Active Travel and Exposure to Air Pollution: Implications for Transportation and Land Use Planning Steve Hankey School of Public and International Affairs, Virginia Tech, 140 Otey Street, Blacksburg, VA 24061 Greg Lindsey Humphrey School of Public Affairs, University of Minnesota, 301 19 th Ave S, Minneapolis, MN 55455 Julian D. Marshall Department of Civil and Environmental Engineering, University of Washington, 201 More Hall, Box 352700, Seattle, WA 98195 INTRODUCTION Designing healthy, livable cities is a common planning goal. Many studies explore single aspects of this goal; for example, how urban form impacts travel behavior, air quality, or environmental injustice. However, little research focuses on linkages among these factors. Here, we integrate results from spatial models of bicycle traffic, pedestrian traffic, and air quality to assess spatial patterns of exposure to air pollution during active travel in Minneapolis, MN. We show that active travel is generally highest where air pollution levels are highest on higher functional class streets and near activity centers and that small changes in choice of routes or in the transportation network may reduce exposure (e.g., moving bicycle facilities away from hightraffic roads). We also show that rates of active travel and levels of air pollution are both higher in low-income neighborhoods with high proportions of non-white residents. Our findings have implications for policies that call for encouraging active travel on high-traffic roads (e.g., Complete Streets) and underscore ongoing efforts to address environmental injustice. RESEARCH APPROACH Our research approach is based on overlaying spatially-refined estimates of urban air quality, bicycle traffic, and pedestrian traffic in Minneapolis, MN to assess exposure during active travel. Our spatial models are based on a statistical-empirical modeling technique called Land Use Regression (LUR) in air quality science and facility-demand modeling in transportation engineering. In this paper, we briefly describe previously published models (Table 1) used to estimate (1) bicycle and pedestrian traffic volumes on street segments and (2) on-road particulate air pollution concentrations. Then, we describe how estimates from those models for all streets in Minneapolis were used to make spatial comparisons. Our work offers insight into exposure patterns in two distinct ways. First, our models describe where people actually bike or walk rather than the likelihood someone will walk or bike based on their location (as is the case for walkability metrics). Second, our models of particulate air pollution concentrations can discern small-scale changes in concentrations (i.e., ~100 meters) that a sparse regulatory monitor network cannot (e.g., the EPA s ambient air quality network). By comparing the output of these models we are able to explore factors of the built environment that may impact overall exposure. Spatial models of bicycle and pedestrian traffic: Our estimates of bicycle and pedestrian traffic were derived from facility demand models that estimate bicycle and pedestrian traffic volumes as a function of land use, demographic, and weather-related variables (Table 1; Hankey et al., 2012). The models use volunteer-based counts of cyclists and pedestrians as the dependent variable and to validate the models. Following validation, the facility demand models were used to estimate bicycle and pedestrian traffic volumes on roads without counts. There were 436 (431) separate counts of cyclists (pedestrians) used to build the models; mean (median) counts were 1
521 (313) for cyclists and 848 (336) for pedestrians. Model goodness-of-fit (pseudo-r 2 ) is 0.48 (0.42) for bicycle (pedestrian) traffic volumes. Spatial models of air quality: Our spatial models of particulate air pollution, which estimate concentrations for all surface streets in Minneapolis (locations matched to bicycle and pedestrian traffic estimates; Table 1), were developed as follows: Measurements of on-road particulate concentrations were collected by cycling prescribed routes in Minneapolis during morning (7-9am) and afternoon (4-6pm) rush-hours. Then, those measurements were joined with land use variables to develop regression models to estimate concentrations at locations without measurements (Hankey and Marshall, 2015). The models predict concentrations for surface streets, not highways, separately for morning and afternoon rush-hours. To match the estimates of bicycle and pedestrian volumes we estimated particulate concentrations at the midpoint of each street segment (mean block length in Minneapolis is ~120 meters); this procedure resulted in 13,569 point estimates (at the midpoint of each block) of on-road