You are here

Comparing the estimated health impacts of long-term exposure to traffic-related air pollution using fixed-site, mobile, and deep learning models

Principal Investigator: 
,

McGill University, Canada

This study will evaluate health impacts of long-term exposures to traffic-related air pollution using exposure estimates from fixed-site and mobile measurement campaigns, as well as deep learning models, in Toronto and Montreal, Canada. The investigators will compare exposure estimates generated by these models to present-day and historical measurements, and to each other. 

Funded under
Status: 
Ongoing
Abstract

Comparing the Estimated Health Impacts of Long-Term Exposures to Traffic-Related Air Pollution Using Fixed-Site, Mobile, and Deep Learning Models

Scott Weichenthal1, Marianne Hatzopoulou2, Marshall Lloyd1, Arman Ganji2, Leora Simon1, Junshi Xu2, Mingqian Zhang2, Keni Mallinen2, Kris Hong3, Pedro Pinheiro3, Alexandra Schmidt1, Eric Lavigne4, Hong Chen4, Paul Villeneuve5, Joshua Apte6, Michael Tjepkema7, Richard T Burnett8

1McGill University, Montreal, Quebec Canada; 2University of Toronto, Toronto, Ontario, Canada; 3Consultant Montreal, Canada; 4Health Canada, Ottawa, Canada; 5Carleton University, Ottawa, Ontario, Canada; 6UC Berkeley, Berkeley, California, USA; 7Statistics Canada, Ottawa, Ontario, Canada; 8Consultant, Ottawa, Ontario, Canada

Background: Traffic pollutants such as ultrafine particles (<0.1 µm; UFPs) and black carbon (BC) vary greatly over small spatial gradients and it remains difficult to evaluate the long-term health impacts of these pollutants. Various exposure assessment methods have been applied to address this challenge to date, including models based on mobile and/or fixed-site monitoring data. However, it is not clear if models based on fixed-site and mobile monitoring data provide similar estimates of long-term spatial contrasts in exposures or if both methods result in comparable estimates of chronic health risks. Moreover, few studies have systematically evaluated the ability mobile and fixed-site models to capture past spatial gradients in traffic-related air pollutants. This is an important question as the validity of exposure models used to estimate long-term health risks relies in part on their ability to capture past exposures. In addition, given the large spatial variability of pollutants such as UFPs and BC, residential exposure estimates may not be optimal and alternative metrics that capture neighborhood-level trends in population mobility may provide more reliable estimates of long-term exposures and associated health risks.    

Methods: Our study has five main objectives. Aim 1 will develop new exposure models for UFPs and BC using mobile monitoring data in Montreal and Toronto, Canada and compare model predictions to fixed-site measurements collected over the same time-period. Aim 2 will evaluate the ability of mobile and fixed-site models to estimate past exposures using previously collected monitoring data. Aim 3 will develop mobility-weighted exposure surfaces that capture neighborhood-level trends in population mobility (and changes in mobility patterns over time) informed by travel survey data collected from random samples of Montreal and Toronto residents every 5-years. Aim 4 will evaluate a novel approach to exposure modelling using aerial imagery to estimate spatial variations in traffic-related air pollutants. We will train new deep convolutional neural networks to predict spatial variations in UFPs and BC using a large database of paired digital images/pollutant concentrations collected during mobile monitoring. Aim 5 will apply the new fixed-site, mobile, and deep learning models to estimate associations between UFPs, BC, and nonaccidental and cause-specific mortality in the Canadian Census Health and Environment Cohorts. This analysis will include 2-million adults followed over a 15-year period (2001-2016) in Montreal and Toronto pooled across multiple Census years (1991, 1996, 2001, 2006).

Results: The first year of our study has focused primarily on mobile data collection for the development of new exposures models for UFPs and BC. Data collection is ongoing and will be completed during summer 2021; hundreds of hours of mobile monitoring data have been collected to date reflecting all days of the week and times of day. Outdoor UFP concentrations along mobile monitoring routes range from less than 5000 particles/cm3 to more than 100,000 particles/cm3 whereas BC concentrations range from less than 1000 ng/m3 to approximately 10,000 ng/m3. New exposure models will be developed in 2022 and epidemiological analyses will be completed in 2023. 

Conclusions. Once completed, our investigation will provide new information on how the choice of exposure model impacts estimates of long-term health risks for UFPs, BC, and NO2/O3.