Temporal and Spatial Variations in the Number of Evacuees in the Great Eastern Japan Earthquake: Nationwide Analysis of Evacuees Using Data Science Methods

Abstract

The dynamics of evacuees due to the Great Eastern Japan Earthquake have been compiled in the “Number of Evacuees in Japan” by the Reconstruction Agency, but due to inconsistent formatting and other problems, comprehensive data analysis has not been conducted. In this study, data-scraping technology was adopted to successfully obtain all data. Thus, the dynamics of evacuees in 47 prefectures over more than 10 years were statistically analyzed. Using time-series clustering, a data science technique, we categorized prefectures with similar evacuee dynamics and discussed their respective trends. In addition, a sparse-modeling approach was utilized to explore the factors that explain the differences in evacuee retention rates among prefectures, and the results suggested that evacuees tend to settle in places where industries are thriving, land productivity is high, and living costs are low. Moreover, change-point analysis detected the timing of changes in evacuee dynamics. Thus, by utilizing data science, the dynamics of evacuees nationwide can be clarified, and it is expected to be applied to evacuee support and disaster prevention in the future.