Dynamic-data-driven agent-based modeling for the prediction of evacuation behavior during hurricanes

Establishing an efficient disaster management strategy against severe natural disasters is essential to mitigate and relieve their catastrophic consequences. In order to understand the situation during such devastating events, it is crucial to incorporate individuals' behaviors and their decision-making processes, which requires an amalgamation of information from various sources such as survey data, information regarding location and intensity of disasters, government's policies, and supplies in the affected region. This work proposes a dynamic-data-driven model for individual decision-making processes capable of tracking people's preference value over time, incorporating dynamic environmental changes using Bayesian updates. An agent-based simulation was used to model each of the components vital to devise an effective disaster management strategy. Moreover, the proposed model allows deriving quantitative relationships among people's evacuations, their demographic information, and risk perception based on environmental changes, including traffic status, gas outage, and government notice. For this study, the authors considered Florida's situations during hurricanes Irma, Michael, and Dorian in 2017, 2018, and 2019. What-if analyses were also conducted to find the best disaster management policy for government agencies to minimize the hurricane's effect, which will help prepare for future disaster situations.