Satellite observations of NO₂ indicate legacy impacts of redlining in U.S. Midwestern cities

Urban housing in the United States is often highly segregated by race and socioeconomic status (Rothstein, 2017). In part, these patterns can be explained by (1) legacies of government policies and practices that prevented non-white and/or immigrant residents from moving into certain neighborhoods through explicit or indirect economic and legal barriers, and (2) active and passive steps made by municipal to federal governments that devalued and neglected neighborhoods composed primarily of residents of color (Woods, 2012; Fishback et al., 2021; Winling and Michney, 2021). “Redlining” was one such practice that categorized neighborhoods in metropolitan areas by assessed risk for home loans (Zenou and Boccard, 2000; Rothstein, 2017; Winling and Michney, 2021). The federally organized and supported Home Owner’s Lending Commission (HOLC) that established this systematic neighborhood categorization was a government program, initiated and funded to increase homeownership rates and economically support homeowners threatened by foreclosure during and after the Great Depression (Rothstein, 2017; Winling and Michney, 2021). The HOLC categorized neighborhoods into 1 of 4 grades. In ascending order of assessed “quality,” they were D, “Hazardous,” outlined in red or “redlined” (which also refers to the overall practice); C, “Definitely declining,” outlined in yellow; B, “Still desirable,” outlined in blue; and A, “Best” outlined in green or “greenlined” (Woods, 2012; Rothstein, 2017). Assessed risk for loans was based not only on the quality of the properties but on the racial, immigrant, and socioeconomic makeup of the neighborhood (Zenou and Boccard, 2000; Winling and Michney, 2021). The presence of non-white, immigrant, and low-income residents often settled a neighborhood in the high-risk category according to the HOLC, which meant loans were denied to residents or offered at higher interest rates (Woods, 2012; Park and Quercia, 2020; Fishback et al., 2021). Residents of so-called “redlined” neighborhoods were therefore less likely to recover financially from the Great Depression, less likely to benefit from intergenerational financial benefits of homeownership, and were more likely to be economically limited to areas in which resources were steadily declining due to decades of underinvestment and neglect at the municipal, state, and federal levels (Park and Quercia, 2020; Fishback et al., 2021).

While the explicit practice of redlining was made illegal over 50 years ago in 1968 with the establishment of the Fair Housing Act, its lasting impacts can be mapped to current social, economic, demographic, and environmental inequalities in communities and cities today (Woods, 2012; Rothstein, 2017; Park and Quercia, 2020). Redlined neighborhoods are still home to disproportionately low-income and non-white residents, while greenlined neighborhoods are broadly composed of higher income and predominantly white residents (Zenou and Boccard, 2000; Mitchell and Franco, 2018). It should also be noted that the Federal Housing Administration and its exclusionary practice of not insuring mortgages for residents of Black neighborhoods, which pre-dated HOLC practices, also contributed substantially to historic and ongoing inequality of homeownership (Fishback et al., 2021). An analysis of 10 major cities found significantly higher rates of cancer, asthma, poor mental health, and lack of health insurance in lower graded urban zones (Nardone et al., 2020c). Recently published studies using remotely sensed and ground-based measurements identified clear environmental differences between redlined and greenlined areas, including higher surface temperatures, more impervious surface cover, and lower canopy cover and green space in redlined neighborhoods (Grove et al., 2018; Hoffman et al., 2020; Moxley and Fischer, 2020; Wilson, 2020; Locke et al., 2021). Urban air quality, which is important to public health and disproportionally burdens neighborhoods with non-white and low-income residents (Miranda et al., 2011; Clark et al., 2014; Hajat et al., 2015; Demetillo et al., 2020; Liu et al., 2021), is not always studied in relation to the historic geographies demarcated via redlining, though there are recent analyses that examine these patterns (Namin et al., 2020; Nardone et al., 2020b). Inequities in air quality and how communities of color are disproportionately impacted by air pollution is observed in both urban and rural regions, as well as across different income levels (Tessum et al., 2019; Tessum et al., 2021). These trends are so persistent that air quality continued to be worse in non-white communities even during lockdown periods early in the COVID-19 pandemic (Kerr et al., 2021). In our study, we seek to address these persistent environmental justice issues in U.S. urban areas and apply remote-sensing methods to examine whether historic policies and their indirect and direct impacts relate to urban air quality in the present day.

