BRDF Estimations and Normalizations of Sentinel 2 Level 2 Data Using a Kalman-Filtering Approach and Comparisons with RadCalNet Measurements

BRDF estimation aims to characterize the anisotropic behaviour of the observed surface, which is directly related to the type of surface. BRDF theoretical models are then used in the normalization of the satellite-derived observations to virtually replace the sensor at the nadir. Such normalization reinforces the homogeneity within and between satellite-derived time series. Nevertheless, the inversion of the necessary BRDF parameters for the normalization requires the implementation of robust methods to account for the noise in the Level 2 surface reflectances caused by the atmospheric correction process. Here, we compare normalized reflectances obtained with a Kalman filtering approach with i/the classical weighted linear inversion and ii/a normalization performed using the coefficients of the NASA-MODIS BRDF MCD43A1 band 2 product. We show, using the RadCalNet nadir-view reflectances, that the Kalman filtering approach is a more suitable approach for the Sen2Cor level 2 and the selected sites. Using the proposed approach, daily global maps of land surface BRDF coefficients and the derived normalized Sentinel 2 reflectances would be extremely useful to the global and regional climate modelling communities and for the world's cover monitoring.

Automated summary

BRDF estimation is essential for characterizing the anisotropic behavior of observed surfaces and normalizing satellite-derived observations. Implementing robust methods to handle noise in the reflectance data caused by atmospheric correction is crucial for accurate normalization. The Kalman filtering approach shows promise for achieving more suitable results compared to classical methods.