Retrieval of bidirectional reflectance distribution function (BRDF) at continental scales from AVHRR data using high performance computing

Abstract

We have used high performance computing techniques to implement three different algorithms to model the Bidirectional Reflectance Distribution Function (BRDF) over and from AVHRR data. AVHRR data from the Pathfinder project has a spatial resolution of 8 km, and four years of (1983-1986) was used in this study. Two of the models are statistical models, where the coefficients are derived from a set of directional reflectances for each solar zenith angle by curve fitting using a least square routine. The third model is semi-empirical, and the coefficients are derived by model inversion and numerical iteration. The semi-empirical model is computationally more expensive compared to the other two. One of the statistical models describes surface BRDF as a continuous temporal function using Fourier techniques. Analysis of the standard errors between observed and modeled reflectances from the temporal model show that the errors were larger in higher latitudes, probably due to interannual variations in surface conditions caused by changing snow cover in these areas. Results from the other two models are similar. The results from this study are expected to provide valuable inputs into BRDF retrieval algorithms proposed for future Earth Observation System (EOS) instruments.