Urban air quality has been explicitly connected to patterns of redlining in a few recent studies (Namin et al., 2020; Gonzalez et al., 2022; Lane et al., 2022), and a growing body of literature has connected these historic policies with increased prevalence of lower health outcomes (described and referenced below). Air quality, as estimated through measures of airborne carcinogens, has been shown to be worse in communities of color and low-income neighborhoods and linked to higher rates of asthma, multiple types of cancer, heat-related illnesses, and respiratory and heart disease in residents of these neighborhoods (Morello-Frosch and Jesdale, 2006; Krieger et al., 2020a; Nardone et al., 2020a; Li et al., 2021). Low birthweight, preterm birth, and allergies are also linked to poor urban air quality (Rumchev et al., 2007; Miranda et al., 2011; Kelly and Fussell, 2015; Nardone et al., 2020a; Krieger et al., 2020b).

Over 3,000 different air pollutants have been identified globally with only a fraction of those investigated for health effects and even fewer monitored in urban areas (Fenger, 2002). The Clean Air Acts of 1970 and 1977 include 6 air pollutants in a category called “Criteria Air Pollutants,” which are monitored and regulated nationally, and often occur at high concentrations (Fenger, 2002; Miranda et al., 2011). This category includes the pollutant we focus on for this study—nitrogen dioxide (NO₂)—as well as carbon monoxide (CO), ozone (O₃), particulate matter, lead, and sulfur dioxide. It has been estimated that more than 50% of the U.S. population lives in air that exceeds federal standards for at least one major air pollutant, making air quality of immediate importance in our rapidly urbanizing world (D’Amato et al., 2010); however, more recent regional legislative policies and air quality regulations may have shifted this estimate to a lower coverage.

A recent study identified that Black and non-white Hispanic communities are more likely to shoulder a disproportionately higher air “pollution burden” than white communities, despite contributing significantly less to pollution production through consumption patterns (Tessum et al., 2019). These results are further supported by recent studies focusing on pollutant exposure differences among HOLC neighborhoods (Gonzalez et al., 2022; Lane et al., 2022). Collectively, recent research emphasizes the important need to prioritize public health and environmental equity across the United States to identify and correct historic and ongoing harms.

However, mapping environmental disparities in air quality can be challenging to track in real time and at high resolution. Efforts to accurately measure patterns and sources of air pollution are limited by the spatial and temporal resolution of air quality monitoring data. Most publicly available continuously monitored air quality datasets (i.e., the U.S. Environmental Protection Agency’s [EPA’s] National Emissions Inventory) often only include one or two sampling locations per city. Namin et al. (2020) used data from the EPA’s National Air Toxic Assessment (NATA) to assess variation of airborne carcinogens between different HOLC neighborhoods and found a correlation between air pollution and HOLC grade, with increasing carcinogens detected in lower graded neighborhoods. NATA provides census-tract level data but is only available as single data point “snapshots” every few years (http://www.epa.gov/national-air-toxics-assessment), thus the lower temporal resolution that may not catch seasonal or year-to-year variability.

To address the discrepancy in resolution, satellite data show promise in examining the spatial and temporal variation of other environmental factors between HOLC neighborhoods within a city, in addition to across cities. Recent examples of this application, including both a smaller study of 3 cities and a larger analysis of 108 cities using satellite estimates, confirm that redlined neighborhoods experience significantly higher land surface temperatures (Hoffman et al., 2020; Wilson, 2020). The recent 2017 launch of the Copernicus satellite Sentinel-5 carrying the tropospheric monitoring instrument (TROPOMI) created a database of high-resolution, near-continuously monitored satellite-derived data of the pollutant NO₂, among others, that can be applied to analyze spatial variation across the globe. Further, increased temporal resolution allows for examinations of intra- and interannual comparisons within and across sites to answer questions about human, industrial, and seasonally mediated shifts in air quality (Goldberg et al., 2019; Goldberg et al., 2021). Recently, these data were applied to address changes due to the disruptions caused by the COVID-19 pandemic (Kerr et al., 2021; Jing and Goldberg, 2022; Tzortziou et al., 2022) and urban inequities (Demetillo et al., 2020; Anenberg et al., 2022). These data create new opportunities for higher resolution air quality monitoring in space and time, within and across cities and regions, providing a robust tool for examining current environmental variation in atmospheric pollutant loads.

Our study seeks to apply these high-resolution data to address how (1) U.S. Midwestern cities vary in the levels of a noxious air pollutant (NO₂) and (2) how this pollutant varies within these cities among historic HOLC neighborhood grades. We do this using satellite-collected TROPOMI data of NO₂ for the entire calendar year of 2019, prior to COVID-influenced shifts in traffic and industry during 2020, which have altered air quality in urban U.S. areas (Kerr et al., 2021; Cooper et al., 2022; Jing and Goldberg, 2022; Tzortziou et al., 2022). We hypothesized that, despite a half century since the elimination of codified neighborhood categorization via HOLC, the historic legacies of redlining and associated policies and practices would still be detected in these data, with higher air pollutant load in red and yellow neighborhoods compared to blue and green neighborhoods. Environmental inequities, detected at the ground or from space, demand close monitoring and immediate resolution to protect residents from current pollutants and the compounded impacts of climate change in these cities.

