Modeling Migration and Global Population Patterns: Combining High-Resolution Household and Natural Hazard Data

This dissertation was carried out as part of the EU-funded COASTMOVE project, which aims to improve our understanding of migration and adaptation in response to sea-level rise. By combining hydrological and behavioral models, the project develops a global framework that can simulate, at the household level, how households respond to coastal flooding, coastal erosion, and saltwater intrusion. Do they migrate, adapt, or do nothing? Insights from this work provide important input for further policy development. This dissertation builds on existing statistical research into the relationship between natural hazards and migration and focuses on improving large-scale migration simulation models. A large body of research has employed statistical methods to analyze the relationship between natural hazards and migration. While these studies provide valuable insights, their findings are often mixed, reflecting the complex nature of migration processes as well as differences in data, spatial scale, and model assumptions. In this dissertation, statistical methods are further developed to refine our understanding of the relationship between natural hazards and migration. Chapter 2 estimates a gravity model of migration to assess the impact of different types of natural hazards on internal migration in the United States, using county-level migration data. The results indicate that hurricanes are associated with the largest increases in out-migration, followed by floods and severe storms. Chapter 3 builds on this analysis by examining how methodological choices affect estimated migration responses. Using the same data, we show that regression coefficients are overestimated when spatial correlation in the data is not accounted for. These findings are relevant from a policy perspective, as they imply that households may be less responsive to natural hazards than suggested by models that ignore spatial dependence. When households remain in high-risk areas, there is a need for policies that support adaptation measures or facilitate relocation to safer locations. In addition to statistical approaches, researchers increasingly use simulation-based methods, such as agent-based models, to study migration and adaptation decisions. These models allow migration decisions to be modeled at the household level, while accounting for household characteristics, learning behavior, and interactions between households, communities, and governments. However, most agent-based models are applied at relatively small spatial scales. Scaling up such models requires detailed microdata on household and individual characteristics, but these data are fragmented and often difficult to obtain at larger scales due to privacy constraints. To address this gap, Chapter 4 develops a global synthetic population dataset, GLOPOP-S, comprising nearly two billion households and over 7.3 billion individuals. The dataset includes key household and individual characteristics, such as settlement type (rural or urban), relative wealth or income, household size and type, age, gender, and education level. GLOPOP-S is constructed using data from the Luxembourg Income Study and the Demographic and Health Surveys. In Chapter 5, the synthetic households are allocated to grid cells at a 30 arc-second (approximately 1 km × 1 km) resolution using high-resolution spatial data on household characteristics. The allocation procedure consists of three steps: an initial allocation based on settlement type, followed by an allocation based on wealth data, and a final refinement using age and/or gender distributions. Besides its use in simulation models, the high-resolution gridded version of GLOPOP-S, referred to as GLOPOP-SG, makes it possible to identify households that are vulnerable to changing environmental conditions within regions and to design more targeted policy responses.