Atmospheric correction of MERIS data for case-2 waters using a neuro-variational inversion

Abstract

One of the difficulties in analyzing the ocean signal provided by satellite ocean color sensors is that it is strongly polluted by atmospheric contributions, which should be removed by an atmospheric correction process.We propose a new methodology, based on spectral optimization in the near-infrared, to simultaneously estimate the contributions generated by atmospheric signals and oceanic particles, which is valid for case-1 and case-2 waters. This approach, denoted NeuroVaria, combines a neural network to model the radiative transfer with a variational algorithm for the spectral inversion.NeuroVaria was applied to MERIS data recorded between August 2003 and September 2005 over the Adriatic Sea, off the Venice Lagoon, for which, in situ measurements of the water-leaving reflectance and aerosol optical thickness were available. We present comparisons between the results obtained using NeuroVaria and the MERIS second reprocessing (Megs7.4), and those derived from in situ measurements. We show that NeuroVaria achieves better estimations of the aerosol optical properties, and improves the atmospheric correction for case-2 waters. Using MERIS multi-spectral images, it was thus possible to detect typical features of the Po River discharge into the northern Adriatic, as well as suspended sediments due to the shoaling of wind waves on their approach to the seashore shallow waters.