Modeling Extreme Hurricane Damage Using the Generalized Pareto Distribution

Abstract

SYNOPTIC ABSTRACTExtreme value distributions are used to understand and model natural calamities, man-made catastrophes and financial collapses. Extreme value theory has been developed to study the frequency of such events and to construct a predictive model so that one can attempt to forecast the frequency of a disaster and the amount of damage from such a disaster. In this study, hurricane damage in the United States from 1900 to 2012 has been studied. The aim of the article is threefold. First, to normalize hurricane damage and fit an appropriate model for the normalized damage data. Secondly, to predict the maximum economic damage from a hurricane in the future by using the concept of return period. Finally, to quantify the uncertainty in the inference of extreme return levels of hurricane losses by using a simulated hurricane series, generated by bootstrap sampling. Normalized hurricane damage data are found to follow a generalized Pareto distribution. It is demonstrated that the standard deviation and coefficient of variation increase with the return period, which indicates an increase in uncertainty with model extrapolation.