ADAPTING TO UNCERTAINITIES: ASSESSING THE IMPACTS OF DAMS, CLIMATE CHANGE, AND RESOURCE SHIFTS ON FARMING LIVELIHOODS IN THE LOWER MEKONG REGION By Snehalata Sainjoo A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Fisheries and Wildlife – Doctor of Philosophy Environmental Science and Policy Program – Dual Major 2024 ABSTRACT The Lower Mekong region (LMR) is experiencing rapid social-ecological changes due to infrastructure development and climate change. The construction of dams along the Mekong River and its tributaries has triggered significant biophysical alterations, including changes in water flow, nutrient cycling, and sediment exchange. These changes have profound implications for the 86 percent of the Lower Mekong Basin (LMB) population who rely on the river for their livelihoods, primarily in agriculture and fisheries. Besides, the region faces an increasing frequency of extreme climate events, particularly droughts and floods. These combined pressures have heightened the vulnerability of resource-dependent communities, further exacerbated by unjust resource governance that often marginalizes them. This dissertation explores the multifaceted challenges faced by resource-dependent communities in the LMR, with a particular focus on their responses to the uncertainties driven by dams, climate change, and resource shifts. The study hypothesizes that most noticeable changes are occurring at the household level, influenced by their perceptions of climate risk and their capacity to access and utilize vital livelihood resources. These shifts exert pressures on land and watershed ecosystems, raising concerns over the sustainability of critical ecosystem services. Adopting a mixed-method approach, this research unfolds across three chapters: Chapter 1 assesses the impacts of dams on various livelihood resources in downstream farming communities at different distances from affected rivers, using unbalanced panel data analysis. The findings reveal a decrease in natural and financial resource accessibility, alongside a positive effect on physical resource accessibility post-dam construction. Communities closer to dams (<10 km) experience more pronounced negative effects on natural resources, while financial resource access improves for those within 20 km. Physical resources show spatial improvements, primarily within 10 km. Although no temporal effects are observed for social resources, spatial effects indicate reduced accessibility near dams. However, communities nearby irrigation dam experience increased social resource access post-dam construction. These findings highlight the spatial and temporal variations in resource impacts, emphasizing key areas for improving Environmental Impact Assessments (EIAs) and adaptation strategies to better support downstream ecosystems and communities. Chapter 2 delves into drought risk perceptions among households in irrigated and flood- pulse communities, exploring the factors explaining variations in risk perception at both household and community levels. Utilizing a mixed-effects model alongside qualitative information from interviews and observations, this chapter reveals the significance of psychological and socio-economic factors, including households' knowledge of drought, perceived ability, affiliation with various organizations, and wealth condition, shaping risk perception. These findings suggest the development of context-tailored risk communication and management strategies, enhancing the adaptive capacity of vulnerable communities. Chapter 3 examines the role of perceived peer effect and formal networks in shaping farmers' adaptive behaviors using influence network modeling. The findings suggest the need for leveraging both formal and informal networks, increasing knowledge about drought and improving economic conditions to build farmers’ capacity to navigate climate-related challenges and adopt practices that ensure the sustainability of resource-dependent farming communities in Cambodia and beyond. Copyright by SNEHALATA SAINJOO 2024 ACKNOWLEDGEMENTS I would like to express my deepest gratitude to the following people, without whom I would not have been able to complete this dissertation and earn this doctorate degree. First, I would like to thank my advisor, Dr. Daniel Boyd Kramer, for his unwavering support and mentorship. Dr. Kramer provided me with the intellectual freedom to explore and lead my own research, fostering an environment where I could confidently pitch and develop my ideas. His encouragement and trust in my abilities have been instrumental in shaping this dissertation, and I deeply appreciate his invaluable contributions. I would also like to extend my sincere thanks to my dissertation committee members, Dr. Maria Claudia Lopez, Dr. Amanda Flaim, and Dr. Mark Axelrod. Their thoughtful advice and expertise have been invaluable in refining this work. To the Mekong Culture WELL project team at Michigan State University and in Cambodia, especially Long Lyvouch, Sopheak Chann, and Alice Beban, thank you for your crucial assistance in field activities and data collection—without which I would not have had the content for my dissertation. I am also grateful to the many local community members who supported this research. Their participation and hospitality made the fieldwork possible, and I am truly grateful of their contributions. A special acknowledgment goes to my husband, Pradip Shrestha. Throughout my PhD journey, his unwavering belief and emotional strength have provided the foundation I needed to persevere through the challenges of this work. I could not have imagined completing this journey without him by my side. To my brother-in-law and mentor, Prabhakhar Shrestha, for his guidance since the beginning of my PhD journey. To my friends, Shubhechchha Sharma and Hiwa Rawal, thank you for your encouragement and for always being there to listen during challenging times. Your support has made a significant difference, and I truly appreciate your presence in my life. Finally, I would like to express my heartfelt gratitude to my family for their unwavering support throughout this journey. Your love, patience, and encouragement have meant everything to me. v TABLE OF CONTENTS INTRODUCTION .......................................................................................................................... 1 CHAPTER I: THE SPATIAL-TEMPORAL IMPACT OF DAMS ON DOWNSTREAM COMMUNITIES’ RESOURCE ACCESSIBILITY: CASE STUDIES FROM CAMBODIA ...... 5 CHAPTER II: HOW DO HOUSEHOLDS ASSESS DROUGHT RISK: INSIGHTS FROM IRRIGATED AND FLOOD-PULSE COMMUNITIES, CAMBODIA ...................................... 35 CHAPTER III: PERCEIVED PEER EFFECT ON FARMERS’ ADAPTIVE BEHAVIORS .... 69 CONCLUSION ............................................................................................................................. 95 REFERENCES ............................................................................................................................. 97 APPENDIX A: MANN-WHITNEY U AND INDEPENDENT T-TEST RESULTS ................ 117 APPENDIX B: MIXED EFFECT MODEL RESULTS ............................................................ 118 APPENDIX C: RESULTS FROM THEMATIC CODING ....................................................... 119 APPENDIX D: RESULTS FROM PROBIT AND POISSON MODELS, AND MARGINAL EFFECTS TESTING .................................................................................................................. 121 vi INTRODUCTION Climate change is arguably the greatest challenge facing the world today, with widespread and far-reaching impacts across various sectors including global water systems, agricultural production, human health, and energy resources. Its effects on river ecosystems are particularly complex and uncertain, further complicating water resource management (Ficklin et al., 2013). Human-induced climate change is expected to alter the hydrological regimes of numerous river systems, influencing water flow and sediment flux (Nguyen & Tran, 2024; B. Shrestha et al., 2013). Water infrastructure projects, such as dams, offer adaptation opportunities to meet the growing water demands for irrigation, industrial, and domestic consumption, while also playing a crucial role in energy production. Hydroelectric power is the largest source of renewable energy, accounting for 68 percent of the renewables, more than twice all other renewables combined (Barasa Kabeyi & Akanni. Olanrewaju, 2023). In addition to its energy benefits, hydroelectric power has a minimal carbon footprint and supports ecosystem services, such as water retention and soil conservation. However, while dams and reservoirs are viewed as instruments of economic gain and water security by many developing governments, their unprecedented development has significant impacts on ecosystems and rural livelihoods, particularly in the lower reaches of these structures (Beck et al., 2012). The Mekong River, a transboundary waterway in Southeast Asia, exemplifies these complex dynamics. The river is characterized by its unique and interconnected hydrological, ecological, and agricultural systems, which supported the livelihoods of approximately 65 million people (MRC, 2019). The upper region of the Mekong includes China and Myanmar, while the Lower Mekong Region (LMR) comprises Thailand, Lao PDR, Cambodia, and Vietnam. In the LMR, about 75% of the inhabitants rely on agriculture, which depend on seasonal rainfall, flood-replenished soils, and fisheries from rivers and lakes (Pokharel et al., 2018).However, these systems are increasingly at risk due to climate change, extensive dam construction, and land-use changes. Given the LMR’s sensitivity to climate extremes and disasters, particularly droughts and floods, the region is highly vulnerable to temperature increases and variations in seasonal rainfall (MRC, 2019b). Furthermore, unclear governance structures for water resource management, coupled with land-use changes, threaten ecosystem health and resilience, potentially 1 exacerbating water and food scarcity at the local level (Eastham et al., 2008). As farmers and fishers adapt, their changing livelihood practices, including resource utilization, may further strain watershed ecology and ecosystem services (Piesse, 2016). This dissertation explores the multifaceted challenges faced by resource-dependent communities and households in the Lower Mekong Region, focusing on their perceived risks and responses to uncertainties driven by dams, climate change, and resource shifts. Through a mixed- method approach, this research addresses these issues across three chapters (overview in Table 1), each focusing on different but interconnected aspects of the problem. Chapter 1 explores the study context by examining the relationship between dam construction and shifts in livelihood resources. It lays the foundation for the research by identifying critical resources—natural, social, physical, and financial—that are significantly impacted post-dam, affecting downstream communities in heterogeneous ways. By analyzing both spatial and temporal effects of dams on communities beyond 10 km from the impacted river sections, this chapter highlights dimensions of dam-related impacts on diverse livelihood resources that are often overlooked in studies focusing on the social effects of dam construction. The findings address the need for a more comprehensive understanding of how resource-based uncertainties, compounded by climate risks, shape household vulnerabilities in these regions. Chapter 2 shifts the focus from dam-induced changes to household perceptions of drought risk, maintaining the shared emphasis on water as a critical resource—whether controlled by dams or threatened by drought conditions. This chapter contributes to the limited literature on drought risk perceptions within complex hydro-agricultural-fisheries systems, where institutional support is often lacking. By exploring the variation in drought risk perception at both household and community levels, it examines how biophysical, psychological, socio- economic, and demographic factors shape these perceptions. The findings provide a deeper understanding of how farmers interpret and respond to climate uncertainties, particularly within the evolving resource landscape outlined in Chapter 1. Finally, Chapter 3 explores household adaptation decisions in relation to social networks, a critical resource impacted by dams (Chapter 1) and an important factor in shaping risk perceptions (Chapter 2). This chapter expands on the limited research concerning how perceived peer influence affects farmers' adaptive behaviors, particularly in regions with limited institutional support. It specifically examines the role of both informal peer influence and formal 2 networks in shaping adaptation decisions, while also considering other socio-economic, psychological, and experiential factors related to climate extremes. By addressing these interconnected factors, the chapter deepens our understanding of the social dynamics that influence adaptation in vulnerable farming communities. Together, these chapters provide the importance of considering both individual and community level factors when addressing the challenges posed by resource shifts and climate variability in the Lower Mekong Region. Table 1: Research Framework: Objectives, Methods, and Contributions to dissertation goal Objectives Study Methods Contribution to population dissertation goal Chapter 1: Farming Unbalanced mixed Identifies the key Assess the spatial and communities effect modeling at resources, particularly temporal impact of dams downstream to community level natural, social, and on downstream the multiple communities’ access to dams in diverse livelihood Cambodia resources financial, affected by dam construction, providing insights into resource- based uncertainties faced by farming households Chapter 2: Households Mixed effect modeling Reveals how household Identify key bio-physical, situated in the at household level and perceived drought risk socio-economic, irrigated and qualitative approach based on individual and psychological factors flood-pulse including narratives drought characteristics shaping drought risk communities, from informal in the context of perception and explore Cambodia interviews and causal irrigation dams and the pattern of drought severity by community type loop diagram at changing flood-pulse community level dynamics of Tonle Sap Lake 3 Table 1 (cont’d) Chapter 3 Rice farming Network influence Explains social dynamics and Examine the households, modeling at household decision-making processes effect of Cambodia level and qualitative underlying farmers' responses approach including to resource shifts and climate thematic coding and extremities, considering narratives from informal socio-economic, interviews and observations psychological, and experiential factors perceived peer influence and informal networks in shaping farm households’ adaptive behaviors The dissertation chapters used a case of Cambodia, which is relevant given its unique position in the LMB. Approximately 86 percent of Cambodia's territory lies within the Mekong Basin, making it heavily reliant on the river for water-related economic activities and domestic supply (Sithirith, 2021) . The Mekong and its connected Tonle Sap system are crucial for Cambodian livelihoods, providing essential resources for agriculture, fisheries, and urban development. Rainfed agriculture, which supports around 63 percent of Cambodia's population, contributes 22 percent of the nation's GDP as of 2022 (MoP, 2022). In addition, freshwater fish from the Mekong system provide up to 80 percent of the animal protein consumed in Cambodia, underscoring the river's pivotal role in ensuring food security (Hortle, 2007). The Tonle Sap, Southeast Asia's largest lake, acts as a natural flood buffer and supports vital fisheries and floodplain agriculture (Manohar et al., 2023). However, Cambodia’s downstream position and its dependence on the Mekong's seasonal flood pulse make the country particularly vulnerable to disruptions in the river's hydrology (Morton & Olson, 2018). This vulnerability is compounded by Cambodia’s limited resources and knowledge to effectively adapt to environmental changes. As such, Cambodia serves as a prime example of the complex interplay between resource dependency, climate vulnerability, and development challenges in the LMB, making it an ideal focus for this research. 4 CHAPTER I: THE SPATIAL-TEMPORAL IMPACT OF DAMS ON DOWNSTREAM COMMUNITIES’ RESOURCE ACCESSIBILITY: CASE STUDIES FROM CAMBODIA 1.1 Introduction In recent decades, dam construction has been prioritized in order to meet energy demands, support agricultural production, and foster industrial growth. Dams can mitigate drought conditions, control downstream flooding, and provide a year-round water supply by inundating upstream wetlands and riparian areas (Mulligan et al., 2020). Acknowledging these multiple benefits, many governments, particularly in the Global South, view dams as instruments for economic gain, water security, and poverty reduction (Wang et al., 2022). However, the rapid development of dams has raised concerns regarding their effects on hydro-ecology and dependent rural livelihoods, particularly for farmers and fishers in the lower reaches of these structures (Beck et al., 2012). For instance, studies have shown that dams can significantly alter river ecosystems, affecting fish migration patterns, sediment transport, and water quality, potentially leading to fish population declines and impacts on both commercial and subsistence fisheries (Arantes et al., 2022). While dams can enhance agricultural productivity through irrigation, they may also lead to changes in downstream water quantity and quality, potentially affecting traditional farming practices. Further, there is great uncertainty over the ability of dams to manage water demand under forecasted climate scenarios, with concerns of inducing drought (Di Baldassarre et al., 2018; S. Huang et al., 2021). The social impacts of dams, including displacement of communities, changes in local economies and cultural values, add more complexity to their assessment (Castro- Diaz et al., 2023; Fung et al., 2019; Richter et al., 2010). This all holds true for dams in Lower Mekong Region (LMR). Dams in the LMR Dam construction has been a national priority to LMR countries to spur economic growth while supporting food and energy security. The regional population is expected to grow by as much as 45 percent over two decades, with a projection of 15 to 25 percent by 2030 at the national level (Pokharel et al., 2018). Parallelling this growth, agricultural land expansion of 19-63 percent is anticipated, increasing water and energy demands and driving the planning of numerous dam projects along various reaches of the Mekong mainstream and its tributaries. 5 Recent studies suggest an unprecedented construction of dams, in total 1,055 in the Mekong including 608 hydropower dams and 447 non-hydropower dams for flood control, water supply, and irrigation (Ang et al., 2023). Hydropower projects from 1 MW and 4,200 MW have led to a dramatic increase in hydropower capacity from 1,242 MW in the 1980s to 69,199 MW post-2020s. Considering only 59 hydropower dams constructed before 2015, gross national economic value increased by around 265 percent (MRC, 2019b). Enticed by such financial incentives, most LMB countries, particularly Cambodia and Lao PDR, facing limited economic opportunities, are encouraged to develop an additional large-scale dams, with highest projected growth post-2020s (+18,223 MW) (Ang et al., 2023). As a developer and investor, China has been involved in this unprecedented rise in dam construction. Through its Belt and Road Initiative (BRI) and other investment channels, China has heavily invested in hydropower infrastructure across Southeast Asia (Urban et al., 2013; Wouters et al., 2024). Numerous of these initiatives are a part of China's broader strategy to safeguard energy resources and advance regional economic integration (Ganeshpandian, 2024). For countries like Laos and Cambodia, Chinese investments offer critical financial support to meet their growing energy demands and economic development goals (Kuik & Rosli, 2023; Siciliano et al., 2016). However these investments also raise concerns about fair resource distribution among communities and environmental sustainability (Soukhaphon et al., 2021a). Large water storage structures, such as dams and reservoirs, are expected to increase water and energy consumption in the region. This growth in demand will, in turn, place additional pressure on land and water resources (Smajgl & Ward, 2013). The situation is made worse by increased climate risks and unclear governance structures for water resource management, leading to inequitable resource distribution across communities. In response, community members are likely to modify farm livelihoods including farming practices and resource utilization, furthering pressure on aquatic ecosystems and their services (Siciliano et al., 2015). Piesse (2016), for example, observed many households increasingly relying on nearby forests for income after losing agricultural land. Similarly, lower fish stocks compel riverine communities to shift their dietary practices. Furthermore, Pokharel et al. (2018) found evidence of large-scale forest conversion to agricultural and pastureland, with increases of 10% and 16% respectively between 1992-2015 in the Mekong Basin. The rising trend of sand mining, as observed by Robert (2017) in Vietnam's Mekong Delta, poses additional risks to riverbed stability 6 and sediment flow. These changes have raised grave concerns over the sustainable use of water resources and the continued provision of ecosystem services to the people in the region, prompting the need for a better understanding of the socio-economic impacts of dams. Existing research and policy gaps The impacts of dams on hydro-ecological systems have been extensively studied in the LMR and other parts of the world. Numerous studies have documented changes in water levels, nutrient and sediment transport, soil and water quality degradation, and aquatic habitat deterioration, all of which affect freshwater vertebrates, fish diversity and populations (Bussi et al., 2021; Kuriqi et al., 2021; J. Li et al., 2013; Maavara et al., 2020; Soukhaphon et al., 2021). However, there remains a significant gap in understanding the social impacts of dams on affected communities over time and across spatial dimensions (Kirchherr et al., 2016). Existing studies have focused on various aspects of livelihoods such as agricultural production and income, health (Dillon and Fishman, 2019), social cohesion (Fung et al., 2019), labor and migration (Calvi et al., 2020), and structural and cognitive aspects of social capital (Mayer et al., 2022). However, these studies have concentrated on resettled and host communities, neglecting the impacts on downstream, resource-dependent communities. Further, the spatial extent of these studies is often limited to areas within a few kilometers of the dam site. Some exceptions include a study by Owusu et al. (2019) who examined the Bui dam’s post-dam effect on fisheries, farming and other livelihoods among downstream communities extending up to 30 km below the dam. Additionally, Fan et al. (2022) conducted a global study that found differential impacts of hydropower dams on greenness, population, economies, and other measures across communities at different impact zones, ranging from less than 5 km to 50 km. While some studies have captured temporal dynamics of dam-induced impacts on different livelihoods, they often fail to provide a comprehensive assessment of all livelihood resources post dam construction. Arthur et al. (2020) captured locals’ responses to changes in different livelihood capitals after the construction of the Bui dam. Similarly, a study by Castro-Diaz et al. (2023) reported the positive and negative effects of hydropower dams on natural, social, human, financial, and physical capital. However, both studies did not empirically test the temporal aspects of those effects. Many Environmental Impact Assessments (EIAs) for dam projects fail to adequately assess downstream effects, cumulative impacts, and the full range of socio-economic 7 consequences (Richter et al., 2010). While they often focus on upstream areas and resettled populations, they frequently overlook downstream communities that depend on river ecosystems for their livelihoods. This gap is concerning, as communities far from dam sites still experience significant disruptions to fisheries, agriculture, and water access (Moore et al., 2010). This issue is particularly relevant in the Mekong region, where EIAs often neglect transboundary impacts on downstream communities (Soukhaphon et al., 2021a). While upstream areas may receive compensation, those downstream are frequently excluded from assessments (Kirchherr et al., 2016). Furthermore, EIAs tend to focus narrowly on environmental factors without adequately considering how dam projects affect different dimensions of livelihoods—such as natural, financial, social, human, and physical capital—which limits policymakers' understanding of the full socio-economic consequences (Baird & Frankel, 2015). In response to these shortcomings in existing policy and research, this study seeks to examine the spatial and temporal impacts of dam impacts on accessibility to diverse livelihood resources. Research Question and Hypotheses Our central research question is how resource accessibility varies across downstream communities at different impact zones (i.e., space) over time (i.e., pre- and post-dam construction)? In our study, "impact zone" refers to areas at different proximities to dammed rivers and therefore likely experiencing different environmental and socio-economic changes. "Resource accessibility" refers to the ability of farming communities to access and utilize various resources namely natural, physical, social, human, and financial capital to support their livelihoods. Specifically, we tested the following hypotheses (Figure 1): H1: Resource accessibility changes significantly over time among downstream communities with variation observed in both pre- and post-dam periods. H2: Proximity to dams significantly influences resource accessibility with closer communities experiencing greater disparities and more pronounced changes in resource accessibility compared to those located farther away. 8 Figure 1: Visualization of the research hypothesis 1.2 Materials and method 1.2.1 Theoretical Framework: Sustainable Livelihood Framework (SLF) The Sustainable Livelihood Framework (SLF) provides a theoretical structure for examining the complex impacts of dams on community livelihoods across spatial and temporal dimensions. This widely used framework assesses key livelihood components and contextual factors, including internal and external influences (also known as vulnerability context), which shape livelihood decisions and outcomes. The SLF is particularly relevant to our study as it helps explain how combinations of diverse livelihood resources (also called capital assets) enable households and communities to adopt various livelihood strategies such as cultivation, inland fishing, aquaculture, diversification, and migration, in response to changing environmental conditions. Access to and the transformation of these resources into different livelihood strategies are critical determinants of farm households’ adaptive capacity (Sok & Yu, 2015). In our study, we emphasize livelihood resources, not only the resources that support farmers' and fishers' livelihoods but those enabling control and agency within a system influenced by state, market, and civil organizations (Allison & Ellis, 2001; Bebbington, 1999; Scoones, 1999). Adopting this framework (Allison & Ellis, 2001; Bhandari, 2013), our study focuses on changes in access to five types of livelihood resources in the context of dam construction and operation: 9 Natural: This includes stocks of natural resources such as land, water, forests, and fish, which are critical for resource-dependent populations. Maintaining these assets at sustainable levels is crucial for long-term livelihood security (Reed et al., 2013). Dams can significantly alter these resources, affecting their quantity and quality across different spatial zones. Physical: Basic infrastructure like roads, markets, water supply systems, schools, and banks build communities' capacity to reduce social and environmental vulnerability (Bebbington, 1999; Y. He & Ahmed, 2022; Scoones, 1999). Dam projects often introduce new infrastructure, potentially changing access patterns across impacted areas. Financial: Measured in terms of cash, credit/debt, and savings, financial resources are essential for pursuing livelihood strategies. Dam construction can alter local economies, affecting income sources and financial stability in various ways across different communities. Human: Encompassing household labor, skills, knowledge, and health, human capital enables participation in discussions and negotiations, influencing development discourses (Bebbington, 1999). Dams can enhance education and foster the development of new skills by creating new social and physical infrastructure, which can vary across different spatial and temporal scales. Social: Comprising networks (formal and informal) and associations, social capital supports building trust and reciprocity, enhancing access to resources and adaptive capacity (Adger, 2003). Dam projects can disrupt or create new social structures, potentially altering community resilience and livelihood security. 