Population-Level Exposure to Particulate Air Pollution during Active Travel: Planning for Low-Exposure, Health-Promoting Cities

Background: Providing infrastructure and land uses to encourage active travel (i.e., bicycling and walking) are promising strategies for designing health-promoting cities. Population-level exposure to air pollution during active travel is understudied. Objectives: Our goals were a) to investigate population-level patterns in exposure during active travel, based on spatial estimates of bicycle traffic, pedestrian traffic, and particulate concentrations; and b) to assess how those exposure patterns are associated with the built environment. Methods: We employed facility–demand models (active travel) and land use regression models (particulate concentrations) to estimate block-level (n = 13,604) exposure during rush-hour (1600–1800 hours) in Minneapolis, Minnesota. We used the model-derived estimates to identify land use patterns and characteristics of the street network that are health promoting. We also assessed how exposure is correlated with indicators of health disparities (e.g., household income, proportion of nonwhite residents). Our work uses population-level rates of active travel (i.e., traffic flows) rather than the probability of walking or biking (i.e., “walkability” or “bikeability”) to assess exposure. Results: Active travel often occurs on high-traffic streets or near activity centers where particulate concentrations are highest (i.e., 20–42% of active travel occurs on blocks with high population-level exposure). Only 2–3% of blocks (3–8% of total active travel) are “sweet spots” (i.e., high active travel, low particulate concentrations); sweet spots are located a) near but slightly removed from the city-center or b) on off-street trails. We identified 1,721 blocks (~ 20% of local roads) where shifting active travel from high-traffic roads to adjacent low-traffic roads would reduce exposure by ~ 15%. Active travel is correlated with population density, land use mix, open space, and retail area; particulate concentrations were mostly unchanged with land use. Conclusions: Public health officials and urban planners may use our findings to promote healthy transportation choices. When designing health-promoting cities, benefits (physical activity) as well as hazards (air pollution) should be evaluated. Citation: Hankey S, Lindsey G, Marshall JD. 2017. Population-level exposure to particulate air pollution during active travel: planning for low-exposure, health-promoting cities. Environ Health Perspect 125:–534; http://dx.doi.org/10.1289/EHP442


Supplemental Material
Population-Level Exposure to Particulate Air Pollution during Active Travel: Planning for Low-Exposure, Health-Promoting Cities Steve Hankey, Greg Lindsey, and Julian D. Marshall

Table of Contents
Summary of land use variables and output of spatial models Table S1. Summary of land use variables Table S2. Descriptive statistics of model outputs "Sweet-spot" neighborhoods by pollutant Figure S1. Neighborhood-types for particle number concentration. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/. Figure S2. Neighborhood-types for black carbon concentration. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/. Figure S3. Neighborhood-types for PM2.5 concentration. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/.    Figure 2 (main text) using a morning rush-hour (7-9am) concentration surface. The maps represent the four categories of neighborhood-type outlined in Table 2 of the main text. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/. Figure S15. Replication of Figure 3 (main text) using a morning rush-hour (7-9am) concentration surface. Trends reflect similar patterns to afternoon rush-hour with more blocks meeting the 15% reduction in BC or PN criteria for the morning surface. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/.

Summary of land use variables and output of spatial models
We explored how the output of our spatial models of bicycle traffic, pedestrian traffic, and particulate air pollution varied by land use variables that may be important for designing health-promoting cities. We stratified our model output by 5 land use variables (two that are commonly cited as effective strategies for increasing "walkability" or "bikeability"; three that were significant in our facility-demand models). Table S1 gives a summary of these variables, how they were assembled, and summary statistics. As shown in Table S1 many locations had a value of "0" for open space and retail area (due to the small buffer size in the facility demand models). We randomized the "0" values for the purpose of plotting trends by quartile. This ensured that quartiles 1 and 2 for those two land use factors generated nearly the same values in the subsequent plots. We calculated summary statistics for our model output. Table S2 shows summary statistics for all 13,604 blocks where spatial estimates were made for each health determinant (active travel; particulate air pollution). All data collected to develop these models were collected in the autumn and during afternoon (4-6pm) rush-hour. As such our results should be interpreted as relevant for that time period and season (see below for a sensitivity analysis that compares the afternoon-based active travel estimates with morning rush-hour particulate concentrations). 9.0 a All measurements and models are for afternoon rush-hour (4-6pm). Descriptive statistics are based on all cityblocks (n=13,604).

"Sweet-spot" neighborhoods by pollutant
We identified four neighborhood types that result from combinations of the high/low quartiles of active travel and particulate concentrations. Figure 2 in the main text shows an aggregate map for city-blocks that meet inclusion criteria for two of the three pollutants (particle number, black carbon, PM 2.5 ). Figures S1-S3 show the same maps but for each pollutant separately. Figure S1. Neighborhood-types for particle number concentration. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/. Figure S2. Neighborhood-types for black carbon concentration. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/. Figure S3. Neighborhood-types for PM 2.5 concentration. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/.

Results of spatial models stratified by land use and transportation network variables
We stratified our spatial estimates of active travel and particulate air pollution by (1) various land use variables (see Table S1), (2) aspects of the transportation network, and (3) metrics of environmental justice. Figures 3 and 4 in the main text show the core findings for these analyses. However, in the following figures we show each factor separately and include PM 2.5 (PM 2.5 demonstrated limited variability and thus was not included in the plots in the main text). Below is a brief summary of each plot:  Figure S4. A plot similar to Figure 3 of the main text (street functional class) but also including values for off-street trails.  Figures S5-S8. A separate plot for each factor in Figure 4 of the main text and including PM 2.5 .  Figure S9. The same as Figure

Comparison of afternoon rush-hour bicycle and pedestrian traffic with morning rush-hour particulate concentrations
As a sensitivity analysis we explored using a morning rush-hour (7-9am) particulate concentration surface as an input to our spatial analysis (i.e., comparing spatial patterns of morning concentrations to the afternoon bicycle and pedestrian traffic estimates). To illustrate the findings of this analysis we replicated Figures 1-3 of the main text using the morning particulate surface (see Figure S13-S15). In general, concentrations were higher in the mornings (as compared to the afternoon) and demonstrated more spatial variability. However, the core findings of our base-case reported in the main text (afternoon concentrations) remained unchanged. The greater spatial variability in morning concentrations mostly exacerbated the core findings from the main text indicating that our results may be conservative when comparing to other times of day. A useful direction for future research would be to explore patterns of exposure for different time periods including times of day and seasons. Figure S13. Replication of Figure 1 (main text) using a morning rush-hour (7-9am) concentration surface. Spatial patterns of each factor (left-panel); plots of the transect (right-panel). The plots follow the transect from point A (left) to point B (right). Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/. Figure S14. Replication of Figure 2 (main text) using a morning rush-hour (7-9am) concentration surface. The maps represent the four categories of neighborhood-type outlined in Table 2 of the main text. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/. Figure S15. Replication of Figure 3 (main text) using a morning rush-hour (7-9am) concentration surface. Trends reflect similar patterns to afternoon rush-hour with more blocks meeting the 15% reduction in BC or PN criteria for the morning surface. Maps were created in ArcMap; underlying street and land use data are from https://gisdata.mn.gov/.