The Creation of a Statistical Model to Probabilistically Predict the Occurrence of Pulse Severe Thunderstorms

A small percentage of single cell thunderstorms meet the NWS criteria for being denoted severe. If severe weather does occur, it is of the “pulse” variety, i.e., short-lived and only marginally severe, but still able to produce damage. In the Southeast United States, these pulse severe storms can produce severe hail, but more often present as wind gusts in the form of wet microbursts. Due to the short-lived nature of the severe events and their relative scarcity, pulse severe storms are associated with short (if any) warning lead times, large false alarm ratios, and small probabilities of detection. The primary goal of this research is to build a statistical model that predicts the probability of pulse severe storms in the Southeast United States.

Storm objects generated from the empirical Probability of Severe model (developed by NOAA and Cooperative Institute for Meteorological Satellite Studies) were used as the foundation to build and track 47,562 nonsevere and 362 severe storm objects across the Southeast during the Summer of 2021. Each object was then populated with 110 variables from Multi-Radar Multi-Sensor, the High-Resolution Rapid Refresh model, and the Geostationary Lightning Mapper. Both a random forest classifier and an XGBoost classifier with feature selection techniques were used to determine the optimal combination of parameters that best identify pulse severe environments. This poster will describe our methodology in detail and present verification metrics of the developed model.