A Preliminary Investigation of the High-Resolution Ensemble Forecast (HREF) for Generating Calibrated Probabilistic Thunderstorm Forecasts

In 2003, the Storm Prediction Center (SPC) began post-processing the National Centers for Environmental Prediction (NCEP) Short-Range Ensemble Forecast (SREF) system to provide operational guidance on the prediction of lightning hazards across the contiguous United States. This guidance relies on physically-based parameters to produce probabilistic forecasts of thunderstorms, which are then calibrated using data from the National Lightning Data Network (NLDN) such that the predicted probabilities from an independent sample are statistically reliable against the verifying NLDN data. Although this method has generally shown both skill and reliability at predicting the occurrence of lightning, the temporal and spatial accuracy of the predictions is limited in part by the inability of the SREF to explicitly resolve convection. Therefore, it is hypothesized that including simulated radar reflectivity and other storm-attribute fields from convection-allowing models (CAMs) may lead to improved probabilistic calibrated thunderstorm predictions. This study serves as a preliminary investigation into the use of the operational CAM ensemble at NCEP, the High-Resolution Ensemble Forecast (HREF), for potentially improving probabilistic lightning prediction. Each HREF-member field was compared to the NLDN cloud-to-ground (CG) lightning dataset between 1 July 2017 and 1 July 2018 to identify which fields most closely correlated with the occurrence of one or more CG lightning flashes. A new variable, known as the Composite Lightning Parameter (CLiP) was then derived from the most correlated fields and optimized using a grid-search technique. Finally, probabilistic thunderstorm forecasts were computed from the HREF using the CLiP and calibrated to be statistically reliable using an isotonic regression model. This method has shown both skill and reliability during preliminary tests, and there is potential for greater improvement through additional machine learning techniques.