气象2026，Vol.52Issue(5)：513-526,14.DOI:10.7519/j.issn.1000-0526.2026.030501
"风清"人工智能天气预报模型业务应用性能评估
Performance Evaluation on Operational Application of the AI-Based Global Short-and Medium-Range Forecasting System—Fengqing Model
摘要
Abstract
In 2024,China Meteorological Administration(CMA),in collaboration with Tsinghua Universi-ty,developed the forecasting model following an innovative"AI-Physics"hybrid approach—Fengqing Model.With the designs such as a multi-scale latent space projection architecture and an energy-conserva-tion loss function,the model has been equipped with global short-and medium-range weather forecasting capabilities and has been applied in forecasting operations.This paper comprehensively evaluates the forecasting ability of Fengqing Model in China and the surrounding areas in 2024 from several metrics such as forecast-ing accuracy and bias distribution.Two kinds of typical synoptic processes,i.e.,typhoon and rainstorm,are also focused on deeply exploring the model's performance in the forecasts of disastrous weather.The results show that the 500 hPa geopotential height forecasts by Fengqing Model maintain a predictive skill beyond 10 days.The root mean square errors of 2 m temperature and 850 hPa temperature are significantly lower than those from the European Centre for Medium-Range Weather Forecasts(ECMWF-IFS),having a maximum improvement of 37.66％.In terms of typical weather processes,Fengqing Model demonstrates its superior performance in typhoon track forecasting to ECMWF-IFS,but it underestesmates the typhoon intensity.In addition,Fengqing Model has a good forecasting ability for rainstorm,with smaller forecast errors for the locations of typhoon rain and Meiyu front rainfall belt.The TS score of rainstorm forecasts in the medium-range(73-168 h lead time)is improved by 43.53％compared to that of ECMWF-IFS fore-casts.Overall,the Fengqing Model presents considerable potential in operational forecasting,although further improvements are needed in activity level and typhoon intensity prediction at medium-and long-range lead times.关键词
风清模型/检验评估/灾害性天气Key words
Fengqing Model/verification and evaluation/disastrous weather分类
天文与地球科学引用本文复制引用
李妮娜,罗华坤,龙明盛,王建民,金荣花,宫宇,代刊,曹勇,聂高臻,林建,韦青,吴海旭.."风清"人工智能天气预报模型业务应用性能评估[J].气象,2026,52(5):513-526,14.基金项目
新疆维吾尔自治区重点研发专项(2022B03027-2)、国家自然科学基金项目(42405019)、中国气象局创新发展专项(CXFZ2025J086)及青年创新团队(CMA2023QN05)共同资助 （2022B03027-2）
