Radar Data Assimilation Impacts in the RAP and HRRR Models

The Rapid Refresh (RAP) and High Resolution Rapid Refresh (HRRR) are hourly-updating NOAA operation weather prediction models that are heavily used by National Weather Service (NWS) forecasters and in many other sectors. A key feature of both models is a radar reflectivity assimilation procedure to initialize areas of ongoing precipitation, leading to a significant reduction of model spin-up and improved very short-range forecasts of precipitation, especially convective systems. The RAP and HRRR radar assimilation procedures have some aspects in common (specification of temperature tendencies from a reflectivity-based diagnose of latent heating), but differ in important aspects. Within the RAP reflectivity assimilation procedure a single application of the latent heating specification is made over a window during the diabatic forward model portion of the digital filter initialization (DFI). Within the HRRR reflectivity assimilation procedure, four 15-minute applications of latent specification are made during a one-hour pre-forecast integration and there is no application of the digital filter. Previous work has shown that both the 13-km and 3-km radar assimilation contribution to the overall HRRR short-range (0-4 h) reflectivity forecast skill.

In this work, we extend the previous work to examine further the forecast impact from both the RAP and HRRR reflectivity assimilation for a highly active respective case study period from May 2013. In particular we will examine the field adjustment to the latent heating for both the RAP and HRRR reflectivity assimilation and document sensitivities to strength of the assumed heating. We will also examine relative and combined impact for a variety different storm scenarios and times of day, ranging from near convective initiation time for tornadic supercells to near mature times for larger-scale convective systems.

We will also examine sensitivity to assimilation of radar radial velocity data for both the RAP and HRRR, including documenting sensitivities to the assumed correlation length scale, data quality control, and use of clear-air data. We are also beginning work to test a storm-scale ensemble data assimilation procedure and time permitting will show preliminary results from this work.