Skip to main content
NCBI home page
ACS AuthorChoice logoLink to ACS AuthorChoice
. 2023 Nov 7;57(48):19532–19544. doi: 10.1021/acs.est.3c03999

Ethnoracial Disparities in Nitrogen Dioxide Pollution in the United States: Comparing Data Sets from Satellites, Models, and Monitors

Gaige Hunter Kerr †,*, Daniel L Goldberg †, Maria H Harris ‡, Barron H Henderson §, Perry Hystad ∥, Ananya Roy ‡, Susan C Anenberg †
PMCID: PMC10702433  PMID: 37934506

Abstract

graphic file with name es3c03999_0004.jpg

In the United States (U.S.), studies on nitrogen dioxide (NO2) trends and pollution-attributable health effects have historically used measurements from in situ monitors, which have limited geographical coverage and leave 66% of urban areas unmonitored. Novel tools, including remotely sensed NO2 measurements and estimates of NO2 estimates from land-use regression and photochemical models, can aid in assessing NO2 exposure gradients, leveraging their complete spatial coverage. Using these data sets, we find that Black, Hispanic, Asian, and multiracial populations experience NO2 levels 15–50% higher than the national average in 2019, whereas the non-Hispanic White population is consistently exposed to levels that are 5–15% lower than the national average. By contrast, the in situ monitoring network indicates more moderate ethnoracial NO2 disparities and different rankings of the least- to most-exposed ethnoracial population subgroup. Validating these spatially complete data sets against in situ observations reveals similar performance, indicating that all these data sets can be used to understand spatial variations in NO2. Integrating in situ monitoring, satellite data, statistical models, and photochemical models can provide a semiobservational record, complete geospatial coverage, and increasingly high spatial resolution, enhancing future efforts to characterize, map, and track exposure and inequality for highly spatially heterogeneous pollutants like NO2.

Keywords: air pollution, nitrogen dioxide, environmental justice, satellites, TROPOMI, land-use regression models, urban air quality, photochemical models

Short abstract

Several data sets provide neighborhood-scale information on nitrogen dioxide pollution but have not been compared. We report how these data sets differ in their coverage, performance, and insights they provide for understanding and responding to environmental injustices.

1. Introduction

Ambient nitrogen dioxide (NO2) pollution is ubiquitous in the urban landscape and associated with numerous health effects, including premature death and adverse birth, respiratory, metabolic, and cardiovascular outcomes.1 While annual average NO2 concentrations in the United States (U.S.) remain well below the current Environmental Protection Agency (EPA) standard of 53 parts per billion by volume (ppbv), last revised in 1971, concentrations in many areas remain far above the current, more stringent World Health Organization air quality guideline value of ∼5.3 ppbv.2

The short distance-decay gradients of NO2 result in extremely heterogeneous spatial patterns within urban areas.3 Neighborhoods with the highest NO2 are often marginalized, racialized, and minoritized communities, presenting environmental justice concerns.2,4−6 The root causes of these injustices include redlining7,8 and other racist zoning practices that result in the disparate siting of polluting industries and transportation infrastructure.5,9,10 Considering higher disease rates and, therefore, air pollution susceptibility, NO2-attributable health risks can be even more inequitably distributed than NO2 concentrations.2

Currently, the EPA collects data on NO2 from roughly 450 in situ monitors intended to characterize near-traffic, urban, and background concentrations. This monitoring network is also used to demonstrate the attainment of U.S. air quality standards and for tracking regional air trends and was not intended to be used for characterizing intraurban pollution heterogeneity, given the sparseness of the monitors. However, many studies have relied on these in situ monitors to characterize NO2 exposure for health and environmental justice applications.12−24 These studies typically use concentrations at the nearest monitor to a given administrative unit (e.g., census tract, ZIP code) as representative of NO2 in that unit or sometimes apply more sophisticated techniques such as requiring that monitors be less than a certain distance from a unit or inverse distance weighting to the unit. The reliance on in situ monitors for these applications continues even with the proliferation of relatively high (∼1 km) spatial resolution, spatially complete satellite, land-use regression model (LUR), or hybrid satellite-model NO2 data sets.25

Spatially complete, high-resolution data sets are increasingly used to explore associations with health outcomes and identify communities overburdened by traffic-related air pollution.3,5,6,10,26,27

Here, we compare NO2 levels and ethnoracial NO2 disparities derived using in situ monitors and three spatially complete data sets, which represent different families of tools (i.e., satellite data, statistical models, and photochemical models) that can be used to characterize NO2. Our goal is not to determine which family of tools or specific data set is most accurate, as the sparse network of in situ observations precludes this type of assessment, and the best data set for a particular application may depend not only on the data set’s accuracy but also the specific research question. Rather, we seek to intercompare different families of NO2 data sets and reconcile differences, which may have profound implications for risk assessments, identification of highly exposed areas for targeted interventions, and our understanding of environmental injustices related to NO2 pollution. Our results can advance our ability to map, identify, and track NO2 and associated ethnoracial disparities.

2. Materials and Methods

2.1. NO2 Data Sets and Meteorological Data

We analyzed four complementary NO2 data sets with measurements or estimates over the contiguous U.S. These four data sets are broadly representative of the different families of tools available for understanding NO2: in situ measurements, photochemical models, statistical models, and remotely sensed observations. Other data sets not considered in our study were not available in 2019, the primary year of our analysis, or fit within our broad families of tools.

Observed annual average surface-level NO2 concentrations from regulatory-grade monitors were derived from the EPA’s Air Quality System database (henceforth “AQS monitors”) for 2010–2019.28 The precise number of AQS monitors in a given year in our study period ranges from 394 (2011) to 475 (2015). We include all AQS monitors, including those located in a near-road environment, in our analysis. Near-road monitoring was initiated toward the beginning of our 2010–2019 study period following the 2010 NO2 National Ambient Air Quality Standards review. The subset of monitors classified as near-road increased from 0% in 2010 to approximately 15%, or 68 of the 465 monitors, in 2019.29 While long considered a gold standard for ground-truthing NO2 concentrations, the chemiluminescence measurement technique employed by most AQS monitors (∼90% in 2019) can lead to chemical interference from other reactive species such as nitric acid, alkyl nitrates, and peroxyacyl nitrates. The impact of this interference, when compared with NO2 measured using a spectroscopic technique, results in an overestimation that can approach 50% in relatively clean areas (less in more polluted areas) and is dependent on the time of day, season, and location.30−32

The second data set we considered represents estimated annual average surface-level NO2 concentrations from a 1 km × 1 km “LUR” scaled with satellite measurements of tropospheric columnar NO2 for 2010–2019. Briefly, the LUR estimates global NO2 concentrations in 2010–2012 as a function of several land-use variables, which include terms related to emission sources (e.g., roadways, power plants, and wildfires), population density, and surface characteristics (e.g., tree cover, impervious surface area, and normalized difference vegetation index), and is further described in Larkin et al.33 These 2010–2012 estimates are scaled to each year in our study period using tropospheric columnar measurements of NO2 from the Ozone Monitoring Instrument (OMI). Additional details are included in Anenberg, Mohegh et al.26 This LUR has enabled retrospective studies on urban NO2 pollution,34 health impact assessments,35 and environmental justice-relevant assessments.36 Since the LUR trains to in situ NO2 observations, the aforementioned biases present in the AQS monitor data could percolate to the concentrations estimated by the LUR. It is beyond the scope of our study to understand where and how the monitor bias (versus bias from other inputs to the LUR or the regression technique itself) impacts LUR estimates. Moreover, it would be difficult to understand this important question since AQS monitors employing the bias-prone chemiluminescence measurement technique are not colocated with other monitors that use less bias-prone techniques.

