Long-term Fine-scale PM2.5 Source Impacts from Major Sources at Monitors across the US

TING ZHANG, Lucas Henneman, George Mason University

     Abstract Number: 536
     Working Group: Aerosols Spanning Spatial Scales: Measurement Networks to Models and Satellites

Abstract
Background. Epidemiological evidence has shown differing adverse health impacts of source-specific ambient fine particles (PM_2.5). Spatiotemporal variability in PM_2.5 source impacts is not measured, warranting the development of consistent approaches to quantify source impact fields across large spatial-temporal domains.

Methods. Our study aimed to estimate the daily PM_2.5 source impacts at monitors during 2011-2020 across the US. PM_2.5 species data from >300 monitoring sites were extracted from two monitoring networks, the Interagency Monitoring of Protected Visual Environments (IMPROVE) and the Chemical Speciation Network (CSN). After comparing the flagged and unflagged PM_2.5 species concentrations, we removed those with unacceptable flags and refilled them after evaluating the performance of four interpolation methods. We clustered monitoring sites separately in two networks with a random forest (RF)-based method. Finally, we carried out multi-site dispersion-normalized positive matrix factorization for daily estimates of PM_2.5 source impacts from a consistent list of sources. Single-site source apportionment was conducted to validate the results from multi-site analyses.

Results. We found that CSN and IMPROVE datasets required refilling for 3.40% and 2.85% of the data, respectively, with higher rates (>5%) for carbonaceous substances. For PM_2.5 species concentration interpolation, the RF-based method had the lowest normalized root mean squared error (0.37±0.05) and outperformed others. Monitoring sites from CSN and IMPROVE networks were grouped into 25 clusters each to resolve primary sources, including gasoline vehicles, diesel vehicles, and biomass burning, for individual clusters and monitors. For example, in a site in Los Angeles, the estimated contributions of the above sources were 3.07%, 13.9%, and 16.0%.

Conclusion. The varied techniques over time at CSN and IMPROVE networks make it challengeable to quantify source impacts nationwide under a consistent framework. Here we identified strategies to streamline this process. Our results have the potential to advance source-specific epidemiology, environmental justice, and regulatory accountability studies.