and K. Rupa Kumar (2005), On the recent changes in surface temperature trends over India, Geophysical Research Letters, 32 (18), doi:10.1029/2005GL023528.

Kuil, L., G. Carr, A. Viglione, A. Prskawetz, and G. Bl¨oschl (2016), Conceptualizing socio- hydrological drought processes: The case of the Maya collapse, Water Resources Research, 52 (8), 6222–6242, doi:10.1002/2015WR018298.

Kumar, M., and S. Ghosh (2006), Rainwater harvesting in India: some critical issues for basin planning and research, Land Use and Water Resources Research, 6, 1–17.

Kumar, M. D., N. Bassi, K. S. Kishan, S. Chattopadhyay, and A. Ganguly (2016), Rejuve- nating Tanks in Telangana, Economic & Political Weekly, lI (34), 30–34. REFERENCES 122

Kustu, M. D., Y. Fan, and A. Robock (2010), Large-scale water cycle perturbation due to irrigation pumping in the US High Plains: A synthesis of observed streamﬂow changes, Journal of Hydrology, 390 (3), 222–244.

Lall, U., T. Heikkila, C. Brown, and T. Siegfried (2008), Water in the 21st century: Deﬁning the elements of global crises and potential solutions, Journal of International Aﬀairs, 61 (2), 1–17.

Langford, K. (1976), Change in yield of water following a bushﬁre in a forest of Eucalyptus regnans, Journal of Hydrology, 29 (January 1939), 87–114.

Lele, S., and M. Sowmyashree (2016), Land use and land cover change in the Arkavathy basin, paper presented in “Adapting to Climate Change in Urbanising Watersheds: National Dissemination Workshop.” New Delhi, 22-23 August, 2016.

Lele, S., V. Srinivasan, P. Jamwal, B. K. Thomas, M. Eswar, and T. M. Zuhail (2013), Water Management In Arkavathy Basin: A situation analysis, Tech. Rep. 1, Ashoka Trust for Research in Ecology and the Environment, Bengaluru.

Liebe, J., N. van de Giesen, and M. Andreini (2005), Estimation of small reservoir storage capacities in a semi-arid environment: A case study in the Upper East Region of Ghana, Physics and Chemistry of the Earth, 30 (6-7 SPEC. ISS.), 448–454, doi:10.1016/j.pce.2005. 06.011.

Liebe, J. R., N. Van De Giesen, M. Andreini, M. T. Walter, and T. S. Steenhuis (2009), Determining watershed response in data poor environments with remotely sensed small reservoirs as runoﬀ gauges, Water Resources Research, 45 (7), W07,410, doi:10.1029/ 2008WR007369.

Lima Neto, I. E., M. C. Wiegand, and J. C. de Ara´ujo(2011), Sediment redistribution due to a dense reservoir network in a large semi-arid Brazilian basin, Hydrological Sciences Journal–Journal des Sciences Hydrologiques, 56 (2), 319–333.

Logsdon, S. D., and D. B. Jaynes (1993), Methodology for Determining Hydraulic Conduc- tivity with Tension Inﬁltrometers, Soil Science Society of America Journal, 57 (6), 1426, doi:10.2136/sssaj1993.03615995005700060005x.

Ludden, D. (1979), Patronage and irrigation in Tamil Nadu: a long-term view, Indian Economic & Social History Review.

Macdonald, N., and A. R. Black (2010), Reassessment of ﬂood frequency using historical information for the River Ouse at York, UK (12002000), Hydrological Sciences Journal, 55 (March 2014), 1152–1162, doi:10.1080/02626667.2010.508873. REFERENCES 123

Machado, M. J., B. A. Botero, J. López, F. Francés, A. D´ıez-Herrero, and G. Benito (2015), Flood frequency analysis of historical flood data under stationary and nonstationary modelling, Hydrology and Earth System Sciences, 19 (6), 2561–2576, doi: 10.5194/hess-19-2561-2015.

Maclure, M., and S. Schneeweiss (2001), Causation of bias: the episcope, Epidemiology, 12 (1), 114–122.

Maeda, E. E., and J. A. Torres (2012), Open Environmental Data in Developing Countries: Who Beneﬁts?, AMBIO, 41 (4), 410–412, doi:10.1007/s13280-012-0283-4.

Malveira, V. T. C., J. C. de Ara´ujo,and A. G¨untner (2012), Hydrological Impact of a High-Density Reservoir Network in Semiarid Northeastern Brazil, Journal of Hydrologic Engineering, 17 (1), 109–117, doi:10.1061/(ASCE)HE.1943-5584.0000404.

Maurer, E. (2014), Extraterrestrial solar radiation calculator, http://www.engr.scu.edu/ ~emaurer/tools/calc_solar_cgi.pl. McFeeters, S. K. (1996), The use of the Normalized Diﬀerence Water Index (NDWI) in the delineation of open water features, International Journal of Remote Sensing, 17 (7), 1425–1432, doi:10.1080/01431169608948714.