Our third data set is remotely sensed tropospheric columnar NO2 measurements from the TROPOspheric Monitoring Instrument (“TROPOMI”) onboard the Sentinel-5 Precursor satellite. TROPOMI provides NO2 observations in the early afternoon (∼1330 h local time) at an unprecedented spatial resolution (5.5 km × 3.5 km since August 2019). In this study, we consider TROPOMI NO2 measurements (version 02.03.0137,38) averaged over 2019. These measurements were screened using a quality assurance flag exceeding 0.75 and oversampled to 0.01 × 0.01° (approximately 1 km × 1 km).39 Cloud and snow cover can both contribute to TROPOMI measurements being discarded given this quality assurance flag, and the annual averages used in study–particularly in the northern portions of the U.S. with lengthy winters–might be more representative of a warm season average.40 Beyond representing a different satellite instrument, TROPOMI NO2 measurements differ from the OMI measurements used to scale the LUR in their native spatial resolution (5.5 × 3.5 versus 24 × 13 km, respectively). At the time of this study, TROPOMI represents the state-of-the-science means of surveilling NO2 from space at an unprecedented spatial resolution. Since its launch in 2017, TROPOMI has aided in studies probing NOX emission sources,40,41 neighborhood-scale NO2 variability,42 and NO2 inequality.7,27

Finally, we considered annual average surface-level NO2 concentrations from a 12 km × 12 km photochemical model, the EPA’s Air QUAlity TimE Series Project (“EQUATES”), from which we obtain modeled concentrations for 2010–2019.43 While this data set has a much coarser spatial resolution than TROPOMI or the LUR, it represents the current frontier of spatial resolution possible for nationwide simulations of photochemical models based on computing limitations. The lowest vertical layer of the 35 layer model, which we refer to as the surface-level, has a nominal height of 19 m above the surface. EQUATES is based on the EPA’s Community Multiscale Air Quality (CMAQ) model (version 5.3.2). This model’s anthropogenic emissions derive from the combination of National Emission Inventory year (2002, 2005, 2008, 2011, 2014, but mostly 2017), day-specific fires, and CMAQ runtime parametrizations of lightning NOX, soil NOX, biogenic VOCs, sea-spray, and wind-blown dust.44 Of particular relevance to this study is how mobile (on road) emissions are estimated within EQUATES. These emissions are generated within the Sparse Matrix Operator Kernel Emissions at the county scale using emission factors from the MOtor Vehicle Emission Simulator and year-specific vehicle miles traveled and vehicle populations and thereafter spatially allocated to the 12 km × 12 km grid cell level using proxies.44 Although both the LUR and EQUATES provide modeled surface-level NO2 concentrations, EQUATES represents a complex, dynamical approach for simulating the physical and chemical processes responsible for ambient NO2, such as transport and chemistry. EQUATES is being used increasingly to support a variety of studies aimed at understanding historical air quality trends,45 characterizing future climate-driven changes in air quality,46 and calculating exposure-relevant air pollution metrics.47 We also consider EQUATES as it represents an estimate from the policy-relevant CMAQ model developed by the EPA Office of Research and Development that simulates the processes involved with and outcomes of pollution control strategies.

2.2. Demographic Data

We obtain annual demographic data from the U.S. Census Bureau’s American Community Survey 5 year estimates48 for 2010–2019 over the contiguous U.S. at the census tract-level. We consider five different ethnoracial categories in our analysis: non-Hispanic White, Black (including both Hispanic and non-Hispanic Black), Asian (including Native Hawaiian and Pacific Islander), Hispanic, and those identified as another race or two or more races (which we refer to as “Other” for brevity). Given the small proportion of the American Indian and Alaska Native population, particularly in the major metropolitan areas that were a focus of this study (0.8% of the U.S. population, 0.5% of the population of the 10 largest MSAs in 2019), we did not include these population subgroups in our study.

2.3. Disparity and Trends Analysis

Our unit of analysis is the census tract, which, in urban areas, can be thought of as roughly the same size as a neighborhood and optimally contains about 4000 residents. The typical area of census tracts (mean = 111.4 km2, median = 5.2 km2, standard deviation = 564.1 km2) is appropriate for the native 1 km × 1 km resolution of the TROPOMI and LUR NO2 data sets,5 whereas considering smaller administrative units (e.g., census block groups or blocks) could bias our results since many individual block groups or blocks would be smaller than the native resolution of the data sets. Considering finer resolutions than census tracts has been shown to only have minor impacts on NO2 disparities; specifically, national ethnoracial disparities only changed by 3% when determined using tract versus block data.49

We pair demographic data at the census tract level with NO2 from the LUR, EQUATES, and TROPOMI by averaging the gridded NO2 to each tract following the methods described by Kerr et al.5 In brief, all grid cells whose centroid is within the cartographic boundaries of a given census tract comprise the tract average.50 If a tract is too small to contain coincident grid cells or has irregular geometry (i.e., 5.2% of tracts when using the 1 km × 1 km LUR and TROPOMI data sets), we inverse distance weight to that tract’s centroid using the closest grid cell and that cell’s surrounding eight cells. Given its 12 km × 12 km resolution, adjacent census tracts may have identical tract-averaged NO2 concentrations from EQUATES if the adjacent tracts are located within the same EQUATES grid cell.

We link AQS monitors to demographic data in census tracts in different ways: principally, we use a nearest-neighbor approach (i.e., NO2 concentrations in a given tract are equal to the concentrations observed at the nearest monitor) but additionally consider fixed-distance cutoffs for this nearest-neighbor approach, an average of nearby monitors within different fixed-distances, and inverse distance weighting with and without fixed-distance cutoffs.

