Probabilistic forecasts of severe convection with a WRF-DART analysis and convection-permitting forecast system

Convection-permitting models provide useful forecast guidance on expected convective mode, but often struggle to accurately forecast timing and location for small-scale and extreme events. Additionally, with horizontal grid spacing on the order of 1-4 km, small-scale convective features and severe weather phenomena are not fully resolved, necessitating the use of severe storm proxies. In this study, a five-member Weather Research and Forecasting (WRF) model ensemble was used to assess the performance of probabilistic convection-permitting model guidance in forecasting severe storms that occurred on 19-20 May 2012 across the central Great Plains. Forecasts were initialized with an ensemble mesoscale analysis from the real-time NCAR WRF-DART ensemble data assimilation system. Model forecasts were subjectively evaluated by comparing radar observations and storm reports to model fields and severe storm proxies such as simulated reflectivity and maximum updraft helicity. Precipitation forecasts were verified against Stage IV analysis using a “neighborhood-based” approach, which has been shown to produce more representative forecast skill scores when verifying high-resolution model output. Results suggested the ensemble system produced a reliable forecast of convection that would have provided useful guidance on storm mode, location and intensity on this day. Severe storm proxy probabilities of maximum updraft were found to be most useful for identifying locations of observed weather hazards, while discriminating hazard types from their associated proxies were less skillful.