ABSTRACT Title of Dissertation: UNDERSTANDING GEOSPATIAL DYNAMICS OF PARASITE MIGRATION AND HUMAN MOBILITY AS FACTORS CONTRIBUTING TO MALARIA TRANSMISSION IN THE GREATER MEKONG SUBREGION Yao Li, Doctor of Philosophy, 2021 Dissertation directed by: Professor Kathleen Stewart, Department of Geographical Sciences. Much effort has been made to control malaria over the past decades in South-East Asia Confirmed cases of Plasmodium falciparum (P.f.) and Plasmodium vivax (P.v.) malaria were reduced by 46%, and mortality by 60%. However, malaria remains a major problem in the Greater Mekong Subregion (GMS) with the emerging resistance to the artemisinins and their partner drugs. This raises concerns that the usefulness of first- line malaria treatments may be diminishing in the GMS, and that drug resistance could spread worldwide. Estimating malaria parasite migration patterns is crucial for malaria elimination as well as understanding the role that human mobility plays in malaria transmission. This dissertation will focus on the GMS, especially Cambodia and Myanmar which have been widely regarded as the epicenter of emerging resistance to artemisinin-based combination therapies. This dissertation is structured as three separate studies that look first at the movement of malaria parasites across a region, and then two studies that focus on human movement and how these movements can lead to increased exposure as well as transmission of malaria. In the first study, a semi- automatic workflow was developed to select the optimal number of demes that will maximize model accuracy and minimize computing time when computing estimated effective migration surfaces. A validation analysis showed that the optimized grids displayed both high model accuracy and reduced processing time compared to grid densities selected in an unguided manner. In the second study, an agent-based simulation model was built to estimate and simulate the daily movements of local populations in Singu and Ann Townships in Myanmar in order to identify how two townships in different parts of Myanmar compared with respect to mobility and P.v. and P.f. positivity. The third study examined mobility patterns of local village populations in Singu Township, Myanmar when they traveled longer distances outside of Singu, and discuss these patterns of regional travel in the context of daily mobility within the township. UNDERSTANDING GEOSPATIAL DYNAMICS OF PARASITE MIGRATION AND HUMAN MOBILITY AS FACTORS CONTRIBUTING TO MALARIA TRANSMISSION IN THE GREATER MEKONG SUBREGION by Yao Li Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park, in partial fulfillment of the requirements for the degree of Doctor of Philosophy 2021 Advisory Committee: Professor Kathleen Stewart, Chair Professor Tatiana Loboda Professor Laixiang Sun Professor Shannon Takala Harrison Professor Robin Puett, Dean?s Representative ? Copyright by Yao Li 2021 Dedication To my grandparents and my parents For your love, support and treasured memories with you ii Acknowledgements The Ph.D. journey was, ultimately, a lonely one. Fortunately, I seldom felt alone in this journey. The professors, classmates, colleagues, and friends I met during this journey provided me with a lot of help and support, making this dissertation possible. First, I?d like to thank my advisor Professor Kathleen Stewart. Her insight and attitude toward science were inspiring me every time. I was lucky to work with her and put my efforts into some exciting projects at the beginning of my Ph.D. career, and since then, I have received enormous help and advice from her. I cannot remember how many times she dragged me out from my ?risky? research ideas and patiently helped me with my confusion. Then, I?d like to thank my committee member, Professor Tatiana Loboda, for providing crucial suggestions on chapter 3 and 4. To Professor Laixiang Sun, for the methodology support for chapter 2. To Professor Shannon Takala Harrison for her invaluable advice on relating my research to the field of global health. And final to Professor Robin Puett for being my dean?s representative. Thanks to all the CGISers in Lefrak 1124, Riccardo Fellegara, Federico Iuricich, Xue Yang, Hai Lan, Xin Xu, Yunting Song, Zhiyue Xia, Guiming, Zhu, and Jeffery Sauer, for your assistance on my research and for creating a great environment in our center. Thanks to my friends Yanjia Cao, Cheng Fu, Junchuan Fan, Xiaoxue Zhou, Aolin Jia, Yuhan Rao, Moving Li, Liangping Sun, and Weiming Chen for helping me in many iii different ways. Special thanks to Haoran Xu and Meihui Wang for being helpful and enjoyable friends. Special thanks to Rongsheng Jiang and Lei Sun for being such exciting and inspiring friends. Special thanks to Zheng Liu , Lei Ma and Jiaming Lu for your help and encouragement to me. Finally, I?d like to thank my parents, Shuxin Li (???) and Huifang Yao (?? ?), my brother, and my grandparents for your unconditional love and care for me. iv Table of Contents Dedication ..................................................................................................................... ii Acknowledgements ...................................................................................................... iii Table of Contents .......................................................................................................... v List of Tables .............................................................................................................. vii List of Figures ............................................................................................................ viii List of Abbreviations .................................................................................................... x Chapter 1: Introduction ................................................................................................. 1 1.1 Background and motivation ................................................................................ 1 1.2 Research objectives ........................................................................................... 10 1.3 Dissertation outline ........................................................................................... 13 Chapter 2: Detecting geospatial patterns of Plasmodium falciparum parasite migration in Cambodia using optimized estimated effective migration surfaces ....................... 16 2.1 Abstract ............................................................................................................. 16 2.2 Introduction ....................................................................................................... 17 2.3 Materials and Methods ...................................................................................... 20 2.3.1 Study area and data collection ................................................................... 20 2.3.2 Computing estimated effective migration surfaces .................................... 21 2.3.3 Clustering based on the distribution of P.f. genomic data ......................... 22 2.3.4 Computing the length of triangle edges and determining the number of demes .................................................................................................................. 24 2.3.5 Evaluation of model accuracy .................................................................... 25 2.4 Results ............................................................................................................... 25 2.4.1 Generating an optimized grid for the study area ........................................ 26 2.4.2 Evaluating the optimized grid .................................................................... 27 2.4.3 Effective migration surfaces using the optimized grid .............................. 30 2.4.4 Comparison between estimated effective migration surfaces and P.f. endemicity and annual parasite incidence in Cambodia ..................................... 31 2.5 Discussion ......................................................................................................... 33 2.6 Conclusions ....................................................................................................... 36 Chapter 3: Understanding spatio-temporal human mobility patterns for malaria control using a multi-agent mobility simulation model .......................................................... 37 3.1 Abstract ............................................................................................................. 37 3.2 Introduction ....................................................................................................... 38 3.3 Materials and Methods ...................................................................................... 42 3.3.1 Study area and data collection ................................................................... 42 3.3.2 Malaria prevalence data ............................................................................. 46 3.3.3 Road networks and environmental data ..................................................... 47 3.3.4 Determining origin and destination locations and travel modes for agents 48 3.3.5 Simulating mobility of local populations ................................................... 51 3.4 Results ............................................................................................................... 54 3.4.1 Simulating mobility patterns using an agent-based model ........................ 54 3.4.2 Simulating mobility patterns using an agent-based model ........................ 55 3.4.3 Road network usage for different occupation groups in two Townships .. 57 v 3.4.4 Comparing the most visited network segments for P.f.and P.v. positive individuals and different occupation groups in the two townships..................... 60 3.4.5 Validating the mobility simulation ............................................................ 65 3.5 Discussion ......................................................................................................... 66 3.6 Conclusions ....................................................................................................... 69 Chapter 4: Analyzing multi-scale mobility to assess malaria exposure and transmission risk in central Myanmar .............................................................................................. 70 4.1 Abstract ............................................................................................................. 70 4.2 Introduction ....................................................................................................... 71 4.3 Materials and Methods ...................................................................................... 74 4.3.1 Study area and travel history data collection ................................................. 74 4.3.2 Road networks and environmental variables ............................................. 76 4.3.3 Calculating regional travel frequency using degree centrality .................. 78 4.3.4 Potential malaria transmission due to regional travel for villages in Singu ............................................................................................................................. 80 4.3.5 Modeling daily mobility patterns to understand the dispersal of imported malaria risk due to regional travel ...................................................................... 81 4.3.6 Ecological niche modeling for estimating P.v. and P.f. malaria prevalence in Singu ............................................................................................................... 82 4.3.7 Computing malaria transmission hotspots in Singu................................... 84 4.4 Results ............................................................................................................... 85 4.4.1 Regional mobility patterns in Singu .......................................................... 85 4.4.2 Exposure risk for regional travel destinations ........................................... 86 4.4.3 Understanding mobility and exposure risks for regional travel destinations ............................................................................................................................. 88 4.4.4 High frequency travel ................................................................................ 88 4.4.5 Regional travel for P.v. and P.f. positive cases .......................................... 89 4.4.6 Possible P.v. and P.f. malaria transmission due to regional travel in Singu ............................................................................................................................. 91 4.4.7 Computing exposure risk for P.v. and P.f malaria within Singu ............... 93 4.4.8 Determining malaria transmission hotspots ............................................... 95 4.5 Discussion ......................................................................................................... 97 4.6 Conclusion ...................................................................................................... 100 Chapter 5: Conclusions ............................................................................................ 102 5.1 Summary of major findings ............................................................................ 102 5.2 Future work ..................................................................................................... 107 Appendices ................................................................................................................ 110 Supplementary materials for Chapter 2: ............................................................... 110 Supplementary materials for Chapter 3: ............................................................... 116 Bibliography ............................................................................................................. 125 vi List of Tables Table 3. 1 Outdoor occupation and travel information for Ann Township and Singu Township 45 Table 3. 2 Datasets used to estimate locations for each occupation 49 Table 4. 1 Regional travel from Singu Township ....................................................... 76 Table 4. 2 Converting Frequency to Fre value ........................................................... 79 Table 4. 3 Selected environmental variables for computing ecological niche modeling ..................................................................................................................................... 83 Table 4. 4 Regional travel and exposure risks for regional travel destinations .......... 88 Table 4. 5 Regional travel for P.v. and P.f. positive cases in Singu ........................... 90 vii List of Figures Figure 1. 1 Categorization of countries as malaria-free, eliminating malaria, or controlling malaria as of 2015 (Lim et al. 2016) .......................................................... 1 Figure 1. 2 Research objectives of the dissertation ..................................................... 11 Figure 2. 1 Sampling locations in Cambodia, Thailand and Vietnam ........................ 21 Figure 2. 2 (a) Decision graphs generated from clustering by fast search and find of density peaks(Rodriguez and Laio 2014) and (b) multidimensional scaling graph generated from extracting density peaks ..................................................................... 23 Figure 2. 3 Density-based clustering of genomic data where (a) five genomic clusters that were detected using the decision graph corresponded geographically to (b) six geographic locations (five locations in Cambodia and one in Vietnam). ................... 26 Figure 2. 4 Scatter plot of EEMS model accuracy for between-demes using the optimized grid. ............................................................................................................ 27 Figure 2. 5 Model performance for 200 to 700 demes for (a) model accuracy (R2 value for migration model accuracy) and (b) computation time. ................................ 29 Figure 2. 6 Estimated migration surface of P. falciparum parasites in Cambodia using the optimized grid. ...................................................................................................... 31 Figure 2. 7 P. falciparum endemicity patterns within Cambodia from the Malaria Atlas Project (Data downloaded from https://map.ox.ac.uk/) (2010) ......................... 32 Figure 3. 1 Study villages and road networks for (a) Ann Township and (b) Singu Township..................................................................................................................... 43 Figure 3. 2 Land cover in Ann and Singu Townships ................................................ 48 Figure 3. 3 Determining the possible working areas (a) and candidate working destinations (b) using spatial mean and its ten nearest points .................................... 50 Figure 3. 4 Hexagon x for aggregating the simulation results .................................... 53 Figure 3. 5 Simulated mobility patterns during morning travel in Singu Township for (a) farmers and (b) loggers .......................................................................................... 55 Figure 3. 6 Simulated travel patterns for (a) motorcyclists and (b) walkers in Ann Township and for (c) motorcyclists and (d) walkers in Singu Township ................... 56 Figure 3. 7 Travel patterns in Ann Township for (a) forest workers and (b) farmers. 58 Figure 3. 8 Travel patterns in Singu Township for (a) farmers, (b) miners and (c) forest workers. ............................................................................................................. 59 Figure 3. 9 Most visited network segments (MVS) in Ann Township for (a) P.v. positives, (b) P.f. positives overlapped with the MVS for forest workers and farmers ..................................................................................................................................... 61 Figure 3. 10 Most visited network segments (MVS) in Singu Township that overlapped with the MVS for forest workers, miners and farmers for (a) P.v. positives, (b) P.f. positives .......................................................................................... 62 Figure 3. 11 Most visited network segments (MVS) in Ann Township that overlapped with the MVS for forest workers and farmers for (a) P.v. positives, (b) P.f. positives ..................................................................................................................................... 64 viii Figure 3. 12 Most visited network segments (MVS) in Singu Township for (a) P.v. positives, (b) P.f. positives overlapped with the MVS for forest workers, miners and farmers ........................................................................................................................ 65 Figure 3. 13 Aggregated travel frequency and selected sites for mobility simulation validation..................................................................................................................... 66 Figure 4. 1 Township destinations for regional travel in Singu .................................. 75 Figure 4. 2 Hexagon x for aggregating the simulation results .................................... 82 Figure 4. 3 Regional travel frequency for villages in Singu ....................................... 86 Figure 4. 4 P.v. and P.f. exposure risk for townships in Myanmar ............................. 87 Figure 4. 5 Regional travel patterns for (a) P.v. positive and (b) P.f. positive individuals in Singu Township ................................................................................... 91 Figure 4. 6 MT for (a) P.v. and (b) P.f. malaria in Singu ........................................... 92 Figure 4. 7 Estimate exposure risk areas for (a) P.v. and (b) P.f. malaria in Singu Township..................................................................................................................... 93 Figure 4. 8 Jackknife test results for (a) P.v. and (b) P.f. malaria exposure risk estimates. ..................................................................................................................... 95 Figure 4. 9 Local travel patterns for (a) those who travelled outside Singu Township and the travelers who went to (b) high P.v. and (c) P.f. risk townships. .................... 96 Figure 4. 10 Potential Malaria transmission hotspots resulting from both local and regional travel in Singu. .............................................................................................. 97 Supplementary Figure 1 Environmental features including protected areas, national parks, wildlife sanctuaries, and locations of waterbodies in Cambodia (2013). ....... 110 Supplementary Figure 2 Topological skeleton and boundary for cluster red ........... 111 Supplementary Figure 3 Topological skeleton and boundary for cluster green ....... 112 Supplementary Figure 4 Topological skeleton and boundary for cluster blue ......... 113 Supplementary Figure 5 Topological skeleton and boundary for cluster cyan ........ 114 Supplementary Figure 6 Topological skeleton and boundary for cluster purple ...... 115 Supplementary Figure 7 Travel frequency hotspots for forest workers in Ann Township................................................................................................................... 116 Supplementary Figure 8 Travel frequency hotspots for farmers in Ann Township . 117 Supplementary Figure 9 Travel frequency hotspots for P.v. positive people in Ann Township................................................................................................................... 118 Supplementary Figure 10 Travel frequency hotspots for P.f. positive people in Ann Township................................................................................................................... 119 Supplementary Figure 11 Travel frequency hotspots for forest workers in Singu Township................................................................................................................... 120 Supplementary Figure 12 Travel frequency hotspots for farmers in Singu Township ................................................................................................................................... 121 Supplementary Figure 13 Travel frequency hotspots for miners in Singu Township ................................................................................................................................... 122 Supplementary Figure 14 Travel frequency hotspots for P.v. positive people in Singu Township................................................................................................................... 123 Supplementary Figure 15 Travel frequency hotspots for P.f. positive people in Singu Township................................................................................................................... 124 ix List of Abbreviations ABM agent-based modeling ACTs artemisinin-based combination therapies AUC the area under the receiver operating characteristic curve EEMS estimated effective migration surfaces GMS Greater Mekong Subregion HMR highly mobile group for regional travel ICEMR International Center of Excellence for Malaria Research IMRT imported malaria risk due to regional travel LISA Local Indicators of Spatial Autocorrelation LLINs long-lasting insecticide-treated nets MATSim Multi Agent Transport Simulation Maxent maximum entropy modeling MCMC Markov chain Monte Carlo MT possible malaria transmission due to regional travel MVS most visited network segments OD origin-destination P.f. Plasmodium falciparum P.v. Plasmodium vivax RDT rapid diagnostic tests RLR relative malaria exposure level at regional scales RTF regional travel frequency usPCR ultrasensitive polymerase chain reaction WHO World Health Organization x Chapter 1: Introduction 1.1 Background and motivation Malaria is an infectious disease caused by Plasmodium parasites transmitted by female Anopheles mosquitoes that impacts millions of people worldwide (Newby et al. 2016) (Figure 1.1). This disease has many symptoms, including fever, headache, chills, and vomiting. The areas that are most seriously impacted by malaria are in the tropical regions of Sub-Sahara Africa and Southeast Asia, although malaria is present in as many as 87 countries around the world (https://www.who.int/news-room/fact- sheets/detail/malaria). There were an estimated 229 million cases of malaria resulting in 409,000 deaths worldwide in 2019 (World Health Organization 2020). Figure 1. 1 Categorization of countries as malaria-free, eliminating malaria, or controlling malaria as of 2015 (Newby et al. 2016) 1 These numbers, however, actually reflect a reduction in the global malaria burden compared with 2010, when there were an estimated 239 million cases of malaria and 1,133,000 deaths due to the development and application of long-lasting insecticide- treated nets (LLINs), rapid diagnostic tests (RDTs), and artemisinin-based combination therapies (ACTs) (World Health Organization 2020). The emergence of artemisinin- resistant falciparum malaria in western Cambodia (Dondorp et al. 2009a), however, and its subsequent appearance in Vietnam and Myanmar (Takala-Harrison et al. 2014; Ashley et al. 2014) , as well as the emergence of resistance to key partner drugs (Spring et al. 2015; Leang et al. 2015; Duru et al. 2015; Chaorattanakawee et al. 2015; Hamilton et al. 2019), raises concerns that the usefulness of first-line malaria treatments may be diminishing in the Greater Mekong Subregion (GMS), and that drug resistance could spread worldwide (Imwong et al. 2017; Kumar et al. 2018). This dissertation will focus on the GMS, especially Cambodia and Myanmar. Cambodia was estimated to have 138,620 clinical cases of malaria in 2019 and has been widely regarded as the epicenter of emerging resistance to artemisinin-based combination therapies (World Health Organization 2020). The overall research question for this dissertation is what are the geospatial dynamics of parasite migration and human mobility in the Greater Mekong Subregion and how do these dynamics contribute to malaria transmission in this region? This dissertation is structured as three separate studies that look first at the movement of malaria parasites across Cambodia, and then two studies that focus on human movement patterns in Myanmar and how these movements can lead to increased risk of exposure as well as transmission of malaria. 