Toward sustainable groundwater management in Arid regions: GIS–AHP integration of DRASTIC models and machine learning–based GQI assessment in the Medenine Aquifer, Tunisia

Abstract

Groundwater vulnerability assessment is critical for sustainable management. However, previous research in arid regions, generally rely on traditional index-based models and their methodological adaptations. Combining conventional parametric approaches and machine learning (ML) methods, the objective of this study is to conduct a comparative assessment of groundwater vulnerability and water quality in the Medenine aquifer (Southeastern Tunisia) covering about 3100 km². Using GIS-based spatial analysis, four vulnerability models were applied: standard DRASTIC, modified DRASTIC, AHP-DRASTIC and AHP-modified DRASTIC. Geological, hydrogeological, and land-use factors were integrated to generate vulnerability maps. Moreover, the Groundwater Quality Index (GQI) was computed and integrated with ML models to analyze the hydrochemical parameters influence on groundwater quality and to explore nonlinear relationships. Based on the standard DRASTIC, the area was divided into low (2%), moderate (80%), and high (18%) vulnerability zones. However, the incorporation of the land use factor in the modified DRASTIC significantly increases the high vulnerability zones significantly to 57%. The AHP-modified DRASTIC model also classifies the area into low (28%), moderate (54%), and high (18%) vulnerability zones. The validation using nitrate concentration shows that the modified DRASTIC best matches contamination hotspots, while AHP-modified DRASTIC gives the most consistent vulnerability pattern. The incorporation of ML-based analysis combined with GQI results is useful for improving water quality assessment and understanding hydrochemical controls. The suggested strategy shows the value of integrating ML techniques and multi-criteria decision analysis for groundwater vulnerability and quality evaluation. The resulting maps represent valuable tools for groundwater protection and management, providing essential guidance for land and water resource management in arid regions.

Graphical Abstract

This graphical abstract delivers a concise visual synthesis of the methodological framework and key findings of the study, showcasing the transition from the classical DRASTIC approach to more advanced and better-performing vulnerability assessment models. The workflow begins with the seven intrinsic DRASTIC parameters, which constitute the foundation of traditional groundwater vulnerability evaluation. To enhance the model’s representativeness for the Midyan Basin, a land-use parameter is introduced to capture geomorphological characteristics and human-induced pressures that influence contamination risk. The Analytical Hierarchy Process (AHP) is then applied to refine the rating and weighting system, providing a more objective, consistent, and data-driven assessment of parameter significance. This methodological enhancement leads to the generation of four vulnerability maps: the standard DRASTIC, the land-use–integrated modified DRASTIC, and two AHP-adjusted variants. Collectively, these outputs offer a clearer and more transparent interpretation of spatial patterns in groundwater vulnerability. The graphical summary visually communicates this improved workflow, emphasizing how the incorporation of land-use information and AHP-based weighting substantially strengthens the accuracy and reliability of vulnerability mapping, ultimately supporting more effective groundwater protection and sustainable resource management. In addition, the framework integrates a Groundwater Quality Index (GQI) to represent observed hydrochemical conditions, which is enhanced by Machine Learning algorithms (GBM, RDF and SVM) to analyze the hydrochemical parameters influence on groundwater quality and to explore nonlinear relationships. This combined approach links vulnerability assessment with data-driven quality prediction, providing a more comprehensive evaluation of groundwater contamination risk.