Regularized kernel-based BRDF model inversion method for ill-posed land surface parameter retrieval

In this paper, we consider the direct solution of the kernel-based bidirectional reflectance distribution function (BRDF) models for the retrieval of land surface albedos. This is an ill-posed problem due to nonuniqueness of the solution and the instability induced by error/noise and small singular values of the linearized system or the linear BRDF model. A robust inversion algorithm is critical for the BRDF/albedo retrieval from the limited number of satellite observations. We propose a promising algorithm for resolving this kind of ill-posed problem encountered in BRDF model inversion using remote sensing data. chniques for robust estimation of BRDF model parameters are needed to cope with the scarcity of the number of observations. We are reminded by Cornelius Lanczosʹ dictum: “Lack of information cannot be remedied by mathematical trickery.” Thus identifying a priori information or appropriate constraints, and the embedding of the information or constraints into the regularization algorithm, are pivotal elements of a retrieval algorithm. We develop a regularization method, which is called the numerically truncated singular value decomposition (NTSVD). The method is based on the spectrum of the linear driven kernel, and the a priori information/constraint is based on the minimization of the l2 norm of the parameters vector. The regularization algorithm is tested using field data as well as satellite data. Numerical experiments with a subset of measurements for each site demonstrate the robustness of the algorithm.