ABSTRACT Ecosystem water use efficiency (WUE) is a key indicator for evaluating carbon–water coupling and sustainability in climate‐sensitive transitional regions. However, how nonlinear responses and shifting controls across environmental gradients shape the spatial heterogeneity of WUE remains unclear. In this study, we analysed the spatiotemporal dynamics of WUE in the Luanhe River Basin (2000–2020) using the site‐validated PMLV2 dataset and further combined interpretable machine learning (XGBoost‐SHAP) with path analysis to explore the potential ecohydrological mechanisms driving WUE variability. The results indicated that WUE exhibited a significant increasing trend (0.026 gC kg −1 H 2 O −1 yr −1 ), primarily driven by increasing gross primary productivity (average contribution: 75.33%). The normalized difference vegetation index and air temperature were identified as the dominant factors controlling the spatial heterogeneity of WUE, jointly dominating 86.60% of the basin, while soil organic carbon density and vapor pressure deficit played dominant roles in specific regions. WUE responses to environmental drivers showed pronounced nonlinear characteristics, with clear thresholds for precipitation (~600 mm), solar radiation (~6000 MJ·m −2 ), air temperature (5.5°C and 9.8°C), and vapor pressure deficit (~0.78 kPa). Further analysis suggested that WUE heterogeneity reflected differences in dominant controls across landscape units. The relative importance of water limitation, energy constraint, soil mediation, and evaporative demand shifted along environmental gradients. Vegetation dynamics acted as a key mediator linking climatic and soil influences on WUE. This study provides a mechanistic perspective on carbon–water coupling in semi‐arid to sub‐humid transitional regions and offers a scientific basis for region‐specific ecosystem management.
Yang et al. (2026) studied this question.