Multi-timescale streamflow projections in human-regulated basins of southwest Iran using CWatM and explainable machine learning

The Karun-Karkheh-Marzi-e-Gharb river basin complex, a strategically vital but data-scarce and human-regulated region in Western Iran characterized by semi-arid conditions and complex topography. This research evaluates future hydroclimatic shifts by integrating the Community Water Model (CWatM) with a dual-benchmark calibration framework. The study utilizes a computationally efficient strategy—reducing runtime by 55%—to enable a 30-member ensemble projection (General Circulation Models) without high-performance computing. To address anthropogenic non-stationarity and data constraints, the model was validated against both a naturalized benchmark (KGE=0.71) and regulated local observations (KGE=0.60). Furthermore, SHAP-based sensitivity analysis, validated by high-accuracy surrogate models, including XGBoost (R2=0.98, RMSE=0.04) and Linear Regression (R2 =0.95, RMSE=0.1), was employed to quantify the influence of physical and anthropogenic drivers on the basin's hydrological response. Projections reveal a consistent hydrograph transformation driven by thermal forcing across all scenarios. Warmer winters are projected to reduce snow storage, advancing peak flow timing toward late winter. Crucially, intensified evapotranspiration is found to outweigh precipitation variability, shifting the basin from a storage-controlled to an evaporation-dominated system. This leads to declining annual water yields and escalating late-century drought extremes. These findings provide a refined, physically-based projection that potentially extends beyond traditional national climatic reports, suggesting that regional water security should prioritize demand-side management and a transition from stationary allocation rules toward adaptive, climate-resilient strategies. • Dual-benchmarks quantify human-induced non-stationarity in data-scarce basins. • Early runoff shifts peak flow timing, impacting long-term reservoir recharge. • Thermal forcing outweighs precipitation in driving late-century drought extremes. • SHAP values reveal a shift from storage-controlled to evaporation-dominated. • Automated pipelines reduced CWatM simulation runtimes by 55%.