An algorithm for the retrieval of albedo form nadir reflectance using prior knowledge

A method to derive land surface albedo from near nadir reflectance and a priori BRDF archetypes is presented. It uses kernel-based bidirectional reflectance distribution function (BRDF) models, but assumes a priori knowledge of underling surface BRDFs can be used according to Bayesian inference theory to yield a posteriori estimations of unknown kernel weights based on a BRDF classification. First, MODIS products are used to determine the best weight between observations and a priori knowledge. Then, a lookup table for the weight is built between sun zenith angle, BRDF classification and spectrum bands. Finally, we evaluate the ability of this method to retrieve albedos through 1577 sets of Polder observations which have near nadir (less than 3 degree) reflectance. The results show that, POLDER albedos retrieved from nadir reflectance and Bayes inversion method agree with criterion albedos retrieved from all POLDER observations. Compare to Lambert albedo, this method could improve accuracy of albedo by 5 percent.