Evaluation of WRF model forecasts of environmental parameters for severe-weather forecasting from the NOAA HWT Spring Experiments

The National Severe Storms Laboratory (NSSL) and the Storm Prediction Center (SPC), under the auspices of the NOAA Hazardous Weather Testbed (HWT), conducts experiments over a multi-week period during the peak severe convective season, from late April through early June. A primary focus of recent Spring Experiments was the examination of convection allowing (Δ = 2-4 km) configurations of the WRF model covering approximately the eastern three-fourths of the U. S. in a simulated severe-weather-forecasting environment. Output from various configurations of the WRF model were provided to the HWT during the experiment by NSSL, the Center for the Analysis and Prediction of Storms (CAPS), the Environmental Modeling Center (EMC), the National Center for Atmospheric Research (NCAR), and, in 2009, the Air Force Weather Agency (AWFA). Evaluating the performance of convection-allowing models is important for encouraging the appropriate use and development of high-resolution models for both research and mesoscale applications.

These simulations were evaluated during the 2008 Spring Experiment based on their ability to predict the location and timing of thunderstorm initiation and evolution and offer useful information on thunderstorm morphology. However, this study focuses on the ability of the convection-allowing models to predict the meso-β and larger scale pre-convective and near-storm environments from the convection-allowing WRF models. A finding from the Spring Experiment participants is that mesoscale and even synoptic scale errors in the environment were often large enough in the convection-allowing model domains to greatly diminish the predictive value of the explicit 18 to 30 h thunderstorm forecasts.

Objective measures of forecast accuracy (RMS and mean error) confirm these results and find that the forecasts of the mesoscale environment (low-level temperature and dewpoint, CAPE, vertical wind shear) from the convection-allowing models tend to be worse overall than the forecasts provided by the operation NAM and GFS models. Some of this degradation in accuracy comes from a significant negative bias in 2-m and 850-hPa temperatures over most of the United States and smaller than observed vertical wind shear and moisture over the central and southern High Plains. It is suggested that the model physical parameterizations, particularly the boundary layer processes used at convection-allowing grid lengths may be contributing to these errors.

Finally, Despite the control member of a 10-member convection-allowing ensemble having larger errors than the NAM, the ensemble mean errors were smaller than the NAM errors overall. The larger errors in the deterministic convection-allowing forecasts compared to the NAM combined with the relatively small errors in the convection-allowing ensemble mean compared to the NAM suggests that the combination of mesoscale (convection-parameterizing) and convection-allowing model configurations is an appropriate avenue to explore for optimizing the use of limited computer resources for severe-weather forecasting applications.