Hailstorm prediction and loss assessment using high-resolution hazard and claims data

This study examines hailstorm hazards and losses in the Netherlands by combining high-resolution radar-based maximum expected hail size (MEHS) data, weather station observations, and reanalysis data with a unique agricultural asset-level insurance loss dataset. We analyse hail climatology to identify the spatial and temporal characteristics of hail events and their impacts on agricultural buildings. Using correlation and regression methods, we assess optimal MEHS thresholds and spatial scales for aggregated damage analysis. The findings indicate that a 1.5 cm MEHS threshold with a one-digit postal code level of aggregation provides the strongest correlation with reported losses. In addition, we utilise random forest models to evaluate the predictive power of underlying meteorological variables linked to convective systems, thus identifying potential proxy variables for hailstorm hazard and associated losses. The random forest evaluation shows that models that use ERA5 data have good overall performance. These models consistently prioritise convective-related variables such as convective inhibition (CIN) and convective available potential energy (CAPE), in addition to the less frequent variables such as (dewpoint) temperature and precipitation. Furthermore, using the entire time series of the period of interest, we can first predict hail days and then use a model to identify days with hail large enough to cause significant damage. These insights help to improve hailstorm risk models and inform strategies to mitigate future impacts.