Abstract：To enhance the forecasting and early warning capabilities for extreme precipitation,the article proposes a DeepSeekbased forecasting and early warning technology system.The system employs an encoderdecoder architecture to design a deep neural network model,combined with intelligent multisource data fusion,threestage transfer learning,and risk index evaluation to achieve regional precipitation prediction and hierarchical early warning.Experiments were conducted using observational data from Hainan Prefecture from 2018 to 2024.Results demonstrate that the model achieved an 81.3% hit rate and a TS score of 0.672,representing a 36.9% improvement in comprehensive forecasting performance compared to the ECMWF model and a 69.9% reduction in positional deviation.These findings validate the technology's potential to assist in disaster prevention and mitigation for extreme precipitation in plateau regions.
张海娥, 钟存, 索南措, 李云霞, 崔玉琰. DeepSeek支持下的极端降水预报预警技术研究[J]. 气象水文海洋仪器, 2026, 43(5): 27-30.
Zhang Haie, Zhong Cun, Suo Nancuo, Li Yunxia, Cui Yuyan. Research on extreme precipitation forecasting and warning technology supported by DeepSeek. Meteorological Hydrological and Marine Instrument, 2026, 43(5): 27-30.
2026, Vol. 43 
