Developing a neural-network-based “BRDF” tool for the UAE coastal and inland zones

The radiation reflected by any observed surface is highly dependent on both sun illumination and satellite observation angles. These two angles are also described, respectively, as incident and reflected angles. The geometry-dependence of surface reflectance is usually corrected by a tailored Bidirectional Reflectance Distribution Function (BRDF). It is the most common tool used to eliminate or to reduce the effects of sun-sensor geometry on the reflected radiation. Generally, BRDFs are derived empirically (or semi-empirically) for a specific land cover by analyzing a large set of observations (training set) made under different illumination and observation angles. This approach involves fitting the model to collected observations and inverting it. A strong BRDF model tailored to specific land cover characteristics of the UAE is especially needed for applications that use data acquired with variable sun-sensor geometry. In this paper, a neural-network-based tool "BRDF" was developed and applied to quantify the effect of sun illumination and SEVIRI-MSG observation angles on measured reflectance for both land (mostly desert) and coastal water pixels in the UAE.