annual 10-meter resolution spatial distribution dataset of mangroves in China from 2000 to 2024

As critical coastal blue carbon ecosystems, the accurate, large-scale, and long-term monitoring of the spatial distribution of mangroves is essential for evaluating the efficacy of coastal ecological conservation. This dataset was constructed using a dual-network collaborative mapping framework that integrates an attention gate (AG) mechanism and sub-pixel (SP) convolution technology, aiming to overcome the fundamental constraints of scale mismatch between non-homologous multi-temporal remote sensing observations and the scarcity of high-fidelity historical training labels. During the production process, a Teacher network based on U-Net++AG was first trained using high-resolution Sentinel-2 products from 2020 to 2024. Subsequently, a Student network (U-Net++AG-SP) embedded with a sub-pixel convolution module was guided via a knowledge transfer strategy to perform end-to-end super-resolution mapping on the SDC30 seamless data cube from 2000 to 2020. The spatial extent of the dataset covers the mangrove distribution areas along the coast of China, with a temporal span from 2000 to 2024 and a consistent spatial resolution of 10 meters across the entire time series.The dataset is stored in vector Shapefile (.shp) format, comprising annual spatial distribution patches of mangroves alongside detailed administrative attribute information. Each record in the data table represents an individual mangrove patch. The attribute fields encompass both patch geometric characteristics and multi-level administrative associations: the area field records the projected area of the patch (in square meters), and the year field identifies the data year. Administrative attributes refer to the National Standard Administrative Division Codes and the GADM database, including Chinese and English names for provincial (NAME_1), prefectural (地级), and county levels (NAME_3), as well as corresponding administrative codes (e.g., code and GID series fields), facilitating cross-scale spatio-temporal statistical analysis for users.Quality control was conducted through confusion matrix evaluations using an independent validation set, achieving an average overall accuracy of 94% and a Kappa coefficient of 0.87 across the entire time series. Potential errors are primarily concentrated in extremely small newly established mangrove seedlings and degraded edges with sparse canopy cover, which are influenced by the physical resolution limits of multi-spectral imagery and spectral fluctuations caused by tidal inundation. Nevertheless, by incorporating guided filter edge-fidelity technology and a distance-weight-based smooth transition fusion strategy, the dataset effectively suppresses image mosaicking artifacts. While ensuring long-term consistency, it precisely delineates the three-phase structural evolution of China's mangroves—decelerated degradation, steady recovery, and accelerated expansion—providing a high-fidelity spatial data foundation for the assessment of coastal ecological conservation.