Integration of Remote Sensing, Machine Learning, and Electrical Resistivity Tomography for Soil Salinity Mapping in Ouargla, Southeastern Algeria

Abstract

Soil salinity is one of the most pervasive land-degradation processes in arid and hyperarid environments. In the Ouargla region of southeastern Algeria, salinisation driven by shallow artesian water tables, intensive irrigation, and extreme evapotranspiration rates severely threatens agricultural sustainability and ecosystem integrity. This thesis presents an integrated methodology combining multitemporal Sentinel-2 satellite remote sensing processed in Google Earth Engine (GEE), unsupervised machine learning classification (K-Means, k = 5), and Electrical Resistivity Tomography (ERT) field surveys to produce validated soil salinity maps over a 34 km2 test area centred on Hassi El Ghanami, Ouargla wilaya. Twelve spectral indices were computed from four annual dry-season and wet-season Sentinel2 composites (2022 to 2025), assembled into a 60-band multi-temporal feature stack, and classified into five salinity zones ranging from non-saline palmeraie to active sabkha. Two ERT profiles across contrasting salinity classes provide subsurface geophysical validation of the satellite classification. Temporal NDSI analysis reveals that irrigated agricultural plots, not the open sabkha background, are the primary zones of active secondary salinisation. The integrated approach is cost-effective, reproducible, and directly transferable to comparable hyperarid oasis environments across the MENA region.