Monitoring and forecasting groundwater storage dynamics in arid Wadi systems using GRACE/GRACE-FO satellite observations and XGBoost machine learning: Wadi El-Assiuti, Egypt (2003–2030)

Abstract

Progressive groundwater depletion in arid regions represents one of the most critical water security challenges of the twenty-first century. Despite intensive agricultural abstraction in Egypt’s Wadi El-Assiuti since the 1990s, a continuous basin-scale assessment of groundwater storage dynamics has remained lacking. This study integrates GRACE/GRACE-FO Terrestrial Water Storage Anomalies (TWSA), GLDAS-2.2 soil moisture data, CHIRPS precipitation, and multi-temporal Sentinel-2/Landsat land-use datasets to derive monthly groundwater storage anomalies (GWSA) and cumulative variations over the period 2003–2024. Long-term trends were evaluated using the Mann–Kendall test and Sen’s slope estimator, while relationships were quantified using Pearson correlation analysis. GRACE-derived estimates were evaluated against available piezometric records. An XGBoost machine learning model was trained on the period 2003–2016, independently validated over 2017–2024, and subsequently applied for a 72-month forecasting horizon extending to 2030. Results indicate statistically significant declining trends in both GWSA (p < 0.001) and cumulative groundwater storage, with an average depletion rate of approximately − 987,000 m3 yr⁻1. Annual rainfall showed weak, non-significant correlation with groundwater storage (p = 0.471) and no monotonic trend (Sen's slope = 2.08 × 10⁻7 mm month⁻1), confirming climatic variability is not the primary depletion driver. In contrast, agricultural land expansion increased by approximately 273% between 2003 and 2024 and showed a strong negative correlation with groundwater storage (r =  − 0.76,  p < 0001), confirming irrigation-driven abstraction as the dominant control on aquifer depletion. The XGBoost model demonstrated high predictive performance (R2 > 0.92) and projected a continued depletion trajectory under current land-use conditions through 2030. Overall, the integrated framework indicates that anthropogenic land-use change is the dominant anthropogenic factor associated with groundwater decline in Wadi El-Assiuti. The approach is fully reproducible using open-access datasets and is readily transferable to data-scarce arid aquifer systems across the Middle East.