Abstract
Understanding and predicting human mobility dynamics in response to external spatiotemporal events represents a fundamental societal challenge. These mobility phenomena operate at various spatial and temporal resolutions, resulting from complex interactions between external trigger events, individual decision-making, demographic heterogeneity, social network effects, and various uncertainties associated with human life. Existing approaches lack a unified framework aimed at capturing such mobility dynamics, particularly when mobility is triggered by rapid, unpredictable external shocks. This dissertation takes a step towards addressing this gap by developing a comprehensive computational framework for modeling event-triggered mobility dynamics in networked populations, with conflict-induced forced migration serving as a critical and challenging instantiation of this broader problem class.
Forced migration due to conflict events or violence remains one of the most pressing and complex humanitarian challenges. The Russian invasion of Ukraine in February 2022 is a recent example that has caused the largest forced migration crisis in Europe since World War II, with millions of people displaced both internally and internationally. Providing support to these affected people requires proper planning from the perspective of policymakers, which involves understanding the mobility of migrants at different stages of displacement. The urgency, scale, and complexity of such displacement make it an ideal testbed for developing robust computational methods that must handle incomplete data, behavioral uncertainty, and rapidly evolving conditions—challenges common to many sociotechnical systems facing external shocks.
This dissertation makes three primary contributions. The first contribution is the development of the STP-GDS framework, a generalization of classical Graph Dynamical Systems that integrates external spatiotemporal events into the dynamic evolution of networked agent-based systems. This formalism enables principled analysis of networked agent dynamics under exogenous triggers and establishes computational complexity results that justify simulation-based approaches. Guided by this insight, the second contribution introduces ABM-TPB, a social theory-guided ABM of forward migration, which simulates daily outflows from a conflict region using real conflict event data and synthetic population models. The model captures temporal, spatial and demographic trends and achieves strong agreement with observed refugee flows from Ukraine during the 2022 Russian invasion. Application of ABM-TPB to two other past conflict settings underscores its generalizability, and various case studies highlight the policy utility of the model. Furthermore, comprehensive validation through automated calibration pipelines and parameter space analysis establishes the model's robustness. Third, it extends the framework to return migration dynamics through ABM-HF, incorporating hazard functions that model contextual factors such as safety conditions, displacement duration, and peer influence. Validated against empirical return flows and generalized across multiple displacement contexts, these models support evidence-based reintegration policies.
Together, these contributions provide a methodological foundation for analyzing event-triggered mobility in networked populations that extends beyond conflict migration to broader applications in disaster response, climate-induced displacement, and socio-economic transitions. By unifying theoretical rigor, behavioral realism, and computational scalability, this work equips researchers and policymakers with tools to anticipate, analyze, and respond to human mobility dynamics under stress.