1.2.2 Study area Relying on Cambodia’s Agricultural Census Data from 2014 and 2019, we considered only dams that were constructed and commissioned between 2013 and 2019. Based on this criterion, we identified three dams from the dataset of the Dams of the Greater Mekong (Mekong Region Futures Institute, 2020): the Ajhang, Battambang, and Lower Sesan 2 dams (Figure 2). Table 2 provides the brief overview of dams. 10 Figure 2: Study dam sites 11 Dams Description Table 2: Dam description Ajhang Ajhang dam, located in Kampong Chhang Province, Cambodia, was constructed between 2014 and 2018 by the Ministry of Water Resources and Meteorology, with funding from the Chinese government. Situated 41 km from the city and 120 km northwest of Phnom Penh, it lies 22.4 km away from the highway. The primary aim of the project was to irrigate 10,300 hectares of land along the left bank of the river, achieved through the construction of the main and branch canals. Additionally, the dam facilitates diversion and junction activities, featuring elements like entrance gates, drainage brakes, and connectors between the left and right banks. The dam stands at a height of 28 meters and utilizes undershot water gates. However, concerns arise due to the absence of a fish pass and the fast, shallow flow over the spillway. The catchment area for the dam is the Stung Baribour river. Source: Green et al. (2019) Battambang Battambang is a multipurpose dam, constructed between 2014 and 2018. It serves various functions including hydropower generation, irrigation system enhancement, flood control, and water supply to Battambang city. Located in Ratanak Mondul district’s Plov Meas commune, it stands at a height of 49.5 meters, creating a reservoir capable of storing 286 million cubic meters of water. With the Sangke River as its catchment area, the dam facilitates water distribution through main and sub-canals. This infrastructure supports the irrigation of 47,000 hectares of wet paddy and 12,000 hectares of dry paddy across three districts: Ratanak Mondul, Banan, and Mong Russey. In terms of power generation, the dam boasts an installed capacity of 24 MW, contributing to an annual energy production of 123 GW. Source: Vida (2017) 12 Table 2 (Cont’d) Lower The Lower Se San 2 Dam, a hydropower dam, was constructed between Sesan 2 2014 and 2017 on the Se San River in Stung Treng Province. With an installed capacity of 480 MW and an average annual energy production of 2311.8 GW, it stands at a height of 45 meters. Located 25 kilometers east of Stung Treng city, its construction and operation have significant environmental and social impacts. It is estimated to displace up to 5,000 people and affect the livelihoods of over 38,675 individuals, including indigenous communities. Furthermore, the dam is likely to disrupt fish migrations, impacting fisheries resources for approximately 78,000 people living in the vicinity. Source: Sifton, (2021) Sampling population Previous works (e.g., Lin and Qi (2019)), estimated the maximum impact of dams on nearby land use and land cover in the upper and lower Mekong River to be within 5 kilometers on average. However, Richter et al. (2010) argue that impacts may extend further downstream, reaching up to 10 kilometers. Expanding this perspective to the global south context, Fan et al. (2022) considered the impacts of hydropower dams, focusing on changes to population, urban development, greenness, and GDP, finding significant effects within a radius of 50 kilometers from dam sites, with the most notable changes occurring within 5 to 20 kilometers. Given these findings, we selected a 20-kilometer downstream river section (referred as “impacted river segment”) to evaluate the impacts of dams on downstream communities. While this acknowledges the immediate effects of dam construction on the surrounding ecosystem, we anticipated substantial shifts in various livelihood resources including natural, social, human, economic, and physical resources, beyond the dams' immediate vicinity. We used a systematic approach to identify potential villages located downstream of each dam (unit of analysis) (Figure 3). Using ArcGIS Pro, we began by applying the "Track Downstream" geoprocessing tool to identify the river segments extending from each dam. This approach aligns with techniques used in dam break analysis and flood inundation mapping studies, where tools like HEC-GeoRAS are employed to process geospatial data for hydraulic modeling (Beza et al., 2023; Hagos et al., 2022). 13 Figure 3: Flow diagram showing the steps taken for study population selection To comprehensively understand the full extent of dam impacts, we created three buffer zones (referred as potential impact zones) around delineated impacted river segments (from step 1). These zones were set at 10 km, 20 km, and extended up to 30 km from the impacted river segments, as outlined in step 3, allowing identification of all villages falling within each zone. To add a vertical dimension, we utilized the SRTM 30m resolution digital elevation model to assign elevation values to each identified village (step 4). We then compared these elevations to the respective dam elevations, classifying villages as either above or below the dam's height, selecting those villages at elevations lower than the dam, as the downstream area of influence (step 4). In total we sampled 433 and 50 villages from 2013 and 2019 respectively within 30 Km from the impacted river segment for analysis (Table 3). Table 3: Sample size Dam Altitude Type Sample village Sample village Ajhang Battambang Lower Sesan 2 (masl) 35 51 59 Total Irrigation MPD HP 2013 255 143 35 433 2019 26 17 7 50 14 1.2.3 Variable description a. Dependent variable We developed 5 indices that capture communities’ accessibility to natural, physical, social, financial and human resources respectively, selecting 22 variables as proxy measures to develop these indices (Table 4) following the process listed in Figure 4. The process of selecting the 22 proxy variables involved several stages starting with initial identification of potential measures from a review of twenty peer reviewed articles. We reviewed studies focused on agrarian livelihoods and adaptation within South and Southeast Asia. Drawing on the SLF, which emphasizes access to natural, financial, human, physical, and social capitals for sustainable livelihood practices, we identified key factors—both barriers and motivators—related to adaptation decisions and adaptive capacity. Figure 4: Flow chart showing the process of variable identification Focusing on this region allows for a context-specific exploration of how farmers adjust their resource accessibility and livelihood strategies in response to environmental change, such as those driven by dam construction. After reviewing the literature, we validated the identified measures using field information collected in 2022 through informal interviews and observations. Finally, we cross-checked these measures against available data from the Cambodia Agricultural Census of 2013 and 2019, finalizing the list of variables for analysis. 15 Table 4: List of variables used for building indices Indices Measures Indicator Unit2 Questions1 Source Natural Access to N1: Engagement in Proportion Did you or any of your household members (Bui & Do, 2021; land for cultivation engage in any crop cultivation activity Kasem & Thapa, various during the last 12 months? livelihood N2: Ownership of land Proportion What is the land tenure? 2011; N. A. Khan et al., 2021; Yang activities N3: Parcel for Proportion land use of the parcel for livestock in the et al., 2020; C. livestock N4: Parcel for aquaculture last 12 months Zhang et al., Proportion land use of the parcel for aquaculture in the 2020) last 12 months N5: Parcel for Proportion Do you have any forest and other wooded forest/wooded land lands that are part of this agricultural holding as of this day? Access to N6: Engagement in Proportion Did your household engage in any own- water fish catching account fishing activity (any catching of fish and aquatic products) in the inland or marine water during the last 12 months? Access to N7: Engagement in Proportion Did your household engage in any own- forest forestry account forestry-related activities in the last 12 months? 16 Human Access to H1: Total household Mean Total Number of Members as of (Abid et al., 2016; Table 4 (cont’d) laborers members available for number day of visit economic activities Access to H2: HH head with Proportion education formal education Access to H3: Age of HH head Years What is the age of HH head? knowledge H4: Access to Proportion Did you receive or access any Shrestha, Chaweewan and Arunyawat, 2017; Salaisook, Faysse and Tsusaka, 2020; Jin et al., 2021; Khan et al., 2021; Vo, Mizunoya and skills agricultural agricultural information that helped and Nguyen, 2021) information you manage the agricultural holding during the last 12 months? Financial Wealth F1: TLU Mean TLU3 (Jin et al., 2016; N. A. Khan et al., 2021; Access to F2: Access to Proportion Did you avail the credit/loan during Naqvi et al., 2020; N. credit credit/loan for the last 12 months for agricultural T. T. Pham et al., 2019; agricultural purpose purpose? Salaisook et al., 2020; Access to F3: Household using Proportion Did you avail the credit/loan during Vo et al., 2021) banking facilities bank the last 12 months? Bank Access to F4: Household using Proportion Did you avail the credit/loan during microfinance Microfinance for the last 12 months from these credit sources? Microfinance 17 Table 4 (cont’d) Physical Access to P1: Household with Proportion Did you or your household use (Jin et al., 2021; N. A. irrigation irrigation facilities irrigation in this holding during Khan et al., 2021; N. T. facilities the last 12 months? T. Pham et al., 2019; Access to P2: Household using Proportion Did you bring and Yang et al., 2020) market market either for sale or sell your agricultural for information produce/products in this nearest market from your holding/house during the last 12 months? Access to P3: Household with Proportion Household members’ currently school members attending attending school? school Social Access to S1: Relatives/friends Proportion Did you avail the credit/loan (Bui & Do, 2021; Jin et informal S2: money lender Proportion during the last 12 months from al., 2021; Kasem & networks for credit these sources? Thapa, 2011; I. Khan et al., 2020; N. T. T. Access to S3: Main source of Proportion where did you receive or access Pham et al., 2019) network for agricultural information- the agricultural information? agricultural farmers information S4: Main source of Proportion agricultural information- Government 18 Note: Table 4 (cont’d) 1. These are the questions used to collect information in the Census of Agriculture in Cambodia 2013 and Inter-Censal 2019 at household level. 2. The dataset was initially recorded at the household level. To conduct community-level analysis, further data processing was performed in R using ` dplyr` package, where household-level variables were aggregated to the community level using appropriate summary statistics (e.g., means, proportions) depending on the nature of the variable. 3. Total Livestock Unit: TLU represents the weighted sum of domestic animals owned. TLU is a standardized metric of total livestock owned using a weighted value for each livelihood species as provided by FAO (2005) in the Cambodian context: 0.65 for cattle, 0.7 for buffalo, 0.1 for sheep and goats, 0.25 for pigs, and 0.01 for chickens/poultry. 19 We applied principle component analysis (PCA) to build indices representing access to the five livelihood capitals for both years 2013 and 2019. PCA is a widely used data reduction technique that identifies patterns in large datasets (Tareq et al., 2021) and is commonly used for developing capital indices (Y. He & Ahmed, 2022; Xu et al., 2023). This method was applied to the selected proxy variables for each capital (Table 4), with the first principal component for each livelihood capital used to create the indices, representing the most significant underlying structure of the data. b. Independent variables Dam effect (X1) To capture the temporal changes between the pre- and post-dam construction periods, we created the dummy variable Year (1 = 2019, 0 = 2013). This variable serves as a key indicator for the shifts in resource accessibility that can be attributed to the dam over time (H1). Spatial-temporal effect (X2) To assess the spatial variation in the dam's impact, we developed two interaction terms, each capturing the interaction between time (pre- and post- dam) and the impact zones (Zones 1 and 2), with Zone 3 as the reference for comparison. These interaction terms evaluate how the dam's effects differ across spatial zones (H2). Additionally, we explored dam effects by its type (X3) (e.g., hydropower, multi-purpose and irrigation). We controlled for other contextual factors including nightlight effect using Night Time Lights data at 2014 and 2019 (Elvidge et al., 2021; VIIRS Nighttime Light (VNL), 2021), which serves as the proxy for socio-economic and infrastructural development (Bargain et al., 2023; Quan et al., 2023; Kocornik-Mina et al., 2020; Singhal et al., 2020) that can influence resource accessibility (Fan et al., 2022). 1.2.4 Analytical approach We employed a multi-step analysis, beginning with testing differences in resource accessibility before and after dam construction using a combination of non-parametric tests, followed by employing a mixed effect model to test our hypotheses. Finally, we also included an analysis for both balanced and unbalanced data to ensure robustness and accuracy in the findings. Preliminary tests Given the non-normality of the data, we first applied non-parametric tests, including Mann-Whitney tests at 95 % confidence intervals for proxy measures of each index (Calvi et al., 2020). Then, we conducted independent t-tests to determine the direction of 20 change with significance level. After developing indices representing different livelihood capitals, we performed Wilcoxon Signed-Rank Test (matched), to each of the indices to evaluate significant differences in resource access pre- and post-dam periods (Abbott et al., 2022). Mixed-Effects Model We employed mixed-effects, unbalanced panel data analysis to assess the effects of the dam. To evaluate the effects on each type of livelihood resource, we developed candidate sets of linear mixed effects models, two models (random and fixed effect) corresponding to each of the five livelihood resources (using the plm package in R) (Figure 5). Given non-normal data distribution of indices value (based on Shapiro-Wilk test using an α level of 0.05), we used log-transformed values for each index value in the model. Figure 5: Models specific to each resource access We applied a random effects model for all five indices considering several factors. First, the random effects approach accounts for unobserved heterogeneity between villages, capturing village-specific effects that may not change over time but still influence resource accessibility. Second, the model allows for time-invariant variables, such as the impact zone (proximity to the dam) and dam type, to remain in the analysis, providing estimates for their coefficients. This is important as it offers insights into how dam proximity and dam function type affect resource access over time. Additionally, the random effects model provides an average effect for each village, giving us a broader view of the dam’s impacts while controlling for both observed and unobserved factors. This approach also ensures that village-level variations and other factors that do not change over time are accounted for, making the model robust. 21 Robustness of the estimates To ensure the robustness of estimates, we ran random effects models with the full sample (unbalanced panel data, N= 433) and common sample (balanced panel data, N=50). We checked any inconsistencies in the results based on effect size (coefficient value) and direction of association, standard error value and confidence level. In case of difference in effect size, we reported the coefficient value with small standard errors and high confidence levels. We reported estimates using percentage changes, derived from the exponentiation of 𝛽. 1.3 Results 1.3.1 Findings of preliminary tests Non-parametric Mann-Whitney tests and independent t-tests revealed significant changes in resource access for several proxy measures across all livelihood capitals before and after the post-dam periods. Specifically, 18 out of 22 measures showed significant differences at the 95% confidence level (p < 0.05) over time. Notably, social, physical, and financial resource access measures exhibited moderate to large effect sizes (Cohen's d ≥ 0.5) (see APPENDIX A Table 10). For instance, access to bank services (Cohen's d = 1.38) and credit (Cohen's d = 0.80), as well as access to informal networks—such as friends (Cohen's d = 1.38) and money lenders (Cohen's d = 0.91)—demonstrated significant shifts. Additionally, access to information from government sources exhibited a particularly pronounced change (Cohen's d = 1.92), indicating substantial alterations in resource access across these categories. The Wilcoxon Signed-Rank Test further validated these results, showing notable differences in all indices over time, with the exception of human capital (p > 0.05), as illustrated in Table 5. 22 Table 5: Summary of proxy indicators for each index and distribution of the indices value, representing access to different livelihood resource, before and after dam construction across 50 matched sampled downstream communities Indices Indicator 2013 2019 Distribution of indices showing median value Mean (SD) Mean (SD) (Green color indicates 2019 and orange indicates 2013) Natural N1: Engagement in cultivation N2: Ownership of land N3: Parcel for livestock N4: Parcel for aquaculture N5: Parcel for forest/wooded land N6: Engagement in fish catching N7: Engagement in forestry 0.841 (0.233) 0.972 (0.049) 0.696 (0.260) 0.001 (0.017) 0.129 (0.243) 0.265 (0.306) 0.002 (0.014) 0.9 (0.01) 0.971 (0.053) 0.632 (0.171) 0.788 (0.306) 0.048 (0.068) 0.247 (0.298) 0.869 (0.196) 23 Human H1: Total household members available for economic activities H2: HH head with formal education H3: Age of HH head H4: Access to agricultural information Financial F1: TLU F2: Access to credit/loan for agricultural purpose F3: Household using bank F4: Household using Microfinance for credit 4.652 (0.654) 0.902 (0.169) 48.598 (4.961) 0.357 (0.114) 1.5 (1.06) 0.175 (0.065) 0.147 (0.203) 0.488 (0.292) Table 5 (cont’d) 4.19 (0.52) 0.805 (0.110) 46.737 (3.374) 0.361 (0.155) 5.664 (16.035) 0.474 (0.165) 0.329 (0.248) 0.525 (0.245) 24 Physical P1: Household with irrigation facilities P2: Household using market either for sale or for information P3: Household with members attending school Social S1: Main source of credit- Relatives/friends S2: Main source of credit- money lender S3: Main source of 0.243 (0.329) 0.218 (0.051) 1.132 (0.300) 0.253 (0.248) 0.201 (0.226) 0.306 agricultural information- (0.310) farmers S4: Main source of 0.573 agricultural information- (0.334) Government Table 5 (cont’d) 0.32 (0.133) 0.006 (0.022) 0.195 (0.110) 0.01 (0.027) 0.047 (0.083) 0.263 (0.174) 0.099 (0.102) 25 These preliminary findings highlight the nuanced impact of temporal changes on different aspects of resource access. However, to understand these shifts in relation to dams, we further explored the causal relationships in the subsequent mixed-effects model discussed in the following section. 1.3.2 Findings of mixed effect model The random effects models using the unbalanced panel data explained a substantial portion of the variance in access to livelihood resources, with adjusted R² values exceeding 0.61 for all resources except financial (APPENDIX B Table 11). In contrast, when using balanced panel data, R² values decreased to below 0.20 for most resources, except for natural resources ((APPENDIX B Table 12). This reduction is primarily due to the smaller sample size and reduced variability in the balanced dataset. We reported results from both balanced and unbalanced panels for comparison (APPENDIX B Table 11 and Table 12); however, we focus on the unbalanced panel estimates ( Figure 6) due to its larger sample size, which provides more precise estimates with lower standard errors. The consistency of coefficient estimates across both panels indicates that our findings are robust. Nonetheless, the increased statistical power in the unbalanced panel allows for more reliable detection of statistically significant effects. Post-dam effect on resource access: The models revealed significant impacts of dams on downstream communities' access to diverse livelihood resources, mainly natural, financial and physical with varying directions of effect. Access to natural resources decreased by 36% (β = -0.44, SE = 0.06, p < 0.001) and financial resources by 32% (β = -0.38, SE = 0.06, p < 0.001) compared to the pre-dam period (2013). Conversely, access to physical resources increased by 31% (β = 0.27, SE = 0.04, p < 0.001). Spatial-temporal changes in resource accessibility We observed significant spatial variation in resource access, particularly for natural and social capitals, across communities located at different impact zones. Communities within 10 kilometers from the dam-impacted river segment (Zone 1) showed 2% less access to natural resources (β = -0.09, SE = 0.03, p < 0.01) compared to the communities located beyond 20 kilometers (Zone 3). Communities situated between 10 to 20 kilometers from the dam-impacted 26 river segment (zone 2) demonstrated 8% less access to social resources (β = -0.08, SE = 0.04, p < 0.05). We did not observe any variation in resource access among communities closer to the dam post- dam construction (impact zones 1 and 2), except for financial resources, when compared during the pre-dam period. The estimates indicate an increase in access to financial resources for communities located within 10 kilometers of the dam-affected river segment (impact zone 1), by approximately 35% (β = 0.3, SE = 0.08, p < 0.001). Similarly, communities situated between 10 to 20 kilometers from the dam-impacted river segment (zone 2) experienced a 34% increase (β = 0.29, SE = 0.09, p < 0.001) compared to the pre-dam period. Standardized estimates Figure 6: Results from mixed effect models 27 Dam type and variation in resource access: When comparing variation in resource access among communities across different dam types, we found a significant 12% decrease in access to social resources for communities located near the Ajhang Irrigation Dam (β = -0.13, SE = 0.03, p < 0.001), compared to those near the Battambang and Lower Sesan 2 dams. However, after dam construction, we found increase in access to social resources in these communities (β = 0.24, SE = 0.09, p < 0.01) compared to the pre-dam period. In contrast, we observed decrease in access to natural resources by 24% (β = - 0.27, SE = 0.07, p < 0.001) post-dam construction. We did not find any significant changes for physical, financial, or human resources based on dam types. 1.4 Discussion This study reveals the complex and multifaceted impacts of dams on downstream communities’ access to different livelihood resources including natural, financial, physical and social. The findings highlight that the impact of dams varied by dam type and that these impacts extend beyond the immediate proximity of the dam structure (i.e. beyond 10 km), contributing to the growing body of literature on the social impacts of dams. 1.4.1 Spatial-temporal changes in resource accessibility Natural resource The most prominent changes observed post-dam construction are a significant decrease in access to natural resources, which aligns with global studies on the impacts of hydropower dams (Arthur et al., 2020; Castro-Diaz et al., 2023; Fan et al., 2022; Nhung, 2017a). Furthermore, our findings reveal spatial variation in natural resource access across communities located in different impact zones. Specifically, communities located closer to dams (<10 km) experience more pronounced negative effects compared to those farther away (>20 km). This pattern corresponds with existing studies that highlight how the effects of dams, on greenness (Fan et al., 2022), changes in forestland, grassland and cultivation land (Lin & Qi, 2019; Zhao et al., 2013), and alterations in water and sediment flow (Richter et al., 2010), are often concentrated in areas nearest to the dam. This reduction in natural resource access is critical, especially in rural communities with predominantly natural resource-based livelihoods (Y. He & Ahmed, 2022). Scholars have reported negative effects on common-pool resources like rivers, forests, and pastureland, leading to water scarcity, degraded soil quality and reduced fish production—factors that directly affect 28 locals’ primary livelihoods, such as cultivation and fishing (Arthur et al., 2020; Owusu et al., 2019). These observations are in accord with our findings, where we found a decrease in the proportion of households using land parcels for livestock and forest-based activities, and fishing, with a corresponding increase in aquaculture and crop cultivation (see APPENDIX A Table 10). This shift indicates not only a change in land-use patterns but also a direct response to reduced access to natural resources such as forest products and fish. While communities continue to access forest resources, the sharp decline in fishing activities underscores the broader environmental changes that have followed dam construction. This mirrors findings from other studies in the LMR where fisheries have suffered due to altered hydrological regimes, impacting both fish populations and community livelihoods (Soukhaphon et al., 2021). The increased reliance on alternative livelihood activities, such as aquaculture, intensified land use, and heightened dependence on forest resources, suggests that communities are adapting to the changing availability of natural resources (Piesse, 2016; Robert, 2017). Similar shifts in local livelihoods post dam construction are reported, for example, after construction of the Kamchay Dam in Cambodia (Siciliano et al., 2015) and the Ghana’s Bui dam in Ghana (Owusu et al., 2019). Financial resource The significant decrease in communities’ access to financial resources post dam construction suggests that dams can negatively impact local economies, especially those that rely heavily on natural resources. The study suggest that natural capital serves as the foundation for securing other resources, including financial, particularly in natural resource-dependent livelihoods such as fishing and farming (Xu et al., 2023). However, when observing the spatial changes in financial resource access over time among downstream communities at different impact zones, our findings reveal an increase in financial resource access for those located within 10 Km (zone 1) and between 10-20 Km of the dam impacted river segment. This contrasting trend suggests that while there is an overall reduction in financial access post-dam construction (for communities within 30 km), certain communities closer to the dam may experience unique opportunities that improve their financial access. Castro-Diaz et al. (2023) also identified both positive and negative effects of hydropower dams on financial resources, suggesting that these impacts can be context-specific. 29 The increase in financial access in impact zones 1 and 2 could be attributed to economic activities such as agricultural intensification and aquaculture, which are likely facilitated by improved access to irrigation infrastructure, banking services, microfinance, and credit (Arthur et al., 2020; Green, 2020; see APPENDIX A Table 10). However, it is important to mention that this financial benefit may be limited to only a few communities (Fung et al., 2019) within these zones that are better positioned to leverage new opportunities arising from infrastructural changes, a factor not explored in this study. Furthermore, unlike the decrease in natural resource access, which is concentrated in areas closer to the dam, the effects on financial resources appear to extend beyond the 10 km radius. This suggests that the economic impacts of dam construction may persist and spread over time, affecting wider geographical area (Goodman, 2024). Physical resource Not surprisingly, there is an increase in access to physical resources after dam construction, which is consistent with most studies that show how dams often bring improved physical infrastructure such as roads, energy access, schools, health, markets, and irrigation systems to nearby communities (Beck et al., 2012b; Castro-Diaz et al., 2023; Hensengerth, 2018). However, there is disproportionality in resource access as our estimates suggest greater physical resource access in communities closer to the dam (Zone 1). This finding aligns with studies suggesting that communities in proximity to infrastructural projects often receive the greatest benefits. For example, Aung, Guido and Stijn (2017) observed differential benefits of the Swar Dam project among paddy farmers in the south-central part of Myanmar due to unfair and untimely distribution of the irrigation water by the irrigation department. They found 50 percent of head and middle end-users experienced sufficient water availability while tail end- uses reported water shortages. Social resource We found a significant effect on access to social resources post- dam construction among downstream communities located within 10-20 km of affected river segments (zone 2), suggesting potential disruptions to their social networks. This concern was raised by Tilt, Braun and He (2009) in their review of dam-induced displacement and resettlement. Existing studies on changes in social capital, though limited, primarily focus on cognitive aspects such as conflict, loss of traditional ceremonies, and participation (Castro-Diaz et al., 2023; Fung 30 et al., 2019). Structural aspects, like relations with neighbors, church attendance, and having relatives in the city, have received less attention (Arthur et al., 2020; Mayer et al., 2022).These studies, mostly centered on resettled and host communities, consistently report negative effects of both large and small dams on social resources. For instance, Arthur et al. (2020) found that communities perceived adverse effects of the Bui Dam on their informal ties and community connections. While there is a lack of research capturing the spatial variation in access to social resources, some scholars suggest delayed and indirect impacts of dams on social aspects for communities farther downstream (Richter et al., 2010). Zone 2 communities might experience these delayed effects more acutely as they adapt to gradual changes in their environment and livelihoods. These impacts can be related to the observed decrease in natural resources in Zone 2 and potential changes in livelihoods, which may lead to a breakdown of traditional social structures and support systems, for example a decrease in availability of relatives/friends and money lenders for credit and agricultural information on average for sampled communities in general (see APPENDIX A Table 10). Human resource The study did not find significant spatial and temporal changes in access to human resources post-dam construction. However, when examining specific measures such as access to household labor for livelihood activities, a significant decrease over time was observed, aligning with findings from previous research. For instance Calvi et al. (2020) reported a decrease in family labor due to outmigration after the Belo Monte dam construction in the Brazilian Amazon. Similarly, Owusu et al. (2019) observed outmigration among low-income groups due to disproportionate distribution of amenities like electricity and improved water supply following Ghana's Bui dam hydroelectricity project. Further studies require considering a broader range of human resource measures. We also recommend exploring migration patterns and their relationships to dam construction. 