Simulation of Population-Based Commuter Exposure to NO2 Using Different Air Pollution Models

We simulated commuter routes and long-term exposure to traffic-related air pollution during commute in a representative population sample in Basel (Switzerland), and evaluated three air pollution models with different spatial resolution for estimating commute exposures to nitrogen dioxide (NO2) as a marker of long-term exposure to traffic-related air pollution. Our approach includes spatially and temporally resolved data on actual commuter routes, travel modes and three air pollution models. Annual mean NO2 commuter exposures were similar between models. However, we found more within-city and within-subject variability in annual mean (±SD) NO2 commuter exposure with a high resolution dispersion model (40 ± 7 µg m−3, range: 21–61) than with a dispersion model with a lower resolution (39 ± 5 µg m−3; range: 24–51), and a land use regression model (41 ± 5 µg m−3; range: 24–54). Highest median cumulative exposures were calculated along motorized transport and bicycle routes, and the lowest for walking. For estimating commuter exposure within a city and being interested also in small-scale variability between roads, a model with a high resolution is recommended. For larger scale epidemiological health assessment studies, models with a coarser spatial resolution are likely sufficient, especially when study areas include suburban and rural areas.


Validation of GIS Model
In order to validate the performance of the GIS commuter model used to simulate the walking and bicycle legs in this project, a small validation study was carried out. Information on commuter route and behavior was collected from 36 subjects of the Swiss Tropical and Public Health Institute. First, participants were asked to estimate the duration and distance of their commuter route between home and work locations, as well as to provide information on their commuter mode and route characteristics. Home and work addresses were also collected. Second, participants showed their exact route in Google Map (true distance). Finally, participants' commuter routes between home and work locations were simulated by means of the GIS model developed for this study. Figure S1 shows the comparison between the routing distance (from the GIS model) and the true distance (measured in Googe Map) and reported distance, respectively, for walking and cycling trips. Results are shown for subjects that only walk or cycle to work (i.e., do not combine various travel modes). There is a high agreement between the true distance and the simulated distance (R 2~1 .0). Comparisons between the reported distance and the modeled distance are similar to the comparisons with the Microcensus data ( Figure S2). Figure S1. Comparison of simulated routing distance against distance measured in Google map as well as reported distance in a pilot study in Basel.

Comparison between the Routing Distance and Reported Distance Provided by the Microcensus Data 2010
Routing distance and reported distance (estimated by the subjects during the interview) were compared ( Figure S2) to avoid detours and ensure the plausibility of the legs. Outliers were checked manually and excluded from the analysis if they did not make sense. Figure S2. Reported versus routing distance. The reported distance was estimated by the subject during the interview.

Comparison of Commuter NO 2 Exposure Estimates along Verified and Fastest Car Trips >3km Between Home and Work/School Locations.
We assessed the difference in commute exposure estimates that occur when modeling NO 2 exposure for car trips along the fastest as opposed to the verified route between home and work/school. A subsample of 91 subjects was selected that only commuted by car between home and work/school and had commuter trips longer than 3 km. The verified car trips correspond to the route that was reported during telephone interviews. The fastest route was subsequently simulated based on the same road network (TeleAtlas MultiNet). The PolluMap model was used to estimate commuter exposure to NO 2 . Comparing the time-weighted subjects' NO 2 commuter exposure estimates of the two approaches resulted in an R-square of 0.5 ( Figure S3). Overall, the mean (±standard deviation) of exposure estimates along the fastest routes (32.2 ± 11.2 µg m −3 ) was similar to the ones along the verified routes (33.3 ± 8.7 µg m −3 ). Absolute differences between the corresponding estimates had mean of 5.5 (±6.1) µg m −3 . Public transport legs (n=695) Figure S3. Subjects' NO 2 commuter exposure (estimated using the PolluMap model) along verified versus fastest car routes between home and work/school locations. Only subjects who commuted by car and had car trips >3 km were included.

Temporal Adjustment
We calculated temporal adjustment factors for each hour of the day separately for main roads and side streets to consider the diurnal pattern of NO 2 levels and differences in hourly traffic volume and composition of vehicles (i.e., with separate counts for personal cars and trucks). Ratios were computed between the annual weekday hourly means and the annual mean concentration measured at the monitoring stations (see Figure S4). While the ratios at the side road were peaked during rush hours, those at the street site were highest during working hours (9-11am; 3-5pm), most likely due to increased truck traffic. Ratios were applied to each commuter leg concentration based on the road classification and start hour of the leg. Table S2 shows the percentage of legs defined as main and side roads separately for study area. Summary statistics of the ratios are provided in Table S3. Figure S4. Number of commuter legs per hour of the day and annual weekday hourly NO 2 means (±standard deviation) at the two fixed stations (main road and side road) used to compute temporal adjustment ratios. The straight lines represent the annual average NO 2 measured at the two fixed stations.

Comparison of the Three NO 2 Models with Measurements
We compared the three models-PROKAS, ESCAPE and PolluMap-to NO 2 measurements from a total of 31 sites within Basel-City from the Swiss study on Air Pollution and Lung and Heart Diseases in Adults (SAPALDIA) (Table S4, Figure S5). These measurements were conducted outside subjects' homes in three biweekly integrated sampling campaigns in 2011 using Passam passive diffusion samplers (Passam AG, Schellenstrasse, Männedorf, Switzerland). To keep the evaluation spatially comparable, we compared the average NO 2 concentrations of each site to the value estimated for the grid value of the three models. Thus, the data do not reflect a proper validation of the models. Note: a Measurements were conducted during three seasons (two-week samples) in 2011 as part of the Swiss SAPALDIA study (Basel-City only); b The bias is calculated as the difference (predicted-measured) NO 2 , shown as mean and standard deviation (sd); c Two sites are classified as regional background sites. Figure S5. Comparison of NO 2 models with measurements of 31 sites from the 2011 SAPALDIA study.