【作者】 马寨璞；

【导师】 章本照； 黄大吉；

【作者基本信息】 浙江大学 ， 流体力学， 2002， 博士

【摘要】 本文以三维陆架海模型(HAMSOM)作为数据同化系统的模型部分，深入研究了数据同化的理论方法和在海洋流场数据处理中的应用，对海洋场的数据同化方法提出了行之有效的改进，并将中国黄、东海区域内的TOPEX／POSEIDON(T／P)数据和理论模式的计算结果相结合，进行同化处理，获得了更接近于真实的结果。 对数据同化问题的研究，尤其针对目前我国海区的数据同化应用研究较少这一情况。本文在前人文献的基础上，进行了方法及应用上的研究。这些成果包括：建立考虑了时间相关的方差矩阵和时空相关的最优插值算法；对卡尔曼滤波算法进行了SVD简化以及建立了显式的状态转移矩阵；将T／P实时卫星数据进行调和分析并与数值模型进行同化处理。具体内容为如下几个方面： 1 建立时间相关的协方差矩阵处理方法 在估值类的同化方法中，观测数据与理论模型结合的桥梁是各自的协方差矩阵。在确定协方差矩阵的时候，以往的做法是将所有不同观测时刻的数据当作同一时刻的数据应用；本文在形成协方差矩阵时，不仅考虑空间相关，而且应用了时间相关，对观测数据的应用依照其得到时间的不同分别处理到协方差矩阵的建立中，并根据其在时间上的相关强弱给予不同的权重值，使得对观测数据的应用更接近实际情况。 2 应用动态最优插值算法进行渤海海表温度同化 传统的最优插值算法没有考虑观测数据的时间错位。本文首先推导了既考虑空间相关，又考虑时间相关的最优插值算法，由于应用了时间相关的协方差矩阵，就使这个算法具有动态处理时间错位的观测数据的能力。随后，本文应用理论模型数据和模拟观测数据进行了并行试验，将动态插值算法应用到渤海海表温度场的计算中，同化结果兼顾理论模型结果和模拟观测结果，在方差意义上显示出最优特性，验证了本文所提出的方法改进的正确性。浙江人学博十学位论文全文摘要3应用SVD分解，推导了简约形式的卡尔曼滤波方法，并直接建立了 显式的状态转移矩阵 卡尔曼滤波法无法直接应用于海洋，其主要原因是增益矩阵的计算量太大。本文根据“最大限度保留原来的主要信息，尽可能地简化计算”的原则，应用‘SVD分解推导了简约形式的卡尔曼滤波方法，得到了该方法的简约表达公式，同时，还得到了截断误差部分的伴随发展公式，它可用于探讨截差本身的动态发展。 另外，本文成功应用反向量表述的方法将通常需要迭代得到的卡尔曼状态转移矩阵表述为可直接计算的显式表达形式。 上述两个方面的改进节省了数据同化卡尔曼滤波法的计算量，提高了效率，对时效性很强的海洋预报计算很有意义。4对T/P数据进行调和分析，并将其结果与理论模型数据进行同化处理， 使结果更接近于真实值 对卫星数据如何进行处理和应用是一个重要的课题。本文首先使用黄、东海沿岸60个验潮站实际观测得到的调和常数进行同化处理，有效地’改进了理论模式的结果。随后，本文将美国T/P卫星数据集的第1一第150周期的数据进行调和分析，获得了黄、东海区域的10条轨道上的MZ分潮的调和常数。在此基础上，本文进一步应用最优插值算法，将理论模型结果和卫星数据调和结果进行同化处理。获得了更接近于真实值的同化结果。更多还原

【Abstract】 In this thesis, studies were carried out on the theoretical schemes of data assimilation and their applications on sea fluid field, the three-dimensional baroclinic primitive equation model-Hamburg Shelf Ocean Model (HAMSOM)-being one part of the three ones that constitute the data assimilation (DA) system. Effective modification on schemes of DA were obtained and applications on TOPEX/POSEIDON (T/P) data results that covering the area of Yellow Sea and East Sea were done. The final results became closer to the objective nature.Studies on DA schemes and their applications were detailed on the basis of existed literatures, the studies of applications on China Sea being rather poor at the moment. The final results obtained in this paper contained that the optimal interpolation scheme highlighted by the covariance that the correlation between different time and the correlation between different place being considered; that the simplification of Kalman Filter with the singular-value decomposition (SVD) and the direct construction of state transition matrix pfeceded with "inverse vector expression"; and that the analysis of T/P data and its blending with theoretical model. They are detailed as follows:1. To construct covariance matrix with time correlation and distance correlation.Both observational data and theoretical model results has its respective errors, upon which both covariance matrix being produced. The matrices are the bridge that linkedd the observational data with the theoretical model results together. When made covariance matrix, the observational data were always treated as one group without considering the time misfit, which did not match the true situation. In this thesis not only the distance, but also the time correlation was taken into account when constructed the covariance matrix, the observational data being treated differently according to its obtaining time and the weight, big or small, being assigned to differently according to its time correlation, strong or weak. All that done mentioned above made the application of data be closer to practical situation.2. Applied the dynamic optimal interpolation to Bohai SST.There is no consideration of time misfit in traditional optimal interpolation. In this thesis, some researches were done on the methodology of optimal interpolation, in which a new form of the formulae was developed that named dynamic optimal interpolation, the time correlation being introduced. With the applications of the new method, the weight matrix of observations can be dealt with dynamically while without losing the merits of the traditional ones, and the time correlation can be taken into account in real time when computing the weight matrix, which overcome the problem of time misfit between observations and forecasts. A test arranged to check lately and the result indicates that thenew method can work efficiently, that the results were closer to the objective nature.3. To produce a new reduced KF scheme with SVD, and construct the state transition matrix directly.Many researchers have employed Kalman Filter (KF) scheme to data assimilation, which manifests its validity in blending the model results and data. However, the major difficulty in applying the KF scheme to data assimilation is its enormous size of the system, which leads to the suboptimum in practical applications. Nowadays, many extended KF schemes have been proposed to solve the manipulation of large size of matrix. All the simplifications of KF scheme followed the same rule, that is, kept the major information most possibly and simplified the calculation possibly. In our study, we adopt another new method to reduce the dimensions of state space variables and then obtain a new way to simplify the calculation of KF gain. With the new method, we not only obtain the results that have the same expression forms as those of literatures existed, but also deduce the new equations that can be used to evaluate the abandoned parts quantitatively, which expanded the application of KF and mad更多还原