Population Influences on Tornado Reports in the United States

Abstract

The number of tornadoes reported in the United States is believed to be less than the actual incidence of tornadoes, especially prior to the 1990s, because tornadoes may be undetectable by human witnesses in sparsely populated areas. We use a hierarchical Bayesian model to simultaneously correct for population-based sampling bias and estimate tornado density using historical tornado report data. The expected result is that F2-F5 compared to F0-F1 tornado reports would vary less with population density. The results agree with this hypothesis for the following population centers: Atlanta, GA; Champaign, IL; Des Moines, IA. However, the results indicated just the opposite in Oklahoma. We speculate the result is explained by misclassification of tornadoes that were worthy of F2-F5 Fujita scale rating but were classified as F0-F1 tornadoes, thereby artificially decreasing the number of F2-F5 and increasing the number of F0-F1 reports in rural Oklahoma.