1.4.2 Dam Type and Resource Access The varying impacts of different dam types on resource access contribute to a nuanced understanding of dam effects. Interestingly, despite an overall decrease in social resource accessibility among sampled communities, we found a 27% increase in access to social resources among communities downstream of the Ajhang Irrigation Dam, compared to those near 31 hydropower and multi-purpose dams. This occurs despite these communities experiencing a slight 2% average decrease in natural resource access. This finding suggests complex social dynamics that may be associated with factors such as increased crop cultivation at the expense of fishing or forest parcels, increased interaction among farmers, and community cooperation due to shared irrigation infrastructure. These results align with Dillon's (2011) study in Mali, which found that small-scale irrigation projects led to increased agricultural production and informal food sharing through collective management practices. The contrasting effects of different dam types highlight the need for context-specific assessments. For instance, Kirchherr, Pohlner and Charles (2016) emphasized that the socio- economic impacts of dams vary significantly based on their purpose, size, and local context. 1.4.3 Policy implications This study contributes to a more nuanced understanding of dam impacts, exploring the differential impact of dams on diverse livelihood resources across downstream communities at different dam proximity. This granularity is often underrepresented in existing studies on the social impacts of dam construction. Our findings provide insights for improving Environmental Impact Assessments (EIAs) associated with dam construction and enhancing adaptation programs, particularly in regions affected by dam construction. Comprehensive, Resource-Specific Assessments The findings on differential impacts across various livelihood capitals (natural, social, physical, and financial) emphasize the need for EIAs to extend the range of resource categories beyond natural and economic. This would ensure that all facets of community livelihoods are protected and supported. Adaptation programs should also be designed with a multi-dimensional approach, targeting the diverse ways communities use diverse resources instead of limiting to natural resources. Incorporating Spatial Differentiation in EIAs The study's identification of distinct zones of impact for diverse resources suggests that future EIAs should move beyond broad, one-size-fits-all approaches. Policy should consider spatial differentiation, ensuring that assessments consider varying effects based on different proximity to dam-impacted rivers and other contextual factors. This would provide more localized data on environmental and social impacts, allowing for tailored mitigation strategies. Such differentiated EIAs would improve accuracy in identifying affected downstream 32 communities and resources, guiding fair compensation mechanisms and targeted interventions to mitigate negative impacts and enhance positive outcomes of dam construction. Long-term Monitoring and Temporal Dynamics Current assessments often focus on short-term impacts, but the temporal dynamics observed in this research indicate the need for ongoing evaluations to fully understand the evolving effects of dams on downstream communities. By institutionalizing long-term monitoring, policymakers can ensure that adaptation measures remain responsive and effective over time. 1.4.4 Limitations and Future Research While our study provides valuable insights into the impacts of dams on livelihood resources, it is important to acknowledge several limitations and propose directions for future research: 1. Unbalanced panel data: We reported estimates from unbalanced panel data. Although the samples were drawn from the same population in both 2013 (488 observations) and 2019 (50 observations), the significant difference in sample sizes raises potential concerns about nonrandom missingness. The smaller sample size in 2019 may introduce bias if certain characteristics are underrepresented. Future research should aim to achieve more balanced sampling across time periods to reduce potential biases and ensure more robust comparisons. 2. Limited sample of dam types: This study includes only three dams, each of a different type. This limited sample may not fully capture the differential impacts across various dam types. Future research should incorporate a larger number and diversity of dams to provide a more comprehensive understanding of type-specific impacts. 3. Limited measures of livelihood resources: Due to data constraints, we used a limited set of measures to construct our resource indices. Future studies should incorporate additional measures such as household energy access, health effects, labor movement, cognitive aspects of social capital (e.g., trust, norms, and shared values). These additions would provide a more nuanced understanding of changes in physical, human and social resources, facilitating better comparisons with existing literature. 4. Need for qualitative research: Our quantitative findings could be complemented by qualitative research exploring the lived experiences of affected communities. In-depth 33 interviews and focus group discussions could provide rich contextual information and help explain some of the quantitative patterns observed. 5. Contextual factors: While we used nightlight intensity as a comprehensive measure of socio-economic conditions, future studies could control for additional contextual factors such as population growth, climate extremities and trends, government policies and interventions. 34 CHAPTER II: HOW DO HOUSEHOLDS ASSESS DROUGHT RISK: INSIGHTS FROM IRRIGATED AND FLOOD-PULSE COMMUNITIES, CAMBODIA 2.1 Introduction The Lower Mekong Region (LMR), including the connected Tonle Sap Lake and its downstream floodplain communities, faces increasing risk to climate change. According to the International Panel on Climate Change (IPCC, 2014), key risks are characterized by high- frequency climate-related hazards, minimal capacity to adapt, and persistent socio-economic vulnerabilities such as poverty (Oppenheimer et al., 2014). These criteria are highly relevant to the LMR where rapid hydrological, ecological, and socio-economic changes are occurring due to dam construction, industrial farming, and forest plantations (Pokharel et al., 2018; Spruce et al., 2020). These developments have increased the vulnerability of ecosystems and dependent livelihoods as well as exacerbating conflict, inequality, economic stresses, and poverty (Baird & Barney, 2017; Beban et al., 2017; Sok & Yu, 2015). Such vulnerable systems and communities are highly susceptible to climate change (Oppenheimer et al., 2014). Of particular concern is the increasing frequency, duration, and severity of droughts over the past two decades, with notable drought events in 2004-2005, 2009-2010, 2016, (Adamson & Bird, 2010; MRC, 2019a) and 2019-2020 (Keovilignavong et al., 2023). Climate projections suggest that rising temperatures, prolonged dry spells, and lower river flows will worsen drought conditions over the next 30 to 90 years, threatening the livelihoods of 70-80 percent of the region’s subsistence farmers and fishers (MRC, 2019). While dams and irrigation infrastructures are intended to mitigate drought risks, irregular water distribution, especially during wet and dry seasons, and the absence of benefit-sharing mechanisms have worsened water scarcity (Nhung, 2017). This has disproportionately affected downstream countries like Cambodia leading to increased crop losses (Yamsiri, 2014), reduced fish yields due to altered water and nutrient flows (Yoshida et al., 2020), and water conflicts within communities (Aung et al., 2017). The populations affected by these changes are likely to perceive high drought risks to their farms and livelihoods, particularly under resource constraints. This perception can influence their responses such as sand mining, exploiting groundwater and land resources (Piesse, 2016; Robert, 35 2017) as well as their acceptance of climate change adaptation measures (Bastakoti et al., 2014; Dang et al., 2014). Some behaviors can worsen drought conditions and exacerbate system vulnerabilities. Thus, it is crucial to understand local perceptions of drought risk, which is limitedly addressed in existing studies of risk perception in relation to climate change, dams and interconnected hydro-agricultural-fisheries system like the LMR. 2.1.1 Risk perception studies Studies of risk perception gained prominence in the 1970s and 1980s when scholars in psychology, for example, Paul Solvic and Daniel Kahneman, identified inconsistencies between the scientific community’s and the public’s understanding of risk and its influence on decision- making. Initially, the focus was on technological hazards such as nuclear power risks (Slovic, 1987). However, the need to understand individuals’ perceptions of risk has since been widely adopted in behavioral science and climate change studies, where false perceptions of lower risk and high security with existing practices can exacerbate vulnerability if individuals fail to adapt (Cardona et al., 2012). A large body of research explores how people perceive risks related to natural hazards such as floods (Botzen et al., 2009; Kellens et al., 2013; Siegrist & Árvai, 2020). Yet, empirical studies assessing perceived drought risk at the household level are limited, and most are concentrated in developed countries such as Spain (Urquijo & De Stefano, 2016) and the Netherlands (Duinen et al., 2015). In the Global South, some studies have examined farmers' perceptions of climate change risks more broadly as seen in Vietnam (Dang et al., 2014) and China (Tang et al., 2013a). These studies, however, tend to approach risk perception from a unidimensional perspective, focusing a limited number of characteristics of such hazards—either the probability of future water scarcity or the severity of its consequences on livelihoods. Research grounded in the psychometric paradigm argues that feelings of dread or affect are primary determinants of public perception and acceptance of risk (Slovic et al., 2012; Tang et al., 2013). This is particularly relevant for hazards like drought which are characterized by their indistinct and uncontrollable nature, and uneven, delayed impacts (Slovic, 2016). Such arguments suggest the need to adopt a more comprehensive approach to risk assessment, incorporating the multi-characteristics of hazards—such as the perceived probability of occurrence, severity of consequences, and emotional responses like fear or worry (Walpole & Wilson, 2021; Wilson et al., 2019). 36 Considerable research has explored the individual correlates that shape risk perception such as socio-economic, demographic, and psychosociological factors, which account for variation at the individual level. However, studies that integrate both individual factors and broader contextual elements—particularly those addressing variability at the community level— are limited, providing limited understanding of drought risk. To address this gap, we examined how households in irrigated and flood-pulse communities in Cambodia assess drought risk using a holistic approach. Specifically, we aimed to answer the following questions: 1. Do households’ perceptions of drought risk vary within and across communities? If so, 2. What explains the heterogeneity in risk perception at the household level and by community type? 2.1.2 Drought and individual characteristics shaping risk perception The notion of drought in tropical monsoon regions like LMR does not fit naturally with conventional perception, generally associated with consistently low and marginal rainfalls in the arid and semi-arid region, such as the Sahel in Africa or central Australia, where drought is more natural and permanent. Further, unlike other climatic hazards, for example floods, drought is a slow-onset phenomenon without a distinct end. It also stays longer and results in cumulative and broader impacts (Bachmair et al., 2016). Thus, drought severity and its impacts depend not only on its duration, intensity, spatial extent and site-specific characteristics (for example, soil properties, rainfall, hydrological flow), but also on socio-economic activities and the adaptive capacity of the social systems it affects (Adamson & Bird, 2010). This unique context of drought in the LMR significantly influences how individuals perceive and respond to drought risks. The slow onset and gradual intensification of drought conditions can lead to delayed recognition of the problem, potentially resulting in inadequate preparedness among local populations. Moreover, the cumulative and wide-ranging impacts of drought in the region affects agriculture, fisheries, and water resources. As populations are repeatedly exposed to drought effects, their perceived risk is likely to be shaped by the interplay between evolving drought conditions and individual factors including situated contexts (Siegrist & Árvai, 2020). Therefore, our primary hypothesis is that households’ risk perception is heterogenous, explained by bio-physical, experiential, socio-psychological, and geographical characteristics. 37 Biophysical Factors Existing studies on flood and landslide risk perception have established a relationship between biophysical properties of the environment and risk perception, for example, the effects of elevation and proximity to water bodies (Botzen et al., 2009; Ho et al., 2008; Kellens et al., 2013). Due to the ambiguous nature of drought, risk is often shaped by exposure to the biophysical characteristics of the environment in which households are embedded. For example, living in areas without external water source creates resource constraints like water scarcity (Tang et al., 2013a; Urquijo & De Stefano, 2016). Thus, we hypothesize that H1: Households located in the areas without an external water supply perceive greater drought risk compared to those with access to such a supply. Experiential Factors Direct experiences with stressors such as financial losses from drought are known to positively influence risk perception and precautionary behavior (Kellens et al., 2013; Tang et al., 2013; Wachinger et al., 2013). Thus, we expect that H2: Households experiencing drought related financial damage in the past perceive greater risk today. Socio-Psychological Factors Planned behavioral theory suggests that individuals collect and process information as they interact with others, shaping their perceptions (Schlüter et al., 2017). Such ties are found to amplify direct experiences with risks (Wachinger et al., 2013). Households’ social interactions can trigger memories of their past experiences of hazard that affected their livelihoods. Such discussions can evoke emotions and awareness that make risks feel more immediate and significant, ultimately heightening perception of risk (Tang et al., 2013a).Thus, we hypothesize that H3: There is a positive relationship between household’s social ties for information and their perception of drought risk. Additionally, households' connections to various local institutions can provide diverse forms of support, such as credit, seeds, and training. This support can enhance households' adaptive capacity to manage risk (Adger, 2003; Cassidy & Barnes, 2012), potentially lowering perceived risk. Thus, we expect that 38 H4: Households’ association to any community organizations lower their perceived risk of drought. Studies have examined the importance of the psychological attributes of household in shaping risk perception. Perceived trust in authorities and confidence in protective measures can lower perceived risks as demonstrated by studies in Vietnam, China, and Spain (Tang et al., 2013a; Urquijo & De Stefano, 2016). Thus, we expect that H5: Household’s trust in existing water management boards/committees and infrastructures controlling drought risk can decrease drought risk perception. Shi et al. (2016) found the positive relationship between the level of knowledge about the causes and consequences of climate change and public concern about climate change. Similarly, Botzen et al. (2009) reported that individuals with little knowledge about the causes of flood events lowered the perceptions of flood risk in Netherlands, also supported by other risk perception studies (Siegrist & Árvai, 2020). Thus, we expect that H6: Households with little knowledge of the causes and consequences of drought (‘subjective knowledge’) have lower drought risk perceptions. A household's perceived ability to manage drought impacts—shaped by their economic and personal resources—can significantly influence their perception of risk (Van Duinen et al., 2015). However, the relationship between perceived ability and risk perception may vary depending on the type of hazard. For instance, a study by Ho et al. (2008) in Taiwan found that while landslides typically affect small areas, affected households often adopt precautionary measures such as evacuation, which they perceive as effective for controlling outcomes, resulting in fewer casualties over time. In contrast, flood-affected households may remain in vulnerable areas and implement some precautionary measures, yet they often face persistent financial losses that diminish their perceived control. This reduction in perceived control can escalate their risk perception. Drought, characterized by its widespread and long-term effects, presents a unique challenge compared to both landslides and floods and therefore its effect on perceived drought risk might vary depending on their circumstances. While perceived ability has been linked to lower perceived risk, household wealth can moderate this relationship. Studies have shown that greater wealth is associated with lower perceived risk for financial investments. Additionally, perceptions of relative wealth, rather than 39 absolute wealth, can influence an individual's willingness to engage in risky behavior (Fehr & Reichlin, 2021), thereby increasing their perceived ability (Slovic, 1987; Slovic & Weber, 2002), which in turn can lower their perceived risk. Thus, we hypothesize that H7: The relationship between perceived ability and risk perception varies across different levels of wealth, with greater wealth enhancing perceived ability over drought risk Geographical factors Evidence suggests variability in risk perception across populations located at different geographical locations, for example, rural vs urban in Netherlands (Wachinger et al., 2013); communities with ground water vs surface water in Spain (Urquijo & De Stefano, 2016), flood prone urban communities with different population sizes in Pakistan (Rana et al., 2020). In our study, we examined the effect on perceived drought risk based on community location, those in proximity to irrigation canals and those in proximity to Tonle Sap Lake. We assume that households located closer to irrigation canals have better access to water for cultivation than those farther from the canals. Therefore, we hypothesize that H8: Households in irrigated communities perceive a lower drought risk compared to households in flood-pulse communities. Demographic factors Risk perceptions vary across different demographic groups, necessitating the inclusion of age, education, and sex as explanatory variables. However, the direction of the association can vary depending on the study context or hazard type. For instance, studies have found negative effects of education on risk perception among populations facing multiple hazards in Pakistan and rural communities facing flood hazards in the Netherlands (Ali et al., 2022; Botzen et al., 2009). On the other hand, some studies have found the positive effects of education on risk perceptions among urban communities facing flood hazards in Pakistan (Rana et al., 2020). There are few studies reporting a negative relationship between age and risk perception (Botzen et al., 2009; Tang et al., 2013a). Similarly, very few studies have established a significant effect of gender on risk perception. For example, Dang et al. (2014) and Ho et al. (2008) have found a lower risk perception among males in Taiwan and Vietnam respectively. However, they did not find an effect of age and education on risk perception. These findings suggest no clear direction. 40 Thus, we did not set any hypotheses regarding associations between these factors and risk perception. However, we do include demographic characteristics as control variables. 2.1.3 Significance of this study This study makes several significant contributions to the risk perception literature in the context of climate change, dams, and connected hydro-agricultural-fisheries systems like the LMR: 1. It adds to the limited literature on drought risk perception by examining various factors shaping household risk perceptions including biophysical, experiential, socio-economic, psychological, geographical, and demographic. 2. It supports a multi-dimensional understanding of risk perception including perceived probability, perceived severity of consequences for farms and livelihoods, and affective feelings. 3. It employs a holistic, mixed-methods approach, combining quantitative analysis to identify factors explaining risk perception variability at household level, and qualitative methods to provide deeper insights into the causal relations between factors explaining heterogeneity by community type. This approach provides a comprehensive understanding of drought risk perception. 2.2 Materials and method 2.2.1 Study area The Tonle Sap, known as the "Great Lake" of Cambodia, is the Southeast Asia's largest freshwater lake. It plays a crucial role in supporting interconnected hydrological, agricultural and fisheries systems and dependent livelihoods (Keskinen, 2006), The lake’s climate is characterized by warm temperatures and distinct seasonal patterns, heavily influenced by seasonal monsoons. The lake has a unique hydrological system resulting from alternating bi- seasonal flows. During the wet season (between June-October), the Mekong River floods and reverses the flow of the Tonle Sap River, replenishing the lake and expanding its surface area sixfold. Whereas during dry season, the Lake feeds water to the Mekong River via the Tonle Sap River. Thus, drought conditions are most common during the dry season, particularly in March and April, often exacerbated in years of low rainfall (Oeurng et al., 2019). The lake’s surface area varies from 2,500 km² during the dry season to 15,000 km² during the wet season (Matsui et al., 2005; MRC/WUP-FIN, 2007). These fluctuations profoundly affect 41 the surrounding floodplains, influencing both agricultural and fisheries productivity in the lake and Mekong Delta. The lake supports the livelihoods of nearly half of Cambodia's population (Bonheur, 2001). Although it is well-known for its exceptional fish production, the lake’s floodplains also significantly contribute to Cambodia’s rice production, accounting for over 90% of paddy cultivation alongside the Mekong Delta region (MRC, 2014). In recent years, the lake ecosystem has been greatly affected by climate change and human activities, particularly dam construction. Evidence of significant environmental changes include more frequent drought conditions, fluctuating water levels, and rising temperatures (Nuorteva et al., 2010; Yoshida et al., 2020), further threatening the swamp forests and fish nurseries that provide a substantial portion of Cambodia's protein (Lovgren, 2020). Fish populations decreased by over 87% between 2003 and 2019 (Seng, 2020), while shifts in water and sediments dynamics have diminished rice yields (World Bank, 2023). These changes have had profound impacts on lake-dependent livelihoods, particularly for fishers in floating villages and rain-fed farmers. Although the Mekong River Commission (MRC) and the government has prioritized drought management and established committees to address these issues, implementation at the community level remains limited due to insufficient information, expertise, and institutional capacity (MRC, 2019). 2.2.2 Study population We studied 14 communities near by Tonle Sap Lake (Figure 7) as part of the Michigan State University (MSU) and University of Nevada, Reno (UNR) collaborative Mekong project. The villages were selected purposively selected based on village history interviews with community leaders (January 2023). Sample selection criteria included 1) proximity to the Achang irrigation dam completed in 2013 and Tonle Sap Lake, 2) majority proportion of households engaged in either farming and fishing, and 3) local livelihood concerns related to current socio-environmental challenges (e.g., land access, farming issues, decline in fisheries, relocation, migration) and a diverse set of responses to those issues in communities. 42 Figure 7: Study area showing sampled irrigated (orange box) and flood-pulse communities (blue box) Eight of our study communities are geographically situated across the Tonle Sap floodplain, hence referred to as flood-pulse communities. Ou Ta Prok, Anglong Reang, Srey Chak, and Kampong Prak are located in the Pursat Province and Doun Sdaeung, Pov Veuy, Peacha Krey, and Peam Ban are situated within Beung Tonle Chhmar Ramsar and Kampong Thom Provinces. Most residents live in floating houses and rely primarily on fishing during the wet season (June-October). However, they also engage in floodplain farming and wage labor activities during the dry season (November-May) and are therefore vulnerable to the rapid ecological and hydrological changes, such as seasonal variation in water and floods (Keskinen, 2006). The other six study communities are at different proximities to the Ajang dam, completed in 2013 and are within 15 Km of Tonle Sap Lake. We refer to these communities as irrigated communities including Trapeang and Tang Thnuem farthest upstream, Tapang and Pou Mreach in intermediate positions, and Chhunk Tru and Seh Slab farthest downstream and nearest Tonle 43 Sap Lake. Community members from the four communities farthest upstream are primarily from the Khmer ethnic group and engage in rice farming. The downstream communities of Chhunk Tru and Seh Slab are comprised of Cham, Vietnam, and Khmer ethnic groups and rely on fishing and wage labor. However, due to relocation efforts, some practice crop farming. 2.2.3 Research design We employed a concurrent mixed methods approach (Figure 8) , collecting both quantitative and qualitative data simultaneously. In this design, quantitative data were gathered to explain household-level understanding of drought risk, while qualitative data provided insights into the risk dynamics by community type. These datasets were analyzed separately and then integrated in the combined Results and Discussion section 2.3, offering a broad understanding of how households perceive drought risk within the broader community dynamics. Figure 8:Concurrent mixed methods research design (Adopted from Guest & Fleming, 2015) 2.2.4 Data collection We collected primary data between May and July 2022 utilizing a household survey, informal interviews, and field observation. Our study is approved by the Institutional Review Board (IRB) of Michigan State University (STUDY00004770). We obtained verbal consent from all participants prior to the start of data collection. We randomly sampled 703 households for the household survey, approximately 50 households from each village. We employed tablets equipped with the Open Data Kit platform (ODK) to collect data (Hartung et al., 2010). A trained research team from Cambodia conducted face-to-face 90 minutes interview with each household head using a pretested structured and 44 semi-structured questionnaire to collect information covering household’s demographic, income and debt, material wealth, experiences to multiple shocks (such as, drought, flood, pest and diseases), drought perceptions, social networks and relations, and livelihoods including detailed information on farming, fishing, aquaculture, and livestock. Simultaneously, the MSU research team collected qualitative information using ten informal interviews with village chief, local officials and local farmers and fishers residing nearby irrigated and flood pulse communities. We followed an exploratory approach with no predetermined selection criteria for participants. We approached individuals in each village who were knowledgeable about community practices and environmental challenges and willing to share their perspective on infrastructure development and related hydrological and ecological changes and their effects on fisheries and farms including changes to rice cropping and fishing practices, climate change, drought, coping and adaptation strategies, and local support. The interview process was flexible and inductive, allowing for open-ended conversations that helped uncover community-specific insights into these themes. 2.2.5 Variable description a. Dependent variable Definition Drawing on existing scholarship pertaining to farmers’ risk perceptions to climate change hazards, we define perceived drought risk as the individual’s subjective judgment based on the interaction of drought and individual characteristics. This conception aligns with the definition outlined in the Intergovernmental Panel on Climate Change (Field et al., 2014). Measurement Existing studies on risk perception–– grounded in different disciplines and focused on different hazard types––apply different approaches to measure risk perception. For instance, Ho et al. (2008) conceptualized risk perception as the interaction between hazard characteristics and stakeholder attributes, usings measures of likelihood, severity (including financial damage, life threat, and dread), knowledge of mitigation measures, and perceived controllability. Other studies on flood risk perception, such as those by Botzen et al. (2009) and Wilson et al. (2019a), have used Likert-scale measures to assess perceived probability of occurrence of hazard, either in absolute or in relative terms, and perceived risk to financial damage. In climate risk studies, perceived risk is often defined through the lens of vulnerability, incorporating measures of sensitivity, exposure, and coping capacity (Conde et al., 2008). There 45 are few studies on drought and water scarcity. Some studies, like Tang et al. (2013), have used survey items related to personal knowledge of resource status and future scarcity to measure perceived risk, while others, such as Urquijo & De Stefano (2016), have used qualitative measures, for example, the drought impacts and resource constraints faced by different water user groups. Duinen et al. (2015) provides a case of drought risk perception in the Netherlands using measures of perceived probability of occurrence of extreme dry periods, severity to financial damage, and dread of drought. This multi-dimensional aspect of risk perception was also empirically tested across other hazard types (behavioral, technological, and climate) and was significant in explaining people’s risk perception and predicting self-protective behavior (Wilson et al., 2019b). Thus, our study adopts this multidimensional framework to assess the risk perception. Mathematically, it is expressed as 𝑃𝑒𝑟𝑐𝑒𝑖𝑣𝑒𝑑 𝑟𝑖𝑠𝑘 = 𝑃 × 𝐶 (𝐶1 + 𝐶2) × 𝐴------------Equation 1 Here, probability (P) denotes households’ perceptions of the likelihood of drought occurring in the future. Consequences (C) refer to their perceptions of the severity of drought in the future in terms of financial damage (C1) and general livelihood effects (C2). Affect (A) refers to households’ concern or emotion regarding future drought events. Detailed information on the survey items used to capture these measures and the corresponding responses are provided in Table 6. Table 6: Survey items for assessing drought risk perceptions Component Questions Source Probability (P) How likely is it that you and your farm (cropland and Adapted from livestock) will experience drought or water shortages in (Ho et al., 2008; the near future?1 Responses in Likert scale (1-5) 1= less likely to 5= Very likely Liu et al., 2018; Van Duinen et al., 2015) 1 We deliberately used both "drought" and "water shortages" in the question to ensure clarity. Previous studies suggest that locals' definitions of drought often vary, with many associating it with changes in water availability or water levels in rivers or reservoirs and hence used water scarcity to incorporate drought risk (Tang et al., 2013a). By including both terms, we aimed to capture a broader understanding and ensure that respondents accurately interpreted the question. 