Abstract
Study region: Johannesburg, South Africa, is a rapidly urbanising metropolis where mixed residential, commercial, and industrial land uses generate highly variable stormwater runoff that threatens downstream water quality. This study monitored multiple storm events across contrasting urban catchments to characterise pollutant dynamics under real-world hydrological conditions. Study focus: The research quantified the co-occurrence of heavy metals (Cu, Fe, Zn) and nutrients (N, P) in stormwater using Inductively Coupled Plasma Mass Spectrometry (ICP-MS) and ion chromatography. Artificial neural network (ANN) models were then applied to predict pollutant concentrations under different land-use and seasonal scenarios. The models performed strongly (R-2 > 0.85) for key pollutants, showing that 76 % of samples exceeded local water quality guidelines for at least one metal. Peak zinc and nitrogen loads were linked to industrial runoff. New hydrological insight: The findings demonstrate that high-density urban areas function as hotspots for simultaneous heavy metal and nutrient pollution, intensifying risks of eutrophication and ecological degradation in receiving waters. By linking ANN-based pollutant prediction with specific land-use classes, the study presents the first transferable framework for integrated stormwater quality management in Johannesburg and similar African megacities, supporting more spatially explicit regulation and prioritisation of pollution control measures.