Data-driven and physics-informed AI for river pollution analysis in Serbia using open geodatabases

Pollution of river systems in Serbia has intensified in recent decades due to rapid urban growth, industrial emissions, agricultural runoff, and insufficient wastewater treatment. Addressing these pressures requires advanced analytical frameworks that combine empirical data with hydrological knowledge. In this research, a data-driven and physics-informed artificial intelligence (AI) approach is applied to evaluate river pollution using open-access geodatabases. The methodology integrates Sentinel-2 imagery, Shuttle Radar Topography Mission (SRTM) Digital Elevation Models, and official hydrological and demographic records together with water quality indicators to construct a comprehensive dataset. Machine learning models were employed to map and predict pollution hotspots, while hydrological rules such as flow accumulation, drainage density, and groundwater–surface water interactions were embedded to improve model reliability. Findings highlight the Danube–Sava–Morava corridor as the most critical pollution zone, particularly near Belgrade, Pančevo, and Niš, where industrial discharges and untreated municipal effluents significantly degrade water quality. Smaller basins such as the Nišava and Timok are especially vulnerable because of limited dilution capacity and ecological fragility. Scenario simulations for 2050 indicate worsening contamination under continued urbanization and climate variability, underscoring the urgent need for basin-scale water management and expanded wastewater treatment infrastructure. The results confirm that coupling open geodatabases with physics-informed AI provides a transparent and reproducible tool for sustainable river basin governance, offering both scientific insights and policy guidance for Serbia’s future water security.