TOWARDS BUILDING DRO UGHT RESILIENCE OF RICE PRODUCTION I N CAMBODIA: FROM A SYSTEM DYNAMIC S PERSPECTIVE By Tum Nhim A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Community Sustainability - Master of S cience 2015 ABSTRACT TOWARDS BUILDING DRO UGHT RESILIENCE OF R ICE PRODUCTION IN CAMBODIA: FROM A SYS TEM DYNAMIC S PERSPECTIVE By Tum N him rice cultivation is associated with high risks and uncertainties in the face of climate change. The projected increases in drought frequency and its uneven distribution over seasons and across places due to climate change c oupl ed with low adaptive capacity of rural farmers speaks to the necessity to build drought resilience across the country. The m ain objectives of this study were to identify sources of drought resil ience at household and commune level s and to find possible ways to improve resilience to drought . To meet the objectives , a system dynamics model for drought resilience was developed. Data from household survey were used to estimate major model input variables through descriptive statistics , to define farm househ old typology through two - step cluster analysis , and to estimate relation between variables using multiple regressions . The results of the study show that access to irrigation is the most important source of resilience at both household and community level s . Improving access to irrigation to the threshold lev el of approximately 40% can help maintain stability and continuous development of rice production over time. Another important source of drought resilience is agricultural diversification such as spatial diversification of paddy lands and varietal diversification. The group of farmers that is resilient to drought is associated with this characteristic as depicted in the linkage between farm typology and model outputs. On the other hand, nonfarm diversific ation such as remittance from migration and local wage s can be other sources of resilience to drought. However, it is to be noticed that the resilient group of farmers is associated with an average degree of dependence on both rice and nonfarm income s , den oting that depending too much on nonfarm income might draw resources away from agriculture. iii ACKNOWLEDGEMENT S First of all, I would like to express my deep thanks to Dr. Laura S ch mit t Olabisi , my helpful advis o r . Sh e gave me not only the new idea and comments regarding the research, but also a warm encouragement from the start to the end of my thesis write - up . Without h er immense thesis. Dr. John Kerr, Dr. Maria Claudia Lopez, and Dr. Jean - Christophe Diepart , who always help me in both academic and social life , especially related to this master thesis research. They also gave me a lot of helpful comments on my thesis writing and assisted me in technical issues. agency. This material is based upon work supported by the United States Agency for International Developm ent, as part of the Feed the Future initiative, under the CGIAR Fund, award number BFS - G - 11 - 00002, and the predecessor fund the Food Security and Crisis Mitigation II grant, award number EEM - G - 00 - 04 - 00013. This stud y have gone so smoothly without management team at MSU such as Dr. Anne Schneller, Theresa Doerr and Kathryn Greenhalgh . (LI) for co - funding this research in Cambodia. Special thanks to Rithy, Visal, and enumerators for helping me with the field work and local authority and villagers for participating in this research. iv TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ..................... v LIST OF FIGURES ................................ ................................ ................................ .................. vi 1. Introduction ................................ ................................ ................................ ........................ 1 1.1. Background and problem statement ................................ ................................ ............ 1 1.2. Research Questions ................................ ................................ ................................ ..... 3 1.3. Study Area ................................ ................................ ................................ ................... 4 2. Review of Literature ................................ ................................ ................................ ........... 6 3. Methods ................................ ................................ ................................ .............................. 9 3.1. Formulation of conceptual model for drought resilience ................................ ............ 9 3.2. Field data collection ................................ ................................ ................................ .. 10 3.3. Development of causal loop diagram and system dynamic model ........................... 12 3.4. Description of important model inputs/variables ................................ ...................... 14 3.5. Data analysis ................................ ................................ ................................ ............. 15 4. Results ................................ ................................ ................................ .............................. 18 4.1. Typology analysis ................................ ................................ ................................ ..... 18 4.2. R ice yield and uses of agricultural inputs ................................ ................................ . 20 4.3. Agricultural investment and rice profit ................................ ................................ ..... 21 4.4. Dependence on agriculture and other nonfarm activities ................................ .......... 24 4.5. Recovery from losses due to drought ................................ ................................ ........ 28 4.6. Drought impacts at community level ................................ ................................ ........ 30 4.7. Model Scenario Tests ................................ ................................ ................................ 31 5 . Discussion ................................ ................................ ................................ ......................... 39 6. Conclusion ................................ ................................ ................................ ........................ 43 7. Recommendations ................................ ................................ ................................ ............ 44 8. Limitat ions of study ................................ ................................ ................................ .......... 45 APPENDICES ................................ ................................ ................................ ......................... 46 AppendiX 1: Multiple regression for yield in 2014 (drought condition) and yield in 2013 (normal condition) ................................ ................................ ................................ ................... 47 AppendiX 2: Regression model for household income and fertilizer used ............................. 50 AppendiX 3: Model equations ................................ ................................ ................................ . 51 AppendiX 4 ................................ ................................ ................................ .............................. 56 BIBLIOGRAPHY ................................ ................................ ................................ .................... 5 7 v LIST OF TABLES Table 1: Demography of study communes ................................ ................................ ................ 4 Table 2: Rice farming in the study communes ................................ ................................ .......... 5 Table 3: Respondent profile ................................ ................................ ................................ ....... 5 Table 4: Main household occupations ................................ ................................ ....................... 5 Table 5: Descriptive statistics of rainfall ................................ ................................ ................. 14 Table 6: Characteristics of farmer groups ................................ ................................ ................ 18 Table 7: Descriptive statistics for agricultural input variables ................................ ................ 47 Table 8: Summary b of regression model for yield 2014 and agricultural inputs ..................... 47 Table 9: ANOVA a of regression model for yield 2014 and agricultural inputs ...................... 48 Table 10: Coefficients a of regression model for yield 2014 and agricultural inputs ............... 48 Table 11: Su mmary b of regression model for yield 2013 and agricultural inputs ................... 48 Table 12: ANOVA a of regression model for yield 2013 and agri cultural inputs .................... 49 Table 13: Coefficients a of regression model for yield 2013 and agricultural inputs ............... 49 Table 14: Descriptive statistics for household income and total fertilizer in 2014 ................. 50 Table 15: Summary of regression model for household income and fertilizer used ............... 50 Table 16: ANOVA of regression model for household income and fertilizer used ................ 50 Table 17: Coefficient of regression mo del for household income and fertilizer used ............. 50 vi LIST OF FIGURES Figure 1: Study area, Battambang province, Cambodia ................................ ............................ 4 Figure 2: Conceptual model for drought resilience ................................ ................................ . 10 Figure 3: Causal loop diagram depicting drought resilience ................................ ................... 12 Figure 4: Clustering model for farm typology ................................ ................................ ......... 18 Figure 5: Farm typology ................................ ................................ ................................ .......... 19 Figure 6: Rice yield for farmers in group 1, 2 and 3 ................................ ................................ 20 Figure 7: Fertilizers used per hectare ................................ ................................ ....................... 21 Figure 8: Agricultural investment by groups in USD ................................ .............................. 22 Figure 9: Net rice income in USD ................................ ................................ ........................... 23 Figure 10: Household income stock ................................ ................................ ........................ 24 Figure 11: Dependence on rice income ................................ ................................ ................... 25 Figure 12: Dependence on remittance from migration ................................ ............................ 25 Figure 13: Dependence on different livelihoo d sources for farmers in group 1 ...................... 26 Figure 14: Dependence on different livelihood sources for farmer group two ........................ 27 Figure 15: Dependence on different livelihood sources for farmers in group 3 ...................... 27 Figure 16: Simulated rice yield and normal rice yield for farmers in group 1 ........................ 28 Figure 17: Simulated rice yield and normal rice yield for farmers in group 2 ........................ 29 Figure 18: Simulated rice yield and normal rice yield for farmers in group 3 ........................ 29 Figure 19: Percentage of yield loss by groups ................................ ................................ ......... 30 Figure 20: Community actual harvested area and percentage of affected households ............ 31 Figure 21: Percentage of yield losses for farmer group 1 under baseline, area increase, and area & fertilizer increase scenarios. ................................ ................................ ......................... 32 Figure 22: Percentage of yield losses for farmer group 3 under baseline, area increase, and area & fertilizer increase scenarios. ................................ ................................ ......................... 32 vii Figure 23: Net ri ce income for farmer group 1 under baseline, area increase, and area & fertilizer increase scenarios. ................................ ................................ ................................ ..... 33 Figure 24: Net rice income for farmer group 3 under baseline, area increase, and area & fertilizer increase scenarios. ................................ ................................ ................................ ..... 33 Figure 25: Actual harvested area under baseline, area increase, and area & fertilizer increase scenarios for Groups 1 and 3. ................................ ................................ ................................ .. 35 Figure 26: Percentage of dr ought - affected households under baseline, area increase, and area & fertilizer increase scenarios for Groups 1 and 3. ................................ ................................ . 35 Figure 27: Effect of increase access to irrigation on actual harvested area scenario ............... 36 Figure 28: Effect of increase access to irrigation on percentage of affected households scenario ................................ ................................ ................................ ................................ .... 37 Figure 29: Effect of adjusted migration rate on actual harvested area scenario ...................... 38 Figure 30: Effect of adjusted migration rate on percentage of affected household scenario ... 38 Figure 31: Model structure ................................ ................................ ................................ ....... 56 1 1. Introduction 1.1. Background and problem statement Cambodia largely depends on agriculture as this sector contributes up to 3 2 per cent of the employs approximately 64 per cent of total labor force (ADB, 2014) . Roughly 80 per cent of the population reside in rural areas and depend primarily on rice cultivation for their livelihoods (USDA - FAS, 2010) . According to USDA - FAS ( 2010) , rice is c ultivated over approximately 85 per cent of the total cultivated area, of which only 14 per cent is irrigated . Being an important source for rural livelihoods, rice is also a staple food for the Cambodian diet and consitutes 65 to 75 per cent of the total daily energy needs (Yu & Fan, 2009) . Despite the i mportance of rice farming for rural livelihood s , the majority of Cambodian farmers still practice traditional farming techniques and grow rice for subistance , just one crop cycle per year (Ros, Nang, & Chhim, 2011) . As rainfed rice production in the wet season accounts for most of t he total rice production in Cambodia , the highly e rratic rainfall pattern s associated with regular occurances of exteme climatic events such as flood and drought (USDA - FAS, 2010) can leave agricultural production of the country with continuing uncertainties in the face of climate change . In the agricultural sector, drought apparently is the most experienced natural shock for Cambodia n farmers (Thomas et al., 2013) . The very liklely shift in rainfall pattern s suggested by climate change projection s (Mcsweeney, New, & Lizcano, 200 8) , coupl ed with low adaptive capacity of rural farmers in almost every province (Yusuf & Francisco, 2009) makes the rice sector in Cambodia highly vulnerable to climate change , which requires building drought resilience. There are several reasons why building drought resilience in Cambodia is crucially important. First, d rought is a recurrent crisis whose accumulative impacts livelihood under pressure over time . The impacts of drought are multifaceted including 2 reduced crop production , reduced income, increased unemployment and migratio ns (Wilhite, Svoboda, & Hayes, 2007) , reduced consumption, selling productive assets and so on (Pandey, 2007) . In general, the indirect impacts surpass the direct ones , which is why attention is usually less paid by both farmers themselves and policy make rs compared to other hazards (Wilhite et al., 2007) . How ever, the accumulative losses caused by drought over time can be a June 1994 drought affected 5 million people (almost 50% of population in Cambodia during that time) and caused economic loss es of 100 million USD (EM - DAT, 2014) . Second, building drought resilience is a climate - smart strategy in response to climate change because drought is n aturally a slow - onset gradual process of accumulating impacts can also allow farmers to have more time to respond to and recover from it. Third, there is evidence that drought events and their magnitudes in Cambodia has increased over time . For instance, counting from 1950 to 2005, the probability of occurrence of a drought event in Cambodia is around 0.34 (Pandey, 2007) , meaning that an average drought could occur once every three year s . If climate change projection s for Cambodia depicting increases in extreme climatic event s such as drought (Mcsweeney et al., 2008) are accurate, building drought resilience in Cambodia is a priority that requires involvement from all relevant stakeholders . B ecause uncertainties of risks imposed by unpredictable natural shocks cannot be completely eliminated (Berkes, 2007) , it is necessary to learn to live with these changes and uncertainties and build capacity to deal with them , while sustaining and enhancing livelhood at the same time. Resilience i parts to anticipate, ab sorb, accommodate, or recover from the effects of a potentially hazardous event in a timely and efficient manner, including through ensuring the 3 (p. 5) . Resilience manifest s at multiple scales such as individual, household, community, and systems levels (Béné, Wood, Newsham, & Davies, 2012) and there are multiple sources of resilience at each of these levels (W. Neil Adger et al., 2011) . Furthermore, in social - ecological system perspective, the responses to a particular risk may adversely impair the capacity of the system to cope with other risks (W. Neil Adger et al., 2011) . Thus, to enhance resilience, i.e. reduce vulnerability, in the face of unexpec ted changes and uncertainties of climate extreme such as drought , for example, it has to understand resilience as a multifaceted and multiscalar concept . 1.2. Research Q uestions The main purpose of this study was to understand if the responses of farmer communit ies exposed to recurrent drought hazard s lead to greater resilience or greater vulnerability . To meet this objective, the following research questions need to be answered. o What are the sources of resilience to drought at household and community le vel s ? o What are the factors and/or processes that make a group of households more or less resilient to drought than another? o How do these factors and/or processes influence drought resilience at the community level ? 4 1.3. Study A rea Figure 1 : Study area, Battambang p rovince, Cambodia Two communes of Bannan district in Battambang p rovince of C ambodia, namely Chaeng Meanchey and Kanteur Mouy , were selected for this study . These communes are crossed by Stung Sangker River , which contain s several streams . The total population of these two communes was 14349 in 2010 , with 2998 familie s (Table 1) , the majority of which have rainfed rice farmin g as primary occupation (Table 2 ) . In average, active labor was approximately 49 per cent of the total population and migration rate was around 3 per cent of the active labor (Table 1). In these communes, the average wet rice yield ranged from 1.0 to 2.5 hectare s and varied from year to year (Table 2) . Table 1 : Demography of study communes Communes Population Number of families A ctive labor Migration rate Chaeng Meanchey 9,296 1,806 48.5% 1.9 % Kanteur Mouy 5,053 1,192 49 .0 % 5 .2 % Total 14349 2998 49.2% 3 .1 % ( NCDD , 2010) 5 Table 2 : Rice farming in the study communes Communes W et rice land area Average yield Rice farming as primary occupation Average rice price Chaeng Meanchey 5062 .00 ha 1.50 tons/ha 80.00 % 0.20 USD/kg Kanteur Mouy 4667 .00 ha 1 .00 tons/ha 79 .00 % 0.20 USD/kg Total 9729 .00 ha 1.25 tons/ha 79.50% ( NCDD , 2010) In these study communes, the average age of the population was about 30 years and 50 per cent of them were aged 25 years or less (Median=25). Moreover, in general people had very low education. For instance, in average people spent 4 years or fewer in scho ol. On the other hand, for the household sample, the average household size was about 5.5. Table 3 : Respondent profile N Min. Avg. Max. Median St.Dev. Age 545 1.00 29.88 86.00 25 .00 19.52 Education 545 0.00 4.36 15.00 4.00 3.59 HH size 99 1.00 5.56 11.00 5.00 2.11 Table 4 show s the main household occupations of the household sample. Given that a farm household may have more than one occupation, based on percentage of responses, the five most important household occupations include rice farming, crop growing, livestock, local wage labor, and migration. The other sources for livelihood include growing vegetable s , small business and salary. Table 4 : Main household occupations N Minimum Maximum Mean Std. Deviation Rice farming 99 0.00 1.00 0.99 0.10 Growing crop 99 0.00 1.00 0.60 0.49 Livestock 99 0.00 1.00 0.54 0.50 Local wage 99 0.00 1.00 0.47 0.50 Migration 99 0.00 1.00 0.36 0.48 Growing vegetable 99 0.00 1.00 0.29 0.45 Small trade 99 0.00 1.00 