For each ethnoracial population subgroup described in Section 2.2, we calculate population-weighted NO2 levels and associated ethnoracial disparities from AQS monitors, the LUR, EQUATES, and TROPOMI across the contiguous U.S. as a whole and in urban tracts, rural tracts, and tracts belonging to metropolitan statistical areas (MSAs). The 378 MSAs in the contiguous U.S., which consist of multicounty clusters that contain a city with at least 50,000 inhabitants, are defined by the Office of Management and Budget and used by the U.S. Census Bureau.51 Census tract NO2 averages from the LUR or TROPOMI are classified as belonging to an MSA if the tract lies in counties within the MSA, and any AQS monitors located within these counties are treated as belonging to the MSA. We refer to MSAs by their colloquial names (“New York-Newark-Jersey City, NY-NJ-PA” = “New York”) and consider the amalgamation of all census tracts within MSAs as “urban” and all tracts outside MSAs as “rural.”

We also consider two alternative definitions of what counties constitute a given city. One is the “urbanized areas” definition from the U.S. Census Bureau.52 Urbanized areas form the urban cores of MSAs, and the counties comprising these urbanized areas largely align with those comprising the MSA. However, counties that lie on the periphery of MSAs are sometimes excluded from urbanized areas. The other definition treats the county within the MSA that contains the governmental center as the city (e.g., New York County, or Manhattan, represents New York, Los Angeles County represents Los Angeles, and Cook County represents Chicago). This approach generally only samples the densely populated urban core rather than suburban areas, and we refer to this definition as “county seat.”

When comparing NO2 levels or ethnoracial NO2 disparities across our four data sets, we use annual average NO2 and demographics from 2019, the most recent year in our study period, and the first full year with complete temporal coverage from TROPOMI. For these analyses, we consider all available AQS monitors. In our analysis of NO2 trends from 2010 to 2019, we use demographics and annual average NO2 concentrations from AQS monitors, the LUR, and EQUATES from this time period, updated annually. However, we calculate trends with two different methods: (1) using data from all available AQS monitors and (2) using only data from monitors that meet the EPA criteria for trend reporting. These criteria require that for a given AQS monitor to be included in trends reporting, it must have ≥4380 hly observations (i.e., ≥50%) in a given year, >75% of valid years during the study period, and not be missing more than two consecutive years of data.53 The methodology also dictates that if a site is missing data for a year, linear interpolation is used to fill in the missing year (unless the missing year is an end year, in which case, this year is replaced with the value of the nearest year). We define relative disparities as the ratio of population-weighted NO2 for each population subgroup to the overall population-weighted average for different aggregations (i.e., all, urban, or rural tracts) and MSAs.

To quantify the fidelity of the spatially complete data sets against in situ observations and intercompare NO2 levels across these spatially complete data sets, we use performance metrics such as the Pearson product–moment correlation coefficient and normalized mean bias (NMB) that have conventionally been used to evaluate model performance.54 We quantify the statistical significance of our results in two different ways. When we examine whether monitored versus unmonitored areas differ in their ethnoracial composition, we rely on the nonparametric two-sided Kolmogorov–Smirnov test and a significance level of α = 0.05. When we quantify the significance of NO2 trends or trends in relative disparities, we use the Wald Test with a t-distribution of the test statistic and the same significance level of α = 0.05.

3. Results

In 2019, the 465 AQS monitors were heavily concentrated in the Northeast, Colorado, and California, whereas broad swaths of the Pacific Northwest, Great Plains, and Southwest were unmonitored (Figure 1a). Two-thirds of census-designated MSAs (249 of 378), which accounted for 19% of the U.S. population, lacked even a single in situ monitor in 2019 (Figure S1). Even MSAs that have a large number of monitors relative to other MSAs are not adequately monitored, given the distance-decay gradients for NO2, which range from <0.5 km in unstable atmospheres to up to 2 km in stable atmospheres.3 For example, Los Angeles, CA and Houston, TX both have 18 monitors, the greatest number of monitors in the ten largest MSAs in the U.S. Yet, the mean distance between each census tract and the nearest AQS monitor in Los Angeles and Houston is 9.4 and 11.3 km, respectively. Approximately 98.8 and 98.9% of tracts are over 1 km from the nearest monitor in Los Angeles and Houston, respectively. While more in situ monitors would be needed in these (and other cities) to fully capture spatial NO2 gradients, the monetary and personnel costs might prohibit such a dense network.

Figure 1.

Figure 1

Open in a new tab

Annual 2019 average NO2 levels provided by (a) AQS monitors, (b) the LUR, (c) EQUATES, and (d) TROPOMI. Note that parts a–c share the same color bar, shown to the right of (b). (e–j) show an intercomparison of the four data sets. In (e–g), the LUR, EQUATES, and TROPOMI were sampled in census tracts containing in situ observations, whereas (h–j) represent all census tract-averaged values. Inset text in (e–j) indicates the sample size (N), the Pearson correlation coefficient (r), the slope (m) and intercept (b) of the orthogonal distance regression, and—if the two compared data sets have the same unit of measurement—NMB.

Across the contiguous U.S., spatial variations in NO2 are largely consistent across the three spatially complete data sets and, where they exist, the AQS monitors (Figure 1a–d). These data sets all indicate higher NO2 levels in urban areas and regions with heavy industry (e.g., California’s Central Valley). However, some differences emerge when examining the three spatially complete data sets. EQUATES simulates low concentrations (<3 ppbv) outside of urban regions, while the LUR estimates that concentrations in some of these regions may be higher (3–6 ppbv).

We next compare the correlation and bias of the three spatially complete data sets with observed NO2 concentrations from AQS monitors, first across the U.S. as a whole and then for urban versus rural environments. Census tract-averaged NO2 levels from the spatially complete data sets exhibit a similar correlation with AQS monitors in tracts containing monitors with EQUATES having the highest correlation (r = 0.77), TROPOMI intermediate (r = 0.74), and the LUR the lowest (r = 0.68) (Figure 1e–g). While EQUATES has a higher correlation with AQS, the magnitude of its underestimation (NMB = −0.30) against AQS monitors is nearly three times larger than the overestimation of the LUR (NMB = 0.11). Despite TROPOMI empirically measuring tropospheric NO2 columnar densities, the LUR statistically modeling surface-level concentrations as a function of land-use terms and satellite data and EQUATES simulating concentrations with meteorological and emission inputs and multiple differential equations, these data sets have a moderate to high degree of correlation with each other (r = 0.66–0.88, Figure 1h–j). The highest correlation among the spatially complete data sets is found for EQUATES and TROPOMI (Figure 1j). This level of agreement could reflect that EQUATES has a similar resolution to the native resolution of TROPOMI prior to oversampling (12 km × 12 km and 5.5 km × 3.5 km, respectively; Section 2.1), whereas the LUR is more likely to have finer-scale NO2 gradients given its 1 km × 1 km resolution.

