Healthy Neighborhoods: Walkability and Air Pollution

1The built environment affects public health in many ways, depending on the interplay between factors such as community design, travel patterns, physical activity, transportation safety, and air and water pollution. This study investigated interactions between a) walkability, a measure of how conducive the built environment is to walking and that predicts physical activity and active transportation, and b) exposure to outdoor air pollution, which is associated with a wide array of negative health impacts.

2Physical inactivity and outdoor urban air pollution are two of the top 15 global causes of health impairment. Ozone (O3), vehicle exhaust, and within-city contrasts in other air pollutants for which traffic is a major contributor are associated with many adverse health outcomes, including cardiopulmonary mortality, atherosclerosis, impaired lung development in children, asthma and asthma exacerbations, reduced lung function, cardiac arrhythmia, and preterm and low-birth-weight babies. Inactivity and insufficient activity (< 2.5 hr/week of moderate-intensity activity, or < 4,000 kJ/week) have been causally linked with heart disease, several cancers, diabetes, and other adverse health impacts and are associated with high body mass index (BMI; overweight and obesity), which can lead to additional effects on health. In polluted U.S. cities, the mortality risk from particulate air pollution is comparable to that for obesity (grade 1 or 2), but less than that for extreme obesity (grade 3). Reducing the average energy imbalance (caloric intake minus metabolic activity) among persons in the United States by approximately 100–165 kcal/day would prevent average weight gain (~ 1 kg/year), which suggests that moderate daily exercise—as little as two or three 10-min walking trips, such as to a bus stop or grocery store—could provide major public health benefits.

3“Walkability” of a neighborhood measures whether community design encourages or inhibits walking. For example, lack of a sidewalk can make walking unsafe, and a disconnected street network can discourage walking. Conversely, having retail stores close to where people live and providing connected streets increases the likelihood that a person will incorporate walking into daily routines.

4Neighborhood design—for example, the layout of buildings, land uses, and streets—can influence walking and other exercise activities, BMI, and overall health ratings, as well as air pollution emissions and exposures (Table 1). The American Academy of Pediatrics, the U.S. Centers for Disease Control, the World Health Organization, and others have called for research on how city design affects walking and other exercise and people’s exposure to air pollution. An important goal is using neighborhood design as a tool for creating cleaner, healthier urban environments.

5Our analyses identify neighborhoods that do especially well (or poorly) for both issues (walkability, air pollution). To our knowledge, this study is the first to compare quantitative estimates for these two neighborhood-scale environmental health attributes. Our findings demonstrate important health impacts of spatial exposure to the built environment.

6We investigated air pollution concentrations and walkability in Metro Vancouver, which is a coastal urban region of 2.2 million people in southwest British Columbia, Canada (average density, 760 people/km2). Pollution and walkability were estimated using geographic information system (GIS) mapping software (ArcGIS; ESRI, Redlands, CA, USA). To our knowledge, our study is the first to quantitatively assess the spatial intersection of walkability and air pollution.

7Vancouver, which sits prominently at the top of international rankings of livable cities (The ), is a useful study region because of the wide range in walkability levels. It is often cited as a well-planned city with walkable neighborhoods and comparatively clean air. However, the suburbs in Vancouver are as unwalkable as the suburbs found in typical sprawling regions.

8Concentrations of nitric oxide (NO), which is an indicator of traffic exhaust and of O3, a regional, secondary pollutant, were estimated for 49,702 (89%) of the 56,099 postal codes in Metro Vancouver. Walkability estimates were then generated for those postal codes. Postal codes in this region are typically one city block-face (one side of a block) or smaller (Metro Vancouver’s average postal code size is 39 people, or 0.05 km2). Because postal codes have roughly equal populations, analyses by postal code are roughly population-weighted analyses. The Canadian census provides a 1-to-5 low-to-high measure of affluence [quintile of annual income per person equivalent (QAIPPE)] for each postal code.