0.12 0.32 Salary 99 0.00 1.00 0.12 0.32 Valid N (listwise) 99 6 2. R eview of Literature Resilience has gained popularity among different communities of scholars and development communities; however, it has yet not received a broadly agreed upon definition. One reason is due to an attempt to broaden the resilience concept from its original narrow definition in Ecology into a more integrative one . The term is rooted in the discipline of Ecology (Holling, 1973) , expanded in the social science s (W.N. Adger, 2000) and further integrated in so cial - ecological system research es (Folke, 2006) . An other reason is because of different disciplinary focus es in research areas . For instance, t he early definition of resilience proposed by Holling ( 1973) places importance on the capacity of an ecosystem to maintain its stability and function in the fa ce of changes and disturbances, while from a social science perspective resilience definition is centered around maintain ing and improving livelihood s , while responding to social and environmental changes through appropriate institution s (W.N. Adger, Kelly, Winkels, H uy, & Locke, 2002) . On the other hand, from a social - ecological system perspective, resilience encompasses not only the amount of disturbances the system can tolerate, while retaining its structure and functioning, but also the capacity to self - organize , learn and adapt (Carpenter, Walker, Anderies, & Abel, 2001) . Different schools of thoug ht also define resilience in different ways. Because h azard research tends to focus more on the magnitude of a and degree of recovery , resilience rests on capacity of social and physical system to minimize disaster impacts , pre - and post - disaster measures, and how fast the system can recover from the hazard (Cutter et al., 2008) . On the other hand, i n the climate change community, r esilience is defined as the ability of a system and its component parts to anticipate, absorb, accommodate, or recover from the effects of a potentially hazardous event in a timely and efficient manner, including through ensuring the preservation, restoration, or improvement of its essential basic structure s and functions (Lavell et al., 2012) . Despite being broadly defined , t his definition 7 encompasses the capacit y of the system to moderate impacts of hazards, degree of recovery from those impacts, and capacity to maintain system stabilit y . For development communities, resilience not only includes similar characteristics as defined by hazard communities, but is also linked to vulnerability reduction , while promoting growth. For instance, in policy and program guidance for building resilience to recurrent cris e s, USAID ( 2012) defines resilience the ability of people , households, communities, cou n tries, and systems to mitigate, adapt to, and recover from shocks and stresses in a manner that reduces chronic vulnerability and facilitates inclusive growth (p. 9) . Despite inconsistencies in using key terms to characterize the attributes of resilience and some differences in disciplinary focus, there are commonalities that can be dr awn from. First, r esilience is characterized by three set s of capacities: absorptive capacity the capacity to moderate impacts of shocks and stresses so as to maintain system stability ; adaptive capacity involving incremental adjustment and social learning based on an understanding of changing conditions; and transformative capacity the capacity to make systemic change in a positive way when the old system is no longer v iable (B éné et al., 2012) . The combination of these three sets of response capacities marks the resilience concept as a paradigm shift f rom the traditional perspective , which believe s that changes in systems should be controlled, to a philosophy accepting that social - ecological systems are adaptive systems and humans have the capacity to learn from, live with , and adapt to uncertainties and unexpected changes . Second, resilience manifest s at multiple scales such as individual, household, community, and systems levels (Béné et al., 2012) and there are multiple sources of resilience at each of these levels (W. Neil Adger et al., 2011) . Furthermore, the responses to a particular risk may adversely impair the capac ity of the system to cope with other risks (Adger et al., 2011) . There are numerous discussions by different communities of scholars about two major approaches to resilience building: specified resilience and general resilience, and on which 8 one is more ap propriate for studies of resilience to hazards. T o understand system resilience in a practical manner as well as to identify measureable indicators of resilience, it is important to consider resilience as context - specific (Walker1a & Carpent er, 2002) . In doing so, it i s necessary to clearly determine which part of the system should be resilient and is resilient to what type(s) of disturbance because system resilience may manifest in one time period at the expense of resilience in the following period , and the resilience at a specific spatial scale may be inherited from a broader scale (Carpenter et al., 2001) . Identifying partic ular aspects of the system that should be resilient to certain kinds of disturbances can help us to discover system feedback loops whose processes can explain the pathways of system resilience in a practical manner (Bennett, Cumming, & Pet erson, 2005) . However, when focusing too much on a particular aspect of system resilience , there is a risk of Thus, to provide the best compromise between differences in defi ning, conceptualizing and operationalizing resilience, there are three factors to be taken into consideration for framing resi lience . First, resilience need s to have a general basic definition that can capture the robustness of this concept in a developmen t context am, & Davies, 2013) . Second, it is important to consider the type (slow or rapid onset) and characteristics (social or natural origin) of hazard s /disturbance s of interest when measuring and framing resilience . For example, the slow - onset disturbance such as drought has very different indicators for measurement from rapid - onset one. It also requires relatively different capac ities of response and allows longer recovery time . Plus, human communities have n o capa cities to halt or completely remove this kind of disturbance, but must learn to live with and to respond to it while continuing and improving their livelihood s . In contrast , disturbance s of social origin such as health shock or chronic poverty depend more on hu man ability to remove these negative impacts . People can eliminate or completely remove them given their own response 9 capacities and the context in which the community is located. Last but not least, it is necessary that there is at least a conceptual link between definition, conceptualization, and measurement of resilience . 3. Methods 3.1. Formulation of conceptual model for drought resilience This study adopts the definit ion of community resilience introduced in Frankenberger, Mueller, Spangler, & Alexander ( 2013) as capacity of a community to absorb change , seize opportunity to improve living standards, and to transform livelihood systems The community in this context refers to a commune which is the lowest level of governance in Cambodia and defined by political b oundary Béné, Wood, Newsham, & Davies ( 2012) as composing of three types of capacity: a bsorptive capacity the capacity to moderate immedi ate impacts of drought through preventative measures and short - term coping strategies; a daptive capacity making proactive and long - term decision s about alternative livelihood strategies through incremental adjustment and social learning based on an unde rstanding of changing conditions; and t ransformative capacity the capacity to make systemic change given the fact that the old one is no longer viable . Community resilience is not simply the sum of these three types of capacities but the outcomes of interplay between them through certain processes of responses . The conceptual framework for this study is based on the synthesis of t wo different approaches: li velihoods approach (Scoones, 1998) focusing on a ccess to and distribution of productive as sets within the community through i nstitution al structures and processes to pursue m ajor livelihood strategies; Disaster Risk R eduction approach (Cutter et al., 2008) placing the importance o n recovery activities and time to respond to drought impacts . For instance, d uring the drought period, the immediate impact s of drought are 10 moderated through absorptive capacity at househo ld and/or community level. If not complete ly absorbed , the remain ing impacts may be further reduced through the long - term adaptatio n mechanism (adaptive capacity). If the residu al impact still remain s beyond community capacity to survive , transformation of the livelihood system is needed. To know when transformation would take place , it is necessary to look at the degree of recovery (measured as a time scale) from the drought impacts . If the recovery time is too long (low degree of recovery) extending alm ost close to the next occurrence of hazard or to the extent that the of the system is required . The conceptual model presented in Figure 2 was modified from Béné et al. (2012), Cutter et al. (2008), Frankenberger et al. (2013) and Scoones, (1998) . Figure 2 : Conceptual model for drought resilience 3.2. Field d ata collection This study used both primary and secondary data . The secondary data such as migration rate, percentage of active labor, and cultivated area were extracted from commune database online ( NCDD , 2010) , while rainfall data was obtained from the provincial department of agriculture. The primary data were data from focus group d iscussion s and household survey s , conducted respectively in Februa ry and April 2015. 11 Two focus group discussion s were conducted in Kanteu Mouy and Chaeng Meanchey communes of Battambang province and in total there were 20 participants including the commune chief, village chief s and farmer representatives . The focus group discussion s collected data about available water sources, drought characteristics and drought impacts on The data from group discussions were used to modify the causal loop diagram developed for drought resilience as well as to design the household questionnaire. The household survey was conducted in the two communes and in total four villages in each commune were selected. T he surveys were administered to the household head or his/ her spouse, while the unit of analysis was the household. T he sample size was 99 and the population was 2998 households . The samp le was purposively selected based on characteristics of farmer s who had different livelihood options such as rice cultivation, crop cultivation , local wage, and migration, and who were affected differently by the 2014 drought. 