The top 10 most populated metropolitan areas in the Midwest were selected for inclusion in our analysis. Cities with populations at or exceeding 350,000 residents were selected with the aim of including cities that would provide increased air pollution variation due to greater intracity urbanization and population density patterns. The selected cities are in decreasing population size, Chicago, IL; Columbus, OH; Indianapolis, IN; Detroit, MI; Greater Kansas City metropolitan area, MO and KS; Milwaukee, WI; Omaha, NE; Minneapolis, MN; Wichita, KS; and Cleveland, OH (Table 1). The Greater Kansas City metropolitan area combines Kansas City, MO, and Kansas City, KS. This combination was necessary because the redlining map available through Mapping Inequality (see below) covered Greater Kansas City and was not divided by state. St. Paul was included as the 11th city in the analysis because although it has a population of less than 350,000 residents, it is the location of our home institution, Macalester College. Population size and geographic area reported represent 2019 data and the population density represent 2010 data, all collected from the U.S. Census Bureau.

Chicago is by far the most populous city included in this study with approximately 2 million residents as of 2019, while all other cities fall below the 900,000 residents mark (Table 1). Wichita, KS; Cleveland, OH; and St. Paul, MN, are the least populated cities at 380,000 residents or fewer each (Table 1). Cities vary widely in population density and spatial area, providing axes of variation to interpret the results of air quality differences and impacts on the population (Table 1).

Historic maps commissioned and created by the HOLC were made available through the University of Richmond’s publicly accessible website Mapping Inequality (Nelson et al., 2021). These maps are composite images of multiple historical redlining maps collected and georectified from various years between 1936 and 1940 (or year not indicated). Percentages of spatial area covered by different HOLC zones for each city were available through the website and used in our analysis (see Table 1). The area of the redlining maps may not exactly correspond to the current geographic area of their corresponding cities since individual cities may have expanded spatially since the redlining maps were made in the mid-1900s. The percentages of each city that fall under each HOLC grade are percentages of the entire redlining map and not the modern city area (Table 1). ArcGIS shapefiles of the maps of each city of interest were downloaded from Mapping Inequality and uploaded to Google Earth Engine for our air quality analyses.

The Copernicus satellite Sentinel-5 Precursor (Sentinel-5P) is an Earth-observing satellite carrying the TROPOMI which monitors atmospheric air quality and ultraviolet radiation, and contributes to climate monitoring (https://sentinel.esa.int/web/sentinel/missions/sentinel-5p). The satellite is designed for a 7-year life span and was launched in October 2017. Sentinel-5P collects data for near real-time atmospheric columnar densities of NO₂. The data became publicly available in July 2018, 8 months postlaunch, and has been continuously updated up to the present. The NO₂ data collected via the TROPOMI sensor is available for direct import into the Google Earth Engine where it can be visualized and downloaded (Gorelick et al., 2017). Nitrogen dioxide was selected for analysis as it is one of the 6 EPA Criteria Air Pollutants, given its impact on human health. The spatial resolution is 3.5 × 7 km² for NO₂, which is a resolution that lends itself effectively for intra-urban studies of air quality. TROPOMI is a polar-orbiting satellite, measuring tropospheric column gasses during its one-a-day overpass time (approximately early afternoon). Given the conditions during this time point, there may be measurements that are not accurately representative of the average daily pollutant load. It should also be mentioned that these data are not a stand in for air quality on the ground experienced by city residents, but there is promise in linking these for more appropriate estimates (Demetillo et al., 2020; Kerr et al., 2021).

The Earth Engine Code Editor (EECodeEditor) administered by Google Earth is an online programming tool to manipulate geospatial data and remotely sensed satellite imagery and information (Gorelick et al., 2017). We imported the publicly accessible HOLC map shapefiles from Mapping Inequality (Nelson et al., 2021) to the EECodeEditor to visualize the historic boundaries of red, yellow, blue, and green (i.e., D, C, B, A) neighborhoods on the current landscapes of the metropolitan areas as they appear in the current day. These shapefiles for each city of interest were imported as a FeatureCollection in the EECodeEditor. The HOLC neighborhood polygons vary in shape and can have a large range of sizes and areas even within a single city. To enhance the representation of the NO₂ data and minimize oversampling of pixels, we merged adjacent, contiguous polygons of the same HOLC grade by dissolving their shared boundaries to produce new, larger polygon of that grade (Table S1). For all analyses, we present data using these larger, merged polygons.

