南京大学学报（自然科学版）2026，Vol.62Issue(3)：395-408,14.DOI:10.13232/j.cnki.jnju.2026.03.006
融合人工智能模型与多源数值预报的广东台风暴雨预报研究
Study on typhoon torrential rainstorm over Guangdong based on multi-source forecasts from numerical and AI models
摘要
Abstract
Typhoon Danas,the 4th typhoon in 2025,made landfall in Zhejiang Province and was weakened into a remnant vortex,and moved southwestward across Fujian into Guangdong.Under the influence of this vortex,Guangdong Province experienced a widespread,extreme rainstorm on July 10-11.The extensive extreme rainstorm blanketed across the Pearl River Delta on July 10,when the center of the remnant vortex was over Eastern Guangdong.However,the torrential rains began to subside on July 11,as the remnant vortex center moved closer to the region.Traditional numerical models exhibited significant deviations in forecasting the location and temporal evolution of this heavy rainfall.To address the uncertainty in the typhoon rainstorm forecast,objective correction algorithms,such as frequency bias correction and the neighborhood method were applied to optimize the Bayesian model averaging(BMA).By leveraging the strengths of both artificial intelligence and traditional numerical models,a novel probabilistic precipitation forecast is developed based on the combination of the multiple forecasts.The probability of torrential rain predicted by the optimized BMA scheme successfully captured the major rainstorm area from the Pearl River Estuary to eastern Guangdong on July 10.On July 11,the likelihood of torrential rain decreased over the Pearl River Delta.The spatial distribution and temporal evolution of heavy rain probabilities exhibited better alignment with observations,effectively addressing the limitations of deterministic forecasts.A systematic verification of typhoon precipitation affecting Guangdong in 2025 shows that the optimized probabilistic precipitation forecast outperforms the ECMWF ensemble forecast in all verification metrics.These findings could serve as a valuable scientific basis for typhoon rainstorm prevention,mitigation,and emergency decision-making.关键词
台风暴雨/降水概率预报/盘古气象模型/贝叶斯模型平均Key words
typhoon rainstorm/precipitation probability forecast/Pangu-weather/Bayesian model averaging分类
天文与地球科学引用本文复制引用
张超,李超,蔡伟源,陈潜,陈训来..融合人工智能模型与多源数值预报的广东台风暴雨预报研究[J].南京大学学报（自然科学版）,2026,62(3):395-408,14.基金项目
国家自然科学基金(42575008),深圳市科技计划(JCYJ20250604184312017) （42575008）
