Further development of GSI–based hybrid data assimilation system for convective scale forecast: Assimilation of Reflectivity Data in GSI-based Hybrid Data Assimilation System Using Three Options of Control Variables for the analysis and prediction of 8 May 2003 Oklahoma City Tornadic Supercell Storm

A series of studies had proven the usefulness of radar data assimilation for the convective storms. However, a nonlinear problem in the radar data assimilation appears when the reflectivity observation is directly assimilated. The gradient of the cost function with respect to the low values of hydrometer variables can be dominantly large, such that the minimization has difficulty converging and therefore could cause an imbalance of the gradient among different control variables. To avoid the ill-behavior, another two options of control variables, which are the logarithm values of hydrometer variables (option-logspace) and the reflectivity computed from the hydrometer variables (option-dBZ), were proposed to compare with the control variables with the original values of hydrometer variables (option-original). These options were derived and implemented in the GSI-based hybrid data assimilation system. The impact of these three options, was explored on the analysis and prediction of the 8 May 2003, Oklahoma City, tornadic supercell storm. The experiments were conducted with WRF ARW model at 2-km convection allowing resolution. A 45-member ensemble was used in the GSI hybrid system. The assimilation started at 2100 UTC on the day of the tornadic event. The initial ensemble was downscaled from ensemble analyses of the mesoscale environment. Radar observation including both reflectivity and radial velocity were assimilated every 5 minutes for a total of 1-hour period. All three options reestablished the tornadic supercell much better than GSI 3DVar. The best results were obtained within option-dBZ. The probabilistic forecast of a strong low-level vorticity derived from the ensemble from option-dBZ followed well with the observed tornado track for both the location and longevity of the storm. In comparison, the probabilities with the other options were lower after 25-45 minute lead times, which is not consistent with reality. The tornadic supercell initialized from the hybrid analysis using option-dBZ maintained the strong updraft and vorticity during the entire 1 h forecast period. In comparison, the supercell storm dissipated around 20 and 45 minutes respectively for the option-logspace and option-original. Detailed diagnostics revealed that option-dBZ more correctly analyzed the hydrometeor fields such as the graupel and ice mixing ratio. Such an analysis of the hydrometeor fields led to constructive interaction of the cold pool, the surface gust front and updraft associated with the mid-level mesocyclone. In contrast, such constructive interaction was missing due to overly extensive and overly constraint hydrometeor analyses in option-logspace and option-original respectively. Compared with other options, one advantage of option-dBZ is its flexibility to be applied for various microphysic schemes. The option-dBZ experiment with both the WSM6 and Thompson microphysic schemes were conducted. Both WSM6 and Thompson produced similar low level vorticity forecast and maintenance of the storm. Thompson scheme better forecasted the reflectivity distribution in the forward-flank region of the storm.