Improving AMBRALS using new GO kernel

Abstract

Kernel-driven BRDF (bi-directional reflectance distribution) model was the core of the AMBRALS, an algorithm for MODIS land surface BRDF and albedo products. In the onboard version of AMBRALS, the LiSparseR Gometrical-Optical (GO) kernel was used. But a new derived kernel----LiTransit kernel is also good at transition from LiSparse kernel to LiDense kernel when zenith angle is large, and accords more to the basic principle of GO model than LiSparseR kernel. Results of validation show: RossThick-LiTransit kernels combination has more stability when extrapolated to large sun zenith angles with LiSparseR kernel. Therefore, we will use the LiTransit kernel instead of LiSparseR kernel in the new version of AMBRALS. The speed requirement of tremendous data processing can't be meet easily, such as MODIS data. Although we can calculate the integration of the kernel beforehand, store up and acquire through look-up table method during retrieving albedo, it's inconvenient for an integrated data processing system. Thus we need to get a simple form of the integration of the kernels. Because the integrations of the kernels are approximately independent on directions than BRDF, it's sufficient to use a polynome dependent on the solar zenith angle to regress the integration of kernel. In this paper, we study the polynome regression of LiTransit kernel to instead LiSparseR, but not affect the systematic of the algorithm at the same time. Comparing the numerical integration of the kernel and the polynome regression result shows the relationship is very well, the polynome can be used in the algorithm directly.