We imported Sentinel-5P data of the tropospheric level column NO₂ density (mol m^–2) between the surface and the top of the troposphere into the EECodeEditor within the bounds of each city. Data were imported for the time period of January 1, 2019, to January 1, 2020. Data from the year 2020 was excluded due to potential abnormal variation in fossil fuel emissions and resulting pollutant levels caused by changes in human industrial and transportation activity triggered by the COVID-19 pandemic (Le Quéré et al., 2020; Tzortziou et al., 2022; Jing and Goldberg, 2022). We imported the offline Sentinel-5P Level 3 NO₂ data product, which filters pixels with a QA >0.75. The NO₂ values from TROPOMI imagery within this time period were bounded and averaged spatially by the larger, merged HOLC polygons (described in the previous paragraph). Mean value calculations were area-weighted for polygons that contain more than one full pixel, and/or is partially (>0.5% within the polygon) covered by a pixel. The centroid of the pixel is not required to fall inside the polygon to be included in this calculation. The date and time stamp for each satellite image was also extracted, as well as the number of pixels per polygon, which is reported in Table S2.

The data for our analyses included the HOLC neighborhood grade of the merged neighborhood polygon, the mean air pollutant density value for the merged HOLC neighborhood, the area of each merged neighborhood polygon (Table S1), the number of pixels included in the polygon that were used to calculate the mean daily value (Table S2), and the date and time of the satellite image from which each data point was collected for all of 2019. Knowing that many cities contained smaller neighborhoods that were either noncontiguous with other same-grade neighborhoods (so they weren’t merged to make a larger polygon) or when merged were still very small in area, we only analyzed data for merged polygons which were greater than 2 km². This filtering by area allowed us to reduce overrepresenting small area neighborhoods in our analyses which limits oversampling of the same pixel. While this approach has the potential to remove some small, isolated, high-emitting small neighborhoods, we believe it allows for greater alignment of pixel area and minimizes bias introduced by these potential neighborhoods. Sample size of all NO₂ observations for all HOLC grades in each city are reported in Table S3. All NO₂ data were converted from mol m^–2 into Dobson Units, where 1 DU = 2.69 × 10¹⁶ molecules cm^–2.

All statistical analysis and graphical plots were created in RStudio (version 1.2.5032), using the programming language R (version 3.6.3) (R Core Team, 2021). Negative NO₂ values which indicated an error in the data collection were excluded from analysis, as were NA values. We conducted a power analysis to determine whether our sampling design met criteria to determine differences between the HOLC-graded neighborhoods. Using the pwr package in R (Champley, 2020) and setting our effect size to 0.10 based on our assumption that satellite-derived air pollutant detection may not yield large value differences among areas in urban regions (compared to the large effect sizes we might assume when comparing across different land use types, for instance), we found a minimum sample size of n = 274 to detect significance among 4 groups. Table S3 shows the sample sizes for our analyzed data, note that only 9 of 44 HOLC grades across all cities’ categories contained less than the suggested sample size from our power analysis. NO₂ followed a natural-log distribution and was natural-log-transformed for all statistical analyses, though plots show untransformed data for clarity.

To assess the influence of historic redlining neighborhood categorization on NO₂ in our all-city dataset analysis, we employed a linear mixed effects model, where NO₂ was set to vary by HOLC grade as a fixed effect (meaning that it was analyzed as the driver variable influencing the outcome) and city was set as a random effect (meaning it is likely to contribute to variance but is not the focused driver of variation in our study). Considering all cities’ data in the two-way ANOVA, we evaluated how the city and HOLC grades influenced the outcomes together. Tukey HSD post hoc analysis tests were performed on the two-way ANOVA results to determine significance among HOLC grades and cities. One-way ANOVA and Tukey HSD tests were performed within each city, considering each cities’ data independently, and significant results are reported. We used R packages multcomp, emmeans, lme4, and the tidy packages to conduct our data and statistical analyses (Hothorn et al., 2008; Wickham, 2021; Lenth, 2022).

We were also interested in assessing how coarse city demographic and geographic variables might be used to explain variance in the observed NO₂ values collected from TROPOMI. To do this, we used a linear mixed-model regression to examine the response of observed NO₂ across cities to the following variables: population size, population density, and geographic area of the city, with city included as a random effect. Each variable was assessed independently to compare the relative explanatory strength of the models. We report the R² and P values of each linear model.

Collectively, our results point to varied NO₂ emissions in Midwestern cities that are often related to both the city itself and the HOLC grades within a city. Analyses reveal the likelihood that legacy impacts of redlining are present as observed through the often-worse air quality estimates in D-grade neighborhoods, supported by other satellite-based estimates of common pollutants that have hazardous impacts on the respiratory health of residents (Demetillo et al., 2020; Kerr et al., 2021). Higher NO₂ levels are related to D-graded neighborhoods in many of the most populous Midwestern cities. Coarse variables of city population size and density also show some positive relationships with NO₂ levels.