McVicar, T. R., M. L. Roderick, R. J. Donohue, L. T. Li, T. G. Van Niel, A. Thomas, J. Grieser, D. Jhajharia, Y. Himri, N. M. Mahowald, A. V. Mescherskaya, A. C. Kruger, S. Rehman, and Y. Dinpashoh (2012), Global review and synthesis of trends in observed terrestrial near-surface wind speeds: Implications for evaporation, Journal of Hydrology, 416, 182–205, doi:10.1016/j.jhydrol.2011.10.024.

Meigh, J. (1995), The impact of small farm reservoirs on urban water supplies in Botswana, Natural Resources Forum, 19 (I), 71–83, doi:10.1111/j.1477-8947.1995.tb00594.x.

Merritt, E. (1984), The identiﬁcation of four stages during micro-rill development, Earth Surface Processes and Landforms, 9 (5), 493–496.

Meter, K. J. V., N. B. Basu, E. Tate, and J. Wyckoﬀ (2014), Monsoon Harvests: The Living Legacies of Rainwater Harvesting Systems in South India, Environmental Science & Technology, 48 (8), 4217–4225, doi:10.1021/es4040182.

Mialhe, F., Y. Gunnell, and C. Mering (2008), Synoptic assessment of water resource variability in reservoirs by remote sensing: General approach and application to the runoﬀ harvesting systems of south India, Water Resources Research, 44 (5), n/a–n/a, doi:10.1029/2007WR006065.

Milly, P. C. D., J. Betancourt, M. Falkenmark, R. M. Hirsch, Z. Kundzewicz, D. P. Let- tenmaier, R. J. Stouﬀer, W. Zbigniew, D. P. Lettenmaier, and R. J. Stouﬀer (2008), REFERENCES 124

Stationarity Is Dead: Whither Water Management?, Science, 319 (5863), 573–574, doi: 10.1126/science.1151915.

Mitchell, B. R. (1971), Abstract of British historical statistics, CUP Archive.

Moench, M. (2002), Water and the potential for social instability: Livelihoods, migration and the building of society, Natural Resources Forum, 26 (3), 195–204, doi:10.1111/0165-0203. 00021.

Molden, D. (2007), Water for food, water for life: a comprehensive assessment of water management in agriculture, Earthscan.

Molden, D., and R. Sakthivadivel (1999), Water Accounting to Assess Use and Productivity of Water, International Journal of Water Resources Development, 15 (1-2), 55–71, doi: 10.1080/07900629948934.

Montanari, A., G. Young, H. Savenije, D. Hughes, T. Wagener, L. Ren, D. Kout- soyiannis, C. Cudennec, E. Toth, S. Grimaldi, G. Blöschl, M. Sivapalan, K. Beven, H. Gupta, M. Hipsey, B. Schaefli, B. Arheimer, E. Boegh, S. Schymanski, G. Di Bal- dassarre, B. Yu, P. Hubert, Y. Huang, A. Schumann, D. Post, V. Srinivasan, C. Har- man, S. Thompson, M. Rogger, A. Viglione, H. McMillan, G. Characklis, Z. Pang, and V. Belyaev (2013), Panta RheiEverything Flows: Change in hydrology and societyThe IAHS Scientific Decade 20132022, Hydrological Sciences Journal, 58 (6), 1256–1275, doi: 10.1080/02626667.2013.809088.

Moore, G. K. (1980), Satellite remote sensing of water turbidity / Sonde de télémesure par satellite de la turbiditéde l’eau, Hydrological Sciences Bulletin, 25 (4), 407–421, doi: 10.1080/02626668009491950.

Mujumdar, P. P., and S. Ghosh (2008), Modeling GCM and scenario uncertainty using a possibilistic approach: Application to the Mahanadi River, India, Water Resources Research, 44 (6), doi:10.1029/2007WR006137.

Mukherji, A., and T. Shah (2005), Groundwater socio-ecology and governance: A review of institutions and policies in selected countries, Hydrogeology Journal, 13 (1), 328–345, doi:10.1007/s10040-005-0434-9.

M¨uldner,G., and M. P. Richards (2007), Stable isotope evidence for 1500 years of human diet at the city of York, UK, American Journal of Physical Anthropology, 133 (1), 682–697.

M¨uller,M. F., M. C. M¨uller-Itten,and S. M. Gorelick (2017), How Jordan and Saudi Ara- bia are avoiding a tragedy of the commons over shared groundwater, Water Resources Research. REFERENCES 125

