How much resolution is enough? Comparing daily WRF–ARW forecasts at 2 and 4 km grid spacing in severe convective weather environments during the 2005 SPC/NSSL Spring Program

The SPC/NSSL Spring Program is an annual collaborative effort designed to explore applied research topics that are of mutual interest to the research and operational forecasting communities. The program is conducted during the peak severe weather season, from mid-April through early June. This year output from three different high-resolution configurations of the WRF model were compared every day during the program. The different configurations of WRF were run remotely and output was collected at the SPC/NSSL facility, where it was displayed within NAWIPS, the operational computational and display software used at the SPC

This study focuses on two of the model configurations. They used the ARW dynamic core and the same physical parameterizations, but they differed in their spatial resolution. Initialization procedures and run-time parameters were set as nearly identical as possible to isolate the impact of resolution on the model forecasts. The first configuration used 4 km horizontal grid spacing with 35 vertical levels. It was run at NCAR. The second used 2 km grid spacing and 51 vertical levels. It was run by CAPS at the Pittsburgh Supercomputing Center. Both configurations were initialized by simply interpolating 00 UTC initial conditions from the Eta model to the high-resolution grids. They used different interpolation routines to accomplish this task and NCAR scientists enhanced the soil moisture initialization using HRLDAS, but the primary difference between these two configurations was the spatial resolution.

This setup provided an unprecedented opportunity to assess the sensitivity of many different severe weather forecasts in convection-allowing models to horizontal resolution. Of particular interest is whether the ~ ten fold increase in computing expense required by the 2 km runs can be justified by added value in the higher resolution forecasts.

Forecasts were evaluated for skill in predicting mesoscale convective structures during the afternoon and evening of the next day (18-30 h forecast period). Both forecasts were available for comparison on about 30 days. Evaluation criteria included the subjective assessments of expert severe weather forecasters and output from objective routines designed with high-resolution data in mind. Results will be presented at the conference and the implications for increasing resolution in operational forecast models will be discussed.