Statistical Prediction of Waterspout Probability

Waterspouts pose a threat to boater safety, and more waterspouts are reported in the Florida Keys than in any other place in the world. Previous research provided qualitative guidance for prediction of atmospheric environments favoring waterspouts. Quantitative techniques are needed to improve forecasting of waterspout threats. Statistical analysis of 1200 UTC sounding-derived parameters from Key West, Florida and Charleston, South Carolina provided a quantitative method for prediction of the probability of waterspouts. National Weather Service Local Storm Reports for June through September from 2006 to 2014 identified waterspout report days. Logistic regression yielded a best model with independent predictors. Three metrics evaluated the model and a benchmark model based on discrimination and difference of means, classification using Receiver Operating Characteristic curves with area under the curve, and calibration with reliability diagrams and Brier Skill Score. Both the Key West, Florida model and Charleston, South Carolina model proved superior to the benchmark via assessment of performance metrics. This quantitative method for prediction of the probability of waterspouts demonstrates skill over a benchmark model for the Florida Keys and Charleston, South Carolina. Future work will involve incorporating this technique into weather forecasting operations and refining it with additional data and analysis.