Direct Assimilation of Radar Data With Ensemble Kalman Filter and Hybrid Ensemble‐Variational Method in the National Weather Service Operational Data Assimilation System GSI for the Stand‐Alone Regional FV3 Model at a Convection‐Allowing Resolution

Capabilities to directly assimilate radar radial velocity (Vr) and reflectivity (Z) data are implemented within the operational GSI data assimilation (DA) framework and coupled with the new stand‐alone regional (SAR) FV3 model. The effectiveness and performance of 3DVar, EnKF, and hybrid En3DVar methods are evaluated with a storm cluster over the U.S. Central Plains at 3‐km grid spacing. During the DA cycles, 3DVar analyses show better fit to Z observations but fastest error growth, while EnKF and pure En3DVar lead to smaller forecast errors. For Vr, EnKF outperforms other methods in both analysis and forecast. Good correspondence with tornado reports is obtained by most experiments for probabilistic forecast of updraft helicity (UH), except for 3DVar which shows insufficient confidence in certain regions. Overall, EnKF and hybrid En3DVar show best forecast skills in terms of composite reflectivity and UH. Tests with more cases are needed to draw more general conclusions, however. Plain Language Summary: The Finite‐Volume Cubed‐Sphere Dynamical Core (FV3) was chosen to serve as the single dynamical core for forecasts at all scale by the National Weather Service in late 2016. A stand‐alone regional (SAR) version of FV3 became available in early 2019 and is planned to replace the current operational 3‐km grid spacing High‐Resolution Rapid Refresh (HRRR) system for convection‐allowing forecasting at NCEP. As a key component in convective‐scale numerical weather forecast (NWP) initialization, effective assimilation of radar data is crucial for desirable performances of the upcoming SAR FV3 model. In this paper, direct radar data assimilation (DA) capabilities implemented within the operational GSI DA framework by Center for Analysis and Prediction of Storms (CAPS), distinguished from the indirect approach applied in current operational HRRR to assimilate reflectivity data and believed to be more beneficial in particular for hydrometeor analyses, are interfaced with SAR FV3 model and the performances of different DA methods implemented are examined through the case study of a Central Plains tornadic storm cluster. Overall, results show good improvement of the forecasts, in terms of evolution of storm structures and occurrences of storm‐related severe hazards, which greatly encourage acceleration of operational adoption of such capabilities. Key Points: Direct assimilation of radar data is implemented in operational GSI system and interfaced with the new stand‐alone regional FV3 modelThe capabilities with GSI 3DVar, EnKF, and hybrid and pure En3DVar are tested with a case of Central Plains tornadic storm clusterOverall, EnKF and hybrid En3DVar perform better than pure En3DVar, with 3DVar being significantly worse [ABSTRACT FROM AUTHOR]