2 The first study aims to build a semi-automatic workflow to select the optimal number of demes that will maximize model accuracy and minimize computing time when generating migration maps using estimated effective migration surfaces (EEMS). Relevant literature about how geospatial modeling methods have been used to analyze spatial patterns of relevance relating to malaria to aid further understanding supports public health interventions for this study. For example, spatial kriging has been applied to study the role of land cover change and deforestation in the transmission of malaria parasites in the Democratic Republic of Congo (M. Carrel et al. 2015) and for predictive mapping of the prevalence of K13 propeller mutations in Myanmar to understand the artemisinin resistance in southeast Asia (Tun et al. 2015). Global Moran?s I statistics and Local Indicators of Spatial Autocorrelation (LISA) statistics have been used for the spatial analysis of multi-level malaria Plasmodium infections using clinical P. falciparum and P. vivax episodes, as well as subclinical P. falciparum and P. vivax infections in Pailin Province, Cambodia (Parker et al. 2017). Boolean operators were applied to a GIS-based database to detect malaria hotspots in India (Srivastava et al. 2009) and bivariate spatial logistic models have been used to understand the association between malaria risk and elevation, annual maximum temperature, rainfall and potential evapotranspiration (PET) in Malawi (Kazembe et al. 2006). Geographically weighted regression was used to identify the spatially varying effects of land use on P. falciparum parasitaemia (Ehlkes et al., 2014). Additionally, related literature about one approach named estimated effective migration surfaces (EEMS) (Petkova, Novembre, and Stephens 2016; ?Https://Github.Com/Dipetkov/Eems? 2018a) were also reviewed for study 1. EEMS 3 used genomic data for a species to visualize the spatial contours of migration and diversity for this species for a given study area. The model broadly assumes isolation- by-distance, whereby genetic similarity and geographic distance are negatively correlated and identifies areas where genetic similarity decays faster than expected for a given geographic distance (low effective migration) and areas where genetic similarity decays more slowly than expected for a given geographic distance (high effective migration). The EEMS toolkit also includes support for identity-by-descent approaches that can capture the recent gene flow migration surfaces, and estimations of migration and population-size surfaces for more recent time scales with the MAPS toolkit (Al-Asadi et al. 2019). For an understanding of historic migration patterns in a region, EEMS is a useful tool and has been used, for example, to understand the population structure of human populations in southern (Uren et al. 2016) and eastern (Brucato et al. 2018) Africa and in Europe (Mathieson et al. 2018), and to visualize barriers and corridors of gene flow associated with human migration in Scandinavia (Martin et al. 2018) and Peru (Harris et al. 2018). EEMS has also been applied to simulate historical gene flow patterns for the gray wolf (Canis lupus) (Rick et al. 2017) and the blunt-nosed leopard lizard Gambelia sila (Richmond et al. 2017), and to investigate the genetic diversity of Atlantic Bluefin tuna in the Mediterranean Sea (Antoniou et al. 2017). In this dissertation, EEMS is used to estimate migration surfaces for Plasmodium falciparum (P.f.), the deadliest human malaria species. EEMS can be a useful tool for malaria elimination programs as it can be used to identify defined geographic areas that can be targeted for interventions. However, it is important to reduce spatial uncertainty in the EEMS contours that can result from applying user- 4 defined parameters, such as the resolution of the spatial grid used in the model. For example, if the spatial grid used to define migration contours is too sparse (i.e., too few demes), then many sampling locations may be assigned to a single deme, reducing model accuracy through excessive smoothing of genomic differences. On the other hand, if the grid is too dense (i.e., too many demes), spatial uncertainty may result from the estimation of parameters for many demes lacking genomic data (Shetty et al. 2019). In addition, the number of demes included in the analysis has a substantial impact on computing time, with computing time scaling cubically with the number of demes. Researchers typically employ either an average of the results obtained from running multiple MCMC iterations using different numbers of demes to infer migration patterns (Petkova, Novembre, and Stephens 2016; Rick et al. 2017; Tsuda et al. 2016; Gopalan et al. 2019), or apply maximum likelihood values to guide selection of the number of demes (Shetty et al. 2019). In both cases, users must run EEMS at multiple grid resolutions, which can be time-consuming. Thus, in the first study of my dissertation, an approach that utilizes a density clustering algorithm to define genomic clusters is developed, and these clusters may then be used to determine an optimized maximum length of triangle edges and grid resolution. This workflow provides a systematic method to select the optimal number of demes that will maximize model accuracy and minimize computing time. The optimized workflow was tested by applying it to estimate geospatial patterns of P.f. migration in Cambodia and bordering regions of Thailand and Vietnam. The results showed that the resulting migration contours corresponded to estimates of malaria endemicity and geographic properties of the region that might be expected to impact malaria parasite migration. 5 The goal of the second study is to understand the spatio-temporal human mobility patterns for malaria control using an agent-based mobility simulation model. For this study, the mobility patterns of local village populations were investigated in two townships in Myanmar. The first was Ann Township, which is located in Rhakine State in the eastern region of Myanmar, where forest workers are a common occupation group, many of whom reach their workplace by foot. The second township was Singu Township, located in Mandalay Division in central Myanmar, where the largest occupation group was farmers, who travelled locally mainly by motorcycle. The literature review of study 2 illustrated the critical gap in knowledge about understanding the relationship between malaria risk and human mobility especially in daily movements, as previous research has identified that human movement contributes to the transmission of malaria on spatial scales that typically exceed the limits of mosquito dispersal (Wesolowski et al. 2012), and human movement has played a significant role in detecting sources for parasite transmission (Ruktanonchai, Smith, and De Leenheer 2016). Studies have showed that human population movement and travel patterns were important for malaria transmission, even in low transmission settings (Silal et al. 2015; Saita et al. 2019). Silal et al (2015) reported that the control of imported infected population inflow relating to mobility will be vital to any future malaria control or elimination strategy (Silal et al. 2015). Understanding patterns of mobility can also help researchers to better identify which population groups may be particularly vulnerable to infection as well as contribute to transmission, that in turn may lead to different mitigation strategies as well as different courses of treatment for these groups (Lynch and Roper 2011). Although most species of Anopheles mosquitoes 6 are mostly active only during the night, researchers in Peru have found that transmission during normal daily human activities can also be significant as the biting hours of Anopheles mosquitoes? do not overlap exactly with nighttime activities, and so can place individuals at risk even during the daytime (Pizzitutti et al. 2018; Forattini 1987; Turell et al. 2008). In addition, more researchers have found that daily movements across areas of differing transmission rates can increase heterogeneity in individual disease risk (Cosner et al. 2009; Hast et al. 2019; D. L. Smith et al. 2007; Stoddard et al. 2009), and this heterogeneity can in turn increase population-level transmission (Dye and Hasibeder 1986; Hast et al. 2019; Prosper, Ruktanonchai, and Martcheva 2012; D. L. Smith et al. 2014). Analysis also showed that local movements played an important role in malaria transimission in malaria endemic areas in Southern Africa (Tessema et al. 2019) and found locations of high risk hotspots to be heavily dependent on human movements (Pizzitutti et al. 2018). Therefore, studying human mobility in a local area can provide opportunities for the spatial targeting of malaria control strategies (Bousema et al. 2012; Carter, Mendis, and Roberts 2000; Ruktanonchai, Smith, and De Leenheer 2016). Recent research has investigated the use of mobile phone call records data, for example to extract the movements of local populations in Northern Namibia to better understand travel and malaria transmission (Tessema et al. 2019), and in Bangladesh, where the number of trips of infected individuals were used to detect whether a relationship existed between movement and malaria transmission (Chang et al. 2019). However, mobility patterns captured in these studies depended on the locations of cell phone towers and were not as capable of detecting detailed mobility patterns in regions 7 with low cell tower density. Futuremore, the methords used in these studies were connectivity analysis and statistical approaches. The connectivity analysis that was used to estimate these movement patterns was not able to represent detailed patterns at local scale and the statistical approaches were not able to capture how relationships between movement and malaria transmission varied across different regions within the study area. As interest in targeted malaria control at local scales increases, more detailed human movement patterns are needed to evaluate relationships between human travel patterns and malaria transmission at finer scales. One approach that has significant potential for simulating human mobility at local scales is agent-based modeling (ABMs). ABMs are designed to simulate activities and movements of individuals, and offer flexibility in modeling factors, such as spatial heterogeneity and stochasticity (N. R. Smith et al. 2018) and can accommodate individual travel behaviors and spatial variations and thus help fill knowledge gaps about transmission heterogeneities important in malaria elimination strategies (Chitnis et al. 2010; N. R. Smith et al. 2018). While different ABM simulation frameworks have been utilized in research, for the second study, Multi Agent Transport Simulation (MATSim), an agent- based, extendable, multi-agent simulation framework implemented in Java that is open- source and can be downloaded from the Internet (Horni, Nagel, and Axhausen 2016). MATSim was used to simulate the spatio-temporal mobility patterns of local villagers from different occupation groups and by different travel modes in both Ann and Singu Townships. Daily travel by local villagers who tested positive for Plasmodium falciparum (P.f.) and Plasmodium vivax (P.v.) malaria in both townships were simulated to identify common mobility patterns. This study 2 in my dissertation sought 8 to identify the key elements for building an ABM model of daily travel to work and revealed estimated spatio-temporal patterns of movement for individuals with different occupations as they traveled to work and home again. By analyzing the spatio-temporal patterns of movement for different occupation groups and different travel modes in the context of travel also by P.f. and P.v. positive cases, we are able to gain a better understanding of how mobility impacts risk of malaria infection as well as possibly contributing to the transmission of malaria in a region with different occupation structures and different modes. The third study of this dissertation extends the work undertaken in Chapter 3 on local mobility, to provide a more comprehensive picture of how both local and regional mobility contribute to increased exposure to possible infection as well as possibly contributing to further transmission of malaria. More precise spatial interventions need to be used to avoid the spread of this resistance (Gwitira et al. 2020; Stresman, Bousema, and Cook 2019; Wangdi et al. 2020; Yao Li et al. 2020; Shetty et al. 2019). To help develop spatial interventions to eliminate Plasmodium malaria, a clearer understanding of how malaria may be over longer distances as well as transmitted locally in receptive areas, and how this pattern of transmission may lead to persistence of the disease is an open research question. One possibility is that malaria infections are increasing by being imported in to areas through human movement. Especially for malaria transmission over longer distances, human movement is important since people usually move at larger spatial scales that exceed the spread limits of mosquito vectors (Chang et al. 2019; Tessema et al. 2019; Wesolowski et al. 2012). Analyzing regional mobility, defined here as the number of trips outside of a person?s township of residence, can 9 contribute to improving our understanding of some of the human-driven steps in the malaria transmission cycle. Previous research showed that there is a strong relationship existed between human movement and malaria transmission based on regional travel (Chang, et al. 2019; Tessema et al. 2019; Hast et al. 2019). However, few studies relating to regional mobility analyze such movement in the context of the exposure and transmission risk and discuss how regional travel may contribute to malaria risk being brought back to be dispersed locally (Sinha et al. 2020; Tatem and Smith 2010; Pindolia et al. 2012; Mukhtar, Munyakazi, and Ouifki 2020; Prosper, Ruktanonchai, and Martcheva 2012). In this research study, travel data collected in Singu Township, Mandeley Division, Myanmar from February to December 2018 in a field survey for a project on evidence and action for malaria elimination in Myanmar. Frequency of travel to locations outside Singu Township was analyzed in the context of risk of exposure of P.v. and P.f. malaria computed for this analysis. In Chapter 4 of this dissertation, Maxent was used to estimate the 1% prevalence of both P.v. and P.f. malaria type across the township. An agent-based model introduced in Chapter 3 was used to simulate local mobility patterns for the travelers to understand how the risk from malaria imported from regional travel may persist and spread at local scales. 1.2 Research objectives The main focus of this dissertation is to understand the geospatial dynamics of parasite migration and human mobility as factors contributing to malaria transmission in the Greater Mekong Subregion. These dynamics involve both the movement of 10 parasites and humans, and will be studied through three separate investigations and research objectives (Figure 1.2). Figure 1. 2 Research objectives of the dissertation The detailed research objectives for my dissertation are the following: The first objective for this dissertation addressed in Chapter 2, is how can estimated effective migration surfaces be optimized to detect geospatial patterns of P.f. parasite migration. Related questions are: 1) How can density clustering be used to detect significant genetic differences in parasite genomic data? 2) How can a geospatial approach be used to translate genomic distances to geographic distances to optimize the number of demes used to generate estimated effective migration surfaces (EEMS)? 3) How does optimization of the grids used for EEMS improve the identification of barriers (geographic areas that restrict the spread of the malaria parasites) 11 and hotspots (geographic areas that represent clusters of malaria parasites of certain genomic types)? This research is undertaken for a study site in Cambodia and border areas of Thailand and Vietnam. The second objective for this dissertation addressed in Chapter 3, is to understand local spatio-temporal human mobility patterns for malaria control using an agent- based mobility simulation model. Related research objectives are: 1) Identify key elements for an agent-based model of local mobility dynamics (i.e., village travel patterns) developed for 2 townships in Myanmar using travel history data collected from two separate research studies, one led by researchers from the International Center of Excellence for Malaria Research (ICEMR) at Duke University and the University of Maryland, and the other a project on evidence and action for malaria elimination in Myanmar by the same team of researchers at Duke University and the University of Maryland. a. road network characteristics, occupation, travel history (time of day, frequency), travel mode, and P.v. and P.f status of individuals. 2) Understand how the daily movement patterns of local adults (over 18 years) may be subject to risk of exposure and contribute to the transmission of malaria in Singu and Ann Townships 12 a. How do movement patterns vary among different travel modes and occupation groups, and how do their patterns compare with the mobility patterns for P.v. and P.f. positive people? b. How do two townships in different parts of Myanmar compare with respect to mobility and P.v. and P.f. positivity? The third objective for this dissertation described in Chapter 4 is to analyze multi- scale mobility to guide malaria control and elimination efforts in Singu Township, Myanmar. Related questions are: 1) What are the patterns of longer distance trips (i.e., trips outside of Singu Township) for local populations residing in Singu Township? 2) What is the relationship between mobility and parasite risk rates for townships outside of Singu Township? 3) How can an ecological niche model be applied to determine local risks of exposure within Singu Township? 4) How were combined regional and local travel used to detect potential malaria transmission hotspots in Singu Township? 1.3 Dissertation outline Chapter 1 describes the motivations and the background of the dissertation. The research questions and research objectives are also introduced in this chapter. The second chapter presents the framework I developed for optimizing the spatial grid used 13 in EEMS and then the geospatial patterns of Plasmodium falciparum parasite migration in Cambodia that were detected using this optimized method. Chapter 3 discusses a multi-agent mobility model that is applied to two townships in Myanmar, Ann Township and Singu Township, to simulate daily mobility patterns for all a set of survey respondents from two projects and were led by researchers from the Myanmar Regional Center of Excellence for Malaria Research (Myanmar ICEMR) at Duke University and the University of Maryland. The principal investigators were Dr. Myaing Myaing Nyunt and Dr. Christopher Plowe formerly with the Duke Global Health Institute The spatio-temporal patterns of movement were analyzed by different occupation groups and different travel modes and compared to the simulated movements of P.f. and P.v. positive cases. Chapter 4 extends this research by analyzing mobility at multiple scales ?local and regional ? where local travel refers to travel within Singu Township, and regional travel refers to trips taken outside of Singu Township and over longer distances. These mobility patterns were analyzed further in the context of regional and local risks of exposure extracted using malaria parasite rates and ecological niche modeling, respectively, to understand the relationship between malaria exposure and human mobility as well as how travel might contribute to increased transmission in the region. Chapter 5 concludes the dissertation by summarizing and discussing the major findings of this dissertation and presenting some possible future research directions for research related to geospatial dynamics and malaria. 14 15 Chapter 2: Detecting geospatial patterns of Plasmodium falciparum parasite migration in Cambodia using optimized estimated effective migration surfaces1 2.1 Abstract Understanding the genetic structure of natural populations provides insight into the demographic and adaptive processes that have affected those populations. Such information, particularly when integrated with geospatial data, can have translational applications for a variety of fields, including public health. Estimated effective migration surfaces (EEMS) is an approach that allows visualization of the spatial patterns in genomic data to understand population structure and migration. In this study, we developed a workflow to optimizing the resolution of spatial grids used to generate EEMS migration maps and applied this optimized workflow to estimate migration of Plasmodium falciparum in Cambodia and bordering regions of Thailand and Vietnam. The optimal density of EEMS grids was determined based on density clustering to define genomic clusters and the spatial distance between genomic clusters. Model accuracy for migration estimates using different grid resolutions was determined by examination of scatter plots of the observed and fitted genetic dissimilarity between demes and R2 estimates. The optimized workflow was applied to genomic data generated from P. falciparum sampled in Cambodia and bordering regions, and 1 This work has been published. The full reference is: Li, Yao, Amol C. Shetty, Chanthap Lon, Michele Spring, David L. Saunders, Mark M. Fukuda, Tran Tinh Hien et al. "Detecting geospatial patterns of Plasmodium falciparum parasite migration in Cambodia using optimized estimated effective migration surfaces." International Journal of Health Geographics 19 (2020): 1-11. 16 migration maps were compared to estimates of malaria endemicity, as well as geographic properties of the study areas, as a means of validating observed migration patterns. Optimized grids displayed both high model accuracy and reduced computing time compared to grid densities selected in an unguided manner. In addition, EEMS migration maps generated for P. falciparum using the optimized grid corresponded to estimates of malaria endemicity and geographic properties of the study region that might be expected to impact malaria parasite migration, supporting the validity of the observed migration patterns. Optimized grids produce EEMS migration maps that are more spatially explicit for a study region and that have reduced spatial uncertainty. This workflow will be useful to a broad range of EEMS users as it can be applied to analyses involving other geographic areas and organisms of interest. 2.2 Introduction Understanding the genetic structure of natural populations provides insight into the demographic and adaptive processes that have affected those populations, such as migration or natural selection. Such information can have important applications in fields such as conservation biology or public health, particularly when integrated with geographic data. For example, geospatial modeling methods have been used to analyze pathogen genetic or genomic data to understand spatial transmission patterns of influenza virus (M. A. Carrel et al. 2010; Lam et al. 2012; Wallace and Fitch 2008; M. Carrel et al. 2012) and typhoid fever (Baker et al. 2011), the sources of imported malaria (Chang, et al. 2019, Ghose, et al. 2019) and dengue infections(Raghwani et al. 2011), and malaria parasite landscape genetics (M. Carrel et al. 2015). 