46 Table 6 (cont’d) Consequences (C): latent variable constructed using two observed measures, C1 and C2 C1: If drought or water shortages occur in the near future, Adapted from how severe will your crop loss be?2 (Van Duinen et Response in continuum of crop damage and/or loss in a scale al., 2015) (1-5) 1: None/insignificant loss 2: less than half 3: Half 4: More than half 5: Almost all C2: When you think about the possibility of drought or water Adapted from shortages in the near future, what are the likely effects on you (Dang et al., and your farm?3Select all that apply. Multiple options: 2014; Udmale et al., 2014) C2.1: Financial loss, might be related to crop loss or market demand/price fluctuation C2.2: Food insecurity C2.3: Threatened health C2.4: Threatened social relationships C2.5: Increased debt C2.6: Reduced or even leave farm activities and look for other alternatives C2.7: Lost opportunity in the labor market C2.8: Even migrate for other alternatives 2 This question is relevant to crop farmers, particularly households residing within the irrigated communities in our study context. However, we argue that responses from households embedded in flood-pulse communities are also relevant as they are seasonally involved in crop farming. 3 A majority of studies used direct and immediate effects, for example, financial loss in general (Botzen et al., 2009; Duinen et al., 2015a; Ho et al., 2008) or in terms of crop loss or income (Osiemo et al., 2021) as the measure of the severity of climate hazards. Whereas few studies capture indirect or long-term severity (Kellens et al., 2013), except for a climate risk study in Vietnam (Dang et al., 2014). Considering the invisible nature of drought with both immediate and long-term effects, in addition to severity of crop loss, we included a question that captures the long- term consequences of drought to livelihoods in general affecting health, social, and financial status. 47 Affect (A) How worried are you about the effects of drought on you and Adapted from your farm in the near future? Response in Likert scale (1-5) (Rana et al., Table 6 (cont’d) 1= Less to 5= High 2020; Van Duinen et al., 2015; Wilson et al., 2019b) Risk perception score To calculate a drought risk perception score, we applied confounding factor analysis (CFA), a widely used approach in risk perception studies, particularly when there is an existing underlying structure for validation (Johnson & Kim, 2023; Walpole & Wilson, 2021; Wilson et al., 2019). Following this scholarship, we developed a three- factor model using two observed variables (P and A) and one latent variable (C), as shown in Figure 9. The observed variables (indicated in square boxes) are those directly measured, while the latent variables (indicated in ellipses) are not directly observed or measured and are inferred from a set of observed variables (C1 and C2). To validate our three-factor modeling approach, we also conducted an exploratory factor analysis. We began by checking the basic assumptions for factor analysis, including the Measure of Sampling Adequacy (MSA) using the Kaiser-Meyer-Olkin (KMO) test, and the significance of the multiple correlations using Bartlett’s test of sphericity. The calculated MSA (0.6) suggests that the sample adequacy is moderate but acceptable (Kaiser, 1974). Significant results at p-value <0.05 from Bartlett’s test of sphericity indicates that the variables included are suitable for factor analysis (Bartlett, 1951). Then, we determined that three factors should be extracted using parallel analysis. All analyses were conducted using the psych package in R. 48 Figure 9: Risk perception model Following statistical validation, we used a three-factor model structure that conceptualizes perceived risk as a function of probability, consequences, and affect. We first built the measurement model associated with "consequences (C)" using observed measures (C1 and C2). We then constructed the risk perception model using a three-factor framework: P, C, and A (Figure 9). We utilized the lavaan package in R (Rosseel, 2012) and employed the weighted least squares mean and variance adjusted (WLSMV) estimator for both models (C.-H. Li, 2016). The goodness-of-fit indices (χ² = 322.699, p < 0.001, CFI = 0.85, TLI = 0.82, RMSEA = 0.09) suggest a reasonable fit between the risk perception model and the data (Hu & Bentler, 1998). Finally, we extracted the standardized risk score using the first principal component eigenvalues. Subsequently, we categorized households' perception of risk into four quintiles, corresponding to low, medium, high, and very high levels of perceived risk. b. Independent variables We included a set of key predictors that could explain the heterogeneity in risk perception at the household level. Details on hypotheses specific to each predictor are provided in sub- section 1.3. 49 Biophysical factors include ‘presence of external water source’, which was measured as a binary variable based on whether the household reported having access to an external water supply for drinking and household use (Yes=1), for example private and communal tap, well, and market store. Experiential factors include households’ previous experience with drought, which was measured as a binary response to whether household members and their farm faced drought in the past few years (Yes = 1, No = 0). Social factors include ‘social ties’ for farming related information and ‘membership to any organizations. § Social ties for information were measured as a binary response to whether households received information related to cultivation, fishing and livestock from formal or informal ties (Yes=1, No=0). § Membership to any organizations was measured as a binary response to whether household members were members to any of the groups or organizations: farm-association-or- cooperative, fish-association-or-cooperative, women's organization, water-use organization, irrigation organization, seed bank, religious group, political group, or credit or finance group (Yes=1, No=0). Economic factors We used a wealth index as a proxy measure of economic condition. The wealth index is constructed using Principal Component Analysis (PCA) (Córdova, 2008; Howe et al., 2008) based on nine observed measures. These measures include the total number of durable assets owned by the household; total number of farming and fishing equipment; access to an improved source of drinking water, an improved source of fuel for cooking and lightening, toilet facilities (1=flush latrine, 0=0 = pit over lake or share or none); amount of land owned (ha); and total livestock units (TLU). A higher value indicates higher wealth. TLU represents the weighted sum of domestic animals owned. TLU is a standardized metric of total livestock owned using a weighted value for each species as provided by FAO (2005) in the Cambodian context: 0.65 for cattle, 0.7 for buffalo, 0.1 for sheep and goats, 0.25 for pigs, and 0.01 for chickens/poultry. Psychological factors include ‘subjective knowledge’, ‘trust’, and ‘perceived ability.’ 50 § Subjective knowledge of drought is measured using an index value created using principal component analysis (PCA) based on measures of drought causes (i.e., insufficient or irregular rainfall, excessive water use, upstream dam construction, and inefficient water distribution) and drought effects (i.e., reduced food supply, decreased income, conflicts over water, lowered river or lake water levels, decreased groundwater availability, and reduced water quality) (Shi et al., 2016). Responses to each cause and consequence, for example, is insufficient rainfall as the cause of drought, was recorded as binary response (Yes = 1, No = 0). Overall, a higher index value indicates greater subjective knowledge of drought within the household. § Trust in government agencies was measured as a binary variable (1 = Yes, 0 = No) based on household responses to whether existing water governance organizations are effective in managing droughts. § Perceived ability was measured using a Likert scale in response to the question, "Earlier, you indicated how drought might affect you and your farm. How confident are you in your ability to deal with those problems?" with responses ranging from 1 = not confident to 5 = very confident (adapted from Gebrehiwot & van der Veen, 2015). We also considered geographical factors (referred as ‘community type’) including household’s proximity to either a nearby irrigation canal (Irrigated=1) or flood-pulse zone (Flooded=0). Further, we controlled for household demographic factors including household head age (years), sex (1=Female, 0=Male), education (any formal education: Yes=1, No=0) and ethnicity (1 = Khmer, 0 = Non-Khmer). Table 7 provides summary statistics for all predictors. Table 7: Summary statistics for the model predictors Predictors Mean/ SD Min Max Frequency (%) 0 0 1 1 X1: Presence of Yes =1 external water No =0 source X2: Social ties Yes =1 No =0 18.3% 81.7% 85.3% 14.7% 51 Table 7 (cont’d) X3: Membership to Yes =1 any organization No =0 X4: Wealth index X5: Previous Yes =1 experience to No =0 drought X6: Subjective knowledge X7: Perceived ability X8: Trust in Yes =1 government No =0 16.6% 83.4% 0 81.9% 18.1% 4.1 1.8 10.7% 89.3% X9: Community Irrigated =1 28.4% type Flooded= 0 71.6% X10: Age (Years) X11: Sex Female =1 Male=0 X12: Education Yes =1 No =0 X13: Ethnicity Khmer =1 45.4 15.6% 84.4% 72.5% 27.5% 93.0% Non-Khmer =0 7.0% 1 1 1 13.4 0 1 -2.6 0 0 1 0 0 20 0 0 0 2.8 1 7.5 5 1 1 86 1 1 1 2.2.6 Analytical approach Quantitative We applied a mixed effect model, a two-level hierarchical linear model (HLM) using the lme4 package in R to examine the effects of the predictors at the household level on household’ perceived risk while accounting for community-level variation. HLM is typically utilized with nested data and accounts for the possibility of covariate effects at different levels on the response variable. For our data, the household risk perception score (i) is nested within communities (j=14). We followed guidelines of the general multilevel model by Steenbergen & Jones (2002) and Snijders & Bosker (2011) for model identification. We began by conducting a random effect 52 ANOVA analysis at the 5% level of significance (Null model). We found the random intercept value significant, indicating significant variability in risk scores across communities. We followed by computing the interclass correlation coefficient (ICC) value. The calculated ICC value of 0.132 suggests that approximately 13.2% of the total variability in the perceived risk score is accounted for at the community level, suggesting strong consideration of a multilevel modelling approach (Hedges & Hedberg, 2007). We proceeded with the random intercept fixed- effect model where we considered the variability between communities, measured by a random intercept, but we fixed the effects of household level predictors. Compared to the typical OLS regression model, model estimates are reliable and address type I errors and ecology fallacy (Snijders & Bosker, 2011; Steenbergen & Jones, 2002). The fitted full model (M1) is specified as 𝑦!" = 𝛾## + ∑ %& $’% 𝛽$ 𝑋$!" + u#" + 𝜀!" ----------Eq (1) Where yij denotes the household level perceived drought risk score, the household level (level 1) predictors are denoted as Xkij (l = 1, 2, . . . , 15), and eijk ∼ N(0, σe2) is the error term. γ00 is the regression intercept or grand mean yij measuring perceived risk while controlling for the effects of all levels 1 predictor variables (see detail list in Table 2). 𝛽$ is the effect for each predictor variables on perceived risk. u0j is random intercept for the community j (level 2) which capture variability between communities; whereas 𝜀!" is the error term for individual i within community j), captures variability at household level. In our study, we included the interaction between two predictors- wealth index (X4) and perceived ability (X7). Thus, the fitted full model with interaction (M2) is specified as 𝑦!" = 𝛾## + ∑ (%& $’% 𝛽$ 𝑋$!" + 𝛽%)(𝑋)!" × 𝑋*!") + u#" + 𝜀!" ----------Eq (2) Where, β11 is unstandardized coefficient for the interaction term 𝑋)!" × 𝑋*!". The significance of the models is tested using a likelihood ratio test (LRT). Thus, each time, the null model is compared with alternative models including a full model with 13 predictors (M1) and full model with an interaction term (M2). We also calculated Akaike’s Information Criteria (AIC) to identify the model that provides the best explanation of households’ perceived risk. Further, we calculated the proportion of variation explained at level 2 compared to our null model, using the equation: 53 𝐴𝑑𝑑𝑖𝑡𝑖𝑜𝑛𝑎𝑙 𝑣𝑎𝑟𝑖𝑎𝑡𝑖𝑜𝑛 𝑒𝑥𝑝𝑙𝑎𝑖𝑛𝑒𝑑 = +,-./ 1.2!.-!,( (45)7+,-./ 1.2!.-!,( (4%) +,-./ 1.2!.-!,( (45) ------Eq (3) Qualitative While the quantitative approach is effective in explaining what explains heterogeneity in risk perception at households’ level, it is limited in capturing the deeper, context-specific reasons behind these variations by community type. To address this limitation, we applied qualitative methods, which allow for a nuanced exploration of the underlying bio-physical, household and institutional factors and causal mechanisms at play. We adopted a thematic approach for coding qualitative data, following recommended thematic analysis procedures (Miles et al., 2014) to extract and label meaningful themes (i.e., code) and concepts from responses to semi-structured questions. The Cambodian research team transcribed (Khmer) and translated (English) all recorded qualitative responses from 3,700 audio files beginning with a subset of 20% of files. Both the Cambodian and Michigan State University (MSU) teams then identified themes and concepts resulting in codes, which were then applied to the subset of audio files to validate code applicability. Over three rounds of collective discussion, iterative coding adjustments, and determinations of high inter-coder reliability, a final set of codes, specified to each open-ended question, were developed, and applied to all responses. We developed summary statements from qualitative data collected using informal interviews by extracting key sentences or paragraphs corresponding to each code. These statements served as anecdotes, suggesting underlying reasons or mechanisms for behaviors noted by participants. Finally, we used narratives from interviews to construct a causal loop diagram (CLD) that integrates various biophysical, institutional, and household-level factors which might explain persistent behavior related to drought severity over time across different community types. CLD is a qualitative modeling approach that is particularly effective for identification of the root causes of recurring problems, known as “system archetypes”(Mirchi et al., 2012). Studies increasingly use CLD in examining system archetypes such as drought or water problems. For instance, CLD was used as a base model to explore the causal interactions, identify archetypes, and explain recurring behaviors in contexts like drought in rainfed agriculture in Iran (Shahbazbegian & Bagheri, 2010), coffee production systems in Vietnam (Y. Pham et al., 2020), and water scarcity in Iran (Barati et al., 2023). 54 In our study, the CLD establishes the preliminary causal hypotheses regarding the perceived trends in drought severity over time, which may help explain the heterogeneity in risk perception across different community types. It is important to note that the CLD represents the subjective nature of community-level responses and perceptions derived from interview narratives, rather than a deductive analysis of all possible factors. As such, while institutional factors are included, broader roles such as government involvement may not be fully represented. We developed two CLDs to represent the irrigated (CLD-I) and flood-pulse communities (CLD-F). Using Stella software, we linked the identified variables, illustrating how one variable affects another. The links are labeled with either a “+” or a “-” based on the direction of the relationship, positive or negative. When a causal link demonstrated a reciprocal relationship, we created a feedback loop, which could be either balancing or reinforcing (Barbrook-Johnson & Penn, 2022). Finally, we assembled all feedback loops into a causal loop diagram to create a visual model of perceived drought severity behavior. 2.3 Results and discussion 2.3.1 Drought risk perception From our random intercept model, we found significant variability in farmers' perception of drought risk both within and across communities. The standard deviation of the random intercept, which measures variability between communities, was estimated at 0.167 (95% CI: 0.073, 0.242; variance = 0.028). This indicates notable differences in risk perception between communities. Additionally, the standard deviation of the residuals, capturing variability within communities, was estimated at 0.642 (95% CI: 0.600, 0.667; variance = 0.412). This highlights substantial differences in risk perception among farmers within the same community. Looking at the importance of individual characteristics on drought risk perception, our Confirmatory Factor Analysis (CFA) results revealed varying degrees of factor loadings across three dimensions. The severity of consequences to households' finances and livelihoods emerged as the most significant factor, with a high loading of 0.80. In contrast, affective responses to drought showed a moderate loading of 0.53, and the perceived probability of drought occurrence exhibited the lowest loading at 0.49. These findings suggest that households place greater emphasis on the potential impacts of drought when shaping their risk perceptions, followed by emotional responses and, lastly, their assessment of the likelihood of drought occurrence. 55 Previous studies have similarly highlighted the strong effect of consequences and emotional factors in forming risk perceptions, while consistently noting the relatively lower influence of perceived probability (Tang et al., 2013b; Wilson et al., 2019). Overall, we found that a majority of households—approximately 72%— fall into the perceived risk categories of high or very high. Figure 10 provides the response to each characteristic of drought. A Likelihood of occurrence of drought B Financial consequences C Livelihood consequences D Affect Figure 10: Households’ response to different component of drought risk 56 Around 67% of households reported that it is somewhat to very likely that they and their farm would experience drought or water shortages in the near future (Figure 10A), and they are extremely worried about effects (Figure 10 D). When considering the severity of future consequences, more than half of the households anticipated substantial crop losses with over 50% expecting losses ranging from half to their entire crop yield (Figure 10B). In terms of effect to livelihood, a majority reported that they might face financial loss (around 96%), food insecurity (around 86%), health related problems (62%) and debt (around 56%) (Figure 10C). Dang et al. (2014) found similar observation among farmers in Vietnam, reporting perceived climate risk to their production, income and physical health. When comparing the average risk perception across different communities, we found that communities located near irrigated dams perceived higher drought risk compared to those in the flood-pulse zone (Figure 11), an observation supported by the findings from our mixed-effect model, discussed in the following section. Figure 11: Distribution of perceived drought risk by household and community type [Note: Pie charge shows the distribution of household risk categories. The size of the pie chart indicates the mean value of perceived risk score with larger pies denoting higher perceived risk. The orange and blue boxes identify irrigated and flooded communities respectively.] 57 2.3.2 Household level shaping factors Table 8 shows the results of three nested models, including a null model and a full model without (M1) and with interaction term (M2). Table 8: Summary estimation results of the linear mixed-effects model (N=703) Predictors Presence of external water source [Yes=1] Previous experience to drought [Yes=1] Null model Model without (M2) -0.04 (0.07) 0.1 (0.07) 0.10*** (0.03) 0.26*** (0.02) -0.06 (0.04) 0 (0.08) -0.52*** (0.07) 0.18* (0.08) 0.09 ((0.08)) 0 (0) -0.02 (0.06) 0.18 (0.11) 0.39** (0.13) Model with interaction (M3) -0.01 (0.07) 0.09 -0.07) 0.09*** (0.03) 0.27*** (0.02) 0.09 (0.06) 0 (0.08) -0.46*** (0.07) 0.17* (0.08) 0.1 (0.07) 0 (0) -0.02 (0.06) 0.2 (0.11) 0.41** (0.13) -0.08*** (0.02) -0.34 (0.31) 1467.25 -716.62 11.562 0 (0.08) 1630.43 -812.21 -0.36 (0.32) 1470.76 -719.38 42.83 Subjective knowledge Perceived ability Wealth index Trust in government [Yes=1] Membership to any organization [Yes=1] Social ties [Yes=1] Sex [Female=1] Age (Yrs.) Education [Yes=1] Ethnicity [Khmer=1] Community type [Irrigated=1] Interaction term (Perceived ability * Wealth index) (Intercept) AIC Log Likelihood LRT χ² (p-value <0.001) Notes: The table reports the unstandardized beta coefficient of the mixed-effects linear regression model with community random effects and fixed effect of predictors at level 1. The standard errors are in parentheses. ***, **, * , ' showing significant at <1%, 1%,and 5%. The likelihood ratio test (LRT) indicated that each subsequent model is statistically significant in comparison to the previous model. Further, there is no difference in proportion of 58 variation explained at level 2 when we compared M1 and M2 to the null model. However, upon comparing AIC, we identified the full model with an interaction term (M2) as the best choice to explain the effect of household level factors on risk perceptions. Starting with psychological factors, we only find significant effects of subjective knowledge and perceived ability on households’ perception of drought risk. Subjective knowledge The coefficient (β = 0.09, p < 0.001) indicates that for every one-unit increase in subjective knowledge about the causes and consequences of drought, perceived risk increases by 0.09, holding all other variables constant. This finding highlights the importance of subjective knowledge in raising awareness about drought risks, which could potentially lead to more proactive adaptation measures. This result is consistent with prior studies highlighting the impact of knowledge, particularly perceived causes of hazards on risk perception. For example, Shi et al. (2016) found that higher levels of knowledge, particularly about the causes of climate change, significantly heightened concern about climate change across several countries including China, the UK, Japan, the USA, Canada, and Switzerland. Similarly, Botzen et al., (2009) reported that a lack of knowledge about the causes of flood events lowered flood risk perception among people in the Netherlands, affecting their mitigation actions, for example purchase of insurance. Our findings contribute to this body of work by reinforcing the argument that enhancing domain-specific knowledge—like understanding the causes and consequences of drought—can influence risk perceptions. While our study did not directly compare these perceptions to expert assessments, the observed relationship supports the idea that better-informed individuals may have risk perceptions that align more closely with expert views, thereby fostering more effective risk communication and adaptive behaviors (Siegrist & Árvai, 2020). Perceived ability Our results suggest that household perceived risk is likely to increase by 0.27 score with every unit increase in households’ perceived ability to handle the impacts of drought in the future (β = 0.27, p < 0.001). This finding contrasts with our initial hypothesis and some previous studies such as Van Duinen et al. (2015) who reported a negative association between perceived control over drought effects and perceived risk in Netherlands. Similarly, Sjöberg (2000) suggested a linear relationship between control over various hazards and risk aversion among a Swedish population. Sjöberg (2000) explains this inverse effect as possibly due to an "overconfidence effect," where individuals with a high sense of control may 59 underestimate risks associated with other factors, leading to what is known as "risk denial." They might believe that their actions or decisions can prevent negative outcomes, leading to a lower perception of risk. However, it is important to note that the relationship between perceived ability and perceived risk can vary depending on hazard type. For example, Ho et al. (2008) found an inverse relationship between a sense of controllability and perceived impact for landslides, where a single effective measure, evacuation, exists. In contrast, for floods, where people take various precautions but still face recurring financial losses, the situation is perceived as less controllable. In our study context, this suggests that a higher perceived ability to manage drought could still be associated with increased perceived risk, especially as drought events become more frequent and damaging. Wealth condition and perceived ability We did not find a direct effect of wealth on perceived risk. However, we did observe a significant moderating effect of wealth on the relationship between perceived ability and perceived risk (Figure 12). Figure 12: The marginal effect of wealth on the relationship between perceived ability and risk perception The interaction plot shows a positive linear relationship between perceived control and perceived risk across all wealth levels. However, this relationship varies with wealth: at lower 60 wealth levels (mean -1 SD), the slope is steeper, indicating a stronger relationship between perceived control and perceived risk. Conversely, at higher wealth levels (mean +1 SD), the slope is gentler, suggesting a weaker relationship. Therefore, as households become wealthier, the positive relationship between perceived control and risk perception may weaken. A possible explanation, as suggested by Slovic (1987, 2000), is that financial resources enable individuals to manage and control risks more effectively, thereby providing safeguards against losses. This, in turn, could reduce their overall perceived risk. Looking at social factors, we find significant effects of social ties for information and association to any organization on households’ perceived risk but with opposite relationships. Social ties for information The estimate suggests that information obtained from both formal and informal social networks positively influences household perception of drought risk, with a 0.17 times higher likelihood compared to those without such ties (β = 0.17, p < 0.05). This finding is consistent with previous studies that demonstrate how connections to social networks increase awareness and heighten risk perception (Dang et al., 2014; Tang et al., 2013a; Wachinger et al., 2013). Specifically, informal ties, such as those with friends, relatives, and neighbors, have been shown to amplify perceived risk by raising awareness and triggering memories of past events, such as droughts, further emphasizing the primary role of social networks in risk communication. Association to any organization In contrast to the effect of social ties, we find a negative association between one's association with different organizations and drought risk perception (β = -0.46, p < 0.001). This may be because, unlike households’ social ties for information, households' associations with various organizations––such as farm or fish cooperatives and credit groups–– might primarily involve accessing material, financial and technological supports. These supports likely enhance their adaptive capacity (Adger, 2003; Bastakoti et al., 2014), thereby increasing their perceived control over impacts (Burnham & Ma, 2017), and as a result, lowering their perceived risk. Geographical factor We found a significant effect of community location on drought risk perception. The estimate suggests that households located in communities near irrigation dams perceive 0.41 times higher risk compared to those in communities in the flood-pulse region. While this finding is novel in risk perception studies, earlier scholars have highlighted the importance of contextual factors in shaping flood risk perception. For instance, Botzen et al., 61 (2009) identified geographical characteristics, such as proximity to a main river and rural location, as key determinants in shaping households' perceptions of flood risk and expected damage. Similarly, in the farming context, studies have found that farmers perceive risk through their exposure to resource constraints like water scarcity (Tang et al., 2013a) or uncertainty with existing water sources, such as surface water (Urquijo & De Stefano, 2016). Considering these findings, it is plausible that in our study context, households located closer to irrigation canals may perceive higher drought risk due to their reliance on irrigation infrastructure As drought events become more frequent, these households though located nearby irrigation canal, may face exacerbated impacts, such as crop loss and increased expenses, as they experience variable and inconsistent distributions of irrigation water. This causal relation is further explored using qualitative information and CLD, as detailed in section 3.3. 2.3.3 Heterogeneity in risk perception by community type Figure 13 and Figure 14 show the potential interconnected biophysical, institutional, and household-level factors that may explain why households in irrigated communities perceive a higher risk compared to those in flood-pulse communities. There are two reinforcing and one balancing loops in CLD-I that explain the recurring pattern of drought severity in the irrigated communities (Figure 13). 