Dissecting the drivers of national terrestrial water storage alterations using non-parametric ensemble learning: The case of Iran

Study region Iran faces water bankruptcy driven by decades of overuse and mismanagement, threatening both its population and ecosystems. Study focus This study quantifies long-term terrestrial water storage loss (TWSL) dynamics across Iran by integrating GRACE satellite observations, extensive ground-based hydroclimatic records, and an explainable Random Forest model enhanced with Nataf-based data augmentation. We used SHapley Additive exPlanations (SHAP) to identify the relative contributions of climatic variability and anthropogenic forcing. New hydrological insights for the region Results show pronounced hydroclimatic change between 1973 and 2023, indicating a significant increase of 2.05°C in air temperature and a 43 mm reduction in average annual precipitation between 1973–2000 and 2001–2019. The average annual streamflow at unregulated stations decreased by 37% during 1998–2018 relative to 1970–1997. Groundwater storage deficit over 2003–2018 was 83 BCM. GRACE data indicate a TWSL of about 370 BCM from 2002 to 2023, with 61% occurring after 2016, reflecting accelerating depletion. SHAP analysis revealed that groundwater dynamics dominate TWSL, explaining 90.5%, 91.9%, and 92.3% of its variability during normal, dry, and wet conditions, respectively, whereas the influence of other factors remains limited under all climatic conditions. This dominance indicates a decoupling of TWSL from short-term climate variability and a transition toward groundwater-controlled storage dynamics. These findings highlight the need for demand management and effective control of groundwater abstraction from aquifers to ensure long-term sustainability.