The practical predictability of convective hazards in a convection-permitting weather reforecast dataset

Abstract:

While convection-permitting (CP) weather forecasts routinely provide useful depictions of convective storm initiation, morphology and intensity, convective hazards (e.g., tornadoes, severe hail, etc.) are more challenging to anticipate with CP guidance since they occur on scales not resolved by typical CP forecast systems. To quantify the variations in the practical predictability of convective hazards, a CP reforecast dataset was generated of ~500 high-impact weather events occurring over the contiguous United States (CONUS) between 2010-2017. For each event, forecasts were generated with 3-km and 1-km horizontal grid-spacing and covered the entirety of the CONUS, partially resolving convection but not hazards. Surrogate diagnostics were used to identify when and where convective hazards would occur, with the forecasts verified using observations of severe hazards (i.e., severe storm reports).

For both the 3-km and 1-km forecasts, variations of forecast skill existed across different environments and regions, often related to the ability of the surrogate fields to identify severe convection. For example, in regions where supercells were prevalent, diagnostics related to mesocyclone strength (e.g., updraft helicity) were skillful predictors of all convective hazards, especially the occurrence of large hail. These diagnostics were less skillful when applied to identifying hazards emanating from other convective modes (e.g., intense wind gusts produced by pulse severe storms), as well as during the overnight hours when skill reached a minimum.

The benefit of reducing grid-spacing from 3-km to 1-km was examined for tornado prediction. The 1-km forecasts were statistically significantly better than the 3-km forecasts when using a surrogate diagnostic related to near-surface vertical vorticity to identify potential tornadic storms. Yet, the benefit was reduced after filtering the 3-km forecasts to remove storms in environments not supportive of tornadoes. This suggests that the finer grid-spacing was beneficial by permitting more intense rotation, better discriminating between weakly and strongly rotating storms, and not due to better placement or depiction of tornadic storms. Thus, much of this benefit could be acquired by using environmental information combined with CP diagnostics without the additional computational expense incurred with 1-km forecasts.