Quantifying Reporting Biases in the SPC Tornado Database

The Storm Prediction Center (SPC) tornado database is an indispensable resource for assessing tornado risk in the United States, and may ultimately prove critical for identifying changes in tornado characteristics due to climate change. Maximizing the value of the database, however, requires accounting for under-reporting in regions where tornadoes are more likely to escape notice. Previous studies examining secular effects on reported tornado frequency have modeled the impact of only a single variable (e.g., population density) or of a combination of variables that are implicitly assumed to have mutually independent effects on reporting frequency. The present study, on the other hand, uses multivariable polynomial regression to make provision for inter-variable effects on tornado reporting bias. This allows us to account for such effects as the expected decrease of the influence of local population density very near NWS Weather Forecast Offices (WFOs).

We model tornado reporting bias as a function of various combinations of the following variables: 1) distance from nearest 100K city (population > 100K), 2) distance from nearest 30K city, 3) distance from nearest WFO, 4) distance from nearest interstate, and 5) local population density. Of the five variables, population density is found to have the least impact on reporting bias. Comparing maps of original (from the SPC database) and corrected (using the modeled biases) tornado frequency demonstrates the importance of accounting for secular effects. As expected, the reporting biases are found to be most pronounced for weaker/briefer tornadoes. This work is ultimately intended to inform the methodology of future tornado climatology studies, and to increase the spatial resolution of tornado hazard models used by forecasters, planners, and insurance/reinsurance companies.