9Walkability captures the proximity between functionally complementary land uses (live, work, play) and the degree of route directness or connectivity between destinations. Our walkability estimates incorporated four parameters: a) net residential density (D), the number of dwelling units per square kilometer of residential land; b) intersection density (I), the number of intersections per square kilometer; c) retail floor area ratio (R), the retail shop floor-area divided by retail land area; and d) land-use mix (M), the evenness (i.e., equality) of floor space among categories of land use. We calculated each parameter for a 1-km network buffer around each postal code and then a relative walkability score (W; unitless) for that postal code as

10where ZD, ZI, ZR, and ZM are statistical Z-scores [unitless (mean ≈ 0; SD ≈ 1)] for D, I, R, and M, respectively. Values for I are large where streets are well connected (e.g., a grid) and small where streets are poorly connected (e.g., cul-de-sacs). Values for R are large for multistoried retail buildings with little surface parking and small for one-story retail buildings with large parking lots (“big-box retail”). Values for M are large where land uses such as residential, retail, office, and entertainment are highly mixed, and small where they are spatially homogeneous. Values for M are calculated based on building area within 27 land-use categories. The data used in Equation 1 were derived from three sources: a) locations for postal code centroids (Postal Code Conversion File, version 4D) and demographic information such as QAIPPE are from the 2001 Canadian Census, b) street network data are from the 2001 CanMap (DMTI Spatial, Markham, Ontario, Canada), and c) land-use data are from the 2001 British Columbia Property Assessment (BC Assessment, Victoria, British Columbia, Canada).

11Parameters in Equation 1 were selected because they have been found in many studies to be predictors of travel patterns and of walking in particular. For example, in Seattle, a 5% increase in the Equation 1 walkability score is associated with a 33% increase in the proportion of people who reported that they walked during a 2-day period.

12For each postal code, we evaluated the annual average concentrations of two pollutants: NO concentrations were estimated using land-use regression (LUR), and O3 concentrations were estimated using spatial interpolation of summer-only (May–September) monitoring data. We employed inverse-distance weighted average of the three nearest monitors; among the postal codes, the mean distance to the nearest O3 monitor is 3.7 km. These two approaches—LUR for NO and interpolation for O3—were selected because extant estimates are available for Metro Vancouver. In addition, the approaches match well with each pollutant’s spatial variability: generally, NO concentrations vary over short spatial scales (roughly one or a few city blocks), whereas O3 varies over long spatial scales (suburbs vs. urban core).

13NO was chosen because it is a primary vehicle-related pollutant and therefore serves as a marker for freshly emitted traffic exhaust, including both gasoline and diesel vehicles. O3 offers a useful comparison with NO because it is a secondary pollutant (i.e., formed in the atmosphere rather than emitted directly), so high concentrations tend to occur regionally downwind of the highest density areas (e.g., in suburbs). High O3 concentrations occur in summer.

14We have previously described and validated the LUR model and compared it against interpolation and mechanistic air dispersion models. LUR is a hybrid empirical-statistical approach that combines concentration measurements with GIS maps, thereby offering a high degree of spatial resolution. Briefly, 116 passive NO samplers were deployed for two 14-day periods at 116 sites in the study area. Mean concentrations during these two periods were successfully validated against annual averages from regulatory monitoring network data. For each measurement site, 55 variables were generated in GIS. Linear regression models of NO were built with the most predictive covariates. The model has an R2 of 0.62 and includes as covariates the number of major roads within 100- and 1,000-m radii circular buffers of the measurement sites, the number of secondary roads within a 100-m buffer, the population density within a 2,500-m radius, and elevation. As described previously, model–measurement comparisons for NO at monitoring station locations indicate reasonable to good agreement (mean bias, absolute bias, and error, 29%, 42%, and 1.6 μg/m3, respectively; model–measurement correlation, 0.7). Bias and absolute bias levels meet the goals (30% and 50%, respectively) and criteria (60% and 75%, respectively) suggested by for particulate matter concentration, but they do not meet the regulatory guidance (15% and 35%, respectively) for peak O3 concentrations.

15In Figure 1 and Table 2, we present walkability levels and air pollution concentrations. Mean values are 0.33 for walkability (unitless), 32.1 for NO (micrograms per cubic meter), and 27.7 for O3 (micrograms per cubic meter). Table 2 also presents results from two spatial analyses: distance from city center (the Vancouver courthouse) and spatial length scale for variability. The former analysis reports the median distance between each postal code and downtown Vancouver, stratifying postal codes by tertile of walkability, NO, or O3. The latter analysis measures the change in location needed to observe a modest change, in this case, one-half of the overall spatial SD in each parameter.

16Suburbs and exurbs tend to have high concentrations of O3, yet low levels of NO and walkability. The reverse holds for downtown areas. Those findings are expected: walkability parameters in Equation 1 are higher downtown than in the suburbs [see Supplemental Material, Figure 1, available online (doi:10.1289/ehp.0900595.S1 via http://dx.doi.org/)], and NO concentrations are elevated near the high density of vehicle emissions downtown. In contrast, O3 is a secondary pollutant formed from chemical reactions of nitrogen oxides (NOx) and volatile organic compounds (VOCs) in the presence of sunlight; as O3 forms, air migrates. In addition, NO reacts with and removes (“titrates”) O3, thereby reducing O3 concentrations where traffic emissions (and NO concentrations) are high.

