*2.2. Field Data Collection*

The field data were obtained from the 2020 forest resources management "one map" annual update data released by the Ningxia Forestry Survey and Planning Institute. Using these data reduces the workload of field surveys, and it provides access to a large amount of information on ground sample plots. Due to the wide distribution of national surface survey plots, not all sample plots can be surveyed on the spot, and there is a certain degree of uncertainty in these data. Therefore, based on previous studies, the NDVI obtained from Sentinel-2 data was used to screen plots and remove outlier data (NDVI < 0.2) [28,29]. In the end, 881 small class data were extracted for the modeling analysis, and took the hectare stock volume of living trees as the unit area FSV of each sample plot.

Random grouping was used to divide the training data and the testing data. Among the 881 sample plots, 530 (about 60%) were used as the training data and 351 (about 40%) were used as the testing data. Table 1 counts the characteristics of the field FSV training data and testing data, respectively.


**Table 1.** Descriptive statistics of the FSV training and testing data.
