Development of GSI-Based EnKF and Hybrid EnVar Data Assimilation Capabilities for Continental-Scale 3-km Convection-Permitting Ensemble Forecasting and Testing via NOAA Hazardous Weather Testbed Spring Forecasting Experiments

Over the past two decades, significant progress has been made with radar data assimilation (DA) using the EnKF technique for convection-allowing ensemble forecasts, where radar became indispensable source of observation. As radar DA becomes a common practice in convective-scale DA, CAPS has been leading the efforts in establishing and testing a more operationally relevant EnKF system to accelerate an operational convection allowing model (CAM)-based ensemble for the US. Sponsored by several NOAA projects, CAPS have been producing a real-time EnKF-based CAM ensemble every spring as parts of the NOAA Hazardous Weather Testbed (HWT) Spring Forecast Experiment (SFE) since 2013. One unique feature of this ensemble is the real time assimilation of operational WSR-88D radar radial wind and reflectivity data at the native model grid resolution (4 km initially, 3 km since 2016). Since 2016, a combination of the NCEP Gridpoint Statistical Interpolation (GSI)-based EnKF system and an EnKF system developed at CAPS was cycled over a 6-hour period from 1800 to 0000 UTC. The GSI-EnKF assimilated all operational data used by the Rapid Refresh except for satellite and aircraft data at hourly intervals while the CAPS EnKF system (for computational efficiency reason) assimilated radar data between 2300 and 0000 UTC at 15 minute intervals. The EnKF DA used 40 members with multiple PBL schemes, and the Thompson microphysics scheme with perturbed graupel density. We believe this effort is unique, in terms of running a EnKF system in real time at a convection-permitting resolution over very large (up to CONUS) domains, assimilating full-volume Doppler radar data. After optimization of the EnKF configuration for the 2017 season, significant improvement in the precipitation forecasting skill is achieved compared to forecasts from the 2016 season. Evaluation results for the 2017 season also show that forecasts from the ensemble mean analyses outperform forecasts initialized from 3DVar/cloud analysis (CA) at 0000 UTC. The 3DVar/CA-initialized forecasts display immediate collapse of analyzed storms and rebuilding them in the first hour while it restores balance among the model variables. On the other hand, storm structure is well maintained in the EnKF-initialized forecast, which suggests that the EnKF analysis is more balanced. The power spectra of mixing ratios also exhibit 3DVar/CA-initialized forecasts going through rapid internal adjustments within the first hour. More results comparing behaviors of 3DVar, EnKF, and En3DVar will be presented at the conference.