Near-real-time probabilistic Hail Detection based on polarimetric radar quantities and environmental conditions using machine learning methods

Hailstorms are among the most damaging convective weather hazards in Europe, causing significant losses to infrastructure, vehicles, agriculture, and energy systems. However, reliably detecting and characterizing hail in real time remains a major challenge.

This study aims to improve existing hail detection approaches by combining polarimetric radar observations and environmental data using machine learning techniques. We utilize dual-polarization radar data from the C-band radar operated at the Karlsruhe Institute of Technology (KIT), which provides key variables, such as reflectivity, differential reflectivity, specific differential phase, and correlation coefficient. Convective cells are identified and tracked using the storm tracking algorithm TRACE3D, which performs object-based tracking and incorporates morphological analyses and pattern recognition to extract features indicative of severe convection, including size, shape, echotop height, temporal development, and motion characteristics. These cell objects are linked to environmental predictors derived from numerical weather prediction models, including Convective Available Potential Energy (CAPE), Lifted Index (LI), and storm-relative helicity (SRH), which are known to strongly influence hail formation. Ground-truth verification is supported by hail size reports from the European Severe Weather Database (ESWD), crowd-sourced observations via the DWD WarnWetter app, and insurance claim records. For the estimation of hail probability and hailstone size, we apply machine learning models such as logistic regression, random forest, and convolutional neural networks (CNNs), using combined radar and environmental features as input.

At the time of the conference, we anticipate presenting first results on model performance, predictor relevance, and detection accuracy. The object-based approach of our study enables integration into operational radar systems and supports more accurate, timely hail warnings for improved risk mitigation in weather-sensitive sectors.