A deep-learning model for gap-filling based on expected variance was developed. OI-SwinUnet achieves good performance reconstructing chlorophyll-a concentration data on the South China Sea. The reconstructed dataset depicts both the spatiotemporal patterns at the seasonal scale and a fast-change process at the weather scale. Reconstructed data show chlorophyll perturbations of individual eddies at different life stages, giving academics a unique and complete perspective on eddy studies.