Sensitivity of Vegetation Greenness to Multi-Depth Soil Moisture on the Mongolian Plateau: Nonlinear Responses Revealed by RF–SHAP

Under the context of climate change, investigating vegetation sensitivity to soil moisture in arid and semi-arid regions is crucial for assessing ecosystem stability. Nevertheless, how vegetation sensitivity varies among different soil depths, whether it exhibits nonlinear threshold responses, and how such sensitivity has evolved over recent decades remain poorly understood in the Mongolian Plateau. This study focuses on the Mongolian Plateau and uses long-term NDVI data (1982–2022) together with ERA5-Land multilayer soil moisture reanalysis products. A random forest model combined with SHAP (Shapley Additive Explanations), an interpretable machine learning approach, was employed to quantitatively evaluate vegetation NDVI sensitivity to soil moisture at different soil depths. The spatial patterns, nonlinear response characteristics, and temporal evolution trends of vegetation sensitivity were systematically analyzed, and the main driving factors underlying sensitivity changes were identified. The results show that NDVI sensitivity to soil moisture exhibits a pronounced vertical differentiation. Sensitivity is highest in the subsurface layer (7–28 cm), followed by the surface layer (0–7 cm), and lowest in the deep layer (28–100 cm). A significant nonlinear relationship exists between NDVI sensitivity and soil moisture, with vegetation responses markedly intensifying under low-moisture conditions, indicating clear threshold effects. Areas with high sensitivity are mainly distributed in typical grassland and desert-steppe regions, reflecting strong water-limitation characteristics. Over the past 40 years, vegetation sensitivity to soil moisture has shown an overall increasing trend, particularly in semi-arid regions, suggesting an intensifying ecosystem response to water stress. SHAP-based attribution analysis further indicates that the NDVI trend is the strongest positive contributing factor across all three soil moisture layers, revealing a positive feedback mechanism of vegetation greening–enhanced water dependence. Meanwhile, improved surface soil moisture reduces vegetation reliance on deeper soil water, reflecting a distinct interlayer substitution effect. This study systematically characterizes the spatiotemporal dynamics and driving mechanisms of vegetation sensitivity from a multilayer soil moisture perspective, providing scientific insight into ecosystem responses to climate change in arid and semi-arid regions, and offering useful implications for regional ecological management and adaptive strategies.