Narayanamoorthy, A. (2014), Groundwater depletion and water extraction cost: some evidence from South India, International Journal of Water Resources Development, 0627 (July), 1–14, doi:10.1080/07900627.2014.935302. Nathan, R. J., and T. A. McMahon (1990), Evaluation of Automated Techniques for Baseflow and Recession Analysis, Water Resources Research, 26 (7), 1465–1473, doi: 10.1029/WR026i007p01465. Neitsch, S. L., J. G. Arnold, J. R. Kiniry, and J. R. Williams (2011), Soil and Water Assessment Tool Theoretical Documentation Version 2009, Texas A&M, Agrilife Research and Extension. Novo, E., J. Hansom, and P. Curran (1989), The effect of viewing geometry and wavelength on the relationship between reflectance and suspended sediment concentration, Interna- tional Journal of Remote Sensing, 10 (8), 1357–1372, doi:10.1080/01431168908903973. Nyssen, J., W. Clymans, K. Descheemaeker, J. Poesen, I. Vandecasteele, M. Vanmaercke, A. Zenebe, M. V. Camp, M. Haile, N. Haregeweyn, J. Moeyersons, K. Martens, T. Ge- breyohannes, J. Deckers, and K. Walraevens (2010), Impact of soil and water conservation measures on catchment hydrological responsea case in north Ethiopia, Hydrological Pro- cesses, 24 (13), 1880–1895, doi:10.1002/hyp.7628. O’Brien, K., R. Leichenko, U. Kelkar, H. Venema, G. Aandahl, H. Tompkins, A. Javed, S. Bhadwal, S. Barg, L. Nygaard, and J. West (2004), Mapping vulnerability to multiple stressors: Climate change and globalization in India, Global Environmental Change, 14 (4), 303–313, doi:10.1016/j.gloenvcha.2004.01.001. Ormerod, K. J., and C. a. Scott (2012), Drinking Wastewater: Public Trust in Potable Reuse, Science, Technology & Human Values, 38 (3), 351–373, doi:10.1177/0162243912444736. Ostrom, E. (2009), A general framework for analyzing sustainability of social-ecological systems., Science (New York, N.Y.), 325 (5939), 419–22, doi:10.1126/science.1172133. Ostrom, E., R. Gardner, and W. J. (1994), Rules, Games, & Common-pool resources, 394 pp., University of Michigan Press. Otsu, N. (1979), A threshold selection method from gray-level histograms, IEEE Transac- tions on Systems, Man, and Cybernetics, 9 (1), 62–66, doi:10.1109/TSMC.1979.4310076. Palmer, M. A., C. A. Reidy Liermann, C. Nilsson, M. Flörke, J. Alcamo, P. S. Lake, and N. Bond (2008), Climate change and the world’s river basins: Anticipating management options, Frontiers in Ecology and the Environment, 6 (2), 81–89, doi:10.1890/060148. Pande, S., and M. Sivapalan (2017), Progress in socio-hydrology: a meta-analysis of challenges and opportunities, Wiley Interdisciplinary Reviews: Water, 4 (4), e1193, doi: 10.1002/wat2.1193. REFERENCES 126

Pebesma, E. J., and R. S. Bivand (2005), Classes and methods for spatial data in {R}, R News, 5 (2), 9–13.

Peel, M. C., and G. Bl¨oschl (2011), Hydrological modelling in a changing world, Progress in Physical Geography, 35 (2), 249–261.

Penny, G., V. Srinivasan, I. Dronova, S. Lele, and S. Thompson (2016), Spatial characterization of long-term hydrological change in the Arkavathy watershed adjacent to Bangalore, India, Hydrology and Earth System Sciences Discussions, 2016, 1–25, doi: 10.5194/hess-2016-562.

Penny, G., V. Srinivasan, A. R., J. Peschel, S. Young, and S. E. Thompson (2017), A process- based hydrologic reconstruction to understand streamﬂow decline in a human-dominated semiarid catchment, Water Resources Research, in review.

Perrin, J., S. Ferrant, S. Massuel, B. Dewandel, J. C. Mar´echal, S. Aulong, and S. Ahmed (2012), Assessing water availability in a semi-arid watershed of southern India using a semi-distributed model, Journal of Hydrology, 460-461, 143–155, doi:10.1016/j.jhydrol. 2012.07.002.

Petrone, K. C., J. D. Hughes, T. G. Van Niel, and R. P. Silberstein (2010), Streamﬂow decline in southwestern Australia, 19502008, Geophysical Research Letters, 37 (11), n/a— -n/a, doi:10.1029/2010GL043102.

Puri, S., and P. Nair (2004), Agroforestry research for development in India: 25 years of experiences of a national program, New Vistas in Agroforestry, pp. 437–452.

QGIS Development Team (2016), QGIS Geographic Information System, Open Source Geospatial Foundation.

R Core Team (2016), R: A Language and Environment for Statistical Computing, R Foun- dation for Statistical Computing, Vienna, Austria.

Ramanathan, V., C. Chung, D. Kim, T. Bettge, L. Buja, J. T. Kiehl, W. M. Washington, Q. Fu, D. R. Sikka, and M. Wild (2005), Atmospheric brown clouds: Impacts on South Asian climate and hydrological cycle, Proceedings of the National Academy of Sciences of the United States of America, 102 (15), 5326–5333.

Reddy, V. R. (2005), Costs of resource depletion externalities: a study of groundwater overexploitation in Andhra Pradesh, India, Environment and Development Economics, 10 (4), 533–556, doi:10.1017/S1355770X05002329.

Reichman, O. J., M. B. Jones, and M. P. Schildhauer (2011), Challenges and opportunities of open data in ecology, Science(Washington), 331 (6018), 703–705. REFERENCES 127

Renganayaki, P., and L. Elango (2013), A review on managed aquifer recharge by check dams: A case study near Chennai , India, International Journal of Research in Engineering and Technology, pp. 416–423.

