Groundwater on the brink: machine learning forecasts under CMIP6 climate trajectories in Kermanshah Plain

Climate change (CC) threatens groundwater sustainability in Iran's arid and semi-arid regions. This study assesses CC impacts on groundwater levels (GWLs) in the Kermanshah Plain (KP), a key sub-basin of the Karkheh Basin, using CMIP6 projections (SSP126, SSP245, SSP585) downscaled with Random Forest (RF) for 2026–2100. Sixty-four piezometers were clustered into five hydrogeological zones (Airport, Kehriz, Chiamaran, Heshilan, Sarkan) using k-means to enable localized assessment. The model demonstrated high accuracy, with Nash–Sutcliffe Efficiency (NSE) values between 0.46 and 0.73 and correlation coefficients up to 0.82. Soil moisture was the dominant predictor of groundwater level fluctuations, with INM-CM5-0 and BCC-CSM2-MR models contributing 85–89% influence in critical clusters, surpassing precipitation and temperature effects. Historical analysis (1991–2014) showed significant groundwater level declines of 0.9 m/year (Airport), 0.8 m/year (Sarkan), and 0.4 m/year (Chiamaran). Future projections indicate widespread depletion across all scenarios, with spring exhibiting the most severe declines. SSP126 projects maximum spring reductions of 6.9 m (Sarkan) and 6.5 m (Airport) by 2076–2100, despite potential long-term recovery. SSP245 forecasts sustained depletion with 3–3.6 m spring declines by 2100. SSP585 predicts the most severe depletion, with spring groundwater level reductions reaching 4.5 m (Sarkan), 3.9 m (Heshilan), 3.3 m (Chiamaran), 2.6 m (Kehriz), and 3.2 m (Airport). Critically, Airport and Kehriz consistently maintain groundwater level depths exceeding 17–21 m across all scenarios, making them priority zones for urgent management. These projections emphasize the need for adaptive groundwater management strategies integrating machine learning (ML) and climate modeling to protect water security in Iran's vulnerable aquifers under CC. This research underscores the urgent need for such strategies in the Kermanshah Plain, advocating for the integration of advanced climate modeling and machine learning for sustainable groundwater planning.