FV3-LAM Convection-Allowing Model Forecasts and Ensemble Consensus Products for the 13th HMT Winter Weather Experiment

During the winter of 2022-2023, the Center for Analysis and Prediction of Storms (CAPS) ran an ensemble of convection-allowing model (CAM) forecasts as part of the 13^1h Annual NOAA Weather Prediction Center Hydrometeorology Testbed (WPC-HMT) Winter Weather Experiment (13^th WWE). For the 13^th WWE, the CAPS CAM ensemble comprised 15 members of the FV3 Limited Area Model (FV3-LAM) at ~3-km grid spacing over the contiguous United States (CONUS). The forecast ensemble and ensemble consensus products were produced for more than 30 days within WWE period (from mid-November to mid-March). Specific forecast days of interest were determined in consultation with NWS forecasters via the HMT meteorologists at the Weather Prediction Center. Ensemble consensus products such as simple mean, probability matched (PM) mean and local-probability matched (LPM) mean were produced from the ensemble. Individual members and ensemble consensus products are evaluated for their ability to improve on winter weather forecasting, particularly snow accumulation.

Another objective was to test possible combinations of physics for use in the future NOAA operational regional prediction system, such as the Rapid Refresh Forecast System (RRFS) CAM ensemble forecasting system, planned to replace current suites of regional operational models.

The CAPS CAM ensemble was designed with a control member and 14 other members with physics variations and/or variations in initial conditions and boundary conditions. The physics variations consisted of different combinations of microphysics, Planetary Boundary Layer (PBL), surface layer physics (Sfc), and Land Surface Model (LSM).

In addition to the suite of ensemble consensus methods a U-net convolutional neural network algorithm was trained to produce the snowfall forecasts from the HREF and 4 members of the CAPS CAM ensemble (HREF+). The U-net method produced probabilistic forecasts for snowfall for 6-h accumulations of 1-inch, 2-inch and 3-inch thresholds for lead times of 6 to 36 h.

Graphics products relevant to the forecasting of winter weather were produced, including surface (2m) temperature, liquid-equivalent precipitation and snowfall accumulation at 6h and 24h intervals, precipitation type paintball plots, and precipitation type plots for each member. Ensemble consensus products and U-Net ML product graphics were also generated. These graphics products are being posted online, and are publicly viewable at https://caps.ou.edu/forecast/realtime/ .

Subjective and objective evaluations of the CAPS ensemble, ensemble consensus products, including machine learning forecasts, implications for developing a RRFS ensemble will be presented at the meeting.