Revive Kumudavathi (2017), Kumudavathi River Rejuvenation Project, http:// revivekumudvathi.org/. Richards, A., and N. Singh (2002), Inter-state Water Disputes in India: Institutions and Policies, International Journal of Water Resources Development, 18 (4), 611–625, doi: 10.1080/0790062022000017437.

Rockstr¨om,J. (2000), Water resources management in smallholder farms in Eastern and Southern Africa: An overview, Physics and Chemistry of the Earth, Part B: Hydrology, Oceans and Atmosphere, 25 (3), 275–283, doi:10.1016/S1464-1909(00)00015-0.

Rockstr¨om,J., W. Steﬀen, K. Noone, and E. Al. (2009), A safe operating space for humanity, Nature, 461 (September), 472–475, doi:10.1038/461472a.

Rodell, M., I. Velicogna, and J. S. Famiglietti (2009), Satellite-based estimates of groundwater depletion in India., Nature, 460 (7258), 999–1002, doi:10.1038/nature08238.

Rodrigues, L. N., E. E. Sano, T. S. Steenhuis, and D. P. Passo (2012), Estimation of small reservoir storage capacities with remote sensing in the Brazilian Savannah region, Water Resources Management, 26 (4), 873–882, doi:10.1007/s11269-011-9941-8.

Roohi, R., and J. Webb (2012), Landsat Image Based Temporal and Spatial Analysis of Farm Dams in Western Victoria., in GSR, Citeseer.

Rose, S., and N. E. Peters (2001), Eﬀects of urbanization on streamﬂow in the Atlanta area (Georgia, USA): a comparative hydrological approach, Hydrological Processes, 15 (8), 1441–1457.

Rossi, A., N. Massei, B. Laignel, D. Sebag, and Y. Copard (2009), The response of the Mississippi River to climate ﬂuctuations and reservoir construction as indicated by wavelet analysis of streamﬂow and suspended-sediment load, 1950–1975, Journal of Hydrology, 377 (3), 237–244.

Roy, A. D., and T. Shah (2002), Socio-ecology of groundwater irrigation in India, Intensive use of groundwater challenges and opportunities, pp. 307–336.

Ruddell, B. L., and P. Kumar (2009), Ecohydrologic process networks: 1. Identiﬁcation, Water Resources Research, 45 (3).

Russell, J. C. (1948), British medieval population, University of New Mexico Press Albu- querque. REFERENCES 128

Saft, M., A. W. Western, L. Zhang, M. C. Peel, and N. J. Potter (2015), The influence of multiyear drought on the annual rainfall-runoff relationship: An Australian perspective, Water Resources Research, 51 (4), 2444–2463. Saft, M., M. C. Peel, A. W. Western, and L. Zhang (2016), Predicting shifts in rainfall-runoff partitioning during multiyear drought: Roles of dry period and catchment characteristics, Water Resources Research. Saghafian, B., H. Farazjoo, B. Bozorgy, and F. Yazdandoost (2008), Flood intensification due to changes in land use, Water Resources Management, 22 (8), 1051–1067. Sakthivadivel, R. (2007), The groundwater recharge movement in India, The agricultural groundwater revolution: Opportunities and threats to development, 3, 195–210. Salehyan, I. (2008), From climate change to conflict? No consensus yet, Journal of Peace Research, 45 (3), 315–326. Savenije, H. H. G. (2001), Equifinality, a blessing in disguise?, Hydrological Processes, 15 (14), 2835–2838, doi:10.1002/hyp.494. Savenije, H. H. G., a. Y. Hoekstra, and P. van der Zaag (2014), Evolving water science in the Anthropocene, Hydrology and Earth System Sciences, 18 (1), 319–332, doi:10.5194/ hess-18-319-2014. Sawunyama, T., A. Senzanje, and A. Mhizha (2006), Estimation of small reservoir storage capacities in Limpopo River Basin using geographical information systems (GIS) and remotely sensed surface areas: Case of Mzingwane catchment, Physics and Chemistry of the Earth, 31 (15-16), 935–943, doi:10.1016/j.pce.2006.08.008. Scaramuzza, P. L., G. Schmidt, J. C. Storey, and J. Barsi (2005), Landsat 7 Scan Line Corrector-Off Gap-Filled Product Gap-Filled Product Development Process, In Proceed- ings of Pecora, 16, 23–27. Scott, C. A. (2011), The water-energy-climate nexus: Resources and policy outlook for aquifers in Mexico, Water Resources Research, 47 (12), doi:10.1029/2011WR010805. Semmens, K. A., M. C. Anderson, W. P. Kustas, F. Gao, J. G. Alfieri, L. McKee, J. H. Prueger, C. R. Hain, C. Cammalleri, Y. Yang, T. Xia, L. Sanchez, M. Mar Alsina, and M. Vélez(2015), Monitoring daily evapotranspiration over two California vineyards using Landsat 8 in a multi-sensor data fusion approach, Remote Sensing of Environment, 185, 155–170, doi:10.1016/j.rse.2015.10.025. Sen, A. (1973), On economic inequality, Oxford University Press. Sene, K. J., and F. A. K. Farquharson (1998), Sampling errors for water resources design: the need for improved hydrometry in developing countries, Water Resources Management, 12 (2), 121–138, doi:10.1023/A:1007952319495. REFERENCES 129

Shah, E. (2003a), Social designs: Tank irrigation technology and agrarian transformation in Karnataka, South India, Orient Longman, Bangalore, India.