The performance of the LUR, EQUATES, and TROPOMI against in situ monitors varies across rural and urban environments. The aforementioned LUR overestimation is present in both urban and rural areas, but the overestimation is more pronounced in rural compared to that in urban areas (NMB = 0.10 versus 0.41, respectively; Figure S2a,d). On the other hand, the magnitude of EQUATES’ underestimation is relatively consistent in both rural and urban areas (Figure S2b,e). When assessing the performance of LUR, EQUATES, and TROPOMI in areas with the highest observed NO2 concentrations, we find a degraded correlation and an underestimation in these highly polluted areas (Figure S2g–l). The degraded performance in highly polluted areas might reflect that monitors could capture peak concentrations near roadways or other NOX sources, while the grid cell nature of the spatially complete data sets would smooth over these fine-scale gradients. Based on these results, we might reasonably place more stock in the performance of these spatially complete data sets in urban rather than rural environments and in areas where NO2 levels are more average or intermediate rather than extreme.

Within individual urban areas, EQUATES and TROPOMI exhibit less spatial variability than LUR (Figure S3). NO2 levels from EQUATES and TROPOMI generally decrease radially outward from the city centers (Figure S3b,c). The LUR largely follows this pattern, but census tracts or clusters of tracts outside of city centers (and often along major roadways) can have local NO2 maxima (Figure S3a). Interdata set differences in NO2 gradients are city-specific and are evident when comparing intracity grid cell percentiles (Figure S4). For example, NO2 level percentiles differ by over 50% among the spatially complete data sets in a large portion of the western half of Dallas and eastern Houston (Figure S4b,c). Conversely, differences in the NO2 level percentiles in New York and Philadelphia among data sets are generally smaller in magnitude and spatial extent. Several factors could be at play in explaining these differences: the incorrect placement or spatial allocation of emissions sources in the LUR and EQUATES could impact their agreement with TROPOMI. Biases in meteorological inputs to EQUATES could lead to NO2 over- or underestimates. Furthermore, TROPOMI might detect reservoirs of NO2 aloft in its measurements of tropospheric columnar NO2, whereas EQUATES and the LUR estimate concentrations at the surface.

AQS monitors, stratified by urban–rural locality, are generally located in census tracts whose racial and ethnic compositions are similar to the overall urban or rural demographics in 2019 (Figure S5). Where statistically significant differences between the ethnoracial composition of monitored tracts versus the composition of all tracts exist, we find that monitored tracts tend to have a greater proportion of ethnoracial minorities. For example, the median proportion of the non-Hispanic White population in urban tracts with monitors is 47.1% versus 65.4% in all urban tracts (Hispanic: 18.7% in monitored urban tracts versus 9.2% in all urban tracts; Other: 8.2% in monitored urban tracts versus 5.4% in all urban tracts). The skew of monitors toward ethnoracial minority communities makes it more likely for these communities to have access to local air quality information but also likely highlights the disparities that led to the siting of monitors in these locations. We stress that these findings hold for the U.S. as a whole, and relying on monitors to understand ethnoracial disparities in geographically limited regions may leave particular population subgroups not monitored or undermonitored.

We find that AQS monitors and the spatially complete LUR, EQUATES, and TROPOMI data sets generally agree on the ordering of least-to-most exposed population subgroups for all census tracts and urban census tracts (Figure 2). Relative disparities in all and urban tracts have a similar magnitude, reflecting that a majority of tracts (81.6%) are classified as urban; however, disparities for all tracts are larger than those for only urban tracts. This result stems from the low NO2 levels in predominantly non-Hispanic White rural tracts that, when pooled into the calculation of relative disparities for all tracts, strengthen the magnitude of these disparities. When considering all census tracts, the four data sets indicate that population-weighted NO2 concentrations for the non-Hispanic White population are ∼5–15% lower than the overall population-weighted average, while concentrations are ∼15–50% higher for the Asian and Other population subgroups. The magnitude of the disparities for all census tracts, characterized as the difference in exposure between the most and least exposed population subgroups, is the smallest with AQS monitors (∼25%), intermediate with the LUR (∼30%), and largest with EQUATES (50%) and TROPOMI (∼60%).

Figure 2.

Figure 2

Open in a new tab

Relative NO2 disparities calculated using (a) the nearest AQS monitor to each census tract and tract-averaged NO2 from (b) the LUR, (c) EQUATES, and (d) TROPOMI in 2019. The dashed black vertical line highlights a value of 1, which indicates that a particular subgroup has the same NO2 levels as the overall population-weighted average for a given aggregation or MSA. Numbers in (a) represent the number of AQS monitors in each aggregation or MSA. Some relative disparities for the Asian population in parts (c) and (d) are out of frame, and the text indicates the magnitude of these disparities.

For rural census tracts, there is virtually no agreement in the ordering of least-to-most exposed subgroups, the magnitude of relative disparities, or whether a particular population subgroup experiences NO2 levels greater than or less than the population-weighted rural average across the different data sets (Figure 2). Both AQS monitors and the LUR indicate that the magnitude of disparities, characterized by the difference between the most versus least exposed population subgroups, is smaller (∼10–20%) than the magnitude of disparities from EQUATES and TROPOMI (40–60%). As was previously discussed, the LUR and EQUATES have biases in opposite directions in rural areas that are larger than those in urban areas (Figure S2), and there is also a relatively small number of rural AQS monitors (i.e., 73) and share of minoritized and marginalized populations outside of cities, which all may also contribute to the inconsistent disparities.

While disparities calculated using the four data sets were relatively similar for all census tracts and urban tracts, disparities in the 10 largest MSAs in the U.S., accounting for 26% of the population in 2019, are substantially different when examined through the lens of the nearest AQS monitor versus through the LUR, EQUATES, or TROPOMI (Figure 2). The LUR, EQUATES, and TROPOMI consistently show that non-Hispanic White populations experience lower NO2 levels than other population subgroups in Figure 2 while defining disparities using the nearest AQS monitor leads to the least exposed subgroup varying from MSA to MSA. In MSAs such as Los Angeles, Chicago, Dallas, and Atlanta, AQS monitors indicate that the Black population experiences the lowest NO2 levels. The range of relative disparities from AQS monitors is attenuated compared to the ranges from three spatially complete data sets. For example, in New York, relative disparities for all population subgroups from AQS monitors are within approximately 10% of each other, while the other data sets indicate a much wider range between the least and most exposed subgroups (30–40%).

We expected, a priori that the coarse resolution of EQUATES (i.e., 12 km × 12 km) would lead to different conclusions regarding relative disparities compared to insights from the higher resolution data sets. Yet, we found that at the national-, urban-, and MSA-level, EQUATES exhibits a high degree of agreement with the other spatially complete data sets. We further explored whether different city boundaries impact this agreement (Figure S6). When boundaries include both the city center (disproportionately Hispanic and non-White with higher NO2) and suburban or peri-urban areas (disproportionately non-Hispanic White with lower NO2), as do the MSA and urbanized area definitions, we draw similar conclusions from the LUR, EQUATES, and TROPOMI regarding which population subgroups are more versus less exposed to NO2 and the magnitude of disparities (Figures 2 and S7).

