Using a multikernel least-variance approach to retrieve and evaluate albedo from limited bidirectional measurements

Abstract

When a multikernel modelling approach is being applied to remotely sensed data, a new criteria — least variance of white-sky albedo — selects semiempirical bidirectional reflectance distribution function (BRDF) kernel combinations better than the more conventional least-squares fitting criteria. The BRDF describes the scattering of light from surface as a function of illumination and view geometry. Semiempirical kernels are nonlinear geometric functions derived from a simplification of physical models of scattering. These are combined linearly to fit observed bidirectional reflectance measurements. White-sky albedo (bihemispherical reflectance) is the integral of directional reflectance for all viewing and illumination directions and is a true surface property. The variance of retrieved white-sky albedo is a function of the noises of measurement, the specific viewing and illumination geometry of the surface scattering measurements, and the number of observations used in the inversion. By selecting the kernel combination that provides the least variance of white-sky albedo, our studies show that a more stable estimate of white-sky and black-sky (directional–hemispherical) albedo is produced.