The High-Resolution Rapid Refresh (HRRR, Benjamin et al. 2009, this symposium)) system is planned for operational implementation over the next few years as high-resolution nest within the upcoming Rapid Refresh system, a replacement of the current operational RUC (Rapid Update Cycle) model. The data assimilation system in the Rapid Refresh (driving the HRRR) will be based on the NCEP GSI (Grid-point Statistics Interpolation) 3DVAR system and the prediction model will be based on the WRF-ARW core.  In this paper, we look one step further ahead toward radar assimilation at 3-km resolution toward a potential subsequent upgrade to the HRRR.

A rare event occurred over Oklahoma in August 2007 when Atlantic tropical storm Erin (2007) re-intensified over western Oklahoma three days after making a landfall. The storm re-developed an eye, an eye wall structure and spiral rain bands after weakening significantly over western Texas, producing strong winds and heavy flooding that claimed several lives and caused extensive property damage. Being over land and covered by the operational WSR-88D radar network, this event provides a good opportunity to test the assimilation of both radial velocity and reflectivity observations of operational radars into high-resolution NWP models, for the prediction of such storms of tropical nature, including the prediction of track, intensity, structure and precipitation. An excellent array of observation networks, including the Oklahoma Mesonet, also provides an opportunity for detailed model verification.

The radar-enhanced GSI system is used to assimilate level-II radial velocity data, and the mosaic reflectivity data produced by NSSL are assimilated within the GSI framework using a complex cloud analysis package. Thirty-minute assimilation cycles of up to 6 hours are performed on the 3-km grid. The operational 13-km RUC analyses provide the initial guess to start the radar data assimilation cycles and provide the boundary conditions.

Various model and GSI configurations are tested. The scale of the background error covariance is tuned for optimal results when analyzing radial velocity data. Microphysics parameterization schemes are found to make significant impact on the location, intensity and rainband structure forecast, so does the radar data assimilation. The differences using different PBL schemes are found to be smaller. A 1-km forecast grid is further nested within the 3-km grid and precipitation forecasts are verified against observations.