Abstract:

The health of Mongolia's grassland system is related to the efficiency of its livestock husbandry and ecological security at home and abroad. Measuring and controlling livestock grazing density is important for maintaining the health of Mongolia's grassland ecosystems and realizing the sustainable development of the livestock industry. The lack of information on spatial grazing density gradients has hindered the advancement of research related to grassland carrying capacity.This study is based on the 2015 gridded livestock of the world (GLW) dataset, population density, soil moisture, annual precipitation, surface temperature and net primary productivity (NPP). Using the Google Earth Engine (GEE) cloud platform to run the random forest regression algorithm, the Mongolian grazing density estimation model was established. The accuracy of the model was tested based on the statistical data of livestock stocks in the province, and combined with the predictor data of different years, the spatial distribution of the grazing density in Mongolia from 2006 to 2020 was simulated. In order to ensure the accuracy of the dataset, three error measurement indexes of decision coefficient (R²), mean absolute error (MAE) and root mean square error (RMSE) were used to verify the dataset. The simulation results showed that the grazing density in Mongolia from 2006 to 2020 was higher in the north and lower in the south. From 2006 to 2010, Mongolia grazing density expanded significantly, and the proportion of grazing density above 5 TLU/km² increased from 0.223% to 51.390%. There was no significant change in grazing density in most areas of Mongolia from 2010 to 2020. The test results showed that the dataset could well realize the spatial simulation of grazing density in Mongolia. The fitting R² of the simulation data in 2006, 2010, 2015 and 2020 with the livestock stocks in Mongolia province were 0.844, 0.734, 0.914 and 0.926, respectively, which passed the significance test. MAE were 5.195, 3.513, 2.336, 3.461, and RMSE were 8.135, 5.257, 4.200, 5.909, respectively. The grazing density dataset in Mongolia provided by this study provides important information support for the sustainable development of grassland ecosystem and the livelihood security of herders in this region.

Key words: livestock grazing density, GEE cloud platform, Random forest regression model, Mongolia