Development of an operational procedure to estimate surface albedo from the SEVIRI/MSG observing system by using POLDER BRDF measurements: I. Data quality control and accumulation of information corresponding to the IGBP land cover classes

Abstract

In this paper, we present an operational procedure for the inversion of kernel-driven bi-directional reflectance distribution function (BRDF) model and further albedo retrieval to be applicable to the SEVIRI/MSG reflectance measurements. Our approach aims at bringing solutions to an ill-posed problem that is when the ratio of extreme eigenvalues of the inverse matrix is of several orders of magnitude due to sparse and irregular angular samplings. The geometric Li-Sparse reciprocal and Roujean kernels are considered in the procedure applied to space-borne Polarization and Directionality of Earth Reflectance (POLDER) sensor data which decomposes as follows: (1) quality control, (2) accumulation of a priori information on model coefficients of directional hemispherical reflectance, and (3) implementation of BRDF model inversion methods based on the biased estimation instead of usual nonbiased least solution due to its large variance. The first two steps are detailed in this paper; the third one appears in a companion paper. The data control procedure consists both in filtering inputs of reflectance data and output of model coefficients. In the first stage, the analysis criteria rely on Fisher statistics. In the second stage, a procedure is applied to red and near-infrared POLDER data corresponding to the 16 classes of International Geosphere Biosphere Program (IGBP) land cover classification. Among the retained criteria, some are based on the BRDF shape. They encompass (1) T-statistics, (2) the bowl shape index (BSI), (3) the dome shape index (DSI), (4) the white sky albedo (WSA, bi-hemispherical reflectance), and (5) the black sky albedo variance (BSA, directional–hemispherical reflectance). Statistical results include mean values and covariance matrix for the spectral BRDF model coefficients. This procedure eliminates about 10% of data measurements, thereby reducing uncertainty in retrieved BRDF model coefficients and WSA.