particulate concentrations, bicycle traffic volumes, and pedestrian traffic volumes across the City to use as an input for our spatial comparison. Model goodness-of-fit (adjusted-r 2 ) was 0.35-0.5 among the pollutants. Table 1. Summary of previously published models used to make spatial estimates Study Dependent variable Season Time-of-day Observations Model output Model adj-r 2 Hankey et al., Bicycle counts Autumn 12-hour Bicycle: 436 12-hour bicycle and Bike: 0.48 2012* Pedestrian counts (6:30am-6:30pm) Pedestrian: 431 pedestrian volumes Ped: 0.42 Hankey and Morning (7-9am) 1,101 locations Particle number (PN) PN: 0.49 Concentrations of Marshall, Autumn and afternoon along mobile Black carbon (BC) BC: 0.35 particulate matter 2015** (4-6pm) rush-hour routes PM2.5: 0.40 *Negative binomial regression; **Stepwise linear regression STUDY FINDINGS Spatial patterns of bicycling, walking, and air quality: Figure 1 includes maps of 12-hour bicycle and pedestrian traffic volumes for all street segments in the City of Minneapolis. As expected, the maps reflect the importance of street functional class (i.e., higher levels of traffic on arterials and collectors) and measures of the built environment (e.g., higher levels of traffic in the downtown area and areas of higher land use mix). These maps also highlight differences between bicycle and pedestrian traffic, namely: (1) off-street trails generally have larger relative volumes for cyclists than for pedestrians and (2) pedestrian traffic is more tightly clustered around retail corridors and activity centers (e.g., near the Central Business District [CBD] and along major transportation corridors) than is bicycle traffic. In general, rates of active travel were greatest near the CBD and decreased as a function of distance from the CBD. We modeled three measures of particulate air pollution: (1) particle number (a proxy for ultrafine particles), (2) black carbon, and (3) fine particulates (PM2.5). We estimated concentrations during morning and afternoon rush-hours (i.e., 3 pollutants 2 times of day = 6 sets of concentration estimates). Figure 1 shows concentration estimates during the morning rush-hour. The spatial variation of the pollutant concentrations follow the expected patterns. Namely, concentrations tend to be elevated in areas where there are dense networks of major roads in conjunction with activity centers (downtown Minneapolis, portions of South Minneapolis), near significant emission sources (industrial and railway areas), and lower in areas that are farther from traffic sources (open space and on low-traffic roads). Although there are differences in the magnitude of spatial variability and the overall level of estimated concentrations for each pollutant and time of day, the overall results seem to track each other fairly well and generally follow intuition. 2 PM2.5
Figure 1. Spatial estimates of active travel and particulate air pollution. Estimates of active travel are for 12-hour (6:30am-6:30pm) traffic volumes; estimates of air pollution are for the morning (7-9am) rush-hour models. We present two analyses to explore interactions among our spatial estimates of active travel and air pollution: (1) a mapping exercise to identify characteristic places in Minneapolis that may be of interest to planners (e.g., healthy [high active travel; low air pollution] or unhealthy [low active travel; high air pollution] neighborhoods) and (2) exploring how aspects of the built environment (e.g., street features, population density, land use mix) and socio-economic metrics (e.g., household income, non-white residents) are correlated with both health determinants. Mapping healthy and unhealthy blocks: A core motivation for combining outputs from our models is to explore patterns of population-level rather than individual-level exposure during active travel. For example, if one is only interested in individual-level exposure (e.g., when choosing a cycling route or estimating exposure for survey participants during travel) then a concentration surface is sufficient information. We found that only a small number of blocks are classified as healthy (1%) and unhealthy (1%); more blocks were classified as active & exposed (12%; high active travel, high air pollution) and inactive & clean areas (12%; low active travel, low air pollution). Healthy blocks were mostly located near, but just outside, of downtown; for bicycles, healthy blocks also included off-street trails near lakes or parkways. The built environment and exposure during active travel: To assess strategies for planning, we explored how active travel and particulate concentrations are correlated with certain aspects of the built environment. We summarize our model results by characteristics of streets and attributes of the built environment to explore how urban design may impact the spatial patterns of these two factors. We focus mainly on commonly cited strategies for improving walkability and bikeability (i.e., increasing population density and land use mix) and attributes of roads such as street functional class. 3
