Bayesian Modeling of Reporting Biases in the SPC Tornado Database

Maximizing the value of tornado climatologies requires accounting for unreported or mischaracterized tornadoes, especially where people and damage indicators are sparse. This study adopts a Bayesian modeling framework to estimate tornado reporting rates over the central United States during 1975-2014. The method addresses some limitations of previous techniques. For example, while most previous studies have used a single covariate - population density – to model tornado reporting rate, we incorporate additional covariates into the analysis, including tree coverage and distance from nearest city. The technique produces spatial maps of tornado reporting rate and expected tornado frequency.

Preliminary results suggest around half of tornadoes that actually occurred in the analysis domain were reported, with reporting rate decreasing rapidly away from urban areas. Reporting biases are especially pronounced earlier in the record and for shorter-track tornadoes, but remain nontrivial even for more recent and longer-track tornadoes. Underestimation of tornado frequency increases with damage rating, underscoring the problem of under-rating tornadoes in rural areas.