Ensemble dataset of historical and future U.S. daily large hail 1960 to 2100

Using XGBoost a machine learning model was trained on large-scale environment variables from the European Centre for Medium-Range Weather Forecasts fifth generation reanalysis (ERA5) (Hersbach et al., 2020) and observed hail reports from the Storm Prediction Center's Storm Events dataset (Schaefer and Edwards, 1999) over the period 2000-2021. The model identifies large-scale environmental controls on grid-scale daily probability of U.S. large hail occurrence (hailstones of diameter greater than 25 mm). The hail model is then applied to 10 members of a bias-corrected 1° resolution Community Earth System Model Large Ensemble data (LENS2, Rodgers et al. 2021) under historical and Shared Socioeconomic Pathway 370 future forcing to produce grid-scale daily large hail probabilities from 1960 - 2100. Hail days (value = 1) are assigned when the probability is 0.6 or higher. Non-hail days (value = 0) are assigned when the probability is less than 0.6. The dimension names in the data file are: ensemble: ensemble member, day: days since January 1, 1960, lat (degrees North): latitude of the grid cell center, lon (degrees East): longitude of the grid cell center, Daily_hail_pred(ensemble, day, lat, lon): hail day (=1) or non-hail day (=0). This dataset was funded by QBE Americas, Inc. We also acknowledge computing support from the Casper system (https://ncar.pub/casper) provided by the NSF National Center for Atmospheric Research (NCAR), sponsored by the National Science Foundation.