Machine learning and remote sensing for modeling groundwater storage variability in semi-arid regions

Abstract

This study investigates the prediction of groundwater Storage in the Rabat-Sale-Kenitra region under climate change conditions using advanced machine learning models. A comprehensive dataset encompassing hydrological, meteorological, and geological factors was meticulously curated and preprocessed for model training. Six regression models Decision Tree, Random Forest, LightGBM, CatBoost, Extreme Learning Machine (ELM), and Artificial Neural Network (ANN) were employed to predict groundwater Storage, with hyperparameters optimized via grid search. The performance of these models was rigorously evaluated using metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). Results demonstrated that the LightGBM model outperformed the others, achieving an impressive testing RMSE of 3.07 and an R² of 0.9997, indicating its robustness in handling large datasets. The Extreme Learning Machine and ANN showed considerable limitations, highlighting the importance of model selection. This research underscores the critical role of advanced machine learning techniques in enhancing groundwater resource management, providing valuable insights for policymakers in developing sustainable strategies to address groundwater challenges in the face of climate variability.