Multi-scale eddy identification and analysis based on deep learning method and ocean color data

Oceanic eddies are ubiquitous phenomena that facilitate the horizontal and vertical transportation of substances under the combination of ocean currents and the Coriolis force, thus they are crucial for the regulation of marine ecosystems. Owing to the spatial resolution limitations of altimeters, the detection of submesoscale eddies remains challenging. While instances exist in which submesoscale eddies have been identified using data such as synthetic aperture radar imagery, they are currently limited to near-shore regions. Additionally, certain deep-learning-based methods failed to delineate the contours of eddies when applied to these data. In this study, we proposed a multi-scale eddy detection neural network algorithm (MED-Net) for eddy identification and segmentation task in chlorophyll fields. The algorithm integrates high resolution ocean color data, digital image processing, artificial intelligence, and multi-scale object detection technologies. The results indicate that MED-Net can accurately identify eddies across multiple scales in chlorophyll fields, outline their shapes, and address certain limitations in traditional eddy datasets. This study has significant implications for identifying cross-scale eddies, facilitating morphological analysis of submesoscale eddies, complementing and correcting eddy datasets. Furthermore, it is beneficial to investigate the multi-scale cascade effect of eddies and understand the ecological processes of submesoscale eddies throughout their entire life cycle.