Bias in Severe Thunderstorm and Tornado Warnings Issued by the National Weather Service in the Doppler Radar Era: A Spatial-Temporal Evaluation

A climatology of severe thunderstorm (damaging wind and/or hail) and tornadoes in the United States has established the location of the areas of highest frequency of occurrence. This climatology was attained through analysis of a basic data source, that of observed events, which carries many associated biases. Among these biases is the requirement that someone be on hand to witness the event no matter what time of the day or night, the assumption that the observer had sufficient visibility to see the event clearly, and whether there was something available on location to damage. In this study I use an alternate database consisting of the number of county severe thunderstorm warnings and tornado warnings issued by the National Weather Service, primarily for the 1995-2004 time window, between the Rocky Mountains and Appalachian Mountains. Because this alternative climatology is based upon the much improved technology available using Doppler radar, it is believed to have fewer and more quantifiable biases for the spatial analysis of severe weather distribution. There are two suspected areas of bias in this alternative data source: 1) population density; and 2) distance a county is from the nearest radar transmitter. The numbers could also vary spatially according to which Weather Service Office (WFO) issued the warning. Regression analysis and statistical tests were used to quantify bias to produce a spatial distribution that is complimentary to the climatology based upon reported events. The primary goal of the study was to identify and quantify the biases, and then develop a spatial pattern that is representative of the actual severe weather threat. Results indicate that bias is frequent and highly variable according to WFO but could not be accurately quantified. The difference in issuance frequency of warnings between those offices which is based on much subjectivity appears more dominant than the biases. The resultant distribution of severe thunderstorm warnings is similar to one that uses reported events. The distribution of tornado warnings remains skewed by the differences between WFOs and is not likely to be representative of the actual tornado threat.