Evaluating severe weather composite indices as diagnostic variables and forecast parameters

Several composite indices (CIs) have been developed to diagnose and discriminate severe weather phenomena, including the energy-helicity index (EHI), supercell composite parameter (SCP), and the significant tornado parameter (STP). This study presents an evaluation of these CIs in the identification of locations in which the so-called “practically perfect” probability (PPP) of severe weather of any type or of a specific type occurring within a specified distance of the location of interest is exceeded. This evaluation compares the CIs as diagnostic variables and forecast parameters. We define a diagnostic variable as one in which its maximum magnitude observed during the evaluation period is considered. On the other hand, a forecast parameter is one in which its observed magnitude a specified time before the evaluation period is considered.

The PPPs are determined using kernel density estimation for various evaluation periods (24-h, 12-h, 6-h, 3-h, and 1-h) and reported on a Lambert conformal 40-km horizontal grid encompassing the conterminous United States. Each day (1200 UTC on the nominal date to 1159 UTC the following day) from 2001–2010 is considered. The fields of the selected CIs are obtained by using North American Regional Reanalysis (NARR) data. The CI magnitudes at each grid point in the domain that features at least one severe report within it from the period 1979–2010 are obtained and compared to the associated PPP. For incremented values of the PPPs and CI magnitudes, binary contingency statistics are computed. If both the CI magnitude and PPP are exceeded, the grid point is counted as a hit. If neither is exceeded, the grid point is counted as a correct null. If the CI magnitude is exceeded, but the PPP is not, the grid point is counted as a false alarm. If the PPP is exceeded, but the CI magnitude is not, the grid point is counted as a miss. Using this approach, the PPP for which the maximum Heidke skill score is observed for a given CI magnitude is determined.

Preliminary findings suggest (1) several CIs exhibit considerable skill in identifying locations with >0.01 PPPs of severe weather of any type or of a specific type, (2) in general, skill is highest for relatively low PPPs and CI magnitudes, (3) both the maximum skill and the PPP for which the maximum skill is observed for a given CI magnitude decrease with increasing temporal resolution, (4) most CIs are highly correlated, resulting in statistically similar skill in identifying PPP exceedance regions, and (5) skill of CIs as forecast parameters is considerably lower than as diagnostic variables.