Abstract View
Cokriging With a Low-cost Air Sensor Network to Estimate Spatial Variation of Brake and Tire-wear Related Heavy Metals and Reactive Oxygen Species in Southern California, United States
JONATHAN LIU, Irish del Rosario, Jonah Lipsitt, Farzan Oroumiyeh, Jiaqi Shen, Suzanne E. Paulson, Beate Ritz, Jason Su, Scott Weichenthal, Yifang Zhu, Michael Jerrett, University of California, Los Angeles
Abstract Number: 474
Working Group: Health-Related Aerosols
Abstract
Due to regulations and technological advancements reducing tailpipe emissions, there is an increased focus on non-exhaust automobile emissions, including break and tire wear particulate matter (PM) which contain heavy metals capable of generating oxidative stress in exposed organisms. Despite potential harms to human health, few studies have modeled the spatial variability of PM from brake and tire wear emissions.
Improvements in electrical engineering, internet connectivity, and an increased public concern over air pollution have led to a proliferation of low-cost air sensor networks such as the PurpleAir monitors, which primarily measure fine particulate matter (PM2.5). While unable to measure PM2.5 constituents, these networks are dense and cover a wide spatial area.
In this study, we model the concentrations of barium, zinc, 2-hour reactive oxygen species (ROS), and black carbon alongside DTT loss and OH formation. We use a cokriging approach, targeting fine particulate matter (PM2.5) constituents, measured across Southern California in two sampling campaigns and incorporating data from the PurpleAir network as a secondary predictor variable. Within the cokriging model, we create an external drift by incorporating a land-use regression (LUR) model. We obtained land-use variables such as traffic, business density, tree canopy cover, and impervious surfaces. We then used a deletion-substitution-addition (DSA) algorithm and K-fold cross-validation to select an optimal model for use within cokriging.
Our final LUR model exhibits good fit for predicting metal mass concentrations, black carbon, DTT loss, and 2-hour ROS (adjusted R2 = 0.50-0.68), and reasonable fit for OH and metal normalized mass concentration (adjusted R2 = 0.37-55). Additionally, we find that several of our LUR models exhibit statistically significant spatial autocorrelation (Moran’s I p: 0.01-0.80) and high correlation with collocated PurpleAir sensor PM2.5 measurements (R = 0.67). In this project we present an exposure surface of brake and tire wear in the Southern California meeting. We also present the results of an evaluation of the ability for a low-cost air sensor network to predict speciated PM2.5.