Predicting groundwater level based on remote sensing and machine learning: a case study in the Rabat-Kénitra region

Groundwater is essential for sustaining water needs, industrial growth, agriculture, and ecosystems, particularly in arid regions. This study uses data from GRACE and MODIS satellites, integrating environmental variables like land surface temperature, soil moisture, terrestrial water storage, precipitation, and vegetation indices to predict groundwater levels in Morocco's Rabat-Sal & eacute; Kenitra region. These environmental variables serve as input parameters, with the output being the predicted groundwater level. Advanced machine learning models, including Gradient Boosting Regression (GBR), Support Vector Regression (SVR), Random Forest (RF), and Decision Tree (DT) were employed to capture the relationships between these variables and groundwater levels. The GBR model showed superior performance with an R2 value of 0.99, a Mean Absolute Error (MAE) of 1.94, and a Root Mean Squared Error (RMSE) of 2.98, significantly improving over traditional methods that struggle with non-linear relationships and data noise. Compared to existing methods, our approach offers enhanced accuracy and robustness due to the GBR model's ability to handle complex and non-linear relationships. This study demonstrates the advantages of integrating diverse environmental datasets with advanced machine learning techniques, improving groundwater management strategies and prediction reliability, especially in regions facing significant water scarcity and climate change impacts.