Unraveling nonlinear ecosystem water use efficiency dynamics and environmental drivers across Iran using remote sensing and XGBoost-SHAP

Understanding how ecosystems regulate the trade-off between carbon assimilation and water loss is essential for assessing ecohydrological stability in water-limited regions. This study investigated ecosystem water use efficiency (WUE) dynamics across Iran from 2001 to 2025 using an integrated framework combining multi-source remote sensing datasets, polynomial trend analysis, optimized eXtreme Gradient Boosting (XGBoost) modeling, SHapley Additive exPlanations (SHAP), and Morris global sensitivity analysis (Morris GSA). A polynomial regression approach was employed to characterize linear and nonlinear WUE trends across Köppen–Geiger climate zones and land cover types. In addition, XGBoost-SHAP-Morris GSA modeling identified dominant climatic and environmental drivers, characterized their nonlinear effects, and quantified WUE sensitivity. The results indicated that valid WUE estimates covered approximately 44% of Iran’s total area, of which 23.75% experienced significant long-term WUE trends, with increases (20.39%) substantially exceeding decreases (3.36%). Linear trends were the dominant pattern, covering 17.1% of this area (72.2% of all significant trends), while nonlinear trends accounted for 6.6% of this area. The magnitude of significant WUE trends ranged from –0.97 to 0.90 g C m −2 mm −1 decade⁻ 1 , with increasing trends mainly distributed across western, southwestern, and northern coastal Iran, particularly along the Zagros and Alborz mountain ranges, where changes were more associated with changes in GPP than ET. Among land cover types, grasslands (11.9%) and croplands (5.8%) showed the largest proportions of significant WUE trends. The XGBoost–SHAP analysis revealed that NDVI (50.98% relative contribution), LAI (19.81%), and solar radiation (17.44%) were the dominant factors explaining the variability of WUE. The Morris GSA indicated that modeled WUE responses were primarily sensitive to NDVI (24.00% relative sensitivity, μ* = 0.827), followed by air temperature (20.35%, μ* = 0.701) and solar radiation (16.52%, μ* = 0.569), whereas LST, soil moisture, and wind speed exhibited comparatively weaker overall sensitivity. This study characterizes WUE dynamics across Iran and identifies key environmental factors associated with WUE variability, providing insights into ecosystem water-use dynamics in water-limited regions.