The Najd region of the Sultanate of Oman experiences critical depletion in its groundwater resources due to the hyper-arid climate, rapid population growth, and high demand for agricultural water. In this research, six deep learning and machine learning models are used to estimate Groundwater Storage Anomaly (GWSA) in the Najd region based on monthly Terrestrial Water Storage Anomaly (TWSA) data obtained from the Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On (GRACE-FO) satellite missions and hydroclimatic variables of the Global Land Data Assimilation System (GLDAS) from April 2002 to February 2026. The models are CNN, ERNN, GRU, LSTM, MLP, and TCN. The mass balance approach is adopted to isolate GWSA by removing soil moisture anomalies from TWSA. Linear interpolation is used to fill 35 missing GRACE months, resulting in a complete 287-month continuous time series. The analysis is limited to the 123 GLDAS cell centers located inside the Najd boundary. These cells represent 14 independent CSR mascon-signal groups, divided into 10 training groups (87 cells), 2 validation groups (18 cells), and 2 testing groups (18 cells), with no group overlap. Eight GLDAS variables are retained, in addition to latitude, longitude, normalized time, and cyclical month. The basin analysis indicates a statistically significant declining GWSA trend of -0.171 cm/year. Among all models, CNN achieves the best independent spatial-validation performance with MAE = 0.902 cm, RMSE = 1.131 cm, R² = 0.798, and KGE = 0.865. The findings provide useful information for groundwater monitoring in the Najd region and similar hyper-arid environments.

Keywords

Artificial Intelligence, Deep Learning, GRACE, GLDAS, Groundwater Storage