Considering the entire dataset of all Midwestern cities, NO₂ varies significantly between HOLC grades (Figure 1; Table 2). A-graded neighborhoods showed the lowest NO₂ levels, followed by B, and then finally D neighborhoods, in which the highest mean values were observed (Figure 1). There were no significant differences among values of C neighborhoods and values of both B and D values (Figure 1). The differences between HOLC grades were highly significant (P < 0.001). All estimates of TROPOMI-derived values of NO₂ ranged between approximately 6 × 10^–5 and 1.35 DU, while the differences between the HOLC grades’ mean NO₂ did not exceed 0.012 DU. The largest difference in values considering the all-city data is observed between D and A grades, approximately 0.012 DU greater NO₂ in D-grade neighborhoods. For context, the interquartile range of the all cities data is 0.096, and the interquartile ranges for each HOLC grade (A, B, C, D) are 0.087, 0.095, 0.095, and 0.101, respectively.

Individual city analyses found NO₂ is significantly different among HOLC grades in about half of the sampled metropolitan areas (5 of 11). NO₂ did not significantly vary between any HOLC grades in Columbus, Detroit, Minneapolis, Omaha, St. Paul, and Wichita (Figure 2; Table 2). Four of those cities had sample size less than 274 in at least one neighborhood grade (Table S3), though there was no absolute trend connecting high or low sample size cities with significance. Of the 5 cities where significant differences among HOLC neighborhoods were detected, D neighborhoods had the highest average NO₂ values (Figure 2, Table 2). Specifically, in Chicago, Cleveland, Greater Kansas City, and Indianapolis, D and A neighborhoods were significantly different from each other with A grades showing the lowest values of NO₂ and D the highest (Figure 2). In Milwaukee, only D and B grades were significantly different, with B having the lowest mean values (Figure 2; Table 2). Chicago, Indianapolis, and Greater Kansas City showed the most distinct differences between HOLC grades of A and D, and the highest percent increases in NO₂ values at 16.2%, 13.0%, and 15.4%, respectively (Figure 2). Chicago anchors this trend with the greatest absolute difference in mean values of NO₂ between D and A neighborhoods at 0.0328 DU (Figure 2), for context the interquartile range considering data from all HOLC grades in Chicago is 0.159. Chicago also reports the highest overall values (in DU) of NO₂ across the study for each of the neighborhood categories (followed by Detroit), roughly double the estimated values of NO₂ of cities such as Cleveland, Columbus, Milwaukee, and Minneapolis, and triple the estimates for Wichita.

The trends in NO₂ found clear differences among HOLC grades only in some of the cities, so we further assessed additional urban demographic and geographic variables that might influence the observed relationships. To do this, we focused on estimates of population size and density, and geographic area. We acknowledge the coarseness of these metrics to compare cities that can be incredibly diverse in their geographies, climates, and industrial and developmental histories, but these metrics may also provide insight into broad controls on NO₂ dynamics across the urban Midwest. Metropolitan population size and density correspond positively but with low predictive power to NO₂ values (Table 3). Chicago’s population is more than double the size of all other cities included in our study and its population density is 1.5× greater (Table 1) and also has higher NO₂ than any other city (Tukey HSD indicates P < 0.0001 in all-city analysis). Metropolitan area does not exhibit a clear predictive relationship to NO₂ values among all cities (Table 3).

Redlining and associated discriminatory practices, whose impacts have been sustained and potentially intensified through direct policies (i.e., highway construction), development patterns (i.e., proximity to industrial activity and pollution), inequitable private lending practices and devalued properties, and political divestment driven by direct and indirect impacts of systemic racism, are observed to relate to modern day air quality in many large Midwestern cities as measured via satellite imagery. HOLC-designated D neighborhoods experience a mean increase in values of NO₂ of 7.3% ± 5.9% compared to A neighborhoods, revealing a legacy of redlining in current day urban air quality. However, a large range of variation is evident among the cities in our study: mean city values ranging from a minimal change of –2.4% in St. Paul to a substantial increase of +16.2% in Chicago. Our study demonstrates the potential of using TROPOMI satellite data to monitor NO₂ levels at high resolution in intra- and inter-urban settings to examine environmental inequities, and we encourage higher resolution, ground-based methods to further identify these persistent inequities.