17 Often estimates of population structure are made without regard to the geographic coordinates of sampling locations and then later interpreted in the context of the geographic information. However, approaches have been developed that model both the spatial and genomic data. One such approach called estimated effective migration surfaces (EEMS)(Petkova, Novembre, and Stephens 2016; ?Https://Github.Com/Dipetkov/Eems? 2018b), uses genomic data for a species to visualize the spatial contours of migration and diversity for this species for a given study area. The model broadly assumes isolation-by-distance, whereby genetic similarity and geographic distance are negatively correlated, and identifies areas where genetic similarity decays faster than expected for a given geographic distance (low effective migration) and areas where genetic similarity decays more slowly than expected for a given geographic distance (high effective migration). The model output is a map of areas of high and low effective migration or diversity for the study region. EEMS has been used to understand the population structure of human populations in southern (Uren et al. 2016) and eastern (Brucato et al. 2018) Africa and in Europe (Mathieson et al. 2018), and to visualize barriers and corridors of gene flow associated with human migration in Scandinavia (Martin et al. 2018) and Peru (Harris et al. 2018). EEMS has also been applied to simulate historical gene flow patterns for the gray wolf (Canis lupus) (Rick et al. 2017) and the blunt-nosed leopard lizard Gambelia sila (Richmond et al. 2017), and to investigate the genetic diversity of Atlantic Bluefin tuna in the Mediterranean Sea (Antoniou et al. 2017). In our previous research, we have applied EEMS and approaches based on identity-by-descent to investigate migration patterns and population structure of 18 Plasmodium falciparum in the Greater Mekong Subregion (Shetty et al. 2019), an area of emerging multidrug resistance being targeted for malaria elimination (World Health Organization 2015). EEMS maps are visually-intuitive and may be useful to malaria elimination programs by identifying defined geographic areas that can be targeted with interventions. However, to be useful for this purpose, it will be important to reduce spatial uncertainty in the EEMS contours that can result from user-defined parameters, such as the resolution of the spatial grid used in the model. For example, if the grid is too sparse (i.e., too few demes), then many sampling locations may be assigned to a single deme, reducing model accuracy through excessive smoothing of genomic differences. On the other hand, if the grid is too dense (i.e. too many demes) spatial uncertainty may result from estimation of parameters for many demes without genomic data (Shetty et al. 2019). In addition, the number of demes included in the analysis has a substantial impact on computing time, with computing time scaling cubically with the number of demes. Researchers typically employ either an average of the results obtained from running multiple Markov chain Monte Carlo (MCMC) iterations using different numbers of demes to infer migration patterns (Petkova, Novembre, and Stephens 2016; Rick et al. 2017; Tsuda et al. 2016; Gopalan et al. 2019), or apply maximum likelihood values to guide selection of the number of demes (Shetty et al. 2019). In both cases, users must run EEMS at multiple grid resolutions, which can be time-consuming. Here we present an approach that utilizes a density clustering algorithm to define genomic clusters, which are then used to determine the optimal maximum length of triangle edges. This workflow provides a systematic method to select the optimal number of demes that will maximize model accuracy and minimize 19 computing time. We tested the optimized workflow by applying it to estimate geospatial patterns of Plasmodium falciparum migration in Cambodia and bordering regions of Thailand and Vietnam, and found that migration contours corresponded to estimates of malaria endemicity and geographic properties of the region that might be expected to impact malaria parasite migration. 2.3 Materials and Methods 2.3.1 Study area and data collection The data used in this chapter is a subset of the P. falciparum genomic data from our previous publication (Shetty et al. 2019), including 28,496 biallelic, genome-wide SNPs from 1,007 samples collected in 35 districts in Cambodia and 8 bordering districts of Thailand and Vietnam between 2008 and 2013 (Figure 1) (Dondorp et al. 2009b; Ashley et al. 2014; Chaorattanakawee et al. 2015; Hien et al. 2012; Bethell et al. 2011). SNPs were either called from whole genome sequences generated as part of the MalariaGEN Plasmodium falciparum Community Project (MalariaGEN Plasmodium falciparum Community Project 2016), or, for samples that did not meet quality control criteria for whole genome sequencing or were not part of the Community Project, were genotyped using a P. falciparum-specific Nimblegen DNA microarray (Jacob et al. 2014) (NIH Gene Expression Omnibus, Accession number: GSE100704. European Variant Archive, Accession PRJEB28530). The same nucleotide positions typed on the microarray were extracted from whole-genome data for analysis, with missingness cut-offs applied as previously described(Shetty et al., 2019). 20 Figure 2. 1 Sampling locations in Cambodia, Thailand and Vietnam 2.3.2 Computing estimated effective migration surfaces EEMS utilizes a grid of regular triangles that cover the study area. The grid is created using two user-defined parameters, a bounding box that defines the geographic area where gene flow will be modeled, and the number of demes, where a deme represents a vertex in the grid. The EEMS toolkit allows the number of demes to vary up to a maximum value of 1000, allowing for different grid resolutions. Genomic data from a given sampling location is assigned to the nearest deme, and the model uses the deme locations to estimate and map effective migration surfaces. EEMS assumes individuals migrate locally between subpopulations (at demes) and that migration rates vary by location. The model also assumes that each subpopulation exchanges migrants only with its neighbors (i.e., a stepping-stone model). For every triangle in the grid, 21 EEMS assigns diversity estimates to demes and migration estimates to triangle edges. Markov chain Monte Carlo (MCMC) methods are employed to estimate both migration and diversity parameters by sampling from their posterior distributions given observed genetic dissimilarities. The matrix of average pairwise genetic dissimilarities is computed using bed2diffs, and EEMS is then run using runeems_snps: a C++ implementation of EEMS for SNP data. For each iteration, two sets of Voronoi tessellations are generated, representing spatial patterns of migration and diversity, respectively. These tessellations are generated based on a user-defined value nseeds that represents the number of Voronoi cell,s and that is also assigned to the grids. The two Voronoi tessellations are independent of each other and are updated with a birth/death process since the number of Voronoi cells is initially unknown. A maximum likelihood method is commonly used to adjust estimates of diversity and migration so that simulated genetic dissimilarity rates fit observed genetic dissimilarity in both cases. 2.3.3 Clustering based on the distribution of P.f. genomic data To optimize the grid, we first applied a density clustering algorithm to define clusters based on parasite genomic data (Rodriguez and Laio 2014). Clustering was performed using densityClust, an algorithm that is an improvement on K-means clustering, and is available as an open source package in R (Thomas Lin Pedersen 2015). This clustering method didn?t require prior knowledge of the desired number of clusters and assumes that cluster centers are remote from points with higher local density and are surrounded by points with low local density. A decision graph (Figure 22 2.2a) and a multidimensional scaling graph (Figure 2.2b) were generated for all the samples, where the x-axis represented the local density pi of sample i and the y-axis represented the distance from the nearest points with a higher density ?i. The distance we used in our research is the genetic distance. The output of the decision graph was used to confirm the number of clusters. Points located in the upper-right quadrant distant from the other points are more likely to be cluster centers (Figure 2.2a). For this approach, all points were treated as cluster centers as long as their pi and ?i were higher than the mean pi and ?i value. After the number of cluster centers and hence, the number of clusters was determined, all the samples were assigned to a cluster based on genetic dissimilarity (Figure 2.2b). (a) (b) Figure 2. 2 (a) Decision graphs generated from clustering by fast search and find of density peaks(Rodriguez and Laio 2014) and (b) multidimensional scaling graph generated from extracting density peaks 23 2.3.4 Computing the length of triangle edges and determining the number of demes The genomic clustering results were used to determine the optimal maximum length of triangle edges. For each cluster, a kernel density map was generated based on sample sizes and locations. A natural breaks classification was used to group each kernel density map into binary categories of high and low sample density, respectively. This classification method minimized the average deviation from the class mean and maximized the deviation from the means of the other groups. It also reduced the variance within classes while maximizing the variance between classes. The category representing the highest sample density was selected to determine the spatial distribution of boundaries that represent each cluster and from which triangle lengths could be established. Since the cluster polygons were often irregularly shaped, the topological skeletons of polygons were used to capture polygon shapes. To represent the spatial distance between clusters, the nearest distance between each pair of topological skeletons was computed. Finally, the shortest distance between each pair of clusters was determined and selected to represent the maximum length of a triangle edge in the grid. To determine the number of demes that optimizes the grid, this optimized triangle edge length was used as input in an inverse function to the EEMS grid generation function with the result that the edge length of the generated EEMS grid is shorter than the optimized triangle edge length. In this way, both genomic and spatial clustering were used to guide the optimization of triangle sizes and the density of demes, i.e., grid resolution, used for generating P. falciparum parasite migration maps. 24 2.3.5 Evaluation of model accuracy While the EEMS toolkit can be used to generate both migration and diversity maps for a bounded region (Petkova, Novembre, and Stephens 2016), in this study, we focused particularly on migration maps, using genomic data from P. falciparum. The EEMS toolkit allows the generation of scatterplots to visualize the correlation between observed and fitted genetic dissimilarity between demes to determine model accuracy for migration maps. Such scatterplots have been used by researchers to evaluate the model accuracy of EEMS contours (Sherpa et al. 2019; Chan and Brown 2019; Berv et al. 2019; Pimenta et al. 2019). For each grid density, R2 was estimated and compared to determine how migration model accuracy varied by the number of demes. 2.4 Results Applying density clustering to the P. falciparum genomic data from Cambodia and bordering sites in Thailand and Vietnam, the decision graph identified five genomic clusters (Figure 3a). Kernel density analysis was applied to generate a map of these clusters and showed the five genomic clusters occupied six different locations (Figure 3b). The six locations included, 1) northwestern Cambodia bordering Thailand (Oddar Meanchey and Preah Vihear Provinces), 2) western Pailin Province on the eastern border with Thailand, 3) south of Tonle Sap Lake in Pursat Province, 4) the adjacent region in southeastern Koh Kong Province, southwestern Kampong Speu Province and the northern part of Kampot Province, 5) eastern Cambodia in an area that overlapped southern Ratanakiri Province and northern Mondulkiri Province, and 6) Bu Dop district, Vietnam (Figure 3b). In Figure 3a, one genomic cluster (colored red) was identified in 25 all six locations, while another cluster (cyan) was found in only two locations, namely the Pailin District in western Cambodia and northern Bu Gia Map National Park in Vietnam. All five genomic clusters were found in Pailin District, whereas only one genomic cluster (red) was found in eastern Cambodia in the area where Ratanakiri Province borders Mondulkiri Province (Figure 3b). (a) (b) Figure 2. 3 Density-based clustering of genomic data where (a) five genomic clusters that were detected using the decision graph corresponded geographically to (b) six geographic locations (five locations in Cambodia and one in Vietnam). 2.4.1 Generating an optimized grid for the study area The shortest distance between cluster centers computed using the topological skeletons that corresponded to the spatial footprint of each genomic cluster was 32.6 km, and was used as the maximum edge length for each triangle in the EEMS grid. The 26 corresponding number of demes was calculated by setting the longest edge length to this value and using the inverse function as described above. Using our workflow, the optimized grid contained 350 demes. The MCMC iteration using this grid resolution was 30 million; burn-in was 29 million; and thinning iteration was 9999. The running time was approximately 13 hours using 64 CPUs on a Linux high-performance network. 2.4.2 Evaluating the optimized grid We investigated migration model accuracy for the optimized grid by examining scatterplots of the observed genetic dissimilarity between demes versus the fitted genetic dissimilarity between demes. The fitted genetic dissimilarly was calculated based on the computed migration between two deme locations. The scatter plots showed a strong linear relationship (R2 value was 0.757) (Figure 2.4). Figure 2. 4 Scatter plot of EEMS model accuracy for between-demes using the optimized grid. 27 To evaluate the optimization strategy, we used EEMS to generate migration surfaces for a range of grid resolutions from 200 to 750 demes, and compared the model accuracy and computing time for these grids with the optimized grid (Figure 2.5). Using a grid of 200 demes had the poorest performance with R2=0.38 (Figure 2.5a). A grid of 400 demes also had a slightly lower R2 value of 0.748 compared to the optimized grid (R2=0.757). And while R2 values appeared to increase for grids with more than 550 demes, these cases were associated with excessive computing times (Figure 2.5b). The computing time for the 350-demes case (approximately 13 hrs) was much less than for 400 demes (28 hrs), 500 demes (39 hrs), 600 demes (73 hrs) and 700 demes (112 hrs). Running the analyses with different grid resolutions indicated that optimizing the number of demes offered the best performance for migration estimates with a significantly reduced computing time. 28 (a) (b) Figure 2. 5 Model performance for 200 to 700 demes for (a) model accuracy (R2 value for migration model accuracy) and (b) computation time. 29 2.4.3 Effective migration surfaces using the optimized grid The optimized triangular grid was used to generate an estimated effective migration surface using parasite genomic data generated from isolates collected in Cambodia and surrounding locations. The migration contours in the resulting migration map (where blue indicates relative high migration and brown indicates lower migration) showed the lowest migration index value was near Tonle Sap Lake, which is the largest inland lake in Cambodia with an area of over 12,876 sq.km, while Koh Kong Province in the southwest showed the highest migration (Figure 2.6). Southwest Cambodia (southern Koh Kong, southern Kampong Speu, Sihanoukville, Kampot, and Takeo Provinces) in general showed high migration relative to other locations, and the border area between eastern Cambodia and Vietnam was also associated with high migration while locations in the border area of northwest Cambodia and Thailand showed lower parasite migration. 30 Figure 2. 6 Estimated migration surface of P. falciparum parasites in Cambodia using the optimized grid. 2.4.4 Comparison between estimated effective migration surfaces and P.f. endemicity and annual parasite incidence in Cambodia We compared the migration surface results generated using the spatially optimized grid (Figure 2.6) with a P. falciparum endemicity map based on P. falciparum parasite rate (PfPR) data from 2010 made available through the Malaria Atlas Project (Gething et al. 2011) (Figure 2.7), and also compared migration estimates with estimates of annual parasite incidence (API) per 1000 for 2013 (Maude et al. 2014). The area of low migration near Tonle Sap Lake coincided with low P. falciparum endemicity. Areas with high P. falciparum migration in both southwestern and northeastern Cambodia were found to match regions with a relatively high prevalence of P. falciparum. This 31 close relationship between P. falciparum migration and endemicity may imply the conditions in these locations are suitable for transmission of parasites infected with P. falciparum within these regions in Cambodia. High P. falciparum migration in northeastern Cambodia was consistent with high API values greater than 20% in the Steung Treng and Ratanakiri Provinces. Figure 2. 7 P. falciparum endemicity patterns within Cambodia from the Malaria Atlas Project (Data downloaded from https://map.ox.ac.uk/) (2010) We also compared P. falciparum migration contours with data available from OpenDevelopment Cambodia on natural protected areas (OpenDevelopment Cambodia 2016) and found that our results corresponded to landcover features in a way we might expect, i.e., contours of high migration coincided with areas having mostly 32 forest landcover. For example, high P. falciparum migration in the northeastern and eastern regions of Cambodia (Figure 2.7), corresponded to heavily forested areas, including a large National Park in Ratanakiri Province as well as four large wildlife sanctuaries in Mondulkiri and Ratanakiri provinces that could have served as habitats for Anopheles mosquitoes. Another area of high P. falciparum migration was located along the border between northwestern Cambodia and Thailand, and within northeastern Kampong Thom and south of Preah Vihear Province (Figure 2.7) that were also forested with wildlife sanctuaries and protected natural habitats. Another area of higher migration was observed in Battambang and Pursat provinces, just west of Tonle Sap Lake. Low migration patterns were more notable in western Cambodia, and a region of low P. falciparum migration extended from southern Laos, across the center of Cambodia, to southern Vietnam. 2.5 Discussion In this study, we developed a framework based on both genomic and spatial clustering to select the optimal number of demes to reduce spatial uncertainty in EEMS migration maps, even in the presence of irregular sampling. EEMS migration contours can vary? sometimes substantially?based on the selected number of demes. Therefore, having a systematic, rational approach to determine grid density will likely be helpful to EEMS users. We were able to test and show that optimized grids displayed both high model accuracy and reduced processing time compared to grid densities selected in an unguided manner. In addition, when we utilized an optimized grid to generate EEMS 33 migration maps for P. falciparum, we found that migration contours corresponded to estimates of malaria endemicity and geographic properties of the study region that might be expected to impact malaria parasite migration. The results of our genomic clustering approach indicated the presence of multiple genomic clusters based on malaria parasite genomic data generated from isolates collected in provinces in western, northwestern, and southwestern Cambodia. This finding is consistent with other studies conducted in the region during the same time frame which found multiple genetic subpopulations of parasites from the same study site that were hypothesized to have originated as founder populations resulting from the emergence of artemisinin resistance (Miotto et al. 2013; 2015). The congruence of these findings suggests that the density clustering approach applied in our optimization framework is accurately capturing known patterns of parasite genetic diversity in the study area. Our results indicated overlap between areas of high P. falciparum migration and hotspots of malaria incidence in eastern Cambodia (Cui et al. 2012) (Sluydts et al. 2014; Organization 2010; Steenkeste et al. 2010), as well as other areas of high malaria endemicity (Gething et al. 2011; Autino et al. 2012a). In Battambang and Pursat Provinces, high P. falciparum migration could result from flooding of the forests around Tonle Sap Lake during the wet season (Sawada et al. 2007), providing habitats for malaria vectors (Obsomer, Defourny, and Coosemans 2007). In Pursat and Preah Vihear Provinces, high P. falciparum migration areas coincided with a high prevalence 34 of multidrug resistance that is known to have emerged and spread in the area during this time frame (Amato et al. 2017). Migration maps generated in this study corroborated major migration barriers for P. falciparum identified in our previous study (Shetty et al. 2019). However, use of the optimized grid allowed detection of a migration barrier in Pailin Province that was not identified in our previous analysis that is consistent with malaria elimination efforts in this area that have contributed to a dramatic decline in clinical malaria incidence [44? 46]. P. falciparum migration barriers in northern Cambodia may have been due to higher urbanization (lower vegetation coverage) north of Tonle Sap Lake as well as Tonle Sap Lake itself, which is a large enough waterbody that it may have served as a barrier to P. falciparum migration. Deforestation of cardamom forests and large-scale land acquisitions in the area corresponded to the southern part of the ring-like contour of low migration and may also have been a contributor to this migration barrier (Milne 2012; Davis et al. 2015). The Mekong River running through southeastern Cambodia as well as the urbanized area of Phnom Penh, may both have contributed to reduced parasite migration in this part of Cambodia. Future research will continue to investigate how spatial granularity of sampling may contribute to uncertainty in EEMS migration maps. For example, data from parasite isolates in this study were geolocated at the district level, which could lead to spatial uncertainty based on aggregation of multiple locations into a single location. Improving local geographical granularity may aid in detecting more detailed migration patterns. Further investigation is also required to improve our understanding of any 35 boundary effects in EEMS analyses, as well as the impact of the assumption of geographic uniformity across a study area implied by use of a uniform grid, since geographic uniformity is not assumed for all studies (Ringbauer et al. 2018), and is an assumption that is likely violated in many settings, including in studies of the malaria parasite as presented here. 2.6 Conclusions We have developed a semi-automatic workflow to select the optimal number of demes that will maximize model accuracy and minimize computing time when generating migration maps using EEMS. The optimized grids produce migration maps that are more spatially explicit for a study region and that have reduced spatial uncertainty, an important consideration if maps are to be used to guide intervention strategies. We tested the optimized EEMS workflow on data generated from parasite isolates collected in Cambodia and bordering regions of Thailand and Vietnam, and found that migration contours corresponded to estimates of malaria endemicity and geographic properties of the region that might be expected to impact malaria parasite migration, supporting the validity of EEMS migration estimates. While in this study, our optimization framework was applied to the malaria parasites, we believe it is generalizable for other study areas and pathogens and can be used to generate optimized grids for more spatially diverse regions. 36 Chapter 3: Understanding spatio-temporal human mobility patterns for malaria control using a multi-agent mobility simulation model 3.1 Abstract As the interest in targeted malaria control at local scales increases, more detailed human movement patterns are needed to evaluate relationships between human travel patterns and malaria transmission at finer scales. In this chapter, a multi-agent mobility simulation model was built to simulate the local movements of study participants between home and their workplaces in two townships in Myanmar. Particular travel characteristics in these townships were captured to gain a better understanding of how mobility impacts risk of malaria infection as well as possibly contributing to the transmission of malaria in a region with different occupation structures and different travel modes. Multi Agent Transport Simulation (MATSim), an agent-based, extendable, multi-agent simulation framework was used to simulate human mobility. A multi-origin, multi-destination algorithm based on Dijkstra's algorithm was applied using the road network and travel distance information in conjunction with land cover data to determine likely destination locations for different occupations. A hexagon- based grid was applied to aggregate the simulation results at intersection points in the travel network and to create a summary view of the simulated paths of movement. An agent-based simulation model was built using a multi agent transport simulation (MATSim) framework to estimate and simulate the daily movements of local populations in Singu and Ann Townships in Myanmar. In terms of travel distance, 37 forest workers were the most mobile occupation group in Ann Township, followed by farmers. However, the situation was different in Singu Township, where farmers were found to be the most mobile group, and forest workers the least mobile occupation group. While forest workers and farmers in Ann Township shared similar most visited network segments (MVS) with P.f. positives, in Singu Township, travel by P.f. positive individuals shared more MVS with forest workers than any other occupation. As other researchers have found that occupation can be a strong predictor for understanding the transmission of malaria (e.g., Dunn, Le Mare, and Makungu 2011; Monroe et al. 2015), our findings reveal important details about the mobility characteristics of different occupation groups and how these vary across space. 