Dynamic Human‐Vehicle Risk Assessment for Urban Flood Evacuations in Island Cities: The MUFE Framework Applied to Haidian Island, Haikou

ABSTRACT Urban flood disasters demand dynamic assessment of population risk, yet most evacuation models fail to capture the multifaceted, time‐varying nature of such events. To address this gap, this study develops the modular urban flood‐evacuation (MUFE) framework, an integrated approach to dynamic urban flood risk assessment. MUFE explicitly models pedestrian and vehicle behaviors under inundation and their interactions with evolving hydrodynamic conditions. The framework is tailored for island cities facing compounded threats from storm surges and sea‐level rise, and is demonstrated on Haidian Island, Haikou, China. The research couples multi‐scenario rainfall‐tide hydrodynamic simulations (via InfoWorks ICM) with an agent‐based model implemented in the open‐source Mesa framework. Mesa enables the explicit representation of pedestrian and vehicle agents operating under perceive‐decide‐act cycles and local information constraints. This integration enables joint analysis of hydrodynamic hazard fields—shaped by dynamic tidal fluctuations—and the behavioral responses of pedestrians and vehicles. The framework is modular, allowing alternative hydrodynamic or data‐driven flood models to be coupled for transparent comparison. Simulation results show that pedestrian hazard exposure varies markedly across space and time, shaped by adaptive evacuation behavior, shelter availability, and tide dynamics. Small vehicles are more susceptible to instability—even under shallow depths—while larger vehicles are more resilient but not immune under severe conditions. These patterns underscore the need for differentiated evacuation planning and targeted infrastructure resilience measures in island cities.