Global submesoscale eddy identification and dataset construction based on deep learning algorithms and chlorophyll-a data

Submesoscale eddies are critical in shaping the marine environment and have been the subject of extensive research. However, global submesoscale eddy datasets with long periods and high spatiotemporal continuity remain scarce due to the spatiotemporal resolution limitations of data sources such as satellite altimeters and Synthetic Aperture Radar. In this study, we developed a submesoscale eddy identification method by integrating digital image processing techniques, deep learning algorithms, and chlorophyll data derived from Sentinel-3A/B. Furthermore, we constructed a global submesoscale eddy dataset spanning three years from January 1, 2020, to December 31, 2022. This dataset covers comprehensive information about eddies, including their boundaries and centers with geographic coordinate details, as well as their associated time, polarity, and radii. The identified eddies were compared with temperature and altimeter data. As indicated by the results, our study addressed some limitations of traditional eddy datasets, such as incorrect detections, missed detections, or the merging of two adjacent eddies into one due to the traditional numerical simulation method based on coarse-resolution altimeters. Finally, we statistically analyzed their characteristics, such as their radii, quantity, and spatiotemporal distributions. This dataset, which achieves daily global coverage at a resolution of 300 meters, will significantly address the deficiencies of existing eddy datasets and advance research on submesoscale eddies. The dataset is available at https://zenodo.org/records/17337527.