Shah, E. (2003b), Social designs: Tank irrigation technology and agrarian transformation in Karnataka, South India, vol. 4, Orient Longman Hyderabad.

Shah, T., A. D. Roy, A. S. Qureshi, and J. Wang (2003), Sustaining Asia’s groundwater boom: An overview of issues and evidence, Natural Resources Forum, 27 (2), 130–141, doi:10.1111/1477-8947.00048.

Shah, T., S. Bhatt, R. K. Shah, and J. Talati (2008), Groundwater governance through electricity supply management: Assessing an innovative intervention in Gujarat, western India, Agricultural Water Management, 95 (11), 1233–1242, doi:10.1016/j.agwat.2008.04. 006.

Shah, T., M. Giordano, and A. Mukherji (2012), Political economy of the energy-groundwater nexus in India: exploring issues and assessing policy options, Hydrogeology Journal, 20 (5), 995–1006, doi:10.1007/s10040-011-0816-0.

Sharda, V. N., R. S. Kurothe, D. R. Sena, V. C. Pande, and S. P. Tiwari (2006), Estimation of groundwater recharge from water storage structures in a semi-arid climate of India, Journal of Hydrology, 329 (1-2), 224–243, doi:10.1016/j.jhydrol.2006.02.015.

Shiva, V., H. C. Sharatchandra, and J. Bandyopadhyay (1981), Social, economic, and ecological impact of social forestry in Kolar, Indian Institute of Management.

Shuttleworth, W. J. (2007), Putting the “vap” into evaporation, Hydrology and Earth System Sciences, 11 (1), 210–244, doi:10.5194/hess-11-210-2007.

Siebert, S., J. Burke, J. M. Faures, K. Frenken, J. Hoogeveen, P. D¨oll,and F. T. Portmann (2010), Groundwater use for irrigation a global inventory, Hydrology and Earth System Sciences, 14 (10), 1863–1880, doi:10.5194/hess-14-1863-2010.

Sivapalan, M. (2003), Prediction in ungauged basins: a grand challenge for theoretical hydrology, Hydrological Processes, 17 (15), 3163–3170, doi:10.1002/hyp.5155.

Sivapalan, M., K. Takeuchi, S. W. Franks, V. K. Gupta, H. Karambiri, V. Lakshmi, X. Liang, J. J. McDonnell, E. M. Mendiondo, P. E. O’Connell, T. Oki, J. W. Pomeroy, D. Schertzer, S. Uhlenbrook, and E. Zehe (2003), IAHS Decade on Predictions in Ungauged Basins (PUB), 2003–2012: Shaping an exciting future for the hydrological sciences, Hydrological Sciences Journal, 48 (6), 857–880.

Sivapalan, M., H. H. G. Savenije, and G. Bl¨oschl (2012), Socio-hydrology: A new science of people and water, doi:10.1002/hyp.8426. REFERENCES 130

Sivapalan, M., M. Konar, V. Srinivasan, A. Chhatre, A. Wutich, C. A. Scott, J. L. Wescoat, and I. Rodr\’\iguez-Iturbe (2014), Socio-hydrology: Use-inspired water sustainability science for the Anthropocene, Earth’s Future, 2 (4), 225–230, doi:10.1002/2013EF000164.

Smerdon, J. E., B. I. Cook, E. R. Cook, and R. Seager (2015), Bridging past and future climate across paleoclimatic reconstructions, observations, and models: A hydroclimate case study, Journal of Climate, 28 (8), 3212–3231.

Soilmoisture Equipment Corp. (2008), Tension inﬁltrometer 2826D20 operating instructions.

Sophocleous, M. (1997), Managing Water Resources Systems: Why “Safe Yield” Is Not Sustainable, doi:10.1111/j.1745-6584.1997.tb00116.x.

Sophocleous, M. (2002), Interactions between groundwater and surface water: the state of the science, Hydrogeology Journal, 10 (1), 52–67, doi:10.1007/s10040-001-0170-8.

Sophocleous, M. (2010), Review: groundwater management practices, challenges, and inno- vations in the High Plains aquifer, USA—lessons and recommended actions, Hydrogeology Journal, 18 (3), 559–575, doi:10.1007/s10040-009-0540-1.

Spencer, J. W. (1971), Solar position and radiation tables for Brisbane (latitude 27. 5/sup 0/S), Tech. rep., Commonwealth Scientiﬁc and Industrial Research Organization, Mel- bourne (Australia).