62 + + Perceived drought Irregular rainfall pattern - Crop/soil productivity - R1 + Irrigation system Perceived ability - + R2 + + + Behavioral changes + Financial losses + Income + Social Network Agricultural inputs + - labor/time Collect crabs/fish Livestock raising + Livelihood diversification B1 Wage activities Off-farm activities Figure 13: CLD-I showing the interrelated factors that reinforce the perceived drought severity/risk in the irrigated communities (source: Field survey, 2022) [In the diagram, red text denotes biophysical factor, blue denotes institutional and green denotes the household factors. Grey text denotes the likely factors connecting dots between factors from interview narration. Arrow with double slash (//) shows the lag effect in the system.] The first reinforcing loop (R1) operates as follows: As farmers perceive an increasing trend in drought severity (‘perceived drought’), they make various changes in cultivation practices, referred to as "behavioral changes." Despite these changes, they experience a decline in ‘crop productivity’, which is likely to heighten their ‘perceived drought’ severity risk. This loop is supported by interview narratives. Locals reported an increasing trend of irregular rainfall patterns and short-term droughts during the wet season over last three years, facing significant crop losses. Around 96 percent of farmers reported experiencing drought in the last year. As a result, the majority have shifted from long-term transplantation to the short-term broadcast method for rice farming during the wet season. This shift has led to subsequent changes in other cultivation practices such as changes in crop varieties and increased application of pesticides, 63 herbicides and fertilizers. However, despite their efforts, they continue to face declining in crop productivity. One interviewed farmer echoed this experience: "When I changed from long term to short term rice, I got less yield. Because of drought, we lost crops, but I think it is better to practices short term as I can assure the yield at least, though the yield is less." (Women, 35, Khum Ponley) This narrative suggests that farmers may continue using the new method, increasing their investment, but in the long run, they might face a decline in soil productivity, losing more crops, which could further increase their perception of drought severity. The second reinforcing loop (R2) connects ‘perceived drought’, the ‘irrigation system’, and ‘behavioral changes.’ Although these communities are located near irrigation canals, only 34 percent of households have access to irrigation water (HH survey, 2022), indicating an unequal distribution of irrigation resources. In this context, farmers—both those who have access to irrigation water and those expecting it in the future—make changes in their cultivation practices. Similar to R1, these new methods and subsequent incremental changes increase their investment. However, as they face declining crop production and recurring losses over time, they may perceive the situation as less controllable (‘perceived ability’ decreases), further amplifying their perception of drought severity. One farmer’s experience reflects this challenge: “Rice farming doesn't provide a good profit because fertilizer is quite expensive, but the price of rice is really cheap. I am in debt now because I make a loss in farming.” The balancing loop (B1) introduces a potential connection based on farmers' narratives about reallocating labor and time to other livelihood activities. Many farmers reported that by using the broadcast method instead of transplantation, they saved time and labor. This shift allows them to engage in additional activities, as illustrated by one farmer's statement: “I changed farming practices from transplantation to broad cast. There is increased application of equipment like tractors to farm, compared to past when I used to use cattle. Because of that I could reduce the labor. I have more free time to do other works, for example, wage labor in construction.” (Woman, 36 years old, Tang Thneum) Referring to this statement, we can expect these farmers might diversify their livelihood activities —such as fish and crab collecting, wage labor, and small business ventures—by 64 reallocating labor time. This diversification could increase their financial resources. which, although they may still experience crop losses, might help them better manage the financial impact of drought. Over time, this could enhance their perceived ability to control the situation, ultimately reducing their perception of drought severity. Unlike CLD-I, we did not identify any feedback loops in CLD-F that explain the behavioral patterns of drought in the flood-pulse communities (Figure 14). However, CLD-F is still useful in understanding the possible disconnection between factors that might explain why perceived drought risk is lower among households that primarily rely on fishing. We find that fishers are aware of climatic and environmental changes, such as irregular rainfall, changes in flooding/water levels, and increases in temperature. They can clearly relate these changes to institutional factors (e.g., government policies restricting fishing) and household factors (e.g., illegal fishing) affecting fish population and diversity. Additionally, some local fishers, consistent with conservation officers, link drought with changes in water levels that affect fish populations. However, when discussing the factors affecting fish availability for catch, the majority pointed to illegal fishing (around 75% based on qualitative response from 212 HH on question “Why do you think the availability of these fish has decreased in the past few years?”), which is likely to be driven by government policies restricting fishing in Tonle Sap Lake. These policies, such as seasonal fishing bans, area closures, or restrictions on fishing gear, were designed to allow fish stocks to recover. However, unintended consequences have emerged as many fisher, particularly those who rely heavily on fishing for their livelihood, have struggled to comply with these restrictions due to a lack of alternative income sources. As a result, these fishers often turn to illegal fishing as a means of maintaining their livelihoods (Gerald Flynn, 2022). This illegal fishing is characterized by practices such as fishing during prohibited seasons, using banned gear, or entering protected areas. Weak enforcement of these restrictions allows these practices to persist (Trenchard, 2023), further depleting fish stocks over time(R1). The narratives do not support any link connecting change in fish harvest for their livelihood and perceived drought severity. 65 Irregular rainfall Temperature (+) - Flooding/ water level Storm + Perceived drought + ++ changes in fish breeding - Dissolved oxygen (-) - Government policy + Illegal fishing + Fish population/ diversity + - - - R1 - Fish harvest - Figure 14: Casual loop diagram showing the interrelated factors impacting fishing and potential link explaining perceived drought risk in Flood-pulse fishing communities We hypothesize that this disconnect may stem from the delayed effects within fisheries, where the impacts of drought on fish populations are not immediately apparent (Brown et al., 2012). This delay might prevent fishers from fully recognizing the long-term consequences of drought on their livelihoods, which could explain their lower perceived risk (Slovic, 2016). Additionally, other pressing stressors, such as illegal fishing, may overshadow the effects of drought. As a result, fishers might overlook or underestimate the connection between declining fish harvests and drought severity, resulting in a relatively lower risk perception. Further research is needed to explore these dynamics in greater depth and provide clearer insights. Overall, these insights are valuable for future research and policy initiatives. Specifically, our findings contribute to the development of context-tailored drought risk communication and management strategies, aligning with objectives outlined in "2021–2030 Basin Development Strategy and Mekong River Commission Strategic Plan 2021–2025 " for drought management in 66 the Mekong region. Such strategies can support vulnerable farming and fishing communities in managing drought risk while addressing potential socioeconomic disparities in the region. 2.4 Way forward In this study, we examined how households—nested in communities among different contexts—understand drought risk, revealing significant variation in risk perception both within and across communities. This variation is shaped by socio-economic and psychological factors, including subjective knowledge (e.g., understanding the causes and consequences of drought), perceived ability to cope with drought impacts, wealth status, social ties for information exchange, association with different organizations, and community characteristics (such as proximity to irrigation canal and flood-pulse zone). Our findings highlight the importance of increasing awareness of drought causes and consequences to improve risk perception, which could support more effective adaptation actions. Additionally, we observe that social ties play a positive role in sharing information, thereby enhancing knowledge and increasing risk perception. Conversely, the negative effect of one’s association with different organizations on risk perception suggests that these associations might support building capacity of household to manage risk, thereby lowering their risk perception. The moderating effect of wealth on perceived ability and risk perception suggests that those with limited financial resources tend to have higher perceptions of risk, as they feel less control over their situation. Conversely, wealthier individuals may perceive greater control, potentially leading them to underestimate risk and overlook necessary adaptation measures. This indicates a need for targeted risk communication and management programs that effectively address drought across all socioeconomic groups. Interestingly, we find heterogeneity in risk perception across communities, with higher perceptions of risk among irrigated communities compared to those located around the Tonle Sap Lake. This difference can be explained by possible causal interactions between various biophysical, institutional, and household-level factors. In the irrigated system, despite interventions such as irrigation and agricultural practices like broadcast rice cultivation, these measures, while seemingly supporting farmers' immediate needs, may unintentionally exacerbate drought experiences over time. Specifically, the limited and disproportionate distribution of irrigation resources, combined with behavioral responses to increasing drought events, contributes to a reinforcing feedback loop. As farmers increase investment in new practices 67 which fail to meet their needs (i.e., crop yield) and neglect underlying issues, this leads to declining crop productivity and accumulating financial losses. This, in turn, diminishes households' control over the drought situation and intensifies their perception of drought severity. Although livelihood diversification could lower severity, the reinforcing patterns of perceived drought severity are likely to persist. This underscores the need for future research to explore the long-term effects of these dynamics and for policymakers to develop strategies that not only address immediate needs but also prevent worsening drought conditions over time. In flood-pulse communities, where fishing is the primary livelihood, the causal loop suggests that delayed recognition of drought's impact on fish harvest, coupled with the pressing stressors like illegal fishing, contributes to lower risk perceptions. This indicates a need for improved awareness and a better connection between environmental changes and their impacts on livelihoods to enhance risk perception and adaptive capacity. Future research should focus on validating the proposed causal loop diagrams and applying statistical models specific to both fisheries and irrigation systems to better understand these dynamics. We recommend using system dynamics modeling, where information is fed into the hypothesized causal loop diagram. This approach allows for the quantification of behavioral patterns over time and helps identify leverage points for policymakers to mitigate risks. Additionally, statistical models can be employed for sensitivity analysis of the system dynamics model, providing further insights into the robustness of the findings and enhancing the effectiveness of proposed interventions. Extending this work will help clarify the long-term effects of the identified feedback loops and offer a more comprehensive understanding of how these communities perceive and respond to environmental changes, ultimately informing more effective risk communication and adaptation strategies. 68 CHAPTER III: PERCEIVED PEER EFFECT ON FARMERS’ ADAPTIVE BEHAVIORS 1.1 Introduction Adverse climate-related events raise concerns for agricultural and other natural resource- dependent livelihoods in the developing world (Easterling et al., 2000; IPCC, 2012). This is particularly true for the Lower Mekong Region (LMR), where climate projections suggest an increase in mean temperature of 0.790C and seasonal rainfall variability, resulting in extreme water flows during wet and dry seasons, intensifying flood, and drought events (Commission, 2010; Eastham et al., 2008). Furthermore, the unprecedented boom in dam construction has triggered rapid hydrological and ecological changes (Pokharel et al., 2018) including novel water, nutrient, and sediment dynamics. These changes affect water and soil quality (Bastakoti et al., 2014; Trung et al., 2020), fish diversity (F. He et al., 2018; Nhung, 2017b) and the livelihoods of agriculture and fisheries dependent populations (Robert, 2017). In particular, small-scale and subsistence farmers in rainfed systems are at substantial risk of water scarcity, reductions in crop yields, increased food prices, and food insecurity risks (IPCC, 2012; MRC, 2019b). Thus, it is imperative to anticipate and respond effectively to minimize farmer vulnerability. Farmers’ decisions (i.e., whether and how) to respond to a changing environment and climate are often referred to as adaptive behaviors (Smit et al., 1999, 2000). In the broad climate change literature, adaptive behaviors include a variety of actions taken by households or communities to reduce the effects of environmental and climate-related changes and can be differentiated by purposefulness (autonomous vs. planned) and timing (proactive vs reactive) (Smit & Pilifosova, 2003). These actions for farmers range from decisions about agricultural management practices (e.g., crop selection, planting time), livelihood strategies (e.g., livelihood diversification) to community-level resilience-building (e.g., seed exchange programs). At the heart of these adaptations is the need for adaptive capacity, which is inextricably linked to the social system in which farmers live and work (Adamson & Bird, 2010) including households and local institutions within their communities. Social systems play a crucial role in supporting farmers' access to and utilization of diverse assets aimed at managing climate risk (Lemos et al., 2016). 69 Social capital can facilitate building adaptive capacity of natural resource dependent households by providing access to resources (Bebbington, 1999). However, it’s the role of social networks—both formal and informal—as embodiments and sources of social capital, that is gaining attention in studies on agricultural innovation (van Rijn et al., 2012), natural resource management (Bodin et al., 2006; Groce et al., 2018; Kramer et al., 2016), resilience, and climate change adaptation (Barnes et al., 2020b; Cassidy & Barnes, 2012; Islam & Walkerden, 2014). Formal networks such as local government agencies, NGOs, private businesses, and cooperatives play a crucial role in communicating climate risk and awareness, distributing resources, and mediating innovation (Bastakoti et al., 2014). Membership and well-established communication channels often characterize these networks. Whereas informal networks, including peer information exchange and labor networks, rely on personal relationships. While both networks may influence farmer adaptive behaviors, informal networks are particularly important where formal support is absent or limited (Adger, 2003; Cassidy & Barnes, 2012; Tran & Rodela, 2019). In such settings, farmers often rely on informal networks for the exchange of knowledge, labor, or money. Previous research has analyzed the effects of informal networks on farmer adaptation using measures such as the number of social ties (Wossen et al., 2013), distance to network members, and proportion of network adaptors (Matuschke & Qaim, 2009). These studies, however, do not capture the dynamic process through which farmers interact with and are exposed to the behaviors of their peers over time, leading to changes in their beliefs, knowledge, and behaviors (Barnes et al., 2020b; Frank, 2011). Furthermore, while these studies acknowledge the significance of both formal and informal networks; they analyze their influences independently, limiting our understanding of possible interaction effects. In response to these limitations, we seek to examine the role of farmers' peer and formal networks in influencing adaptive behaviors in response to climate risks in the LMR. Our central question is whether and to what extent perceptions of peer effect influence farmers' adaptation strategies in response to climate risk. Additionally, we explore how formal networks affect the interplay between perceived peer effect and farmers’ adaptive behaviors. Our study contributes to the existing literature on climate change adaptation in several ways. First, we expand the limited knowledge on how perceived peer effects influence farmers’ adaptive behaviors in regions with limited institutional support by quantifying this process 70 through social influence models. Second, we examine how the impact of perceived peer effect varies depending on the presence of ties with formal networks, introducing interaction effects. Finally, we integrate qualitative methods to validate and extend our quantitative findings, enhancing the robustness of our analysis. In the following sections, we outline our conceptual framework, describe the materials and methods used, present the results, and discuss their implications, including limitations, and suggest directions for future research. 1.2 Conceptual framework Our conceptual framework integrates established behavioral theories including bounded rationale and planned behavior (Schlüter et al., 2017), and social influence network theory (Friedkin & Johnsen, 2011) and incorporates empirical findings from climate change adaptation and social network studies (Figure 15). Figure 15: Conceptual framework: Influence of social setworks and other socio-physiological, economic and demographic factors on farmers' adaptation choice We assume that farmers' decision-making to manage climate risks starts with building knowledge—either through personal experiences or social learning. This knowledge encompasses perceived causes and consequences of climate risks and potential adaptation options. Farmers acquire this knowledge by being exposed to external environments and climate 71 extremes and by attempting various strategies to mitigate risks over time (Ha Vo et al., 2021; Jiao et al., 2020). When farmers’ uncertainty is bounded by incomplete information and limited cognitive capacity, they tend to rely on experience and learning from their social networks (Taberna et al., 2023). Social influence network theory describes how individuals update their beliefs and behaviors through interaction and exposure to their peers, including friends and relatives. Additionally, association to formal ties—including NGOs, government agencies, community-based or private organizations—can enhance learning (Arora, 2012) and facilitate adaptation by offering technical and other material support as seen in Thailand (Bastakoti et al., 2014), Vietnam (Bui & Do, 2021) and Ghana (Abdallah et al., 2014). In making decisions about farm practices, other factors also come into play. Planned behavior theory suggests that behaviors are influenced by behavioral attitudes––shaped by expectation, for example increase in crop yield trust in government bodies management strategies (Cologna & Siegrist, 2020; L. Li et al., 2023) and their perceived ability to manage the risk using appropriate measures, also called perceived adaptive capacity (Grothmann & Patt, 2005). Even if farmers develop adaptation options and intentions, their actual behaviors may be influenced by perceived behavioral controls, that is, their aggregated belief about the control factor (for example, assets) and perceived power of those factors to manage risk. Economic factors such as wealth, access to credit (Barnes et al., 2020a; Cassidy & Barnes, 2012; Yang-jie et al., 2014) and land tenure security (Abdallah et al., 2014; Jiao et al., 2020; Yaméogo et al., 2018) are the crucial control factors . Further, household characteristics like age (Ha et al., 2023; Ma et al., 2022), education and gender of household head, and household size (Abdallah et al., 2014; Belay & Fekadu, 2021; Ha Vo et al., 2021) can also influence the adaptation decision. 1.3 Materials and methods 1.3.1 Study sites and population We studied farming households located in the communities near the Tonle Sap Lake in Cambodia (Figure 2). The Tonle Sap, known as the "Great Lake" of Cambodia, is a large freshwater body that, until recently, experienced area fluctuations from 2,500 km² during the dry season to 15,000 km² during the wet season (Matsui et al., 2005; MRC/WUP-FIN, 2007). The lake shares a unique, interconnected hydrological and agricultural ecosystem with the Mekong River via the Tonle Sap River (Keskinen, 2006a) and supports the livelihoods of nearly half of 72 Cambodia’s population (Bonheur, 2001). Although it is well known for its exceptional fish production, the lake’s floodplains also significantly contribute to Cambodia's rice production. Figure 16: Study communities around the Tonle Sap Lake The communities around Tonle Sap Lake are part of a socio-economic landscape characterized by diversity in ethnicity (i.e., Khmer, Cham, and Vietnamese), livelihoods, resource access, poverty levels, and vulnerability. The predominant livelihoods are fishing for those living near the lake and rice cultivation for those living farther away. Each is highly susceptible to fluctuations in the lake's flood pulse dynamics (MRC/WUP-FIN, 2007). Other local challenges include rapid population growth, government-led development initiatives like dam construction, and climate change (Nuorteva et al., 2010; Yoshida et al., 2020), heightening the sensitivity of the agricultural sector by extending drought periods, changing flood dynamics, and promoting unjust water governance, which has reduced rice yields (World Bank, 2023), exacerbating indebtedness and food insecurity. Our sample comprises households from eight villages, purposively selected based on village history interviews with community leaders (January 2023). The selection criteria included 1) proximity to the Achang irrigation dam completed in 2013 and Tonle Sap Lake, 2) proportion 73 of households engaged in farming and fishing, and 3) local livelihood concerns related to current socio-environmental challenges (e.g., land access, farming issues, decline in fisheries, relocation, migration) and a diverse set of responses to those issues in communities. Four communities, Chhnuk Tru, Ou Ta Prok, Peam Bang, and Srey Chak, are located on the floodplain of Tonle Sap Lake. Residents of these communities typically engage in fishing during the wet season (June- October), farming during the dry season (November-May) and seasonal wage labor activities. The other four villages, Tang Thnuem, Tang Trapeang, Tapang, and Pou Mreach, are located at different distances downstream of the Achang irrigation dam and within 15 km of Tonle Sap Lake and primarily comprise rice farmers. We considered households mainly engaged in rice production as the primary unit of analysis and therefore included 213 of 400 households sampled from eight communities. We then identified and excluded cases with missing survey responses, resulting in a final sample size of 198 households. 1.3.2 Data collection We collected data using household surveys, informal in-depth interviews, and participatory observation methods between May 2022 and June 2022 using the tablet-based Open Data Kit platform (ODK) (Hartung et al., 2010). Our study is approved by Institutional Review Board (IRB) of Michigan State University under the reference number STUDY00004770. We obtained oral consent from all participants prior the start of the study. The consent form, detailing the study's purpose, data usage, confidentiality measures, and the right to refuse participation, was explained to all participants prior to obtaining their approval. A trained, Cambodian research team conducted 90 minutes face-to-face surveys with household heads. They used a pretested structured and semi-structured questionnaire to collect cross-sectional and retrospective information on household demographics, income and debt, material wealth, food security, drought, social networks and relations, resilience, and livelihoods including detailed information on farming, fishing, and aquaculture. Following the survey, the Cambodian and MSU teams conducted informal interviews with local villagers in six communities near the Achang irrigation dam to understand recent farm-related challenges including rice cropping practices, infrastructure development, climate change, coping and adaptation strategies, and local support. 74 Network Data Collection Definitions Using theoretical frameworks and empirical results from recent scholarship on social network analysis and climate change adaptation (Barnes et al., 2020b; Matuschke & Qaim, 2009), we define a social network as a source of social capital that facilitates adaptation to environmental and climate change through the exchange of resources, knowledge, practices, and norms within interconnected ecological and social systems. We focus on social ties among farmers in informal peer or ego-centric networks (i.e., friends, family members, or neighbors) centered on an individual farmer (i.e., the ‘ego’) for information exchange. Furthermore, we considered three formal networks including non-governmental organizations (NGOs), government agencies, and private vendors, representing more structured relationships that can also facilitate adaptation in a rural, agricultural context (Adger, 2003). Methods Given the challenges associated with conventional network data collection methods, we adopted an alternative approach better suited to our study context. Previous studies have used methods such as the 'name-generator technique' (Isaac, 2012; Isaac et al., 2014), the 'first-name-cue method' (Matouš et al., 2013), and snowball methods (Albizua et al., 2021) to collect network data. These methods presented challenges in our study context. First, these approaches require the names of individuals from whom the surveyed farmers seek information, followed by interviewing those listed individuals. However, in our study communities, residents use nicknames instead of formal names, posing difficulties in accurately identifying peers. Second, these methods require collecting network data from all nominated individuals within a defined village boundary. However, in many agricultural systems, including rural Cambodia, farmers rely on close ego networks and connections beyond village boundaries (Bandiera & Rasul, 2006; Conley & Udry, 2001) as indicated in our village histories. Instead, we collected complete egocentric network data from each sampled household. Each household head noted up to five people from whom they seek information and advice about farming. We refer to these individuals as informal network partners. This approach has been shown to provide reliable information on individuals with whom respondents interact regularly (Marsden 2011, pp. 382-383). For each network partner, household heads were asked to provide information about their network partners including location, the nature of their relationship, and frequency of interactions. They were also asked to report whether their network partners made behavioral changes recently regarding crop selection and farm management (i.e., intercropping, 75 irrigation, fertilizers, and pesticides). Ma et al. (2022) applied a similar approach to collect information on climate change behavior on the relatives, friends, neighbors, and government of respondents in China. This approach has some limitations. Most obviously, we did not obtain first-hand information from network partners. However, social learning theory and previous studies suggest that while farmers may not have detailed knowledge about other farmers’ practices (e.g., amount of fertilizer or pesticide applied), they do have general information of their peers’ practices through interaction and/or observation (Conley & Udry, 2001), providing reliable information on adaptive practices within egocentric networks of farmers. Lastly, we also collected information on formal network sources from which households obtain information and advice on farming. Measures of peer effect We used perceived peer effect (PPE) as the measure of the informal peer network effect, which combines the perceived frequency of interactions and the adaptive behaviors of those peers within respondents’ egocentric networks. This concept is grounded in the discussion of dynamic interactions within social networks (Manski, 2000) and influence modeling for natural resource management (Frank, 2011). Both highlight that individuals’ behaviors are influenced by their network with a time lag, as they interact, observe, and gradually adopt the behaviors of their peers. Also, prior studies identify the role of social networks and network partners’ behaviors on adaptation (Matuschke & Qaim, 2009; Moritz et al., 2024; Wossen et al., 2013). By considering both the frequency of interactions over 12 months and the adaptive behaviors of peers, our measure captures the gradual process through which peer influence affects farm activities in their networks. We use "perceived peer effect," an adapted version of the term "network exposure" used by previous scholars in social influence models (Barnes et al., 2020b; Frank, 2011), because it captures the farmer respondents’ self-reported ties, levels of interaction, and adaptive behaviors of their peers. While this measure is less common in social network and climate change studies, its use is growing in social network studies in the fields like education, particularly in examining risk behaviors and substance use among adolescents and young adults (Hofer et al., 2024; Schuler et al., 2019). Studies have even found that perceived peer behaviors can have a greater influence than actual peer behaviors in shaping individual decisions (Watts et al., 2024). 