Assessing Groundwater Salinization Using Spatial Machine Algorithm Techniques

Abstract Groundwater underpins livelihoods in arid regions yet remains vulnerable to climate variability and intensive abstraction. This study evaluates groundwater salinity susceptibility in the Wilayat of Barka, Oman, using ensemble learning—Random Forest (RF) and Extreme Gradient Boosting (XGBoost)—applied to a multi-decadal dataset (1985–2021). Inputs combine Landsat-derived land-use/land-cover (LULC) change, precipitation and temperature records, groundwater level and salinity observations, and LiDAR-based elevation, slope, and hydrological connectivity. Anthropogenic and geomorphological controls are represented by well density and distance to drainage. Model performance was strong: XGBoost achieved R² = 0.99 and MAPE = 0.011, while RF achieved R² = 0.97 and MAPE = 0.084. We produced five electrical-conductivity classes, from freshwater to very high salinity, and mapped spatial hotspots. XGBoost emphasized vulnerability in the north and northeast; RF highlighted the northwest, central, and southeastern sectors. Variable importance consistently favored groundwater level, followed by elevation, temperature, and rainfall. The integrated remote-sensing, GIS, and machine-learning workflow is reproducible and scalable, enabling routine salinity surveillance where monitoring networks are sparse. Findings provide actionable evidence for prioritizing protection, guiding well licensing, and targeting recharge and demand-management interventions. More broadly, the framework supports adaptive groundwater governance as hydroclimatic pressures intensify across data-limited drylands and enables evidence-based water policy decisions. Graphical Abstract A Compact RS–GIS–ML Pipeline Summarizes inputs, methods, and Outcomes for Groundwater Salinity Susceptibility in Wilayat Barka (area ≈ 166.663 km²). Multi-date Landsat Scenes (1985, 1990, 2000, 2013, 2021; TM/ETM+/OLI) Feed Image Segmentation and Random Forest Classification To Derive LULC and Change across 1985–1990, 1990–2000, 2000–2013, and 2013–2021. Vegetation Cover Dynamics Are Depicted with Directional Distribution (standard Deviation ellipse) To Show Dominant Spread from 1985 To 2021. Susceptibility Modeling Integrates LULC, rainfall, temperature, elevation, slope, Proximity To Drainage, Well density, and Groundwater-table Decline Using Random Forest and XGBoost. Outputs Include Five electrical-conductivity Classes (fresh, low, moderate, High, Very High) and Spatial Hotspot maps. Both Models Indicate Concentrated Salinity along the Northern Coastal belt, with Secondary Pockets Centrally and in the northwest; Agriculture Expansion Is Evident To the south. Model Skill Is High across algorithms; Groundwater Level Is the Leading driver, while Terrain and Climate Variables Modulate Risk. It Emphasizes a workflow—satellite Archives Plus Ancillary hydro-environmental data—to Monitor and Manage Groundwater Salinity in Arid settings. Policy Takeaways: Prioritize Surveillance in Northern Hot zones; Refine Well-licensing and Pumping controls; Target Managed Aquifer Recharge and Drainage improvements; Align land-use Planning with Salinity Risk To Protect Agricultural productivity, Ecosystem health, Community resilience, and long-term Food Security.