A method for MERIS atmospheric correction based on the spectral and spatial observation

This study aims at developing an autonomous atmospheric correction method, that is exploiting the information content in the image of the satellite considered. The signal recorded by the sensor contains information relative both to the atmosphere and the surface. Aerosol characteristics are the most difficult to evaluate because they vary rapidly with time and space. The spectral variation of the radiance signal, when enough sampled by the sensor, generally allows decoupling aerosol effects from that of the surface. However, on non vegetated areas or region with low vegetation amount, the decoupling is more difficult. For this reason, we propose to use the spatial variation of the signal to better constrain the decoupling process, assuming that the aerosol vary over typically scales of few tenth of kilometers, while the surface varies at shorter distances. A data base was created using Radiative Transfer Model simulations. It contains the satellite top of atmosphere (TOA) reflectances along with the corresponding aerosol optical thickness (AOT). The method is applied to the MERIS sensor, a neural network is then trained at relating AOT to reflectance TOA. Conclusions are drawn on its advantages and limits and possible application to other sensors.