Nitrogen dioxide data, as collected from the global-monitoring TROPOMI satellite sensor, show promise for measuring intraurban, neighborhood-level differences (Goldberg et al., 2019; Demetillo et al., 2020; Kerr et al., 2021; Cooper et al., 2022). This tool allows us to examine air quality in areas that may have previously been underrepresented in monitoring projects due to lack of resources, including historically redlined neighborhoods. NO₂ is among the most abundant outdoor air pollutants in urban areas (D’Amato et al., 2010) and is associated with increased risk of asthma, stroke, hospitalization for respiratory and cardiac issues, and susceptibility to respiratory infections (Kelly and Fussell, 2015). Anenberg et al. (2022) shows that although urban levels of NO₂ have been decreasing over the past few decades, current levels in high-income countries such as the United States still contribute significantly to prevalence of pediatric asthma, and even moderate declines in urban NO₂ levels can decrease this burden. Our findings show large within-HOLC-grade variation, which indicates other influences on NO₂, including time, day of the week, and climate (Goldberg et al., 2021). While the magnitude of the HOLC-grade variation may seem small, we believe these significant differences point to important trends in inequities of air quality burdens that necessitate ground-based quantification and signify potential exposure disparities that relate to health.

NO₂ from human activity mostly originates from fossil fuel combustion (Fenger, 2002), though recent studies have identified agriculture as a major emitter of NO₂ with influences on public health (Clark et al., 2019; Hill et al., 2019; Goldberg et al., 2021). In the U.S. Midwest, even the largest metropolitan areas are rarely far from intensive, industrial farming of corn and soy, so this agricultural contribution may have more influence in Midwestern versus coastal population centers (Hill et al., 2019). However, given the decay gradients of NO₂ from pollutant sources, the agricultural signal may not extend to urban regions in a significant manner; future examinations of urban-to-rural gradients could address these varied contributions (Demetillo et al., 2020). Our data does show higher absolute values of NO₂ in Chicago and Detroit, nearly triple those observed in Wichita, suggesting diverse influences of geography, climate, and industrial histories. It should be noted that NO₂ is not directly emitted from fuel combustion, rather it forms from the oxidation of nitric oxide (NO), which is produced from incomplete combustion of fuels (Fenger, 2002). However, since this reaction occurs rapidly, NO₂ is formed close to the point source of NO pollution, and thus still highly correlates with areas of high fossil fuel emissions (Fenger, 2002).

NO₂ may also indirectly affect air quality and public health by increasing ground-level ozone (O₃), a secondary pollutant formed from the reaction of NO₂ with other precursor compounds emitted by human industrial activity (Tu et al., 2007). Ozone has been identified as one of the two most damaging air pollutants to human health (Miranda et al., 2011; Hajat et al., 2015; Kelly and Fussell, 2015), and also one of the most prevalent outdoor air pollutants in urban areas, making it of great importance for urban public health (D’Amato et al., 2010). The relationship of fossil fuel emissions and air quality is further complicated because the presence of NO may decrease already present O₃ by oxidizing it to NO₂, so areas of high vehicle traffic may have lower levels of O₃ and higher levels of NO₂ depending on the dynamics of NO and NO₂ in the atmosphere (Fenger, 2002). In the future, this dynamic might change further as climate change increasingly pushes air temperatures in urban areas beyond historic norms to new extremes. Specifically, climate change may spur higher formation rates of O₃, increasing the frequency of high O₃ “events” in which levels rapidly accumulate beyond a hazardous threshold (D’Amato et al., 2010), and also increasing the number of peak heat days that coincides with higher NO₂ levels (Goldberg et al., 2021).

Our study found that NO₂ varies significantly between HOLC grades when considering all 11 sampled Midwestern cities. Overall, the greatest difference in NO₂ appears between D-grade neighborhoods with the highest values and A-grade neighborhoods with the lowest, and this variation translates to a mean increase of +7.3% among cities. Chicago exhibited the highest overall levels of NO₂ compared with all other cities, potentially attributable to its large population and congested vehicle traffic. Kerr et al. (2021) demonstrate disproportionately higher burdens of high NO₂ due to vehicle traffic in non-white and low-income neighborhoods—demographic factors which are often represented in redlined areas—even during the periods of COVID-19 lockdowns, when traffic was greatly reduced overall. We also observed a general trend that cities with significant differences in NO₂ between A- and D-grade neighborhoods tended to have the proportion of neighborhoods strongly skewed toward C and D grades, with around 70%–90% of the HOLC designated city area given a C or D grade. Though, Detroit and Wichita also have a similar neighborhood skew, but do not exhibit significant differences in NO₂. It may be that cities that were comprised of more C and D neighborhoods were subject to different development and investment/divestment patterns than the other Midwestern cities, including more extreme industrialization and intracity highway construction, or already contained more C and D neighborhoods due to preexisting industry and/or existing demographics (Rothstein, 2017). We encourage city-specific multidisciplinary research to examine the drivers and causes of NO₂ in the past, now, and into the future to more completely address these issues.

