Calibration of PM2.5 Mass Concentrations Used in the Pittsburgh Aerosol Research and Inhalation Epidemiology Study. Bilonick, R., A., Connell, D., P., Talbott, E., O., Xue, T., & Rager, J., R. Atmospheric Environment, 115:325-339, 5, 2015.
Calibration of PM2.5 Mass Concentrations Used in the Pittsburgh Aerosol Research and Inhalation Epidemiology Study [pdf]Paper  Calibration of PM2.5 Mass Concentrations Used in the Pittsburgh Aerosol Research and Inhalation Epidemiology Study [link]Website  abstract   bibtex   
Fifteen different types of PM2.5 mass concentration samplers were used by seven different monitoring networks at 47 locations in the Pittsburgh, Pennsylvania, region from 1999 to 2008. The samplers included Federal Reference Method (FRM) samplers, speciation samplers, tapered element oscillating microbalance (TEOM) samplers, and others. The different measurement principles used in these designs tended to lead to systematic differences (biases) when measuring the same quantity, and to differences in the typical size of random errors (imprecision) introduced by each type of sampler. Bias can take different forms either as a constant bias or as a non-constant (scale) bias, which depends on the size of the quantity being measured. The objective of the work presented here was to simultaneously calibrate the measurements made by these different samplers to remove relative biases (both constant and non-constant) so that all of the available PM2.5 data could be used interchangeably to develop exposure estimates for a retrospective epidemiology study. In order to accomplish this, we used linked temperature-stratified structural equation models, nonlinear regression models, and nonlinear mixed effects models. Applying these methods we constructed a comprehensive measurement error model that included both systematic error and random error components, and derived calibration equations that can be applied to place all of the PM2.5 mass concentration measurements on the same scale. The FRM sampler was used as the reference scale although the parameter estimates are invariant to this choice. Results showed that: (1) 50 °C TEOM samplers tended to show a large downward bias relative to the FRM sampler at low temperatures, and the magnitude of this bias decreased according to a nonlinear (sigmoidal) pattern with increasing temperature, (2) speciation samplers and other integrated samplers generally showed smaller biases relative to the FRM sampler that were not temperature-dependent, and (3) FRM samplers tended to be more precise than non-FRM samplers. These results are consistent with our previous work focusing on just a single monitoring site. Results are also presented here for several types of samplers that were not part of our previous study, including 30 °C TEOM, FDMS TEOM, and beta attenuation monitors.

Downloads: 0