Modeling hurricane evacuation/return under compound risks

Abstract

Disasters faced by human society are becoming more frequent and complex, raising a need to model the combinations of different types of disasters, such as hurricanes and pandemics. In this paper, we explore various modeling options for predicting aggregated individual evacuation metrics under the compound risks drawn by COVID-19 and Hurricane Ida (2021) using large-scale location-based services data. For each model, we compare its performance with other options and analyze the SHapley Additive exPlanation (SHAP) values to understand the impact of different explanatory variables on the model outcome. The results suggest that the COVID-19 factors marginally enhance the modeling of evacuation rates and distance, holding similar importance to traditionally recognized factors such as the percentage of senior people and vehicle ownership. Further analysis also suggests the impact of COVID-19 factors diminishes with distance from the coast. Moreover, we observed that the contributions of COVID-19 factors increase significantly when their values reach extreme levels, both very low and very high, suggesting that evacuation patterns were notably impacted under these conditions. Our findings contribute to understanding the impacts of various factors on evacuation patterns during Hurricane Ida, inform model selection for predicting critical evacuation/return metrics, and enrich the knowledge base of evacuation modeling in scenarios involving compound risks.