Estimation and Prediction of Hydrological Variables Using Machine Learning Algorithms for Groundwater Management: ErfoudRadier Station in Morocco

Accurate prediction and estimation of missing values for hydrological and meteorological variables are crucial for the sustainable management of water resources in arid and semi-arid environments, particularly in Morocco. This work aims to evaluate the effectiveness of several machine learning algorithms for the prediction and estimation of missing mean temperature values at the ErfoudRadier station in the GZR basin. Five Machine Learning algorithms were applied and compared: Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), and K-Nearest Neighbors (KNN). Performance was evaluated on test data using evaluation metrics including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the Coefficient of Determination (R2). The comparative analysis highlights the superiority of the ANN and SVM models. The ANN model exhibits the lowest errors (RMSE = 2.954; MAE = 1.863) and a coefficient of determination nearly similar to that of the SVM (R2 = 0.878), while the SVM achieves the highest coefficient of determination with R2 = 0.8822. The RF (R2 = 0.820) and KNN (R2 = 0.821) models show more modest performance, and the Decision Tree (DT) produces the highest errors (RMSE = 4.765), confirming its limited suitability for modeling this type of continuous data. The results confirm the effectiveness of machine learning models, particularly ANN and SVM, for modeling hydrological and meteorological variables at the Erfoud station, significantly improving the reliability of forecasts and enabling more informed water management.