12 3.3. Development of causal loop diagram and system dynamic model Figure 3 : Causal loop diagram depicting drought resilience Causal loop diagram ming is an important tool in system dynamics . It depicts a set of feedback loops that explain complex interactions between actors, action of responses or information . The research questions , thus, can be depicted through diagram above (Figure 3 ). This causal loop diagram was developed based on contextual understanding of the study area and the nature of the problem, information from focus group discussion, and survey data. There are three major feedback loop s in this diagram. In th e first feedback loop , four agricultural inputs (whether irrigating in drought year, whether using improved varieties, fertilizer used per hectare, ratio of agricultural labor to household size) positively affect rice yield. If any of these inputs are increased, r ice yield will go up, w hich then increase s the a nnual rice harvest. The more rice harvest , the more rice is 13 sold , which then increases household income . When household income increases , more fertilizer is used , which will then increase rice yield . This is a reinforcing feedback loop . However, in the second feedback loop, the increased use of agricultural inputs are associated with increased costs, for example, cost of irrigating, cost of buying seeds for those who use improved varieties, and cost of buying fertilizer. When the costs of agricultural inputs are increased, the household income is reduced, which then leads to less money available for investing in agricultural input use. This is the balancing feedback loop. In the third feedback loop, out - migrat ion brings back remittance s , adding to household income. When farmers have more income, they are more likely to invest in agricultural inputs, which then increase s rice yield. When rice yield increase s , farmers see a good oppor tunity to invest in agricultu re and they attract agricultural labor, which then reduce s number of out - migrants from the community. This makes a balancing feedback loop. A system dynamic s model was developed based on this causal loop diagram following the developed conceptual model/framework for drought resilience (Figure 2) to address the above research questions. The system dynamic model that was developed is described in A ppendices 3 and 4. System dynamic s model ing (SDM) is a methodology as well as a tool . The rationale for using this modelling approach for aiding decision making processes is three fold. First , the model makes it possible to identify the potential thresholds beyond which the current state of system in consideration tips into the new state which can change the function and structure of the system -- f or example, the point at which households transform their current livelihood or production system to a new one in response to a shock . Second, SDM can help figure out which underlying factors contribute to enhancing or eroding system resilience, from which we can provide insights into how adaptive and transformative capacity can be built to enhance the system resilience to unexpected shocks. For instance, the model allows us to test 14 which parameters/variables are very sensitive to change of the outcome variables being investigated. On top of that, the robustness of using SDM is that it creates new opportunities for understanding the degree of recovery and re - organization after disturbances to a system . For example, the model simulation enables comparison between different communities regarding how fast or slow they recover from a drought hazard. 3.4. Description of important model inputs /variables o D rought and normal year definition Table 5 : Descriptive statistics of rainfall N Valid 18 Missing 0 Mean 1277.49 Median 1296.10 Std. Deviation (SD) 203.21 Minimum 907.10 Maximum 1707.40 Percentiles 25 1079.32 75 1439.27 A drought year in this study is defined as the year that has rainfall value below (Mean - 1*SD) mm, which is equal to 1074.3 mm in this case, while a normal year , the non - drought year, is the year that has rainfall value higher than this. Mean annual rainfal l is the long - term average value. From this assumption, for the 17 years of rainfall data available the probability of drought occurrence is 0.28, which is very close to the overall probability of drought occurrence in Cambodia, as indicated by (Pandey, 2007) . That is to say it is likely that drought occurs every three year s . 2014 is the most severe drought year according to the rainfall data and data from survey, while 2013 is a normal year, the year when the commune s received enough rain for rice cultivation. 15 o Yield estimation for drought and normal year Because agricultural inputs respond differently to yield in normal and drought year s , two different yield - input relationships for normal and drought year were estimated. In this study, annual rice yield was estimated using multiple regression, where yield is the dependent variable and four agricultural inputs (agricultural labor, access t o irrigation, amount of fertilizer used, and varieties used) are independent variables. The details for estimation of multiple regressions are presented in Appendix 1. o Relationship between household income and total fertilizer used This relationship was e stimated using linear regression based on the assumption that the more household income the farm households earn , the more likely they invest in agriculture, i.e. in fertilizer. This is indicated in the causal loop diagram. The details for es timation of th is relationship are presented in Appendi x 2. o Other variables The other variables represented in the model and their relations/equation s are listed in Appendi x 3. 3.5. Data analysis Data analysis ha s two parts. The first one involved constructing relations between variables. Both qualitative and quantitative data were used in this study. Qualitative data included data from focus group discussion and in - depth interview s , while quantitative data were da ta from household survey and data from commune database online ( NCDD , 2010) for Cambodia. Data from focus group discussion s and in - depth interview s provided contextual understanding abo ut puts and drought responses, which is very useful for informing the casual loop diagram and model structure development. Data from in - depth interview and group discussion were also used to explain reasons behind certain ou tput s produced by model simulation. Survey data and data from commune database online 16 were very important sources for model input variables and estimation of some important relationships between variables. Survey data analysis was performed using descripti ve statistics for understanding major model input variables, multiple regression for estimating relationship between rice yield and usage of agricultural inputs, and cluster analysis for defining farm household typology on which the model simulation scenar io was run. Farm typology was defined based on the attributes of agricultural inputs a household possesses, using two - step clustering method, a built - in function in SPSS. This model (Figure 5) used Log - Bayesian Criterion (BIC) for clustering criterion. The input variables for the model were of categorical and continuous. Evaluation fields such as rice yield in normal and drought years and percentage of yield loss due to drought were also included in the model, but they were not used as clustering criteria. The rationale for developing the farm typology was to understand which group of farmers perform better or worse off during normal and drought years, and under what processes. This classification was us ed as criteria to determine input variables as well as to run the simulation. The second part involved analysis and interpretation of model simulation outputs by comparing different groups of farmers as defined in a typology classification run on certain a spects such as rice yield, net rice income, household income, yield recovery, dependence on farm and nonfarm income, and percentage of yield loss. At the commune level, actual harvested area and percentage of drought affected household s were investigated b ased on the simulation. Four main scenarios were also created and tested through model simulation. The first set of scenarios was at group level (farmer groups) and included : 1) increase of cultivated area and 2) increase of fertilizer used. These two adaptation strategies were the main responses of farmers as indicated in household survey results. The second group of 17 scenario s was at the commune level and included: 1) increase of percentage of access to ir rigation and 2) adjustment of migration rate around the current rate . The analysis rest s on the linkage between the farm typology analysis and model simulation outputs, which can be interpreted or explained through the conceptual model/framework for drough t resilience. Recalling the conceptual model for drought resilience, the actual impact s of drought received by a farmer, a group of farmers or a community depend on their capacity of responses , drought characteristics and existing conditions of the study c ommunes . The actual impacts are represented by outcome variables for resilience or vulnerability at household and group level s such as household income, net rice income, percentage of yield losses, dependence on farm and nonfarm income, and degree of yield recovery, and at the community level by community actual harvested area and percentage of drought - affected households. The values of the outcome variables were obtained from model outputs. On the other hand, the existing conditions/attributes of farm hous ehold and community were explained through the results of a typology model whose inputs include agricultural input use and farm characteristics. Understanding patterns of agricultural input use and their association with farm characteristics through this t ypology classification is a core for understanding coping, adaptation, and transformation processes of farmers in response to drought, which can be a guide for explanation of linkages between this typology analysis and model simulation outputs. 