Predicting Future Hail Risk under Climate Change: A Covariate-Based Model Leveraging Machine Learning

Hailstorms are the costliest natural hazard in the United States, causing over $30 billion in insured losses annually. Understanding how the frequency and size of hail might change under varying environmental conditions is crucial for developing effective mitigation strategies in a warming climate. As dynamical downscaling efforts are unable to span greater than decadal periods, typically at end century, we leverage low-dimensional statistical approaches to extrapolate how risk from hail may change in the future. To explore potential changes in future extreme events, global climate models (GCMs) offer valuable insights into how environmental conditions favorable for such extremes may shift over time. To explore such projections, we leverage ERA5 reanalysis data to build a predictive model that links environmental covariates to hail size through the parameters of the Generalized Extreme Value (GEV) distribution. By identifying key covariates that drive hail occurrences, we employ a random forest machine learning approach to predict the GEV parameters, which then allow us to estimate the expected hail return distribution for varying hail sizes under different conditions. To apply this covariate-based model in the context of future climate scenarios, we incorporate data from GCMs drawn from the CMIP6 project, representing a range of potential climate pathways at fixed intervals in the near and far future. However, the environmental conditions of the GCM may contain considerable biases in variables of interest. Distributional bias correction is therefore applied to the GCM data to ensure the model's predictions are consistent with any potential biases. Applying these bias-corrected future climate data, we predict the GEV parameters under specific environmental conditions for two future time periods: 2030-2065 and 2066-2100. These periods are analyzed under two emissions scenarios, SSP 4.5 and SSP 8.5, to assess the impact of moderate and high greenhouse gas concentrations on hailstorm activity. By estimating the return levels of hail size for each scenario and time-period, we can provide insight into the potential changes in hail risk relevant to design standards that could arise in a warming world.