Grassland Aboveground Biomass Estimation Dataset Based on Sentinel-2 Time Series

During the field survey, the center of each subplot was precisely located using a GPS receiver (Garmin MAP 60CS, positioning accuracy ±3 m). A total of more than 650 AGB plot observations were obtained in this survey. After cloud masking and spatial registration of multi-temporal satellite images, plot data with missing or distorted pixel values due to cloud cover were excluded.

Finally, 625 valid in-situ AGB data were selected for subsequent analysis. The measured AGB values ranged from 9.186 to 256.400 g/m², indicating good representativeness of the samples. The table contains 626 rows and 3 columns of data, with the first row as column labels for longitude, latitude, and AGB value, respectively.2、Monthly Composite Sentinel-2 Imagery.All available Sentinel-2 images covering the entire study area from 7 orbits during the vegetation growing season (April–October) from 2017 to 2019 were obtained from the Copernicus Open Access Hub.

To ensure a sufficient number of valid samples, images with cloud cover exceeding 30% were excluded, and a total of 57 high-quality Sentinel-2 images were selected. Ten bands were used for subsequent processing and analysis, including four visible and near-infrared (NIR) bands (10 m spatial resolution), four red edge bands, and two shortwave infrared (SWIR) bands (20 m spatial resolution). The Sen2Cor processor (version 2.11.0) was used for batch atmospheric correction and resampling of all images.

Atmospheric correction converted the top-of-atmosphere reflectance to surface reflectance using a radiative transfer model, and resampling upscaled the 20 m resolution bands to 10 m using bilinear interpolation. On this basis, all images from the same month of the same year were mosaicked to generate seven periods of monthly composite images covering April–October. All composite images were then masked and clipped using the study area vector boundary to construct the Sentinel-2 surface reflectance time series dataset for the growing season.3、PCA-Transformed Imagery.Principal Component Analysis (PCA) was used to reduce the dimensionality of the Sentinel-2 time series data.

The first three principal components retained at least 98% of the variation information in the Sentinel-2 data.4、Model.The model code developed in this study was used.