18 4. Results 4.1. Typology analysis Figure 4 : Clustering model for farm typology The following figure(Figure 5 ) shows detail ed information for each cluster. The input variables (predictors) in the list were arranged from top down by order of importance , and in this case access to irrigation in dro ught year and varieties used were the two most imp ortant factors that differen t iate group s of farmers . Other input variables such as those from non - farm dimension s were tested in the model, but they w ere not significant. Fro m this classification, there were three major group s of farmers , which can be summarized in the following table. Table 6 : Characteristics of farmer groups Farm Typology Characteristics Group 1 Least diversification in agriculture , medium - scale farmers without access to irrigation Group 2 Most diversification in agriculture , large - scale farmers with access to irrigation Group 3 Fair diversification in agriculture , medium - scale farmers without access to irrigation Group 1 had the lowest degree of diversification in agriculture , with small land size and withou t access to irrigation as their paddy fields were far from irrigable water sources. This group of farmers own ed just about one to two plots of p addy land (lea s t spatial diversification of paddy field s ) and use d only late - duration varieties which is the traditional variety that has low yield compared to the medim and shor t - duration varieties (least varietal diversification) . 19 Group 3 had very si mila r characteristics to Group 1 except that this gro up used two or more varieties (varieties diversification). Group 3 constitute d almost half of the total households in the community. Last, Group 2 was the most diversifed in agriculture . Farmers in Group 2 o wn ed two or more large paddy plots (spatial diversification of paddy lands) , use d two or more varieties (varietal diversification) and had access to irrigation as their paddy fields were very close to water sources. This group constitute s around one - third of the total households in the community. Figure 5 : Farm typology 20 4.2. Rice yield and uses of agricultural inputs Figure 6 indicates rice yield fluctuation over years from 2010 to 2050 generated from the model simulation. The oscillation resulted from changes in agricultural inputs over time, especially the changes in water availability for rice cultivation due to rainfall variability. The graph lines peak in normal years and hit their lowest value in dr ought years. In general, farmers in group 2 do better in drought year s , meaning their yield in drought year s is higher than the other two groups. This is because this group has access to and irrigates paddy field s when they face drought. However, in normal year s , farmers in group 1 and 3 perform better in terms of yield. This is probably because these two groups of farmers use a higher amount of fertilizer per hectare (Figure 7 ). From in - depth interview s , the reason that farmers in group number 2 use less a mount of fertilizer might be because they own a relatively large cultiva ted area, which requires a hectare in average due to financial constraint s . The other reason includes soil q uality. The more fertile the soil is, the less fertilizer they use. Figure 6 : Rice yield for farmers in group 1, 2 and 3 21 Figure 7 : Fertilizer s used per hectare 4.3. Agricultural investment and rice profit Figure 8 below depicts agricultural investment stock over time by farmer gro ups. It is noticed that investment in agriculture is compose d of costs of irrigation, seeds, labor, and fertilizers. From the graph it is clear that in general farmers in g roup 2 invest more in agriculture over the long - term , followed by farmers in group 3 and group 1. This is because farmers in Group 2 have relatively large area of paddy land, irrigate their rice fields if they face d rought, and buy rice seeds in average every two to three years . On the other hand, farmers in group 1 spend less overall even though they generally apply higher amount s of fertilizer per hectare. This is because they have relatively small land, do not irrigate or buy seed s as they use only lat e - duration varieties which are reserved from the previous cultivating season. Farmers in group 3 have intermediate investment rates . 22 Figure 8 : Agricultural investment by groups in USD Net rice income is equal to the gross rice income minus the production costs (cost of agricultural input) . As can be seen from Figure 0 9 , farmers in group 2 have higher net rice income over time, followed by farmers in group s 3 and 1. In addition, farmers in group 2 can maintain profit over time fr om rice sale s even in drought years. On the other hand, farm ers in group s 3 and 1 do not make any profit in drought years as net income goes down to zero during these years. There is also relation obse rved in Figure 08 and 0 9 between investment in agriculture and rice profit. All in all, it can be inferred from these two figures that , for farmers who have access to irrigation (group 2 in this case), the more the y invest in their rice production, the more profit they can make d espite the regular occurrence of drought over time . in the agricultural inputs put them at risk of making no profit in the drought years . 23 Figure 9 : Net rice income in USD Household income stock in Figure 11 was calculated by the sum of farm income and nonfarm income. Farm income includes rice income, income from other agricultural cultivation (vegetables and crops), while nonfarm income is the sum of local wage, remittance from migration, and other regular nonfarm income. It can be seen in Figure 10 that at the first 10 year of simulation the household income of the three groups of farmers does not differ much but it deviate s more after tha t period. Also, farmers in group 2 are still better off compared to the other two groups. F armers in group 1 have relatively higher household income compared to those in group 3, despite their lower rice income. This is because the y earn more from other no nfarm sources. 24 Figure 10 : Household income stock 4.4. Dependence on agriculture and other nonfarm activities Dependence on agricultural and non - agricultural income s in this case was calculated as the percentage of each individual income source contribution to the total household income. As shown in Figure 11 , farmers in group 2 depend more on rice income than the other two groups. Fo r this group, rice income share s almost 50 per cent of the total household income in normal years a nd approximately 40 per cent in drought years. Farmer s in group 1 depend less on rice income . Rice income for this group shares around 30 per cent of total income in normal years, and only 5 per cent in dr ought years. For farmers in group 3 , the share of r ice income in normal and drought years fluctuates around 20 to 40 per cent. 25 Figure 11 : Dependence on rice income Figure 1 2 shows that dependence of farmers on remittance from migration does not differ much among the three groups of farmers. The share of remittance from migration to the total household income for the three groups range s from 1 to 8 per percent and overal l , this dependence slightly increases over time . Figure 12 : Dependence on remittance from migration 26 Figure 13 illustrates the dependence of farmer s in group one on each income source. It can be seen from this figure that the farmers in this group depend a lot on nonfarm income which contribute s up to about 55 to 70 per cent of the total income in normal and drought years . The share of rice income to th e total income account s for around 10 to 25 per cent, while that of other agricultural income shares around 15 per cent only. It is to be noticed that this group of farmer s is the least diversified among the three and their household income stock accumulat ed over time is lower than the other two groups . Figure 13 : Dependence on different livelihood sources for farmer s in group 1 For farmers in group 2 as shown in Figure 14 , rice income and nonfarm income are equally important. Bot h sources of income contribute around 45 per cent of the total household income, while the share from other agricultural income account s for around 10 per cent. T his group of farmers is the most agricultural ly diversifed compared to the other two groups , invest s in agriculture more, and can still generate rice income during drought year s . 27 Figure 14 : Dependence on different livelih ood sources for farmer group two For farmers in group 3 (Figure 15 ) , the share s of rice income , other agrciltural income , and nonfarm income to total household income do not differ much and range from 25 to 35 per cent . Particularly, t his group of farmers weigh s the importance of other agricultural income higher than the other two groups. It seems that they do not place relative importance on any single source of income , all sources of income share the sam e weight. Figure 15 : Dependence on different livelih ood sources for farmer s in group 3 28 4.5. Recovery from losses due to drought Figure 16 illustrate s to what extent yearly simulated rice yield deviates from the normal rice yield for group 1 . In general, the simulated rice yield fluctuates around the normal yield over time. However, during normal years their rice yi eld goes just above the normal yield, while in drought years , their rice yield go es down far below the normal yield . This means that farmers in group 1 tend to lose more due to drought than they gain in yields o ver time . They are able to recover from yield loss in drought years , but can go just about the normal yield. Figure 16 : Simulated rice yield and normal rice yield for farmer s in group 1 For farmer s in group 2 , the simulated rice yield also fluctuates around the normal yield (Figure 17 ) . However, this group of farmers has more balance between normal and drought years in terms of yield. It seems that the magnitude of yield losses in drought years does not exceed what they gain in normal years in a long - term perspective . They can recover from drought and make more yield improvement beyond the normal yield. This would be because their normal yield is lower than that of the other group s. Despite this, based on previous results, this group of farmers still make s profit from r ice in drought years, while the other two do not. 29 Figure 17 : Simulated rice yield and normal rice yield for farmer s in group 2 Particularly, over time farmers in group 3 cannot reach the no rmal yield they first achieved under optimal conditions (Figure 18 ) . This might be because their normal yield is already high and even higher than the other two groups, which make s it difficult to maintain that yield level over time. However, this group perform s better than group 1 in drought years in terms of yield based on the above results. Figure 18 : Simulated rice yield and normal rice yield for farmer s in group 3 30 The percentage of yield losses due to drought in this study was calculated by the formula 100*[normal yield - simulated yield]/normal yield . Figure 19 below shows the percentage of yield loss due to drought over time by farmer groups. When the values go below zero, it indicates a yield gain, and the peaks indicate the percentage of yield loss in drought years. The result s show that in general all of the three groups lose yield during drought years , but group 2 lose s yield much less than the other two. Figure 19 : Percentage of yield loss by groups 4.6. Drought impacts at community level Figure 20 below shows the model simulation results at the commune level. As can be seen from this figure, the actual harvested area decreased and percentage of drought - affected households increased over time. This indicates that the commun ity does not have enough capacity to maintain the available cultivated land over time because the magnitude of drought impacts is larger than what the community can do to moderate the impacts, i.e. to irrigate their paddy lands. 