A Machine Learning Based High-Resolution Large Hail Climatology for the Contiguous United States

Over the past decade, hail has been responsible for most financial losses associated with severe convective storms, with costs steadily increasing. As the population grows and increasingly invests in vulnerable infrastructure, the risk of economic damage from large hail also rises. Accurate hazard assessment is therefore essential for effective mitigation.

While recent machine learning (ML) approaches have enabled the development of hail climatologies for hazard assessment, many existing datasets remain limited by regional biases and coarse spatial or temporal resolution.

This study aims to produce a novel, high-resolution large hail (> 2.5 cm in diameter) hazard climatology for the Contiguous United States (CONUS) using the gradient-boosted decision tree algorithm XGBoost. Our model will integrate multiple remotely sensed hail proxy observations, including hourly satellite-derived brightness temperature, radar-reflectivity, and lightning counts, along with key hail-related environmental parameters derived from the ERA5 reanalysis. The high resolution of the remotely sensed data will enable us to train the ML model at an unprecedented grid spacing of 4 km × 4 km, enhancing the spatial detail of large hail climatologies derived in previous datasets.

We will assess model performance using held out test data and compare regional calibration across different climate regimes in the CONUS. Our research will also explore which predictors are most influential for large hail estimation and how effectively the model captures spatial heterogeneities in hail occurrence.

The resulting dataset will represent a refined tool for hail risk assessment in the CONUS. Moreover, with globally available satellite and reanalysis data, this framework also holds the potential for expanded application in other regions around the world.