集合降维变分同化中的初始扰动和局地化  

Initial Perturbation and Localization in Ensemble-Based Reduced-Dimensional Variational Assimilation Method

作　　者：希爽[1,2] XI Shuang(Center for Earth System Modelling and Prediction of CMA,Beijing 100081;National Satellite Meteorological Center,Beijing 100081)

机构地区：[1]中国气象局地球系统数值预报中心,北京100081 [2]国家卫星气象中心,北京100081

出　　处：《气象科技》2022年第5期670-676,共7页Meteorological Science and Technology

基　　金：国家自然科学基金项目(41905031);国家重点研发计划项目(2017YFC1501603)共同资助。

摘　　要：集合降维变分同化方法ERDVar不需要求解切线性模式和伴随模式,不仅能减少同化计算量,而且能够提供“流依赖”的背景误差协方差矩阵。本文提出用NMC初始扰动生成方法和分区同化方案,来解决初始扰动样本生成问题和全球同化局地化问题,最终实现将ERDVar应用到全球中期数值预报模式T106L19。试验结果表明:(1)使用ERDVar方法能够有效提取真实增量信息,提高全球同化精度。(2)用NMC方法产生的扰动样本反映预报误差结构特征,在预报过程中不容易衰减,同化后至少使预报误差降低10%。(3)与全球ERDVar同化试验相比,分区ERDVar同化试验各变量平均的均方根误差降低14%,计算代价进一步降低。分区ERDVar方法和NMC样本的联合应用使同化改进效果更稳定。The Ensemble-based Reduced Dimension Variational(ERDVar) assimilation method can not only reduce the computational cost without solving the tangential model and adjoint model but also provide the “follow dependent” background error covariance matrix.The NMC(National Meteorogical Center,USA) perturbation method and Regional ERDVar(R-ERDVar) are proposed to resolve the initial perturbation and localization in this article.Finally,ERDVar has been applied to the Global Medium-range Numerical Weather Prediction Model T106 L19.The results show that:(1) It is effective to obtain higher accuracy in assimilation using ERDVar,as the information of true innovations is extracted.(2) The NMC initial perturbations reflect the structure of forecast errors and cannot decay easily during forecast subsequently,with at least 10% reduction on forecast errors in ERDVar experiments.(3) Compared with the global ERDVar experiments,there is a 14% reduction for all variable RMSE on average in R-ERDVar experiments,with smaller computational cost.Farther more,the combination use of the R-ERDVar method and NMC perturbation samples can make improvements more stable.

关 键 词：集合降维变分同化方法 初始扰动 局地化 全球中期数值预报模式T106L19 

分 类 号：P456.7[天文地球—大气科学及气象学]