76 This focus on perception is critical because, as research in agriculture shows, farmers' perceptions of what their peers are doing or expect them to do can be as important as actual behaviors. For instance, Vasquez et al. (2019) found that perceived norms around antimicrobial use influenced farmers' intentions on New York dairy farms. Similarly, Qiu, Zhong and Huang (2021) showed that perceived peer behavior affected decisions on land protection in China. Qiao et al. (2022) also observed that perceived peer behavior (as they use farmers’ self-reported behavior of their surrounding farmers) influence farmers' decisions on green production. These findings underscore that perceptions of peer behavior can be influential in decision-making processes. 1.3.3 Covariate selection Initially, we considered 18 covariates suggested by behavioral theories and a review of empirical studies explaining adaptive behaviors to climate change (see Figure 15). To account for the explanatory power and sensitivity of the Poisson model given our sample size of 198, we employed a multimodal inference approach using the MuMIn package in R (Bartoń, 2024) to further identify key covariates. From this process, we identified 12 covariates which included our variables of interest (i.e., perceived peer effect and the interaction of perceived peer effect with formal ties), that appeared significant across the top 20 models with high predictive power based on AIC and holds high weights in the relative importance analysis. For consistency and to examine the robustness of results, we applied the same covariates in our Probit model. We performed a power analysis to confirm the appropriateness of the selected number of variables for each model (Pseudo R2 = 0.1-3, power = 0.84-0.99, at p-value = 0.05). Measures of selected covariates We considered three types of formal networks, ties to government bodies, NGOs, and private vendors, coded as dummy variables with binary responses where ‘1’ denotes the presence (‘Yes’) and ‘0’ the absence (‘No’) of the respective ties. We used two measures of economic condition: amount of land owned in hectares (‘land ownership’) and a wealth index derived from principal component analysis (PCA) (Córdova, 2008; Howe et al., 2008) using proxy measures (total number of durable assets owned by the household, access to improved source of drinking water, improved source of fuel for cooking and lightening, improved toilet facilities, the amount of land owned (ha) and Total Livestock Unit (TLU)). We used change in crop productivity compared to past few years as a proxy measure of 77 farmers’ expected return from adaptive behavior, coded as dummy variable where ‘1’ denotes decrease in crop productivity and ‘0’ denotes same or increase in crop productivity. We also included the total number of extreme events faced by farming households (‘multiple shocks’) as a measure of psychological factor. Further, we controlled for the household head’s age. Table 9 provides summary statistics for model covariates and expected associations. Table 9: Summary statistics for the model covariates Variables Frequency Mean Std. Min Max Expected Perceived PPE-intensity peer effect PPE- irrigation use PPE- fertilizer use PPE - pesticide use Ties- government Yes =1 15.2% No =0 84.8% Formal network effect Dev. 4.37 5 4.06 1.87 5.30 2.30 5.10 2.29 0 1 1 1 0 20 8 8 8 1 sign + + + + ± Ties- NGO 0 1 ± Yes =1 23.2% No =0 76.8% Ties- private 0 1 ± Yes =1 14.1% No =0 85.9% Psychological previous experience to 2.25 0.89 multiple shocks crop productivity Decreased =1 Increased/ same=0 49 % 51 % 0 0 3 1 Economic land ownership (ha) wealth index 1.3 -0.03 1.1 1.2 0.1 10 -2.37 3.67 Demographic Age (Yrs) 48.18 12.62 22 86 + – + + ± 78 1.3.4 Analytical approach We adopted a mixed-method approach, integrating both statistical methods and qualitative approaches for data collection and analysis. a. Quantitative We used network influence models to measure and understand how farmers’ behaviors are affected by perception of peer behaviors within their egocentric networks (Bodin et al., 2006). Influence models have been used to examine the network effect on fisher’s decisions to enforce rules of sea tenure (Stevens et al., 2015), hunters’ harvest decision (Kramer et al., 2016); and households’ adaptive and transformative actions related to climate change (Barnes et al., 2020b). We applied a two-step modelling approach. First, we analyzed the influence of a farmers’ formal and informal social networks on the extent of adoption of adaptive behaviors (i.e., adaptation intensity) or the total count of all adaptation behaviors adopted. We also modeled the effect of networks on farmers’ specific adaptive behaviors (i.e., adaptation choices) such as crop and farm management. To determine whether there was any clustering effect and need for hierarchical linear modeling approach, we conducted a random effects ANOVA analysis at a significance level 5%, followed by computing the interclass correlation coefficient (ICC) value (Snijders & Bosker, 2011; Steenbergen & Jones, 2002). The ANOVA results suggested no significant variability in adaptation intensity across communities. Further, an ICC value of 0.02 indicated only 2% of variance in adaptation intensity is accounted for at the village level (Hedges & Hedberg, 2007). These results suggest no clustering effect; thus, our modeling accounts for only household-level factors. Social Network and Adaptation Intensity Adaptation intensity represents the cumulative count of nine adaptive behaviors adopted by each farmer. We identified these behaviors through cross-tabulation, utilizing the significance of Pearson Chi-square (p < 0.05) and Cramer's coefficient (Φc), with a focus on significant moderate to strong associations (0.3 > Φc > 0.5) (Cohen, 1988). This approach ensures that the included behaviors are not just random or coincidental, resulting in index values that accurately reflect farmers’ adaptive behaviors. It is important to note that this measure captures only the diversity of adaptation actions, which may not directly indicate the intensity of resource 79 investment or commitment to each practices. Although this framing is consistent with prior research (Jiao et al., 2020; N. A. Khan et al., 2021), we acknowledge it as a study limitation, as the number of adaptations may not fully represent the true intensity or sustainability of the adaptation efforts. Further, we did not assess the adaptive behaviors of farmers in terms of their usefulness, sustainability, or whether they were positive or negative. Instead, our focus was on documenting the range of adaptive practices employed by farmers in response to climate challenges, without making normative judgments about which adaptations are preferable or more effective in the long term. Because adaptation intensity followed a Poisson distribution with a dispersion parameter (δ ) of 1, we used the basic Poisson model for our analysis (Hoffmann, 2016), which is well suited to situations when the mean is equal to the variance, as was the case with our data. Assuming that a farmer i made pth changes in his/her farm management practices over the past few years (t – t1), we mathematically express this as 𝑙𝑛(𝜆!(-7-!)) = 𝛽# + ∑ 8 8’% 𝛽8𝑥8! + 𝑒! (1) In the model, 𝑙𝑛(𝜆!(-7-!)) is the natural logarithm of the mean rate parameter for the Poisson distribution corresponding to the adaptation intensity for the ith farmer (i= 1,…,n) over the period (t-t1). In our study, we interpret λ as the average extent of adaptation by a farmer. 𝛽# is the intercept term, xpi (p = 1, 2, . . ., p) denotes household level covariates (Table 1), 𝛽8 are the coefficients associated with covariates, and 𝑒i ∼ N (0, σe 2) is the error term. Expanding equation 1 in terms of social network analysis using an influence model as guided by Friedkin and Johnsen (2011) and Frank, (2011), ln(𝜆!(-7-!)) = 𝛽# + 𝛽% ∑ ( !"’% 𝑤!!"(-7%)7- 𝑌9:;.1!,2!"(-7%) ( / ∑ 𝑥!!" !" + ∑ 8 8’% 𝛽8𝑥8! + 𝑒! (2) In the model, farmer i, interacting with his/her egocentric network partner i' in last 12 months, denoted as 𝑤!!"(-7%)7- . 𝑌9:;.1!,2!"(-7%) is the total number of farm related changes made by network partner i'. 𝑥!!" is the total number of partners in their ego-centric network. Then, ∑ is the egocentric network effect (PPE), which 𝑤!!"(-7%)7- 𝑌9:;.1!,2!"(-7%) ( / ∑ 𝑥!!" !" ( !"’% represents the farmers’ perceived exposure to the practices in one’s network via interaction over time, explained by 𝛽%. 80 We used Maximum Likelihood Estimation (MLE) to estimate regression coefficients in the Poisson model. MLE, well suited for count data, allows for precise estimation of intercept (β₀), and covariate coefficients (βₚ) in Equation (2)'s network analysis expansion. This optimization aligns our parameter estimates with data distribution, enhancing analysis validity and reliability. Further, we calculated incident rate ratio (IRR), derived from the exponentiation of 𝛽, to explain the multiplicative effect of covariates on adaptation intensity. To enhance interpretability, we reported the IRR value as a percentage, calculated as (IRR-1) multiplied by 100. We used robust standard errors for the estimates to address heteroscedasticity. Social Network and Adaptation Choices We employed a probit model to analyze the influence of egocentric networks on farmers’ adaption choices (i.e., increases in fertilizer, pesticide, and irrigation), specified as follows: 𝑦!"(-7-!))’ <#= 0.1), suggesting that government ties do not change relationship between perceived peer effect and adaptation intensity (Figure 20c). Figure 20: Probability of adopting multiple adaptive behaviors depending on the perceived peer exposure and one’s ties to formal networks Looking at other control factors, farmers’ previous experience with multiple shocks, as expected, increases adaptation intensity by 9.3 percent for each additional climate shock (IRR = 1.093, p- value <0.1). Alternatively, wealthier households have lower adaptation intensities, a decrease of 7.5 percent for each additional unit increase in household wealth (IRR=0.925, p-value <0.1). As expected, farmers experiencing a decrease in crop productivity are less likely to intensify adaptive measures, around 15 percent lower compared to those experiencing an increase or no change in crop productivity. Further, as the household head ages, there may be a modest decline (0.7%) in the propensity to adopt multiple adaptive measures (IRR = 0.993, p-value <0.05). Contrary to previous studies, our analysis revealed no significant effects of land ownership on adaptation intensity. 87 1.4.4 Social network and adaptation choices Figure a presents the probit model results for distinct adaptation choices (i.e., increase in irrigation, pesticide, and fertilizer use) excluding interaction term (APPENDIX D Table 15 for details). Figure 21: Probit model results: Main effects of variables (Average Marginal Effects) Unlike perceived peer effect on adaptation intensity, peer networks did not affect farmers’ decisions to increase irrigation and fertilizer use. However, peer networks positively influenced the likelihood of increasing pesticide use by 5.2 percent (AME =0.052, p-value <0.001). Ties to NGOs significantly influenced farmers’ behaviors concerning irrigation, pesticide and fertilizer use, albeit differently. The likelihood of farmers intensifying irrigation use increased by around 21 percent among households with ties to NGOs (AME= 0.211, p-value <0.01). Conversely, households’ ties to NGOs are associated with a decreased likelihood of intensifying pesticide use by 20 percent (AME= 0.205, p-value <0.05) and fertilizer use by 15 percent (AME = -0.153, p-value <0.1) compared to those without it. Ties with private vendors positively affected the decision to increase fertilizer use (AME = 0.199, p-value <0.05). Finally, ties to government agencies did not affect any farmer adaptation choices. Consistent with the findings of the adaptation intensity model, our adaptation choice model showed significant interaction effects, specifically formal ties with NGOs and private 88 vendors. Ties with NGOs appeared to moderate the perceived peer effect, reducing the likelihood of using pesticides and fertilizer use by around 10 and 7 percentage points, respectively (Figure 22b and Figure 22c). On the contrary, NGO ties positively influenced perceived peer effect on irrigation practices by 5.7 percentage points higher compared to those without such ties (Figure 22a, and APPENDIX D Table 17). Figure 22: Probability of taking a particular adaptive behavior depending on the peer network exposure and one’s tie to formal networks Private vendor ties, similar to NGO ties, moderate the perceived peer effect leading to a decrease in pesticide use among households with such connections (Figure 22d) and unlike the 89 effect of NGO ties, households with private vendor ties decrease the likelihood of adopting irrigation practices by around 14 percent (Figure 22e). In addition to network effects, land ownership was a significant factor influencing farmers' adaptation behaviors, particularly in increasing pesticide use such that for each additional hectare, households increased their pesticide use by 6 percent. Unlike the effects on adaptation intensity, our analysis did not find any significant effect of previous experience to multiple shocks, change in crop productivity, wealth index and age of household head on farmers’ adaptation choices. 1.5 Discussion Rice farming households across communities undertook a variety of incremental adaptation behaviors, either singly or in combination, including increased use of fertilizers, herbicides, and insecticides, and adjustment in planting times (Dapilah & Nielsen, 2020). These adaptations are not isolated responses but are intertwined within the broader transformation of rice farming systems. This transformation is characterized by a shift from conventional labor- intensive planting methods to mechanized broadcast techniques in the context of increasing climate events, mainly drought and water infrastructure developments like irrigation systems (APPENDIX B Table 13), as evident in other parts of Cambodia (W. N. Green, 2020). In addition to institutional arrangements and climate change, we found evidence that household-level factors, including social networks significantly shape the adaptive behaviors of farmers. Most farmers seek farming-related information and advice from their informal peer networks. Adger characterizes peer networks as a form of bonding capital, operating within a community, and thriving in contexts marked by limited or absent governmental support (Adger, 2003). While some connect to formal networks, for example, NGOs, for training, technological assistance, credit, and marketing services, this is limited to few households, depending on farm location and their contribution to rice farm production. Our findings suggest the importance of peer networks in influencing farmers' adaptive behaviors, explained by perceived peer effect. This finding suggests a dynamic behavioral pattern: As farmers increasingly interact with their personal network partners, these relationships become a conduit for sharing knowledge and experiences about practices and outcomes. Over time, these exchanges may shape their beliefs and behaviors, culminating in the adoption of farm practices. This pattern is corroborated by narratives from interviews. Local farmers frequently 90 reported observing and interacting with their close neighbors and adopting their practices on rice variety selection and planting times as seen similarly in other agricultural contexts. In experimental study in Uzbekistan, Moritz et al. (2024) found that observational learning within small, close peer groups significantly influenced farmers' adaptation choices regarding Index Insurance and Savings. Additionally, the lagged innovation choices of neighbors continued to shape farmers' behaviors in subsequent seasons. Similarly, Ma et al. (2022) observed a significant positive effect of adaptive behaviors among relatives, friends, and neighbors on farmers’ responses to climate change in China. This pattern is also evident among pineapple farmers in Ghana (Conley & Udry, 2001) and hybrid seed adopting farmers in India (Matuschke & Qaim, 2009). Looking at the perceived peer effect on a particular behavior, we find a significant association limited to pesticide use. One explanation is farmers’ higher perceived urgency in addressing weed and pest-related challenges. As most farmers faced pest problems, pesticide use could be widely discussed and promoted within social networks. Also, farmers with limited resources are more likely to embrace practices that provide immediate feedback and noticeable impact in terms of preventing large production losses (Chèze et al., 2020). Our results showed varying perceived peer effect on adaptive behaviors in the presence or absence of formal network ties, particularly NGOs and private vendors. Households with such ties tend to reduce their adaptive behaviors as they perceive increasing exposure to their informal peers’ behaviors. Matouš et al. (2013) observed similar result in their study of Ethiopian farmers who become less receptive to information and conservation practices from extension agents as the proportion of communal ties in their network grows. NGO interaction effects differ by behavior type. Ties with NGO for farming-related information synergizes the effect of peer influence on the use of irrigation. As indicated in interview responses, NGOs actively engage with these farmers by providing valuable training, providing credit support, and assisting with marketing services. This engagement seems to encourage these farmers to invest more in irrigation practices, potentially increasing crop yields amidst escalating drought severity. NGOs also provide support in marketing and selling farm products, helping farmers repay loans and safeguard livelihoods. However, farmers with NGO ties are less likely to increase fertilizer and pesticide usage compared to those without ties, 91 perhaps attributable to NGOs' focus on promoting sustainable practices. Locals reported receiving training in the use and timing of fertilizer and pesticides. These observations suggest nuanced effects between formal and informal networks on adaptation choices and intensity. However, these networks may have distinct roles in various stages of adaptation, especially in contexts where institutional support is limited. Interview responses indicate that NGOs provide knowledge on practices such as broadcast planting, crop varieties, application of fertilizers and pesticides through a small number of local community members. As farmers who received support begin to realize benefits, other farmers with weaker external networks receive this information. This process of information diffusion can be explained in light our interaction plots. We observe that while the expected number of adaptive behaviors is initially higher among farmers with formal ties, this changes as the perceived peer effect to network partners’ behaviors increases. Specifically, beyond a certain threshold of perceived peer exposure (say 5), the likelihood of adopting adaptive behaviors increases among farmers without formal ties. This means that as farmers increasingly interact with more adopters in their community over time, they become more receptive to different ideas and practices. This suggests the pivotal role of informal networks in facilitating the diffusion of adaptive behaviors as highlighted in previous study in India (Matuschke & Qaim, 2009). It is worth noting that while our model does not directly measure these temporal effects, we assume that these trends could emerge over time based on the observed relationship between formal and peer network interaction and predicted adaptive behavior. Future research with time-series data could provide more insight into the temporal dynamics of this process. Besides network effects, our results indicate the importance of both psychological and economic factors in shaping farm households' decisions, particularly adaptation intensity. We found that farmers' inclination to make multiple adjustments is not solely a response to a single shock but rather a response to prior experiences with multiple shocks including drought, pest outbreaks, food price fluctuations, and floods, which is particularly relevant in highly vulnerable and climate-sensitive communities (Oppenheimer et al., 2014; Smit & Wandel, 2006). As expected, when farmers experience crop losses and perceive greater risk (Le Dang et al., 2014), they are more likely to adapt with greater intensity and use strategies aimed at protecting and supporting crop yields. 92 This observation is consistent with previous studies that have identified individual experiences with past damages as a key predictor of self-protective or adaptive actions in response to climatic hazards (Barnes et al., 2020b; Y. Huang et al., 2024; Jiao et al., 2020). Notably, we found a negative effect of wealth on adaptation intensity. This suggests that wealthier households may lean towards risk-averse strategies, possibly due to their existing investments in farming practices with high returns (W. Zhang, 2017). Conversely, it is also plausible that wealthier households, though have high risk tolerance, may adopt fewer adaptation behaviors, possibly relying on their financial resources to mitigate risks or engage in less climate-sensitive, off-farm activities as a means of buffering against climate impacts (Adnan et al., 2020). We did not find any significant association between the size of owned farmland and farm households' decisions to adaptation intensity. However, the extent of farmland ownership emerges as a significant factor motivating farmers to adopt practices such as increased use of pesticide. This finding is consistent with prior studies in different contexts, for example, adoption of soil and water conservation techniques among maize farmers in Ghana (Abdallah et al., 2014) and the use of soil management practices such as chemical fertilizer among rice farmers in Northern Ghana (Donkoh & Awuni, 2011). This suggests households with larger owned farm size are more likely to investment in practices like pesticide use, getting benefit from maximum production with lower input costs per unit area (Y. Huang et al., 2024). While our study provides valuable information, we acknowledge some limitations. First, we did not make any normative judgements on the usefulness or positive or negative or long- term sustainability of farmer adaptive behaviors. Second, influence-based network analysis demands longitudinal data for causal inferences, but we relied on retrospective data based on memory recall which could introduce bias. Third, our study primarily examined informal peer information networks and a limited number of formal networks. Further exploration of other informal networks, such as labor networks, could provide additional insights on adaptive decision-making processes. While our study focuses on climate risk management, our finding on perceived peer effect may have broader applications beyond the issue of climate change. The influence mechanisms driving behavior change through peer interaction and exposure to perceived peer behaviors are likely relevant in other domains, such as health and education where behavior 93 modification is desired. Future research could test this and explore how perceived peer-based interventions might be applied to promote health literacy and safety practices (Coman et al., 2020; Simoni et al., 2011). 3.6 Conclusion Our findings identify the important role that perceptions of peer effect play in supporting farm households’ learning and leveraging adaptive behaviors to manage climate risks, particularly in communities where formal support is limited or non-existent. Our findings highlight the interplay between formal and informal networks in influencing adaptation decisions, underscoring the need for future research to understand how different types of social networks interact and affect the diffusion of innovation and adaptive capacity in agricultural communities. This knowledge also provides guidance for policymakers and planners in designing adaptation programs that enhance the adaptive capacity of farm communities. Additionally, we recognize the psychological and economic factors as both barriers and entry points for interventions aimed at supporting farmers. Farmers’ experiences with multiple shocks can heighten their perceptions of livelihood risks and motivate them to intensify their adaptive responses. Factors such as wealth, land ownership, and the age of households also play a role in adaptation decisions, offering insights for those aiming to strengthen climate resilience in agriculture and empower farmers to navigate the challenges of a changing climate to ensure the sustainability of resource-dependent farming communities in Cambodia and beyond. 94 CONCLUSION This dissertation advances our understanding of the dynamics of challenges and local responses to resource and climate uncertainties in the context of dam construction. By identifying the processes and factors that hinder or motivate adaptation decisions among resource-dependent households, this work adds to the growing body of knowledge on adaptation in vulnerable regions. The mixed-method approach, which blends quantitative and qualitative methods, provides a comprehensive understanding of the multi-faceted nature of adaptation decisions process explored across three chapters. Moreover, the use of household and community-level case studies is crucial for capturing the localized impacts of climate change and resource management and diversity of responses to environmental stressors. The insights gained from the three chapters offer both theoretical advancements and practical applications for improving the adaptive capacity of vulnerable farming and fishing communities. Chapter 1 provides a more nuanced understanding of dam impacts, particularly on natural, social, and financial livelihood resources in downstream farming communities. The findings reveal the varied impacts of dam construction on downstream communities, particularly on natural, social, physical and financial livelihood resources. Natural resource access decreases near dams, while physical resource access improves within 30 km after dam construction. Though there is financial resource access among downstream communities withing 30 Km, finding suggests some improvement among communities within 20 Km post-dam construction period. Communities’ nearby irrigation dam experience increased social resource access post- dam, but overall, spatial and temporal variations highlight the complexity of dam impacts on local resources. These findings emphasize the need for better Environmental Impact Assessments (EIAs) and adaptation strategies to address the diverse effects of dams on downstream ecosystems and livelihoods. Future research should include a larger variety of dams to develop a more comprehensive understanding of type-specific impacts. Additionally, exploring qualitative insights on measures, like migration patterns, social cohesion and their relationship to dam construction, could complement these quantitative findings. Chapter 2 improves our understanding of drought risk perceptions among households within complex hydro-agricultural-fisheries systems, emphasizing the need for context-specific risk communication strategies. This chapter identifies key psychological and socio-economic 95 factors—including households' drought knowledge, perceived adaptive capacity, organizational affiliations, and wealth status—as significant influences on risk perception. The proposed causal loop diagram (CLD) suggests a potential recurring and intensifying pattern of drought severity related to irrigation infrastructure, institutional and households’ response to the environmental changes. These findings contribute to ongoing regional initiatives such as the 2021–2030 Basin Development Strategy and Mekong River Commission Strategic Plan 2021–2025, aiming to strengthen agricultural and fishing systems in the face of increasing drought severity. Future research should focus on validating the CLD and applying system dynamics modeling to quantify behavioral patterns over time. This approach will help identify leverage points for policymakers to mitigate risks, offering further insights into the dynamics of drought response. Finally, chapter 3 expands the limited knowledge on how perceived peer effects influence farmers' adaptive behaviors, particularly in regions with limited institutional support. By quantifying this process through social influence models, the chapter highlights the interplay between formal and informal networks in shaping adaptation decisions, which might affect the adaptation diffusion process in agricultural communities. Further, the chapter identifies the significant psychological and economic factors—such as farmers' experiences with multiple shocks, wealth, and land ownership—to leverage for interventions aimed at strengthening farmers capacity to manage climate risk in Cambodia and beyond. Future research should explore how different types of social networks interact and influence the diffusion of innovations and adaptive capacity in agricultural communities, offering a deeper understanding of the social mechanisms that drive adaptation decisions. Overall, this dissertation underscores the critical role of understanding the local challenges, their perception of the resource scarcity and climate uncertainties and role of different social ties in the adaptation processes, supporting local livelihoods while sustaining the agro-fisheries ecosystem in the Lower Mekong Region. It provides actionable insights for designing policies and programs that enhance adaptive capacity, particularly in resource- dependent communities, making it a vital contribution to the field of sustainable development and climate adaptation. 96 REFERENCES Abbott, K. M., Zaidel, P. A., Roy, A. H., Houle, K. M., & Nislow, K. H. (2022). Investigating impacts of small dams and dam removal on dissolved oxygen in streams. PloS One, 17(11), e0277647. https://doi.org/10.1371/journal.pone.0277647 Abdallah, A.