When we evaluated each city individually, comparing NO₂ across HOLC grades, we found D neighborhoods to have the highest levels of NO₂ in cities that exhibited significant variation across grades. In Chicago, the NO₂ is 16% greater in D neighborhoods compared to A neighborhoods. NO₂ is greater in D neighborhoods by over 10% in 4 out of 5 cities which showed significant differences (Chicago, Indianapolis, Greater Kansas City, and Cleveland). Differences of above 10% for 4 of the 11 cities—as high as 16% for Chicago—are not marginal shifts, especially considering this is an analysis of a single year, 2019, nearly a century after neighborhoods were categorized by the HOLC, and over 50 years after the Fair Housing Act. These data provide evidence that D neighborhoods in large Midwestern cities experience a larger burden of NO₂ pollution than A neighborhoods, demonstrating a modern legacy of redlining and associated inequities that is observed in persistently unequal air quality. Previous studies also support the ability of TROPOMI-derived NO₂ air pollution data to detect differences at high spatial resolutions (Goldberg et al., 2019; Demetillo et al., 2020; Kerr et al., 2021). Demetillo et al. (2020) emphasizes that TROPOMI data can accurately estimate relative differences in ground-level NO₂ values, but may not precisely determine absolute values of NO₂. We suggest our results be used to identify persistent areas of air quality inequality, and encourage higher resolution ground-based data be obtained to quantify absolute values and describe and predict individual and community health impacts. It is important to note that a lack of significant differences NO₂ as estimated by TROPOMI among HOLC neighborhoods does not necessarily imply equity in air quality on the ground; we strongly encourage ground-based monitoring to more accurately reveal absolute variation experienced by residents.

Detectable from space, NO₂ pollution poses an uneven public health threat in neighborhoods also facing higher temperatures under climate change due to a mix of increased impervious surface area and lower greenspace (Demetillo et al., 2020; Hoffman et al., 2020; Moxley and Fischer, 2020; Namin et al., 2020). Trees and other vegetation can directly reduce atmospheric NO₂ by uptaking and metabolizing NO₂ molecules through their leaf stomata (Nowak et al., 2014; Grote et al., 2016). On average, urban trees can reduce NO₂ levels by 15% (Rao et al., 2014). Grade A neighborhoods can have as much as 25% more tree canopy cover than D neighborhoods, and this may directly contribute to lower NO₂ values (Locke et al., 2021). These trends taken together emphasize the need for increased tree canopy cover in historically redlined neighborhoods to reduce NO₂ pollution as well as decrease land surface temperatures (Hoffman et al., 2020). However, there are documented issues with relying on tree planting alone to improve air quality, as trees can also contribute to increased pollen and the production of biogenic volatile organic compounds that can lead to ozone (Eisenman et al., 2019). NO₂ is also directly connected with fossil fuel combustion, particularly in industry and transportation (Chen et al., 2007; Goldberg et al., 2019); D-grade neighborhoods frequently have more and denser highways and factories due to selective planning and policies to expand these developments in the D-grade areas (Rothstein, 2017; Bullock et al., 2018; Demetillo et al., 2020; Moxley and Fischer, 2020).

Population and population density often correlate with increases in traffic and other sources of fossil fuel emissions, and therefore may be predictive of NO₂ levels (Ou et al., 2016; Goldberg et al., 2021). Our results showing potential relationships between any of the coarse urban geographic variables support the need for higher spatial resolution measurements to detect predictive trends and establish effective working models for understanding urban air quality and how it varies within and across cities (Lamsal et al., 2013; Zhu et al., 2014). While satellite sensing is a powerful tool, the resolution discrepancies in urban areas demand that ground-based sensor networks be installed to fully understand the nuances of each region and what controls these pollutants within and across redlined neighborhoods so we can work effectively toward environmental justice in these cities.

We demonstrate the potential of TROPOMI-collected data for monitoring and evaluating NO₂ air pollution across neighborhoods in cities historically categorized by HOLC grades in the U.S. Midwest. This is an advancement in monitoring air quality in the United States and globally since ground-based sensors within an area are often limited by number, spatial distribution, or by the frequency of temporal data collection. These methods can and should be applied to additional cities representing different climates, biomes, population densities, and development histories, as well as those impacted differently from current and projected climate change. Exploring the industrial, economic, and social histories of individual cities, including the ones we focused on in this study, will also allow for the understanding of historical drivers of observed inequalities.

