Bayesian estimation of the likelihood of extreme hail sizes over the United States

Abstract

Large hail causes significant economic losses in the United States each year. Despite these impacts, hail is not typically included in building and infrastructure design standards, and assessments of hazards from extreme hail size remain limited. Here, we use a novel approach and multiple hail size datasets to develop a new Generalized Extreme Value model through a Bayesian framework to identify large hail-prone regions across the country at 0.25° × 0.25°. This model is smoothed using Gaussian process regression for nationwide estimation of return likelihood. To contextualize local risk, hazard returns intersecting high-population exposure centers are compared. Fitted extreme value models suggest earlier work likely underestimates the hail hazard. Especially for higher return periods, the Bayesian approach is found to better model very rare hail occurrences than traditional approaches. This provides a framework for appreciating underlying risk from hail and motivates mitigative approaches through improving design standards.