Verification of the AIFS-WRF hybrid model for precipitation forecasting in a flood period of South China

A Weather Research and Forecasting Model (WRF)-based hybrid system, driven by Artificial Intelligence Forecasting System (AIFS) global fields and constrained by nudging, is evaluated for one-month precipitation prediction during the 2025 South China flood period. Three downscaling experiments driven by global fields serve as benchmarks. The forecast results are verified with traditional verification and the neighborhood method. The results show that the AIFS-WRF hybrid model significantly improves the performance of precipitation forecasting from the heavy rain to extreme rainstorm magnitude. The hybrid model shares similar forecast capabilities with traditional numerical weather prediction (NWP) regional models in short-term forecasts but shows more advantages from the lead time of 4 days. The hybrid model outperforms the NWP regional models and global models with a forecast time gain of 1–2 days and 1–3 days on the heavy rain and rainstorm magnitude, respectively; and exhibits a gain of more than 4 days over the global models on the heavy rainstorm magnitude. A nudging constraint is necessary in the hybrid model. The distribution and intensity of precipitation belts forecasted by the hybrid model are improved when potential temperature is applied in the nudging constraints. 针对当前人工智能(AI)气象模型降水能力不足的问题, 本研究使用以AIFS模型数据为驱动,融入谱张弛逼近(spectral nudging)约束的WRF混合预报模式, 以提升降水预报效果.本研究在2025年华南汛期开展了为期一个月的降水预报试验, 采用传统检验方法与邻域法, 将混合模式与传统数值模式的预报结果进行对比检验.研究发现, 混合模式在大雨至特大暴雨各量级的降水预报性能均有显著提升, 并且在4天及以上预报时效中展现出明显优势.本研究证实了将AI环流场与传统数值模式相耦合的可行性, 同时为极端降水天气预报领域提供了新的思路与方向.