Some practical considerations for the first generation of operational convection-allowing NWP: How much resolution is enough?

During the 2005 NOAA Hazardous Weather Testbed Spring Experiment (formerly known as the SPC/NSSL Spring Program) two different high-resolution configurations of the WRF-ARW model were used to produce 30 h forecasts five days a week for a total of 7 weeks. These configurations used the same physical parameterizations and the same input dataset for initial and boundary conditions, differing primarily in their spatial resolution. The first set of runs used 4 km horizontal grid spacing with 35 vertical levels while the second used 2 km grid spacing and 51 vertical levels.

This setup provided an unprecedented opportunity to assess the sensitivity to spatial resolution in the upper end of the convection-allowing range of grid-spacing, during many different severe-weather events. Of particular interest was whether the ~ ten fold increase in computing expense required by the 2 km runs could be justified by added value in the higher resolution forecasts. In this study, we examine and compare these forecasts from several different perspectives. First, we provide a visual examination of simulated reflectivity fields from selected convective events, highlighting the differences that might be detected by an operational forecaster – differences between the two model runs and the ways that both differ from observed reflectivity fields. Next, we present the results of subjective assessments of forecast skill, based on daily ratings assigned by panels of experts during the Spring Experiment. Then, we move on to objective measures of skill. These measures are based on time-averaged behavior characteristics of the models rather than selected points in time and space. For example, we examine the mean diurnal trends of simulated reflectivity and accumulated precipitation fields, as compared with observations. We compare the size distributions of individual reflectivity and precipitation entities, or “storms”, and we look at measures of storm rotation. Further, we look at traditional verification statistics such as equitable-threat and bias scores.

In general, we find that meteorological fields from the two model configurations behave much more like each other than like observations. The 2 km forecasts provide more detailed structures and appear to provide more realistic depictions of supercell-like storm configurations, both of which are intriguing to severe weather forecasters, but neither configuration shows much skill in predicting these small-scale features. On the scales where they show higher levels of skill – the scale of mesoscale convective features – the forecasts are often quite similar. The implications of these results, i.e., the value added by doubling resolution in this context, will be discussed at the conference.