As others have demonstrated in cities throughout the USA over the period following the EPA Clean Air Acts [25], we saw marked overall city-wide reductions in PM2.5 and NO2 annually averaged levels from 2000 through 2016 in Boston, Nashville, and Detroit. However, within these cities, between-area pollution disparities persisted. Urban areas in the lowest redlining score category (A) were exposed to higher NO2 levels compared to areas with historical redlining (D and C) (Table 3). The most consistently significant between-area differences in NO2 levels occurred in the first half of 2000–2016. Between-area within-city differences in PM2.5 were less consistent. Longitudinal patterns in the magnitude and duration of NO2 pollution disparities were city-specific. Over the 2000–2016 period in Boston, disparities in NO2 levels in D-scored compared to A-scored areas remained relatively constant over time. In contrast, in Detroit the gap in NO2 levels between historically D-scored and A-scored areas decreased, and in Nashville, the gap in NO2 levels increased.

Deindustrialization with Urban Renewal: Detroit

In addition to the persistence of redlining-based disparities in resources and economic opportunities including home ownership, the census data confirm that in all areas, central Detroit experienced an overall population decline with a corresponding loss of income, increasing rates of poverty, and decreasing home values [26]. This likely reflects the progressive decentralization, relocation, and decline of the local auto and associated industries to suburban areas and the loss of local resources and industry-related jobs that had begun prior to the 2000–2016 period that we studied [27]. New economic resources to revitalize central Detroit resulted in two new sports stadiums, new restaurants and service industries, and some new housing but not necessarily in new industry or employment opportunities adequate to support local communities like the auto industry had done [28]. Against the background of regional (from the USA or Canada) pollution sources, a reduction in redlined-area–specific industrial and other local sources, the hollowing out of many neighborhoods, combined with new traffic patterns may have resulted in a more even city-wide dispersion of traffic-related pollution [26].

Do Redlined Area-Level Increases in Wealth Lead to Pollution Reduction? Boston vs. Nashville

In contrast to Detroit, Boston and Nashville experienced an increase in the proportion of college-educated individuals, median household income, and home values in historically redlined areas. Boston Metropolitan neighborhood demographic, housing, and income changes during the decades of interest in our report were mapped interactively and described in detail in a Harvard University Joint Center for Housing Studies report. The report used the U.S. Census data to summarize the complex Boston demographic changes between 1990 and 2016 as including “(1) growing racial and ethnic diversity but continued isolation within the region; (2) increased affluence but rising levels of income inequality; (3) somewhat more concentrated poverty; (4) declining amounts of modest-cost housing; (4) rising numbers of cost-burdened renters and homeowners; and (5) gentrification or stagnation in low-income tracts” [29]. The continued isolation that they describe in Boston resulted in part from post-1930s “block-busting” and other discriminatory practices that reinforced continued area-specific segregation of Black and other minoritized communities in historically redlined areas [12].

Overlapping with demographic change, between 1991 and 2006/7, the “new” Central Artery/Tunnel “Big Dig” that depressed high burden traffic in parts of Boston was completed. Its effects on neighborhoods and beneficial changes in near-roadway pollution were superimposed on effects of many remaining very old established transportation patterns [29]. The greatest density of traffic and pollution sources remained in central Boston, with its major highways, bus routes, local road traffic, railroad stations and lines (both commuter and freight), and nearby city port and airport activities. Thayer and colleagues have shown evidence of inequitable distribution of near roadway ultra-fine particle concentrations in central Boston [30]. However, they also point out that over time pockets of central Boston have developed with a “concentration of high-income housing coincident with high pollution levels.” In the midst of these demographic and transportation pattern changes in Boston, our data show that the relative gap in pollution between historically redlined and non-redlined neighborhoods remained similar from 2000 through 2016. This pollution disparity may have contributed to persistent health inequities [31,32,33].

In Nashville, we observed a widening difference in pollution between historically redlined and non-redlined areas over time. While Nashville is historically a railroad hub, and was located near coal-fired power plants, traffic appears to be the growing source of air pollution, with some highways traveling through urban neighborhoods [34]. The well-documented inflow of new wealth, entertainment and service industries to Nashville may have brought new traffic and other NO2 sources to historically redlined areas [35]. Simultaneously, the proportion of Black residents in those areas decreased, possibly due to displacement as housing grew less affordable (as documented by the Nashville Affordable Housing Task Force), or as work for long-time residents became less available [36, 37].

Area and City Differences in Which Criteria Pollutants Are Linked to Historical Redlining

The specific mix of air pollutant exposures linked to historical redlining varies by the specific mixtures of local pollution sources, patterns of roadways, siting of local industrial sources, and city-wide and regional contributions to exposures, as well as the method of measurement or estimation of pollutant levels [10, 38]. Thayer, Brugge and colleagues have shown the added insight gained by measuring near-roadway pollution that may not be among the criteria pollutants used by the EPA for regulation [30]. Studying the vicinities around NYC schools from 2000 to 2018, Jung and colleagues found that while pollution trended lower overall, the reduction over time in pollution was smaller for schools in redlined areas [14]. However, in NYC, the effects of redlining were seen specifically for PM2.5, traffic-related PM components (black carbon), and the primary emission NO but not for NO2 [14].

