Model-Based Neural Network to Retrieve Ancillary Information About Sea Oil Slicks

In this study, a model-based neural network approach is proposed to retrieve ancillary parameters related to oil pollutants at sea. The proposed methodology consists of two pillars. First, an electromagnetic scattering model is used to generate radar backscatter for slick-free and slick-covered sea surface at variance of incidence angle, faction of water into the oil and oil thickness. Then, these radar backscatter values are combined to generated a metric adopted fro the retrieval process, namely the damping ratio. Second, an artificial neural network is first trained on the simulated damping ratio DR and then applied to actual synthetic aperture radar imagery to retrieve oil thickness and seawater volume fraction. Results, obtained processing synthetic aperture radar scenes collected during the Deep Water Horizon oil accident by the L-band uninhabited aerial vehicle synthetic aperture radar (National Aeronautics and Space Administration - Jet Propulsion Laboratory), show the soundness of the proposed methodology.