If cities are only considered to be the county seat (Figure S6), there is substantially lower agreement across the three spatially complete data sets, with little consensus in the ordering of least-to-most exposed population subgroups in several cities (Figure S7). Across the LUR, EQUATES, and TROPOMI data sets, however, the magnitudes of disparities in this densely populated core are generally smaller than those found when cities are defined as MSAs or urbanized areas. Additionally, the higher resolution spatially complete data sets (i.e., the LUR and TROPOMI) indicate that in certain urban areas, some ethnoracial minority groups may face lower population average NO2 levels than the non-Hispanic White population (e.g., the Black populations in Dallas and Washington, DC; Figure S7b,c).

We considered additional spatial interpolation approaches to estimate NO2 disparities from AQS monitors using the two largest MSAs in the U.S. (New York, NY and Los Angeles, CA) as examples, but we find that these approaches cannot reproduce the ethnoracial NO2 disparities found with LUR or TROPOMI data (Figure S8, Text S1). A cross-validation holdout analysis of AQS monitors and LUR data further highlights how the performance of these spatial interpolation approaches degrades as a function of distance between AQS monitors, whereas LUR performance is largely independent of distance as it pertains to the holdout analysis (Figure S9, Text S2).

Up to this point, we have compared ethnoracial NO2 disparities among the four data sets in 2019. We next explore how the LUR and EQUATES, like AQS monitors, can track disparities over time but with the added advantage that they provide estimated NO2 concentrations in every census tract of the contiguous U.S. Currently, TROPOMI cannot provide the same rich temporal data record offered by the LUR, EQUATES, or AQS monitors, but its continued operations are likely to increase its suitability for such longitudinal studies. Long-term NO2 decreases recorded by AQS monitors used for trends reporting (−0.29 ppbv yr–1) and from the LUR and EQUATES sampled at census tracts containing these monitors (−0.36 and −0.42 ppbv yr–1, respectively) agree within approximately 30% (Figure 3a), indicating that these spatially complete data sets are capturing the long-term decreases recorded by in situ monitors. Roughly 40% of the AQS monitors were discarded in order to meet the trends reporting criteria summarized in Section 2.3. Interestingly, when we compute trends using observations from all AQS monitors, we find decreases of only 0.09 ppbv yr–1. The magnitude of these decreases is only one-third the magnitude found with monitors used for trends reporting, likely due to the systematic addition of new monitors in areas with higher NO2 levels where trends may differ from the overall macro-level decreases.

Figure 3.

Figure 3

Open in a new tab

(a) Annual average NO2 concentrations averaged over all AQS monitors and all LUR or EQUATES census tracts (solid lines; “All”) and only at monitors that meet the trends reporting criteria and colocated census tracts (dashed lines; “Trends”). (b,c) Annual average population-weighted NO2 concentrations from the LUR and EQUATES for the U.S. population as a whole and different ethnoracial population subgroups. Note that TROPOMI is available only from May 2018 and, therefore, is not included. Inset text indicates trends, characterized by the slope of the least-squares linear regression in units of ppbv yr–1. The legend for (a) differs for (b,c), and all trends are significant.

The LUR and EQUATES also enabled us to track NO2 for different population subgroups during the 2010s. Notwithstanding the bias between these two data sets, both exhibit good agreement in the rate of change for different population subgroups and the magnitude of absolute disparities (i.e., the difference in concentrations for individual population subgroups and “All” in Figure 3b,c). While all population subgroups experienced decreases in their exposure to NO2 in the 2010s, the greatest reduction in exposure to NO2 and absolute disparities occurred for the Asian and Other ethnoracial categories. The rate of NO2 change for these two subgroups was 1.4–2.2 times larger (depending on whether we consider the LUR or EQUATES) than the rate of change for the non-Hispanic White population. Moreover, reductions in absolute disparities for the Asian and Other categories were up to four times larger than the reduction in absolute disparities for the non-Hispanic White population.

Despite substantial decreases in the level of NO2 and absolute disparities in the 2010s, changes in relative disparities were much less pronounced. Trends in relative disparities for all population subgroups were significant when defined using the LUR but only significant for the Black and White population subgroups when using EQUATES. For example, relative disparities for the Black population decreased by <5% (LUR: 1.17 in 2010, 1.13 in 2019; EQUATES: 1.23 in 2010, 1.20 in 2019). While both the LUR and EQUATES indicate that the Black population experienced decreases in relative disparities, the sign of trends in relative disparities from these two data sets differs for other population subgroups. For those identified as another race or two or more races (“Other” in Figure 3), the LUR indicates that relative disparities decreased by 8.2% (1.31 in 2010, 1.20 in 2019), while EQUATES suggests that relative disparities increased by 2.0% (1.30 in 2010, 1.32 in 2019).

4. Discussion

Our study documents how spatially complete, high-resolution satellite and modeling data sets provide a relatively consistent characterization of ethnoracial NO2 disparities regarding both the ordering of least-to-most exposed population subgroups and the magnitude of NO2 disparities. The sparse coverage of AQS monitors, however, yields different patterns of disparities compared to those found with the LUR, EQUATES, and TROPOMI, even when considering several different spatial interpolation methods. Interpolation of in situ monitor observations does not capture the gradients that affect marginalized communities overburdened by traffic-related pollution or agree with the literature on NO2 inequality.4−6,10 Currently, it is not possible to determine whether the LUR, EQUATES, or TROPOMI are closer to the truth: these three data sets have similar performance against AQS monitors. While AQS monitors are not sited densely enough to assess NO2 at a scale commensurate with its spatial variations and which of the three spatially complete data sets has the most realistic fine-scale NO2 gradients, there are some fixed and mobile monitoring campaigns (although not for 2019) in select urban areas that might aid future work on this topic.55−57

The rich spatiotemporal resolution of the LUR and EQUATES provides a means to track long-term NO2 trends for different ethnoracial population subgroups, and we find that NO2 decreases for racialized and minoritized subgroups have decreased the overall exposure of these subgroups and narrowed absolute disparities in NO2 inequality, though relative disparities have persisted (Figure 3b,c). Moreover, despite greater absolute NO2 decreases for racialized and minoritized subgroups, population subgroups ranked by their NO2 levels have remained consistent over time and across spatially complete, independent data sets. These persistent national-level disparities reflect inequity in pollution exposure across multiple geographic scales; Liu and Marshall58 previously reported that intraurban, regional, and state-level disparities each explain substantial portions of national ethnoracial disparities in NO2 exposure in the U.S.

