Evaluation of Severe Weather Forecasts from Real-Time Medium-Range Convection-Allowing Ensembles over the Conterminous United States

During both spring 2023 and 2024, we produced two sets of experimental, real-time, medium-range (3-8 days) convection-allowing ensemble (CAE) forecasts with variable-resolution global models that had 3-km horizontal cell spacing over the conterminous United States (CONUS). One ensemble had 10 members, produced 8.5-day forecasts, and used GFDL's Continental System for High-Resolution Prediction on Earth-to-Local Domains (C-SHiELD), a prediction model based on the finite-volume cubed-sphere dynamical core. The second ensemble had 5 members, produced 5.5-day forecasts, and used NSF NCAR's Model for Prediction Across Scales (MPAS). These demonstrations collectively represented the first real-time, medium-range CAE guidance over the CONUS and the first real-time CAE forecasts produced with variable-resolution global numerical weather prediction models. Probabilistic medium-range surrogate severe weather forecasts based on 2-5-km updraft helicity (UH) were objectively verified against observed severe weather reports over the CONUS. For optimal UH thresholds and smoothing length scales, the MPAS and C-SHiELD ensembles produced skillful severe weather forecasts relative to climatology through 5 and 8 days, respectively. Moreover, both sets of medium-range ensemble forecasts typically had higher skill scores than 3-8-day Storm Prediction Center (SPC) outlooks, although comparison of SPC and CAE forecasts should be interpreted cautiously. Numerous forecast examples demonstrate strengths of the medium-range CAE forecasts, as well as some weaknesses. Overall, results suggest great promise of medium-range CAEs for severe weather forecasting applications over the CONUS. SIGNIFICANCE STATEMENT: In spring 2023 and 2024, we executed the first real-time, medium-range (3-8 days), high-resolution ensemble forecasts over the conterminous United States (CONUS). The forecasts over the CONUS had 3-km horizontal cell spacing and were produced by two very different modeling systems. This study shows that probabilistic medium-range severe weather forecasts from both ensembles were skillful and suggests that high-resolution medium-range ensembles may be useful guidance tools for severe weather forecasters.