Characterizing the spatial variability of local and background concentration signals for air pollution at the neighbourhood scale

Vehicle emissions represent a major source of air pollution in urban districts, producing highly variable concentrations of some pollutants within cities. The main goal of this study was to identify a deconvolving method so as to characterize variability in local, neighbourhood and regional background concentration signals. This method was validated by examining how traffic-related and non-traffic-related sources influenced the different signals. Sampling with a mobile monitoring platform was conducted across the Greater Toronto Area over a seven-day period during summer 2015. This mobile monitoring platform was equipped with instruments for measuring a wide range of pollutants at time resolutions of 1 s (ultrafine particles, black carbon) to 20 s (nitric oxide, nitrogen oxides). The monitored neighbourhoods were selected based on their land use categories (e.g. industrial, commercial, parks and residential areas). The high time-resolution data allowed pollutant concentrations to be separated into signals representing background and local concentrations. The background signals were determined using a spline of minimums; local signals were derived by subtracting the background concentration from the total concentration. Our study showed that temporal scales of 500 s and 2400 s were associated with the neighbourhood and regional background signals respectively. The percent contribution of the pollutant concentration that was attributed to local signals was highest for nitric oxide (NO) (37-95%) and lowest for ultrafine particles (9-58%); the ultrafine particles were predominantly regional (32-87%) in origin on these days. Local concentrations showed stronger associations than total concentrations with traffic intensity in a 100 m buffer (p:0.21-0.44). The neighbourhood scale signal also showed stronger associations with industrial facilities than the total concentrations. Given that the signals show stronger associations with different land use suggests that resolving the ambient concentrations differentiates which emission sources drive the variability in each signal. The benefit of this deconvolution method is that it may reduce exposure misclassification when coupled with predictive models.