3.2 Introduction Controlling malaria across the globe has been a stated goal of the WHO (World Health Organization 2019) with confirmed cases of malaria being reduced by 37%, and mortality by 60% between 2000 and 2015 (Cibulskis et al. 2016). Acquiring knowledge about the relationship between human mobility and risk of malaria transmission assists malaria elimination efforts by providing decision support for identifying sources and sinks of imported malaria infections (Wesolowski et al. 2012). Understanding patterns of mobility can also help researchers to better identify which population groups may be particularly vulnerable to infection as well as contribute to transmission, that in turn may lead to different mitigation strategies as well as different courses of treatment for these groups (Lynch and Roper 2011). Researchers have found that daily movements across areas of differing transmission rates can increase heterogeneity in individual 38 disease risk (Cosner et al. 2009; Hast et al. 2019; D. L. Smith et al. 2007; Stoddard et al. 2009), and this heterogeneity can in turn increase population-level transmission (Dye and Hasibeder 1986; Hast et al. 2019; Prosper, Ruktanonchai, and Martcheva 2012; D. L. Smith et al. 2014). Because this heterogeneity in disease risk drives transmission, malaria tends to persist in localized areas of self-sustaining transmission (Bousema et al. 2010) as humans and mosquitoes both transport parasites (Bousema et al. 2012; Wesolowski et al. 2012). Recent studies have showed even in low transmission settings, human population movement and travel patterns were important for malaria transmission (Saita et al. 2019; Silal et al. 2015). Silal et al (2015) reported that the control of imported infected population inflow relating to mobility will be vital to any future malaria control or elimination strategy (Silal et al. 2015). Understanding patterns of mobility can also help researchers to better identify which population groups may be particularly vulnerable to infection as well as contribute to transmission, that in turn may lead to different mitigation strategies as well as different courses of treatment for these groups (Lynch and Roper 2011). Although many species of Anopheles mosquitoes are mostly active during the night, researchers in Peru have found that transmission during normal daily human activities can also be significant as the biting hours of Anopheles mosquitoes? do not overlap exactly with nighttime activities, and so can place individuals at risk even during the daytime (Pizzitutti et al. 2018; Forattini 1987; Turell et al. 2008). In addition, researchers have found that daily movements across areas of differing transmission rates can increase heterogeneity in individual disease risk (Cosner et al. 2009; Hast et al. 2019; D. L. Smith et al. 2007; Stoddard et al. 2009), and 39 this heterogeneity can in turn increase population-level transmission (Dye and Hasibeder 1986; Hast et al. 2019; Prosper, Ruktanonchai, and Martcheva 2012; D. L. Smith et al. 2014). Analysis has also showed that local movements have played an important role in malaria transimission in malaria endemic areas in Southern Africa (Tessema et al. 2019) and have found that locations of high risk (hotspots of malaria) may be heavily dependent on human movements (Pizzitutti et al. 2018). Therefore, studying human mobility over a local area can provide opportunities for the spatial targeting of malaria control strategies (Bousema et al. 2012; Carter, Mendis, and Roberts 2000; Ruktanonchai, Smith, and De Leenheer 2016). Recent research has investigated the use of mobile phone call records data, for example to extract the movements of local populations in Northern Namibia to better understand travel and malaria transmission (Tessema et al. 2019), and in Bangladesh, where the number of trips of infected individuals were used to detect whether a relationship existed between movement and malaria transmission (Chang, et al. 2019). However, mobility patterns captured in these studies depended on the locations of cell phone towers and were not as capable of detecting detailed mobility patterns in regions with low cell tower density. Furthermore, the methods used involved connectivity analysis and statistical approaches. The connectivity analysis used to estimate these movement patterns was not able to represent detailed patterns at local scale, and the statistical approaches were not able to capture how relationships between movement and malaria transmission varied across different regions within the study area. As interest in targeted malaria control at local scales increases, more detailed human movement patterns are needed to evaluate relationships between human travel patterns 40 and malaria transmission at finer scales. One approach that has significant potential for simulating human mobility at local scales is agent-based modeling (ABMs). ABMs are designed to simulate activities and movements of individuals, and offer flexibility in modeling factors, such as spatial heterogeneity and stochasticity (N. R. Smith et al. 2018) and can accommodate individual travel behaviors and spatial variations and thus help fill knowledge gaps about transmission heterogeneities important in malaria elimination strategies (Chitnis et al. 2010; N. R. Smith et al. 2018). While different ABM simulation frameworks have been utilized in research, for the second study, Multi Agent Transport Simulation (MATSim), an agent-based, extendable, multi-agent simulation framework implemented in Java that is open-source and can be downloaded from the Internet (Horni, Nagel, and Axhausen 2016). MATSim was used to simulate the spatio-temporal mobility patterns of local villagers from different occupation groups and by different travel modes in both Ann and Singu Townships. Daily travel by local villagers who tested positive for Plasmodium falciparum (P.f.) and Plasmodium vivax (P.v.) malaria in both townships were simulated to identify common mobility patterns. This study in my dissertation sought to identify the key elements for building an ABM model of daily travel to work and revealed estimated spatio-temporal patterns of movement for individuals with different occupations as they traveled to work and home again. By analyzing the spatio-temporal patterns of movement during the dry season for different occupation groups and different travel modes in the context of travel also by P.f. and P.v. positive cases, we are able to gain a better understanding of how mobility impacts risk of malaria infection as well as possibly contributing to 41 the transmission of malaria in a region with different occupation structures and different modes. 3.3 Materials and Methods 3.3.1 Study area and data collection The study areas for this research included Ann and Singu Townships in Myanmar (Figure 3.1). Ann Township is mostly rural with nearly 70% high forest coverage, while Singu Township has less than 25% of its area covered by forest (the remaining 75% is cropland and bare land). Both townships are at low elevation of 354m and is 264m for Ann and Singu respectively. With respect to occupation structure, Ann Township has approximately 63.3% of its population working in forestry and agricultural-related areas (Ministry of Immigration and Population 2015), while for Singu Township, approximately 50.9% are working in the forestry and agricultural related sectors, and 13.2% in mining and quarrying areas (Ministry of Immigration and Population 2015). 42 (a) (b) Figure 3. 1 Study villages and road networks for (a) Ann Township and (b) Singu Township Travel history data for Ann Township was collected from individuals in five villages from August 2018 to May 2019 who were participants in a study led by researchers from the International Center of Excellence for Malaria Research (ICEMR) at Duke University and the University of Maryland. As part of their enrollment process, individuals completed a travel history survey collected using a standardized case report form. The form collected information on demographics (gender, age, occupation), travel history (time of day for travel to and from occupations, typical travel distances to their occupation, travel mode) and information on other non-work activities (e.g., 43 household chores involving trips to water). The travel history data of 1000 individuals were collected and 464 of them were adults (over 18 years) and travelled during the dry season. Most of the workers in the study sample were outdoor workers who reported traveling from 1 to 5 miles. Most reported walking to their work destinations (89%), while a smaller group reported using motorcycles (6%) (Table 3.1). For Singu Township, similar travel survey data were collected from 5564 participants in thirty-seven villages from February 2018 to December 2018 in a field survey for a separate project on evidence and action for malaria elimination in Myanmar led by the same team at Duke University that was focused on investigating approaches for developing and implementing improved strategies for identifying and eliminating malaria transmission hotspots, as well as strengthening capacity and political will for elimination in Myanmar. Among the 5564 participants, there were 2741 adult individuals who traveled during dry season. Most of the participants in Singu Township reported being outdoor workers, and most reported traveling less than 1 mile to their occupation (Table 3.1). For travel modes, 59% reported walking to their work destinations, while 34% reported using motorcycles (Table 3.1). 44 Table 3. 1 Outdoor occupation and travel information for Ann Township and Singu Township Ann Township Singu Township Plantation workers 59% 13% Farmers 30% 37% Main outdoor occupations Loggers 3% 5% Mine workers 0.2% 27% Less than 1 mile 18 % 55 % Travel distance 1 to 5 miles 62 % 40 % Larger than 5 miles 20 % 5 % 5 am to 9 am 97 % 97 % Travel to work time 9 am to 12 pm 3 % 3 % 4 pm to 8 pm 81 % 75 % 12 pm to 4 pm 7 % 14 % Travel back home time 9 am to 12 pm 11 % 8 % Other 1 % 3 % 45 3.3.2 Malaria prevalence data In addition to travel history information, each study participant was screened from August 2018 to May 2019 for P.f. and P.v malaria using ultrasensitive polymerase chain reaction (usPCR) tests. The primary aim of the screening process is not to obtain an unbiased measurement of malaria prevalence, which is already known to be low in many sites. Rather, it is to identify and enroll in the study the target number of cases of subclinical malaria infection, which are expected to be rare. People and locations suspected or known to harbor malaria was targeted for screening, based on: RDT and usPCR cross-sectional prevalence data from our ongoing separately-funded malaria surveillance activities; RDT-based malaria incidence data from previous and ongoing research and surveillance activities; interviews with local health providers and community leaders; and results of initial screening. During the screening process, household and workplace contacts of suspected or confirmed malaria cases were traced/ invited to find more infected individuals until enrollment targets were reached. This targeted screening means that the overall sampling framework was neither random nor fully representative of the larger communities. However, whenever possible, comprehensive or representative sampling was tried to be carried out to provide unbiased prevalence estimates of P.f. and P.v. infection for specific locations or populations within the larger study areas (e.g. sampling all individuals in a village). While sampling strategies were necessarily at locale-specific (e.g. clinic-based, school- based, military-based or household-based), at each location the basis for sampling was carefully documented so that we can ascertain when unbiased estimates of prevalence can be made for specific locations or populations. Each site had a central location for 46 screening, enrollment and follow-up and satellite locations, e.g. clinics or schools, where study activities were occurred. Based on the screen results, fifty-eight individuals from Ann Township were infected with P.v. malaria (5.8%) while forty- five individuals (4.5%) were infected with P.f. malaria. In Singu Township, there were 304 individuals (5.5%) were infected with P.v. malaria and 71 individuals (1.3%) infected with P.f. malaria. 3.3.3 Road networks and environmental data Detailed road networks for both Ann and Singu Township were digitized from high resolution remote sensing imagery provided via ESRI?s Community Maps Program for 2018 and 2019 respectively (Figure 3.1). Additional environmental data were also collected for both Townships from imagery products including cropland land cover sourced from Global Food Security Support Analysis Data (GFSAD) 2 , a managed forests data layer determined using Global Forest Change (GFC) data 3 (forests that were disturbed between 2000 and 2016 OR showed forest gain) and natural forest areas were extracted using GFC data (forests with forest cover greater than 40%) (Figure 3.2). 2 https://www.usgs.gov/centers/wgsc/science/global-food-security-support-analysis-data-30-m-gfsad 3 Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova, S. A., Tyukavina, A., ... & Kommareddy, A. (2013). High-resolution global maps of 21st-century forest cover change. science, 342(6160), 850-853. 47 Figure 3. 2 Land cover in Ann and Singu Townships 3.3.4 Determining origin and destination locations and travel modes for agents To simulate occupation-related movement patterns of villagers, origin-destination locations were determined for use in the ABM. Origin locations corresponded to the centroid of each village. Destinations refer to representative workplace locations for the occupations identified in the case report forms. This study focused on four occupation types including, plantation workers, miners, farm workers, and loggers. The plantation workers and loggers were combined as forest workers since the landcover data currently used not capable of differentiate the destination difference for loggers and plantation workers (Table 3.2). 48 Table 3. 2 Datasets used to estimate locations for each occupation Occupation Selected location Data source Mine worker Mining areas Dataset from Conservation Ecology Center, Smithsonian Conservation Biology Institute4 Forest worker (plantation Managed forest Global Forest Change (GFC) data5 worker) Forest worker (logger) Forest areas Global Forest Change (GFC) data5 Farm worker Crops land Global Food Security Support Analysis Data (GFSAD)6 To determine the landcover region related to the various occupations modeled by each agent, different datasets were used (Table 3.2). For the ABM, the set of possible workplace destination locations were determined by overlaying these landcover regions with an estimated region of travel detected using road network data as well as the reported travel distances provided by respondents in the case report forms. To detect these regions, an origin-destination (OD) matrix based on Dijkstra's algorithm was 4 LaJeunesse Connette, Katherine J., et al. "Assessment of mining extent and expansion in Myanmar based on freely-available satellite imagery." Remote Sensing 8.11 (2016): 912. 5 Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova, S. A., Tyukavina, A., ... & Kommareddy, A. (2013). High-resolution global maps of 21st-century forest cover change. Science, 342(6160), 850-853. 6 https://www.usgs.gov/centers/wgsc/science/global-food-security-support-analysis-data-30-m-gfsad 49 applied using the road network and travel distance information to create travel catchment areas for the agents. The catchment areas for each agent were overlayed on the landcover regions and treated as the estimated work location area for each agent (Figure 3.3a). A fishnet grid was then generated based on this estimated work area (Figure 3.3b). The spatial mean of all the fishnet grid points located in the possible work area was selected as an initial work destination location (Figure 3.3b). In addition to the spatial mean, the ten nearest points to this point were also used as candidate work place destinations. These candidate locations were used in the ABM simulation to retrieve an optimized workplace destination for an agent. (a) (b) Figure 3. 3 Determining the possible working areas (a) and candidate working destinations (b) using spatial mean and its ten nearest points 50 3.3.5 Simulating mobility of local populations Three different travel modes (walking, motorcycle) were modeled in the simulations, and estimated trip trajectories for all agents in all villages and both townships were computed and mapped. Trip statistics were calculated including estimated travel speed, duration, and distance travelled within one day. For the time ranges collected from the survey (e.g., 5:00am-9:00am, 12:00pm-4:00pm), a Latin Hyper Cube sampling method was used to generate the go-to-work time and return- from-work time for this agent. This process is done through the R package ?lhs?. The generated time is used as the initial input for the simulations. These random times were updated during the simulation to reflect multi-agent travel patterns. For this research, agents represented individuals that resided in the villages and traveled to work on a daily basis based on different occupations (e.g., plantation workers, farmers, loggers, etc.). To invoke travel modes in the ABM, a set of rules were defined: ? Individuals travelled to their occupation and home again using the road network, ? Individuals travelled using the shortest path to the estimated location for their occupation, ? Every individual was assumed to have only one place of work. The simulated behaviors included travel to work and travel home from work. MATSim was used to simulate the mobility of villagers traveling to their occupations 51 in both townships. The decisions about how to travel between home and work locations for the mobility simulation were made as a plan. MATSim uses a scoring function (Charypar and Nagel 2005) to calculate the actual performance of a plan and utilizes the Vickrey model (Vickrey 1969; Arnott, De Palma, and Lindsey 1993) for road congestion. Originally, this formulation was established to model departure time choices, however, studies have shown that the MATSim function is also appropriate for modeling further choice dimensions (Charypar and Nagel 2005; Horni, Nagel, and Axhausen 2016; Ziemke, Nagel, and Bhat 2015). The utility of a plan Splan is computed as the sum of all activity utilities Sact,q : ?? ???1???????? = ???=0 ????????,?? (Equation 3.1) N refers to the number of activities. Trip q refers to a trip that follows activity q. MATSim uses a re-planning process to make sure all the agents have the highest score, which means that each agent undertakes the most reasonable plan during the simulation. For our simulation, the re-planning process included two parts: 1) re-plan the time of departure and 2) re-plan the occupation-based destinations. ? Re-plan the departure time Alter the previous randomly generated initiation time t0 5 minutes earlier or later for each individual and generate a new time t1. Then, re-run the scoring function to calculate an updated score ????1????????. If ?? ??1 ???????? > ?? ??0 ?? ????????, update t0 with t1, otherwise if ?? 1 ???????? < ????0????????, keep t0. 52 ? Re-plan the work destinations The 10 candidate points closest to spatial mean d0 from the set of possible occupation locations will be used as work travel destinations in the simulation process. The score ????1???????? will be recalculated for the updated set of possible destinations d ?? ??1. If ?? 1 0???????? > ??????????, update d0 with d1, otherwise if ?? ??1 < ????0???????? ????????, keep d0. To create one single view of all the simulated paths of movement, all paths for all times ranges were aggregated. Any inconsistencies in travel frequency values that exist at each intersection point in the road network must be smoothed. To do this, an approach that used a hexagon-based grid was applied to aggregate the simulation results at each intersection point in the road network (Figure 4). Hexagons reduce sampling bias that can arise due to edge effects of the grid shape and also capture curves in the road network more naturally than e.g., square grids (Grewar 2015). The travel frequency for hexagon x is calculated using equation (3.2): Figure 3. 4 Hexagon x for aggregating the simulation results 53 ?????? = ? ???? ? ?????? (Equation 3.2) ?? = ?? ?? ?? ?? ?? (Equation 3.3) ???????????? Where ?????? is the length of the road segment i within hexagon x, ?????????????? is the length of the road segment i, ?????? is the travel frequency on a road segment i. The times an agent crosses into a hexagon region contributes to travel frequency counts for this region. 3.4 Results 3.4.1 Simulating mobility patterns using an agent-based model To derive the daily mobility patterns for villagers traveling to work in the dry season in both Ann and Singu Townships, a python and an R script was written to prepare the data for running MATSim. The model was simulated for 3000 iterations and simulated travel for three occupations groups for Singu Township (farmers, forest workers, and mine workers) and two occupation types for Ann Township (farmers, forest workers) and two travel modes (foot and motorbike) for both. Generally, for both townships, individuals are more likely to travel between 5:00am to 9:00am (Figure 3.5). 54 (a) (b) Figure 3. 5 Simulated mobility patterns during morning travel in Singu Township for (a) farmers and (b) loggers 3.4.2 Simulating mobility patterns using an agent-based model Here the travel frequency patterns were used to represent the mobility patterns. The travel patterns associated with using different travel modes were analyzed and compared based on the ABM simulations. Simulated results showed that in Ann Township, walkers traveled further on the road network (219 km) (Figure 3.6a) compared with motorcyclists (171 km) (Figures 3.6b). Note that over 89% of individuals in Ann Township reported that they walked as compared to 6% who reported using a motorcycle. In Singu Township, the majority of individuals also reported walking to their occupation destinations (59 %) (Figure 3.6c), and traveled an 55 estimated 375 km, in comparison to the travel distances estimated for motorcyclists (34 %) that was approximately 408 km of the road network (Figure 3.6d). (a) (b) (c) (d) Figure 3. 6 Simulated travel patterns for (a) motorcyclists and (b) walkers in Ann Township and for (c) motorcyclists and (d) walkers in Singu Township 56 3.4.3 Road network usage for different occupation groups in two Townships Simulated mobility analysis showed that in Ann Township, forest workers were the most mobile group, based on an analysis of travel frequency (defined as the number of times an agent crosses into a hexagon) traveling on more than 192 km of the road network (Figure 3.7a). This travel frequency was the highest out of all occupation groups in Ann Township. Farmers were estimated to have the second highest road network usage and traveled an estimated 160 km in Ann Township (Figure 3.7b). Farmers used the roads around Taung Chauk, Ge Laung, Pyung Thay and Tan Chaung villages. Fewer farmers were estimated to travel between Taung Chauk and Ge Laung for daily work as the travel frequencies of farmers were very low along the roads that connected these two villages (Figure 3.7b). The same situation held for Thin Pan and Tan Chaung villages. 57 (a) (b) Figure 3. 7 Travel patterns in Ann Township for (a) forest workers and (b) farmers. In Singu Township, the situation was a little different as in general travel distances were greater than those in Ann Township. In Singu, farmers were estimated to travel the farthest on the road network, traveling nearly 300 km in total distance on local roads (Figure 3.8a). Miners were second to farmers with respect to mobility, traveling over 266 km (Figure 3.8b), while the total distance traveled by forest workers was estimated at about 201 km (Figure 3.8b). 58 (a) (b) (c) Figure 3. 8 Travel patterns in Singu Township for (a) farmers, (b) miners and (c) forest workers. 59 3.4.4 Comparing the most visited network segments for P.f.and P.v. positive individuals and different occupation groups in the two townships To compare the mobility patterns and routes traveled by P.v. or P.f. positive individuals for all occupations in both townships with the most visited network segments (MVS) modeled as hexagons by occupation, a Getis-Ord Gi* hotspot analysis was conducted using the results of the simulation. The MVS for P.v. positive cases were along the roads near Taung Chauk village (Figure 8a). In addition, the roads that connected Ge Laung and Pyaung Thay villages, and the roads located in the north of Tan Chaung were also estimated to be most visited travel routes for P.v. positive individuals (Figure 3.9a). The MVS for agents representing P.f. positive cases in Ann Township were similar to the MVS for P.f. positives, with additional of roads between Thin Pan and Tan Chaung in eastern Ann Township (Figure 3.9b). Comparing these mobility patterns to the patterns by occupation, both farmers and forest workers in Ann Township shared most of their MVS with P.v. positives with an overlap of approximately 75.9% and 69.0% respectively (Figure 3.9a). Forest workers shared the most MVS with P.f. positives with an overlap of approximately 53.1% (Figure 3.9b) while farmers similarly shared 50% of their MVS with P.f. positives (Figure 3.9b). 