Exploration of the NSSL Maximum Expected Size of Hail (MESH) Product for Verifying Experimental Hail Forecasts in the 2014 Spring Forecasting Experiment

Since the 2012 Spring Forecasting Experiment (SFE), near real-time objective verification has been performed in order to complement and enhance the evaluation activities conducted in the Hazardous Weather Testbed (HWT). Similar to prior years, daily activities in the 2014 SFE included an evaluation of the probabilistic forecast products created by the participants on the previous day. For the first time, though, the Storm Prediction Center (SPC) desk produced individual hazard (tornado, wind, hail) probabilistic forecasts for three-hour time windows instead of one probabilistic forecast for total severe. Spatial plots of forecasts and time-matched observations were created with the ability to display the computed statistic alongside the appropriate product on webpages. While the preliminary local storm reports (LSR) served as the primary verification dataset, images from other observation sources were also made available for subjective comparisons. Of these, radar-derived maximum expected size of hail (MESH) from the National Severe Storms Laboratory (NSSL) served as a valuable surrogate for the occurrence of hail given the scarcity of LSRs in low-density population areas.

One of the goals after the conclusion of the 2014 SFE was to explore gridded MESH fields as an alternative and comparison to traditional LSRs. To nearly mimic the procedure during the SFE using severe storm reports, “practically perfect” [PP] hindcasts (Brooks et al. 1998) for MESH were constructed by applying a two-dimensional Gaussian smoother to values above an appropriate threshold (e.g., 1 inch) within 25 miles (i.e. 40-km radius of influence [ROI]) of a 1-km x 1-km grid box. The probabilities from the full-period (20-hr) experimental hail forecasts for Day 1 were compared directly to those obtained from the MESH PP hindcast (i.e., 20-hourly maximum MESH) by calculating the fractions skill score (FSS; Schwartz et al. 2010). In addition, contingency table forecast verification metrics (e.g., CSI) were calculated based on a set probabilistic threshold (e.g., 5%). The value of using MESH as a verification dataset will be judged on both a visual assessment of PP areas against those created from LSRs as well as the scores obtained from the statistical forecast metrics.