TROPOMI collects data on other air pollutants, including CO and O₃, both of which are listed in the EPA’s 6 Criteria Air Pollutants. Current coarseness of those data—assessment of the total rather than tropospheric atmospheric column in the former, and pixels of very large size in the latter—prevent using them from accurately estimating local, ground-level concentrations. As satellite detecting capabilities become even higher resolution, these and other pollutants can be tracked by similar monitoring methods. Our study also applied an area-based cutoff to better align with NO₂ pixel size and minimize over-sampling issues; however, this approach limits understanding more about the presence and impact of potential isolated, high-emissions neighborhoods that could strongly affect residents’ health. To better incorporate data from these smaller neighborhoods and to effectively scale from ground to satellite, we suggest further exploration of high-resolution ground-based sensor networks that consider HOLC neighborhood histories for future investigations. The Pandonia Global Network (http://pandonia-global-network.org), a ground-based sensor network used to ground-truth tropospheric data collected from satellites, is an active and extremely promising initiative to further understand how to collect accurate, real-time, open-data on air pollutants that are important to human health (Judd et al., 2020; Verhoelst et al., 2021). However, the distribution and locations of these sensors is not ubiquitous, and not yet useful for within-city comparisons in the U.S. Midwest (Judd et al., 2020; Verhoelst et al., 2021).

We also encourage examining variation across seasons and day of week, as patterns and type of human activities may vary significantly, and temperature changes across seasons—particularly in the seasonal Midwest—may influence the chemistry of the formation of these compounds in the air (Goldberg et al., 2021). Further, assessing interannual variations and patterns may inform predictions of the long-term influence of climate change on urban air pollution. Finally, employing oversampling methods within the HOLC neighborhood polygons could allow for greater signal-to-noise ratios in the TROPOMI data (Lamsal et al., 2013; Zhu et al., 2014).

Despite variation in the development history, size and economy between the 11 large Midwestern cities, there is a consistent pattern of higher NO₂ in D-grade neighborhoods based on our study using satellite data collected for the calendar year of 2019. This is strong evidence of a legacy of environmental inequalities tied to the policy of federally commissioned redlining that persisted under decades of political marginalization and systemic racism (Rothstein, 2017). While we did not directly investigate the composition of the demographic data, it is likely D-grade neighborhoods are still primarily comprised of low-income residents and residents of color, since this pattern is seen in many cities today (Mitchell and Franco, 2018). Recent research that integrates demographic data with air quality data shows that pollution burdens in historically redlined neighborhoods fall disproportionately on Black and Hispanic residents (Demetillo et al., 2020; Kerr et al., 2021; Gonzalez et al., 2022; Lane et al., 2022) and are likely to influence the health of these residents (Nardone et al., 2020b; Nardone et al., 2020c; Anenberg et al., 2022). These relationships support the responsibility of city, state, and federal governments to rectify these historic and present-day inequalities and harms. Our results warrant further research identifying the specific sources and causes of concentrated air pollution within individual cities that contribute to modern day inequalities, which can then be targeted by government policies, and reparations for residents, a practice which is gaining traction at local levels (Brown et al., 2021). Based on our findings, we suggest future research look toward establishing high-temporal resolution sensors on residences, schools, and in public spaces in cities historically categorized by the HOLC, which would be key to understanding pollution burden on urban residents and addressing these burdens through public health support, city planning interventions, and financial reparations.

The following datasets were generated for our study and are openly available:

Hrycyna, Elizabeth; Mergenthal, Jennings G.A.; Noor, Saiido; Heskel, Mary A. 2022. Satellite observations of NO₂ indicate legacy impacts of redlining in U.S. Midwestern cities. Available at https://doi.org/10.5281/zenodo.6536071.

The supplemental files for this article can be found as follows:

Tables S1–S3. Docx

We would like to acknowledge Profs. Mary Montgomery of Macalester College and Sarah Fortner of the Science Education Resource Center of Carleton College for their roles in providing feedback as senior honors thesis committee members. We also deeply appreciate Macalester College for funding the collaborative summer undergraduate research fellowship and Wallace Fund faculty research grant support that led to this study. Publication of this article was funded by the Macalester College Dewitt Wallace Library Open Access Fund. Finally, we thank the creators of Mapping Inequality for making shapefile data openly available and sparking so many needed conversations and studies about redlining.

We, as an author group, have no competing interests to declare.

• Contributed to conception and design: EH, MAH.

• Contributed to acquisition of data: EH, SN, MAH.

• Contributed to analysis and interpretation of data: EH, MAH, SN, JGAM.

• Drafted and/or revised the article: EH, JGAM, SN, MAH.

• Final approval of the version to be published: EH, JGAM, SN, MAH.

How to cite this article: Hrycyna, E, Mergenthal, JGA, Noor, S, Heskel, MA. 2022. Satellite observations of NO₂ indicate legacy impacts of redlining in U.S. Midwestern cities. Elementa: Science of the Anthropocene 10(1). DOI: https://doi.org/10.1525/elementa.2022.00027

Domain Editor-in-Chief: Detlev Helmig, Boulder AIR LLC, Boulder, CO, USA

Associate Editor: Maria Val Martin, School of Biosciences, The University of Sheffield, Sheffield, UK

Knowledge Domain: Atmospheric Science