A Verification Method for Precipitation Forecast Based on Process Identification and Its Application

To address the double‐penalty issue in traditional verification metrics for precipitation forecasting when evaluating high‐resolution models and the limitation of disconnection between evaluation results and practical applications, this study proposes a verification method for precipitation forecasting. This method enables automatic identification and tracking of precipitation processes and can flexibly adapt to the requirements of identifying precipitation processes at various spatiotemporal scales by adjusting precipitation threshold, scale parameters, and matching threshold. On this basis, a multidimensional verification framework encompassing process existence, temporal and spatial attributes, as well as traditional verification scoring metrics is constructed. This framework allows both multidimensional verification for individual precipitation events and statistical tests of forecast biases for multiple events over a long period. An application of this method was conducted on precipitation processes in Xinjiang, China, in 2024. The results show that the traditional threat score (TS) severely underestimates model’s performance due to the double‐penalty problem, whereas the proposed metrics related to spatial and temporal attributes indicate that the forecasts are close to observations. Statistical tests further reveal the systematic biases of precipitation forecasts from both the European Centre for Medium‐Range Weather Forecasts (ECMWF) and the China Meteorological Administration Mesoscale Model (CMA‐MESO). Both models overestimate the proportion of southern‐route precipitation systems in Xinjiang, China, predict later start times and shorter durations by about 6 h, and systematically underestimate the volumetric precipitation by 25%–40%. The proposed method in this study overcomes the limitations of point‐to‐point statistical verification, providing a targeted diagnostic tool for optimizing flood warning timing sequence and improving the model’s representation of physical processes.