Health Significance of City-Wide and Between-Area, Within-City NO2 and PM2.5 Levels

Being exposed to NO2 and PM2.5 at the overall and area-specific levels that we estimated for Boston, Nashville, and Detroit has clinical relevance. In Project Viva, a Boston birth cohort studied during the same period (2000–2016), lower levels of lung function and greater airway inflammation (FENO) in teens were associated with higher NO2 exposures averaged over the first year of life and up to the teen years [39]. The median NO2 exposures in that study were 33.1 ppb (interquartile range [IQR], 10.4) and 24.5 ppb (IQR, 8.9 ppb) through early teens. In a 2000–2016 study of Medicare beneficiaries throughout the contiguous USA, higher levels of NO2 and PM2.5 were associated with more hospitalizations for cardiovascular disease, coronary heart disease, and cerebrovascular disease [40]. Associations were stronger at the lower end of the exposure distributions, which for NO2 was a median level of 10.6 ppb (IQR, 5.6 ppb). No threshold has been found for these adverse pollution health effects. These and other studies suggest that by 2016, while NO2 levels had improved, they still were relevant for health effects. For example, an annual average of 21.6 ppb in Nashville for the D areas has potential clinical relevance, as does the significant difference in annual average ppb exposures between the D and the A areas.

Study Strengths and Limitations

Our study contributes a longitudinal perspective to the literature, demonstrating cross-sectional city-specific disparities in air pollution exposures for historically redlined neighborhoods [10, 14, 16, 17]. While our findings may not be fully generalizable to other redlined U.S. cities, some lessons learned may be relevant. Our study is limited by a focus on two regulated criteria pollutants, PM2.5 and NO2, without complementary area-level data on toxic particle-level components or other gases that may have local sources. Cross-sectional city-specific studies have demonstrated that historically redlined communities have also had disparities in the traffic particle component black carbon, ultrafine particles, volatile organic compounds, and sulfur dioxide [11, 16, 17]. Nevertheless, the redlining-based pollution disparities we describe have high relevance to health risk [11, 16, 41, 42]. Higher air pollution exposures to either PM2.5 or NO2 at levels below EPA regulatory thresholds can lead to greater risk of adverse respiratory and cardiovascular outcomes [43,44,45].

While we did not have detailed longitudinal data on pollution sources to account for the city-specific trends and changes in pollution within and between areas with A through D redlining grades, the models estimating PM2.5 and NO2 over time took into account the change in those sources. The overall decline in air pollution in the past several decades (recently slowed in some parts of the country by the increase in pollution from wildfires [46]) is generally attributed to environmental (including traffic) regulation, enforcement of pollution controls, and deindustrialization. Superimposed are regional and local influences on pollution, some constant (geography) and others changing over time—climate, urban land use, siting of polluting industries, roads, traffic policies, public and private vehicle emissions controls. These factors merit study for defining opportunities to reduce urban pollution.

One challenge in our geospatial approach was alignment between historical HOLC maps, modern census tracts, and 1-km2-grid air pollution estimates. To conduct our analysis at the census tract level, we calculated a redlining score, which may have contributed to potential misclassification of redlining exposure. For Boston and Detroit, we also created an “ungraded” category representing locations within our urban perimeter not graded by HOLC maps that, at the time of HOLC mapping, represented non-residential areas (e.g., rural or sparsely populated areas or industrial or commercial areas). In our Boston and Detroit data, compared to historically redlined D areas, ungraded and A-graded areas had higher wealth and lower NO2 air pollution. As they were estimated from well-validated spatio-temporal prediction models, we acknowledge there is some error associated with the predicted pollution levels analyzed in this study. However, the magnitude of this error is certainly much less than that reported in those modeling efforts [47], as we analyze levels averaged over both space (census-tract) and time (annual averages).

We relied on census-based classifications of race, ethnicity, and socioeconomic indicators, but other area-level measures reflect structural inequities, some but not all of which may be linked to historical redlining. The census may undercount some marginalized populations. Persistent racial segregation and under-resourcing of minoritized communities with siting of polluting industries in redlined D neighborhoods have also resulted from other policies and practices, some existing long before HOLC map development. As a well-documented federally sponsored housing policy enacted across the country with accessible records and explicitly racist determinations, redlining associations with air pollution are easier to document than the effects of racial covenants or other discriminatory policies more difficult to measure or geographically localize. In a nationwide study, Lane and colleagues found that while air pollution disparities were larger by redlining category than by race and ethnicity, racial and ethnic disparities in pollution exposures persisted within each HOLC category [10]. Defining these disparities may require more spatial resolution—at the block rather than the tract level.