Haze compensation and atmospheric correction for Sentinel-2 data

Kurzfassung

Sentinel-2 data bring the opportunity to analyze landcover at a high spatial accuracy together with a wide swath. Nevertheless, the high data volume requires a per granule analysis. This may lead to border effect (difference in the radiance/reflectance value) between the neighboring granules during atmospheric correction. If there is a high variation of the aerosol optical thickness (AOT) across the granules, especially in case of haze, the atmospherically corrected mosaicked products often show granule border effects. To overcome this artifact a dehazing prior the atmospheric correction is performed. The dehazing compensates only for the haze thickness keeping the AOT fraction for further estimation and compensation in the atmospheric correction chain. This approach results in a smoother AOT map estimate and a corresponding bottom of atmosphere (BOA) reflectance with no border artifact. Digital elevation model (DEM) is employed allowing a better labeling of haze and a higher accuracy of the dehazing. The DEM analysis rejects high elevation areas where bright surfaces might erroneously be classified as haze, thus reducing the probability of misclassification. An example of a numeric evaluation of the atmospheric correction products (AOT and BOA reflectance) is given. It demonstrates a smooth transition between the granules in the AOT map leading to the proper estimate of the BOA reflectance data. The dehazing and atmospheric correction are implemented in the DLR's ATCOR software.