A new event-scale evaluation framework suggests underestimated skill of WRF-HAILCAST model in climatological hail simulations

Abstract Hail simulation remains challenging in convection-permitting models, and diagnosing bias pathways is essential for targeted improvement. We developed a framework for evaluating the performance of WRF–HAILCAST in hail simulations from the perspective of mesoscale convection. Across 1,502 hail events in China, the model reproduced hail-swath morphology in 50.87% of cases. Among these cases, 53.4% also showed hail-size distributions consistent with observations. Biases emerged at distinct stages of the simulation chain. In 30.36% of events, displacement of the simulated convective system shifted the entire hail footprint, indicating an upstream positional error. Extent biases were not primarily caused by a prolonged hail lifecycle. Instead, both overestimation and underestimation of hail extent were tied more closely to storm realization and hail diagnosis than to CAPE and shear backgrounds, as they depended on where WRF generated mature precipitating cores and where HAILCAST initiated and grew hail. This event-scale framework offers a new perspective on WRF–HAILCAST evaluation and reveals bias pathways often obscured in cell-based verification, tracing large-sample biases across the hail-simulation chain, from large-scale forcing and the mesoscale convective environment to storm organization, hail initiation, and growth.