Ensemble Probabilistic Severe Weather Prediction Using Convection-Allowing Models and Deep Generative Models

An ensemble post-processing method is developed for the probabilistic prediction of severe weather (tornadoes, hail, and wind gusts) over the Conterminous United States (CONUS) out to 24 hours. The method combines Conditional Generative Adversarial Networks (CGANs), a type of deep generative model, with a Convolutional Neural Network (CNN) to post-process convection-allowing model (CAM) forecasts. The CGANs are designed to create pseudo ensemble members from deterministic CAM forecasts (see figure), and their outputs are processed by the CNN for probabilistic estimations. The method is tested using High-Resolution Rapid Refresh (HRRR) forecasts as inputs and the Storm Prediction Center (SPC) severe weather reports as targets. Based on the post-processing experiment in 2021-2022, the method produced skillful severe weather predictions with up to 20% Brier skill score (BSS) increases compared to other neural-network-based reference methods. The BSS increase is primarily contributed from the northeastern United States and the Great Plains, where severe weather is observed frequently. For the evaluation of uncertainty quantifications, the method is overconfident but produces meaningful ensemble spreads that can distinguish good and bad forecasts. The quality of CGAN outputs is also evaluated. Results show that the CGAN outputs behave similarly to a numerical ensemble; they preserved the inter-variable correlations and the contribution of influential predictors as in the original HRRR forecasts. This work is the first that applies CGANs to post-process CAM forecasts, and it is one of only a few studies that performed ensemble predictions of severe weather using neural networks.