Mechanisms Disturbing Forest Water Yield Stability in the Upper Yellow River Under Environmental Stress: Insights From Explainable Machine Learning

ABSTRACT Forests in the Upper Yellow River Basin (UYRB) function as critical headwater ecosystems where water‐yield stability (WYS) is a priority management concern given accelerating climate change and intensifying anthropogenic pressures. However, current assessments often treat the basin as a homogeneous whole, thereby obscuring the distinct hydrological stability mechanisms of forest ecosystems. Furthermore, reliance on linear assumptions or single, pre‐determined models limits the ability to capture complex nonlinear dynamics and introduces structural uncertainties. To bridge these knowledge gaps, we focused specifically on forest ecosystems using multi‐source data (2000–2020) and a comparative modeling framework—systematically evaluating XGBoost, Random Forest, and CatBoost to identify the optimal model annually. We further elucidated spatiotemporal WYS dynamics and controls using shapley Additive Explanations (SHAP), partial dependence plots (PDPs), and bivariate dependence plots (BDPs). WYS was primarily regulated by external environmental gradients rather than intrinsic variability. Stable zones were concentrated in western Aba Prefecture, whereas unstable zones predominated in northern Haibei Prefecture. Precipitation (PRE) emerged as the dominant driver, while other variables (TEMAVE, TEMMAX, DEM, DRS, FVC, LAI, and HFP) exerted significant nonlinear threshold effects. Key thresholds included PRE = 709.88 mm, FVC = 48%, TEMAVE = 3.08°C, TEMMAX = 10.37°C, DEM = 3498 m, LAI = 0.58–1.38, and HFP = 5.54. These quantitative benchmarks inform adaptive, spatially explicit forest and water management strategies in the region.