We will describe the development of a high-resolution, gridded forecast guidance product for warm season cloud-to-ground (CG) lightning in Florida. Four warm seasons of analysis data from the 20-km Rapid Update Cycle (RUC) and lightning data from the National Lightning Detection Network are used to examine relationships between observed atmospheric parameters and the spatial and temporal patterns of CG lightning over Florida. The most important RUC-derived parameters then are used in a perfect prognosis (PP) technique to develop equations producing 3-hourly spatial probability forecasts for one or more CG flashes, as well as the probability of exceeding various flash count percentile thresholds. Binary logistic regression is used to develop the equations for one or more flashes, while a negative binomial (NB) model is used to predict the amount of lightning, conditional on one or more flashes occurring. When applied to the dependent sample of RUC analyses, the equations show forecast skill over a model containing only persistence and climatology (L-CLIPER).

We also evaluate the lightning forecast scheme when applied to output from three mesoscale models during an independent test period (the 2006 warm season). The evaluation is performed using output from NCEP's 13-km RUC, the NCEP 12-km NAM-WRF, and local runs of WRF for a domain over South Florida that were initialized with NCEP 1/12^th degree sea-surface temperatures (SST) and data from the Local Analysis and Prediction System (LAPS) (WRF-LAPS). During the most active lightning period (1800-2059 UTC), the three models forecast between 80-90% of the lightning events having one or more flashes, and between 30-60% of the events with flash counts meeting or exceeding the 95^th percentile. Of the three mesoscale models, WRF-LAPS generally produces the best verification scores during 1800-2059 UTC. Forecasts from all three mesoscale models generally show positive skill with respect to L-CLIPER and persistence through the 2100-2359 UTC period, demonstrating that the PP scheme is model independent. Although the exact timing and placement of forecast lightning is not perfect, there generally is good agreement between the forecasts and their verification, with most of the observed lightning occurring within the higher forecast probability contours.