Derivation and validation of a new kernel for kernel-driven BRDF models

摘要

Kernel-driven bidirectional reflectance (BRDF) models have recently been widely used for mapping albedo with multiangle remote sensing data such as ASAS or temporal AVHRR sequences. An Ambrals algorithm will be used to produce global BRDF and albedo products in the coming EOS era using multiangle reflectance data from the MODIS and MISR. Its operational version currently uses a combination of Ross-Thick and Li-Sparse-Reciprocal kernels which has been validated favorably over other kernels or combinations. To further improve the ability of extrapolation of the Ambrals kernel combination with better physical sense while keeping its data-fitting ability, a new kernel, Li-Transit, is derived and suggested to replace Li-Sparse-Reciprocal kernel in next version of Ambrals. We tested the new kernel combination against the current one and a few alternatives using 29 field collected BRDF data sets. The results show similar data fitting ability and more reliable extrapolation for albedo mapping. A test is also done by using the new combination and the current one to produce temporal albedo change maps of New England of U.S.A. using AVHRR images. Presented also is our recent study on scaling effect of Helmholtz principle of reciprocity, and discussion on application of a priori knowledge in kernel- driven BRDF model inversion.