60 (a) (b) Figure 3. 9 Most visited network segments (MVS) in Ann Township for (a) P.v. positives, (b) P.f. positives overlapped with the MVS for forest workers and farmers In Sing Township, travel routes in the northeastern area of the township were hotspots for travel by P.v. positive individuals. This also included the region near Nat Taung and Ta Line Ngoke villages, and the region near Kone Aei and Mya Thi Da and Mya Kan Thar villages (Figure 3.10a). The MVS for P.f. positive individuals was the region near Mar Lae, War Boe, Pyin Chaung and Za Loke Gyi villages in Singu Township (Figure 3.10b). Forest workers shared the most MVS with P.v. positives in Singu with an overlap of 67.3% (Figure 3.10a). Farmers had the second most similar MVS patterns compared to the P.v. positives in Singu Township with an overlap of 59.3% (Figure 3.10a). Miners only shared 35.1% of their MVS with P.v. positives in Singu Township (Figure 61 3.10a). Forest workers in Singu Township had the most similar mobility patterns to the P.f. positives with approximately 63% of MVS overlapping (Figure 3.10b). Farmers had mobility patterns that differed the most from those of P.f. positives with less than 13 % of overlap between MVS for farmers and P.f. positives, and most of these were located in central and southern parts of the Township (Figure 3.10b). Approximately 34.4% of these hotspot routes overlapped with the hotspot routes of P.f. positives (Figure 3.10b). (a) (b) Figure 3. 10 Most visited network segments (MVS) in Singu Township that overlapped with the MVS for forest workers, miners and farmers for (a) P.v. positives, (b) P.f. positives 62 In Ann Township, looking specifically at the MVS for farmers who were P.f. positive, these cases overlapped 34.0% of the MVS for all individuals reporting as farmers. The situation was similar for farmers who were P.v. positive, where 24.1% of their MVS overlapped with the MVS for all the farmers in Ann Township. For forest workers, the degree of overlap among MVS was higher, with 52.8% overlap between MVS for forest workers and those who were P.f. positive, and 42.2% for P.v positive forest workers. Locations near Pyaung Thay, Ge Laung and Tan Chaung were the MVS locations where both farmers and forest workers were either P.f. or P.v. positive. Walkers may have been at higher risk of being infected through mosquito biting during their travels compared with those who used motorcycles. The MVS for forest workers and farmers who traveled by foot in Ann Township shared a similar high degree of overlap of MVS with those who were infected with P.v. malaria with percentages of 93.1% and 96.6% respectively. P.f. positive cases shared more than 70% of their MVS with forest workers who traveled by foot compared with 60% for farmers who traveled by foot (Figure 3.11). These results suggest that most farmers in Ann Township were likely at less risk for either P.f. or P.v. malaria. The higher degree of overlap between the MVS for forest workers overall with those who were P.f. positive or P.v. positive suggests that it was the forest workers who were at greater risk in Ann Township for both P.f. and P.v. malaria. 63 (a) (b) Figure 3. 11 Most visited network segments (MVS) in Ann Township that overlapped with the MVS for forest workers and farmers for (a) P.v. positives, (b) P.f. positives For Singu Township, a similar situation was found where the MVS for P.f. positive farmers had an overlap of 25.8% of the MVS for all farmers, while a higher degree of overlap of 35.4% was found with P.v. positive farmers. The MVS for P.f. positive forest workers had an overlap of 42.5% with the MVS for all forest workers. This was very similar to the overlap for forest workers infected with P.v. malaria, which was 42.2%. When we looked further at walkers within each occupation group, forest workers who traveled by foot shared the most (48.6%) MVS with the MVS of P.f. positive, compared with farmers (23.4%) and miners (17.1%) who traveled by foot. Farmers and forest workers who traveled by foot shared the most MVS with P.v. positives with an overlap of 49.6% and 33.6%, respectively, compared with miners who travel by foot (only 13.6% overlap). These findings suggest that in Singu 64 Township, forest workers, more so than farmers, may be the group most at risk from both P.f. and P.v. malaria, especially the forest workers who traveled by foot (Figure 3.12.). (a) (b) Figure 3. 12 Most visited network segments (MVS) in Singu Township for (a) P.v. positives, (b) P.f. positives overlapped with the MVS for forest workers, miners, and farmers 3.4.5 Validating the mobility simulation To validate the ABM simulation, a survey that included a set of questions about local travel for ten different locations was designed for Singu Township. Singu was selected as this township had the largest number of villages as well as P.v. and P.f. cases among the two townships. The survey questions included asking local residents to describe the traffic at three different times (7am, 1pm, 4:30pm) and at 10 different 65 locations in the township. They were also asked to provide information on what kind of travel mode was the most frequent on these roads and locations to help us validate the travel modes simulation. The survey was sent to a group of local experts in Singu Township who were familiar with patterns of local traffic. They were asked to categorize traffic into four categories, very light, light, medium, and busy. Values from 0 to 3 were assigned to each category. For the ten locations, the mean value was calculated based on the gradings provided. To compare the simulated mobility with the mean value for each location, a natural breaks classification was applied to the aggregated results using four classes. The class value of the hexagons near each of the ten locations were compared with the mean value to determine the degree of match. The results found that seven of ten points matched the survey results giving an overall accuracy of 70% for the ABM simulation (Figure 3.13). Figure 3. 13 Aggregated travel frequency and selected sites for mobility simulation validation 3.5 Discussion Agent-based models have been proposed as an effective approach to model human mobility associated with the spread of viruses. Previous research involving ABMs in 66 infectious disease epidemiology included comparisons of vaccination strategies to address a deliberate bioterrorist introduction of smallpox (Tracy, Cerd?, and Keyes 2018), contact tracing and quarantine to reduce measles transmission (Enanoria et al. 2016), and hygiene procedures to reduce Clostridium difficile infection transmission in health care settings (Codella et al. 2015), and vaccination strategies against influenza pandemics, including their impact on health care personnel (Lee et al. 2010; Tracy, Cerd?, and Keyes 2018). Researchers have also used ABMs to understand the relationship between the spreading of innovations and human movements in the ancient world (Djurdjevac Conrad et al. 2018). The use of local surveys often capture accurate records of an individual?s position in space combined with an annotated description of the purposes of each trip (Liang et al. 2013; Schneider et al. 2013). As ABM models characterizes each agent with a variety of variables that are considered relevant to model different scenarios, they need realistic data from which to create agents that effectively capture human behavior. In this chapter, an agent-based simulation model was built using a multi agent transport simulation (MATSim) framework to estimate and simulate the daily movements of local populations in Singu and Ann Townships in Myanmar in order to identify how two townships in different parts of Myanmar were similar or different with respect to mobility and P.v. and P.f. positivity. The simulation model allowed us to generate spatial patterns of mobility that were not available using only the case report forms. The key elements for building an ABM model of daily travel that captures mobility for local scales were identified and included travel time, travel mode, travel origins and destinations. The mobility patterns for different modes and occupation 67 groups were simulated and provided insights on the spatial extent of travel, as well as the estimated routes travelled by local village populations. These patterns were compared with the mobility patterns of P.v and P.f positive individuals. In terms of travel distance, forest workers were the most mobile occupation group in Ann Township, followed by farmers. However, the situation was different in Singu Township, where farmers were the most mobile group, and forest workers were the least mobile occupation group. Analysis of the simulated results used a hexagon-based approach to generate a continuous surface from which mobility paths could be extracted. This analysis showed that both forest workers and farmers in both townships shared the same most visited network segments (MVS) with P.v. positive individuals. In Ann Township, forest workers and farmers shared the similar MVS with P.f. positives, however, in Singu Township, travel by P.f. positive individuals shared more MVS with forest workers than any other occupation. As other researchers have found that occupation can be a strong predictor for understanding the transmission of malaria (Dunn, Le Mare, and Makungu 2011; Monroe et al. 2015), our findings reveal important details about the mobility characteristics of different occupation groups and how these vary across space. Future research will focused on including family and school-related trips, and social visits were not taken into consideration. Furthermore, trips by women and children could also be a focus for future research. 68 3.6 Conclusions An agent-based simulation model was built using a multi agent transport simulation (MATSim) framework to estimate and simulate the daily movements of local populations in Singu and Ann Townships in Myanmar. The key elements for building an ABM model of daily travel that captures mobility for local scales were identified and included travel time, travel mode, travel origins and destinations, and occupation. The simulation model allowed us to generate spatial patterns of mobility that were not available from the case report forms. The workflow described for this research and applied to simulate the human mobility is useful for other infectious diseases in addition to malaria, for example, Zika and Dengue fever. 69 Chapter 4: Analyzing multi-scale mobility to assess malaria exposure and transmission risk in central Myanmar 4.1 Abstract There were an estimated 229 million cases of malaria resulting in 409,000 deaths worldwide in 2019 (World Health Organization 2020). While these numbers actually reflect a reduction in the global malaria burden, the impact of interventions on malaria prevalence has recently slowed. In this chapter, we use data on both regional mobility (i.e., travel outside a township of residence) and local mobility (i.e., travel inside a township of residence) from a survey conducted within Singu Township to investigate how human mobility patterns may be contributing to both exposure risk and transmission for both Plasmodium vivax (P.v.) and Plasmodium falciparum (P.f) malaria. In particular this research focuses on the spatial distribution and frequency of regional mobility using degree centrality and agent-based modeling. The analysis of regional mobility extends the work undertaken in Chapter 3 on local mobility, and taken together provides a more comprehensive picture of how both local and regional mobility contribute to increased exposure to possible infection and the potential transmission of malaria. The possibility of being exposed to P.v. and P.f. malaria for locations outside Singu Township were calculated using corresponding five-year- average parasite rates reported by the Malaria Atlas Project from 2015-2019 respectively (Weiss et al. 2019; Battle et al. 2019). To estimate exposure risk within Singu Township, a maximum entropy approach for species distribution and ecological niche modeling was used with a set of remotely sensed environmental variables and 70 usPCR test results. The risk regions for both within and outside of Singu also provides information for assessing possible risk of transmission. The spatial distribution of regional travel for three different groups was examined, including individuals that were P.v. positive, P.f. positive and individuals with negative usPCR tests, and we found that regional mobility patterns were different among these groups. A logistical regression analysis was applied to detect the relationship between the mobility patterns of P.f. and P.v. positive cases with the risk level for their regional travel destinations. The local daily movement patterns for people from high, medium, or low risk regions within Singu were also studied to understand how any potential imported malaria risk may be spreading locally. Potential malaria transmission hotspots were also detected to represent areas with the highest potential of contributing to transmission of malaria in Singu. 4.2 Introduction The emergence of artemisinin-resistant falciparum malaria in western Cambodia (Dondorp et al. 2009a), and its subsequent appearance in Vietnam and Myanmar (Takala-Harrison et al. 2014; Ashley et al. 2014), as well as the emergence of resistance to key partner drugs (Spring et al. 2015; Leang et al. 2015; Duru et al. 2015; Chaorattanakawee et al. 2015; Hamilton et al. 2019) has raised concerns that the usefulness of first-line malaria treatments may be diminishing in the Greater Mekong Subregion (GMS), and that drug resistance could spread worldwide (Imwong et al. 2017; Kumar et al. 2018). More precise spatial interventions need to be used to avoid the spread of this resistance (Gwitira et al. 2020; Stresman, Bousema, and Cook 2019; 71 Wangdi et al. 2020; Yao Li et al. 2020; Shetty et al. 2019). To help develop spatial interventions to eliminate Plasmodium malaria, a clearer understanding of how malaria may be transmitted locally as well as over longer distances in receptive areas, and how this pattern of transmission could lead to persistence of the disease is an open research question. One possibility is that malaria infections are increasing by being spread through human movement. Especially for malaria transmission over longer distances, human movement is important since people usually move at larger spatial scales that exceed the spread limits of mosquito vectors (Chang, et al. 2019; Wesolowski et al. 2012). Analyzing regional mobility, defined here as the number of trips outside of a person?s township of residence, can contribute to improving the understanding of some of the human-driven steps in the malaria transmission cycle. Few studies relating to regional mobility analyze such movement in the context of the exposure and transmission risk and discuss how regional travel may contribute to malaria risk being brought back to be dispersed locally (Sinha et al. 2020; Tatem and Smith 2010; Pindolia et al. 2012; Mukhtar, Munyakazi, and Ouifki 2020; Prosper, Ruktanonchai, and Martcheva 2012). In this research study, the travel data were collected from participants in thirty- seven villages in Singu Township, Mandeley Division, Myanmar from February to December 2018 in a survey for a project relating to evidence and action for malaria elimination in Myanmar. Frequency of travel to locations outside Singu Township was analyzed in the context of the risk to being exposed to P.v. and P.f. malaria that was computed for this analysis. Statistical methods were applied to determine the relationship between regional mobility and P.v. and P.f. prevalence. A multi-agent 72 based model introduced in Chapter 3 was used to simulate local mobility patterns for the regional travelers to understand how the risk from malaria may be imported as a result of regional travel patterns and how malaria may persist and spread at local scales. To quantify malaria risk level at regional scales, two datasets, P.v. parasite rate 2000-2019 and P.f. parasite rate 2000-2019 were used in the research (Battle et al. 2019; Weiss et al. 2019). These two datasets have been used for malaria risk in other studies, for example, for understanding malaria morbidity in Africa (Alegana, Okiro, and Snow 2020), transmission comparison between P.v. and P.f. malaria globally (Price et al. 2020) and systematic analysis research for the global burden of disease in 2017 (Stanaway et al. 2019). Five-year-average parasite rates from 2015 to 2019 were used to represent the risk of exposure to both P.v. and P.f. malaria for townships outside of Singu Township. To detect areas of exposure risk for P.v. and P.f. malaria within Singu Township, a machine learning technique named maximum entropy modeling (Maxent) was used. Maxent is a species distribution prediction model based on maximum entropy theory (Phillips and Dud?k 2008; West et al. 2016). Entropy is a fundamental concept in information theory and can be described as a measure of how much ?choice? is involved in the selection of an event (Claude E. Shannon 1948; Claude Elwood Shannon 2001; Phillips 2005; Phillips et al. 2017; Phillips and Dud?k 2008; Elith et al. 2011). The theory underlying Maxent is that the best approach to estimate an unknown probability distribution is to ensure that the approximation satisfies any constraints on the unknown distribution, such the distribution should have maximum entropy (Jaynes 1957). Maxent has been well used in predicting the spatial distribution of species like Alternanthera philoxeroides (Yan et al. 2020), Cunninghamia lanceolata (Yingchang 73 Li et al. 2020) and Dermacentor marginatus (Song et al. 2020). Maxent outperformed most of the species distribution methods especially when the sampled occurrence data is limited (Yan et al. 2020; Yi et al. 2017; Saatchi et al. 2008). In this chapter, a malaria risk surface representing 1% prevalence of P.v. and P.f. malaria was generated using a maximum entropy approach for species distribution and ecological niche modeling to extract any identified malaria exposure risk locations for Singu. 4.3 Materials and Methods 4.3.1 Study area and travel history data collection The study area for this research was located in Singu Township, a township of Pyinoolwin District, Mandalay Division, Myanmar. Mobility data were collected via a travel survey of participants in thirty-seven villages from February to December 2018 as part of a project investigating approaches for developing and implementing improved strategies for identifying and eliminating potential malaria transmission hotspots, as well as strengthening capacity and political will for elimination in Myanmar (Figure 4.1). The travel survey (introduced in Chapter 3) included questions about longer distance travel, e.g., how many times did you travel outside of your township within the last year and to which townships did you usually travel within the last year, and which mode did you use when you traveled outside your township last year. The survey also included questions about short distance travel as described in Chapter 3 including for example, occupation and travel distance to work. 74 Figure 4. 1 Township destinations for regional travel in Singu From the travel history data, we had 1,810 people who reported traveling outside of Singu Township. Approximately 66.7% traveled regionally less than 2 times a year, suggesting that the majority of the participants did not travel out of their township frequently. Among the regional travelers, 79 (4.4%) of them were infected with P.v. malaria while 16 (0.9%) were infected with P.f. malaria (Table 4.1). 75 Table 4. 1 Regional travel from Singu Township Number of Percentage travelers Once a year 457 25.2% 2 times a year 665 36.7% Regional travel More than 4 times a 542 30.0% frequency year Every week 119 6.6% Every day 27 1.5% P.v. positive 79 4.4% P.f. positive 16 0.8% Infection status Negative 1716 94.8% Failed 1 0.0% 4.3.2 Road networks and environmental variables Detailed road networks for Singu Township were digitized from high resolution remote sensing imagery provided via ESRI?s Community Maps Program for 2018 and 2019 respectively (Figure 4.1). Additional environmental data were also collected from imagery products including cropland land cover sourced from Global Food Security Support Analysis Data (GFSAD), a managed forests data layer for 2016 determined using Global Forest Change (GFC) data on forests that were disturbed between 2000 and 2016 OR showed forest gain for 2016 and natural forest areas for 2016 were extracted using GFC data (forest cover greater than 40%). These dataset were used to generate Land Cover Land Use Map for the study area, the detailed methods were published in 2021 (Chen et al. 2021). 76 Precipitation data was extracted using the data from Climate Forecast System V2 2018. The MODIS MOD11A2 (Wan et al. 2015a) and MYD11A2 (Wan et al. 2015b) 8-day Land Surface Temperature/Emissivity datasets were used as the sources of our LST variable. Daytime LST as reported by MOD11A2 and MYD11A2 from the same 8-day period were merged into a single data layer. The same procedure was applied to nighttime LST. Then, we resampled the Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) to 1-km to reconcile with the LST data. The resampled DEM was used to normalize the LST data to the sea level based on the lapse rate (air temperatature decreases for 0.6 ?C for every 100 m of increase in elevation) as followed: LSTsea level=LST + 0.65 * (elevation1 km / 100) The normalized LST was then overlaid with the original 30 m SRTM DEM and 30-m LST was calculated as followed: LST30 m= LSTsea level ? 0.65 * (elevation30 m / 100) Relative Humidity (RH): The MODIS MOD07 (Borbas et al. 2016a) and MYD07 (Borbas et al. 2016b) Atmosphere Profiles Product was used as the sources of our RH variable. Specifically, air temperature at 700 hpa, surface pressure, and total column precipitable water vapor were extracted from the MOD07 and MYD07 data. For each variable, swath data acquired on the same day for the same locations were averaged to obtain the daily mean values, which were subsequently further averaged over every 8-day window to derive the 8-day mean values. RH was then calculated for each 8-day time step following the empirical equations below: 77 Tsea level= T700 hpa + 0.65 * (elevation700 hpa / 100) T30 m = Tsea level ? 0.65 * (elevation30 m / 100) MR=107.34 * p ? 0.0004 1 ???? = 1 ? 1.844 ? 10?4 ? ln( ???? ? ?? 273.15 0.6113 ? (???? + 0.622)) e = 0.6113 ? exp( 5423 ? ( 1 ? 1 )) 237.15 ???? es = 0.6113 ? exp( 17.3694?(??30???273.15)) ??30???35.86 RH = e/es where MR: mixing ratio; p: total column precipitable water vapor; Td: dew point temperature; e: actual water vapor pressure; es: saturation water vapor pressure; 4.3.3 Calculating regional travel frequency using degree centrality To quantify the regional travel frequency (RTF) between a village v and regional travel destination (i.e., a trip outside the township) t, an undirected graph between village v and township t were built. Each node in this graph represents a village or a township. Multiple edges were expected between the two nodes since usually multiple individuals will travel between a certain between village v and a certain township t. The edges were weighted using regional travel frequency and a metrics named degree centrality, was used to represent the RTF between a village and township. In weighted 78 networks the degree centrality is calculated as the sum of weights assigned to the node?s direct connected edges. Here degree centrality means the number of individuals traveled between village v and township t. Specially the following equation was used: ???????? = ???? ?? ???? ?? = ??????? ? ?????????? (Equation 4.1) ???? = ?1, ( individual ?? traveled ???? township ??)???? 0, (individual ?? never traveled ???? township ??) (Equation 4.2) ???????????? represents the regional travel frequency from individual i to township t in the last year, while individual i comes from village v. which is defined as, Table 4. 