31 Figure 20 : Community actual harvested area and percentage of affected households 4.7. Model Scenario Test s a) Cultivated Area and Fertilizer Increase Scenarios For scenario 1, a model test was performed to see how adjustment s of agricultural inputs at the household or farm group level have effects on percentage of yield loss and net rice income of farmer s in group 1 and gro up 3 and also to observe how these change s affect the community. This test is to show if the attributes of farmers in group 2 (cultivated area and fertilizer used), who perform well, can be a lesson for group s 1 and 3. First, cultivated area s of farmers in group 1 and 3 w ere increase d to 4.4 ha , similar to the cultivated area of farmer s in group 2 . T he amount of fertilizer per hectare was also increased proportionally, by approximately 20%. The results of these two simulations were compared with those of the initial condition s . Line N o 1 in Figure s 2 1 and 2 2 represent s the baseline condition for group 1 and 3 , r espectively. L ine N o 2 show s the simulation in which cultivated area was increased , and Line N o 3 the simulation in which both cultivated area and fertilizer were increased for farmer 32 group s 1 and 3 . As can be seen from these two figure s , there was almost no improvement for either group of farmers in terms of reducing percentage of yield loss due to drought. Figure 21 : P ercentage of yield losses for farmer group 1 under bas eline, area increase, and area & fertilizer increase scenarios. Figure 22 : Percentage of yield losses for farmer group 3 under base line, area increase, and area & fertilizer increase scenarios. Similarly , Figure s 23 and 24 show the effects of the same scenario test on net rice income for farmer group s 1 and 3. Through increasing cultivated area and amount of fertilizer used by 33 these two groups of farmer s , they could make more profit from rice cultivation in normal years only. In drought years, they woul d risk making no profit at all as net rice income goes to zero, but they make more investment s in this scenario because of increased land area . Figure 23 : Net rice income for farmer group 1 under bas eline, area increase, and area & fertilizer increase scenarios. Figure 24 : Net rice income for farmer group 3 under bas eline, area increase, and area & fertilizer increase scenarios. C hanges at the community level were also observed under these scenarios . Figure 25 and 26 indicate that increase in cultivated area and amount of fertilizer used for group s 1 and 3 have 34 a negative impact on the community, i.e. reducing actual harvested area and increasing number of drought - affected household s . This indicates that the characteristics that make farmers in group 2 better off do not necessarily make the other groups better off, and adjustment s made at the household/farm group level may have a negative impact on the community. These results are consistent with the above results in terms of magnitude of yield loss and net rice income. Previous results indicate that f armer groups 1 and 3 cannot make profit from rice cultivation in drought years due to high yield loss. So, if these groups of farmers invest in rice cul tivation by increasing amount of fertilizer or increasing both fertilizer amount and cultivated area as tested in these scenarios , they may loss more and more when facing drought. To larger extent, if drought frequencies increase, the magnitude of yield lo ss will be intensified too and as a result , negative impacts of these adjustments can be seen at commune level. In addition, there are also evidences from in - depth interview that support this finding. For instance, there were some farmers that increased ri ce cultivated land in 2014 by renting the l and from others and these farmers reported the huge losses due to severe drought in 2014. On the other hand, some other farmers mentioned that adding fertilizer just after drought period as a way to recover as wel l as to boost rice growth did not make rice recovered or boosted as they expected. In return, it burnt the rice and made the situation even worse. 35 Figure 25 : A ctual harvested area under bas eline, area increase, and area & fertilizer increa se scenarios for Groups 1 and 3 . Figure 26 : P ercentage of drought - affected households under bas eline, area increase, and area & fertilizer increase scenarios for Groups 1 and 3. 36 b) Increase access to irrigation a nd adjusted migration scenarios In scenario 2, tests were performed to see how adjustment s at the community level affect actual harvested area and percentage of households affected by drought. o Effect of access to irrigation Figure 27 below compares community actual harvested area for the current condition of access to irrigation and the condition in which more households would have access to irrigation. Line N o 1 in this figure indicates the community actual harvested area for the current condition i n which only 22% have access to irrigation. This line depicts the actual harvested area decreasing over time as drought continues to occur in the future, which indicates that the community is vulnerable to drought. L ine N o 2 depicts a scenario in which acc ess to irrigation in the community is increased to 40% . This signifies the threshold level of irrigation which can maintain stability of continuous rice cultivation over time. Thus, if access to irrigation is increased to around 50% or more, it is more likely that the community is better off in the face of drought. Figure 27 : Effect of increase access to irrigation on actual harvested area scenario 37 Similarly, in Figure 28, line N o 1 indicates the percentage of drought - affected household s for the current condition in which only 22% have access to irrigation. Line N o 1 depicts the percentage of drought - affected households in creasing over time as drought continues to occur in the future, which indicates the community is vulnerable to drought. If access to irrigation was increased to 40 per cent as indicated in line N o 2, the percentage of drought - affected household s increases much more slowly . Thus, if access to irrigation is increased to around 50% or more, it is more likely that the community is more resilient to drought. Figure 28 : Effect of increase access to irrigation on p ercentage of affected household scenario o Effects of migration For this scenario test, migration rate was adjusted around the current migration rate of 3% (line N o 1) by first increasing to 5% (line N o 2) and then decreasing to 1% (line N o 3) . The results show that the more migrations, the less actual harvested area (Figure 28) and the higher the percentage of drought - affected household s (Figure 29) . This shows the negative impacts of migration on resilienc e of rice production at community level. 38 Figure 29 : Effect of adjusted migration rate on actual harvested area scenario Figure 30 : Effect of adjusted migration rate on percentage of affected household scenario 39 5. Discussion The typology analysis depicts three major groups of farmers who differ from one another by whether they have access to irrigation and their associated degree of diversification in agriculture . The first group of farmers was the one that had the lowest deg ree of diversification and no access to irrigation because their paddy fields were located far away from irrigable water sources. They had only one or two paddy plots, used only one type of rice variet y (late - duration varieties), and owned relatively small cultivated area. This group constitutes 28% of the population. Looking at their agriculture and livelihood outcomes from model outputs, this group of farmers had the highest yield in normal years as they used more fertilizers than the other two, but in ge neral their total investment in inputs was lower than the other two groups, and as a result their net rice income was relatively low. In addition, they obtained the lowest yield in drought years, which makes them lose much more than the others when facing drought. After facing drought, they could recover their yield to just above their normal yield, but over time it seems they lose yield more frequently than gaining yield . T his group of farmers depends a lot on nonfarm income for their livelihood, which ac counted for around 60 - 75% of total income , while rice income and income from other agricultural crops were less important. Thus, it can be inferred from the results of t he t ypology analysis and model simulation that this group of farmers is on the vulnerab ility / resilience threshold. This group has room for improvement towards resilience by utilizing their resources from nonfarm activities. The second group of farmers was the most diversified in agriculture and had access to irrigation because their paddy fi elds were located next to or very close to irrigable water sources. They owned relatively large cultivated area and had both varietal diversification (using two or more varieties) and spatial diversification of paddy lands (having more than 40 two plots ) . Thi s group makes up around 27% of the population. Considering their agriculture and livelihood outcomes from model outputs, this group of farmers had the lowest yield in normal years as they use less fertilizer than the other two groups, and they own relatively large cultivating area , but they had the highest yield in drought year s , which made them lose much less than the first and the third groups in drought year s . Clearly, this group benefited from access to irrigation which allowed them to irrigate their paddy fields when facing drought. In general, their total investment in inputs was higher than the other two groups, and as a