Magnetic fingerprints and AI reveal how traffic pollution hides in city soils

Every time a car brakes, a tyre scuffs against asphalt, or an engine burns fuel, it releases a invisible cloud of microscopic particles enriched with toxic heavy metals. Those particles drift onto nearby soils, settle, and stay. Now a team of researchers working in the rapidly growing city of Minna, in north-central Nigeria, has shown that these contaminated particles carry a second, hidden signature: they are magnetic. By reading that magnetic signal with handheld instruments and validating it with machine learning, the scientists have built one of the most complete pictures yet of how traffic pollution moves through, and becomes locked into, urban roadside soils.

The study, published in Environmental Earth Sciences, is built around a conceptual model the authors call the Source-Pathway-Sink framework. Traffic is the source, releasing ferrimagnetic particles — minerals such as magnetite and maghemite formed in combustion and mechanical wear — together with heavy metals including lead, cadmium, zinc, copper and chromium. Air, runoff and resuspended dust act as the pathway, spreading the contaminants laterally away from the road. The soil itself is the sink, and its physical properties determine whether pollutants stay near the surface or creep toward groundwater. What makes the work remarkable is that it quantifies all three stages with a combination of environmental magnetism, geochemistry, geotechnical testing, spatial statistics and artificial intelligence.

The fieldwork was conducted along six major road corridors in Minna, a city of intensifying traffic and rapid urban expansion built on the Precambrian rocks of the Nigerian Basement Complex. The researchers laid out transects perpendicular to each roadway, taking magnetic susceptibility readings every ten metres out to roughly one hundred metres from the tarmac, and then drilled vertical profiles down to seventy centimetres to track how the contamination behaves with depth. In total, forty-four sampling locations provided the dataset for the statistical and machine-learning analyses.

The headline finding is spatial and strikingly consistent: magnetic susceptibility, denoted χlf, spikes immediately beside the road and decays exponentially with distance. The team fitted decay curves to each corridor and found attenuation constants between 0.0757 and 0.1838, with steeper declines indicating that particles drop out of the air close to the kerb, and gentler ones showing wider lateral dispersal. Interpolated maps revealed hotspots clustered along the busiest corridors, while longitudinal profiles along the roads showed localized peaks at junctions, markets and congested commercial strips — places where idling engines and braking vehicles dump extra particulate load onto the verge.

Crucially, the magnetic signal tracks the chemistry. Low-frequency susceptibility correlated strongly with lead concentrations (r = 0.78), and moderately with zinc, confirming that the ferrimagnetic grains and the traffic-derived metals travel together. Cadmium, copper and chromium showed far weaker links, pointing to additional sources such as waste handling and the natural geological background. Contamination factors told a similar story: cadmium was classified as very highly contaminated, with a mean value of 8.36 and isolated maxima reaching 22, while lead and copper showed moderate enrichment. Yet the overall Pollution Load Index averaged just 0.45 — below the critical threshold of one — demonstrating that contamination in Minna is intense but patchy, concentrated in discrete near-road hotspots rather than blanketing the city.

The most provocative part of the study is its use of machine learning to prove that a cheap magnetometer can substitute for expensive laboratory chemistry, at least for screening purposes. The researchers trained two models — a Random Forest ensemble and a Simple Linear Regression — to predict the Heavy Metal Pollution Index, a weighted composite of all five metals, from magnetic and soil data. Under five-fold cross-validation, the simple linear model achieved an extraordinary R² of 0.9697, and the Random Forest’s feature-importance analysis assigned surface magnetic susceptibility a score of 0.9948, dwarfing every geotechnical variable combined. In other words, nearly all the predictive information about cumulative heavy-metal pollution in these soils is encoded in a single, rapidly measurable magnetic property.

Vertical profiling added a second dimension to the story. Magnetic correlations with the surface signal fell steadily from ρ = 0.87 at ten centimetres depth to just 0.19 at seventy, showing that the traffic-derived contamination is largely trapped near the top of the profile. The reason is geotechnical: Minna’s roadside soils are dominated by fine-grained, clay-rich lateritic materials with low permeability, high plasticity and strong adsorption capacity. These soils resist downward water flow and chemically bind metal ions, acting as a natural lid on the pollution.

To quantify this retention behaviour, the team invented a new metric, the Vertical Retention Capacity Index, which combines plasticity index, liquid limit, inverse permeability, soil classification and porosity into a single depth-resolved score. The VRCI peaked at roughly fifteen centimetres and declined systematically below, allowing the researchers to define three subsurface zones: a high-retention layer from the surface to thirty centimetres, a transitional zone to forty centimetres, and a low-retention domain beneath. Statistical validation was emphatic — one-way ANOVA returned F = 1784.51, the Kruskal-Wallis test H = 63.18, and Spearman correlation confirmed retention falls significantly with depth. From a remediation standpoint, this means the upper thirty centimetres of hotspot soils are where excavation or stabilization efforts should focus first.

Frequency-dependent magnetic measurements helped separate the anthropogenic signal from natural noise. Values of χfd% were generally low, consistent with coarse, multi-domain grains typical of vehicle emissions rather than the fine superparamagnetic grains produced by natural soil-forming processes. Permeability showed a moderate negative correlation with susceptibility, reinforcing the picture that less permeable soils hoard contaminants while coarser materials let them migrate. Principal component analysis cleanly separated the dataset into an anthropogenic contamination component — dominated by susceptibility, pollution index, lead and zinc — and a soil hydraulic component defined by permeability and porosity, giving multivariate weight to the dual-control interpretation.

The broader significance is practical as much as scientific. Laboratory-based heavy-metal analysis is slow and costly, which is one reason contamination monitoring lags behind urban growth across much of sub-Saharan Africa. A framework that can screen entire road networks with a handheld susceptibility meter, validated by algorithms that quantify exactly how much the magnetic signal can be trusted, offers city planners a rapid, non-destructive first line of defence. The authors caution that their retention indices capture physical controls but not chemical ones such as pH, organic matter and cation exchange capacity, and that results from a single city must be tested elsewhere. But the convergence of magnetic, geochemical, geotechnical and machine-learning evidence in Minna makes a compelling case that the dirt beside the world’s roads is not just polluted — it is measurably, predictably, magnetically polluted, and now we know how to read it.

Subject of Research: Traffic-derived heavy-metal contamination and its source, transport and retention in urban roadside soils, assessed using environmental magnetism and machine learning.

Article Title: Source-pathway-sink controls on heavy-metal contamination in urban soils: Insights from environmental magnetism and machine learning

Article References: Source-pathway-sink controls on heavy-metal contamination in urban soils: Insights from environmental magnetism and machine learning. (n.d.). https://doi.org/10.1007/s12665-026-13114-w

Image Credits: AI Generated

DOI: 10.1007/s12665-026-13114-w

Keywords: heavy metals, urban soil, environmental magnetism, magnetic susceptibility, machine learning, traffic pollution, roadside contamination, soil geotechnics, pollution index, Nigeria, source-pathway-sink, contamination hotspots