观测分批变分资料同化技术及其应用

The variational data assimilation and ensemble Kalman filter(EnKF) are the principle methods of current data assimilation including their hybrid method. Each of these methods has its own excellences and insufficiencies, for example, variational data assimilation could not take the advantage of the flow-dependent background error covariance. The reason for this is that variational data assimilation is "global fit", which will encounter large matrix computation and storage, thus the assimilation technology could not be enforceable by the use of flow-dependent background error covariance under the circumstance of a large member of observation. However, the refined analysis requires background error covariance can represent the actual weather flow pattern, in that the background error covariance is the most critical factor of controlling propagation characteristics about the observation information. Thus, this program proposes the Batch-wise Data Variational Assimilation. It is pointed out that the observation is divided into multiple batches and assimilated by each batch. The flow-dependent background error covariance can be applied in the framework of variational assimilation because of the small number of data in each batch. This project mainly includes three parts. Firstly, Batch-wise Data Variational Assimilation technology can be established. Secondly, reasonable estimates for background error covariance can be found, which represents the actual weather flow pattern characteristics. Thirdly, practicality of established data assimilation technology can be analyzed.

当前资料同化的主要方法是变分资料同化方法和集合卡尔曼滤波资料同化方法及其它们的混合方法。它们有各自的优点和不足，其中，变分资料同化方法的不足之处是不能较好地运用"依流型而变"的背景误差协方差，究其原因：变分资料同化方法是"全局拟合"，在观测数量多的情形下，如果运用"依流型而变"的背景误差协方差，会遇到超大矩阵的计算和储存，从而导致不可以实施。然而，精细的分析要求背景误差协方差能够代表实际天气流型，因为它是控制观测信息传播特征的关键因子。所以，本项目研究"观测分批变分资料同化技术"，是将观测分成多个批次，逐批进行同化，由于每批同化的资料数量较少，就可以在变分同化框架中应用"依流型而变"的背景误差协方差。项目主要内容：（1）建立观测分批变分资料同化的技术理论；（2）探讨合理估计背景误差协方差的技术，使其能够代表实际天气流型的特征；（3）对所建立的资料同化技术的实用性和能力进行分析。

常见资料同化的主要方法是变分资料同化方法和集合卡尔曼滤波资料同化方法及其它们的混合方法。它们有各自的优点和不足，其中，变分资料同化方法的不足之处是不能较好地运用"依流型而变"的背景误差协方差，究其原因：变分资料同化方法是"全局拟合"，在观测数量多的情形下，如果运用"依流型而变"的背景误差协方差，会遇到超大矩阵的计算和储存，从而导致不可以实施。然而，精细的分析要求背景误差协方差能够代表实际天气流型，因为它是控制观测信息传播特征的关键因子。所以，本项目研究"观测分批变分资料同化技术"，是将观测分成多个批次，逐批进行同化，由于每批同化的资料数量较少，就可以在变分同化框架中应用"依流型而变"的背景误差协方差。项目按照计划完成了相关研究内容并取得了预期成果：研发了观测分批变分同化技术，建立了可以实际应用的“多网格观测分批集合同化系统”；开发了集合扰动生成及集合预报技术，特别是研发了集合成员概率优选技术，克服坏样本对背景误差协方差和平均值得影响；研发了雷达、卫星等多源观测资料应用技术,针对台风和暴雨的高影响天气过程进行资料同化技术的应用试验，取得了较好的应用效果，表明项目研发的技术具有较好的性能和实用性；初步统计项目完成了16篇学术论文。.