Forecasting Groundwater Storage Dynamics in an Arid Region Using Hybrid Machine Learning Under Climate and Land-Use Change: Al-Arish Basin, Sinai

Abstract

A satellite-based machine learning framework was developed to reconstruct and forecast groundwater storage dynamics in the arid Al-Arish Basin, Sinai Peninsula, using GRACE/GRACE-FO terrestrial water storage anomalies (TWSA). The framework integrates bias-corrected regional climate projections and dynamically evolving land-use and land-cover (LULC) scenarios, relying exclusively on globally available datasets to overcome the absence of in-situ groundwater observations. A hybrid ElasticNet–Random Forest model was used to combine long-term and short-term groundwater dynamics. Independent validation (2013–2015) achieved a basin-wide RMSE of 1.72 cm, NRMSE of 11.5%, and showed negligible annual bias (−0.13 cm), indicating strong temporal stability. Projections for 2025–2100 indicate sustained groundwater depletion under both RCP 2.6 and RCP 8.5, with basin-average losses of ∼27.4 cm and ∼34.4 cm, respectively. Depletion is most pronounced before mid-century and spatially concentrated in the eastern basin, emphasizing the importance of early intervention for sustainable groundwater management in arid regions.