We found substantial similarity across patterns and magnitudes of disparities determined with EQUATES and the higher spatial resolution data sets at the national, MSA, and urbanized area levels (Figures 2 and S6–S7). Clark et al.49 compared NO2 disparities calculated using data inputs at five administrative scales (state, county, census tract, census block group, and census block) and found that using data at the county- and state-levels underestimated disparities compared to those calculated using tract and finer data inputs. The 12 km × 12 km resolution of EQUATES lies between the county and tract scales investigated by Clark et al.,49 and the similarity we observed across the EQUATES, LUR, and TROPOMI-based estimates of disparities for MSAs and urbanized areas suggests that the 12 km scale data do not similarly underestimate disparities across metropolitan regions compared to the 1 km data sets (Figure S7). The lack of agreement in estimates of disparities across the spatially complete data sets within urban core areas (“county seat” in Figures S6 and S7), however, suggests that more work is needed to fully characterize finer-scale disparities between and within urban core neighborhoods (Figure S7). Recent mobile monitoring and modeling efforts59,60 have produced hyperlocal (i.e., subkilometer) NO2 observations or estimates, which may also prove useful for diagnosing disparities over these smaller scales.

Our finding that AQS monitors are relatively evenly distributed among the demographic landscape or, in some cases, are located in tracts with a higher percentage of ethnoracial minorities (Figure S5) might give the impression that relying on AQS monitors for studies that stratify by race or ethnicity would provide a realistic sampling of the population. However, we have clearly shown that these monitors are heavily concentrated in urban areas and limited geographic areas of the U.S. (Figure 1a), and interpolating monitors to cover the population yields meaningfully different disparities (Figure 2). Our analyses of monitors’ paucity (Figure S1) and demographic representativeness (Figure S5) build on several similar studies. Miranda et al.61 similarly found that a majority of counties in the U.S. lack access to information from in situ monitors on local ozone and fine particulate matter (PM2.5) pollution. However, in contrast to our findings regarding the placement of AQS NO2 monitors, others have shown that AQS and low-cost PM2.5 monitors tend to be in areas with a lower percentage of minority populations and higher socioeconomic status relative to unmonitored areas.62,63

Despite their complete spatial coverage, the LUR, EQUATES, and TROPOMI data sets have important limitations and known biases. LUR models are trained to monitors, tend to perform well where monitors sample a broad spectrum of land-use patterns, and may contain any biases present in monitor data, such as those due to the previously discussed measurement techniques that most AQS monitors use. Additionally, land-use terms used in the LUR and emission inventories for EQUATES can be outdated or are prone to errors related to how proxies are used to allocate emissions. Harkey et al.64 showed that near-surface NO2 is sensitive to many meteorological parameters and most sensitive to boundary layer height. Yet, boundary layer heights in CMAQ, the model upon which EQUATES is based, are not always simulated correctly when compared with ceilometer measurements,65 which can impact EQUATES’ simulation of NO2. TROPOMI retrievals are influenced by a priori data (e.g., vertical NO2 profile shapes, surface reflectivity, etc.) that generally have a lower spatial resolution than the oversampled tropospheric NO2 product used in our study. Potential biases in the satellite products used directly or indirectly in our study (i.e., TROPOMI, OMI) may vary seasonally, which could influence annual averages and estimated disparities. TROPOMI also measures total tropospheric columnar NO2 levels and thus reflects a combination of NO2 at the surface and in the free troposphere. However, the short atmospheric lifetime of NO2 minimizes its vertical and horizontal dispersion relative to other pollutants, and a relatively strong relationship between surface-level concentrations and tropospheric levels has been demonstrated.39 Additionally, TROPOMI measures NO2 at a single snapshot during the early afternoon, so any NOX emitted at other times may not be captured by TROPOMI and could impact our results if the locations of these emissions follow ethnoracial gradients. We expect that current and anticipated geostationary satellites, such as Tropospheric Emissions: Monitoring of Pollution (TEMPO) or Geostationary Extended Observations (GeoXO), capable of observing NO2 during all daytime hours, may shed light on this important issue. All of these limitations may contribute to the interdata set differences documented herein (e.g., Figures 1, S3, and S4). However, further exploring these limitations is beyond the scope of this study. We posit that some of these limitations (e.g., incorrect model simulation of boundary layer height) are less relevant to impact the ability of the different data sets to detect ethnoracial NO2 disparities as they are unlikely to co-vary with demographic patterns. We encourage the developers of these data sets to consider these limitations and how they relate to population-level exposure and disparities in future updates to their products.

All four families of data sets—in situ measurements, remotely sensed observations, statistical model estimates, and photochemical model estimates—are needed and contribute different insights regarding NO2 pollution and ethnoracial disparities. However, in situ measurements historically have carried the most weight in policy discussion and related action despite known measurement technique biases (Section 2.1). Looking ahead, augmenting AQS data with the high resolution, spatially complete NO2 data sets similar to those used in this study can provide a broader toolkit to monitor air quality, ensure that no segments of the American population are omitted from air quality assessments due to sparse monitor coverage, and allow for distributional analyses (Table 1). Already, satellite imagery from OMI, the predecessor satellite instrument to TROPOMI, has been used by the EPA to qualitatively illustrate the success of the Clean Air Act.66 Higher resolution satellite NO2 data sets are relatively new, precluding their use to assess long-term trends. However, TROPOMI’s continued operations and the launch of new instruments, such as NASA’s geostationary TEMPO mission, will further enhance our ability to understand NO2 from space by observing NO2 with higher temporal coverage (i.e., all daytime hours). Considering both LUR and satellite data (as well as population characteristics) may aid in identifying areas needing further investigation, guide the placement of additional in situ monitoring to accurately measure exposure to vulnerable populations, and inform pollution mitigation. Machine learning and other data fusion techniques that combine the strengths of these different families of data sets might take advantage of the ground truth provided by in situ observations, complete spatial coverage and semiempirical nature of satellite data, and the complete spatial coverage and ability to model different meteorological and emissions scenarios of statistical and geophysical models. Already, such techniques have leveraged the strengths afforded by these different data sets to estimate NO267,68 as well as other air pollutants, such as PM2.5.69

Table 1. Advantages and Disadvantages of the Four NO2 Data Sets from This Study and Potentially Appropriate Applications.

