Urban land use regression models: can temporal deconvolution of traffic pollution measurements extend the urban LUR to suburban areas?

Abstract

Land use regression (LUR) models that predict intra-urban variability of traffic-related air pollution are frequently used to inform exposure assessments in epidemiological studies. Spatially extending models built in city centres to unmeasured surrounding areas could be beneficial as it increases the size of area available for exposure assessment, which can improve the statistical power of health studies. However, past studies have shown poor performance when these models were transferred to unmeasured areas. In this study, ambient concentrations of black carbon (BC), ultrafine particles (UFP), nitric oxide (NO) and nitrogen dioxide (NO2) with response times ranging from 1 s to 20 s were measured in the summer using a mobile laboratory. These ambient concentrations were temporally deconvolved into local, neighbourhood- and regional-background time-series signals. The temporally resolved signals were used to develop resolved LUR models (i.e. local, neighbourhood-background and regional-background models) while unresolved LUR models were developed from total ambient concentrations (i.e., measurements that were not temporally resolved). Both resolved and unresolved LUR models displayed modest R2 values when compared to models developed from fixed sites which may be due to the mobile nature of our data. External validation of the resolved and unresolved models within the models' original geographic domain showed comparable performance for BC (R2 for resolved vs. unresolved: 0.44 vs. 0.47), UFP (0.34 vs. 0.35), NO (0.18 vs. 0.19) and NO2 (0.42 vs.0.40). However, the resolved models were better able to assess exposure than the unresolved models when they were spatially extended to suburban areas bordering the urban area: UFP (R2: 0.42 vs. 0.29), NO2 (0.36 vs 0.26), NO (0.21 vs 0.14) and BC (0.32 vs 0.30). Furthermore, the resolved models were better able to predict the observed variability in suburban areas with both similar and different land uses to the urban area.