Main content area

Accumulation of Traffic-Related Trace Metals in Urban Winter-Long Roadside Snowbanks

Moghadas, S., Paus, K. H., Muthanna, T. M., Herrmann, I., Marsalek, J., Viklander, M.
Water, air, and soil pollution 2015 v.226 no.12 pp. 404
alkalinity, cadmium, chromium, copper, data collection, electrical conductivity, lead, linear models, nickel, prediction, principal component analysis, regression analysis, snow, sodium, total suspended solids, traffic, water quality, zinc, Norway, Sweden
Accumulations of mass loads of selected chemicals in roadside snowbanks were studied at five sites with various traffic densities in the city of Trondheim (Norway) by collecting snow samples throughout the winter period and analyzing them for 13 water quality constituents: pH, electrical conductivity (EC), alkalinity, Cl, Na, total suspended solids (TSS), Cd, Cr, Cu. Ni, Pb, W, and Zn. The resulting dataset was then supplemented by similar data collected earlier in the city of Luleå (Sweden). Regression analyses for individual sites indicated linear trends in unit-area constituent accumulations with time (0.65 < R ² < 0.95) and supported the assumption of linearity in further analyses. Principal component analysis (PCA) of the combined Luleå/Trondheim data revealed cause-effect relationships between the chemical mass loadings (TSS and trace metals) and three predictors: snow age (snow residence time (SRT)), traffic density (annual average density of traffic (AADT), and cumulative traffic volume (CTV = SRT × AADT). Cl and Na loads, originating from road salt applications in Trondheim only, did not display this trend. Two types of parsimonious models for predicting trace metal accumulations in winter-long roadside snowbanks were developed: (a) a linear regression model using CTV as a single predictor and predicting metal accumulations with a moderate certainty (0.37 < R ² < 0.66) and (b) multiple regression models using SRT, AADT, and snow water equivalent (SWE) as predictors. The latter models indicated good correlations between the mass loads and the predictors (0.64 < R ² < 0.77) and produced slightly better prediction accuracies (0.44 < R ² < 0.67) than the simpler model.