A hybrid deep learning and data assimilation method for model error estimation

作     者：Ziyi PENG Lili LEI Zhe-Min TAN 

作者机构：Key Laboratory of Mesoscale Severe WeatherMinistry of Educationand School of Atmospheric SciencesNanjing UniversityNanjing 210023China Department of Atmospheric and Oceanic SciencesFudan UniversityShanghai 200438China 

出 版 物：《Science China Earth Sciences》 (中国科学（地球科学英文版）)

年 卷 期：2024年第67卷第12期

页      面：3655-3670页

核心收录：

学科分类：07[理学] 070601[理学-气象学] 0706[理学-大气科学] 

基　　金：supported by the National Key R&D Program of China(Grant No.2023YFF0804803) the Fundamental Research Funds for the Central Universities-Cemac“GeoX”Interdisciplinary Program(Grant No.020714380207) 

主　　题：Data assimilation Deep learning Model error 

摘      要：Forecast errors of numerical weather prediction consist of model errors and the growth of initial condition errors,while the initial condition is often optimized based on short-term *** it is difficult to untangle the initial condition error and model error,but it is essential to infer model errors not just for prediction but also for data assimilation(DA).A hybrid deep learning(DL)and DA method is proposed here,aiming to correct model *** uses a convolutional neural network(CNN)to extract characteristics of initial conditions and forecast errors,and then provides estimations for model *** CNN-based model error estimation method can consider the model error resulted from inaccurate model parameters,or simultaneously consider the model error and initial condition *** on the Lorenz05 model,offline and online experiments demonstrate that the CNN-based model error estimation method can effectively correct model errors resulted from inaccurate model parameters,including the forcing F,coupling coefficient c,and relative scale *** both online and offline model error estimations,simultaneously considering model errors and initial condition errors are beneficial to infer the model errors,compared to considering model errors ***,using the observations to verify the forecasts has advantages over using the analyses,to estimate the model *** observations can also achieve a faster convergence of model error estimation with online DA than using analyses.