Detection of three-dimensional structures of oceanic eddies using artificial intelligence

Oceanic mesoscale eddies play an important role in transports of heat, freshwater, mass in the ocean, therefore understanding three-dimensional structure of oceanic eddies is of significance to climate study and oceanic applications. However, detection of three-dimensional (3D) structures is a big challenge though many algorithms of sea surface 2D eddy detection are developed. In this study, we present a novel approach by using 3D U-Net residual architecture (3D-U-Res-Net) to identify 3D structure of oceanic eddies. The sensitivity tests to input variables are conducted to optimalize the input setting. Trained by 3D eddy data provided by a kinetic eddy detection method, the AI-based method can identify different kinds of eddy vertical structures and moreover can dig out more eddy information in deeper layers. This study has significant implications for the further application of the AI-based algorithm in oceanic study.