17NO has a relatively short length scale (0.4 km), which indicates high spatial heterogeneity: changing location by a small distance can yield a comparatively large change in NO concentration. O3 has a longer length scale (8 km) because it is a regional pollutant and because estimations are derived from spatial interpolation of the monitoring data. The length scale for walkability is between that for NO and O3. Length scales (kilometers) for the Equation 1Z-scores vary significantly (and are between the NO and the O3 length scales): ZD, 5.6; ZI, 0.8; ZR, 4.8; ZM, 0.6. Variability among length scales highlights the possibility that certain areas might avoid high pollution levels yet have moderate or high walkability.

18Figure 2 presents average values for parameters as a function of distance from downtown. Walkability and NO levels decline sharply in the first 6 km from downtown, whereas O3 increases consistently along the urban-to-rural gradient. The high walkability area 16–19 km from downtown represents older satellite cities (e.g., New Westminster). The slight increase in walkability approximately 4 km from downtown derives from increases in intersection density. Importantly, locations approximately 4–8 km from downtown avoid the peak concentrations of NO and O3, yet they still have above-average walkability levels.

19A plot analogous to Figure 2, but for the Equation 1Z-scores [see Supplemental Material, Figure 1(doi:10.1289/ehp.0900595. S1)], reveals that intersection density has an overall maximum approximately 5 km from downtown and a local maximum approximately 18 km from downtown (which causes the walkability increase at 16–19 km in Figure 2). In contrast, for the remaining three Z-scores (net residential density, retail floor area ratio, land-use mix), values are high downtown, decline rapidly for the first 5–10 km, and then are relatively constant (roughly between 0 and −0.5) for distances > 10 km from downtown.

20Among the Equation 1Z-scores, NO concentrations are most correlated with residential density (Pearson r = 0.53) and least correlated with intersection density (r = 0.24). This finding suggests that “cars follow people, not roads”: neighborhoods with moderate or low residential density may avoid traffic exhaust yet still offer high intersection density (thereby improving walkability). Consistent with Figure 1, walkability is correlated with NO concentrations (r = 0.49) and inversely correlated with O3 concentrations (r = −0.70). The Supplemental Material contains additional pairwise comparisons and statistical models of the three attributes (walkability, NO, and O3), as well as correlations between those attributes and traffic counts [modeled peak morning traffic—automobile and truck—within several radii, 100–1,000 m, of each postal code centroid]. As expected, traffic is correlated with NO (r = 0.37–0.68) and walkability (r = 0.18–0.64) and inversely correlated with O3 (r = −0.14 to −0.59); correlations are larger in magnitude for automobile traffic than for truck traffic and for larger radii than for smaller radii. Among Equation 1Z-scores, traffic is least correlated with intersection density and most correlated with residential density (automobile traffic) and land-use mixing (truck traffic).

21Figure 3 presents distributions of postal codes by walkability and pollution tertile. Fewer than 4% of postal codes have high walkability yet low NO concentrations, whereas 24% of postal codes have high walkability and low O3 concentrations. (If attributes were uncorrelated, each entry in Figure 3 would be 11%.) The proportion of postal codes with low walkability and low NO is 6 times greater for highest-income than for lowest-income postal codes (22% vs. 4%). This finding is consistent with expectations, because on average neighborhoods are more affluent in suburbs than near the city center. Affluent postal codes tend to be less walkable and have lower NO; the reverse holds for poor postal codes.

22Figure 3 also illustrates postal codes in the most and least desirable tertile for each attribute: “Sweet-spot” postal codes are high walkability, low pollution; “sour-spot” postal codes are the opposite (low walkability, high pollution). We visited several sweet-spot neighborhoods. Our informal examination indicated urban-form attributes consistent with results presented above: high street connectivity, mixed land uses nearby, absence of large parking lots near retail space, sidewalks, and, in general, active streets with many walkers.

23Table 3 presents distributions of postal codes by income for specific walkability and pollution tertiles. Although NO and walkability values follow relatively monotonic patterns with income, O3 does not. High O3 occurs most commonly for middle- and upper-middle income postal codes, and low O3 occurs most commonly for lowest and highest income postal codes.

