WCD-05. Advancing Radar Data Assimilation in the HRRR/RRFS

Abstract
Numerical weather prediction (NWP) is an initial/boundary-value problem. Skillful NWP forecasts require accurate initial conditions (IC). For modern-day NWP forecast models IC information is obtained by combining a previous “guess” (supplied by a previous forecast) with contemporary meteorological observations in a process called “data assimilation” (DA). For higher-resolution models that can represent thunderstorms on the model grid (termed Convection Allowing Models, “CAM”s), necessary information for initializing smaller meteorological entities such as thunderstorms are missing from the larger-scale driving models. For convective storms in particular, spatial distribution of hydrometeors, temperature, moisture, and updraft/downdraft intensity within the convective cloud is required for good ICs. One of the only, and arguably the best, available tools to supply this observational information is weather radar; but weather radar only provides some of the information (i.e., total condensed moisture and one dimension of the wind distribution within a storm). This is where DA techniques come in to supply the remaining information. For the past decade or so, the standard method of using radar data was a temperature tendency specification that included a somewhat ad-hoc algorithm based on radar reflectivity. While this method has been successful at improving initial model states for CAM forecasts, the deficiencies in this method have been exposed, and further improvements are sought. Improved radar reflectivity DA methods have been developed, tested, and shown in both idealized and real-data case studies to provide improved forecasts of convective storms in CAMs.GSL collaborators have provided an updated version of one such advanced DA method that uses a mathematically rigorous algorithm to better update the model fields consistent with the physics of convective storms. This method adjusts not only all hydrometeor content and temperature, but also the wind components in such a way that all fields are approximately dynamically balanced, which results in substantially reduced model shock and improved retention of newly assimilated data upon resumed model integration. However, there remain questions as to how best to configure this method. We will present a summary of the various reflectivity DA configurations and examples of additional improvements to case study forecasts as well as the results of a larger-sample forecast experiment.