The main source of Sentinel-2 reflectivity data is downloaded from ESA. The rest of the auxiliary data sources are on-site measurement data and are generated independently.

The Sentinel-2 reflectivity data were not further processed because they had already been processed by ESA. The vegetation biomass data used for calculation of this model uses a 10m × 10m standard sample square, and the collection methods are standard branch collection method and full collection method.

This dataset is based on comparing various machine learning methods such as random forest, gradient regression, and support vector machines, and selecting random forest algorithm for biomass inversion calculation. The data resolution is 10m. The training samples of the Follow-up Forest Algorithm include measured data sets from Hami City in Turpan, Aibi Lake Basin, and the northern edge of the Tarim River Basin in southern Xinjiang. The collection of surface biomass is carried out using standard branches or full harvest methods, and forests are carried out using forest resource survey specifications. There are null values in some areas of the physical data, mainly because the cloud amount in this area is too large in the month, and Sentinel-2 satellite cannot obtain effective images to accurately retrieve the soil salinity in this area, so there are null values.