In this study, ten retrospective cases from 2015 and 2016 are used to test potential improvements to the convection allowing ensemble assimilation and forecast system . The performance of an ensemble forecasting system depends upon the ability of the ensemble to represent all sources of uncertainty, including initial condition errors and model error. The ensemble of analyses produced by the hybrid DA system can automatically sample the errors of the initial conditions. How to optimally sample the errors in the model still remains to be addressed. In this study, we look to address the following questions: how is a multi-core ensemble compared to a single core multi-physics or stochastic physics ensemble given that multiscale initial condition errors are already sampled by a state-of-the-art DA method?
A single 3-km horizontal resolution domain covering the CONUS is used throughout, with hourly assimilation of conventional observations as well as 20-min sub-hourly assimilation of radar DA for the final hour of cycling. In addition to single-model single-physics (SMSP) configurations, ensemble forecast experiments compared multi-model (MM) and multiphysics (MP) approaches. Stochastic physics was also applied to MP for further comparison. These five ensemble configurations are compared subjectively and quantitatively within the 18-h free forecast period. Neighborhood-based verification of precipitation and composite reflectivity showed each of these model error techniques to be superior to SMSP configurations. Comparisons of MM and MP approaches had mixed findings. The MM approach had better overall skill in precipitation forecasts, however MP ensembles had reduced ensemble mean error of other fields, particularly near the surface. The MM experiment had the largest spread in precipitation, and for most hours in other fields; however, rank histograms and spaghetti contours showed significant clustering of the ensemble distribution. MP plus stochastic physics was able to significantly increase spread in time to be competitive with MM by the end of the forecast. The results generally suggest a MM approach is best for early forecast lead times up to 6-12 hours, while a combination of MP and stochastic physics approaches is preferred for forecasts beyond 12 hours. In addition to the above comparisons in NAMRR and HRRR-based contexts, preliminary performance comparisons of ensemble forecasts run using FV3-SAR, NOAA’s newest regional model, initialized by the multiscale EnVar analyses will be also be presented.