Multi-Temporal Ecosystem Service Classification Products for Xinjiang from 2000 to 2023

Two non-vegetation categories, “bare land” and “others,” were also included. Sample data were derived from the 2021 National Forest and Grassland Resources Survey Database, followed by data cleaning, consistency checks with multi-source land-cover products, manual visual correction, and validation with field-collected samples. Spectral features, terrain (DEM), meteorological variables (temperature, precipitation, potential evapotranspiration), and texture features were integrated into the classification process, which was ultimately completed using the Random Forest algorithm.Pixel values in the classification results range from 0 to 9, corresponding to background fill value, desert, cropland, wind-break/sand-fixation forest, wind-break/sand-fixation grassland, wind-break/sand-fixation shrubland, water-conservation forest, water-conservation grassland, water-conservation shrubland, and other land types.

Using a 30% hold-out sample for validation, all four periods achieved an overall accuracy above 85%; the 2023 product reached an overall accuracy of 83% based on field-sample validation. No systematic missing data are present.The dataset is stored in GeoTIFF format and can be opened and processed using ENVI, ArcGIS, QGIS, and other software. It is suitable for scientific research and practical applications such as analyzing the spatiotemporal evolution of ecosystem services in Xinjiang, ecological protection planning, and ecological management decision-making.