Correction of Directional Effects in Sentinel-2 and -3 Images with Sentinel-3 Time Series and Dart 3D Radiative Transfer Model.

The anisotropy of land surfaces reflectance of greatly impacts the acquisition and analysis of satellite images. Normalization methods using semi empirical kernel-driven models commonly remove these directional effects. Here, we present a physically-based approach to assess the accuracy of 2 normalization methods of Sentinel-2 (S2) and 3 (S3) images. A ReBeLS-based process fits the parameters of a kernel-driven model with S3 and S2 images simulated by the DART RT model in agricultural and urban areas. Here, normalization methods (c-factor, NDVI-disaggregation) hardly validate their assumptions: surface type and orientation have a great influence. We show the potential of RT models to assess and improve normalization methods.