Modeling and predicting evacuation flows during hurricane Irma

EPJ Data Science

Vanessa Frias-Martinez¹  Lingzi Hong² 

[1] College of Information Studies, University of Maryland College Park, 20742, College Park, Maryland, United States;UMIACS, University of Maryland College Park, 20742, College Park, Maryland, United States;College of Information, University of North Texas, 76201, Denton, Texas, United States;

关键词: Evacuation;    Disaster;    Geotagged tweets;    Environmental factors;    Prediction model;    Data-driven methods;    Decision-making;   

DOI  :  10.1140/epjds/s13688-020-00247-6

来源: Springer

【 摘 要 】

Evacuations are a common practice to mitigate the potential risks and damages made by natural disasters. However, without proper coordination and management, evacuations can be inefficient and cause negative impact. Local governments and organizations need to have a better understanding of how the population responds to disasters and evacuation recommendations so as to enhance their disaster management processes. Previous studies mostly examine responses to evacuations at the individual or household level by using survey methods. However, population flows during disasters are not just the aggregation of individuals’ decisions, but a result of complex interactions with other individuals and the environment. We propose a method to model evacuation flows and reveal the patterns of evacuation flows at different spatial scales. Specifically, we gathered large-scale geotagged tweets during Hurricane Irma to conduct an empirical study. First, we present a method to characterize evacuation flows at different geographic scales: the state level, considering evacuation flows across southern states affected by Irma; the urban/rural area level, and the county level. Then we demonstrate results on the predictability of evacuation flows in the most affected state, Florida, by using the following environmental factors: the destructive force of the hurricane, the socioeconomic context, and the evacuation policy issued for counties. Feature analyses show that distance is a dominant predictive factor with counties that are geographically closer generally having larger evacuation flows. Socioeconomic levels are positively related to evacuation flows, with popular destinations associated to higher socioeconomic levels. The results presented in this paper can help decision makers to better understand population evacuation behaviors given certain environmental features, which in turn will aid in the design of efficient and informed preparedness and response strategies.

【 授权许可】

CC BY