Speaker
Description
The assessment of groundwater potential in Mediterranean climate regions, such as the Inaouene watershed in northern Morocco, is a strategic priority for sustainable water resource management. In this study, we explore the application of machine learning algorithms to improve the predictive mapping of groundwater potential zones. Sixteen environmental and geological factors influencing groundwater occurrence were identified, extracted, and analyzed using Geographic Information System (GIS) tools. These factors were then integrated into a Python-based modeling workflow developed in the Spyder environment.
Four supervised classification algorithms—Random Forest (RF), Decision Trees (DT), Naive Bayes (NB), and Probabilistic Neural Network (PNN)—were applied to the dataset. The performance of each model was evaluated using standard metrics, including the confusion matrix, Kappa index, and overall accuracy. The Random Forest algorithm demonstrated superior performance, achieving an accuracy of 91.4% and a Kappa index of 0.88, indicating excellent agreement between predicted and observed groundwater potential zones. Decision Trees and PNN achieved moderate results, with accuracies of 83.6% (Kappa = 0.76) and 85.1% (Kappa = 0.72), respectively, while Naive Bayes was less effective (accuracy = 78.2%, Kappa = 0.69).
These results show and prove the robustness and reliability of the Random Forest model for delineating high-potential groundwater areas. The study highlights the advantages of combining GIS-based spatial analysis with advanced machine learning techniques for predictive hydrogeological mapping, providing valuable insights for water resource planning and management in similar climatic regions. Those results could be compared with geophysical data for mapping the potential groundwater areas.
