Validation of Kernel-Driven Semiempirical Models for the Surface Bidirectional Reflectance Distribution Function of Land Surfaces

Abstract The seniempirical, kernel-driven Ambrals BRDF model was developed for correcting and studying view and illuinination angle effects of a wide variety of land covers in remote sensing applications. This model, also scheduled for use in producing a global bidirectional reflectance distribution function and albedo data product from EOSMODI and MISR data, is validated in this article by demonstrating its ability to model 27 different multiangular data sets well, representing major types of land cover. The selection of the kernels used in the model is shown to relate to land cover type, and the inversion accuracy to be good in nearly all cases: the correlation coefficient between modeled and observed reflectances is larger than 0.9 for about half o f the data sets and larger than 0.70 in all but two cases where the observations are irregular. The average root mean squared error of the inversions is 0.034. A new kernel modeling the sun zenith angle dependence of multiple scattering is introduced and shown to improve fits for dense vegetation. Operation of the Ambrals model is demonstrated by applying it to an ASAS image on a per-pixel basis.