Spencer, T., O. Slaymaker, and C. Embleton-Hamann (2009), Landscape, landscape-scale processes and global environmental change: synthesis and new agendas for the twenty- ﬁrst century, Geomorphology and global environmental change, Cambridge U. Press, Cam- bridge, pp. 403–423.

Srinivasan, V., S. Thompson, K. Madhyastha, G. Penny, K. Jeremiah, and S. Lele (), PLACEHOLDER TITLE, Water Resources Research, In review(This issue).

Srinivasan, V., E. F. Lambin, S. M. Gorelick, B. H. Thompson, and S. Rozelle (2012a), The nature and causes of the global water crisis: Syndromes from a meta-analysis of coupled human-water studies, Water Resources Research, 48 (10), 1–16, doi:10.1029/ 2011WR011087.

Srinivasan, V., K. C. Seto, R. Emerson, and S. M. Gorelick (2012b), The impact of urbanization on water vulnerability: A coupled human–environment system approach for Chennai, India, Global Environmental Change, doi:10.1016/j.gloenvcha.2012.10.002.

Srinivasan, V., K. C. Seto, R. Emerson, and S. M. Gorelick (2013), The impact of urbanization on water vulnerability: A coupled human-environment system approach for Chennai, India, Global Environmental Change, 23 (1), 229–239, doi:10.1016/j.gloenvcha.2012.10.002. REFERENCES 131

Srinivasan, V., S. Thompson, K. Madhyastha, G. Penny, K. Jeremiah, and S. Lele (2015), Why is the Arkavathy River drying? A multiple-hypothesis approach in a data- scarce region, Hydrology and Earth System Sciences, 19 (4), 1905–1917, doi:10.5194/ hessd-12-25-2015.

Srinivasan, V., M. Sanderson, M. Garcia, M. Konar, G. Bl¨oschl, M. Sivapalan, M. Sander- son, M. Garcia, M. Konar, G. Bl¨oschl, and M. Sivapalan (2016), Prediction in a socio- hydrological world, Hydrological Sciences Journal Journal des Sciences Hydrologiques, 00 (00), 1–8, doi:10.1080/02626667.2016.1253844.

Srinivasan, V., M. Konar, and M. Sivapalan (2017), A dynamic framework for water security, Water Security, 19, 4225, doi:10.1016/j.wasec.2017.03.001.

Steﬀen, W., P. J. Crutzen, and J. R. McNeill (2007), The Anthropocene: Are Hu- mans Now Overwhelming the Great Force of Nature?, Ambio, 36 (8), 614, doi:10.1579/ 0044-7447(2007)36.

Steﬀen, W., K. Richardson, J. Rockstr¨om,S. E. Cornell, I. Fetzer, E. M. Bennett, R. Biggs, S. R. Carpenter, W. de Vries, C. A. de Wit, C. Folke, D. Gerten, J. Heinke, G. M. Mace, L. M. Persson, V. Ramanathan, B. Reyers, and S. Sorlin (2015), Planetary boundaries: Guiding human development on a changing planet, Science, 347 (6223), 1259,855– 1259,855, doi:10.1126/science.1259855.

Steﬀens, K. J., and K. J. Franz (2012), Late 20th-century trends in Iowa watersheds: an investigation of observed and modelled hydrologic storages and ﬂuxes in heavily managed landscapes, International Journal of Climatology, 32 (9), 1373–1391.

Stocker, T. F., and C. C. Raible (2005), Climate change: water cycle shifts gear, Nature, 434 (7035), 830–833, doi:10.1038/434830a.

Suresh, T. (2001), An urban water scenario: A case study of the Bangalore metropolis, Karnataka, India, Regional management of water resources: Proceedings of a symposium held during the Sixth IAHS Scientiﬁc Assembly at Maastricht, The Netherlands, July 2001, pp. 97–104.

Suweis, S., J. A. Carr, A. Maritan, A. Rinaldo, S. Suweis, J. A. Carr, A. Maritan, A. Rinaldo, and P. D. Odorico (2015), Resilience and reactivity of global food security, Proceedings of the National Academy of Sciences, 112 (34), E4811–E4811, doi:10.1073/pnas.1512971112.

Thatheyus, A., D. P. Dhanaseeli, and P. Vanitha (2013), Inter - State Dispute over Water and Safety in India: The Mullaperiyar Dam, a Historical Perspective, American Journal of Water Resources, 1 (2), 10–19, doi:10.12691/ajwr-1-2-2.

Theis, C. (1940), The source of water derived from wells, Civil Engineering. REFERENCES 132

Thompson, S., M. Sivapalan, C. J. Harman, V. Srinivasan, M. R. Hipsey, P. Reed, A. Mon- tanari, and G. Bl¨oschl (2013a), Developing predictive insight into changing water systems: use-inspired hydrologic science for the Anthropocene, Hydrology and Earth System Science, 17, 5013–5039, doi:10.5194/hess-17-5013-2013.

Thompson, S. E., C. J. Harman, a. G. Konings, M. Sivapalan, a. Neal, and P. a. Troch (2011), Comparative hydrology across AmeriFlux sites: The variable roles of climate, vegetation, and groundwater, Water Resources Research, 47, W00J07, doi:10.1029/2010WR009797.

