Hybrid land use regression modeling for estimating spatio-temporal exposures to PM2.5, BC, and metal components across a metropolitan area of complex terrain and industrial sources.

Land use regression (LUR) modeling has become a common method for predicting pollutant concentrations and assigning exposure estimates in epidemiological studies. However, few LUR models have been developed for metal constituents of fine particulate matter (PM2.5) or have incorporated source-specific dispersion covariates in locations with major point sources. We developed hybrid AERMOD LUR models for PM2.5, black carbon (BC), and steel-related PM2.5 constituents lead, manganese, iron, and zinc, using fine-scale air pollution data from 37 sites across the Pittsburgh area. These models were designed with the aim of developing exposure estimates for time periods of interest in epidemiology studies. We found that the hybrid LUR models explained greater variability in PM2.5 (R2 = 0.79) compared to BC (R2 = 0.59) and metal constituents (R2 = 0.34-0.55). Approximately 70% of variation in PM2.5 was attributable to temporal variance, compared to 36% for BC, and 17-26% for metals. An AERMOD dispersion covariate developed using PM2.5 industrial emissions data for 207 sources was significant in PM2.5 and BC models; all metals models contained a steel mill-specific PM2.5 emissions AERMOD term. Other significant covariates included industrial land use, commercial and industrial land use, percent impervious surface, and summed railroad length.