result their net rice income was relatively high and their accumulative household income was also higher. After facing droug ht, they could recover their yield to a level well above their normal yield and over time they were gaining more frequently than losing yield. On the other hand, this farmer group depended a lot on both rice income and nonfarm income for their livelihood, which equally made up around 40 - 50% of the total household income, while income from other agricultural crops w as much less significant. Thus, it can be inferred from the results of the typology analysis and model simulation that this group of farmers is r esilient to drought. The third group of farmers was the one that has the average degree of diversification in agriculture and had no access to irrigation because their paddy fields were located far away from irrigable water sources. They had only one paddy plot, owned relatively small cultivated area, but used two or more rice varieties. This group accounted for 45% of the population. Looking at their agriculture and livelihood outcomes from model outputs, this group of farmers had intermediate yield compar ed to the other two groups, and intermediate amount s of fertilizer used, total investment in inputs, and net rice income. However, in a long - term perspective, their accumulative household income was the lowest among all groups. After facing drought, they could not recover their yield to the level of their normal year and over time they were always losing yield. T his farmer group depended equally on nonfarm income, 41 rice income and income from other agricultural crops for their livelihood ; each of these sour ces accounted for around 20 - 40% of the total household income. Thus, it can be inferred from the results of the typology analysis and model simulation that this group of farmers is vulnerable to drought. In summary , the third group of farmers is vulnerable to droug ht, while the first group is on the vulnerability/ resilience threshold . These two groups constitute 73% of the total farm households. Only a small proportion of farmers (27%) is resilient to drought. At the community level of aggregation from the farm group level, model simulation outputs indicate that the community actual harvested area, the area remain ing after drought impacts, decreased over time and the percentage of household s affected by drought increas ed over time. Model results from both t he farm group level and the community level are consistent, showing that this community is not resilient to drought as the majority of farmers are vulnerable to drought or at the threshold of vulnerability. Thus, farmers from group s one and three can adap t resilience strategies from farmer s in group two. To demonstrate this, scenario tests were performed by modifying agricultural inputs at the group level to see if any changes in drought response would manifest for these two groups and for the whole community. For instance, cultivated area for farmers in group one and three was increased to 4.4 ha, the same as that of group two, the resilient group, and then their amount of fertilizer used per hectare was also in creased proportionally (20%). The results show that there was no improvement for either group 1 or group 3 in terms of yield loss reduction or and net rice income improvement . At the community level , these scenarios actually made the situation worse , decre asing actual harvested area and increasing percentage of drought - affected household s . This is because the current major adaptation strategies such as increasing fertilizer amount and cultivated area in drought year are not effective enough to cope with dro ught, but lead to a reverse outcome. These results are also confirmed with 42 answers from some farmers during in - depth interview. Those who cultivated rice on large area of land tend to loss more than those who cultivate rice on smaller area in drought year, Another set of scenario test s was also performed to see how changes at the community level can improve resilience. For instance, first, the percentage of households having access to irrigation was increase d from the current condition of 22.68% to 40% , and results show that the community actual harvested area and percentage of drought - affected households attained stability over time. This indicates that increasing access to irrigation to about 40% of househo lds can help maintain stability of rice production of the community in the face of drought and increasing to more than this would make the community more resilient to drought. Second, the migration rate was modified around its current rate of 3%, i.e. incr eased to 5%, and then decreased to 1%. The results indicate that out - migration had negative effects on rice production in a long - term perspective . It decreased community actual harvested area and increased number of drought - affected households. 43 6. Conclusio n From results and discussions, access to irrigation is the most important source for resilience at both the household and community level. Improving access to irrigation to the threshold level of approximately 40% or more can help maintain stability and continuous development of rice production over time . Another important source of drought resilience is agricultural diversification such as spatial diversification of paddy lan ds and varietal diversification. The group of farmers that is resilient to drought, i.e. those who can recover from yield loss due to drought, maintain profit from rice cultivation in both normal and drought year and have minimal yield losses due to drought, is associated with this characteristic as seen in the linkage between fa rm typology and model outputs. On the other hand, nonfarm diversification such as remittance from migration and local wage can be other sources of resilience to drought . However, it is to be noticed that t he resilient group is associated with an average de gree of dependence on both rice and nonfarm income , denoting that depending too much on nonfarm income might draw resources away from agriculture . 44 7. Recommendation s From the results of the study, the following recommendations were proposed: - Improving acces s to irrigation to approximately 40 % can help maintain stability and continuous development of rice production over time. However, more investigation need s to be ma de so as to identify which group should be provided irrigation access . Institutions are also required to manage the irrigation systems . - While nonfarm activities are important sources of resilience, i mproving local employment is a way to strengthen resilience to drou ght in a long - term perspective because it can minimize the possibility that farmers would move away from agriculture through other nonfarm activities such as out - migration . 45 8. Limitations of study For the results of the study, the following limitation s need to be considered: - The relation between yield and agricultural input use estimated through multiple regression is not strong, which may overestimate or underestimate the yield as simulated in the model . - The relationship between drought and migration was not well understood from the survey responses . Hence , drought - induced migration was not included as a dynamic process in the casual loop diagram . 46 APPENDICES 47 Appendi X 1: Multiple regression for yield in 201 4 (drought condition) and yield in 2013 (normal condition) Table 7 : Descriptive statistics for agricultural input variables Max Mean Min Standard Deviation Percentile 75 Column N % Yield in 2014 2400.00 968.93 89.29 588.08 1333.33 Yield in 2013 4000.00 1889.88 50.00 831.34 2500.00 Fertilizer per ha in 2014 166.67 74.95 .00 45.16 100.00 Fertilizer per ha in 2013 166.67 71.03 .00 42.47 100.00 Ratio of agri. labor to HH size 1.00 .63 .13 .22 .80 Whether irrigating in 2014 No 77.4% Yes 22.6% Whether using improved varieties 2014 No 14.9% Yes 85.1% Whether using improved varieties 2013 No 20.0% Yes 80.0% a) Regression model for y ield in 2014 and a gricultural inputs Table 8 : Summary b of regression model for yield 2014 and agricultural inputs Model R R Square Adjusted R Square Std. Error of the Estimate Durbin - Watson 1 .488 a .238 .204 510.38893 1.673 a. Predictors: (Constant), Whether irrigating in 2014, Ratio of agricultural labor to HH size, Whether using improved varieties 2014, Fertilizer per ha in 2014 b. Dependent Va riable: Yield in 2014 48 Table 9 : ANOVA a of regression model for yield 2014 and agricultural inputs Model Sum of Squares df Mean Square F Sig. 1 Regression 7169535.145 4 1792383.786 6.881 .000 b Residual 22923723.894 88 260496.862 Total 30093259.039 92 a. Dependent Variable: Yield in 2014 b. Predictors: (Constant), Whether irrigating in 2014, Ratio of agricultural labor to HH size, Whether using improved varieties 2014, Fertilizer per ha in 2014 Table 10 : Coefficients a of regression model for yield 2014 and agricultural inputs Model Unstandardized Coefficients Standardized Coefficients t Sig. Correlations B Std. Error Beta Zero - order Partial Part (Constant) 137.899 223.219 .618 .538 Fertilizer per ha in 2014 3.791 1.209 .301 3.136 .002 .355 .317 .292 Ratio of agricultural labor to HH size 342.466 237.436 .135 1.442 .153 .146 .152 .134 Whether using improved varieties 2014 292.693 150.037 .184 1.951 .054 .215 .204 .182 Whether irrigating in 2014 298.679 132.479 .220 2.255 .027 .329 .234 .210 b) Regression for y ield in 2013 and a gricultural inputs Table 11 : Summary b of regression model for yield 2013 and agricultural inputs Model R R Square Adjusted R Square Std. Error of the Estimate Durbin - Watson 1 .341 a .116 .064 774.08397 2.137 a. Predictors: (Constant), Whether irrigating in 201 4 , Ratio of agricultural labor to HH size, Whether using improved varieties 2013, Fertilizer per ha in 2013 b. Dependent Va riable: Yield in 2013 49 Table 12 : ANOVA a of regression model for yield 2013 and agricultural inputs Model Sum of Squares df Mean Square F Sig. 1 Regression 5282530.631 4 1320632.658 2.204 .078 b Residual 40146801.408 67 599205.991 Total 45429332.039 71 a. Dependent Variable: Yield in 2013 b. Predictors: (Constant), Wh ether using improved varieties 2013, Whether irrigating 2014, Ratio of agricultural labor to HH size, Fertilizer per ha in 2013 Table 13 : Coefficients a of regression model for yield 2013 and agricultural inputs Model Unstandardized Coefficients Standardized Coefficients t Sig. Correlations B Std. Error Beta Zero - order Partial Part (Constant) 1328.995 365.586 3.635 .001 Fertilizer per ha in 2013 4.822 2.249 .259 2.144 .036 .193 .253 .246 Whether irrigating in 2014 - 533.545 231.718 - .279 - 2.303 .024 - .183 - .271 - .264 Ratio agricultural labor