基于扩散滤波的多尺度三维变分研究

摘要

将扩散方程引入三维变分分析,揭示了传统3D-VAR不能有效提取多尺度观测信息的根本原因,即观测导致的目标函数梯度在空间的不连续分布.将扩散滤波融入基于梯度的最优化算法,发展了基于扩散滤波的多尺度3D-VAR.海表面温度数据同化试验结果表明,新方法可从长波至短波有效地提取多尺度的观测信息.%The diffusion equation is introduced into 3D-VAR analysis in this paper. As it reveals, the root cause of the inefficiency of the traditional 3D-VAR scheme in extracting multi-scale information resolved by observations lies in the discontinuous distribution of gradient of the cost function caused by the inhomogeneous distribution of observations. A multi-scale 3D-VAR scheme is developed by incorporating diffusion filtering into gradient-based minimization algorithm. The new scheme is applied to a two-dimensional idealized SST assimilation experiment with real AVHRR SST observations, which shows a superior performance and the multi-scale observational information,and from longer to shorter wavelengths, such observational informationcan be extracted successively by this scheme.