This New Investigator Award study seeks to provide novel exposure–response functions for the effects of long-term exposure to ultrafine particles on several mortality and morbidity outcomes, while adjusting for exposures to other traffic-related air pollutants, road traffic noise, and socioeconomic status. The study makes use of a new Danish cohort of 650,000 adults.
Poster abstract for HEI Annual Conference 2023
Performance Evaluation of a Google Street View-based LUR model for Prediction of Ultrafine Particles in COpenhagen Ultrafine Particles and Health (COUPH) Study
Heresh Amini1,2,3 and COUPH collaborators
1Department of Public Health, University of Copenhagen, Copenhagen, Denmark; 2Harvard TH Chan School of Public Health, Boston, United States, 3Icahn School of Medicine at Mount Sinai, New York, United States
Background. Land use regression (LUR) modeling based on off-road fixed-site monitoring is a widely used method for modeling and estimating fine scale long-term spatial variation of air pollution, especially for applications in epidemiological studies. More recently, LUR models have been developed based on intensively repeated on-road mobile monitored data and mixed-effects modeling. The objective of this work was to 1) monitor ultrafine particles (UFP) across selected residences in Copenhagen, Denmark, and 2) evaluate the performance of a Google Street View-based LUR model developed for Copenhagen for prediction of UFPs across residences.
Methods. We monitored particle number concentration (PNC) at 37 residences in two ~72-hour-periods per site (warm and cold season). Temporal adjustment was done based on monitoring at a reference site from May 29, 2021, to May 29, 2022. The estimated long-term means at residences were then compared to a mixed-effects model based on Google Street View measurements (Google-MM) in Copenhagen that predicted 2019 long-term mean PNC levels.
Results. Annual mean (SD) PNC at the reference site was 4715 (3001) pt/cm3, while annual means at across monitoring sites were slightly higher with 5201 (804) pt/cm3, ranging from 3715 to 6583 pt/cm3 at individual sites. Mean (SD) based on the Google-MM PNC for 37 residential monitored sites was 11,804 (5423) pt/cm3, ranging from 4422 to 30,956 pt/cm3. The annual mean PNC from the Google-MM model predictions was about two times higher than the estimated annual mean PNC across residential measurements. The estimated annual mean PNC at 27 residential sites with two complete measurements was not correlated with PNC from Google-MM (Spearman’s correlation coefficient: -0.01). The un-adjusted season-specific measurements were better correlated with the Google-MM PNC. Campaign 1 (warmer period) PNC at 37 sites, before temporal adjustment, was positively and poorly correlated with Google-MM PNC (0.27), while correlation in Campaign 2 (colder period) PNC at 32 sites and Google-MM PNC was 0.29.
Conclusions. This is the first study to compare Google-MM PNC predictions in Copenhagen to monitored residential PNC. Very low correlation was found between monitored PNC and predicted PNC from the Google-MM LUR model in Copenhagen.