Global mapping of pharmaceutical ecological risk in rivers using machine learning: drivers, hotspots, and compounded water stress

Pharmaceuticals serve as a double-edged sword: while essential for health care, they are also ubiquitous pollutants that threaten ecosystem. The vast scale of global surface waters makes comprehensive pharmaceutical monitoring impractical, resulting in a significant knowledge gap regarding associated risks. In this study, we developed an interpretable machine-learning framework that integrates SHapley Additive exPlanations (SHAP) with structural equation modeling (SEM) to quantify how multiple drivers jointly shape pharmaceutical risk, thereby overcoming the interpretability limitations of traditional black-box models. Applying this framework, we generated a probability map of ecological risk from pharmaceuticals in global river network at 2 km resolution, and identified the pharmaceutical pollution hotspots in the Central United States, Southern China, Northern India, Central/Western Europe and Southern Brazil. The SHAP-SEM analysis revealed hydrogeological factors (26.3 %) as primary determinants, while socioeconomic (17.1 %) and climatic (11.2 %) factors acted as exogenous influences, indirectly affecting pharmaceutical risk through cascading impacts on livestock production, wastewater generation, and agricultural practices. Furthermore, to identify rivers under compound pressures, we mapped the co-occurrence of pharmaceutical risk (> 5 % of species affected), poor water quality (fecal coliform > 200 cfu/100 mL), and high water scarcity (water scarcity index = 1). This analysis highlighted high-concern rivers facing multiple stressors, including the Yellow River in China, the Sutlej River in Pakistan, and the River Nile in Egypt. Our adaptable framework offers a spatially explicit foundation for targeted surveillance and mitigation strategies, and is extensible to other emerging contaminants.