2 Converting Frequency to Fre value Frequency Fre value Do not travel 0 regionally Once a year 1 2 times a year 2 More than 4 times a 3 year Every week 4 79 Every day 5 ?????????? represents the regional travel frequency from village v to township t. 4.3.4 Potential malaria transmission due to regional travel for villages in Singu Five-year-average parasite rates from 2015-2019 were used to represent the risk of exposure to both P. v. and P. f. malaria for townships outside of Singu Township. For all townships that served as a travel destination for survey participants, we used mean parasite rate values within each township to represent an exposure risk value for each township. A quantile classification that allocates an equal number of values into groups and is particularly useful for areas of similar size, was used to map the data into three categories to represent the relative malaria exposure level at regional scales (RLR) in Myanmar. For P.v., the categories were low exposure risk (parasite rates between 0.1% and 0.7%), medium exposure risk (parasite rates between 0.7% and 0.9%) and high exposure risk (parasite rates between 0.9% and 2.3%). For P.f., the categories were also included low exposure risk (parasite rates between 0.0% and 0.2%), medium exposure risk (parasite rates between 0.2% and 0.7%) and high exposure risk (parasite rates between 0.7% and 2.6%). Note that these levels of exposure risk for P.v. or P.f. are relative to P.v. or P.f. malaria prevalence across Myanmar, so relative medium and relative high risk. To quantify possible malaria transmission due to regional travel at individual level, the potential malaria transmission for an individual i in village v, ??????????, is defined as, 80 ?????????? = ???? ? ?????????? (Equation 4.3) ???? is the regional risk level for township t based on the five-year-average parasite rate for t, and township t is the regional travel destinations for individual i. ?????????? is the regional travel frequency for individual i from village v. Then the potential malaria transmission due to regional travel between a village v and travel destination t at village level ???????? is defined as: ?????? ???? = ????????? (Equation 4.4) ?????????? represents the possible malaria transmission due to regional travel at individual level from village v to township t. 4.3.5 Modeling daily mobility patterns to understand the dispersal of imported malaria risk due to regional travel The daily movement patterns of these travelers within Singu were simulated using a multi-agent mobility simulation model presented in Chapter 3. To create a single view of all the simulated paths of movement (both local and regional), we first aggregated all paths for all times ranges. Then a hexagon-based grid was used to aggregate the simulation results at each intersection point in the road network (Figure 4.2). As discussed in Chapter 3, hexagons reduce sampling bias that can arise due to edge effects of the grid shape and also capture curves in the road network more naturally than e.g., square grids (Grewar 2015). To estimate the dispersal of imported malaria risk due to regional travel (IMRT), the travel frequency hexagon for each regional traveler was 81 weighted using its possible malaria transmission due to regional travel (MTI). IMRT was defines as Equation 4.5 and 4.6. Figure 4. 2 Hexagon x for aggregating the simulation results ???????? ???? = ? ???? ? ?????? ? ???????? (Equation 4.5) ?? ???? = ???? ?? (Equation 4.6) ???????????? Where ?????? is the length of the road segment i within hexagon x, ?????????????? is the length of the road segment i, ?????? is the travel frequency on a road segment i. ?????????? is the potential malaria transmission due to regional travel for this individual. 4.3.6 Ecological niche modeling for estimating P.v. and P.f. malaria prevalence in Singu To detect areas of exposure risk for P.v. and P.f. malaria within Singu Township, we used a species distribution estimation method named maximum entropy modeling (Maxent) to estimate the 1% prevalence of each malaria type across the township. Entropy is a fundamental concept in information theory and can be described as a measure of how much ?choice? is involved in the selection of an event (Claude E. Shannon 1948; Claude Elwood Shannon 2001; Phillips 2005; Phillips et al. 2017; Phillips and Dud?k 2008; Elith et al. 2011). A distribution with higher entropy involves 82 more choices (i.e., it is less constrained). Therefore, the maximum entropy principle used by Maxent produces probability maps that show the likelihood of an occurrence of a species within a region (Claude E. Shannon 1948; Claude Elwood Shannon 2001). Table 4. 3 Selected environmental variables for computing ecological niche modeling Environmental variables Landcover (remote sensing group from ICEMR project ) Population density (Landscan, 2019) Elevation (The Shuttle Radar Topography Mission DEM) Slope (calculated from SRTM DEM) Topographic wetness index (calculated from SRTM DEM) Precipitation (Climate Forecast System V2, 2018) Mean relative humidity in 2018 (derived using MODIS product, 2018) Maximum relative humidity in 2018(derived using MODIS product, 2018) Minimum relative humidity in 2018 (derived using MODIS product, 2018) Standard deviation of relative humidity in 2018 (derived using MODIS product, 2018) Mean nighttime land surface temperature in 2018 (derived using MODIS product, 2018) Maximum nighttime land surface temperature in 2018 (derived using MODIS product, 2018) Minimum nighttime land surface temperature in 2018 (derived using MODIS product, 2018) Standard deviation of nighttime land surface temperature in 2018 (derived using MODIS product, 2018) Mean daytime land surface temperature in 2018 (derived using MODIS product, 2018) Maximum daytime land surface temperature in 2018 (derived using MODIS product, 2018) Minimum daytime land surface temperature in 2018 (derived using MODIS product, 2018) Standard deviation of daytime land surface temperature in 2018 (derived using MODIS product, 2018) To estimate malaria risk areas using Maxent, the P.v. and P.f. prevalence rates for the locations of the 37 villages were calculated from the available usPCR data and used in the model. An additional eighteen environmental variables were also included in the model to estimate the spatial distribution of areas with a higher than 1% P.v. and P.f. prevalence rate. A leave-one-out cross validation process was applied in this process. 83 Two estimated results, one each for P.v. and P.f. malaria were generated by averaging the results from the cross validation process. In this study, the estimated value at a location represented the probability of having a higher than 1% P.v. and P.f. prevalence at this location. The exposure risk areas for P.v. and P.f. malaria were extracted by applying a threshold named maximizing the sum of sensitivity and specificity to the probability maps (Liu, White, and Newell 2013). Finally, local exposure risk maps for both types of malaria were generated using hexagon grids. The values for each hexagon were calculated based on the percentage of risk area within a hexagon. 4.3.7 Computing potential malaria transmission hotspots in Singu Two malaria transmission hotspots analyses, one each for P.v. and P.f. malaria were computed by combining the exposure risk surface with the mobility pattern analysis results at both regional and local scales. For the regional analysis, two IMRT map for P.v. and P.f. malaria were calculated for all the individuals who, based on the simulation, traveled regionally in Singu. Then the IMRT maps were scaled between 0 and 1 respectively. This step was undertaken to highlight the dispersal of any possible imported malaria due to individuals traveling outside of Singu (e.g., to a hotspot) and then returning home. The multi-agent mobility model from Chapter 3 was used to simulate the travel frequency for all the trips by adults within Singu Township and summarized using hexagon grid maps. The values by hexagon were multiplied by the exposure risk values and were also scaled between 0 and 1 to determine an overall estimate of exposure risk for both P.v. and P.f. due to local mobility. This represented the amount of potential 84 malaria transmission due to local travel within Singu. The results for potential transmission due to regional and local mobility were averaged to produce a final surface of estimated malaria transmission due to regional and local mobility in Singu. Finally, a Getis-Ord Gi* hotspot analysis was applied to this estimated surface to identify malaria transmission hotspots in Singu. 4.4 Results 4.4.1 Regional mobility patterns in Singu The regional travel frequency (RTF) was computed for each village in Singu Township. To more easily visualize the spatial distribution of RTF, villages in Singu were clustered into three groups (northern, central and southern). Based on our results, most villages had an RTF that was between 2 and 273. High RTFs were associated with villages in the northern region of Singu. The three most visited townships were Thabeikkyin (RTF=782), Shwebo (RTF=761) and Mandalay (RTF=511) township, probably because these townships are adjacent to Singu (Figure 4.3). 85 Figure 4. 3 Regional travel frequency for villages in Singu 4.4.2 Exposure risk for regional travel destinations The spatial distributions for exposure risk for P.v. and P.f. malaria outside Singu were different (Figure 4.4). Based on the computed estimates for risk outside of Singu Township based on P.v. parasite rates, 156 out of 330 townships (or 47%) were classified as low exposure risk, while 83 (25%) and 91 (28%) were medium and high risk respectively. For P.f., 40% (131 townships) were relative low P.f. risk townships, while approximately 34% (111) townships were classified as medium risk, and 27% (88) were high risk based on our analysis. While central Myanmar tended to have a relative high P.v. exposure risk, the risk for P.f. in this region tended to be low to medium. While the estimates for northern Myanmar were low to medium P.v. risk, 86 these locations had an the estimated exposure risk for P.f. that was relatively high. Townships along the eastern coast of Myanmar were relative high risk for both P.v. and P.f. malaria and this was the case also for the southwest border between Myanmar and Thailand. Among the 68 regional travel destinations, 29% (20) of them had a high P.v. risk level while only 11.8%(8) had a high P.f. risk level. Note here the risk levels (high, medium and low) were relative, it doesn?t mean a township has to have a high prevalence to be classified as high exposure risk, but rather it means its prevalence was high compared with other townships in Myanmar. Figure 4. 4 P.v. and P.f. exposure risk for townships in Myanmar 87 4.4.3 Understanding mobility and exposure risks for regional travel destinations Approximatley 28% took trips to townships with a relative high P.v. exposure risk, while only 1.5% of regional travel was to high P.f. risk townships. Based on estimated P.v. exposure risk, most trips (45.5 %) involved traveling to low risk townships, while P.f. travel was mostly (65.6%) to townships classed as medium relative exposure risk (Table 4.4). The most visited township with a relative high P.v. exposure risk was Shwebo Township. Bamauk was the most visited township that had a high P.f. risk level. Table 4. 4 Regional travel and exposure risks for regional travel destinations Relativ Number of Percentage e risk travelers High P.v. risk 520 28.7% Medium P.v. risk 446 24.6% Malaria risk levels for Low P.v. risk 824 45.5% regional travel High P.f. risk 27 1.5% destinations Medium P.f. risk 1188 65.6% Low P.f. risk 595 32.9% 4.4.4 High frequency travel We identified a highly mobile group for regional travel (HMR). This HMR group included 149 individuals (8.2% of the total group) where 27 (1.5% of the total group) traveled regionally every day and 119 (6.6% of the total group) of them traveled out of the township every week. Individuals in this group have the potential to be a catalyst 88 of malaria transmission between their villages and the townships they visited and possibly exporting malaria risk to the destination townships, although in this study, only four individuals were P.v. positive and two were P.f. positive. Over 20% of HMR traveled to high P.v. risk townships suggesting there could be a possibility of importing P.v. from these visited townships back to their own villages. In this study sample, no individuals from the HMR group traveled to high P.f. risk townships implying the probability of importing P.f. to Singu Township was low. The top three visited townships for HMR were Thabeikkyin, Shwebo and Mandalay, destinations for 83.9 % of this highly mobile group. These three townships were also the most visited townships among all the regional destinations. 4.4.5 Regional travel for P.v. and P.f. positive cases Of the P.v. positive cases, 46.8% travelled regionally to low P.v. risk townships suggesting that exporting P.v. malaria to these corresponding regions was not a determinant factor for the P.v. situation for these townships. Approximately 18.7% of P.f. positive individuals visited regional destinations with relatively low P.f. risk townships highlighting also that the exported P.f. malaria risk to these townships may not be that significant such as to increase P.f. risk in these townships. Nearly 19% of P.v. positive participants traveled to relative high P.v. risk townships. Based on this, there is a possibility that there could be further contributions to P.v. infections due to this pattern of mobility. However, there were few to no trips by P.f. positive participants to high P.f. risk townships outside of Singu. Singu Township may be sufficiently distant 89 from these locations to keep this travel-related risk low compared to other drivers (Table 4.5). Table 4. 5 Regional travel for P.v. and P.f. positive cases in Singu Risk level for Regional Proportion destinations High P.v. risk 15 (19.0%) Malaria risk for regional travel Medium P.v. risk 27 (34.2%) destinations (P.v. positive) Low P.v. risk 37 (46.8%) High P.f. risk 0 (0.0%) Malaria risk for regional travel Medium P.f. risk 13 (81.3%) destinations (P.f. positive) Low P.f. risk 3 (18.7%) Individuals who were P.v. positive in Singu reported traveling to 15 townships, with about 33.3% of these townships being adjacent to Singu. P.f. positive cases traveled to 9 townships, with approximately 55.6% being adjacent to Singu. The most visited township by both P.v. and P.f. positives was Thabeikkyin Township (directly north of Singu). Overall, trips by P.v. positive individuals tended to be to townships involving longer distances than those who were P.f. positive. 90 (a) (b) Figure 4. 5 Regional travel patterns for (a) P.v. positive and (b) P.f. positive individuals in Singu Township 4.4.6 Possible P.v. and P.f. malaria transmission due to regional travel in Singu The pairwise possible malaria transmission due to regional travel (MT) between the three groups of villages for both P.v. and P.f. cases were computed. Then the MT values were classified into three categories to represent relative high, low and medium MT. For P.v. malaria, the high MT has a value of no less than 511, the medium MT has a value between 1 and 511 and the low MT has a value equals to 1. For P.f. malaria, the high MT has a value of no less than 273, the medium has a value between 1 and 273 and the low MT is equals to 1. Based on this analysis, nearly 86% of the regional 91 trips had a medium MT for P.v. malaria and only the group in the northern part of Singu had a high MT. Only the group of villages in the north had a high MT for P.f. malaria. (a) (b) Figure 4. 6 MT for (a) P.v. and (b) P.f. malaria in Singu 92 4.4.7 Computing exposure risk for P.v. and P.f malaria within Singu Based on the Maxent results for computing ecological niche, the risk area (i.e., area of 1% prevalence probability) for P.v. exposure was mainly located in the central and northeastern areas of Singu, while the potential risk areas for P.f. exposure were mainly located in the northeast (Figure 4.7). This analysis estimated that P.v. was the more common Plasmodium malaria in Singu Township with northeastern Singu having a relatively high exposure risk for both P.v. and P.f. malaria. To validate the Maxent results, a leave-one-out cross validation process was applied and the area under the receiver operating characteristic curve (AUC) was used to quantify the accuracy of the Maxent results. This validation revealed that the average test AUC for the P.v. risk region was 0.896, and its standard deviation was 0.145, while the average test AUC and standard deviation for P.f. risk region was 0.810 and 0.115 respectively. (a) (b) Figure 4. 7 Estimate exposure risk areas for (a) P.v. and (b) P.f. malaria in Singu Township. 93 To evaluate the importance of each environmental variable in the Maxent analysis, a jackknife test was applied. The annual precipitation variable was dropped since precipitation rates were homogeneous across the Township. Typically, the most important environmental variable should have the highest gain when used in isolation, and model returns should also indicate this with a long blue bar in the model output (Figure 4.8). It was expected that the gain from the model would decrease the most when this variable was omitted, and this variable should have had the shortest green bar (Figure 4.8). For P.v. malaria risk prediction, the Jackknife results indicated that landcover data had the highest gain when used alone. Landcover data also decreased the gains the most when it was ignored in the prediction, suggesting this variable appeared to contribute information that wasn't present in the other variables. Other variables that provided useful information for estimating the ecological niche of P.v malaria included, road density, population density, and slope. Population density provided the most information for P.f. malaria prediction when used alone, and also decreased the gain the most when excluded, i.e., the model found that it contributed more useful information than other variables. Road density, standard deviation for land surface temperature at night were also among the top three variables that contributed to the estimated P.f. malaria risk areas. 94 (a) (b) Figure 4. 8 Jackknife test results for (a) P.v. and (b) P.f. malaria exposure risk estimates. 4.4.8 Determining the potential malaria transmission hotspots To understand how regional travel could play a role in importing malaria into Singu Township, we analyzed simulated travel patterns of these travelers based on their daily travel to and from an occupation in Singu (Chapter 3). The simulation results were aggregated using the hexagon grid approach (Chapter 3). There were 872 adults who reported traveling outside Singu, and for whom we simulated local work travel patterns within Singu including local travel frequency. Based on the simulation results, individuals traveled an average of 72 km during a day inside their township. Travelers with high RLR for P.v. traveled an average of 39.6 km daily while individuals with high P.f. RLR, the distance was lower, close to 18 km. Travelers with high P.v. RLR traveled throughout Singu Township while the most visited road segments for travelers with high P.f. RLR formed a corridor in the central Singu (Figure 4.9). 95 (a) (b) (c) Figure 4. 9 Local travel patterns for (a) those who travelled outside Singu Township and the travelers who went to (b) high P.v. and (c) P.f. risk townships. Both potential P.v. malaria and P.f. transmission hotspots included 13 villages with 11 of these villages being shared between the two infection types. For an area with a higher relative chance of malaria exposure, we would expect both higher mobility as well malaria risk in this area. 96 + + + Figure 4. 10 Potential Malaria transmission hotspots resulting from both local and regional travel in Singu. 4.5 Discussion For the data used in this research, there were more P.v. cases than P.f.. Nearly 30% of regional travel destinations were classified as having relative high P.v. risk compared with only 11.8% for P.f. malaria, signifying that at present, P.v. appears to be generally the more common malaria species in Myanmar, especially in Singu Township. These findings are consistent with the results from recent research that has reported that the malaria epidemiology has been changing recently in the GMS, with a rise in P.v. malaria (Brashear et al. 2020). P.v. not only has become the predominant Plasmodium species in recent years in GMS (Baird 2017; N. Li et al. 2013), but also has caused several malaria outbreaks (Zhou et al. 2016; Yingchang Li et al. 2020; Geng et al. 2019). 97 Most participants in the Singu study didn?t travel to other townships frequently implying that their possibility of transmitting P.v. and P.f. malaria risk is limited. We didn?t detect a significant linear relationship between regional travel frequency, RLR and Plasmodium malaria infection. For this reason, regional travel frequency alone cannot explain or determine whether individuals were infected with either P.v. or P.f. malaria. This is not unreasonable since malaria infection is a complex process and the risk of exposure to malaria at local scales is also a critical component to malaria infection (Dejon-Agobe et al. 2019; World Health Organization 2020). Our analysis found that Singu could act as a source for P.v. and P.f. malaria as we do see people infected with P.v. or P.f. traveling regionally to 24 different townships. Thabeikkyin could be the most possible malaria sink for Singu, as it was the mostly visited township for P.v. and P.f. positives from Singu. The northern villages in Singu were most likely to serve as a potential malaria transmission area as they had the highest RTF and MT, suggesting that the northern villages in Singu had the ability to transmit malaria to other townships as well as possibly transmit malaria into Singu. Shwebo was a possible source for regional P.v. malaria transmission for Singu, and Bamauk was a possible source for regional P.f. malaria transmission for Singu as they both had relative high parasite risk levels for P.v and P.f. respectively, and were regularly visited by Singu residents. Landcover data was the most important environment variable in the estimation of the ecological niche for P.v. malaria. This is reasonable since landcover plays a critical role in identifying high risk regions for malaria, according to Ostfeld et al. (2005) who 98 reported that using more explicit landcover information could improve the understanding and prediction of the disease risk (Curran et al. 2000; Ostfeld, Glass, and Keesing 2005; 2005). Population density was the most important variable for P.f. malaria ecological risk estimation, and population density was often found to be an important predictor of malaria risk in several studies (see for example, Kabaria et al. 2017). Other important variables we detected were, road density, slope and standard deviation for land surface temperature at night. According to previous research, road density can be a strong predictor for predicting malaria risk (Hahn et al. 2014). Slope has also been found to be very important for malaria prevalence detection for when it is combined with precipitation levels at a certain location, it may influence the dispersion of malaria. Flat areas may be more prone to accumulate water, creating dam rainwater and increasing the risk of malaria (Ferrao et al. 2018). Open, treeless habitats have a lower malaria transmission risk compared with forest sites (Ferrao et al. 2018). And warmer land surface temperatures that might enable more rapid mosquito and parasite development may eventually contribute to malaria risk (E. Mbunge et al. 2021; Sudre et al. 2013). The potential malaria transmission hotspots detected for P.v. and P.f. malaria showed the most risky and vulnerable regions in Singu. Individuals in the hotpot regions may not only have a potential of being exposed to relative higher P.v. or P.f. malaria from Singu travelers, but also shared the local travel routes from travelers returning from high RLR townships. These hotpot regions can be sentinel sites for monitoring malaria in Singu, as these hotspots may be more sensitive to malaria change than other regions in Singu. Either an increase in risk within Singu or in other townships 99 visited by residents from Singu could possibly influence malaria prevalence in these regions. 4.6 Conclusion This chapter examined mobility patterns of local village populations in Singu Township, Myanmar when they traveled longer distances outside of Singu. These travel patterns were analyzed further in the context of daily local mobility patterns within the township. Regional travel patterns (i.e., travel outside Singu) were extracted using degree centrality. The travel frequency between a village located within Singu and a township outside, was estimated using both the number of people who traveled between these two locations and their travel frequency as reported in the travel surveys. Local movements were simulated using the agent-based model described in Chapter 3. Malaria risk surfaces for P.f. and P.v. malaria were generated to account for potential malaria exposure by travelers, using the average annual parasite rates between 2015- 2019 as determined by the Malaria Atlas Project. Local malaria risk (i.e., within Singu) for P.f. and P.v. malaria was estimated using ecological niche modeling techniques. A level-one-out cross validation process was applied to assess the accuracy of this model and the importance of each of the environmental variables used in the model. Travel patterns for individuals who undertook regional trips as well as those who were P.v. and P.f. positive were identified. Potential malaria transmission due to regional travel was quantified and hotspots were detected for P.v. and P.f. malaria showing regions estimated to be more vulnerable to malaria transmission due to both 100 local and regional travel in Singu. These locations may be sentinel sites for monitoring malaria in Singu. 