This study will assess the effects of metals from nontailpipe emissions on asthma and lung function in the most recent cohort of the Children’s Health Study in Southern California (recruited during 2002-2012), using available filters with particulate matter samples. The investigators will estimate exposure to several pollutants and transportation noise and evaluate the roles of socioeconomic status, green space, physical activity, diet, and stress.
Poster abstract for HEI Annual Conference 2022
Exposure modeling of traffic-related non-tailpipe factors in Southern California
Meredith Franklin1,2, Xiaozhe Yin2, Masoud Fallah-Shorshani2, Rob McConnell2, Scott Fruin2
1University of Toronto, Toronto, Ontario, Canada; 2University of Southern California, Los Angeles, California, USA
Background. With the implementation of regulatory standards, tailpipe traffic emissions have decreased substantially over the last several decades. On the other hand, non-tailpipe emissions such as noise and particulate matter (PM) arising from brakes, tires, and resuspended road dust remain unregulated and have surpassed particulate tailpipe emissions in much of Southern California. Due to limitations in routine monitoring of either speciated PM or noise, there have been few studies in which non-tailpipe environmental exposures can be accurately assessed over a large urban area.
Methods. We coupled over 150 variables including meteorological, roadway and traffic characteristics and dispersion model estimates with two sets of samples: 1) A-weighted, equivalent noise (LAeq in decibels, dB) data collected on hour-long foot journeys around 16 locations throughout Long Beach, CA; and 2) quasi-ultrafine (PM0.2), accumulation mode fine (PM0.2-2.5), and coarse (PM2.5-10) particulate matter elements collected at 220 locations over two seasons throughout 8 communities across Southern California. Machine learning and land use regression (LUR) models were trained and tested to predict exposure surfaces of noise and 24 elemental components representing a variety of non-tailpipe sources.
Results. For noise, among all machine learning models, extreme gradient boosting had the best results in validation tests (leave-one-route-out R2 = 0.71, root mean square error (RMSE) 4.5 dB; 5-fold R2 = 0.96, RMSE 1.8 dB). Local traffic volume was the most important predictor of noise; road features, land use, and meteorology including humidity, temperature, and wind speed also contributed to model accuracy. For fine and coarse copper, iron, and zinc (as non-tailpipe vehicle emissions) LUR models with 17 to 36 predictor variables including meteorology; distance to different classifications of roads; intersections and off ramps within a given buffer distance; truck and vehicle traffic volumes; and near-roadway dispersion model estimates produced the best predictions. Our models effectively captured spatial gradients in the metallic portion of PM, particularly near roadways for the non-tailpipe emissions, and appeared to make better use of meteorological variables then other similar studies’; LUR R2 ranged from 0.76 to 0.92, some of the highest reported in the literature.
Conclusions. With a novel, on-foot mobile noise measurement method and one of the largest spatially resolved datasets of measured PM elements in the US we were able to use machine learning and LUR to generate small-scale exposure estimates of non-tailpipe traffic emissions and noise. These results will allow important and novel epidemiological investigations to follow.