基于集合Kalman滤波的同化策略研究

发布时间：2018-10-10 19:57

【摘要】：近年来,气候变化对人类生存影响日益突出,为了加强对陆地、大气以及海洋的监测预测,全球对地观测系统计划(GEOSS)和全球环境与安全监测计划(GMES)等相继被提出。新的数学研究成果不断被引入数据同化算法中,标志着作为连接观测数据和模型模拟预测的关键桥梁的数据同化算法已经得到了迅速的发展。基于集合的数据同化方法近年来得到广泛的重视和研究,已经逐步试验在业务大气数据同化系统中来替代变分类方法。集合Kalman滤波方法高度依赖于集合的大小,在集合数较小的情况下,会出现欠采样、协方差低估、滤波发散和远距离的虚假相关等问题,导致其是一种次优滤波。局地化技术可以有效改善小集合数带来的相关问题。在强非线性Lorenz-96模型基础上,研究有无局地化技术处理的效果差异,从而探讨小集合数条件下的局地化技术的优劣性;提出一种基于功率谱密度(PSD)判断集合数据同化效果的办法。本论文的主要工作概括如下:(1)在有限集合数下,采用Kalman增益值和PSD可以评价同化效果,结合局地化技术,可以获得效率更高的同化算法。(2)局地化技术不仅能消除背景场误差协方差矩阵的虚假相关,还可以增加背景场误差协方差矩阵的秩。(3)协方差局地化方法在更新集合均值和集合扰动上具有较强的鲁棒性。研究结论有助于背景场误差协方差的精细分析和估计。本文通过一系列数值实验对CL和LA两种局地化方法进行了参数敏感性实验,观察不同的观测误差和集合数对同化效果的影响,从而得出了局地分析的思想能有效地解决虚假相关等问题,为实现集合数据同化的业务化提供了有价值的参考。
[Abstract]:In recent years, the impact of climate change on human survival has become increasingly prominent. In order to enhance the monitoring and prediction of land, atmosphere and ocean, the Global Earth observation system (GEO) Program (GEOSS) and the Global Environment and Safety Monitoring Program (GMES) have been put forward one after another. New mathematical research results have been introduced into the data assimilation algorithm, which indicates that the data assimilation algorithm, which is the key bridge between observation data and model simulation prediction, has been developed rapidly. The data assimilation method based on set has been paid more and more attention in recent years, and has been gradually tested to replace the variable classification method in the operational atmospheric data assimilation system. The ensemble Kalman filtering method is highly dependent on the size of the set. When the number of sets is small, problems such as under-sampling, underestimation of covariance, filtering divergence and long distance false correlation will lead to its suboptimal filtering. The localization technique can effectively improve the related problems caused by the small set number. On the basis of strong nonlinear Lorenz-96 model, this paper studies the difference in the effect of local processing with or without localization, and discusses the superiority and inferiority of localization under the condition of small set number, and puts forward a method to judge the effect of data assimilation based on power spectral density (PSD). The main work of this paper is summarized as follows: (1) under the finite set number, the assimilation effect can be evaluated by using Kalman gain value and PSD, combined with localization technique. A more efficient assimilation algorithm can be obtained. (2) Localization can not only eliminate the false correlation of background field error covariance matrix. It can also increase the rank of background field error covariance matrix. (3) Covariance localization method is robust in updating set mean and set perturbation. The results are helpful to the precision analysis and estimation of background field error covariance. In this paper, a series of numerical experiments have been carried out on the CL and LA localization methods, and the effects of different observation errors and the number of sets on the assimilation effect have been observed. It is concluded that the idea of local analysis can effectively solve the problem of false correlation and provide a valuable reference for realizing the business of data assimilation.
【学位授予单位】：西北师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN713

【相似文献】