24The prevalences of overall sweet- and sour-spot postal codes in Table 3 are 1.7% and 4.6%, respectively. [If the attributes were uncorrelated, the prevalence would be 3.7% (i.e., 3−3).] Sweet-spot postal codes are heavily skewed toward high incomes (68% are highest income, QAIPPE = 5; only 3% are lowest income, QAIPPE = 1) and tend to be near but not at the city center (mean ± SD distance from city center, 6 ± 1 km). Sour-spot postal codes are predominantly from the middle three income quintiles and tend to be spread far from the city center (mean ± SD distance from city center, r= 22 ± 11 km).

25For postal codes with high walkability, we compared Equation 1Z-scores for low pollution versus not low pollution and found that the former group has a higher average score for ZR (1.7 vs. 1.0) and lower scores for the three other parameters (ZI, 0.7 vs. 1.1; ZM, 0.5 vs. 0.9; ZD, −0.3 vs. 0.3). Thus, what separates sweet-spot postal codes from other highly walkable postal codes is proximity to shops with limited parking per retail floor area.

26As discussed below, we investigated differences in air pollution and walkability for neighborhoods near versus not near mass transit by comparing attributes for postal codes inside and outside of 0.25-, 0.5-, and 1-mile circular buffers around light-rail (SkyTrain) stations. On average, the near-station postal codes tend to have higher walkability [mean difference (unitless) ± SD, 3 ± 1], higher NO (30–40%), and lower O3 (10%) than do areas not near rail stations. The difference in walkability is mainly attributable to differences in net residential density and land-use mixing.

27We explored urban spatial patterns for three environmental health attributes (walkability; NO, a primary pollutant with high concentrations near traffic and other combustion sources; and, O3, a secondary pollutant). Our results help elucidate spatial variations in community design, exposure to air pollution, and income. The methods presented provide new planning tools that can help identify where increased levels of physical activity may occur because of higher levels of walkability and where traffic emissions and O3 are least and most heavily concentrated. Walkability and air pollution have independently been associated with several adverse health outcomes, including physical inactivity, heart disease, mortality, and atherosclerosis, but health studies have not yet investigated interactive or joint effects. Our investigation builds on previous research and highlights the complexities of built environment and health relationships.

28We observed spatial differences among the three attributes. On average, conditions are better in the urban core than in the suburbs for O3 and walkability; the opposite is true for NO. We further identified neighborhoods that had reasonable levels for walkability and pollution. Sweet-spot locations, representing less than 2% of postal codes, are concentrated near but not at the city center. Most of them are higher income, indicating that they are highly desirable locations. In contrast, sour-spot postal codes are far from the city center, are spread more widely, and are occupied mostly by middle-income groups, although the degree of income segregation is less than for the sweet spots. Our finding that income is correlated with O3 concentrations but is inversely correlated with primary pollutant (NO) concentrations is consistent with similar findings in Southern California. Our results add to the literature on environmental justice aspects of air pollution, including on exposure to traffic-related pollution. Results in North America suggest that for NO and other primary pollutants, low-income and nonwhite populations face a disproportionate share of the burden of urban air pollution.

29Length-scale calculations highlighted that the aspects of the built environment studied here exhibit differing spatial patterns. For example, intersection density is a somewhat local attribute (shorter length scale), whereas retail floor area ratio is more regional. These age findings are important for several reasons. They suggest that when considering multiple attributes of the built environment, analysts may be able to identify areas trading off one attribute for the other, as well as sweet- and sour-spot locations. This attribute is useful for policy evaluation and for identifying locations meriting further investigation. In addition, people travel a finite distance each day. [In one study, which was not conducted in Vancouver, researchers found that the typical daily maximum distance from home is approximately 5 km.] During each travel day, individuals may experience a wider range of values for some aspects of the built environment than for others. For example, based on the length-scale results presented above, an individual willing to walk 2 km is likely to reach locations with different levels of land-use mixing but with similar retail floor area ratios.

