A deep learning-based mesoscale eddy subsurface temperature inversion model

Satellites provide a unique opportunity to study the surface expression of mesoscale eddies due to their better spatiotemporal coverage compared to in situ observing systems. However, the three-dimensional structures of mesoscale eddies are still unclear due to the low spatial and temporal resolution of in-situ data. To better understand the vertical structure of mesoscale eddies, deep learning (DL) based mesoscale eddy subsurface temperature inversion model is developed in this study. Combined with the measured data and satellite sea surface data (sea surface height, sea surface temperature), the DL-based model was used to develop the inversion model of the subsurface temperature of mesoscale eddies. The DL-based model can accurately retrieve the subsurface temperature within eddies from the remote sensing data, solves the sparse targeted observations on mesoscale eddies, and help to fully reveal the characteristics variations of three-dimensional structure within mesoscale eddies.