Assessment of Groundwater Potential in a Mountainous Area Using Machine Learning and GIS, Rherhaya Basin, High Atlas, Morocco

Abstract

Mountainous regions are vital for recharging aquifers in plain areas downstream, but understanding the geological, hydrogeological, and climatic factors is crucial to comprehend groundwater processes in these regions. Several parameters, including lithology, topography, secondary porosity, geological structures, and climatic conditions, affect the potential of groundwater in mountainous aquifers. Traditional groundwater modeling tools face several challenges in handling large amounts of real-time data, such as extracting useful features, quantifying uncertainty, and identifying links between different variables. Recent technological advances in artificial intelligence, particularly machine learning, provide solutions for hydrogeological research and applications. This paper focuses on modeling potential zones of groundwater sources using various methodologies based on GIS, spatial remote sensing, and machine learning. The study evaluated three models, Random Forest, Support Vector Machine, and Logistic Regression, in identifying potential groundwater zones in the Rherhaya watershed. More than 200 localized spring points were needed to ensure efficient model learning. The Support Vector Machine model demonstrated the highest performance during the 70/30% split, with a ROC-AUC of 84.4% for the test data. The study identified four critical conditioning factors of groundwater potentiality, including Topographic position index, River Distance, Valley Depth, and Plane Curvature. The models also highlighted the distance to rivers as a significant factor, particularly in the upstream portion of the watershed. The very low potentiality class occupied the largest area (over 32%), followed by low (between 24 and 29%), moderate (12–19%), high (10–14%), and very high (only 9–12%) classes. Only the Support Vector Machine model predicted that 12% of the catchment area had high potential for groundwater resources, indicating its superior performance in identifying high-potential zones. The results offer valuable insights that can aid decision-makers in effectively managing water resources in vulnerable areas.