Gap-Filling of Highly Incomplete Daily Chlorophyll-a Remote Sensing Time-Series Data Over the Eastern China Seas via Spatiotemporal-Periodicity Aware Tensor Completion

The application of daily Chlorophyll-a (Chla) concentration remote sensing data is inevitably affected by large gaps caused by clouds, sunlight, and other undesirable conditions. Existing methods face challenges in addressing large data gaps due to their limited ability to leverage spatiotemporal correlation and periodic features. This study presents a spatiotemporal–periodicity aware tensor completion method for gap-filling highly incomplete daily Chla datasets over the eastern China seas. The method first performs periodic-aware tensor restructuring to align interannual seasonal patterns, then embeds spatial and temporal features through graph Laplacian, fractional temporal differencing, and periodic circulant transforms, followed by adaptive low-rank tensor completion with mode-wise weight optimization. Experiments using 10 years of MODIS-Aqua daily Chla data over the eastern China seas, with an average missing rate of 87%, demonstrate that the proposed method consistently achieves high reconstruction accuracy across multiple years, with correlation coefficients above 0.97, RMSDs below 0.43 mg/m3, and MAPDs under 15.2%. The method remains robust even under missing rates up to 95%. Compared with linear interpolation, DINEOF, DINCAE, CP, and Tucker, the proposed method achieves the highest overall accuracy and superior spatial-temporal consistency, particularly under complex missing patterns. Additionally, ablation results confirm the complementary contributions of the proposed spatial, temporal, and periodic transform operations. The findings indicate that the proposed method provides a promising approach for filling large data gaps in daily Chla remote sensing dataset, which could enhance the application of Chla data in marine ecosystem-related studies.