气象2025，Vol.51Issue(4)：389-399,11.DOI:10.7519/j.issn.1000-0526.2025.032801
临近气象预报大模型"风雷"V1版本检验及个例评估
Verification and Case Evaluation of the"Fenglei"V1 Meteorological Nowcasting Model
摘要
Abstract
Traditional extrapolation techniques,such as the optical flow method,are the main objective methods currently used for nowcasting severe convective weather.These methods fail to represent the gen-eration,dissipation,and evolution of convective systems,resulting in limited forecast validity periods.In 2024,the China Meteorological Administration released China's first AI-based meteorological nowcasting model"Fenglei"V1(hereafter referred to as"Fenglei")."Fenglei"can generate 3 h extrapolation fore-casts based on composite radar reflectivity.The results of quantitative verification on the 2023 data show that"Fenglei"outperforms the traditional optical flow extrapolation algorithms in objective verification scores,with more significant advantages for the forecasts exceeding a lead time of 1 h.Its verification scores decline relatively slowly and flatly,having relatively small Biases within the 3 h forecast lead time.Its TS score for severe and hazardous echo systems has been improved by 33％compared to the optical flow extrapolation algorithm.Case evaluations on the 2024 severe convective events of different scales reveal that"Fenglei"can accurately forecast the generation,dissipation,and evolution of convective systems within a certain forecast lead time.It shows the forecasting capability that traditional methods lack for thunderstorm trend evolution,effectively extending the extrapolation lead time.Thus,"Fenglei"can pro-vide reliable AI-based objective forecast products for the nowcasting of severe convection.关键词
人工智能/风雷/临近预报/检验评估Key words
artificial intelligence(AI)/Fenglei/nowcasting/verification and evaluation分类
天文与地球科学引用本文复制引用
盛杰,龙明盛,王建民,金荣花,张小雯,代刊,张小玲,关良,杨波,张育宸,邢蓝翔..临近气象预报大模型"风雷"V1版本检验及个例评估[J].气象,2025,51(4):389-399,11.基金项目
国家自然科学基金面上项目(42175001)、中国气象局气象能力提升联合研究专项(23NLTSZ002)和中国气象局青年创新团队(CMA2023QN06)共同资助 （42175001）
