The SPC/NSSL Spring Program is a collaborative exercise held in Norman, OK during the peak severe convective weather season. It brings together a variety of meteorologists from research and operational communities to investigate specific applied research problems and to promote interactions between the two communities. Subjective verification of NWP models has been an integral part of the Spring Program for several years. Previous experience has shown that it can be a valuable tool for comparing model performance when certain standards are established. During each Spring Program, subjective verification procedures have been systematic, quantitative, and performed by a diverse group of model users; performance criteria have been clearly defined and a variety of meteorological events have been included. These standards help make the subjective verification datasets credible and useful.

Previous subjective verification exercises have focused on mesoscale models with parameterized deep convection. During the 2003 Spring Program, convection resolving numerical forecasts were included for the first time. In particular, explicit predictions of convection were generated with the WRF model, using a relocatable domain with 3 km grid spacing that was centered over the area of most concern to the SPC on a daily basis. Over this same domain, these high-resolution WRF forecasts were subjectively verified and compared with forecasts from 4 different mesoscale models: a coarser resolution configuration of WRF (12 km grid spacing), the operational Eta model, the EtaKF (an experimental version of the Eta routinely used at the SPC), and the new non-hydrostatic mesoscale model (NMM).

Each day the different models were evaluated and assigned verification ratings based on their predictions of convective activity, with an emphasis on severe convection. Furthermore, the two versions of the WRF model were compared on an hourly basis for three specific characteristics of convective activity: timing of initiation, convective evolution, and convective mode (e.g., isolated cells vs. convective lines, etc.).

Several important inferences can be drawn from the resulting subjective verification dataset. The data provide unique information about the value of model output to human forecasters, information that cannot be gleaned from summaries of traditional verification measures. The results have important implications for tailoring model output to best suit the needs of operational forecasters.