Thompson, S. E., M. Sivapalan, C. J. Harman, V. Srinivasan, M. R. Hipsey, P. Reed, a. Mon- tanari, and G. Bl¨oschl (2013b), Developing predictive insight into changing water systems: use-inspired hydrologic science for the Anthropocene, Hydrology and Earth System Sci- ences, 17 (12), 5013–5039, doi:10.5194/hess-17-5013-2013.

Tiwari, V. M., J. Wahr, and S. Swenson (2009), Dwindling groundwater resources in northern India, from satellite gravity observations, Geophysical Research Letters, 36 (18), L18,401, doi:10.1029/2009GL039401.

Tol, R. S. J., and S. Wagner (2010), Climate change and violent conﬂict in Europe over the last millennium, Climatic Change, 99 (1), 65–79.

Tripathi, S., V. Srinivas, and R. S. Nanjundiah (2006), Downscaling of precipitation for climate change scenarios: A support vector machine approach, Journal of Hydrology, 330 (3- 4), 621–640, doi:10.1016/j.jhydrol.2006.04.030.

United Nations (2013), World Population Prospects: The 2012 Revision.

United Nations (2016), The Sustainable Development Goals Report 2016, The Sustainable Development Goals Report, vol. 2016, 1–56 pp., United Nations, doi:10.18356/3405d09f-en.

Vaidyanathan, A., C. for Science, and I. Environment (New Delhi (2001), Tanks of South India, Centre for Science and Environment.

Van Meter, K., S. E. Thompson, and N. B. Basu (2016a), Human Impacts on Stream Hy- drology and Water Quality, in Stream Ecosystems in a Changing Environment, p. 441, Elsevier.

Van Meter, K. J., N. B. Basu, E. Tate, and J. Wyckoﬀ (2014), Monsoon harvests: The living legacies of rainwater harvesting systems in South India, Environmental Science and Technology, 48 (8), 4217–4225, doi:10.1021/es4040182.

Van Meter, K. J., N. B. Basu, D. L. McLaughlin, and M. Steiﬀ (2016b), The socio- ecohydrology of rainwater harvesting in India: understanding water storage and release dynamics at tank and catchment scales, Hydrology and Earth System Sciences Discussions, 20, 2629–2647, doi:10.5194/hessd-12-12121-2015. REFERENCES 133

Viglione, A., G. Di Baldassarre, L. Brandimarte, L. Kuil, G. Carr, J. L. Salinas, A. Scolobig, and G. Blöschl (2014), Insights from socio-hydrology modelling on dealing with flood risk–roles of collective memory, risk-taking attitude and trust, Journal of Hydrology, 518, 71–82. Vogel, R. M., U. Lall, X. Cai, B. Rajagopalan, P. K. Weiskel, R. P. Hooper, and N. C. Matalas (2015), Hydrology: The interdisciplinary science of water, Water Resources Research, 51 (6), 4409–4430, doi:10.1002/2015WR017049. Vörösmarty, C., D. Lettenmaier, C. Levêque, M. Meybeck, C. Pahl-Wostl, J. Alcamo, W. Cosgrove, H. Grassl, H. Hoff, P. Kabat, F. Lansigan, R. Lawford, and R. Naiman (2004), Human transforming the Global Water System, Eos, 85 (48), 509–520, doi: 10.1029/2004EO480001. Vörösmarty, C. J. (2000), Global Water Resources: Vulnerability from Climate Change and Population Growth, Science, 289 (5477), 284–288, doi:10.1126/science.289.5477.284. Vörösmarty, C. J., P. B. McIntyre, M. O. Gessner, D. Dudgeon, A. Prusevich, P. Green, S. Glidden, S. E. Bunn, C. A. Sullivan, C. R. Liermann, Others, and P. M. Davies (2010), Global threats to human water security and river biodiversity., Nature, 467 (7315), 555– 561, doi:10.1038/nature09549. Vörösmarty, C. J., C. Pahl-Wostl, S. E. Bunn, and R. Lawford (2013), Global water, the anthropocene and the transformation of a science, Current Opinion in Environmental Sustainability, 5 (6), 539–550, doi:10.1016/j.cosust.2013.10.005. Wada, Y., L. P. H. van Beek, C. M. van Kempen, J. Reckman, S. Vasak, and M. F. P. Bierkens (2010), Global depletion of groundwater resources, Geophysical Research Letters, 37 (20), doi:10.1029/2010GL044571. Wada, Y., L. P. H. Van Beek, and M. F. P. Bierkens (2012), Nonsustainable groundwater sustaining irrigation: A global assessment, Water Resources Research, 48 (1), doi:10.1029/ 2011WR010562. Wagener, T., M. Sivapalan, J. McDonnell, R. Hooper, V. Lakshmi, X. Liang, and P. Kumar (2004), Predictions in ungauged basins as a catalyst for multidisciplinary hydrology, Eos, Transactions American Geophysical Union, 85 (44), 451–457. Wagener, T., M. Sivapalan, P. a. Troch, B. L. McGlynn, C. J. Harman, H. V. Gupta, P. Kumar, P. S. C. Rao, N. B. Basu, and J. S. Wilson (2010), The future of hydrology: An evolving science for a changing world, Water Resources Research, 46 (5), n/a–n/a, doi:10.1029/2009WR008906. Walker, B., C. S. Holling, S. R. Carpenter, and A. Kinzig (2004), Resilience, Adaptability and Transformability in Social ecological Systems, Ecology and Society, 9 (2), 5, doi: 10.1103/PhysRevLett.95.258101. REFERENCES 134

