融合DEM与FY-4A数据的ECMWF预报产品深度学习订正方法  

Deep learning based correction of ECMWF forecast products with fusion of DEM and FY-4A data

作　　者：谈玲[1] 刘巧 夏景明[2] TAN Ling;LIU Qiao;XIA Jingming(School of Computer Science,Nanjing University of Information Science and Technology,Nanjing 210044,China;School of Artificial Intelligence,Nanjing University of Information Science and Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学计算机学院,南京210044 [2]南京信息工程大学人工智能学院,南京210044

出　　处：《气象学报》2024年第4期539-553,共15页Acta Meteorologica Sinica

基　　金：国家重点研发计划(2021YFB2901990)。

摘　　要：精准的数值天气预报是精细化气象公共服务和商业服务的重要前提。欧洲中期天气预报中心(European Center forMedium Weather Forecasting,ECMWF)预报产品在全球被广泛采用,但始终存在系统预报误差。针对数值天气预报中的误差和多源数据融合中的非线性映射等问题,设计了一个ECMWF数值预报产品的深度学习订正模型(Numerical Forecast CorrectionNetwork,NFC-Net)。NFC-Net引入了FY-4A卫星观测数据、数字高程模型数据(Digital Elevation Model,DEM)和ERA5历史实况数据订正预报结果,利用多源数据空间分辨率对齐模块、时空特征提取模块解决多源异构数据特征的提取与融合问题,并通过UNet网络实现ECMWF预报产品的订正。为了评估所提算法的性能,利用NFC-Net对ECMWF产品中的2 m气温和10 m风速两个天气要素开展订正试验,并将试验结果与ECMWF预报结果、ANO方法订正结果、Convlstm方法订正结果、Fuse-CUnet方法订正结果和ERA5实况进行对比。结果显示,NFC-Net模型订正的2 m气温和10 m风速的均方根误差(Root Mean Squared Error,RMSE)分别较ECMWF预报产品下降49.71%和50.86%。表明NFC-Net模型可以充分利用多源数据有效改善复杂地形条件下的订正结果。NFC-Net模型可用于订正ECMWF预报结果,显著提升数值天气预报的精度。Accurate numerical weather prediction is an important prerequisite for refined public and commercial meteorologicalservices.ECMWF forecast products are widely used around the world,where as systematic forecast errors always exist.As acorrection of numerical prediction products,multi-source data fusion can effectively reduce prediction errors,which is also a typicalhigh-dimensional nonlinear mapping problem.Due to the heterogeneity of geographic data and ground truth data and satellite data,itis necessary to establish a mechanism to fully extract and utilize effective information from these data while avoid noise andredundancy of the information.In recent years,deep learning methods have been extensively applied to data post-processing inmeteorological field.Aiming at errors in numerical weather prediction and the nonlinear mapping problem in multi-source datafusion,this study designs a correction deep learning model NFC-Net for ECMWF numerical prediction products,which mainlyincludes a multi-source data spatial resolution alignment module,a spatiotemporal feature extraction module,and a UNet correctionmodule.NFC-Net optimizes and corrects the forecast results by integrating multi-source data such as FY-4A satellite data,DEM,andERA5 historical truth data,and utilizes multi-source data spatial resolution alignment module and spatiotemporal feature extractionmodule to achieve feature extraction and fusion for multi-source heterogeneous data.At the same time,this paper also proposes aspatial resolution alignment algorithm based on convolutional neural networks(UPS-MSR algorithm)and a dual self-attentionmechanism(DSA).The UPS-MSR algorithm uses up-sampling and multi-scale residual networks to achieve grid alignment ofmeteorological and geographic data with different resolutions,which can effectively avoid information loss.The DSAConvlstmnetwork embedded in DSA module can balance the spatiotemporal correlation and element correlation when extracting features fromhigh-dimensional meteorological information.To evaluate t

关 键 词：数值天气预报 误差订正 深度学习 多源数据融合 注意力机制 

分 类 号：P435[天文地球—大气科学及气象学]