Comparative Analysis of ANN and SVM for Groundwater Potential Mapping in Karst Terrains of Southwestern Iran

Amid escalating water scarcity and the pressing need for sustainable water management, especially in arid and semi-arid regions, this study emphasises the importance of developing precise and efficient geospatial methods to evaluate groundwater potential in complex karst landscapes. This research focuses on Khuzestan Province in southwestern Iran, employing advanced Machine Learning (ML) techniques—namely, Artificial Neural Networks (ANN) and Support Vector Machines (SVM)—to map groundwater potential zones. The goal is to enhance resilience and promote sustainable water resource management in the region. A comprehensive array of topographic, geological, hydrographic, edaphic, and meteorological data was collected, processed, and integrated into a Geographic Information System (GIS) database to establish key conditioning factors for predictive modelling. After conducting a spatial multicollinearity analysis, the selected input variables included elevation, slope, aspect, multiple topographic indices, relief energy, heat load index, drainage density, lithostratigraphic units, fracture density, land use/cover, NDVI, and precipitation. Hydrogeological data, such as water-table depth and spring locations, obtained from official records, were also integrated to assess the performance of modelling outputs. Two predictive models—using ANN and SVM—were developed to generate groundwater potential maps for the study area. Both models demonstrated high predictive accuracy, highlighting unique strengths in capturing the complex spatial patterns of karst environments. This methodological approach shows promise as a reliable, globally applicable framework for groundwater potential mapping in similar karst regions. By offering valuable insights for hydrogeologists and policymakers, this approach supports enhanced groundwater exploration strategies and fosters sustainable water management in water scarcity regions.