All Times EDT
Recovering the Compensation Strategies for Tornado-Induced Property Losses via Statistical Manifold Learning (310010)*Thilini Vasana Mahanama, Texas Tech University
Pushpi Paranamana, University of Notre Dame
Dimitri Volchenkov, Texas Tech University
Keywords: Risk assessment, tornado property losses, statistical manifold learning
In this study, we examine the relationship between property losses caused by a tornado to its physical parameters (tornado path length and width) as it potentially could be useful in risk assessment strategies. By defining tornado scales based on physical parameters, we study the probability distributions of property losses. We observe that no single probability distribution describes the compensations in all tornado scales, and the dynamics of statistics are too complex to interpret. Therefore, we learn a statistical manifold based on these statistics to delineate the variations. We quantify the distances between statistics using Kolmogorov-Smirnov's distance and visualize them using classical multidimensional scaling. We obtain a statistical manifold by outlining the inherent patterns with respect to physical parameters. Taking the cell densifications into account, we define a curvature coefficient to explain the dynamics of statistics. We identify the physical parameters for potential extreme tornado events. Using this study, we were able to recover the compensation strategies for tornado-induced losses reported in the United States: private insurance and state funding.