A land use regression model for ultrafine particles in Vancouver, Canada

• Main Author: Abernethy, Rebecca
• Language: English
• Published: University of British Columbia 2012
• Online Access: http://hdl.handle.net/2429/41895

Background and Aims: Epidemiologic studies have associated adverse health outcomes with exposure to traffic-related air pollutants, principally NO₂, at levels below those showing effects in controlled exposure studies. This suggests the importance of related outdoor air contaminants, such as ultrafine particles (UFP) (<0.1µm in diameter). Presently, no UFP monitoring exists in North America and little information is available regarding UFP spatial distributions. We measured particle number concentrations (PNC) in Vancouver to develop a land use regression (LUR) model for use in epidemiologic studies and to identify important sources of UFP. Methods: During a two-week sampling period in spring 2010, PNC were measured with portable condensation particle counters (CPC) for 60-minutes at eighty locations used previously to characterize spatial variability in nitrogen oxides. Continuous PNC measuring occurred at four additional locations to assess temporal variation. LUR modeling was conducted using 135 geographic predictors, including: road length, vehicle density, intersection and bus stop density, land use type, fast food restaurant density, population density and others, following previously developed methods. A novel buffer approach incorporated meteorologic data through wedge-shaped wind roses from measurements made during PNC sampling, in addition to circular buffers. Results: The range of measured (60-minute median) PNC across locations varied 70-fold (range: 1500 – 105 000 particles/cm³, mean [SD] = 18 200 [15 900] particles/cm³). Correlations of PNC with concurrently measured two-week average NO₂, NO and NOX concentrations at the same sites were 0.64, 0.65 and 0.70. A model (R² = 0.48, leave-one-out cross validation R² = 0.32) predicting PNC included length of truck routes within 50m, density of fast food locations within 200m and ln-distance to the nearest port. LUR models created with wind rose shaped buffers had lower predictive power than models with circular buffers (R² = 0.29 – 0.34). Conclusions: Measured PNC was highly variable across the Metro Vancouver region and correlated with nitrogen oxides. Geographic predictors explained a smaller proportion of variability in PNC than found previously for nitrogen oxides, suggesting some common sources and additional unknown factors influencing PNC spatial variability. This represents the first LUR model for UFP in North America.