The Rapid Refresh (RR) mesoscale analysis and prediction system assimilates national mosaic radar reflectivity data via a diabatic digital filter initialization procedure. This procedure, initially developed and applied within the Rapid Update Cycle (RUC) system, is key for improvements of both short-range RR convective forecasts and the High Resolution Rapid Refresh (which runs as a nest within the RR). In order to quantify the impact of the radar data assimilation, as well as other model configuration changes, we have developed a comprehensive reflectivity and precipitation verification package. This package, which runs in real-time on all RR and HRRR parallel cycles and has been applied to various retrospective test cases, is crucial for quickly evaluating possible improvements.

The multi-scale verification package first maps native precipitation and reflectivity fields to several coarser grid resolutions (facilitating “neighborhood verification”), then computes standard verification scores and creates plots of hits, misses, and false alarms for several thresholds on each resolution grid. The verification data are then loaded into a database and a web interface that allows many statistical measures (CSI, bias, RMS, coverage area, etc.) to be displayed and overlaid with a wide variety of user input stratification specifications. The verification system allows isolation of patterns to model errors (diurnal, lead time, etc.), quickly giving feedback on changes made to the various real-time and retrospective forecast cycles.

In this presentation, we will report on the application of this system to quantify the impact of the RR reflectivity data assimilation on the RR and HRRR forecasts of reflectivity and precipitation. In addition to evaluating real-time runs (parallel RR at EMC, ESRL primary and development RRs, and ESRL primary and developmental HRRRs), we have evaluated several retrospective runs. Key issues we are examining with regard to the radar data assimilation are 1) documenting the impact for the DFI-radar assimilation in both the RR and the HRRR as a function of forecast length and time-of-day, 2) quantifying the sensitivity of the impact to the strength of the latent heating based temperature tendency. In particular, we are conducting experiments to see if we can improve forecast skill by adding stability and convective life-cycle dependencies to the latent heating specification. At the conference, we will present the results and subsequent changes made to the radar assimilation algorithm.