Spatiotemporal Evolution and Driving Forces of Desertification in Typical Semi‐Arid Regions of China Based on an <scp>ML</scp> Model

ABSTRACT Land desertification, a major form of ecosystem degradation, poses severe ecological and socio‐economic challenges worldwide. The Xiliao River Basin, located in the semi‐arid agro‐pastoral transition zone of northern China, has long suffered desertification, making it crucial to clarify its dynamics for ecological restoration. In this study, a Desertification Difference Index (DDI) was reconstructed at 5‐year intervals from 1985 to 2025 by integrating Landsat imagery on Google Earth Engine (GEE) with an Albedo–NDVI feature space. Fifteen natural and socio‐economic indicators were incorporated into an XGBoost model with SHAP interpretability to quantify nonlinear contributions and factor associations. Results show: (1) desertification intensified from 1985 to 2000, with degraded land expanding 2.74% annually, but reversed in 2001–2025, when recovery exceeded degradation at 3.01% annually; (2) strong spatial heterogeneity was evident, with degradation concentrated in the Horqin Sandy Land, while Chifeng and Tongliao showed recovery; (3) the model achieved high accuracy (test R 2 = 0.98). SHAP analysis indicated irrigation water use (21.2%), precipitation (15.7%), and grazing (13.5%) dominated 1985–2000, whereas precipitation (26.5%), cropland reduction (16.8%), and soil moisture (14.7%) were key in 2001–2025. These results reveal interactive mechanisms and temporal shifts between climate and human activities, supporting targeted ecological management in semi‐arid regions.