-H., Michael, A., & Samuel, A. D. (2014). Smallholder adoption of soil and water conservation techniques in Ghana. African Journal of Agricultural Research, 9(5), 539– 546. https://doi.org/10.5897/ajar2013.7952 Abid, M., Schilling, J., Scheffran, J., & Zulfiqar, F. (2016). Climate change vulnerability, adaptation and risk perceptions at farm level in Punjab, Pakistan. Science of the Total Environment, 547, 447–460. https://doi.org/10.1016/j.scitotenv.2015.11.125 Adamson, P., & Bird, J. (2010). The Mekong: A drought-prone tropical environment? International Journal of Water Resources Development, 26(4), 579–594. https://doi.org/10.1080/07900627.2010.519632 Adger, W. N. (2003). Social capital, collective action, and adaptation to climate change. Economic Geography, 79(4), 387–404. https://doi.org/10.1111/j.1944- 8287.2003.tb00220.x Adnan, K. M. M., Ying, L., Ayoub, Z., Sarker, S. A., Menhas, R., Chen, F., & Yu, M. M. (2020). Risk management strategies to cope catastrophic risks in agriculture: The case of contract farming, diversification and precautionary savings. Agriculture (Switzerland), 10(8), 1– 16. https://doi.org/10.3390/agriculture10080351 Albizua, A., Bennett, E. M., Larocque, G., Krause, R. W., & Pascual, U. (2021). Social networks influence farming practices and agrarian sustainability. PLOS ONE, 16(1), e0244619. https://doi.org/10.1371/journal.pone.0244619 Ali, A., Rana, I. A., Ali, A., & Najam, F. A. (2022). Flood risk perception and communication: The role of hazard proximity. Journal of Environmental Management, 316, 115309. https://doi.org/10.1016/j.jenvman.2022.115309 Allison, E. H., & Ellis, F. (2001). The livelihoods approach and management of small-scale fisheries. In Marine Policy (Vol. 25, pp. 377–388). http://www.fao.org/fi/projects/sflp/index.html Ang, W. J., Park, E., Tran, D. D., Loc, H. H., & Pokhrel, Y. (2023). Dams in the Mekong: A comprehensive database, spatiotemporal distribution, and hydropower potentials. https://doi.org/10.5194/essd-2023-417 Arantes, C. C., Laufer, J., Pinto, M. D. D. S., Moran, E. F., Lopez, M. C., Dutka‐Gianelli, J., Pinto, D. M., Chaudhari, S., Pokhrel, Y., & Doria, C. R. C. (2022). Functional responses of fisheries to hydropower dams in the Amazonian Floodplain of the Madeira River. Journal of Applied Ecology, 59(3), 680–692. https://doi.org/10.1111/1365-2664.14082 97 Arora, S. (2012). Farmers’ Participation in Knowledge Circulation and the Promotion of Agroecological Methods in South India. Journal of Sustainable Agriculture, 36(2), 207– 235. https://doi.org/10.1080/10440046.2011.620231 Arthur, J. L., Murray, G., Rollins, R., Dearden, P., & Stahl, A. (2020). Differential impacts of dam construction on livelihoods in Ghana. African Geographical Review, 39(3), 267– 281. https://doi.org/10.1080/19376812.2020.1719367 Aung, T. O., Guido, V. H., & Stijn, S. (2017). Differential impacts of an irrigation project: Case study of the Swar Dam Project in Yedashe, Bago region of Myanmar. Journal of Development and Agricultural Economics, 9(7), 178–189. https://doi.org/10.5897/JDAE2017.0824 Bachmair, S., Stahl, K., Collins, K., Hannaford, J., Acreman, M., Svoboda, M., Knutson, C., Smith, K. H., Wall, N., Fuchs, B., Crossman, N. D., & Overton, I. C. (2016). Drought indicators revisited: The need for a wider consideration of environment and society. Wiley Interdisciplinary Reviews: Water, 3(4), 516–536. https://doi.org/10.1002/wat2.1154 Baird, I. G., & Barney, K. (2017). The political ecology of cross-sectoral cumulative impacts: Modern landscapes, large hydropower dams and industrial tree plantations in Laos and Cambodia. Journal of Peasant Studies, 44(4), 884–910. https://doi.org/10.1080/03066150.2017.1289921 Bandiera, O., & Rasul, I. (2006). Social networks and technology adoption in Northern Mozambique. The Economic Journal, 116(514), 869–902. Barasa Kabeyi, M. J., & Akanni. Olanrewaju, O. (2023, November 14). Hydropower in the Sustainable Energy Mix. Proceedings of the International Conference on Industrial Engineering and Operations Management. 2nd Australian International Conference on Industrial Engineering and Operations Management, Melbourne, Australia. https://doi.org/10.46254/AU02.20230247 Barati, A. A., Pour, M. D., & Sardooei, M. A. (2023). Water crisis in Iran: A system dynamics approach on water, energy, food, land and climate (WEFLC) nexus. Science of The Total Environment, 882, 163549. https://doi.org/10.1016/j.scitotenv.2023.163549 Barbrook-Johnson, P., & Penn, A. S. (2022). Causal Loop Diagrams. In P. Barbrook-Johnson & A. S. Penn, Systems Mapping (pp. 47–59). Springer International Publishing. https://doi.org/10.1007/978-3-031-01919-7_4 Bargain, O. B., Caldeira, É., & Vincent, R. C. (2023). Shine a (night)light: Decentralisation and Economic Development in Burkina Faso [Application/pdf]. 51 p. https://doi.org/10.3929/ETHZ-B-000604542 Barnes, M. L., Wang, P., Cinner, J. E., Graham, N. A. J., Guerrero, A. M., Jasny, L., Lau, J., Sutcliffe, S. R., & Zamborain-Mason, J. (2020a). Method- Social determinants of adaptive and transformative responses to climate change. Nature Climate Change, 10(9), 823–828. https://doi.org/10.1038/s41558-020-0871-4 98 Barnes, M. L., Wang, P., Cinner, J. E., Graham, N. A. J., Guerrero, A. M., Jasny, L., Lau, J., Sutcliffe, S. R., & Zamborain-Mason, J. (2020b). Social determinants of adaptive and transformative responses to climate change. Nature Climate Change, 10(9), 823–828. https://doi.org/10.1038/s41558-020-0871-4 Bartlett, M. S. (1951). The Effect of Standardization on a χ 2 Approximation in Factor Analysis. Biometrika, 38(3/4), 337. https://doi.org/10.2307/2332580 Bartoń, K. (2024). MuMIn: Multi-model inference R package version 1.5. 2013 [Computer software]. https://orcid.org/0000-0001-5562-8274 Bastakoti, R. C., Gupta, J., Babel, M. S., & van Dijk, M. P. (2014). Climate risks and adaptation strategies in the Lower Mekong River basin. Regional Environmental Change, 14(1), 207–219. https://doi.org/10.1007/s10113-013-0485-8 Beban, A., So, S., & Un, K. (2017). From Force to Legitimation: Rethinking Land Grabs in Cambodia. Development and Change, 48(3), 590–612. https://doi.org/10.1111/dech.12301 Bebbington, A. (1999). Capitals and capabilities A framework for analysing peasant viability, rural livelihoods and poverty in the Andes A background paper for: Policies that work for sustainable agriculture and regenerating rural economies Series Editor: Simon Croxton A resear. 1–54. Beck, M. W., Claassen, A. H., & Hundt, P. J. (2012a). Environmental and livelihood impacts of dams: Common lessons across development gradients that challenge sustainability. International Journal of River Basin Management, 10(1), 73–92. https://doi.org/10.1080/15715124.2012.656133 Beck, M. W., Claassen, A. H., & Hundt, P. J. (2012b). Environmental and livelihood impacts of dams: Common lessons across development gradients that challenge sustainability. International Journal of River Basin Management, 10(1), 73–92. https://doi.org/10.1080/15715124.2012.656133 Belay, D., & Fekadu, G. (2021). Influence of social capital in adopting climate change adaptation strategies: Empirical evidence from rural areas of Ambo district in Ethiopia. Climate and Development, 13(10), 857–868. https://doi.org/10.1080/17565529.2020.1862741 Berry, W. D., Demeritt, J. H. R., Esarey, J., Berry, W. D., State, F., Demeritt, J. H. R., & Texas, N. (2010). Models: Is a Product Term Essential? Testing for Interaction in Binary Logit and Probit. American Journal of Political Science, 54(1), 248–266. Beza, M., Fikre, A., & Moshe, A. (2023). Dam Breach Modeling and Downstream Flood Inundation Mapping Using HEC-RAS Model on the Proposed Gumara Dam, Ethiopia. Advances in Civil Engineering, 2023, 1–15. https://doi.org/10.1155/2023/8864328 Bhandari, P. B. (2013). Rural livelihood change? Household capital, community resources and livelihood transition. Journal of Rural Studies, 32, 126–136. https://doi.org/10.1016/j.jrurstud.2013.05.001 99 Bodin, Ö., Crona, B., & Ernstson, H. (2006). Social networks in natural resource management: What is there to learn from a structural perspective? Ecology and Society, 11(2). https://doi.org/10.5751/ES-01808-1102r02 Bonheur, N. (2001). Tonle Sap Ecosystem and Value. Technical Coordination Unit for Tonle Sap, Ministry of Environment. http://www.mekoninfo.org/ Botzen, W. J. W., Aerts, J. C. J. H., & Van Den Bergh, J. C. J. M. (2009). Dependence of flood risk perceptions on socioeconomic and objective risk factors. Water Resources Research, 45(10), 1–15. https://doi.org/10.1029/2009WR007743 Brown, C. J., Fulton, E. A., Possingham, H. P., & Richardson, A. J. (2012). How long can fisheries management delay action in response to ecosystem and climate change? Ecological Applications, 22(1), 298–310. https://doi.org/10.1890/11-0419.1 Bui, H. T. M., & Do, T. A. (2021). Choice of adaptation strategies to climate change among farm households in mountainous areas of Northeastern Vietnam. GeoJournal, 0123456789(2005). https://doi.org/10.1007/s10708-021-10544-w Burnham, M., & Ma, Z. (2017). Climate change adaptation: Factors influencing Chinese smallholder farmers’ perceived self-efficacy and adaptation intent. Regional Environmental Change, 17(1), 171–186. https://doi.org/10.1007/s10113-016-0975-6 Bussi, G., Darby, S. E., Whitehead, P. G., Jin, L., Dadson, S. J., Voepel, H. E., Vasilopoulos, G., Hackney, C. R., Hutton, C., Berchoux, T., Parsons, D. R., & Nicholas, A. (2021). Impact of dams and climate change on suspended sediment flux to the Mekong delta. Science of The Total Environment, 755, 142468. https://doi.org/10.1016/j.scitotenv.2020.142468 Calvi, M. F., Moran, E. F., Silva, R. F. B. da, & Batistella, M. (2020). The construction of the Belo Monte dam in the Brazilian Amazon and its consequences on regional rural labor. Land Use Policy, 90(September 2019), 104327. https://doi.org/10.1016/j.landusepol.2019.104327 Cardona, O. D., Van Aalst, M. K., Birkmann, J., Fordham, M., Mc Gregor, G., Rosa, P., Pulwarty, R. S., Schipper, E. L. F., Sinh, B. T., Décamps, H., Keim, M., Davis, I., Ebi, K. L., Lavell, A., Mechler, R., Murray, V., Pelling, M., Pohl, J., Smith, A. O., & Thomalla, F. (2012). Determinants of risk: Exposure and vulnerability. Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation: Special Report of the Intergovernmental Panel on Climate Change, 9781107025, 65–108. https://doi.org/10.1017/CBO9781139177245.005 Cassidy, L., & Barnes, G. D. (2012). Understanding household connectivity and resilience in marginal rural communities through social network analysis in the village of Habu, Botswana. Ecology and Society, 17(4). https://doi.org/10.5751/ES-04963-170411 Castro-Diaz, L., García, M. A., Villamayor-Tomas, S., & Lopez, M. C. (2023). Impacts of hydropower development on locals’ livelihoods in the Global South. World Development, 169, 106285. https://doi.org/10.1016/j.worlddev.2023.106285 100 Chèze, B., David, M., & Martinet, V. (2020). Understanding farmers’ reluctance to reduce pesticide use: A choice experiment. Ecological Economics, 167(January 2019), 106349. https://doi.org/10.1016/j.ecolecon.2019.06.004 Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences. Academic Press. https://doi.org/10.4324/9780203771587 Cologna, V., & Siegrist, M. (2020). The role of trust for climate change mitigation and adaptation behaviour: A meta-analysis. Journal of Environmental Psychology, 69, 101428. https://doi.org/10.1016/j.jenvp.2020.101428 Coman, M. A., Marcu, A., Chereches, R. M., Leppälä, J., & Van Den Broucke, S. (2020). Educational Interventions to Improve Safety and Health Literacy Among Agricultural Workers: A Systematic Review. International Journal of Environmental Research and Public Health, 17(3), 1114. https://doi.org/10.3390/ijerph17031114 Commission, M. R. (2010). Impacts of climate change and development on Mekong flow regimes First assessment—2009. 29. Conde, C., Kulkarni, J., Nyong, A., & Pulhin, J. (2008). Climate Change and Vulnerability. Earthscan. Conley, T., & Udry, C. (2001). Social Learning Through Networks: The Adoption of New Agricultural Technologies in Ghana. American Journal of Agricultural Economics, 83, 668–673. https://doi.org/10.1111/0002-9092.00188 Córdova, A. (2008). Methodological note: Measuring relative wealth using household asset indicators. Americas Barometer Insights, 6, 1–9. Dang, H., Li, E., Nuberg, I., & Bruwer, J. (2014). Farmers’ Perceived Risks of Climate Change and Influencing Factors: A Study in the Mekong Delta, Vietnam. Environmental Management, 54, 331–345. https://doi.org/10.1007/s00267-014-0299-6 Dapilah, F., & Nielsen, J. Ø. (2020). Climate change extremes and barriers to successful adaptation outcomes: Disentangling a paradox in the semi-arid savanna zone of northern Ghana. Ambio, 49(8), 1437–1449. https://doi.org/10.1007/s13280-019-01275-x Di Baldassarre, G., Wanders, N., AghaKouchak, A., Kuil, L., Rangecroft, S., Veldkamp, T. I. E., Garcia, M., Van Oel, P. R., Breinl, K., & Van Loon, A. F. (2018). Water shortages worsened by reservoir effects. Nature Sustainability, 1(11), 617–622. https://doi.org/10.1038/s41893-018-0159-0 Dillon, A. (2011). The Effect of Irrigation on Poverty Reduction, Asset Accumulation, and Informal Insurance: Evidence from Northern Mali. World Development, 39(12), 2165– 2175. https://doi.org/10.1016/j.worlddev.2011.04.006 Dillon, A., & Fishman, R. (2019). Dams: Effects of Hydrological Infrastructure on Development. Annual Review of Resource Economics, 11(1), 125–148. https://doi.org/10.1146/annurev- resource-100518-093913 101 Donkoh, S. A., & Awuni, J. A. (2011). Adoption of farm management practices in lowland rice production in Northern Ghana. Journal of Agricultural and Biological Science, 2(4), 84– 93. http://www.udsspace.uds.edu.gh/handle/123456789/1049 Duinen, R. V., Filatova, T., Geurts, P., & Veen, A. (2015a). Empirical analysis of farmers’ drought risk perception: Objective factors, personal circumstances, and social influence. Risk Analysis : An Official Publication of the Society for Risk Analysis, 35 4, 741–755. https://doi.org/10.1111/risa.12299 Duinen, R. V., Filatova, T., Geurts, P., & Veen, A. V. D. (2015b). Empirical Analysis of Farmers’ Drought Risk Perception: Objective Factors, Personal Circumstances, and Social Influence. Risk Analysis, 35(4), 741–755. https://doi.org/10.1111/risa.12299 Easterling, D. R., Meehl, G. A., Parmesan, C., Changnon, S. A., Karl, T. R., & Mearns, L. O. (2000). Climate Extremes: Observations, Modeling, and Impacts. 289(September), 2068– 2075. Eastham, J., Mpelasoka, F., Ticehurst, C., Dyce, P., Ali, R., & Kirby, M. (2008). Mekong River Basin Water Resources Assessment: Impacts of Climate Change. CSIRO: Water for a Healthy Country Flagship Report Series, August, 153. https://doi.org/ISSN: 1835-095X Elvidge, C. D., Zhizhin, M., Ghosh, T., Hsu, F.-C., & Taneja, J. (2021). Annual Time Series of Global VIIRS Nighttime Lights Derived from Monthly Averages: 2012 to 2019. Remote Sensing, 13(5), 922. https://doi.org/10.3390/rs13050922 Fan, P., Cho, M. S., Lin, Z., Ouyang, Z., Qi, J., Chen, J., & Moran, E. F. (2022). Recently constructed hydropower dams were associated with reduced economic production, population, and greenness in nearby areas. Proceedings of the National Academy of Sciences, 119(8), e2108038119. https://doi.org/10.1073/pnas.2108038119 Fehr, D., & Reichlin, Y. (2021). Perceived Relative Wealth and Risk Taking (CESifo Working Paper No. No. 9253). Center for Economic Studies and ifo Institute (CESifo). https://hdl.handle.net/10419/245434 Ficklin, D. L., Stewart, I. T., & Maurer, E. P. (2013). Tation. PLoS ONE, 8(8), e71297. https://doi.org/10.1371/journal.pone.0071297 Field, C. B., Barros, V. R., Dokken, D. J., Mach, K. J., & Mastrandrea, M. D. (Eds.). (2014). Emergent Risks and Key Vulnerabilities. In Climate Change 2014 Impacts, Adaptation, and Vulnerability (pp. 1039–1100). Cambridge University Press. https://doi.org/10.1017/CBO9781107415379.024 Frank, K. A. (2011). Social Network Models for Natural Resource Use. In Social networks and natural resource management: Uncovering the social fabric of environmental governance (pp. 180–205). Friedkin, N. E., & Johnsen, E. C. (2011). Social Influence Network Theory: A Sociological Examination of Small Group Dynamics. Cambridge University Press. 102 Fung, Z., Pomun, T., Charles, K. J., & Kirchherr, J. (2019). Mapping the social impacts of small dams: The case of Thailand’s Ing River basin. Ambio, 48(2), 180–191. https://doi.org/10.1007/s13280-018-1062-7 Ganeshpandian, P. (2024). Dams, hegemony and beyond: China’s hydro-stability in the evolving world order. Discover Global Society, 2(1), 9. https://doi.org/10.1007/s44282-024-00036- w Gebrehiwot, T., & van der Veen, A. (2015). Farmers Prone to Drought Risk: Why Some Farmers Undertake Farm-Level Risk-Reduction Measures While Others Not? Environmental Management, 55(3), 588–602. https://doi.org/10.1007/s00267-014-0415-7 Gerald Flynn, V. S. (2022). Fisheries crackdown pushes Cambodians to the brink on Tonle Sap lake. Mongabay. Goodman, L. (2024). Illuminating trade-offs: The socio-economic impacts of dam construction in the Global South. Cambridge Centre for Energy, Environment and Natural Resources Governance. Green, A., Droggers, P., Simons, G., Giesen, W., Balen, B. van, Hortle, K., Hunink, J., Oeurng, C., Bonvongsar, T., & Yem, D. (2019). Detailed Surface Water Resources Assessment for the Tonle Sap and Mekong Delta River Basins, Cambodia [TA-7610 CAM]. Asian Development Bank. https://www.futurewater.nl/wp- content/uploads/2020/07/SWRA_FinalReport.pdf Green, W. N. (2020). Financial landscapes of agrarian change in Cambodia. Geoforum, February, 0–1. https://doi.org/10.1016/j.geoforum.2020.02.001 Groce, J. E., Farrelly, M. A., Jorgensen, B. S., & Cook, C. N. (2018). Using social-network research to improve outcomes in natural resource management. 33(1), 53–65. https://doi.org/10.1111/cobi.13127 Grothmann, T., & Patt, A. (2005). Adaptive capacity and human cognition: The process of individual adaptation to climate change. Global Environmental Change, 15(3), 199–213. https://doi.org/10.1016/j.gloenvcha.2005.01.002 Guest, G., & Fleming, P. (2015). Mixed Methods Research (pp. 581–610). https://doi.org/10.4135/9781483398839.n19 Ha, T. M., Boulom, S., Yang, F., Voe, P., Dao, C. D., Le, T. T. L., Dang, X. P., Hoang, T. T. H., Veu, T., Ngy, S., & Ha, D. A. (2023). Factors influencing farmers’ climate change adaptation in Southeast Asia: A comparative study from Vietnam, Laos, and Cambodia. APN Science Bulletin, 13(1), 40–49. https://doi.org/10.30852/10.30852/sb.2023.2101/ Ha Vo, H., Mizunoya, T., & Dinh Nguyen, C. (2021). Determinants of farmers’ adaptation decisions to climate change in the central coastal region of Vietnam. Asia-Pacific Journal of Regional Science, 5, 327–349. https://doi.org/10.1007/s41685-020-00181-5 103 Hagos, Y. G., Andualem, T. G., Mengie, M. A., Ayele, W. T., & Malede, D. A. (2022). Suitable dam site identification using GIS-based MCDA: A case study of Chemoga watershed, Ethiopia. Applied Water Science, 12(4), 69. https://doi.org/10.1007/s13201-022-01592-9 Hartung, C., Lerer, A., Anokwa, Y., Tseng, C., Brunette, W., & Borriello, G. (2010). Open Data Kit:Tools to Build Information Services for Developing Regions. Proceedings of the 4th ACM/IEEE International Conference on Information and Communication Technologies and Development, 1–12. https://doi.org/10.1145/2369220.2369236 He, F., Bremerich, V., Zarfl, C., Geldmann, J., Langhans, S. D., David, J. N. W., Darwall, W., Tockner, K., & Jähnig, S. C. (2018). Freshwater megafauna diversity: Patterns, status and threats. Diversity and Distributions, 24(10), 1395–1404. https://doi.org/10.1111/ddi.12780 He, Y., & Ahmed, T. (2022). Farmers’ Livelihood Capital and Its Impact on Sustainable Livelihood Strategies: Evidence from the Poverty-Stricken Areas of Southwest China. Sustainability, 14(9), 4955. https://doi.org/10.3390/su14094955 Hedges, L. V., & Hedberg, E. C. (2007). Intraclass Correlation Values for Planning Group- Randomized Trials in Education. Educational Evaluation and Policy Analysis, 29(1), 60– 87. https://doi.org/10.3102/0162373707299706 Hensengerth, O. (2018). South-South technology transfer: Who benefits? A case study of the Chinese-built Bui dam in Ghana. Energy Policy, 114, 499–507. https://doi.org/10.1016/j.enpol.2017.12.039 Ho, M., Shaw, D., Lin, S., & Chiu, Y. (2008). How Do Disaster Characteristics Influence Risk Perception? Risk Analysis, 28(3), 635–643. https://doi.org/10.1111/j.1539- 6924.2008.01040.x Hofer, M. K., Robillard, C. L., Legg, N. K., & Turner, B. J. (2024). Influence of perceived peer behavior on engagement in self-damaging behaviors during the transition to university. Journal of Research on Adolescence. https://doi.org/10.1111/jora.12933 Hoffmann, J. P. (2016). Regression models for categorical, count, and related variables: An applied approach. In Regression Models for Categorical, Count, and Related Variables: An Applied Approach. University of California Press. Hortle, K. G. (2007). Consumption and the yield of fish and other aquatic animals from the Lower Mekong Basin (MRC Technical Paper No. 16, p. 87). Mekong River Commission. Howe, L. D., Hargreaves, J. R., & Huttly, S. R. A. (2008). Issues in the construction of wealth indices for the measurement of socio-economic position in low-income countries. Emerging Themes in Epidemiology, 5, 1–14. https://doi.org/10.1186/1742-7622-5-3 Hu, L., & Bentler, P. M. (1998). Fit indices in covariance structure modeling: Sensitivity to underparameterized model misspecification. Psychological Methods, 3(4), 424–453. https://doi.org/10.1037/1082-989X.3.4.424 104 Huang, S., Zhang, X., Chen, N., Li, B., Ma, H., Xu, L., Li, R., & Niyogi, D. (2021). Drought propagation modification after the construction of the Three Gorges Dam in the Yangtze River Basin. Journal of Hydrology, 603, 127138. https://doi.org/10.1016/j.jhydrol.2021.127138 Huang, Y., Long, H., Jiang, Y., Feng, D., Ma, Z., & Mumtaz, F. (2024). Motivating factors of farmers’ adaptation behaviors to climate change in China: A meta-analysis. Journal of Environmental Management, 359. https://doi.org/10.1016/j.jenvman.2024.121105 IPCC. (2012). Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation. In Summary for Policymakers. In: Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation (p. 551). International Panel on Climate Change. Islam, R., & Walkerden, G. (2014). How bonding and bridging networks contribute to disaster resilience and recovery on the Bangladeshi coast. International Journal of Disaster Risk Reduction, 10(PA), 281–291. https://doi.org/10.1016/j.ijdrr.2014.09.016 Jiao, X., Zheng, Y., & Liu, Z. (2020). Three-stage quantitative approach of understanding household adaptation decisions in rural Cambodia. International Journal of Climate Change Strategies and Management, 12(1), 39–58. https://doi.org/10.1108/IJCCSM-01- 2019-0004 Jin, J., Kuang, F., Wan, X., He, R., & Ning, J. (2021). Exploring the effects of sociocognitive factors on the adaptation behaviour of farmers in rural Southwest China. Climate and Development, 13(2), 164–172. https://doi.org/10.1080/17565529.2020.1740078 Jin, J., Wang, W., & Wang, X. (2016). Adapting agriculture to the drought hazard in rural China: Household strategies and determinants. Natural Hazards, 82(3), 1609–1619. https://doi.org/10.1007/s11069-016-2260-x Johnson, B. B., & Kim, B. (2023). COVID-19 risk perception measures: Factoring and prediction of behavioral intentions and policy support. Journal of Risk Research, 26(11), 1191–1212. https://doi.org/10.1080/13669877.2023.2264301 Kaiser, H. F. (1974). An index of factorial simplicity. Psychometrika, 39(1), 31–36. https://doi.org/10.1007/BF02291575 Kasem, S., & Thapa, G. B. (2011). Crop diversification in Thailand: Status, determinants, and effects on income and use of inputs. Land Use Policy, 28(3), 618–628. https://doi.org/10.1016/j.landusepol.2010.12.001 Kellens, W., Terpstra, T., & De Maeyer, P. (2013). Perception and Communication of Flood Risks: A Systematic Review of Empirical Research. Risk Analysis, 33(1), 24–49. https://doi.org/10.1111/j.1539-6924.2012.01844.x Keovilignavong, O., Nguyen, T. H., & Hirsch, P. (2023). Reviewing the causes of Mekong drought before and during 2019–20. International Journal of Water Resources Development, 39(1), 155–175. https://doi.org/10.1080/07900627.2021.1967112 105 Keskinen, M. (2006a). The Lake with Floating Villages: Socio-economic Analysis of the Tonle Sap Lake. International Journal of Water Resources Development, 22(3), 463–480. https://doi.org/10.1080/07900620500482568 Keskinen, M. (2006b). The lake with floating villages: Socio-economic analysis of the Tonle Sap Lake. International Journal of Water Resources Development, 22(3), 463–480. https://doi.org/10.1080/07900620500482568 Khan, I., Lei, H., Shah, I. A., Ali, I., Khan, I., Muhammad, I., Huo, X., & Javed, T. (2020). Farm households’ risk perception, attitude and adaptation strategies in dealing with climate change: Promise and perils from rural Pakistan. Land Use Policy, 91(August 2019), 104395. https://doi.org/10.1016/j.landusepol.2019.104395 Khan, N. A., Qiao, J., Abid, M., & Gao, Q. (2021). Understanding farm-level cognition of and autonomous adaptation to climate variability and associated factors: Evidence from the rice-growing zone of Pakistan. Land Use Policy, 105(March 2020), 105427. https://doi.org/10.1016/j.landusepol.2021.105427 Kirchherr, J., Pohlner, H., & Charles, K. J. (2016). Cleaning up the big muddy: A meta-synthesis of the research on the social impact of dams. Environmental Impact Assessment Review, 60, 115–125. https://doi.org/10.1016/j.eiar.2016.02.007 Kocornik-Mina, A., McDermott, T. K. J., Michaels, G., & Rauch, F. (2020). Flooded Cities. American Economic Journal: Applied Economics, 12(2), 35–66. https://doi.org/10.1257/app.20170066 Kramer, D. B., Mitterling, A., & Frank, K. A. (2016). Understanding Peer Influence in Hunter Harvest Decisions Using Social Network Theory and Analysis. Human Dimensions of Wildlife, 21(5), 414–426. https://doi.org/10.1080/10871209.2016.1183243 Kuik, C.-C., & Rosli, Z. (2023). Laos-China infrastructure cooperation: Legitimation and the limits of host-country agency. Journal of Contemporary East Asia Studies, 12(1), 32–58. https://doi.org/10.1080/24761028.2023.2274236 Kuriqi, A., Pinheiro, A. N., Sordo-Ward, A., Bejarano, M. D., & Garrote, L. (2021). Ecological impacts of run-of-river hydropower plants—Current status and future prospects on the brink of energy transition. Renewable and Sustainable Energy Reviews, 142, 110833. https://doi.org/10.1016/j.rser.2021.110833 Le Dang, H., Li, E., Nuberg, I., & Bruwer, J. (2014). Farmers’ Perceived Risks of Climate Change and Influencing Factors: A Study in the Mekong Delta, Vietnam. Environmental Management, 54(2), 331–345. https://doi.org/10.1007/s00267-014-0299-6 Lemos, M. C., Lo, Y. J., Nelson, D. R., Eakin, H., & Bedran-Martins, A. M. (2016). Linking development to climate adaptation: Leveraging generic and specific capacities to reduce vulnerability to drought in NE Brazil. Global Environmental Change, 39, 170–179. https://doi.org/10.1016/j.gloenvcha.2016.05.001 106 Li, C.-H. (2016). Confirmatory factor analysis with ordinal data: Comparing robust maximum likelihood and diagonally weighted least squares. Behavior Research Methods, 48(3), 936–949. https://doi.org/10.3758/s13428-015-0619-7 Li, J., Dong, S., Peng, M., Yang, Z., Liu, S., Li, X., & Zhao, C. (2013). Effects of damming on the biological integrity of fish assemblages in the middle Lancang-Mekong River basin. Ecological Indicators, 34, 94–102. https://doi.org/10.1016/j.ecolind.2013.04.016 Li, L., Jin, J., He, R., Kuang, F., Zhang, C., & Qiu, X. (2023). Effects of social capital on farmers’ choices of climate change adaptation behavior in Dazu District, China. Climate and Development, 15(2), 110–121. https://doi.org/10.1080/17565529.2022.2061403 Lin, Z., & Qi, J. (2019). A new remote sensing approach to enrich hydropower dams’ information and assess their impact distances: A case study in the Mekong River Basin. Remote Sensing, 11(24). https://doi.org/10.3390/rs11243016 Liu, D., Li, Y., Shen, X., Xie, Y., & Zhang, Y. (2018). Flood risk perception of rural households in western mountainous regions of Henan Province, China. International Journal of Disaster Risk Reduction, 27, 155–160. https://doi.org/10.1016/J.IJDRR.2017.09.051 Lovgren, S. (2020). Cambodia’s biggest lake is running dry, taking forests and fish with it. National Geographic. Ma, J., Zhou, W., Guo, S., Deng, X., Song, J., & Xu, D. (2022). The influence of peer effects on farmers’ response to climate change: Evidence from Sichuan Province, China. Climatic Change, 175(1–2). https://doi.org/10.1007/s10584-022-03463-3 Maavara, T., Chen, Q., Van Meter, K., Brown, L. E., Zhang, J., Ni, J., & Zarfl, C. (2020). River dam impacts on biogeochemical cycling. Nature Reviews Earth & Environment, 1(2), 103–116. https://doi.org/10.1038/s43017-019-0019-0 Manohar, S., Downs, S., Shaikh, S., Mak, S., Sok, S., Graham, E., Miachon, L., & Fanzo, J. (2023). Riverine food environments and food security: A case study of the Mekong River, Cambodia. Bulletin of the World Health Organization, 101(02), 140–148. https://doi.org/10.2471/BLT.22.288830 Manski, C. F. (2000). Economic analysis of social interactions. Journal of Economic Perspectives, 14(3), 115–136. https://doi.org/10.1257/jep.14.3.115 Matouš, P., Todo, Y., & Mojo, D. (2013). Roles of extension and ethno-religious networks in acceptance of resource-conserving agriculture among Ethiopian farmers. 