Mapping the spatial heterogeneity of nitrogen species in road dust through integrated multi-source remote sensing and...

Nitrogen accumulation in road dust constitutes a significant yet underexplored threat to environmental quality, particularly in regions where rapid urbanization intersects with intensive agriculture. The spatial distribution patterns and governing factors of individual nitrogen species within this environmental matrix remain poorly characterized. This study presents an integrated framework coupling field-based chemical speciation analysis, multi-source remote sensing, and interpretable machine learning to quantify and map the spatial distribution of total nitrogen (TN), nitrate-nitrogen (NO₃^--N), ammonium-nitrogen (NH₄⁺-N), and organic nitrogen (Org-N) in road dust across South Tianjin, China. Laboratory analysis of 41 road dust samples collected along urban-rural transects revealed pronounced spatial heterogeneity, with TN concentrations spanning 289-3782 mg/kg; the highest values were consistently associated with agricultural land use. Org-N constituted the dominant nitrogen fraction, accounting for more than 88% of TN across all land-use categories. Predictive models constructed with the Extra Trees algorithm demonstrated strong generalization capability, yielding cross-validated coefficients of determination (R²) of 0.891 for TN, 0.887 for Org-N, 0.846 for NH₄⁺-N, and 0.806 for NO₃^--N. Among all predictors, spatial distance from the urban core emerged as the most influential variable, explaining 32-46% of model variance across nitrogen species. Model interpretation via SHapley Additive exPlanations (SHAP) and permutation importance further identified surface runoff, precipitation, and population density as potential drivers of nitrogen variability, while Monte Carlo uncertainty analysis indicated greater predictive uncertainty for Org-N relative to inorganic nitrogen forms. Spatially explicit predictions delineated agricultural districts in eastern and southern South Tianjin as critical nitrogen hotspots. The proposed remote sensing-machine learning framework effectively discriminates between diffuse agricultural nitrogen inputs and localized urban emission sources, offering actionable guidance for source-specific pollution management and sustainable land-use governance in rapidly urbanizing regions.