AAAR 34th Annual Conference
October 12 - October 16, 2015
Hyatt Regency
Minneapolis, Minnesota, USA
Abstract View
Nature and Sources of Measurement Error in the USEPA PM2.5 Chemical Speciation Network
KELSEY HADDAD, Li Du, Jay Turner, Washington University in St.Louis
Abstract Number: 577 Working Group: Source Apportionment
Abstract The USEPA PM2.5 Chemical Speciation Network (CSN) was established to provide insights into the chemical composition of ambient PM in urban areas across the United States. These data have been used for State Implementation Plan (SIP) development for PM$_(2.5) nonattainment areas, source apportionment modeling, and health effects studies. Understanding the nature and sources of measurement error in CSN data is important to inform its proper use in all of these applications.
Measurement error in the CSN was examined using collocated data that is routinely collected at six sites. The data were censored to include only those samples with both concentration values >3xMDL. Root mean square (RMS) precision and percentile precision were calculated using the collocated data. These top-down estimates were compared to the predicted precision which is a bottom-up estimate generated from the reported measurement uncertainties. For most site-species combinations, both the RMS and percentile precisions were greater (less precise) than the predicted precision. The differences were especially large for species such as Ca, Fe and Si which are typically associated with crustal material. There was also large site-to-site variability in the top-down precision estimates for these species. Bias between the collocated samplers proved to be a significant contributor to the imprecision with long (often one year or more) periods of persistent bias. For example, for calcium the site-specific maximum in the one-year rolling mean bias ranged from 6% to 53% across the six sites while for sulfur the range was 3% to 11%. This suggests that for some species pooling the collocated precision data across the sites might not represent the precision at a given site within the network. Potential sources of bias include inadequate maintenance of sampler inlets, flow rate calibration error and flow rate deviations from the setpoint even when the flow rate is properly measured.