Development and Testing of 3D- and 4D-EnVar Subhourly Data Assimilation in the HRRR

The current operational configuration of the 3-km High-Resolution Rapid Refresh (HRRR) features an hourly update cycle based on a downscaled RAP initialization and a one-hour pre-forecast. In an effort to take advantage of sub-hour observational data, testing of 15-minute update cycles during the one-hour pre-forecast was conducted using both 3D- and 4D-EnVar. Initial findings indicate that on average, both 3D- and 4D-EnVar result in a closer fit to the obs (lower RMSE and bias) when compared to hourly 3D-EnVar for traditional verification metrics at analysis time. For 1-, 3-, and 6-hr forecasts, the sub-hourly DA provides an overall neutral impact. However, analysis of radar reflectivity during the free forecast indicates a potential qualitative improvement of placement and intensity of small-scale precipitation features for the sub-hourly runs, in particular for the 4D-EnVar simulations. In addition, a comparison of spread was conducted between the GFS ensemble, currently used in the 3DEnVar operational HRRR DA system, and a prototype 3-km WRF-based ensemble. Much more detail and evolution in mesoscale features was apparent in the 3-km WRF-based ensemble between forecast hours, suggesting that its use in DA could greatly impact forecast results. Further analysis will focus on assessing changes within the three backgrounds used in 4D-EnVar to illustrate how this evolution impacts the final analysis and forecast.