Climate-change projections of hazardous convective weather using an environment-informed, convection-permitting, dynamical downscaling ensemble

Convection-permitting dynamical downscaling (CPDD) allows for an explicit representation of the storm-scale generators of tornadoes, hail, severe thunderstorm winds, and locally heavy precipitation. Possible changes in such hazardous convective weather (HCW) due to human-induced climate change are therefore projected with higher confidence using CPDD than with analyses of relatively coarse global climate models (GCM). However, the computational resources necessary for CPDD are significant and therefore CPDD-based future projections of HCW have tended to be based on a single experiment, and thus absent of uncertainty measures otherwise determined with an ensemble of experiments via an ensemble of GCMs. Herein we present “environment-informed” CPDD as a means to efficiently generate a CPDD ensemble driven by different GCMs. This variant of CPDD is applied only to a subset of days and geographical domains over which the meteorological conditions potentially favor supercell thunderstorms, which are the most frequent generators of significant HCW. The temporal and geospatial occurrence of supercells over the United States is demonstrated from the perspective of environment-informed CPDD as applied to eight different GCMs and ERA5 reanalysis. Such occurrences vary considerably from downscaled GCM to GCM, thus demonstrating the value of an ensemble. Based on the ensemble mean, future supercell occurrence is projected to be most frequent over an area centered on the Missouri Bootheel. An earlier-start to the annual cycle of HCW risk is also projected.

The CPDD ensemble is also used to inform future projections of tornado, hail, and severe wind occurrences. Such occurrences are based on proxies developed using storm reports and CPDD simulations driven by ERA5. Consistent with the supercell projections, we find that tornado, hail, and severe wind occurrences generally tend to increase in the future over the central U.S., and decrease in the future over the southern half of Texas.

We will discuss how this methodology might be applied across Europe, and also how it forms the basis for machine learning applications.