Neuro-variational inversion of ocean color imagery

This paper presents a neuro-variational method to invert satellite ocean color signal. The method is based on a combination of neural networks and classical variational inversion. The radiative transfer equations are modeled by neural networks whose input are the oceanic and atmospheric parameters and output the top of the atmosphere reflectance at several wavelengths. The procedure consists in minimizing a quadratic cost function which is the distance between the satellite observed reflectance and the neural network computed reflectance, the control parameters being the oceanic and atmospheric parameters. The method allows us to retrieve atmospheric and oceanic parameters. We present a feasibility experiment. We show we can retrieve Chl-a with an error of 19.7% if we can obtain a perfect knowledge of three atmospheric parameters. Finally, an inversion of one SeaWiFS image is presented. The Chl-a give coherent spatial structures.