How Skillful are the CMIP6 Models in Capturing Severe Thunderstorm Environments Over the United States?

Severe thunderstorms are one of the costliest natural phenomena in the United States due to their high frequency and damaging sub-perils, accounting for nearly $30 Billion dollars, annually, of insured loss. Understanding how these extreme events might change is an essential component of mitigating and adapting to a future, warmer climate. Global climate model (GCM) simulations are widely employed to explore changes in future climate; however, present-day GCMs generally have a horizontal grid-spacing that is inadequate to resolve thunderstorms explicitly. Nevertheless, these model simulations can be used to analyze how the environments favorable for severe storms are likely to change, or as forcing models for dynamical/statistical downscaling experiments. The environmental conditions of GCM simulations may contain considerable bias in variables of interest, such as convective available potential energy (CAPE) or wind shear, necessitating a thorough evaluation of the GCMs before they are employed to analyze future projections.

The present study examines severe weather proxies including CAPE, convective inhibition (CIN), 0-1 km and 0-6 km wind shear in 12 GCMs from the Coupled Model Intercomparison Project phase 6 (CMIP6) and evaluate their skill relative to two independent reanalysis products – ERA5 and MERRA2. We also analyze composite parameters such as significant tornado parameter (STP) and supercell composite parameter (SCP) to evaluate the co-variability of component parameters. The analysis reveals that while most models show a reasonable skill in capturing the annual and seasonal distribution of observed parameters, significant bias exists within the ensemble. For instance, several models tend to overestimate CAPE values considerably, specifically over the oceanic regions. This characteristic is very likely associated with the errors arising from the boundary layer and convective parameterization schemes employed in the respective models. The mixed biases among the ensemble in component parameters (CAPE, CIN, and wind shear) consequently impact the skill of composite parameters. Additionally, we compare the distributions of vertical profiles of temperature, humidity, and wind in the models with rawinsonde profiles from Integrated Global Radiosonde Archive version 2 (IGRA2) during convective and non-convective periods to identify the biases in basic model variables. The findings from this study will offer valuable information to choose the right set of models for downscaling applications, and for investigating the future changes in severe storm environments over the US in general.