Estimating aerosol parameters above the ocean from MERIS observations using topological maps

Abstract

A topological Neural Network (Kohonen map) was used here to retrieve the aerosol optical properties over ocean case 1 waters from the MEdium Resolution Imaging Spectrometer (MERIS) observations. The Kohonen map was first used to aggregate similar spectral MERIS signatures in 400 pertinent groups without any a priori information on the corresponding optical properties of atmosphere and ocean. For this purpose, more than 148.000 spectra taken from 85 MERIS images collected in 2003 and 2004 over the Mediterranean Sea were processed to train the Kohonen map. The 400 groups were then labelled using synthetic spectra (of which the corresponding optical properties are known) generated by the MOMO (Matrix Operator MethOd) code. This algorithm was successfully applied to a set of 5 MERIS images representative of the conditions prevailing in the Eastern Mediterranean. The good spatial homogeneity of the level‐2 atmospheric products – the atmospheric reflectance and the aerosol optical thickness (AOT) – is a first indication of the potential of the presented method. Ground based measurements of the AOT provided by the AERONET (AErosol RObotic NEtwork) program, collected on a small island (Lampedusa) in the Mediterranean Sea, validated our method for the five selected days.