Real-Time Validation of Operational Ocean Models Via Eddy-Decting Deep Neural Networks

Surface currents provided, in real time, by operational ocean models often differ from each other but also from satellite altimetry observations, especially in terms of mesoscale dynamics. Eddies, which play a dominant role on circulation at the regional scale, have a signature on both altimetry maps and satellite imagery, such as sea surface temperature. Combining these independent signatures allows for a highly reliable detection of reference eddies. To this end, we build a convolutional neural network capable of detecting the contours of mesoscale eddies on SST maps in real time. Combined with a standard eddy detection algorithm applied to altimetry maps, we were able to locate and identify with high accuracy more than 900 eddies, in the Mediterranean Sea, over a period of 6 months, and use them as a reference for numerical model validation. We compare as a case study the performance of two operational models: MERCATOR and MFS.