5903. https://doi.org/10.1080/14735903.2012.751701 Matsui, S., Keskinen, M., Sokhem, P., & Masahisa Nakamura. (2005). Tonle Sap: Experience and Lessons Learned Brief. Tonle Sap, 1, 407–419. Matuschke, I., & Qaim, M. (2009). The impact of social networks on hybrid seed adoption in India. Agricultural Economics, 40, 493–505. https://doi.org/10.1111/j.1574- 0862.2009.00393.x 107 Mayer, A., Lopez, M. C., Leturcq, G., & Moran, E. (2022). Changes in Social Capital Associated with the Construction of the Belo Monte Dam: Comparing a Resettled and a Host Community. Human Organization, 81(1), 22–34. https://doi.org/10.17730/1938-3525- 81.1.22 McCabe, C. J., Halvorson, M. A., King, K. M., Cao, X., & Kim, D. S. (2022). Interpreting Interaction Effects in Generalized Linear Models of Nonlinear Probabilities and Counts. Multivariate Behav Res., 57, 243–263. https://doi.org/10.1080/00273171.2020.1868966 Miles, M. B., Huberman, A. M., & Saldana, J. (2014). Qualitative Data Analysis: A Methods Source book. Sage Publications. Mirchi, A., Madani, K., Watkins, D., & Ahmad, S. (2012). Synthesis of System Dynamics Tools for Holistic Conceptualization of Water Resources Problems. Water Resources Management, 26(9), 2421–2442. https://doi.org/10.1007/s11269-012-0024-2 Mize, T. (2019). Best Practices for Estimating, Interpreting, and Presenting Nonlinear Interaction Effects. Sociological Science, 6, 81–117. https://doi.org/10.15195/v6.a4 Moore, D., Dore, J., & Gyawali, D. (2010). The World Commission on Dams + 10: Revisiting the Large Dam Controversy. Water Alternatives, 3(2), 3–13. Moritz, L., Kuhn, L., Bobojonov, I., & Glauben, T. (2024). Assessing Peer Influence on Farmers’ Climate Adaptations: An Experimental Adoption of Index Insurance and Savings in Uzbekistan. Journal of Agricultural and Applied Economics, 56(1), 1–20. https://doi.org/10.1017/aae.2023.38 Morton, L. W., & Olson, K. R. (2018). The Pulses of the Mekong River Basin: Rivers and the Livelihoods of Farmers and Fishers. Journal of Environmental Protection, 09(04), 431– 459. https://doi.org/10.4236/jep.2018.94027 MRC. (2019a). MRC-Drought Management Strategy for the Lower Mekong Basin 2020-2025 (p. 71 pp.). MRC. (2019b). State of the Basin Report 2018. Mekong River Commission/Vientiane/Lao PDR. https://doi.org/ISSN 1728:3248 MRC/WUP-FIN. (2007). Hydrological , Environmental and Socio-Economic Modelling Tools for the Lower Mekong Basin Impact Assessment (Issue 5, p. 126). Mekong River Commission and Finnish Environment Institute Consultancy Consortium. http://www.eia.fi/wup-fin/wup- fin2/publications.htm Mulligan, M., Van Soesbergen, A., & Sáenz, L. (2020). GOODD, a global dataset of more than 38,000 georeferenced dams. Scientific Data, 7(1), 31. https://doi.org/10.1038/s41597- 020-0362-5 Naqvi, S. A. A., Ul Hassan, R. H., Wu, W., Shah, A. A., Makhdum, M. S. A., & Shah, S. A. R. (2020). Synergy between adaptations and resilience of livelihood from climate change vulnerability: A group-wise comparison of adapters and non-adapters. PLoS ONE, 15(8 August), 1–21. https://doi.org/10.1371/journal.pone.0236794 108 Nguyen, Q. H., & Tran, V. N. (2024). Temporal Changes in Water and Sediment Discharges: Impacts of Climate Change and Human Activities in the Red River Basin (1958–2021) with Projections up to 2100. Water, 16(8), 1155. https://doi.org/10.3390/w16081155 Nhung, P. T. (2017a). Impacts of Hydropower Development on Natural Resource Accessibility and the Livelihoods of Local People: The Case of Quang Nam Province in Vietnam. ASEAN-Canada Research Partnership Working Paper Series. Nhung, P. T. (2017b). Impacts of Hydropower Development on Natural Resource Accessibility and the Livelihoods of Local People: The Case of Quang Nam Province in Vietnam (Issue 7). https://www.jstor.org/stable/resrep17179 Nuorteva, P., Keskinen, M., & Varis, O. (2010). Water, livelihoods and climate change adaptation in the tonle sap lake area, Cambodia: Learning from the past to understand the future. Journal of Water and Climate Change, 1(1), 87–101. https://doi.org/10.2166/wcc.2010.010 Oeurng, C., Cochrane, T. A., Chung, S., Kondolf, M. G., Piman, T., & Arias, M. E. (2019). Assessing Climate Change Impacts on River Flows in the Tonle Sap Lake Basin, Cambodia. Water, 11(3), 618. https://doi.org/10.3390/w11030618 Oppenheimer, M., Campos, M., Warren, R., Birkmann, J., Luber, G., O’Neill, B., & Takahashi, K. (2014). IPCC-WGII-AR5-19. Emergent Risks and Key Vulnerabilities. Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, 1039–1099. Osiemo, J., Ruben, R., & Girvetz, E. (2021). Farmer Perceptions of Agricultural Risks; Which Risk Attributes Matter Most for Men and Women. Sustainability, 13(23), 12978. https://doi.org/10.3390/su132312978 Owusu, K., Asiedu, A. B., Yankson, P. W. K., & Boafo, Y. A. (2019). Impacts of Ghana’s Bui dam hydroelectricity project on the livelihood of downstream non-resettled communities. Sustainability Science, 14(2), 487–499. https://doi.org/10.1007/s11625-018-0588-8 Pham, N. T. T., Nong, D., & Garschagen, M. (2019). Farmers’ decisions to adapt to flash floods and landslides in the Northern Mountainous Regions of Vietnam. Journal of Environmental Management, 252(July), 109672. https://doi.org/10.1016/j.jenvman.2019.109672 Pham, Y., Reardon-Smith, K., Mushtaq, S., & Deo, R. C. (2020). Feedback modelling of the impacts of drought: A case study in coffee production systems in Viet Nam. Climate Risk Management, 30, 100255. https://doi.org/10.1016/j.crm.2020.100255 Piesse, M. (2016). Livelihoods and Food Security on the Mekong River. Future Directions International, May, 1–7. Pokharel, Y., Burbano, M., Jacob, R., Kang, H., Sridhar, V., & Hyndman, D. W. (2018). A Review of the Integrated Effects of Changing Climate, Land Use, and Dams on Mekong River Hydrology. Water, 10(Figure 1), 1–25. https://doi.org/10.3390/w10030266 109 Poulton, P. L., Dalgliesh, N. P., Vang, S., & Roth, C. H. (2016). Resilience of Cambodian lowland rice farming systems to future climate uncertainty. Field Crops Research, 198, 160–170. https://doi.org/10.1016/j.fcr.2016.09.008 Qiao, D., Xu, S., Xu, T., Hao, Q., & Zhong, Z. (2022). Gap between Willingness and Behaviors: Understanding the Consistency of Farmers’ Green Production in Hainan, China. International Journal of Environmental Research and Public Health, 19(18), 11351. https://doi.org/10.3390/ijerph191811351 Qiu, W., Zhong, Z., & Huang, Y. (2021). Impact of perceived social norms on farmers’ behavior of cultivated land protection: An empirical analysis based on mediating effect model. International Journal of Low-Carbon Technologies, 16(1), 114–124. https://doi.org/10.1093/ijlct/ctaa043 Quan, X., Song, X., Miao, J., Huang, C., Gao, F., Li, J., & Ying, L. (2023). Study on the substitutability of nighttime light data for SDG indicators: A case study of Yunnan Province. Frontiers in Environmental Science, 11, 1309547. https://doi.org/10.3389/fenvs.2023.1309547 Rana, I. A., Jamshed, A., Younas, Z. I., & Bhatti, S. S. (2020). Characterizing flood risk perception in urban communities of Pakistan. International Journal of Disaster Risk Reduction, 46, 101624. https://doi.org/10.1016/j.ijdrr.2020.101624 Reed, M. S., Podesta, G., Fazey, I., Geeson, N., Hessel, R., Hubacek, K., Letson, D., Nainggolan, D., Prell, C., Rickenbach, M. G., Ritsema, C., Schwilch, G., Stringer, L. C., & Thomas, A. D. (2013). Combining analytical frameworks to assess livelihood vulnerability to climate change and analyse adaptation options. Ecological Economics, 94, 66–77. https://doi.org/10.1016/j.ecolecon.2013.07.007 Richter, B. D., Postel, S., Revenga, C., Lehner, B., & Churchill, A. (2010). Lost in Development’s Shadow: The Downstream Human Consequences of Dams. 3(2), 14–42. Robert, A. (2017). A River in Peril: Human Activities and Environmental Impacts on the Lower Mekong River and Its Delta. 9157. https://doi.org/10.1080/00139157.2017.1374794 Rosseel, Y. (2012). lavaan: An R Package for Structural Equation Modeling. Journal of Statistical Software, 48(2). https://doi.org/10.18637/jss.v048.i02 Salaisook, P., Faysse, N., & Tsusaka, T. W. (2020). Reasons for adoption of sustainable land management practices in a changing context: A mixed approach in Thailand. Land Use Policy, 96(April), 104676. https://doi.org/10.1016/j.landusepol.2020.104676 Schlüter, M., Baeza, A., Dressler, G., Frank, K., Groeneveld, J., Jager, W., Janssen, M. A., McAllister, R. R. J., Müller, B., Orach, K., Schwarz, N., & Wijermans, N. (2017). A framework for mapping and comparing behavioural theories in models of social- ecological systems. Ecological Economics, 131, 21–35. https://doi.org/10.1016/j.ecolecon.2016.08.008 110 Schuler, M. S., Tucker, J. S., Pedersen, E. R., & D’Amico, E. J. (2019). Relative influence of perceived peer and family substance use on adolescent alcohol, cigarette, and marijuana use across middle and high school. Addictive Behaviors, 88, 99–105. https://doi.org/10.1016/j.addbeh.2018.08.025 Scoones, I. (1999). Sustainable Rural Livelihoods: A Framework for Analysis. In Development (Basingstoke) (Vol. 42, Issue 2, pp. 57–63). https://doi.org/10.1057/palgrave.development.1110037 Seng, S. (2020). How Local People Use Their Indigenous Knowledge to Respond to Floods and Droughts: A Case Study of Tonle Sap Lake Community, Cambodia [Open Access Te Herenga Waka-Victoria University of Wellington]. https://doi.org/10.26686/wgtn.17145953 Shahbazbegian, M., & Bagheri, A. (2010). Rethinking assessment of drought impacts: A systemic approach towards sustainability. Sustainability Science, 5(2), 223–236. https://doi.org/10.1007/s11625-010-0110-4 Shi, J., Visschers, V. H. M., Siegrist, M., & Arvai, J. (2016). Knowledge as a driver of public perceptions about climate change reassessed. Nature Climate Change, 6(8), 759–762. https://doi.org/10.1038/nclimate2997 Shrestha, B., Babel, M. S., Maskey, S., Van Griensven, A., Uhlenbrook, S., Green, A., & Akkharath, I. (2013). Impact of climate change on sediment yield in the Mekong River basin: A case study of the Nam Ou basin, Lao PDR. Hydrology and Earth System Sciences, 17(1), 1–20. https://doi.org/10.5194/hess-17-1-2013 Shrestha, R. P., Chaweewan, N., & Arunyawat, S. (2017). Adaptation to climate change by rural ethnic communities of Northern Thailand. Climate, 5(3). https://doi.org/10.3390/cli5030057 Siciliano, G., Urban, F., Kim, S., & Dara Lonn, P. (2015). Hydropower, social priorities and the rural–urban development divide: The case of large dams in Cambodia. Energy Policy, 86, 273–285. https://doi.org/10.1016/j.enpol.2015.07.009 Siciliano, G., Urban, F., Tan-Mullins, M., Pichdara, L., & Kim, S. (2016). The Political Ecology of Chinese Large Dams in Cambodia: Implications, Challenges and Lessons Learnt from the Kamchay Dam. Water, 8(9), 405. https://doi.org/10.3390/w8090405 Siegrist, M., & Árvai, J. (2020). Risk Perception: Reflections on 40 Years of Research. Risk Analysis, 40(S1), 2191–2206. https://doi.org/10.1111/risa.13599 Sifton, J. (2021). Underwater: Human rights impacts of a China belt and road project in Cambodia. Human Rights Watch. Simoni, J. M., Franks, J. C., Lehavot, K., & Yard, S. S. (2011). Peer interventions to promote health: Conceptual considerations. American Journal of Orthopsychiatry, 81(3), 351– 359. https://doi.org/10.1111/j.1939-0025.2011.01103.x 111 Singhal, A., Sahu, S., Chattopadhyay, S., Mukherjee, A., & Bhanja, S. N. (2020). Using night time lights to find regional inequality in India and its relationship with economic development. PLOS ONE, 15(11), e0241907. https://doi.org/10.1371/journal.pone.0241907 Sithirith, M. (2021). Downstream State and Water Security in the Mekong Region: A Case of Cambodia between Too Much and Too Little Water. Water, 13(6), 802. https://doi.org/10.3390/w13060802 Sjöberg, L. (2000). Factors in Risk Perception. Risk Analysis, 20(1), 1–12. https://doi.org/10.1111/0272-4332.00001 Slovic, P. (1987). Perception of Risk. Science, 236(4799), 280–285. https://doi.org/10.1126/science.3563507 Slovic, P. (2016). The perception of risk. Scientists Making a Difference: One Hundred Eminent Behavioral and Brain Scientists Talk about Their Most Important Contributions, August 2016, 179–182. https://doi.org/10.1017/CBO9781316422250.040 Slovic, P., Finucane, M., Peters, E., & Macgregor, D. G. (2012). Risk as feeling: Some thoughts about affect, reason, risk and rationality. The Ethics of Technological Risk, 24(2), 163– 181. https://doi.org/10.4324/9781849772990 Slovic, P., & Weber, E. U. (2002). Perception of Risk Posed by Extreme Events. https://www.ldeo.columbia.edu/chrr/documents/meetings/roundtable/white_papers/slovic _wp.pdf Smajgl, A., & Ward, J. (2013). The Water-Food- Energy Nexus in the Mekong Region: Assessing Development Strategies Considering Cross-Sectoral and Transboundary Impacts. © Springer Science+Business Media New York. https://doi.org/10.1007/978-1-4614-6120-3 Smit, B., Burton, I., Klein, R. J. T., & Street, R. (1999). The Science of Adaptation: A Framework for Assessment. Mitigation and Adaptation Strategies for Global Change, 4(3), 199–213. https://doi.org/10.1023/A:1009652531101 Smit, B., Burton, I., Klein, R. J. T., & Wandel, J. (2000). An Anatomy of Adaptation to Climate Change and Variability. Climatic Change, 45(1), 223–251. https://doi.org/10.1023/A:1005661622966 Smit, B., & Pilifosova, O. (2003). Adaptation to Climate Change in the Context of Sustainable Development and Equity. Sustainable Development, 8, 879–912. Smit, B., & Wandel, J. (2006). Adaptation, adaptive capacity and vulnerability. Global Environmental Change, 16(3), 282–292. https://doi.org/10.1016/j.gloenvcha.2006.03.008 Snijders, T. A., & Bosker, R. J. (2011). Multilevel analysis: An introduction to basic and advanced multilevel modeling. Sage Publications. 112 Sok, S., & Yu, X. (2015). Adaptation, resilience and sustainable livelihoods in the communities of the Lower Mekong Basin, Cambodia. International Journal of Water Resources Development, 31(4), 575–588. https://doi.org/10.1080/07900627.2015.1012659 Soukhaphon, A., Baird, I. G., & Hogan, Z. S. (2021a). The Impacts of Hydropower Dams in the Mekong River Basin: A Review. Water, 13(3), 265. https://doi.org/10.3390/w13030265 Soukhaphon, A., Baird, I. G., & Hogan, Z. S. (2021b). The impacts of hydropower dams in the mekong river basin: A review. Water (Switzerland), 13(3), 1–18. https://doi.org/10.3390/w13030265 Spruce, J., Bolten, J., Mohammed, I. N., Srinivasan, R., & Lakshmi, V. (2020). Mapping Land Use Land Cover Change in the Lower Mekong Basin From 1997 to 2010. Frontiers in Environmental Science, 8. https://doi.org/10.3389/fenvs.2020.00021 Steenbergen, M. R., & Jones, B. S. (2002). Modeling Multilevel Data Structures. American Journal of Political Science, 46(1), 218. https://doi.org/10.2307/3088424 Stevens, K., Frank, K. A., & Kramer, D. B. (2015). Do Social Networks Influence Small-Scale Fishermen ’ s Enforcement of Sea Do Social Networks Influence Small-Scale Fishermen ’ s Enforcement of Sea Tenure ? March. https://doi.org/10.1371/journal.pone.0121431 Taberna, A., Filatova, T., Hadjimichael, A., & Noll, B. (2023). Uncertainty in boundedly rational household adaptation to environmental shocks. Proceedings of the National Academy of Sciences, 120(44), e2215675120. https://doi.org/10.1073/pnas.2215675120 Tang, J., Folmer, H., & Xue, J. (2013a). Estimation of awareness and perception of water scarcity among farmers in the Guanzhong Plain, China, by means of a structural equation model. Journal of Environmental Management, 126, 55–62. https://doi.org/10.1016/j.jenvman.2013.03.051 Tang, J., Folmer, H., & Xue, J. (2013b). Estimation of awareness and perception of water scarcity among farmers in the Guanzhong Plain, China, by means of a structural equation model. Journal of Environmental Management, 126, 55–62. https://doi.org/10.1016/j.jenvman.2013.03.051 Tareq, M., Abdel-Razzaq, A. I., Rahman, M. A., & Choudhury, T. (2021). Comparison of weighted and unweighted methods of wealth indices for assessing SOCIO-ECONOMIC status. Heliyon, 7(2), e06163. https://doi.org/10.1016/j.heliyon.2021.e06163 Tilt, B., Braun, Y., & He, D. (2009). Social impacts of large dam projects: A comparison of international case studies and implications for best practice. Journal of Environmental Management, 90, S249–S257. https://doi.org/10.1016/j.jenvman.2008.07.030 Tran, T. A., & Rodela, R. (2019). Integrating farmers’ adaptive knowledge into flood management and adaptation policies in the Vietnamese Mekong Delta: A social learning perspective. Global Environmental Change, 55(May 2018), 84–96. https://doi.org/10.1016/j.gloenvcha.2019.02.004 113 Trenchard, T. (2023). Cambodia’s Tonlé Sap Lake is running out of fish. Geographical. https://geographical.co.uk/news/fishing-crisis-cambodia-tonle-sap Trung, L. D., Duc, N. A., Nguyen, L. T., Thai, T. H., Khan, A., Rautenstrauch, K., & Schmidt, C. (2020). Assessing cumulative impacts of the proposed Lower Mekong Basin hydropower cascade on the Mekong River floodplains and Delta – Overview of integrated modeling methods and results. Journal of Hydrology, 581, 122511. https://doi.org/10.1016/j.jhydrol.2018.01.029 Udmale, P., Ichikawa, Y., Manandhar, S., Ishidaira, H., & Kiem, A. S. (2014). Farmers׳ perception of drought impacts, local adaptation and administrative mitigation measures in Maharashtra State, India. International Journal of Disaster Risk Reduction, 10, 250–269. https://doi.org/10.1016/j.ijdrr.2014.09.011 Urban, F., Nordensvärd, J., Khatri, D., & Wang, Y. (2013). An analysis of China’s investment in the hydropower sector in the Greater Mekong Sub-Region. Environment, Development and Sustainability, 15(2), 301–324. https://doi.org/10.1007/s10668-012-9415-z Urquijo, J., & De Stefano, L. (2016). Perception of drought and local responses by farmers: A perspective from the Jucar river Basin, Spain. Water Resources Management, 30(2), 577– 591. https://doi.org/10.1007/s11269-015-1178-5 Van Duinen, R., Filatova, T., Geurts, P., & van der Veen, A. (2015). Empirical Analysis of Farmers’ Drought Risk Perception: Objective Factors, Personal Circumstances, and Social Influence. Risk Analysis, 35(4), 741–755. https://doi.org/10.1111/risa.12299 van Rijn, F., Bulte, E., & Adekunle, A. (2012). Social capital and agricultural innovation in Sub- Saharan Africa. Agricultural Systems, 108, 112–122. https://doi.org/10.1016/j.agsy.2011.12.003 Vasquez, A. K., Foditsch, C., Dulièpre, S.-A. C., Siler, J. D., Just, D. R., Warnick, L. D., Nydam, D. V., & Sok, J. (2019). Understanding the effect of producers’ attitudes, perceived norms, and perceived behavioral control on intentions to use antimicrobials prudently on New York dairy farms. PLOS ONE, 14(9), e0222442. https://doi.org/10.1371/journal.pone.0222442 Vida, T. (2017, January 13). Battambang dam 98 percent complete. Khmer Times. https://www.khmertimeskh.com/60218/battambang-dam-98-percent-complete/ VIIRS Nighttime Light (VNL) (Version 2.1). (2021). [Dataset]. https://eogdata.mines.edu/products/vnl/ Vo, H. H., Mizunoya, T., & Nguyen, C. D. (2021). Determinants of farmers’ adaptation decisions to climate change in the central coastal region of Vietnam. Asia-Pacific Journal of Regional Science, 5(2), 327–349. https://doi.org/10.1007/s41685-020-00181-5 Wachinger, G., Renn, O., Begg, C., & Kuhlicke, C. (2013). The risk perception paradox- implications for governance and communication of natural hazards. Risk Analysis, 33(6), 1049–1065. https://doi.org/10.1111/j.1539-6924.2012.01942.x 114 Walpole, H. D., & Wilson, R. S. (2021). A Yardstick for Danger: Developing a Flexible and Sensitive Measure of Risk Perception. Risk Analysis, 41(11), 2031–2045. https://doi.org/10.1111/risa.13704 Wang, J., Walter, B. A., Yao, F., Song, C., Ding, M., Maroof, A. S., Zhu, J., Fan, C., McAlister, J. M., Sikder, S., Sheng, Y., Allen, G. H., Crétaux, J.-F., & Wada, Y. (2022). GeoDAR: Georeferenced global dams and reservoirs dataset for bridging attributes and geolocations. Earth System Science Data, 14(4), 1869–1899. https://doi.org/10.5194/essd-14-1869-2022 Watts, L. L., Hamza, E. A., Bedewy, D. A., & Moustafa, A. A. (2024). A meta-analysis study on peer influence and adolescent substance use. Current Psychology, 43(5), 3866–3881. https://doi.org/10.1007/s12144-023-04944-z Wilson, R. S., Zwickle, A., & Walpole, H. (2019a). Developing a Broadly Applicable Measure of Risk Perception. Risk Analysis, 39(4), 777–791. https://doi.org/10.1111/risa.13207 Wilson, R. S., Zwickle, A., & Walpole, H. (2019b). Developing a Broadly Applicable Measure of Risk Perception. Risk Analysis, 39(4), 777–791. https://doi.org/10.1111/risa.13207 World Bank. (2023). Cambodia Economic Update—Post-COVID-19 Economic Recovery: Special Focus—From Spending More to Spending Better: Toward Improved Human Development Outcomes (English) (No. 182137; p. 21). World Bank. http://documents.worldbank.org/curated/en/099051523221517821/P17734003f2bcf02b0a 89500f61b1f3ff7b Wossen, T., Berger, T., & Mequaninte, T. (2013). Social network effects on the adoption of sustainable natural resource management practices in Ethiopia. International Journal of Sustainable Development & World Ecology, 20(6), 477–483. https://doi.org/10.1080/13504509.2013.856048 Wouters, P., Daza-Clark, A. M., & Devlaeminck, D. J. (2024). China’s transboundary hydropower development at home and abroad: Exploring the regulatory interface between international water law and international economic law. Frontiers in Climate, 5, 1302103. https://doi.org/10.3389/fclim.2023.1302103 Xu, Z., Qayum, M., Afzal, J., & Aslam, M. (2023). Availability and access to Livelihood capital assets for development of sustainable Livelihood strategies of fishermen: A case study of Manchar Lake Pakistan. Heliyon, 9(12), e22549. https://doi.org/10.1016/j.heliyon.2023.e22549 Yaméogo, T. B., Fonta, W. M., & Wünscher, T. (2018). Can social capital influence smallholder farmers’ climate-change adaptation decisions? Evidence from three semi-arid communities in Burkina Faso, West Africa. Social Sciences, 7(3), 1–20. https://doi.org/10.3390/socsci7030033 Yamsiri, T. (2014). Water management in Thailand: Dams and the voice of the affected and displaced people (pp. 235–251). SciencesPo. 115 Yang, L., Chien, H., Chen, Y., & Wu, Z. (2020). Land-use change for maize in china and its driving forces: A multilevel modelling approach based on household data. Transformations in Business and Economics, 19(3), 206–234. Yang-jie, W., Ji-kun, H., & Jin-xia, W. (2014). Household and Community Assets and Farmers ’ Adaptation to Extreme Weather Event: The Case of Drought in China. Journal of Integrative Agriculture, 13(4), 687–697. https://doi.org/10.1016/S2095-3119(13)60697-8 Yoshida, Y., Lee, H. S., Trung, B. H., Tran, H. D., Lall, M. K., Kakar, K., & Xuan, T. D. (2020). Impacts of mainstream hydropower dams on fisheries and agriculture in lower mekong basin. Sustainability (Switzerland), 12(6), 1–21. https://doi.org/10.3390/su12062408 Zhang, C., Jin, J., Kuang, F., Ning, J., Wan, X., & Guan, T. (2020). Farmers’ perceptions of climate change and adaptation behavior in Wushen Banner, China. Environmental Science and Pollution Research, 27(21), 26484–26494. https://doi.org/10.1007/s11356- 020-09048-w Zhang, W. (2017). Household risk aversion and portfolio choices. Mathematics and Financial Economics, 11(3), 369–381. https://doi.org/10.1007/s11579-017-0184-1 Zhao, Q., Liu, S., Deng, L., Dong, S., Yang, Z., & Liu, Q. (2013). Determining the influencing distance of dam construction and reservoir impoundment on land use: A case study of Manwan Dam, Lancang River. Ecological Engineering, 53, 235–242. https://doi.org/10.1016/j.ecoleng.2012.12.050 116 APPENDIX A: MANN-WHITNEY U AND INDEPENDENT T-TEST RESULTS Table 10: Preliminary test results showing temporal variation in proxy measures of diverse livelihood resources Independent t -test over time Null hypothesis: population means equal for both groups (2013 and 2019) Resource Measures Indicators Unit of measurement Access to land Natural Access to water Access to forest Access to laborers Access to education Access to knowledge and skills Enagagement in cultivation Ownership of land Parcel for livestock Parcel for aquaculture Parcel for forest/wooded land Engagment in fish catching Engagment in forestry Proportion Proportion Proportion Proportion Proportion Proportion Proportion Total household members available for economic activities Number Education of HH head Experiences in farming Proportion Years Results from non- parametric test 2013 2019 P-value Direction Not significant 0.841 (0.233) 0.9 (0.01) <0.001 Significant Significant Significant Significant 0.972 (0.049) 0.971 (0.053) 0.696 (0.260) 0.632 (0.171) 0.934 0.021 Same Decrease 0.001(0.017) 0.788(0.306) <.001 Increase 0.129 (0.243) 0.048 (0.068) <.001 Decrease Not significant 0.265 (0.306) 0.247 (0.298) Significant 0.002(0.014) 0.869(0.196) Significant Significant 4.652 (0.654) 4.19(0.52) 0.902 (0.169) 0.805 (0.110) 0.68 <.001 <.001 <.001 Increase Decrease Decrease Significant 48.598 (4.961) 46.737 (3.374) <.001 Decrease Access to agricultural information Proportion Not significant 0.357 (0.114) 0.361 (0.155) Wealth TLU TLU Access to credit Access to credit/loan for agricultural purposeProportion Access to banking facilities Access to microfinance Household using bank for credit Proportion Proportion Access to irrigation facilities Household with irrigation facilities Proportion Access to market Household using market either for sale or for information Proportion Physical Access to school Household with members attending schoolProportion Access to informal networks for credit Relatives/friends money lender Proportion Proportion Significant Significant Significant Significant Significant Significant Significant Significant Significant 1.5(1.06) 5.664 (16.035) 0.175 (0.065) 0.474 (0.165) 0.147 (0.203) 0.329 (0.248) 0.488(0.292) 0.525 (0.245) 0.243 (0.329) 0.32 (0.133) 1.132 (0.300) 0.195 (0.110) 0.253 (0.248) 0.01 (0.027) 0.201 (0.226) 0.047 (0.083) 0.218 (0.051) 0.006 (0.022) <.001 Decrease Human Financial Social 0.855 0.072 <.001 <.001 0.322 0.002 Increase Increase Increase Increase Increase <.001 <.001 <.001 0.135 <.001 Decrease Decrease Decrease Decrease Access to network for agricultural information Main source of agricultural information- farmers Proportion Not signficant 0.306 (0.310) 0.263 (0.174) Proportion Main source of agricultural information- Government Significant 0.573 (0.334) 0.099 (0.102) 117 Estimated effect size (Cohen's d) 0.329 0.013 0.290 0.363 0.455 0.061 0.623 0.782 0.683 0.439 0.030 0.366 1.387 0.804 0.138 0.306 0.534 0.415 1.379 0.906 0.173 1.922 APPENDIX B: MIXED EFFECT MODEL RESULTS Table 11: Mixed effect model results (N= 483) Variables Dam effect Impact zone 1 Impact zone 2 Dam type [IRR =1] Nightlight Interaction [dam effect * Impact zone1] Interaction [dam effect * impact zone 2] Interaction [dam effect * dam type=IRRI] R2 Adj. R2 Natural Social Human Financial Physical B EXP (B) B EXP (B) B EXP (B) B EXP (B) B EXP (B) -0.44*** (0.06) -0.09** (0.03) -0.02 (0.02) 0.02 (0.02) 0.03* (0.01) 0.05 (0.08) 0.13 (0.08) -0.27*** (0.07) 0.64 0.91 0.98 1.02 1.03 0.08 (0.08) 0.06 (0.04) -0.08* (0.04) -0.13*** (0.03) 0.02 (0.02) 1.08 1.06 0.92 0.88 1.02 1.05 -0.02 0.98 1.14 0.76 0.13 (0.12) 0.24** (0.09) 1.14 1.27 0.11’ (0.05) 0.01 (0.04) 0.01 (0.03) -0.01 (0.03) 0.01 (0.01) -0.13 (0.08) -0.02 (0.08) 0.06 (0.07) 1.12 1.01 1.01 0.99 1.01 0.88 0.98 1.06 -0.38*** (0.06) -0.03 (0.03) -0.03 (0.02) 0.02 (0.02) 0.01 (0.01) 0.3*** (0.08) 0.29*** (0.09) 0.02 (0.07) 0.68 0.97 0.97 1.02 1.01 1.35 1.34 1.02 0.27*** (0.04) 0.05* (0.02) 0.02 (0.02) 0.03 (0.02) 0 (0.01) -0.04 (0.06) 0.01 (0.07) -0.05 (0.05) 1.31 1.05 1.02 1.03 1.00 0.96 1.01 0.95 0.7 0.69 0.61 0.6 0.91 0.9 0.14 0.13 0.79 0.79 Note: value in parentheses is Standard Error (SE) ***p < 0.001; **p < 0.01; *p < 0.05 118 Table 12: Mixed effect model results (N= 100) Variables Dam effect Impact zone 1 Impact zone 2 Dam type [IRR =1] Nighlight Interaction [dam effect * Impact zone=1] Interaction [dam effect * Impact zone=2] Natural -0.50*** (-0.07) -0.13 (-0.08) -0.08 (-0.09) -0.01 (-0.07) 0.02 (-0.01) 0.07 (-0.1) 0.18' (-0.11) Social 0.08 (-0.09) 0.09 (-0.09) -0.07 (-0.1) -0.17* (-0.07) 0.01 (-0.01) -0.05 (-0.13) 0.12 (-0.14) Human 0.12* (-0.06) -0.01 (-0.06) 0.08 (-0.07) -0.01 (-0.05) 0.01 (-0.01) -0.13 (-0.08) -0.05 (-0.09) Financial -0.40** (-0.15) -0.04 (-0.15) -0.01 (-0.16) 0.01 (-0.13) 0 (-0.02) 0.3 (-0.22) 0.27 (-0.23) -0.23** Interaction [dam effect * dam (-0.08) type=IRRI] R2 0.7 Adj. R2 0.67 Note: Value is unstandardized Beta coefficient and value in parentheses is Standard Error (SE) ***p < 0.001; **p < 0.01; *p < 0.05 0.28** (-0.11) 0.27 0.2 0.03 (-0.18) 0.13 0.05 0.06 (-0.07) 0.19 0.12 Physical 0.27*** (-0.05) 0.03 (-0.05) -0.05 (-0.05) 0.06 (-0.04) 0 (-0.01) -0.03 (-0.07) 0.06 (-0.08) -0.07 (-0.06) 0.45 0.4 APPENDIX C: RESULTS FROM THEMATIC CODING Table 13: Responses to the question “Why did you make those farm related changes” (Household survey, 2022) Themes Increase rice yield/get better rice yield Increase in insects/pest/weeds Decrease in soil fertility/maintain soil fertility Drought/no rain Change in rice plantation method Facilitate farming/save time Changes in water availability Irrigation canal nearby farm No money Responses (%) 15.6 50.6 48.1 16.9 45.5 28.6 13 2.6 9.1 119 Table 14: Cross-tabulation between responses on “What changes have you made in your farming practices over past few years” and “Why did you make those farm related changes?” Reasons Respondents reporting increased use of these practices (%) Increase rice yield/get better rice yield Increase in insects/pest/weeds Decrease in soil fertility/maintain soil fertility Drought/no rain Change in rice plantation method Facilitate farming/save time Changes in water availability Irrigation canal nearby farm No money Note: Irrigation Pesticide Fertilizer 0.86 7.76 3.45 4.31 5.17 5.17 5.17 1.72 0.00 13.79 51.72 41.38 12.07 37.93 23.28 2.59 0.00 1.72 18.10 46.55 46.55 12.07 36.21 19.83 5.17 0.86 6.90 Planting time 13.79 New farm equipment 4.31 41.38 37.07 12.07 43.10 25.86 10.34 0.86 6.03 14.66 10.34 4.31 16.38 14.66 2.59 1.72 0.86 1. 2. Highlighted cells denote the significant association based on chi-square at p-value <0.05 Color denotes the strength of association based on Cramer value Moderate Strong Note: Moderate strength refers to value between 0.3 to 05 and strong refers to value greater than 0.5 (Cohen, 1988) 120 APPENDIX D: RESULTS FROM PROBIT AND POISSON MODELS, AND MARGINAL EFFECTS TESTING Table 15: Results from Poisson and Probit models (N=198) Variables Perceived peer exposure (PPE) Ties-NGO Ties-Government Ties-Private PEE*Ties-NGO PPE*Ties- Government PPE*Ties-Private Previous experience to multiple shocks Crop productivity Land ownership Wealth index Age AIC log-likelihood Nagelkerke pseudo-R2 Adaptation intensity (IRR) (Model a) 1.055*** (1.012) 1.384* (1.146) 1.579** (1.174) 1.858*** (1.154) 0.947* (1.022) 0.966 (1.023) 0.927*** (1.018) 1.093' (1.049) 0.844* (1.076) 1.090 (1.058) 0.925* (1.033) 0.993* (1.003) 744.996 -359.498 0.219 Adaptation choice (AME) Irrigation use (Model b) 0.019 (0.012) 0.211** (0.068) 0.084 (0.079) 0.119 (0.093) 0.062* (0.029) 0.06 (0.047) -0.099 * (0.044) 0.013 (0.026) -0.052 (0.043) 0.018 (0.016) -0.027 (0.02) -0.001 (0.002) 136.129 -55.0647 0.344 Pesticide use (Model c) 0.052*** (0.014) -0.205* (0.083) 0.091 (0.089) 0.118 (0.090) -0.023 (0.033) 0.004 (0.034) -0.039 (0.029) -0.016 (0.042) -0.099 (0.064) 0.060' (0.032) -0.014 (0.028) -0.003 (0.003) 248.125 -111.06 0.271 Fertilizer use (Model d) 0.012 (0.016) -0.153' (0.085) 0.033 (0.098) 0.199* (0.093) -0.045 (0.032) 0.0002 (0.039) -0.014 (0.031) -0.071 (0.044) -0.034 (0.069) 0.046 (0.034) -0.033 (0.030) -0.003 (0.003) 279.205 -126.603 0.105 Notes: ***, **, * , ' showing significant at <1%, 1%, 5%, and 10% probability level, respectively; robust standard errors are in parentheses. Table 16: Results for how adaptation intensity is associated with informal and formal networks: tests of Average Marginal effects (AMEs) and second differences (N = 198) Formal network Informal peer network Ties- NGO Perceived peer exposure Ties- Government Perceived peer exposure First difference in AME Without 0.103** 0.078* With -0.059 -0.027 Perceived peer exposure Note: ‘***’, ‘**’, ‘*’, ‘'’ refers to p-value <0.001, <0.01, <0.05 and <0.1 respectively 0.100** Private -0.157’ Second difference 0.162' 0.105 0.257** 121 Table 17: Results for farmers’ adaptation decision (specific behavior) are associated with informal and formal network: tests of average marginal effects (AMEs) and second differences (N = 198) Model Formal network Informal peer network First difference in AME Without With 0.062* 0.047 -0.099*** 0.004 0.015 0.038* Private Pesticide Irrigation Ties- NGO Ties- NGO Ties- Government Perceived peer exposure Perceived peer exposure Perceived peer exposure Perceived peer exposure Perceived peer exposure Perceived peer exposure Perceived peer exposure Perceived peer exposure Perceived peer exposure Note: ‘***’, ‘**’, ‘*’, ‘'’ refers to p-value <0.001, <0.01, <0.05 and <0.1 respectively Ties- Government Ties- Government Ties- NGO 0.061*** 0.078*** 0.067*** Fertilizer 0.0295' Private Private 0.014 0.017 -0.023 0.0002 -0.039 -0.045 0.0002 -0.014 Second difference -0.057' -0.045 0.138*** 0.101** 0.061 0.108*** 0.074' 0.014 0.031 122