Transferring Household Evacuation Choice Behavioral Models to Create a Digital Twin for Future Storm Responses: Opportunities and Challenges

Abstract

This research studied whether and how hurricane evacuation behavioral models, estimated with survey data collected from previous storms, could be used in predicting evacuation patterns in a new storm setting (with an anticipation of being able to conduct real-time simulation in the future). With publicly available data, this study first created synthetic populations for the study region (i.e., New Orleans, Louisiana) by year. Findings from this process show that: 1) simulating evacuation behavior can only be done for storms that have occurred between 2013 and two years back from the current year; and 2) it might not be appropriate to use population data of a different year to simulate evacuation behavior in a current storm year owing to population migration. With synthetic populations created for 2021, this study simulated household hurricane evacuation-related choices in Hurricane Ida with behavioral models estimated before, which facilitates discussions about model transferability. It was found that lognormal distance function parameters in the evacuate/stay and departure timing joint choice model, and destination risk perception values in the destination choice model are the two most critical factors that need to be updated. Both factor updates are related to storm characteristics and can be completed with live storm feeds, which indicates real-time data input is indispensable in improving prediction accuracy. This study highlights data challenges for real-time simulation, helps improve the usefulness of estimated statistical models in practical applications, and emphasizes the importance of considering human components (including demographic profiles and choice behavior) in creating digital twins.