to HH size 442.605 397.981 .131 1.112 .270 .117 .135 .128 Whether using improved varieties 2014 11.781 247.810 .006 .048 .962 - .004 .006 .005 a. Dependent Variable: Yield in 2013 50 Appendi X 2 : Regression model for household income and fertilizer used Table 14 : Descriptive statistics for household income and total fertilizer in 2014 N Minimum Maximum Mean Std. Deviation HH income 94 125.00 16631.00 2649.9176 3104.03410 Total fertilizer in 2014 99 .00 1799.00 311.4343 277.64252 Valid N (listwise) 94 Table 15 : Summary of regression model for household income and fertilizer used R R Square Adjusted R Square Std. Error of the Estimate .511 .261 .252 168.116 The independent variable is HH income. Table 16 : ANOVA of regression model for household income and fertilizer used Sum of Squares df Mean Square F Sig. Regression 817867.834 1 817867.834 28.938 .000 Residual 2317555.976 82 28262.878 Total 3135423.810 83 The independent variable is HH income. Table 17 : Coefficient of r egression model for household income and fertilizer used Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta HH income .060 .011 .511 5.379 .000 (Constant) 159.951 28.310 5.650 .000 51 Appendi X 3: Model e quations Rice yield if Whether_drought_year=1 then (316.045+5.103*Amount_of_fertilizer_used_Kg_per_Ha+412.571*Whether_irrigating_in_dr ought_year+150.715*Whether_use_improved_varieties+120.952*Ratio_of_agricultural_labo rs_to_HH_size) else (1328.995+4.822 *Amount_of_fertilizer_used_Kg_per_Ha - 533.545*Whether_irrigating_in_drought_year+11.781*Whether_use_improved_varieties+44 2.605*Ratio_of_agricultural_labors_to_HH_size) Whether_drought_year if Time_series__of_rainfall<1074.3 then 1 else 0 Amount_of_fertilizer_used_Kg_per_Ha Total_fertilizer__used/Cultivated__area Whether_irrigating_in_drought_year if Whether_drought_year=1 then 1 else 0 Whether_use_improved_varieties [1=yes; 0=no] Ratio_of_agricultural_labors_to_HH_size (Number_of__ agricultural_labor/Number__of_HHs)/HH_size Rice harvest Rice_yield*Cultivated__area Rice consumption HH_size*Rice_consumption_per_capita/Rice_to_rice_milled__conversion_factor Rice to rice milled conversion factor: 0.64 Rice consumption per capita : 1 50 [kg] HH size: 5.5 52 Rice sale Rice_stock - Rice__consumption - Rice_seeds Rice seeds if Whether_use_improved_varieties=0 then Rice_seeds_used__kg_per_Ha*Cultivated__area else 0 Investment in agriculture Total_production_cost_per_ha*Cultivated__area Rice income Rice_sale*Price__USD_per_kg Price USD per kg : 900/4000 [USD] Investment in agriculture Total_production_cost_per_ha*Cultivated__area Cultivated area: (4.59, 3.48, 3.88) [m 2 ] Total production cost per ha (Cost_of_labor_per_ha+Cost_of_fert ilizer_per_Ha+Cost_of_seeds_per_Ha+Cost_of_traction _means_per_ha+Cost_of_irrigation_per_ha) Cost of fertilizer per ha Amount_of_fertilizer_used_Kg_per_Ha*Cost_of_fertilizer_per_Kg Cost of fertilizer per kg : 0.75 [USD] Cost of seeds per ha if Whether_use_improved_varieties=0 then 0 else Cost_of_seeds_per_Kg*Amount_of_seeds__Kg_per_Ha/3 Amount of seeds per ha: 100 [kg] Cost of seeds per kg: 0.375 [USD] Cost of irrigation per ha if Whether_drought_year=0 then 0 else if Whether_irrigating _in_drought_year=0 then 0 else 30 Cost of labor per ha : 288000/4000 [USD] 53 Cost of traction means per ha : 50 [USD] HH income Farm_income + Nonfarm_income Farm income Other_agricultural_income + Rice_income Other agricultural income : (1109.43, 395.05, 362.17) [USD] Nonfarm income Other_nonfarm__income+Local_wage+Remittance__from_migration Other nonfarm income: (1121.17, 688.62, 297.29) [USD] Local wage: (218.7, 119.71, 232.71) [USD] Remittance from migration: Remittance_per_capita* Avg_number_of_migrants_per_HH Avg number of migrants per HH Out_migrant_stock/Number__of_HHs Remittance per capita Community_remittance/Out_migrants Community remittance: 23933 [USD] Out migrants stock: 216 Out migrants Migration_rate *Number_of__agricultural_labor Migration rate: 1.30/100 Number of agricultural labor : 6984 Number of HHs Population/HH_size Population: 14349 5 4 People becoming active Population*Rate_of_becoming__active_labor Rate of becoming active RANDOM( - 3.69 /100,3.53/100) Migration: Out migrants Death Population*Death_rate Death rate: 7.78/1000 Birth Population*Birth_rate Birth rate: 30/1000 Community drought affected area (1/100)*Community_actual_harvested__area*(Percent_loss_per_year[Group_1]*30.7/100+P ercent_loss_per_year[Group_2]*28/100+Percent_loss_per_year[Group_3]*41.3/100) Community actual harvested area: 4667+5062 [m 2 ] Percent loss per year: 100*(Normal_yield - R ice_yield)/Normal_yield Normal yield: (2045.39, 2135.22, 1895.45) Community irrigated area Community_actual_harvested__area*Percentage_of_HHs_irrigating_when_drought Percentage of HHs irrigating when drought: if Whether_drought_year=1 then 22.68 /100 else 0 Number of HHs affected by drought Community_drought__affected_area/(Cultivated__area[Group_1]*30.7/100+Cultivated__are a[Group_2]*28/100+Cultivated__area[Group_3]*41.3/100) 55 Number of HHs irrigating when drought Community_irrigated_area/(Cultivated__area[Group_1]*30.7/100+Cultivated__area[Group_ 2]*28/100+Cultivated__area[Group_3]*41.3/100) Percentage of affected HHs if 100 - 100*Total_number_of_none_affected_HHs/Number__of_HHs<0 then 0 else 100 - 100*Total_number_of _none_affected_HHs/Number__of_HHs Total number of none affected HHs: 2998 Time series of rainfall If Time <2015 then Historical__rainfall else Projected__rainfall Projected rainfall RANDOM(907.1,1707.4,0.1) Historical rainfall TIME 56 Appendix 4 Figure 31 : Model structure 57 BIBLIOGRAPHY 58 BIBLIOGRAPHY ADB. (2014). Key Indicators for Asia and the Pacific 2014. Retrieved from http://www.adb.org/sites/default/files/publication/43030/cam.pdf Adger, W. N. (2000). Social and ecological resilience: are they related? Progress in Human Geography , 24 (3), 347 364. doi:10.1191/030913200701540465 (2011). Resilience implications of policy responses to climate change. Wiley Interdisciplinary Reviews: Climate Change , 2 (5), 757 766. doi:10.1002/wcc.133 Adger, W. N., Kelly, P. M., Winkels, A., Huy, L. Q., & Locke, C. (2002). Migration, Remittances, Livelihood Trajectories, and Social Resilience. AMBIO: A Journal of the Human Environment , 31 (4), 358. doi:10.1639/0044 - 7447(2002)031[0358:MRLTAS]2.0.CO;2 ewsham, A., & Davies, M. (2013, March 1). Making the most of resilience. IDS In Focus Policy Briefings . Institute of development studies. Retrieved from http://eprints.soas.ac.uk/18033/1/IF32.pdf Béné, C., Wood, R. G., Newsham, A., & Davies, M. (2012). Res ilience: New Utopia or New Tyranny? Reflection about the Potentials and Limits of the Concept of Resilience in Relation to Vulnerability Reduction Programmes. IDS Working Papers , 2012 (405), 1 61. doi:10.1111/j.2040 - 0209.2012.00405.x Bennett, E. M., Cumming , G. S., & Peterson, G. D. (2005). A Systems Model Approach to Determining Resilience Surrogates for Case Studies. Ecosystems , 8 (8), 945 957. doi:10.1007/s10021 - 005 - 0141 - 3 Berkes, F. (2007). Understanding uncertainty and reducing vulnerability: lessons fro m resilience thinking. Natural Hazards , 41 (2), 283 295. doi:10.1007/s11069 - 006 - 9036 - 7 Carpenter, S., Walker, B., Anderies, J. M., & Abel, N. (2001). From Metaphor to Measurement: Resilience of What to What? Ecosystems , 4 (8), 765 781. doi:10.1007/s10021 - 001 - 0045 - 9 Cutter, S. L., Barnes, L., Berry, M., Burton, C., Evans, E., Tate, E., & Webb, J. (2008). A place - based model for understanding community resilience to natural disasters. Global Environmental Change , 18 (4), 598 606. doi:10.1016/j.gloenvcha.2008.07. 013 EM - DAT. (2014). EM - DAT: The OFDA/CRED International Disaster Database. Brussels: Université catholique de Louvain. Retrieved from http://www.emdat.be/result - country - profile Folke, C. (2006). Resilience: The emergence of a perspective for social ecologi cal systems analyses. Global Environmental Change , 16 (3), 253 267. doi:10.1016/j.gloenvcha.2006.04.002 59 Frankenberger, T., Mueller, M., Spangler, T., & Alexander, S. (2013). Community Resilience: Conceptual Framework and Measurement Feed the Future Learning Agenda. Rockville, Md.: Westat . Holling, C. S. (1973). RESILIENCE AND STABILITY OF ECOLOGICAL SYSTEMS. Annual Review of Ecology and Systematics , 4 , 1 23. IPCC, 2012: Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation . A Special Report of Working Groups I and II of the Intergovernmental Panel on Climate Change [Field, C.B., V. Barros, T.F. Stocker, D. Qin, D.J. Dokken, K.L. Ebi, M.D. Mastrandrea, K.J. Mach, G. - K. Plattner, S.K. Allen, M. Tignor, and P.M. Midgley (eds.)]. Cambridge University Press, Cambridge, UK, and New York, NY, USA, 582 pp. Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation. In Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation (pp. 25 64). Cambridge: Cambridge University Press. doi:10.1017/CBO9781139177245 Mcsweeney, C., New, M., & Lizcano, G. (2008). UNDP Climate Change Country Profiles C ambodia . National Committee for Sub - National Democratic Development. (2010). Commune Database Online. Retrieved December 29, 2014, from http://db.ncdd.gov.kh/cdbonline/home/index.castle mechanisms. International Rice Research Notes . Retrieved from http://agris.fao.org/agris - search/search.do?recordID=qr2007000068 Ros, B., Nang, P., & Chhim, C. (2011). Case of Cambodia . Scoones, I. (1998). Sustainable Rural Livelihoods: A Framework for Analysis. IDS. Thomas, T. S., Ponlok, T., Bansok, R., Lopez, T. De, Chiang, C., Phirun, N., & Chhun, C. (2013). Cambodian Agriculture: Adaptation to Climate Change Impact . Retrieved from http://www.ifpri.org/p ublication/cambodian - agriculture USAID. (2012). Building Resilience to Recurrent Crisis: USAID POLICY AND PROGRAM GUIDANCE . Retrieved from http://www.usaid.gov/sites/default/files/documents/1866/Policy %26 Program Guidance - Building Resilience to Recurren t Crisis_Dec 2012.pdf USDA - FAS. (2010). CAMBODIA: Future Growth Rate of Rice Production Uncertain. Retrieved January 2, 2015, from http://www.pecad.fas.usda.gov/highlights/2010/01/cambodia/ 60 Walker1a, B., & Carpenter, S. (2002). Resilience management in soc ial - ecological systems: a working hypothesis for a participatory approach. Conservation . Retrieved from http://ovc.uoguelph.ca/personal/ecosys/ecocourse/Walkeretal.pdf Wilhite, D. A., Svoboda, M. D., & Hayes, M. J. (2007). Understanding the complex impa cts of drought: A key to enhancing drought mitigation and preparedness. Water Resources Management , 21 (5), 763 774. doi:10.1007/s11269 - 006 - 9076 - 5 Yu, B., & Fan, S. (2009). Rice Production Responses in Cambodia . Yusuf, A. A., & Francisco, H. (2009). Climate Change Vulnerability Mapping for Southeast Asia .