"Another One Bites the Dust" : Artificial Neural Networks Application and Source Tracking of Polycyclic Aromatic Hydrocarbons and Ecological Risk Posed by Urban Road Dust

Road dust acts as both a vector and a source of multiple urban pollutants, including polycyclic aromatic hydrocarbons (PAHs), which can be transported into surface waters via runoff, contributing to widespread environmental contamination. This study integrates chemical analyses, magnetic measurements, Artificial Neural Network modeling, source tracking, and ecological risk assessment to evaluate PAHs contamination in Warsaw’s urban road dust and its potential ecological and human health risks. A total of 206 road dust samples were collected across diverse urban locations, encompassing variations in traffic intensity, building height, urban layout, and municipal heating activity. Samples were characterized by particle size fractions and magnetic properties, including magnetic susceptibility, saturation magnetization, and remanent magnetization. A subset of 57 samples focusing on the fine fraction (<0.2 millimeters) was analyzed for sixteen priority PAHs compounds, markers of combustion-derived pollution.

Total PAHs concentrations (∑16PAHs) in the fine fraction ranged from below the limit of quantification to twelve milligrams per kilogram, with an average of 3.5 milligrams per kilogram. The most abundant compounds were acenaphthene, fluorene, and phenanthrene, while high molecular weight PAHs accounted for approximately fifty-five percent of total PAHs. Diagnostic isomer ratios, including indeno[1,2,3-cd]pyrene to the sum of indeno[1,2,3-cd]pyrene and benzo[ghi]perylene, indicated traffic-related pyrogenic sources dominated. Magnetic susceptibility normalized to fine particle proportion (χWN) correlated strongly with total and high molecular weight PAHs (r = 0.79, R² ≈ 0.63), confirming its utility as a rapid, non-destructive proxy for organic contamination. Multivariate analyses revealed distinct pollution patterns, grouping PAHs by shared sources and physicochemical behavior. Elevated PAHs levels occurred in the city center, while peripheral areas had lower concentrations.

Artificial Neural Network models predicted PAHs concentrations from magnetic properties, particle size fractions, traffic intensity, building height, urban layout, and municipal heating patterns. A consolidated model across all samples achieved moderate performance (correlation coefficients ~0.45–0.57), whereas models stratified by municipal heating activity performed significantly better. Neural networks for periods of inactive heating yielded high correlation coefficients (0.91–0.94) and low root mean square errors, indicating strong predictive capability and stability. Sensitivity analysis identified building height and heating-related factors as most influential. Combining neural network predictions with isomer ratio diagnostics allowed source tracking of PAHs, distinguishing contributions from traffic, combustion heating, and urban structural influences. Magnetic proxies, particle size, and urban parameters efficiently identified PAH pollution hotspots in dense urban areas.

Ecological risk assessment using MERM-Q showed most samples fell into the low-risk category, with highest values in the city center. These results provide quantitative insights into potential ecological and human health risks posed by traffic-related PAHs, highlighting road dust as both a local pollutant and a vector transporting contaminants into broader urban environments and surface waters. This methodology enables rapid identification of pollution hotspots, supports targeted mitigation strategies, and informs urban planning, traffic management, and municipal heating policies to reduce environmental and health hazards. By combining predictive modeling with source apportionment, this study offers a robust framework for monitoring and managing hazardous organic pollutants in cities.

This research was funded in whole by the National Science Centre, Poland under grant number 2021/43/D/ST10/00996.