Assessment of the Statistical Characteristics of Thunderstorms Simulated with the WRF Model Using Convection-Permitting Resolution

The short-term (0-8 hour) prediction of thunderstorms and their characteristics—organization, severity, orientation (for squall lines), storm spacing—is vital air traffic flow management. Nowcasting systems based on radar observations have skill at predicting storm characteristics in the 0-2 hour time frame; however, this skill rapidly declines owing to the difficulties of nowcasting storm initiation and evolution. High-resolution convection-permitting simulations with numerical weather prediction (NWP) models, which no longer parameterize the convection, are able to resolve storm structures and organization; however, the skill of NWP (when comparing pixel-by-pixel at fine scales) remains rather low owing to intensity and phase (space and time) errors. Note that assimilation of radar data improves the model forecasts at short lead times, although at this time data assimilation cannot outperform extrapolation forecasts with lead times of less than 2 hours. While the skill of NWP in predicting the actual location of storms may be poor, the high resolution runs have apparent skill in predicting the timing, evolution and organization of storms. The goal of this study is to determine whether high resolution NWP has skill at predicting the statistical characteristics of storms relevant to air traffic flow management at lead times > 3 hours (e.g., relevant for strategic planning). Data from the Weather Research and Forecasting – Advanced Weather Research (WRF-ARW) model run in real time twice per day (00 and 12 UTC) at 4 km resolution on the NCAR supercomputer are used to evaluate the ability of convection-permitting simulations to produce realistic storm structures and their evolution. In this study we focus on the model's ability to simulate characteristics of a range of storms types that occurred in the Southeastern US during the entire month of July of 2006. The TITAN (Thunderstorm Identification Tracking and Nowcasting) software has been extended to determine storm characteristics (e.g., storm coverage, size distribution, orientation, spacing, interconnectedness, etc) in model data and the observations (using the WSR-88D national mosaic provided by WSI). The model tended to have overall skill at predicting the location of storm areas and their propagation speed; however, significant phase errors existed. The model tended to overpredict the number of small “air-mass” type storms while the mean spacing between the storms was realistic. A detailed comparison of the modeled and observed convective storm characteristics will be given in the presentation.