Both air pollution and active travel are correlated with street functional class (Figure 2). The clear implication for exposure is that most cycling and walking occurs on high-traffic roads that are also the most polluted (an important outlier is bicycle traffic on off-street trails). This finding makes sense since most utilitarian cyclists and pedestrians are using infrastructure to get to destinations (i.e., mostly located on high-traffic roads). However, it may be possible to make small shifts in the transportation network to reduce exposure to air pollution. One question that arises is what impact shifting active travel from major roads to adjacent local roads may have on exposure. To shed light on that question, we summarized our model estimates by: (1) major roads and (2) local roads at specific distances from the nearest major road. Active travel and particulate concentrations are highest on major roads and decreased steadily as the distance from a major road increased. The median block size in Minneapolis is ~120 meters which is located in the second bin for local roads in Figure 2 (100-200 meters). Therefore, shifting traffic one block corresponds to an estimated average decrease in morning [afternoon] exposure concentrations of 21 [12]% for particle number, 15 [20]% for black carbon, 7 [3]% for PM2.5. Figure 2. Median active travel traffic volumes and air pollution concentrations by street functional class (top - panels) and distance from a major road (bottom-panels). Concentrations normalized to morning, major roads values. 4
Environmental justice and exposure during active travel: One aspect of achieving a healthy city for all residents is the equitable distribution of access to health-promoting places. In the environmental justice literature two metrics commonly used to explore equity issues for exposure to environmental hazards are household income and proportion of non-white residents. Following those studies we stratified our model estimates by those two factors. Bicycle and pedestrian traffic was highest for low-income blocks with high proportions of non-whites. Particulate concentrations decreased with higher household income and lower proportions of non-white residents. The differences in concentrations between high- and lowincome areas as well as between high and low proportions of non-white residents were largest for particle number concentrations. These trends in particulate concentrations coupled with the increased volumes of bicycle and pedestrian traffic suggest that a significant share of the air pollution exposure burden while walking and biking is likely occurring in low-income, nonwhite neighborhoods. CONCLUSIONS Here, we described spatial patterns of particulate air pollution and active travel in Minneapolis, MN. Relatively few blocks were classified as healthy (1%; high active travel, low air pollution) or unhealthy (1%; low active travel, high air pollution). Spatial patterns of active travel and air pollution by street functional class suggest that minor shifts to the transportation network may reduce overall exposure; for example: (1) moving cyclists away from pollution by strategically locating bicycle infrastructure on low-traffic roads and (2) moving pollution away from pedestrians by shifting the location of emission sources (e.g., bus routes or stops). This finding may be of interest where policy prescriptions may encourage active travel on high-traffic roads (e.g., Complete Streets); i.e., providing cyclists and pedestrians a choice of low-exposure route may be useful. We found the highest rates of active travel are in neighborhoods with low-income and with a high proportions of non-whites. Air pollution concentrations too are highest in those neighborhoods. Our results underscore the importance of equitable access to health-promoting neighborhoods. REFERENCES Hankey, S., Lindsey, G., Wang, X., Borah, J., Hoff, K., Utecht, B. & Xu, Z. (2012). Estimating use of non-motorized infrastructure: models of bicycle and pedestrian traffic in Minneapolis, MN. Landscape and Urban Planning, 107(3), 307-316. doi: 10.1016/j.landurbplan.2012.06.005 Hankey, S. and Marshall, J.D. (2015). Land use regression models of on-road particulate air pollution (particle number, black carbon, PM2.5, particle size) using mobile monitoring. Environmental Science and Technology, 49(15), 9194-9202. doi: 10.1021/acs.est.5b01209 5