GRAPES全球三维变分同化系统——基本设计方案与理想试验  

GRAPES Global 3D-Var System—Basic Scheme Design and Single Observation Test

作　　者：庄世宇[1] 薛纪善[1] 朱国富[1] 赵军[2] 朱宗申 

机构地区：[1]中国气象科学研究院数值预报研究中心和灾害天气国家重点实验室,北京100081 [2]国防科技大学计算机研究所,长沙410073 [3]国家气象中心,北京100081

出　　处：《大气科学》2005年第6期872-884,共13页Chinese Journal of Atmospheric Sciences

基　　金：十五国家重点科技攻关计划项目2001BA607B

摘　　要：中国气象局数值预报研究中心开发了一个全球和区域统一框架的格点三维变分资料同化系统(GRAPES3D-Var)。作者给出全球版本的方案设计,采用单点观测试验检验方案设计的合理性和正确性。全球方案选择流函数、速度势、非平衡高度和相对湿度作为分析求解的控制变量,用线性平衡方程作为质量场和风场的动力约束,通过预调方法解决背景误差协方差矩阵阶数过大和难以求逆的问题,其中垂直变换和物理变换以及观测算子设计等方面与区域方案相同,但预调的水平变换采用球谐函数谱滤波表示背景误差协方差的水平相关模型,避免了递归滤波在高纬和极区因相关尺度过大无法实施的问题。另外,设计了一个使极区插值和差分计算完全闭合的网格以及相应的算法,解决了极点分析问题。单点观测理想试验结果表明,GRAPES 3D-Var系统能够合理给出全球任何地区的分析。With the necessity of improving both medium range and mesoscale numerical weather prediction （NWP） in China, especially solving the problem of merging the remote sense observations into NWP, a 3D variational data assimilation （3D-Var） system with a unified framework for both global and regional model has been developing since April 2001. It is a subsystem of Global and Regional Assimilation and Prediction System （GRAPES） in Center for Numerical Prediction and Research （CNPR）. GRAPES 3D-Vat is a grid analysis model with flexible horizontal Arakawa A grid and vertical pressure levels, in which the horizontal grid is designed to make differential and interpolation algorithms easily implemented over the polar region for global version. For the purpose of achieving the ultimate goal of four-dimensional data assimilation （4D-Var） algorithms in the future, the incremental method is used to enable the implementation of 3D-Vat on a different grid and resolution to that of full model. In order to reduce the tremendous cost of minimization and make it easier to be converged, a preconditioning transform or control variable transform is performed. The preconditioning transform generally consists of horizontal, vertical and change of physical variable transforms. As for the horizontal transform, the recursive filter is used for regional version, while the harmonic spectral filter is used for global version to approximately represent the horizontal background error correlations. The vertical transform is applied via a projection from eigenvector of Empirical Orthogonal Functions （EOFs） of the vertical component of background error onto model levels Finally, the physical variable transformation converts control variables to model variables. The control variables used in the current version are streamfunction, velocity potential, unbalanced mass variable and humidity variable. This univariate and multivariate experiments with single observation over different area are performed to test the validity of the

关 键 词：全球/区域同化预报系统 全球三维变分资料同化系统 谱滤波 理想试验 

分 类 号：P468[天文地球—大气科学及气象学]