Modeling horizontal and vertical variation in intraurban exposure to PM2.5 concentrations and compositions

• Vertical variation was observed for PM2.5 and Si.
• Sampling height above ground was a significant predictor in PM2.5 and Si models.
• Vertical mixing of particles from multiple sources was likely for Cu, Fe, and Zn.
• The R2 value of the LUR model was 0.65 for PM2.5.
• The R2 value of the LUR model were >0.69 for Cu, Fe, Mn, Ni, Si, and Zn.

Land use regression (LUR) models are increasingly used to evaluate intraurban variability in population exposure to fine particulate matter (PM2.5). However, most of these models lack information on PM2.5 elemental compositions and vertically distributed samples. The purpose of this study was to evaluate intraurban exposure to PM2.5 concentrations and compositions for populations in an Asian city using LUR models, with special emphasis on examining the effects of having measurements on different building stories. PM2.5 samples were collected at 20 sampling sites below the third story (low-level sites). Additional vertically stratified sampling sites were set up on the fourth to sixth (mid-level sites, n=5) and seventh to ninth (high-level sites, n=5) stories. LUR models were built for PM2.5, copper (Cu), iron (Fe), potassium (K), manganese (Mn), nickel (Ni), sulfur (S), silicon (Si), and zinc (Zn). The explained concentration variance (R2) of the PM2.5 model was 65%. R2 values were >69% in the Cu, Fe, Mn, Ni, Si, and Zn models and <44% in the K and S models. Sampling height from ground level was a significant predictor in the PM2.5 and Si models. This finding stresses the importance of collecting vertically stratified information on PM2.5 mass concentrations to reduce potential exposure misclassification in future health studies. In addition to traffic variables, some models identified gravel-plant, industrial, and port variables with large buffer zones as important predictors, indicating that PM from these sources had significant effects at distant places.