101 Chapter 5: Conclusions 5.1 Summary of major findings This dissertation aims to understand dynamics related to malaria in the GMS using geospatial methods. The dynamics include parasite migration dynamics and human movement dynamics. Theories and methods from the fields of GIScience, population genetics, and public health were used to achieve this goal. The major findings of this dissertation are discussed in the following paragraphs. In the first study (chapter 2), a semi-automatic workflow was developed to select the optimal number of demes that will maximize model accuracy and minimize computing time when generating migration maps using EEMS. An R package named ?DemeNumberSelector? was developed to realize this workflow. Optimization is important as using an optimized grid reduces spatial uncertainty in the resulting EEMS migration maps, even in the presence of irregular sampling. EEMS migration contours can vary?sometimes substantially?based on the selected number of demes. Therefore, having a systematic, rational approach to determine grid density will be helpful to EEMS users. A validation analysis showed that the optimized grids displayed both high model accuracy and reduced processing time compared to grid densities selected in an unguided manner. Furthermore, when an optimized grid was utilized to generate EEMS migration maps for P. falciparum, the migration contours were found to correspond well to the mapped estimates of malaria endemicity (Gething et al. 2011; Autino et al. 2012a) (Gething et al. 2011; Autino et al. 2012b) and geographic properties of the study region (e.g., Battambang and Pursat Provinces) that have been shown to impact malaria 102 parasite migration. The results of the genomic clustering approach using malaria parasite genomic data generated from isolates collected in provinces in western, northwestern, and southwestern Cambodia indicated the presence of multiple genomic clusters. This finding is consistent with other studies conducted in the region during the same time frame that also found multiple genetic subpopulations of parasites from the same region that were hypothesized to have originated as founder populations resulting from the emergence of artemisinin resistance (Miotto et al. 2013; 2015). The congruence of these findings suggests that the density clustering approach applied in our optimization framework accurately captured known patterns of parasite genetic diversity in the study area. The results from the first study found overlap between areas of high P. falciparum migration and hotspots of malaria incidence in eastern Cambodia, as well as other areas of high malaria endemicity. In Battambang and Pursat Provinces, high P. falciparum migration could result from flooding of the forests around Tonle Sap Lake during the wet season as discussed by Sawada et al. (2007), providing habitats for malaria vectors as suggested by Obsomer et al. (2007). In Pursat and Preah Vihear Provinces, high P. falciparum migration areas coincided with a high prevalence of multidrug resistance that is known to have emerged and spread in the area during this time frame. Migration maps generated in this study corroborated major migration barriers for P. falciparum identified in our previous study (Shetty et al. 2019). However, use of the optimized grid allowed detection of a migration barrier in Pailin Province that was not identified in our previous analysis, and that is consistent with malaria elimination efforts in this area that have contributed to a dramatic decline in clinical malaria incidence. P. falciparum 103 migration barriers in northern Cambodia may have been due to higher urbanization (lower vegetation coverage) north of Tonle Sap Lake as well as Tonle Sap Lake itself, which is a large enough waterbody that it may have served as a barrier to P. falciparum migration. Deforestation of cardamom forests and large-scale land acquisitions in the area corresponded to the southern part of the ring-like contour of low migration and may also have been a contributor to this migration barrier. The Mekong River running through southeastern Cambodia as well as the urbanized area of Phnom Penh, may both have contributed to reduced parasite migration in this part of Cambodia. The optimized grids produce migration maps that are more spatially explicit for a study region and that have reduced spatial uncertainty, an important consideration if maps are to be used to guide intervention strategies. While in this study, the optimization framework was applied to the malaria parasites, it is expected to be generalizable for other study areas and pathogens and can be used to generate optimized grids for more spatially diverse regions. In the second study (Chapter 3), an agent-based simulation model was built using a multi agent transport simulation (MATSim) framework to estimate and simulate the daily movements of local populations in Singu and Ann Townships in Myanmar in order to identify how two townships in different parts of Myanmar were similar or different with respect to mobility and P.v. and P.f. positivity. The simulation model allowed us to generate spatial patterns of mobility that were not available using only the case report forms. The key elements for building an ABM model of daily travel that captures mobility for local scales were identified and included travel time, travel mode, travel origins and destinations. The mobility patterns for different modes and 104 occupation groups were simulated and provided insights on the spatial extent of travel, as well as the estimated routes travelled by local village populations. These patterns were compared with the mobility patterns of P.v and P.f positive individuals. In terms of travel distance, forest workers were the most mobile occupation group in Ann Township, followed by farmers. However, the situation was different in Singu Township, where farmers were the most mobile group, and forest workers were the least mobile occupation group. Analysis of the simulated results used a hexagon-based approach to generate a continuous surface from which mobility paths could be extracted. This analysis showed that both forest workers and farmers in both townships shared the same most visited network segments (MVS) with P.v. positive individuals. In Ann Township, forest workers and farmers shared the similar MVS with P.f. positives, however, in Singu Township, travel by P.f. positive individuals shared more MVS with forest workers than any other occupation. As other researchers have found that occupation can be a strong predictor for understanding the transmission of malaria (e.g., Dunn, Le Mare, and Makungu 2011; Monroe et al. 2015), our findings reveal important details about the mobility characteristics of different occupation groups and how these vary across space. The third study (Chapter 4) examined mobility patterns of local village populations in Singu Township, Myanmar when they traveled longer distances outside of Singu. These longer trips represented regional travel and this type of travel was studied together with daily mobility patterns within the township. Regional travel 105 patterns (i.e., travel outside Singu) were extracted using degree centrality. The travel frequency between a village within Singu and a township outside was estimated using both the number of people who traveled between these two locations and their travel frequency as reported in the travel history survey. Local movement was simulated using the agent-based model built in the second study (Chapter 3). Malaria risk surfaces for P.f. and P.v. malaria were generated to account for the risk of malaria exposure for regional travelers using the average annual parasite rates between 2015-2019 as determined by the Malaria Atlas Project (Weiss et al. 2019; Battle et al. 2019).. Local malaria risk (i.e., within Singu Township) for P.f. and P.v. malaria was estimated using ecological niche modeling that is based on maximum entropy theory. A level-one-out cross validation process was applied to assess the importance of each of the environmental variables used in the model and the accuracy of this model for modeling exposure risk. Travel patterns for individuals who undertook regional trips as well as those who were P.v. and P.f. positive were also identified. The potential malaria transmission due to regional travel were quantified and potential malaria transmission hotspots were detected for P.v. and P.f. malaria showing regions estimated to be at most risk with respect to potential malaria transmission due to both local and regional travel in Singu. These hotspot regions may be sentinel sites for monitoring malaria in Singu, as these hotspots are more sensitive to malaria change than other regions in Singu. 106 5.2 Future work There are several topics that deserve further investigation. For example, future research could investigate how the spatial granularity of sampling may contribute to uncertainty in EEMS migration maps. The impact of how an irregular distribution of sample locations across a region may impact the robustness of EEMS migration patterns, as well as the impact of varying numbers of samples at deme locations could be investigated to help EEMS users interpret gene flow patterns across a study area. Also, further study of the accuracy of estimations near the edge of grids used in EEMS analyses could also be further studied and tested to better understand the role that boundary effects play in the Markov Chain Monte Carlo estimation. This dissertation discussed how travel to and from work on a road network could be simulated, while other kinds of travel such as trips using boats or motorcycles that travel off road could also be considered in future studies. In this study, only trips related to work were simulated, while other trips, including family and school-related trips, and social visits were not taken into consideration. Trips by women and children could also be a focus for future research. To have a detailed understanding of daily travels for individuals in a region with malaria, a future research direction could be to include more activities in the mobility simulation. As part of this dissertation research, ecological niche modeling using Maxent was used to estimate the malaria risk surface within Singu Township. Future research could extend this estimation to a larger region (beyond a single township) to understand the effectiveness of using Maxent to estimate the ecological niche for larger regions. Monte 107 Carlo simulation could be used to increase the information for Maxent estimation when sampling of P.f., for example, is low. In addition, collecting data on mobility with more detailed geographical information e.g., data collected using mobile devices, would benefit both a regional mobility simulation and the malaria risk determination for modeling mobility over different distances. 5.3 Implications of Research Outcomes for the Global Malaria Elimination Agenda The method developed in study 1 (Chapter 2) can be used in other study areas which have a malaria burden. Since spatial uncertainty relating to migration contours is reduced, parasite migration hotspots can be used as a reference to detect possible high malaria pandemic areas. And these possible high malaria pandemic areas are where the malaria elimination tools and control steps (e.g., malaria drugs) could be applied. For example, detected parasite migration hotspots could be investigated with higher priority than other regions. This detection process is especially important if there are limited medical resources available or there is a need to reduce the spread of the artemisinins-resistant malaria parasites. The frameworks designed for study 2 (Chapter 3) and 3 (Chapter 4) simulated daily movements as well as longer distance movements for surveyed individuals in Myanmar. The potential malaria transmission hotspots due to local and regional mobility can also be detected through the frameworks. These hotspots areas can serve as sentinel sites for monitoring malaria as these hotspots may be more sensitive to malaria change than other regions. These locations may also be useful for delivering healthcare services as they are places where local populations may have higher 108 potential of both exposure and transmission to malaria. This may contribute to improved healthcare services for malaria that are more efficient and effective. 109 Appendices Supplementary materials for Chapter 2: Supplementary Figure 1 Environmental features including protected areas, national parks, wildlife sanctuaries, and locations of water bodies in Cambodia (2013). 110 Supplementary Figure 2 Topological skeleton and boundary for cluster red 111 Supplementary Figure 3 Topological skeleton and boundary for cluster green 112 Supplementary Figure 4 Topological skeleton and boundary for cluster blue 113 Supplementary Figure 5 Topological skeleton and boundary for cluster cyan 114 Supplementary Figure 6 Topological skeleton and boundary for cluster purple 115 Supplementary materials for Chapter 3: Supplementary Figure 7 Travel frequency hotspots for forest workers in Ann Township 116 Supplementary Figure 8 Travel frequency hotspots for farmers in Ann Township 117 Supplementary Figure 9 Travel frequency hotspots for P.v. positive people in Ann Township 118 Supplementary Figure 10 Travel frequency hotspots for P.f. positive people in Ann Township 119 Supplementary Figure 11 Travel frequency hotspots for forest workers in Singu Township 120 Supplementary Figure 12 Travel frequency hotspots for farmers in Singu Township 121 Supplementary Figure 13 Travel frequency hotspots for miners in Singu Township 122 Supplementary Figure 14 Travel frequency hotspots for P.v. positive people in Singu Township 123 Supplementary Figure 15 Travel frequency hotspots for P.f. positive people in Singu Township 124 Bibliography Al-Asadi, Hussein, Desislava Petkova, Matthew Stephens, and John Novembre. 2019. ?Estimating Recent Migration and Population-Size Surfaces.? PLOS Genetics 15 (1): e1007908. https://doi.org/10.1371/journal.pgen.1007908. Alegana, Victor A., Emelda A. Okiro, and Robert W. Snow. 2020. ?Routine Data for Malaria Morbidity Estimation in Africa: Challenges and Prospects.? BMC Medicine 18: 1?13. Amato, Roberto, Pharath Lim, Olivo Miotto, Chanaki Amaratunga, Dalin Dek, Richard D Pearson, Jacob Almagro-Garcia, et al. 2017. ?Genetic Markers Associated with Dihydroartemisinin?Piperaquine Failure in Plasmodium Falciparum Malaria in Cambodia: A Genotype?Phenotype Association Study.? The Lancet Infectious Diseases 17 (2): 164?73. https://doi.org/10.1016/S1473-3099(16)30409-1. Antoniou, Aglaia, Panagiotis Kasapidis, Georgios Kotoulas, Constantinos C. Mylonas, and Antonios Magoulas. 2017. ?Genetic Diversity of Atlantic Bluefin Tuna in the Mediterranean Sea: Insights from Genome-Wide SNPs and Microsatellites.? Journal of Biological Research-Thessaloniki 24 (1): 3. Arnott, Richard, Andre De Palma, and Robin Lindsey. 1993. ?A Structural Model of Peak-Period Congestion: A Traffic Bottleneck with Elastic Demand.? The American Economic Review, 161?79. Ashley, Elizabeth A., Mehul Dhorda, Rick M. Fairhurst, Chanaki Amaratunga, Parath Lim, Seila Suon, Sokunthea Sreng, Jennifer M. Anderson, Sivanna Mao, and 125 Baramey Sam. 2014. ?Spread of Artemisinin Resistance in Plasmodium Falciparum Malaria.? New England Journal of Medicine 371 (5): 411?23. Autino, Beatrice, Alice Noris, Rosario Russo, and Francesco Castelli. 2012a. ?Epidemiology of Malaria in Endemic Areas.? Mediterranean Journal of Hematology and Infectious Diseases 4 (1): e2012060?e2012060. https://doi.org/10.4084/MJHID.2012.060. ???. 2012b. ?Epidemiology of Malaria in Endemic Areas.? Mediterranean Journal of Hematology and Infectious Diseases 4 (1). Baird, J. Kevin. 2017. ?Asia-Pacific Malaria Is Singular, Pervasive, Diverse and Invisible.? International Journal for Parasitology 47 (7): 371?77. Baker, Stephen, Kathryn E Holt, Archie C A Clements, Abhilasha Karkey, Amit Arjyal, Maciej F Boni, Sabina Dongol, et al. 2011. ?Combined High- Resolution Genotyping and Geospatial Analysis Reveals Modes of Endemic Urban Typhoid Fever Transmission.? Open Biology 1 (2): 110008?110008. https://doi.org/10.1098/rsob.110008. Battle, Katherine E., Tim CD Lucas, Michele Nguyen, Rosalind E. Howes, Anita K. Nandi, Katherine A. Twohig, Daniel A. Pfeffer, Ewan Cameron, Puja C. Rao, and Daniel Casey. 2019. ?Mapping the Global Endemicity and Clinical Burden of Plasmodium Vivax, 2000?17: A Spatial and Temporal Modelling Study.? The Lancet 394 (10195): 332?43. Berv, Jacob S., Leonardo Campagna, Teresa J. Feo, Ivandy Castro-Astor, Camila C. Ribas, Richard O. Prum, and Irby J. Lovette. 2019. ?Genomic Phylogeography of the White Crowned Manakin Pseudopipra Pipra (Aves: 126 Pipridae) Illuminates a Continental-Scale Radiation out of the Andes.? BioRxiv, 713081. Bethell, Delia, Youry Se, Chanthap Lon, Stuart Tyner, David Saunders, Sabaithip Sriwichai, Sea Darapiseth, Paktiya Teja-Isavadharm, Phisit Khemawoot, and Kurt Schaecher. 2011. ?Artesunate Dose Escalation for the Treatment of Uncomplicated Malaria in a Region of Reported Artemisinin Resistance: A Randomized Clinical Trial.? PloS One 6 (5): e19283. Bousema, Teun, Chris Drakeley, Samwel Gesase, Ramadhan Hashim, Stephen Magesa, Frank Mosha, Silas Otieno, Ilona Carneiro, Jonathan Cox, and Eliapendavyo Msuya. 2010. ?Identification of Hot Spots of Malaria Transmission for Targeted Malaria Control.? The Journal of Infectious Diseases 201 (11): 1764?74. Bousema, Teun, Jamie T. Griffin, Robert W. Sauerwein, David L. Smith, Thomas S. Churcher, Willem Takken, Azra Ghani, Chris Drakeley, and Roly Gosling. 2012. ?Hitting Hotspots: Spatial Targeting of Malaria for Control and Elimination.? PLoS Med 9 (1): e1001165. Brashear, Awtum M., Qi Fan, Yubing Hu, Yuling Li, Yan Zhao, Zenglei Wang, Yaming Cao, Jun Miao, Alyssa Barry, and Liwang Cui. 2020. ?Population Genomics Identifies a Distinct Plasmodium Vivax Population on the China- Myanmar Border of Southeast Asia.? PLoS Neglected Tropical Diseases 14 (8): e0008506. Brucato, Nicolas, Veronica Fernandes, St?phane Mazi?res, Pradiptajati Kusuma, Murray P. Cox, Joseph Wainaina Ng?ang?a, Mohammed Omar, Marie-Claude 127 Simeone-Senelle, Coralie Frassati, and Farida Alshamali. 2018. ?The Comoros Show the Earliest Austronesian Gene Flow into the Swahili Corridor.? The American Journal of Human Genetics 102 (1): 58?68. Carrel, Margaret A., Michael Emch, R. Todd Jobe, Aaron Moody, and Xiu-Feng Wan. 2010. ?Spatiotemporal Structure of Molecular Evolution of H5N1 Highly Pathogenic Avian Influenza Viruses in Vietnam.? PloS One 5 (1): e8631. Carrel, Margaret, Jaymin Patel, Steve M. Taylor, Mark Janko, Melchior Kashamuka Mwandagalirwa, Antoinette K. Tshefu, Ananias A. Escalante, et al. 2015. ?The Geography of Malaria Genetics in the Democratic Republic of Congo: A Complex and Fragmented Landscape.? Social Science & Medicine 133 (May): 233?41. https://doi.org/10.1016/j.socscimed.2014.10.037. Carrel, Margaret, Xiu-Feng Wan, Tung Nguyen, and Michael Emch. 2012. ?Highly Pathogenic H5N1 Avian Influenza Viruses Exhibit Few Barriers to Gene Flow in Vietnam.? EcoHealth 9 (1): 60?69. Carter, Richard, Kamini N. Mendis, and Donald Roberts. 2000. ?Spatial Targeting of Interventions against Malaria.? Bulletin of the World Health Organization 78: 1401?11. Chan, Kin Onn, and Rafe M. Brown. 2019. ?Linking Patterns of Genetic Variation to Processes of Diversification in Malaysian Torrent Frogs (Anura: Ranidae: Amolops): A Landscape Genomics Approach.? BioRxiv, 628891. Chang, Hsiao-Han, Amy Wesolowski, Ipsita Sinha, Christopher G Jacob, Ayesha Mahmud, Didar Uddin, Sazid Ibna Zaman, Md Amir Hossain, M Abul Faiz, 128 Aniruddha Ghose, et al. 2019. ?Mapping Imported Malaria in Bangladesh Using Parasite Genetic and Human Mobility Data.? ELife 8 (April): e43481. https://doi.org/10.7554/eLife.43481. Chang, Hsiao-Han, Amy Wesolowski, Ipsita Sinha, Christopher G. Jacob, Ayesha Mahmud, Didar Uddin, Sazid Ibna Zaman, Md Amir Hossain, M. Abul Faiz, and Aniruddha Ghose. 2019. ?Mapping Imported Malaria in Bangladesh Using Parasite Genetic and Human Mobility Data.? Elife 8: e43481. Chaorattanakawee, Suwanna, David L. Saunders, Darapiseth Sea, Nitima Chanarat, Kritsanai Yingyuen, Siratchana Sundrakes, Piyaporn Saingam, Nillawan Buathong, Sabaithip Sriwichai, and Soklyda Chann. 2015. ?Ex Vivo Drug Susceptibility and Molecular Profiling of Clinical Plasmodium Falciparum Isolates from Cambodia in 2008-2013 Suggest Emerging Piperaquine Resistance.? Antimicrobial Agents and Chemotherapy, AAC. 00366-15. Charypar, David, and Kai Nagel. 2005. ?Generating Complete All-Day Activity Plans with Genetic Algorithms.? Transportation 32 (4): 369?97. Chen, Dong, Varada Shevade, Allison Baer, Jiaying He, Amanda Hoffman-Hall, Qing Ying, Yao Li, and Tatiana V. Loboda. 2021. ?A Disease Control- Oriented Land Cover Land Use Map for Myanmar.? Data 6 (6). https://doi.org/10.3390/data6060063. Chitnis, Nakul, Allan Schapira, Thomas Smith, and Richard Steketee. 2010. ?Comparing the Effectiveness of Malaria Vector-Control Interventions through a Mathematical Model.? The American Journal of Tropical Medicine and Hygiene 83 (2): 230?40. 129 Cibulskis, Richard E, Pedro Alonso, John Aponte, Maru Aregawi, Amy Barrette, Laurent Bergeron, Cristin A Fergus, et al. 2016. ?Malaria: Global Progress 2000 - 2015 and Future Challenges.? Infectious Diseases of Poverty 5 (1): 61? 61. https://doi.org/10.1186/s40249-016-0151-8. Codella, James, Nasia Safdar, Rick Heffernan, and Oguzhan Alagoz. 2015. ?An Agent-Based Simulation Model for Clostridium Difficile Infection Control.? Medical Decision Making 35 (2): 211?29. Cook, Jackie, Nico Speybroeck, Tho Sochanta, Heng Somony, Mao Sokny, Filip Claes, Kristel Lemmens, et al. 2012. ?Sero-Epidemiological Evaluation of Changes in Plasmodium Falciparum and Plasmodium Vivax Transmission Patterns over the Rainy Season in Cambodia.? Malaria Journal 11 (1): 86. Cosner, Chris, John C. Beier, Robert Stephen Cantrell, D. Impoinvil, Lev Kapitanski, Matthew David Potts, A. Troyo, and Shigui Ruan. 2009. ?The Effects of Human Movement on the Persistence of Vector-Borne Diseases.? Journal of Theoretical Biology 258 (4): 550?60. Cui, Liwang, Guiyun Yan, Jetsumon Sattabongkot, Yaming Cao, Bin Chen, Xiaoguang Chen, Qi Fan, et al. 2012. ?Malaria in the Greater Mekong Subregion: Heterogeneity and Complexity.? Tackling The Malaria ?End Game?: Regional Needs And Challenges For Successful Malaria Elimination 121 (3): 227?39. https://doi.org/10.1016/j.actatropica.2011.02.016. Curran, Paul J., Peter M. Atkinson, Giles M. Foody, and Edward J. Milton. 2000. ?Linking Remote Sensing, Land Cover and Disease.? Advances in Parasitology 47: 37?80. 130 Davis, Kyle Frankel, Kailiang Yu, Maria Cristina Rulli, Lonn Pichdara, and Paolo D?Odorico. 2015. ?Accelerated Deforestation Driven by Large-Scale Land Acquisitions in Cambodia.? Nature Geoscience 8 (10): 772. Dejon-Agobe, Jean Claude, Ulysse Ateba-Ngoa, Albert Lalremruata, Andreas Homoet, Julie Engelhorn, Odilon Paterne Nouatin, Jean Ronald Edoa, Jos? F. Fernandes, Meral Esen, and Yoanne Darelle Mouwenda. 2019. ?Controlled Human Malaria Infection of Healthy Adults with Lifelong Malaria Exposure to Assess Safety, Immunogenicity, and Efficacy of the Asexual Blood Stage Malaria Vaccine Candidate GMZ2.? Clinical Infectious Diseases 69 (8): 1377?84. Djurdjevac Conrad, Nata?a, Luzie Helfmann, Johannes Zonker, Stefanie Winkelmann, and Christof Sch?tte. 2018. ?Human Mobility and Innovation Spreading in Ancient Times: A Stochastic Agent-Based Simulation Approach.? EPJ Data Science 7: 1?22. Dondorp, Arjen M., Fran?ois Nosten, Poravuth Yi, Debashish Das, Aung Phae Phyo, Joel Tarning, Khin Maung Lwin, Frederic Ariey, Warunee Hanpithakpong, and Sue J. Lee. 2009a. ?Artemisinin Resistance in Plasmodium Falciparum Malaria.? New England Journal of Medicine 361 (5): 455?67. ???. 2009b. ?Artemisinin Resistance in Plasmodium Falciparum Malaria.? New England Journal of Medicine 361 (5): 455?67. Dunn, Christine E., Ann Le Mare, and Christina Makungu. 2011. ?Malaria Risk Behaviours, Socio-Cultural Practices and Rural Livelihoods in Southern 131 Tanzania: Implications for Bednet Usage.? Social Science & Medicine 72 (3): 408?17. Duru, Valentine, Nimol Khim, Rithea Leang, Saorin Kim, Anais Domergue, Nimol Kloeung, Sopheakvatey Ke, Sophy Chy, Rotha Eam, and Chanra Khean. 2015. ?Plasmodium Falciparum Dihydroartemisinin-Piperaquine Failures in Cambodia Are Associated with Mutant K13 Parasites Presenting High Survival Rates in Novel Piperaquine in Vitro Assays: Retrospective and Prospective Investigations.? BMC Medicine 13 (1): 305. Dye, Christopher, and G?nther Hasibeder. 1986. ?Population Dynamics of Mosquito- Borne Disease: Effects of Flies Which Bite Some People More Frequently than Others.? Transactions of the Royal Society of Tropical Medicine and Hygiene 80 (1): 69?77. E. Mbunge, M. N. Sibiya, R. C. Millham, and S. Takavarasha. 2021. ?Micro-Spatial Modelling of Malaria Cases and Environmental Risk Factors in Buhera Rural District, Zimbabwe.? In 2021 Conference on Information Communications Technology and Society (ICTAS), 52?58. https://doi.org/10.1109/ICTAS50802.2021.9394987. Ehlkes, Lutz, Anne Caroline Krefis, Benno Kreuels, Ralf Krumkamp, Ohene Adjei, Matilda Ayim-Akonor, Robin Kobbe, et al. 2014. ?Geographically Weighted Regression of Land Cover Determinants of Plasmodium Falciparum Transmission in the Ashanti Region of Ghana.? International Journal of Health Geographics 13 (1): 35. https://doi.org/10.1186/1476-072X-13-35. 132 Elith, Jane, Steven J. Phillips, Trevor Hastie, Miroslav Dud?k, Yung En Chee, and Colin J. Yates. 2011. ?A Statistical Explanation of MaxEnt for Ecologists.? Diversity and Distributions 17 (1): 43?57. Enanoria, Wayne TA, Fengchen Liu, Jennifer Zipprich, Kathleen Harriman, Sarah Ackley, Seth Blumberg, Lee Worden, and Travis C. Porco. 