NO2 data set strengths weaknesses potentially appropriate applications for policy-relevant end-users
monitors direct measurements sparsely and unevenly distributed monitoring attainment of National Ambient Air Quality Standards
  many monitoring sites have a long-term data record prone to biases based on measurement techniques ground-truthing satellite and model data sets
  relatively low technical barriers to the usage of data costly (>$10,000)  
satellites semiempirical observations retrievals are prone to interference from clouds and surface albedo characterizing relative differences in NO2 levels across population groups or regions when surface-level concentrations are not needed
  high spatial resolution (∼1 km) achieved through oversampling techniques temporal resolution sacrificed through oversampling guiding placement of future AQS monitors or measurement campaigns (e.g., targeted mobile measurements)
  valuable tool to understand recent shocks to air pollution (e.g., COVID-19 pandemic) measures total tropospheric columnar rather than surface-level concentrations, so satellite-derived NO2 cannot be used in conventional health studies identifying potential hotspots and new emission sources
  (for OMI only) record since 2005 enables long-term trend analysis technical skills need to preprocess and analyze data estimating urban NOX emissions
    presently most satellites are polar-orbiting, providing ∼one snapshot per day in the early afternoon  
    the spatial resolution of satellite instruments prior to TROPOMI (launched in 2017) is nearly 1 order of magnitude coarser than TROPOMI’s resolution  
photochemical and statistical models increasingly high spatial resolution (statistical models: 50 m to 1 km; photochemical models: 10 km) computationally expensive to generate reporting long-term trends
  estimates surface-level concentrations, useable for health impact assessments ability to model recent year(s) may be hampered by lag to acquire up-to-date model inputs estimating NO2-attributable disease burdens and associated disparities
  useful for assessing long-term trends technical skills need to preprocess and analyze data (for photochemical models only) quantifying source sector contributions to ambient NO2
    (statistical models only) driven by monitor placement  
Open in a new tab

Decades-long decreases in ambient NO2 pollution continued in the 2010s, recorded by in situ AQS monitors, a photochemical model, and a high-resolution LUR. Larger decreases for minoritized populations have reduced absolute ethnoracial NO2 disparities, though relative gaps have persisted. Asian, Hispanic, Black, and multiracial populations, as well as those who do not identify with any of the census-designated racial categories, remained the most exposed population subgroups over time, suggesting that more targeted, place-based policies and interventions beyond the Clean Air Act and its amendments would be needed to further reduce NO2 inequality. The sparse AQS monitoring network is inadequate to monitor temporal trends in pollution inequality and capture neighborhood-scale NO2 disparities. Future characterizations of fine-scale NO2 variations may further elucidate whether modeled or remotely sensed data are more accurate in their assessment of NO2 levels. To this end, efforts to observe NO2 at a high spatiotemporal resolution are underway through mobile monitoring, such as campaigns in San Francisco, CA, and surrounding areas,59 and dense urban monitoring networks, such as the dense network of sensors in Chicago, IL from Microsoft Research’s Project Eclipse.70 In the meantime, the spatially complete modeled and remotely sensed NO2 data sets represent a leading, state-of-the-science means to surveil ambient NO2 and analyze distributional effects and should be used more widely in tandem with surface-level monitors by researchers and regulators.

Acknowledgments

We thank six anonymous referees for their review of our study. This study has not been formally reviewed by the Environmental Protection Agency (EPA), and views expressed in this document are solely those of the authors and do not necessarily reflect those of the Agency. The LUR data set used in this study can be accessed at disc.gsfc.nasa.gov/datasets/SFC_NITROGEN_DIOXIDE_CONC_1/summary,71 EQUATES at doi.org/10.15139/S3/F2KJSK,43 AQS monitors at aqs.epa.gov/aqsweb/airdata/download_files.html,28 and TROPOMI at 10.6084/m9.figshare.12909878.v6.72 The authors thank all those responsible for the creation and maintenance of these data sets. We gratefully acknowledge the computing resources provided on the High-Performance Computing Cluster operated by Research Technology Services at the George Washington University.73

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.3c03999.

  • Methods for spatially interpolating in situ monitor data; methods for cross-validation holdout analysis, number of in situ monitors in urban areas; data set performance in urban, rural, and high versus low pollution environments; data set intercomparison in largest urban areas; demographics in monitored areas; disparities using different definitions of urban areas; disparities from in situ monitors using different distance and interpolation thresholds; and cross-validation holdout analysis (PDF)

This study was funded by grants from the NASA Health and Air Quality Applied Sciences Team (#80NSSC21K0511) and the NASA Atmospheric Composition Modeling and Analysis Program (#80NSSC23K1002). We further acknowledge funding from the Environmental Defense Fund, supported by a gift from Signe Ostby, Scott Cook, and the Valhalla Foundation.

The authors declare the following competing financial interest(s): Kerr reports that he has served as a consultant for Environmental Defense Fund, Department of Justice, and California Air Resources Board. Anenberg reports that she has served as a consultant for the Environmental Defense Fund, Department of Justice, and Environmental Integrity Project. The remaining authors report no conflicts of interest relevant to this article.

Supplementary Material

es3c03999_si_001.pdf (2.8MB, pdf)

