Spatial-Temporal Pattern Analysis of Grassland Yield in Mongolian Plateau Based on Artificial Neural Network

Accurate and timely estimation of grass yield is crucial for understanding the ecological conditions of grasslands in the Mongolian Plateau (MP). In this study, a new artificial neural network (ANN) model was selected for grassland yield inversion after comparison with multiple linear regression, K-nearest neighbor, and random forest models. The ANN performed better than the other machine learning models. Simultaneously, we conducted an analysis to examine the spatial and temporal characteristics and trends of grass yield in the MP from 2000 to 2020. Grassland productivity decreased from north to south. Additionally, 92.64% of the grasslands exhibited an increasing trend, whereas 7.35% exhibited a decreasing trend. Grassland degradation areas were primarily located in Inner Mongolia and the central Gobi region of Mongolia. Grassland productivity was positively correlated with land surface temperature and precipitation, although the latter was less sensitive than the former in certain areas. These findings indicate that ANN model-based grass yield estimation is an effective method for grassland productivity evaluation in the MP and can be used in a larger area, such as the Eurasian Steppe.

Automated summary

In this study, a new artificial neural network (ANN) model was selected and compared with other machine learning models to accurately estimate grass yield in the Mongolian Plateau (MP). The ANN performed better and showed that grassland productivity decreased from north to south, with 92.64% of grasslands exhibiting an increasing trend. Areas of grassland degradation were primarily located in Inner Mongolia and the central Gobi region of Mongolia. The findings suggest that the ANN model-based grass yield estimation is effective for evaluating grassland productivity in the MP and can be applied more widely.