Abstract:To address the issue that existing radar-based precipitation extrapolation techniques inadequately capture thunderstorm evolution, leading to systematic biases in heavy rainfall forecasting, this study proposes a heavy rainfall extrapolation and correction technology that integrates thunderstorm evolution characteristics. Building upon the storm identification (SCIT) algorithm, a thunderstorm development trend discrimination model (P-value) is constructed by integrating key factors such as strong gradient zones, new cell generation, and echo area changes for different thunderstorms. By further incorporating composite reflectivity, optical flow vectors, and a dynamic Z–I relationship, physical constraints and corrections are applied to the heavy rainfall extrapolation process.Verification during the 2025 main flood season demonstrates that the corrected product outperforms the SWAN system in terms of TS scores for both 20 mm/h and 50 mm/h intensity levels in the 0–1 hour forecast, with the advantage increasing at higher intensity levels. In the 1–2 hour forecast, the corrected product shows improved scores for both the 20 mm/h and 50 mm/h levels compared to SWAN, yet its own score for the 50 mm/h level remains very low, indicating essentially no forecasting capability for this intensity.. A typical case study further confirms that the proposed method more accurately captures the precipitation characteristics of rapidly moving and developing convective systems, effectively mitigating the "underestimated intensity and limited coverage" issue of traditional extrapolation methods. This technology shows strong potential for operational application. Future work will focus on enhancing its adaptability in complex terrain and data-sparse regions, integrating environmental parameters and machine learning methods to improve the discrimination of thunderstorm initiation and dissipation, and promoting the fusion of deep learning with traditional extrapolation to further enhance the robustness and generalizability of the algorithm.