30Policies designed to improve one attribute may hinder other attributes. For instance, modest reductions in NO emissions can increase O3 locally (because NO titrates O3) and either increase or decrease O3 regionally (depending on whether O3 chemistry is limited by NOx or VOCs). In areas with low walkability, people often have high levels of driving and of vehicle emissions per person, but if activities and emissions are dispersed, then concentrations of vehicle emissions may be low. Conversely, walkable neighborhoods may exhibit reduced per-capita vehicle use and emissions yet elevated traffic congestion, emissions, and concentrations if activities are highly concentrated. Ideally, one would understand all important impacts before recommending a policy action such as alternative growth patterns or transportation investments. Our research highlights that high NO exposures may occur where physical activity is encouraged through active transportation and in low-income areas. More work is needed to understand how to avoid that outcome, especially for susceptible subpopulations such as youth and the elderly. Policy options include siting residential buildings (especially schools, daycare centers, and assisted living facilities) back from major transportation corridors, rules or incentives to reduce high-emitting vehicles in urban centers, indoor air cleaners, and high-rise buildings that distance inhabitants from ground-level pollution. Shifts in the built environment can provide an intervention strategy for improving environmental health. Policies that provide affordable housing in walkable areas with cleaner air can help to offset the trend of low-income households experiencing worse than average primary air pollution; access to employment and other opportunities is an important aspect of overcoming disadvantages faced by traditionally underserved groups. Our approach would allow policy makers to evaluate and spatially optimize changes to the built environment that would yield the greatest health benefit per dollar.

31Walkable neighborhoods (and also neighborhoods served by mass transit) may allow people to reduce their daily travel distance, thereby decreasing vehicle emissions of NO and other O3 precursors. Improving air pollution and walkability will require changes in technologies, such as reducing emissions from motor vehicles, and also in urban design (e.g., land use mixing, mass transit).

32Our results presented above indicate that near-transit neighborhoods are more walkable, have lower O3 concentrations, and have somewhat higher (30–40%) NO concentrations than do other neighborhoods. The influence of building configuration on air pollution, although not directly explored here, may be an important aspect of this comparison. The increased density near rail stations is often accommodated via taller buildings. High-rise buildings elevate occupants above ground-level emissions. For primary pollutants, concentrations aloft (> 10–25 m, or three to five stories; ) can be several times lower than at ground level, which could more than compensate for the 30–40% ground-level NO concentration difference identified here. However, the urban street canyons formed by tall buildings reduce dilution rates and increase near-ground NO concentrations. In contrast, because of O3 titration by ground-level NO emissions in downtown areas, O3 concentrations are often greater aloft than at ground level.

33Additional limitations of the present study include the following: the walkability measure lacks data on sidewalks and parks; O3 concentration estimates do not include on-roadway O3 titration by NO; and, although the walkability score has been correlated with actual walking levels in several other cities, we have not yet completed that research in Vancouver (excluding our informal field examinations). Although this study focuses on neighborhood design (i.e., urban form; see Equation 1) as a predictor of walking levels (e.g., trip frequency, duration, or distance; percentage of people walking more than a threshold), other factors are also important, including, weather, topography, crime, pedestrian safety, noise, pollution, and personal preferences. Similarly, several factors affect the relationship between urban layout and pollutant emissions and concentrations. Limitations in extrapolating from our results to other situations include that other pollutants may not track spatially with NO and O3. Unique characteristics of Vancouver include the following: Vancouver is in a coastal air shed bounded by mountains, winds are predominantly from ocean to land, and walkability is relatively high when compared with other North American cities. In this study, we evaluated spatial variability, but future research is needed that considers temporal variability such as diurnal, weekly, and seasonal variations in NO and O3 concentrations and walkability and changes over time in urban form. Future research could address these limitations, investigate ways to target or optimize built-environment interventions, and document changes over time or measure the effectiveness of specific policies or actions. Other research could also include data on actual physical activity levels and pollutant exposures across urban form and address the health trade-offs between exposure to air pollution and physical activity for different age and income groups. Comparative research that documents the disparities in air-pollution exposure and walkability across income groups, while also identifying and testing strategies for improvement, is needed. Finally, an important next step toward maximizing the ability of urban form to impact health is the development and refinement of trip-planning tools that incorporate the various health-related attributes to allow individuals to reduce exposures during travel. offers a first step toward this type of tool.

34Our investigation explores potential environmental health impacts of neighborhood design. We found that neighborhoods with high walkability tend to have high levels of primary traffic-related pollution (NO) but low concentrations of O3. Attributes of the built environment evaluated here have differing spatial patterns, but all exhibit an urban–rural gradient. High-walkability and high-NO neighborhoods tend to be low-income neighborhood, whereas neighborhoods with high O3 tend to be middle income. Neighborhoods that exhibit low pollution and high walkability are rare and tend to be high income and located near to but not at the city center. Neighborhoods with high pollution and low walkability are far from the city center.

35The environmental attributes we studied are associated with several adverse health outcomes. The results emphasize that various aspects of environmental quality exhibit differing spatial patterns. Our analyses could usefully be applied to other cities, employed to track changes over time resulting from urban development and redevelopment, or used to design areas that are low in pollution exposure and that promote physical activity through increased walkability.