Abstract
When forecasting it is essential to give the public advanced warning of any incoming or possible severe weather events, so the public can take appropriate action. Thus, forecasters must hurriedly examine pertinent data, recognize any threats, and issue any watches, warning, or special weather statements to the public as quickly as possible. The National Oceanic and Atmospheric Administration (NOAA) and the Cooperative Institute for Meteorological Satellite Studies (CIMSS) helped answer that need by developing the NOAA/CIMSS Probability of Severe (ProbSevere) Model which is a probabilistic guidance system that communicates the level of certainty of the possibility that convective storm cells will produce severe weather in the future. We sampled 297 storms from March through July 2018 to determine how well calibrated the ProbSevere model is to Central and South Florida. We found by producing a Receiver Operating Characteristic (ROC) curve that the parameter Environmental MUCAPE had negative skill in the model while the parameters ProbSev and Environmental EBS showed some skill. The ROC curve for the parameter Flash Rate showed some skill, but, in the range of 21 to 40 flashes per minute the model had negative skill. This means that forecasters when using this model should not use the parameter Env. MUCAPE when deciding whether to issue a weather statement. Meteorologists also need to be wary of issuing a weather statement based on Flash Rate between 21 and 40 flashes per minute. The results of the reliability curve produced for the parameter ProbSev showed a positive slope, indicating that the forecasts have some reliability. However, the slope is less than the perfect reliability diagonal in thresholds 30 to 100%. The threshold of 90 to 100 % had the highest reliability being the closest to the perfect reliability diagonal. This indicates that forecasters can have confidence that storms will produce severe weather when issuing a weather statement for a storm that has a ProbSev of 90 to 100%. Future work is needed to see how these results change by lowering the threshold of storms looked at for the false alarm cases along with changes in sample size, and season.