Predictive modeling of diverse factors impacting regional soil erosion degree with machine learning.

Soil erosion is an obstacle in the process of maintaining and restoring the ecological environment in desert steppe landforms. We studied soil erosion in typical desert steppe landform areas in the northern foothills of the Yinshan Mountains. We used differences in altitude maps from different years to construct a dataset for soil erosion research, combined with data on precipitation, human activities, land cover, topography, and soil properties. We employed a convolutional neural network to learn and predict ASED and SEDD in different regions of the Yinshan Mountains. The trained model provided satisfactory prediction performance. We also determined the contributions of different factors and years to soil erosion prediction. Human activities had a greater impact on ASED prediction, while the impact of various factors on SEDD prediction was more balanced. Topography and precipitation factors had a relatively small effect on ASED prediction but played a prominent role in SEDD prediction. This study provides a new approach to exploring the mechanism of different factors on soil erosion in the Yinshan Mountains' northern foothills. [ABSTRACT FROM AUTHOR]