Walvoord, M. A., and R. G. Striegl (2007), Increased groundwater to stream discharge from permafrost thawing in the Yukon River basin: Potential impacts on lateral export of carbon and nitrogen, Geophysical Research Letters, 34 (12).

Wang, D., and X. Cai (2009), Detecting human interferences to low ﬂows through base ﬂow recession analysis, Water resources research, 45 (7), doi:10.1029/2009WR007819.

Wang, D., and M. Hejazi (2011), Quantifying the relative contribution of the climate and direct human impacts on mean annual streamﬂow in the contiguous United States, Water Resources Research, 47 (10).

Weiss, H., and R. S. Bradley (2001), What drives societal collapse?, Science, 291 (5504), 609–610.

Whipkey, R. Z. (1965), Subsurface stormﬂow from forested slopes, Hydrological Sciences Journal, 10 (2), 74–85.

Whitlock, C. H., L. R. Poole, J. W. Usry, W. M. Houghton, W. G. Witte, W. D. Morris, and E. a. Gurganus (1981), Comparison of reﬂectance with backscatter and absorption parameters for turbid waters., Applied optics, 20 (3), 517–522, doi:10.1364/AO.20.000517.

Wickham, H. (2009), ggplot2: Elegant Graphics for Data Analysis, Springer-Verlag New York.

Womack, J. M., and Others (2012), Evaluation of the hydrologic eﬀects of stock ponds on a prairie watershed, Ph.D. thesis, Montana State University-Bozeman, College of Engineer- ing.

Wooding, R. A. (1968), Steady Inﬁltration from a Shallow Circular Pond, Water Resources Research, 4 (6), 1259–1273, doi:10.1029/WR004i006p01259.

Xiankun, Y. (2014), Reservoir delineation and cumulative impacts assessment in large river basins: A case study of the Yangtze River Basin, Ph.D. thesis, National University of Singapore.

Xu, Y. D., B. J. Fu, and C. S. He (2013), Assessing the hydrological eﬀect of the check dams in the Loess Plateau, China, by model simulations, Hydrology and Earth System Sciences, 17 (6), 2185–2193.

Yair, A., and H. Lavee (1985), Runoﬀ generation in arid and semiarid zones, in Hydrological Forecasting, edited by M. Anderson and T. Burt, January, pp. 183–220, John Wiley \& Sons.

Yang, D., D. L. Kane, L. D. Hinzman, X. Zhang, T. Zhang, and H. Ye (2002), Siberian Lena River hydrologic regime and recent change, Journal of Geophysical Research: Atmospheres, 107 (D23). REFERENCES 135

Yitzhaki, S. (1998), More Than a Dozen Alternative Ways of Spelling Gini, Research on Economic Inequality, 8, 13–30, doi:10.1007/978-1-4614-4720-7.

Young, S., J. Peschel, G. Penny, S. Thompson, and V. Srinivasan (2017), Robot-Assisted Measurement for Hydrologic Understanding in Data Sparse Regions, Water, 9 (7), 494.

Zalasiewicz, J. A. N., and M. Williams (2010), The New World of the Anthropocene, Envi- ronmental Science and Technology, 44 (7), 2228–2231.

Zeng, R., and X. Cai (2014), Analyzing streamﬂow changes: irrigation-enhanced interaction between aquifer and streamﬂow in the Republican River basin, Hydrology and Earth System Sciences, 18 (2), 493–502, doi:10.5194/hess-18-493-2014.

Zhang, D. D., P. Brecke, H. F. Lee, Y.-Q. He, and J. Zhang (2007), Global climate change, war, and population decline in recent human history, Proceedings of the National Academy of Sciences, 104 (49), 19,214–19,219.

Zhang, D. D., H. F. Lee, C. Wang, B. Li, Q. Pei, J. Zhang, and Y. An (2011), The causality analysis of climate change and large-scale human crisis, Proceedings of the National Academy of Sciences, 108 (42), 17,296–17,301.

Zhu, Z., and C. E. Woodcock (2012), Object-based cloud and cloud shadow detection in Landsat imagery, Remote Sensing of Environment, 118, 83–94, doi:10.1016/j.rse.2011.10. 028.

Zlinszky, A., and G. Tim´ar(2013), Historic maps as a data source for socio-hydrology: A case study of the Lake Balaton wetland system, Hungary, Hydrology and Earth System Sciences, 17 (11), 4589–4606, doi:10.5194/hess-17-4589-2013.