A Statistical Procedure to Forecast Warm Season Lightning over Portions of the Florida Peninsula

Abstract

Sixteen years of cloud-to-ground lightning data from the National Lightning Detection Network and morning radiosonde-derived parameters are used to develop a statistical scheme to provide improved forecast guidance for warm season afternoon and evening lightning for 11 areas of the Florida peninsula serviced by Florida Power and Light Corporation (FPL). Logistic regression techniques are used to develop equations predicting whether at least one flash will occur during the noon?midnight period in each area, as well as the amount of lightning that can be expected during this same period, conditional on at least one flash occurring. For the amount of lightning, the best results are achieved by creating four quartile categories of flash count based on climatology, and then using three logistic equations and a decision tree approach to determine the most likely quartile. A combination of forward stepwise screening and cross validation are used to select the best combination of predictors that are most likely to generalize to independent data. Results show the guidance equations to be superior to persistence on both the dependent dataset and during cross validation. The greatest skill scores are achieved for predicting whether at least one flash will occur, as well as predicting the number of flashes to within one quartile of that observed. These results demonstrate that the equations possess forecast skill and will provide useful guidance for the probability and amount of lightning in each of the 11 FPL service areas.