Global Land Use Regression Model for Nitrogen Dioxide Air Pollution

• Authors: Andrew Larkin, Oregon State University; Jeffrey A. Geddes, Boston University; Randall V. Martin, Dalhousie University; Qingyang Xiao, Emory University; Yang Liu, Emory University; Julian D. Marshall, University of Washington; Michael Brauer, University of British Columbia; Perry Hystad, Oregon State University
• Journal Title: Environmental Science and Technology
• Volume: Volume 51, Number 12
• Publisher: American Chemical Society | 2017-06-20, Pages 6957-6964
• Type of Work: Article | Post-print: After Peer Review
• Publisher DOI: http://doi.org/10.1021/acs.est.7b01148
• Subjects: Health Sciences, Public Health; Physics, Atmospheric Science; Engineering, Environmental
• Keywords: Science & Technology; Technology; Life Sciences & Biomedicine; Engineering, Environmental; Environmental Sciences; Engineering; Environmental Sciences & Ecology; EXPOSURE ASSESSMENT; LUNG-FUNCTION; NO2; URBANIZATION; METAANALYSIS; RISK

Abstract

Nitrogen dioxide is a common air pollutant with growing evidence of health impacts independent of other common pollutants such as ozone and particulate matter. However, the worldwide distribution of NO2 exposure and associated impacts on health is still largely uncertain. To advance global exposure estimates we created a global nitrogen dioxide (NO2) land use regression model for 2011 using annual measurements from 5,220 air monitors in 58 countries. The model captured 54% of global NO2 variation, with a mean absolute error of 3.7 ppb. Regional performance varied from R2= 0.42 (Africa) to 0.67 (South America). Repeated 10% cross-validation using bootstrap sampling (n = 10,000) demonstrated a robust performance with respect to air monitor sampling in North America, Europe, and Asia (adjusted R2 within 2%) but not for Africa and Oceania (adjusted R2 within 11%) where NO2 monitoring data are sparse. The final model included 10 variables that captured both between and within-city spatial gradients in NO2 concentrations. Variable contributions differed between continental regions, but major roads within 100 m and satellite-derived NO2 were consistently the strongest predictors. The resulting model can be used for global risk assessments and health studies, particularly in countries without existing NO2 monitoring data or models.