A1676
Title: Bayesian source apportionment of PM2.5 species data collected over space and time
Authors: Veronica Berrocal - University of California, Irvine (United States) [presenting]
Abstract: The health burden associated with particulate matter exposure is now well-documented and recognized. Recently, various research efforts have been undertaken to better understand the health impacts of individual components of fine particulate matter, or PM2.5. While evidence in this regard is still forming, it appears clear that some of the species of PM2.5 are particularly noxious to human health. Thus, the reduction of PM2.5 species concentration is fundamental to protecting public health. To ensure that mitigation efforts are well-directed and more effective, it is important to understand what are the major sources of PM2.5 species. A Bayesian hierarchical model is proposed to perform source apportionment of six PM2.5 components observed in California in the year 2021. Adopting a functional framework, the observed log concentration of the 6 PM2.5 components is represented using a Bayesian latent factor model with site-specific latent source profiles modeled to be spatially dependent and linked to a global profile. The proposed model is applied to simulated data, successfully retrieving the pollution sources and the contribution of the sources to each pollutant. The model is also applied to the observed concentration of aluminum, sulfur, organic carbon, elemental carbon, nitrate and sulfate in California during the year 2021, identifying 4 major sources.
