A hybrid land use regression/AERMOD model for predicting intra-urban variation in PM2.5

Characterizing near-source spatio-temporal variation is a long-standing challenge in air pollution epidemiology, and common intra-urban modeling approaches [e.g., land use regression (LUR)], do not account for short-term meteorological variation. Atmospheric dispersion modeling approaches, such as AERMOD, can account for near-source pollutant behavior by capturing source-meteorological interactions, but requires external validation and resolved background concentrations. In this study, we integrate AERMOD-based predictions for source-specific fine particle (PM2.5) concentrations into LUR models derived from total ambient PM2.5 measured at 36 unique sites selected to represent different source and elevation profiles, during summer and winter, 2012-2013 in Pittsburgh, Pennsylvania (PA). We modeled PM2.5 emissions from 207 local stationary sources in AERMOD, utilizing the monitoring locations as receptors, and hourly meteorological information matching each sampling period. Finally, we compare results of the integrated LUR/AERMOD hybrid model to those of the AERMOD + background and standard LUR models, at the full domain scale and within a 5 km(2) sub-domain surrounding a large industrial facility. The hybrid model improved out-of-sample prediction accuracy by 2-10% over LUR alone, though performance differed by season, in part due to within-season temporal variability. We found differences up to 10 mu g/m(3) in predicted concentrations, and observed the largest differences within the industrial sub-domain. LUR underestimated concentrations from 500 to 2500 m downwind of major sources. The hybrid modeling approach we developed may help to improve intra-urban exposure estimates, particularly in regions of large industrial sources, sharp elevation gradients, or complex meteorology (e.g., frequent inversion events), such as Pittsburgh, PA. More broadly, the approach may inform the development of spatio-temporal modeling frameworks for air pollution exposure assessment for epidemiology. (C) 2016 Elsevier Ltd. All rights reserved.