Atmospheric environment (Oxford, England : 1994) | 11 Apr 2018
CW Milando and SA Batterman
The development of accurate and appropriate exposure metrics for health effect studies of traffic-related air pollutants (TRAPs) remains challenging and important given that traffic has become the dominant urban exposure source and that exposure estimates can affect estimates of associated health risk. Exposure estimates obtained using dispersion models can overcome many of the limitations of monitoring data, and such estimates have been used in several recent health studies. This study examines the sensitivity of exposure estimates produced by dispersion models to meteorological, emission and traffic allocation inputs, focusing on applications to health studies examining near-road exposures to TRAP. Daily average concentrations of CO and NOx predicted using the Research Line source model (RLINE) and a spatially and temporally resolved mobile source emissions inventory are compared to ambient measurements at near-road monitoring sites in Detroit, MI, and are used to assess the potential for exposure measurement error in cohort and population-based studies. Sensitivity of exposure estimates is assessed by comparing nominal and alternative model inputs using statistical performance evaluation metrics and three sets of receptors. The analysis shows considerable sensitivity to meteorological inputs; generally the best performance was obtained using data specific to each monitoring site. An updated emission factor database provided some improvement, particularly at near-road sites, while the use of site-specific diurnal traffic allocations did not improve performance compared to simpler default profiles. Overall, this study highlights the need for appropriate inputs, especially meteorological inputs, to dispersion models aimed at estimating near-road concentrations of TRAPs. It also highlights the potential for systematic biases that might affect analyses that use concentration predictions as exposure measures in health studies, e.g., to estimate health impacts.
* Data courtesy of Altmetric.com