Atmospheric Correction in Sentinel-2 Simplified Level 2 Product Prototype Processor: Technical Aspects of Design and Implementation

This paper presents the scientific and technical aspects of the Level 2A (atmospheric/topographic correction) for the Sentinel-2 Simplified Level 2 Product Prototype Processor (S2SL2PPP). Design aspects are partly fixed by the ESA as main customer. Together with the alternative atmospheric correction system MACCS, the developed chain based on ATCOR is used for the estimation of the following products: Atmosphere type, Bottom of atmosphere reflectance (including cirrus detection and correction), Aerosol optical thickness, and Water vapor. Being a mono-temporal correction chain ATCOR requires a selection of the spectral bands for the estimation of Aerosol type, Aerosol optical thickness based on the dense dark vegetation method and Water vapor based on the atmospherically pre-corrected differential absorption method as well as an estimation of the best parameter set for these methods. The parameter set was estimated by a sensitivity analysis on a simulated top and bottom of atmosphere radiance/reflectance data based on radiative transfer simulations. The aerosol type is estimated by the comparison of the path radiances ratio to the ground truth path radiances ratio for the standard atmospheres, namely rural, urban, maritime, and desert. Aerosol optical thickness map and Water vapor map are initially estimated on the 20m pixel size data, then the maps are interpolated to the pixel size of 10m and the 10m reflectance data are estimated. The cirrus cloud map is created by the cirrus 1.38 µm band thresholding to the thin, medium, thick cirrus and cirrus clouds. Cirrus compensation is performed by correlating the cirrus band reflectance to the reflective region bands and subtraction of the cirrus contribution per band. Validation of the chain is performed given the top of atmosphere data (as input) and bottom of atmosphere products (the reference). Estimated reflectance is assessed given the ground truth reflectance, Aerosol optical thickness is validated given the AERONET measurements, cirrus correction is validated using a pair of Landsat-8 scenes acquired for the same area with a small time difference. One scene is contaminated by cirrus cloud that has to be restored, while the other is cirrus free and used as reference. A comparison of the estimated products is also performed with an alternative atmospheric correction chain – FLAASH. The software is developed using the Interactive Data Language (IDL) and python. This paper presents the scientific and technical aspects of the Level 2A (atmospheric/topographic correction) for the Sentinel-2 Simplified Level 2 Product Prototype Processor (S2SL2PPP). Design aspects are partly fixed by the ESA as main customer. Together with the alternative atmospheric correction system MACCS, the developed chain based on ATCOR is used for the estimation of the following products: Atmosphere type, Bottom of atmosphere reflectance (including cirrus detection and correction), Aerosol optical thickness, and Water vapor. Being a mono-temporal correction chain ATCOR requires a selection of the spectral bands for the estimation of Aerosol type, Aerosol optical thickness based on the dense dark vegetation method and Water vapor based on the atmospherically pre-corrected differential absorption method as well as an estimation of the best parameter set for these methods. The parameter set was estimated by a sensitivity analysis on a simulated top and bottom of atmosphere radiance/reflectance data based on radiative transfer simulations. The aerosol type is estimated by the comparison of the path radiances ratio to the ground truth path radiances ratio for the standard atmospheres, namely rural, urban, maritime, and desert. Aerosol optical thickness map and Water vapor map are initially estimated on the 20m pixel size data, then the maps are interpolated to the pixel size of 10m and the 10m reflectance data are estimated. The cirrus cloud map is created by the cirrus 1.38 µm band thresholding to the thin, medium, thick cirrus and cirrus clouds. Cirrus compensation is performed by correlating the cirrus band reflectance to the reflective region bands and subtraction of the cirrus contribution per band. Validation of the chain is performed given the top of atmosphere data (as input) and bottom of atmosphere products (the reference). Estimated reflectance is assessed given the ground truth reflectance, Aerosol optical thickness is validated given the AERONET measurements, cirrus correction is validated using a pair of Landsat-8 scenes acquired for the same area with a small time difference. One scene is contaminated by cirrus cloud that has to be restored, while the other is cirrus free and used as reference. A comparison of the estimated products is also performed with an alternative atmospheric correction chain – FLAASH. The software is developed using the Interactive Data Language (IDL) and python.