2016. ?The Effect of Contact Investigations and Public Health Interventions in the Control and Prevention of Measles Transmission: A Simulation Study.? PloS One 11 (12): e0167160. Ferrao, Joao L., Sergio Niquisse, Jorge M. Mendes, and Marco Painho. 2018. ?Mapping and Modelling Malaria Risk Areas Using Climate, Socio- Demographic and Clinical Variables in Chimoio, Mozambique.? International Journal of Environmental Research and Public Health 15 (4). https://doi.org/10.3390/ijerph15040795. Forattini, Oswaldo Paulo. 1987. ?Exophilic Behavior of Anopheles Darlingi Root in a Southern Region of Brazil.? Revista de Sa?de P?blica 21: 291?304. Geng, Jinting, Pallavi Malla, Jiaqi Zhang, Shiling Xu, Cuiying Li, Yan Zhao, Qinghui Wang, Myat Phone Kyaw, Yaming Cao, and Zhaoqing Yang. 2019. ?Increasing Trends of Malaria in a Border Area of the Greater Mekong Subregion.? Malaria Journal 18 (1): 1?10. Gething, Peter W., Anand P. Patil, David L. Smith, Carlos A. Guerra, Iqbal RF Elyazar, Geoffrey L. Johnston, Andrew J. Tatem, and Simon I. Hay. 2011. ?A New World Malaria Map: Plasmodium Falciparum Endemicity in 2010.? Malaria Journal 10 (1): 378. 133 Gopalan, Shyamalika, Richard EW Berl, Gillian Belbin, Christopher Gignoux, Marcus W. Feldman, Barry S. Hewlett, and Brenna M. Henn. 2019. ?Hunter- Gatherer Genomes Reveal Diverse Demographic Trajectories Following the Rise of Farming in East Africa.? BioRxiv, 517730. Grewar, J. D. 2015. ?USING HEXAGON GRIDS FOR DATA COLLATION IN GEOGRAPHIC INFORMATION SYSTEMS.? CONTINUING EDUCATION PRESENTERS. Gwitira, Isaiah, Munashe Mukonoweshuro, Grace Mapako, Munyaradzi D. Shekede, Joconiah Chirenda, and Joseph Mberikunashe. 2020. ?Spatial and Spatio- Temporal Analysis of Malaria Cases in Zimbabwe.? Infectious Diseases of Poverty 9 (1): 1?14. Hahn, Micah B., Ronald E. Gangnon, Christovam Barcellos, Gregory P. Asner, and Jonathan A. Patz. 2014. ?Influence of Deforestation, Logging, and Fire on Malaria in the Brazilian Amazon.? PLOS ONE 9 (1): e85725. https://doi.org/10.1371/journal.pone.0085725. Hamilton, William L., Roberto Amato, Rob W. van der Pluijm, Christopher G. Jacob, Huynh Hong Quang, Nguyen Thanh Thuy-Nhien, Tran Tinh Hien, Bouasy Hongvanthong, Keobouphaphone Chindavongsa, and Mayfong Mayxay. 2019. ?Evolution and Expansion of Multidrug-Resistant Malaria in Southeast Asia: A Genomic Epidemiology Study.? The Lancet Infectious Diseases 19 (9): 943?51. Harris, Daniel N., Wei Song, Amol C. Shetty, Kelly S. Levano, Omar C?ceres, Carlos Padilla, V?ctor Borda, David Tarazona, Omar Trujillo, and Cesar Sanchez. 134 2018. ?Evolutionary Genomic Dynamics of Peruvians before, during, and after the Inca Empire.? Proceedings of the National Academy of Sciences, 201720798. Hast, Marisa, Kelly M. Searle, Mike Chaponda, James Lupiya, Jailos Lubinda, Jay Sikalima, Tamaki Kobayashi, Timothy Shields, Modest Mulenga, and Justin Lessler. 2019. ?The Use of GPS Data Loggers to Describe the Impact of Spatio-Temporal Movement Patterns on Malaria Control in a High- Transmission Area of Northern Zambia.? International Journal of Health Geographics 18 (1): 1?18. Hien, Tran Tinh, Nguyen Thanh Thuy-Nhien, Nguyen Hoan Phu, Maciej F. Boni, Ngo Viet Thanh, Nguyen Thuy Nha-Ca, Cao Quang Thai, Pham Van Toi, Phung Duc Thuan, and Laura Merson. 2012. ?In Vivo Susceptibility of Plasmodium Falciparum to Artesunate in Binh Phuoc Province, Vietnam.? Malaria Journal 11 (1): 355. Horni, Andreas, Kai Nagel, and Kay W. Axhausen. 2016. The Multi-Agent Transport Simulation MATSim. Ubiquity Press London. Hoyer, Stefan, Sokomar Nguon, Saorin Kim, Najibullah Habib, Nimol Khim, Sarorn Sum, Eva-Maria Christophel, et al. 2012. ?Focused Screening and Treatment (FSAT): A PCR-Based Strategy to Detect Malaria Parasite Carriers and Contain Drug Resistant P. Falciparum, Pailin, Cambodia.? PLOS ONE 7 (10): e45797. https://doi.org/10.1371/journal.pone.0045797. ?Https://Github.Com/Dipetkov/Eems.? 2018a. March 21, 2018. ????.? 2018b. March 21, 2018. 135 Imwong, Mallika, Thuy Nhien Nguyen, Rupam Tripura, Tom J. Peto, Sue J. Lee, Khin Maung Lwin, Preyanan Suangkanarat, et al. 2015. ?The Epidemiology of Subclinical Malaria Infections in South-East Asia: Findings from Cross- Sectional Surveys in Thailand?Myanmar Border Areas, Cambodia, and Vietnam.? Malaria Journal 14 (1): 381. Imwong, Mallika, Kanokon Suwannasin, Chanon Kunasol, Kreepol Sutawong, Mayfong Mayxay, Huy Rekol, Frank M Smithuis, et al. 2017. ?The Spread of Artemisinin-Resistant Plasmodium Falciparum in the Greater Mekong Subregion: A Molecular Epidemiology Observational Study.? The Lancet. Infectious Diseases 17 (5): 491?97. https://doi.org/10.1016/S1473- 3099(17)30048-8. Jacob, Christopher G., John C. Tan, Becky A. Miller, Asako Tan, Shannon Takala- Harrison, Michael T. Ferdig, and Christopher V. Plowe. 2014. ?A Microarray Platform and Novel SNP Calling Algorithm to Evaluate Plasmodium Falciparum Field Samples of Low DNA Quantity.? BMC Genomics 15 (1): 719. Jaynes, Edwin T. 1957. ?Information Theory and Statistical Mechanics.? Physical Review 106 (4): 620. Kazembe, Lawrence N., Immo Kleinschmidt, Timothy H. Holtz, and Brian L. Sharp. 2006. ?Spatial Analysis and Mapping of Malaria Risk in Malawi Using Point- Referenced Prevalence of Infection Data.? International Journal of Health Geographics 5 (1): 41. https://doi.org/10.1186/1476-072X-5-41. 136 Kumar, Sahil, T. R. Bhardwaj, D. N. Prasad, and Rajesh K. Singh. 2018. ?Drug Targets for Resistant Malaria: Historic to Future Perspectives.? Biomedicine & Pharmacotherapy 104: 8?27. Lam, Tommy Tsan-Yuk, Hon S. Ip, Elodie Ghedin, David E. Wentworth, Rebecca A. Halpin, Timothy B. Stockwell, David J. Spiro, Robert J. Dusek, James B. Bortner, and Jenny Hoskins. 2012. ?Migratory Flyway and Geographical Distance Are Barriers to the Gene Flow of Influenza Virus among North American Birds.? Ecology Letters 15 (1): 24?33. Leang, Rithea, Walter RJ Taylor, Denis Mey Bouth, Lijiang Song, Joel Tarning, Meng Chuor Char, Saorin Kim, Benoit Witkowski, Valentine Duru, and Anais Domergue. 2015. ?Evidence of Plasmodium Falciparum Malaria Multidrug Resistance to Artemisinin and Piperaquine in Western Cambodia: Dihydroartemisinin-Piperaquine Open-Label Multicenter Clinical Assessment.? Antimicrobial Agents and Chemotherapy 59 (8): 4719?26. Lee, Bruce Y., Shawn T. Brown, George W. Korch, Philip C. Cooley, Richard K. Zimmerman, William D. Wheaton, Shanta M. Zimmer, John J. Grefenstette, Rachel R. Bailey, and Tina-Marie Assi. 2010. ?A Computer Simulation of Vaccine Prioritization, Allocation, and Rationing during the 2009 H1N1 Influenza Pandemic.? Vaccine 28 (31): 4875?79. Li, Nana, Daniel M. Parker, Zhaoqing Yang, Qi Fan, Guofa Zhou, Guoping Ai, Jianhua Duan, Ming-chieh Lee, Guiyun Yan, and Stephen A. Matthews. 2013. ?Risk Factors Associated with Slide Positivity among Febrile Patients in a 137 Conflict Zone of North-Eastern Myanmar along the China-Myanmar Border.? Malaria Journal 12 (1): 1?8. Li, Yao, Amol C. Shetty, Chanthap Lon, Michele Spring, David L. Saunders, Mark M. Fukuda, Tran Tinh Hien, Sasithon Pukrittayakamee, Rick M. Fairhurst, and Arjen M. Dondorp. 2020. ?Detecting Geospatial Patterns of Plasmodium Falciparum Parasite Migration in Cambodia Using Optimized Estimated Effective Migration Surfaces.? International Journal of Health Geographics 19: 1?11. Li, Yingchang, Mingyang Li, Chao Li, and Zhenzhen Liu. 2020. ?Optimized Maxent Model Predictions of Climate Change Impacts on the Suitable Distribution of Cunninghamia Lanceolata in China.? Forests 11 (3): 302. Liang, Xiao, Jichang Zhao, Li Dong, and Ke Xu. 2013. ?Unraveling the Origin of Exponential Law in Intra-Urban Human Mobility.? Scientific Reports 3 (1): 1?7. Liu, Canran, Matt White, and Graeme Newell. 2013. ?Selecting Thresholds for the Prediction of Species Occurrence with Presence?only Data.? Journal of Biogeography 40 (4): 778?89. Lynch, Caroline, and Cally Roper. 2011. ?The Transit Phase of Migration: Circulation of Malaria and Its Multidrug-Resistant Forms in Africa.? PLoS Medicine 8 (5). MalariaGEN Plasmodium falciparum Community Project. 2016. ?Genomic Epidemiology of Artemisinin Resistant Malaria.? Elife 5: e08714. 138 Martin, Alicia R., Konrad J. Karczewski, Sini Kerminen, Mitja I. Kurki, Antti-Pekka Sarin, Mykyta Artomov, Johan G. Eriksson, T?nu Esko, Giulio Genovese, and Aki S. Havulinna. 2018. ?Haplotype Sharing Provides Insights into Fine-Scale Population History and Disease in Finland.? The American Journal of Human Genetics 102 (5): 760?75. Mathieson, Iain, Song?l Alpaslan-Roodenberg, Cosimo Posth, Anna Sz?cs?nyi-Nagy, Nadin Rohland, Swapan Mallick, I?igo Olalde, Nasreen Broomandkhoshbacht, Francesca Candilio, and Olivia Cheronet. 2018. ?The Genomic History of Southeastern Europe.? Nature 555 (7695): 197. Maude, Richard J., Chea Nguon, Po Ly, Tol Bunkea, Pengby Ngor, Sara E. Canavati de la Torre, Nicholas J. White, et al. 2014. ?Spatial and Temporal Epidemiology of Clinical Malaria in Cambodia 2004?2013.? Malaria Journal 13 (1): 385. https://doi.org/10.1186/1475-2875-13-385. Milne, Sarah. 2012. ?Grounding Forest Carbon: Property Relations and Avoided Deforestation in Cambodia.? Human Ecology 40 (5): 693?706. Ministry of Immigration and Population. 2015. ?The 2014 Myanmar Population and Housing Census.? Ministry of Immigration and Population Nay Pyi Taw. Miotto, Olivo, Jacob Almagro-Garcia, Magnus Manske, Bronwyn MacInnis, Susana Campino, Kirk A. Rockett, Chanaki Amaratunga, Pharath Lim, Seila Suon, and Sokunthea Sreng. 2013. ?Multiple Populations of Artemisinin-Resistant Plasmodium Falciparum in Cambodia.? Nature Genetics 45 (6): 648. Miotto, Olivo, Roberto Amato, Elizabeth A Ashley, Bronwyn MacInnis, Jacob Almagro-Garcia, Chanaki Amaratunga, Pharath Lim, et al. 2015. ?Genetic 139 Architecture of Artemisinin-Resistant Plasmodium Falciparum.? Nature Genetics 47 (3): 226?34. https://doi.org/10.1038/ng.3189. Monroe, April, Obed Asamoah, Yukyan Lam, Hannah Koenker, Paul Psychas, Matthew Lynch, Emily Ricotta, Sureyya Hornston, Amanda Berman, and Steven A. Harvey. 2015. ?Outdoor-Sleeping and Other Night-Time Activities in Northern Ghana: Implications for Residual Transmission and Malaria Prevention.? Malaria Journal 14 (1): 1?12. Mukhtar, Abdulaziz YA, Justin B. Munyakazi, and Rachid Ouifki. 2020. ?Assessing the Role of Human Mobility on Malaria Transmission.? Mathematical Biosciences 320: 108304. Newby, Gretchen, Adam Bennett, Erika Larson, Chris Cotter, Rima Shretta, Allison A. Phillips, and Richard GA Feachem. 2016. ?The Path to Eradication: A Progress Report on the Malaria-Eliminating Countries.? The Lancet 387 (10029): 1775?84. Obsomer, Val?rie, Pierre Defourny, and Marc Coosemans. 2007. ?The Anopheles Dirus Complex: Spatial Distribution and Environmental Drivers.? Malaria Journal 6 (1): 26. OpenDevelopment Cambodia. 2016. ?OpenDevelopment Cambodia.? 2016. Organization, World Health. 2010. ?Malaria in the Greater Mekong Subregion: Regional and Country Profiles.? Ostfeld, Richard S., Gregory E. Glass, and Felicia Keesing. 2005. ?Recent Developments in Remote Sensing for Environmental Monitoring.? Trends in 140 Ecology & Evolution 20 (6): 328?36. https://doi.org/10.1016/j.tree.2005.03.009. Parker, Daniel M., Rupam Tripura, Thomas J. Peto, Richard J. Maude, Chea Nguon, Jeremy Chalk, Pasathorn Sirithiranont, Mallika Imwong, Lorenz von Seidlein, and Nicholas J. White. 2017. ?A Multi-Level Spatial Analysis of Clinical Malaria and Subclinical Plasmodium Infections in Pailin Province, Cambodia.? Heliyon 3 (11): e00447. Petkova, Desislava, John Novembre, and Matthew Stephens. 2016. ?Visualizing Spatial Population Structure with Estimated Effective Migration Surfaces.? Nature Genetics 48 (1): 94?100. Phillips, Steven J. 2005. ?A Brief Tutorial on Maxent.? AT&T Research 190 (4): 231?59. Phillips, Steven J., Robert P. Anderson, Miroslav Dud?k, Robert E. Schapire, and Mary E. Blair. 2017. ?Opening the Black Box: An Open?source Release of Maxent.? Ecography 40 (7): 887?93. Phillips, Steven J., and Miroslav Dud?k. 2008. ?Modeling of Species Distributions with Maxent: New Extensions and a Comprehensive Evaluation.? Ecography 31 (2): 161?75. Pimenta, Jo?o, Alexandra M. Lopes, Angel Carracedo, Miguel Arenas, Ant?nio Amorim, and David Comas. 2019. ?Spatially Explicit Analysis Reveals Complex Human Genetic Gradients in the Iberian Peninsula.? Scientific Reports 9 (1): 7825. 141 Pindolia, Deepa K., Andres J. Garcia, Amy Wesolowski, David L. Smith, Caroline O. Buckee, Abdisalan M. Noor, Robert W. Snow, and Andrew J. Tatem. 2012. ?Human Movement Data for Malaria Control and Elimination Strategic Planning.? Malaria Journal 11 (1): 1?16. Pizzitutti, Francesco, William Pan, Beth Feingold, Ben Zaitchik, Carlos A. ?lvarez, and Carlos F. Mena. 2018. ?Out of the Net: An Agent-Based Model to Study Human Movements Influence on Local-Scale Malaria Transmission.? PLOS ONE 13 (3): e0193493. https://doi.org/10.1371/journal.pone.0193493. Price, Ric N., Robert J. Commons, Katherine E. Battle, Kamala Thriemer, and Kamini Mendis. 2020. ?Plasmodium Vivax in the Era of the Shrinking P. Falciparum Map.? Trends in Parasitology. Prosper, Olivia, Nick Ruktanonchai, and Maia Martcheva. 2012. ?Assessing the Role of Spatial Heterogeneity and Human Movement in Malaria Dynamics and Control.? Journal of Theoretical Biology 303: 1?14. Raghwani, Jayna, Andrew Rambaut, Edward C. Holmes, Vu Ty Hang, Tran Tinh Hien, Jeremy Farrar, Bridget Wills, et al. 2011. ?Endemic Dengue Associated with the Co-Circulation of Multiple Viral Lineages and Localized Density- Dependent Transmission.? PLOS Pathogens 7 (6): e1002064. https://doi.org/10.1371/journal.ppat.1002064. Richmond, Jonathan Q., Dustin A. Wood, Michael F. Westphal, Amy G. Vandergast, Adam D. Leach?, Lawrence R. Saslaw, H. Scott Butterfield, and Robert N. Fisher. 2017. ?Persistence of Historical Population Structure in an Endangered 142 Species despite Near?complete Biome Conversion in California?s San Joaquin Desert.? Molecular Ecology. Rick, Jessica A., Ron A. Moen, John D. Erb, and Jared L. Strasburg. 2017. ?Population Structure and Gene Flow in a Newly Harvested Gray Wolf (Canis Lupus) Population.? Conservation Genetics, 1?14. Ringbauer, Harald, Alexander Kolesnikov, David L. Field, and Nicholas H. Barton. 2018. ?Estimating Barriers to Gene Flow from Distorted Isolation by Distance Patterns.? Genetics, genetics. 300638.2017. Rodriguez, Alex, and Alessandro Laio. 2014. ?Clustering by Fast Search and Find of Density Peaks.? Science 344 (6191): 1492?96. Ruktanonchai, Nick W., David L. Smith, and Patrick De Leenheer. 2016. ?Parasite Sources and Sinks in a Patched Ross?Macdonald Malaria Model with Human and Mosquito Movement: Implications for Control.? Mathematical Biosciences 279: 90?101. Saatchi, Sassan, Wolfgang Buermann, Hans Ter Steege, Scott Mori, and Thomas B. Smith. 2008. ?Modeling Distribution of Amazonian Tree Species and Diversity Using Remote Sensing Measurements.? Remote Sensing of Environment 112 (5): 2000?2017. Saita, Sayambhu, Wirichada Pan-Ngum, Suparat Phuanukoonnon, Patchara Sriwichai, Tassanee Silawan, Lisa J. White, and Daniel M. Parker. 2019. ?Human Population Movement and Behavioural Patterns in Malaria Hotspots on the Thai?Myanmar Border: Implications for Malaria Elimination.? Malaria Journal 18 (1): 1?14. 143 Sawada, Haruo, Makoto Araki, Nick A. Chappell, James V. LaFrankie, and Akira Shimizu. 2007. Forest Environments in the Mekong River Basin. Vol. 169. Springer. Schneider, Christian M., Vitaly Belik, Thomas Couronn?, Zbigniew Smoreda, and Marta C. Gonz?lez. 2013. ?Unravelling Daily Human Mobility Motifs.? Journal of The Royal Society Interface 10 (84): 20130246. Shannon, Claude E. 1948. ?A Mathematical Theory of Communication.? The Bell System Technical Journal 27 (3): 379?423. Shannon, Claude Elwood. 2001. ?A Mathematical Theory of Communication.? ACM SIGMOBILE Mobile Computing and Communications Review 5 (1): 3?55. Sherpa, St?phanie, Michael GB Blum, Thibaut Capblancq, Tristan Cumer, Delphine Rioux, and Laurence Despr?s. 2019. ?Unravelling the Invasion History of the Asian Tiger Mosquito in Europe.? Molecular Ecology 28 (9): 2360?77. Shetty, Amol C., Christopher G. Jacob, Fang Huang, Yao Li, Sonia Agrawal, David L. Saunders, Chanthap Lon, et al. 2019. ?Genomic Structure and Diversity of Plasmodium Falciparum in Southeast Asia Reveal Recent Parasite Migration Patterns.? Nature Communications 10 (1): 2665. https://doi.org/10.1038/s41467-019-10121-3. Silal, Sheetal Prakash, Francesca Little, Karen Irma Barnes, and Lisa Jane White. 2015. ?Hitting a Moving Target: A Model for Malaria Elimination in the Presence of Population Movement.? PloS One 10 (12): e0144990. Sinha, Ipsita, Abdullah Abu Sayeed, Didar Uddin, Amy Wesolowski, Sazid Ibna Zaman, M. Abul Faiz, Aniruddha Ghose, M. Ridwanur Rahman, Akramul 144 Islam, and Mohammad Jahirul Karim. 2020. ?Mapping the Travel Patterns of People with Malaria in Bangladesh.? BMC Medicine 18 (1): 1?17. Sluydts, Vincent, Somony Heng, Marc Coosemans, Karel Van Roey, Charlotte Gryseels, Lydie Canier, Saorin Kim, Nimol Khim, Sovannaroth Siv, and Vanna Mean. 2014. ?Spatial Clustering and Risk Factors of Malaria Infections in Ratanakiri Province, Cambodia.? Malaria Journal 13 (1): 387. Smith, David L., F. Ellis McKenzie, Robert W. Snow, and Simon I. Hay. 2007. ?Revisiting the Basic Reproductive Number for Malaria and Its Implications for Malaria Control.? PLoS Biology 5 (3): e42. Smith, David L., T. Alex Perkins, Robert C. Reiner Jr, Christopher M. Barker, Tianchan Niu, Luis Fernando Chaves, Alicia M. Ellis, Dylan B. George, Arnaud Le Menach, and Juliet RC Pulliam. 2014. ?Recasting the Theory of Mosquito-Borne Pathogen Transmission Dynamics and Control.? Transactions of the Royal Society of Tropical Medicine and Hygiene 108 (4): 185?97. Smith, Neal R., James M. Trauer, Manoj Gambhir, Jack S. Richards, Richard J. Maude, Jonathan M. Keith, and Jennifer A. Flegg. 2018. ?Agent-Based Models of Malaria Transmission: A Systematic Review.? Malaria Journal 17 (1): 1?16. Song, Ruiqi, Ying Ma, Zhengxiang Hu, Yingke Li, Min Li, Lijiang Wu, Caishan Li, Erjiala Dao, Xinli Fan, and Yunwei Hao. 2020. ?MaxEnt Modeling of Dermacentor Marginatus (Acari: Ixodidae) Distribution in Xinjiang, China.? Journal of Medical Entomology 57 (5): 1659?67. 145 Spring, Michele D., Jessica T. Lin, Jessica E. Manning, Pattaraporn Vanachayangkul, Sok Somethy, Rathvicheth Bun, Youry Se, Soklyda Chann, Mali Ittiverakul, and Piyaporn Sia-ngam. 2015. ?Dihydroartemisinin-Piperaquine Failure Associated with a Triple Mutant Including Kelch13 C580Y in Cambodia: An Observational Cohort Study.? The Lancet Infectious Diseases 15 (6): 683?91. Srivastava, Aruna, BN Nagpal, PL Joshi, JC Paliwal, and AP Dash. 2009. ?Identification of Malaria Hot Spots for Focused Intervention in Tribal State of India: A GIS Based Approach.? International Journal of Health Geographics 8 (1): 30. https://doi.org/10.1186/1476-072X-8-30. Stanaway, Jeffrey D., Andrea Parisi, Kaushik Sarkar, Brigette F. Blacker, Robert C. Reiner, Simon I. Hay, Molly R. Nixon, Christiane Dolecek, Spencer L. James, and Ali H. Mokdad. 2019. ?The Global Burden of Non-Typhoidal Salmonella Invasive Disease: A Systematic Analysis for the Global Burden of Disease Study 2017.? The Lancet Infectious Diseases 19 (12): 1312?24. Steenkeste, Nicolas, William O Rogers, Lucy Okell, Isabelle Jeanne, Sandra Incardona, Linda Duval, Sophy Chy, et al. 2010. ?Sub-Microscopic Malaria Cases and Mixed Malaria Infection in a Remote Area of High Malaria Endemicity in Rattanakiri Province, Cambodia: Implication for Malaria Elimination.? Malaria Journal 9 (April): 108?108. https://doi.org/10.1186/1475-2875-9-108. Stoddard, Steven T., Amy C. Morrison, Gonzalo M. Vazquez-Prokopec, Valerie Paz Soldan, Tadeusz J. Kochel, Uriel Kitron, John P. Elder, and Thomas W. Scott. 146 2009. ?The Role of Human Movement in the Transmission of Vector-Borne Pathogens.? PLoS Neglected Tropical Diseases 3 (7): e481. Stresman, Gillian, Teun Bousema, and Jackie Cook. 2019. ?Malaria Hotspots: Is There Epidemiological Evidence for Fine-Scale Spatial Targeting of Interventions?? Trends in Parasitology 35 (10): 822?34. Sudre, Bertrand, Massimiliano Rossi, Wim Van Bortel, Kostas Danis, Agoritsa Baka, Nikos Vakalis, and Jan C Semenza. 2013. ?Mapping Environmental Suitability for Malaria Transmission, Greece.? Emerging Infectious Diseases 19 (5): 784?86. https://doi.org/10.3201/eid1905.120811. Takala-Harrison, Shannon, Christopher G. Jacob, Cesar Arze, Michael P. Cummings, Joana C. Silva, Arjen M. Dondorp, Mark M. Fukuda, Tran Tinh Hien, Mayfong Mayxay, and Harald Noedl. 2014. ?Independent Emergence of Artemisinin Resistance Mutations among Plasmodium Falciparum in Southeast Asia.? The Journal of Infectious Diseases 211 (5): 670?79. Tatem, Andrew J., and David L. Smith. 2010. ?International Population Movements and Regional Plasmodium Falciparum Malaria Elimination Strategies.? Proceedings of the National Academy of Sciences 107 (27): 12222?27. Tessema, Sofonias, Amy Wesolowski, Anna Chen, Maxwell Murphy, Jordan Wilheim, Anna-Rosa Mupiri, Nick W. Ruktanonchai, Victor A. Alegana, Andrew J. Tatem, and Munyaradzi Tambo. 2019. ?Using Parasite Genetic and Human Mobility Data to Infer Local and Cross-Border Malaria Connectivity in Southern Africa.? Elife 8: e43510. 147 Thomas Lin Pedersen. 2015. ?DensityClust.? 2015. https://github.com/thomasp85/densityClust. Tracy, Melissa, Magdalena Cerd?, and Katherine M. Keyes. 2018. ?Agent-Based Modeling in Public Health: Current Applications and Future Directions.? Annual Review of Public Health 39: 77?94. Tsuda, Yoshiaki, Jun Chen, Michael Stocks, Thomas K?llman, J?rn Henrik S?nsteb?, Laura Parducci, Vladimir Semerikov, Christoph Sperisen, Dmitry Politov, and Tiina Ronkainen. 2016. ?The Extent and Meaning of Hybridization and Introgression between Siberian Spruce (Picea Obovata) and Norway Spruce (Picea Abies): Cryptic Refugia as Stepping Stones to the West?? Molecular Ecology 25 (12): 2773?89. Tun, Kyaw M., Mallika Imwong, Khin M. Lwin, Aye A. Win, Tin M. Hlaing, Thaung Hlaing, Khin Lin, Myat P. Kyaw, Katherine Plewes, and M. Abul Faiz. 2015. ?Spread of Artemisinin-Resistant Plasmodium Falciparum in Myanmar: A Cross-Sectional Survey of the K13 Molecular Marker.? The Lancet Infectious Diseases 15 (4): 415?21. Turell, Michael J., Michael R. Sardelis, James W. Jones, Douglas M. Watts, Roberto Fernandez, Faustino Carbajal, James E. Pecor, and Terry A. Klein. 2008. ?Seasonal Distribution, Biology, and Human Attraction Patterns of Mosquitoes (Diptera: Culicidae) in a Rural Village and Adjacent Forested Site near Iquitos, Peru.? Journal of Medical Entomology 45 (6): 1165?72. Uren, Caitlin, Minju Kim, Alicia R. Martin, Dean Bobo, Christopher R. Gignoux, Paul D. van Helden, Marlo M?ller, Eileen G. Hoal, and Brenna M. Henn. 148 2016. ?Fine-Scale Human Population Structure in Southern Africa Reflects Ecogeographic Boundaries.? Genetics 204 (1): 303?14. Vickrey, William S. 1969. ?Congestion Theory and Transport Investment.? The American Economic Review 59 (2): 251?60. Wallace, Robert G., and Walter M. Fitch. 2008. ?Influenza A H5N1 Immigration Is Filtered out at Some International Borders.? PloS One 3 (2): e1697. Wangdi, Kinley, Sara E. Canavati, Thang Duc Ngo, Thu Minh Nguyen, Long Khanh Tran, Gerard C. Kelly, Nicholas J. Martin, and Archie CA Clements. 2020. ?Spatial and Temporal Patterns of Malaria in Phu Yen Province, Vietnam, from 2005 to 2016.? The American Journal of Tropical Medicine and Hygiene 103 (4): 1540?48. Weiss, Daniel J., Tim CD Lucas, Michele Nguyen, Anita K. Nandi, Donal Bisanzio, Katherine E. Battle, Ewan Cameron, Katherine A. Twohig, Daniel A. Pfeffer, and Jennifer A. Rozier. 2019. ?Mapping the Global Prevalence, Incidence, and Mortality of Plasmodium Falciparum, 2000?17: A Spatial and Temporal Modelling Study.? The Lancet 394 (10195): 322?31. Wesolowski, Amy, Nathan Eagle, Andrew J. Tatem, David L. Smith, Abdisalan M. Noor, Robert W. Snow, and Caroline O. Buckee. 2012. ?Quantifying the Impact of Human Mobility on Malaria.? Science 338 (6104): 267?70. West, Amanda M., Sunil Kumar, Cynthia S. Brown, Thomas J. Stohlgren, and Jim Bromberg. 2016. ?Field Validation of an Invasive Species Maxent Model.? Ecological Informatics 36: 126?34. 149 World Health Organization. 2015. ?Strategy for Malaria Elimination in the Greater Mekong Subregion (2015-2030).? 2015. ???. 2019. ?World Malaria Report 2019.? 2019. http://www.who.int/malaria/media/world-malaria-report-2018/en/. ???. 2020. ?World Malaria Report 2020: 20 Years of Global Progress and Challenges.? Yan, Huyong, Lei Feng, Yufei Zhao, Li Feng, Di Wu, and Chaoping Zhu. 2020. ?Prediction of the Spatial Distribution of Alternanthera Philoxeroides in China Based on ArcGIS and MaxEnt.? Global Ecology and Conservation 21: e00856. Yi, Yujun, Xi Cheng, Zhifeng Yang, Silke Wieprecht, Shanghong Zhang, and Yingjie Wu. 2017. ?Evaluating the Ecological Influence of Hydraulic Projects: A Review of Aquatic Habitat Suitability Models.? Renewable and Sustainable Energy Reviews 68: 748?62. Zhou, Guofa, Eugenia Lo, Daibin Zhong, Xiaoming Wang, Ying Wang, Sameer Malla, Ming-chieh Lee, Zhaoqing Yang, Liwang Cui, and Guiyun Yan. 2016. ?Impact of Interventions on Malaria in Internally Displaced Persons along the China?Myanmar Border: 2011?2014.? Malaria Journal 15 (1): 1?10. Ziemke, Dominik, Kai Nagel, and Chandra Bhat. 2015. ?Integrating CEMDAP and MATSIM to Increase the Transferability of Transport Demand Models.? Transportation Research Record 2493 (1): 117?25. 150