Evaluation of Real-time Convection-Allowing FV3 Ensemble Forecasts from the 2022 HWT Spring Forecasting Experiment to Inform RRFS Ensemble Design

As part of the transition to a unified forecast framework, NOAA and the National Weather Service are developing the Rapid Refresh Forecasting System (RRFS)—a rapidly-updating convection-allowing NWP ensemble using the limited area FV3 NWP model (FV3-LAM) that is part of the Unified Forecasting System (UFS) framework. The RRFS will produce ensemble forecasts at convection-allowing (≈ 3 km) grid spacing and will provide hourly-updated deterministic and ensemble forecasts over North America for up to several days, replacing the suite of regional NWP forecasts currently run by NOAA.

To help inform and optimize the ensemble design of the experimental RRFS ensemble, the Center for the Analysis and Prediction of Storms (CAPS) at the University of Oklahoma ran daily real-time FV3-LAM ensemble forecast experiments at 3 km grid spacing daily during the 2022 Hazardous Weather Testbed (HWT) Spring Forecast Experiment (SFE) from May through early June of 2022. These experiments included 21 forecasts grouped into 3 sub-ensembles: (1) 5 forecasts which share an initial condition obtained from a prototype RRFS hybrid EnVar analysis produced by NOAA GSL but have different physics configurations; (2) 10 forecasts with various physics configurations and diversity in initial conditions provided via ensemble Kalman filter initial conditions from GSL; (3) 5 forecasts using a subset of configurations from group 2 with additional stochastic physics perturbations produced using the SPPT, SKEB and SHUM perturbation schemes. One baseline forecast starting from the GFS analysis was also produced. Lateral boundary conditions were obtained from GFS and GEFS for the physics-perturbation-only and fully-perturbed ensembles, respectively. Physics schemes tested include Thompson and NSSL microphysics; MYNN, Shin-Hong, and TKE-EDMF PBL schemes; and Noah, NoahMP, and RUC land surface models.

CAPS FV3 ensemble forecasts from the 2022 HWT SFE are evaluated and verified in terms of member-by-member performance, the relative performance of sub-ensembles with different perturbation strategies, and the performance of these sub-ensembles and the CAPS FV3 ensemble in aggregate compared against other operational ensemble NWP systems, including HREF and GEFS. The Model Evaluation Tools (MET) package is used as the basis for these evaluations. Precipitation forecasts on days 1, 2, and 3 are a particular focus for evaluation, verified against Stage-IV precipitation data. Forecasts of surface and upper-air variables are also considered, verified against the Unrestricted Mesoscale Analyses (URMA) and sounding data. The operational utility of varied perturbation strategies, and potential implications for RRFS ensemble design, will be discussed.