Integrating SAR Remote Sensing, Optical Imagery, and Gravimetric Data with Machine Learning for Fractured Aquifer Delineation in a Semi-Arid Mountain Region

Delineating fractured aquifer units in structurally complex semi-arid terrains remains a critical challenge for groundwater resource management. In the Haut Atlas Central (Béni Mellal region, Morocco), tectonic lineaments exert dominant control on preferential groundwater flow, yet subsurface characterization is severely constrained by sparse field data. This study develops a multi-source, data-driven framework that integrates Sentinel-1 SAR, Sentinel-2 optical imagery, and Bouguer anomaly gravimetric data using machine learning algorithms to map and characterize fractured aquifer systems.

Tectonic lineaments were extracted from Sentinel-1 and ALOS PALSAR imagery using the PCI Geomatica LINE module with optimized parameters. Lithological units were classified using a comparative ML approach that combined Random Forest (RF), Support Vector Machine (SVM), and XGBoost, trained on multi-source feature stacks derived from SAR backscatter, Sentinel-2 spectral indices, and geological map data. Bouguer anomaly data from the Bureau Gravimétrique International (BGI) were processed in Oasis Montaj for regional–residual separation and structural edge detection (THDR, Tilt Angle, Analytic Signal), enabling subsurface lineament correlation. Spring emergence points served as proxy observations for hydrogeological validation.

SAR-based lineament extraction yielded 1,706 tectonic lineaments. RF lithological classification achieved an overall accuracy of 0.867 (Kappa 0.807). Spatial analysis revealed a 70% fault–spring source correlation within a 1 km buffer, and six conceptual hydro-structural models were derived linking fault intersection density to spring emergence patterns.

The proposed framework demonstrates that combining SAR-derived structural mapping, ML-based lithological classification, and gravimetric subsurface analysis enables effective prospecting for fractured aquifers in data-scarce mountain environments. Comparative evaluation of RF, SVM, and XGBoost will identify the most transferable approach for similar semi-arid hydrogeological contexts across North Africa.