References

  1. HEI Panel on the Health Effects of Long-Term Exposure to Traffic-Related Air Pollution. Systematic Review and Meta-Analysis of Selected Health Effects of Long-Term Exposure to Traffic-Related Air Pollution; Special Report 23; Health Effects Institute: Boston, MA, 2022. https://www.healtheffects.org/system/files/hei-special-report-23_1.pdf (accessed July 06, 2022).
  2. Kerr G. H.; Martin R. V.; van Donkelaar A.; Brauer M.; Bukart K.; Wozniak S.; Goldberg D. L.; Anenberg S. C.. Increasing Disparities in Air Pollution Health Burdens in the United States. under review. [Google Scholar]
  3. Demetillo M. A. G.; Navarro A.; Knowles K. K.; Fields K. P.; Geddes J. A.; Nowlan C. R.; Janz S. J.; Judd L. M.; Al-Saadi J.; Sun K.; McDonald B. C.; Diskin G. S.; Pusede S. E. Observing Nitrogen Dioxide Air Pollution Inequality Using High-Spatial-Resolution Remote Sensing Measurements in Houston, Texas. Environ. Sci. Technol. 2020, 54 (16), 9882–9895. 10.1021/acs.est.0c01864. [DOI] [PubMed] [Google Scholar]
  4. Clark L. P.; Millet D. B.; Marshall J. D. National Patterns in Environmental Injustice and Inequality: Outdoor NO2 Air Pollution in the United States. PLoS One 2014, 9 (4), e94431 10.1371/journal.pone.0094431. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Kerr G. H.; Goldberg D. L.; Anenberg S. C. COVID-19 Pandemic Reveals Persistent Disparities in Nitrogen Dioxide Pollution. Proc. Natl. Acad. Sci. U.S.A. 2021, 118 (30), e2022409118 10.1073/pnas.2022409118. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Liu J.; Clark L. P.; Bechle M. J.; Hajat A.; Kim S.-Y.; Robinson A. L.; Sheppard L.; Szpiro A. A.; Marshall J. D. Disparities in Air Pollution Exposure in the United States by Race/Ethnicity and Income, 1990–2010. Environ. Health Perspect. 2021, 129 (12), 127005. 10.1289/EHP8584. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Hrycyna E.; Mergenthal J. G. A.; Noor S.; Heskel M. A. Satellite Observations of NO2 Indicate Legacy Impacts of Redlining in U.S. Midwestern Cities. Elementa: Science of the Anthropocene 2022, 10 (1), 00027. 10.1525/elementa.2022.00027. [DOI] [Google Scholar]
  8. Lane H. M.; Morello-Frosch R.; Marshall J. D.; Apte J. S. Historical Redlining Is Associated with Present-Day Air Pollution Disparities in U.S. Cities. Environ. Sci. Technol. Lett. 2022, 9, 345–350. 10.1021/acs.estlett.1c01012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Tessum C. W.; Paolella D. A.; Chambliss S. E.; Apte J. S.; Hill J. D.; Marshall J. D. PM2.5 Polluters Disproportionately and Systemically Affect People of Color in the United States. Sci. Adv. 2021, 7, 22. 10.1126/sciadv.abf4491. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Demetillo M. A. G.; Harkins C.; McDonald B. C.; Chodrow P. S.; Sun K.; Pusede S. E. Space-Based Observational Constraints on NO 2 Air Pollution Inequality From Diesel Traffic in Major US Cities. Geophys. Res. Lett. 2021, 48 (17), e2021GL094333 10.1029/2021GL094333. [DOI] [Google Scholar]
  11. Ritz B.; Wilhelm M.; Zhao Y. Air Pollution and Infant Death in Southern California, 1989–2000. Pediatrics 2006, 118 (2), 493–502. 10.1542/peds.2006-0027. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Mann J. K.; Balmes J. R.; Bruckner T. A.; Mortimer K. M.; Margolis H. G.; Pratt B.; Hammond S. K.; Lurmann F. W.; Tager I. B. Short-Term Effects of Air Pollution on Wheeze in Asthmatic Children in Fresno, California. Environ. Health Perspect. 2010, 118 (10), 1497–1502. 10.1289/ehp.0901292. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Bell M. L.; Ebisu K. Environmental Inequality in Exposures to Airborne Particulate Matter Components in the United States. Environ. Health Perspect. 2012, 120 (12), 1699–1704. 10.1289/ehp.1205201. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Malig B. J.; Pearson D. L.; Chang Y. B.; Broadwin R.; Basu R.; Green R. S.; Ostro B. A Time-Stratified Case-Crossover Study of Ambient Ozone Exposure and Emergency Department Visits for Specific Respiratory Diagnoses in California (2005–2008). Environ. Health Perspect. 2016, 124 (6), 745–753. 10.1289/ehp.1409495. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Smith G. S.; Van Den Eeden S. K.; Garcia C.; Shan J.; Baxter R.; Herring A. H.; Richardson D. B.; Van Rie A.; Emch M.; Gammon M. D. Air Pollution and Pulmonary Tuberculosis: A Nested Case-Control Study among Members of a Northern California Health Plan. Environ. Health Perspect. 2016, 124 (6), 761–768. 10.1289/ehp.1408166. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Ware L. B.; Zhao Z.; Koyama T.; May A. K.; Matthay M. A.; Lurmann F. W.; Balmes J. R.; Calfee C. S. Long-Term Ozone Exposure Increases the Risk of Developing the Acute Respiratory Distress Syndrome. Am. J. Respir. Crit. Care Med. 2016, 193 (10), 1143–1150. 10.1164/rccm.201507-1418OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Tallon L. A.; Manjourides J.; Pun V. C.; Salhi C.; Suh H. Cognitive Impacts of Ambient Air Pollution in the National Social Health and Aging Project (NSHAP) Cohort. Environ. Int. 2017, 104, 102–109. 10.1016/j.envint.2017.03.019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Tallon L. A.; Manjourides J.; Pun V. C.; Mittleman M. A.; Kioumourtzoglou M.-A.; Coull B.; Suh H. Erectile Dysfunction and Exposure to Ambient Air Pollution in a Nationally Representative Cohort of Older Men. Environ. Health 2017, 16 (1), 12. 10.1186/s12940-017-0216-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Eum K.-D.; Kazemiparkouhi F.; Wang B.; Manjourides J.; Pun V.; Pavlu V.; Suh H. Long-Term NO2 Exposures and Cause-Specific Mortality in American Older Adults. Environ. Int. 2019, 124, 10–15. 10.1016/j.envint.2018.12.060. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Li Z.; Xu X.; Thompson L. A.; Gross H. E.; Shenkman E. A.; DeWalt D. A.; Huang I.-C. Longitudinal Effect of Ambient Air Pollution and Pollen Exposure on Asthma Control: The Patient-Reported Outcomes Measurement Information System (PROMIS) Pediatric Asthma Study. Academic Pediatrics 2019, 19 (6), 615–623. 10.1016/j.acap.2019.03.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Thilakaratne R. A.; Malig B. J.; Basu R. Examining the Relationship between Ambient Carbon Monoxide, Nitrogen Dioxide, and Mental Health-Related Emergency Department Visits in California, USA. Sci. Total Environ. 2020, 746, 140915. 10.1016/j.scitotenv.2020.140915. [DOI] [PubMed] [Google Scholar]
  22. Lee E. K.; Romeiko X. X.; Zhang W.; Feingold B. J.; Khwaja H. A.; Zhang X.; Lin S. Residential Proximity to Biorefinery Sources of Air Pollution and Respiratory Diseases in New York State. Environ. Sci. Technol. 2021, 55 (14), 10035–10045. 10.1021/acs.est.1c00698. [DOI] [PubMed] [Google Scholar]
  23. McKeon T. P.; Anil Vachani; Penning T. M.; Hwang W.-T. Air Pollution and Lung Cancer Survival in Pennsylvania. Lung Cancer 2022, 170, 65–73. 10.1016/j.lungcan.2022.06.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Liao K.; Huang X.; Dang H.; Ren Y.; Zuo S.; Duan C. Statistical Approaches for Forecasting Primary Air Pollutants: A Review. Atmosphere 2021, 12 (6), 686. 10.3390/atmos12060686. [DOI] [Google Scholar]
  25. Anenberg S. C.; Mohegh A.; Goldberg D. L.; Kerr G. H.; Brauer M.; Burkart K.; Hystad P.; Larkin A.; Wozniak S.; Lamsal L. Long-Term Trends in Urban NO2 Concentrations and Associated Paediatric Asthma Incidence: Estimates from Global Datasets. Lancet Planet. Health 2022, 6 (1), e49–e58. 10.1016/S2542-5196(21)00255-2. [DOI] [PubMed] [Google Scholar]
  26. Dressel I. M.; Demetillo M. A. G.; Judd L. M.; Janz S. J.; Fields K. P.; Sun K.; Fiore A. M.; McDonald B. C.; Pusede S. E. Daily Satellite Observations of Nitrogen Dioxide Air Pollution Inequality in New York City, New York and Newark, New Jersey: Evaluation and Application. Environ. Sci. Technol. 2022, 56, 15298–15311. 10.1021/acs.est.2c02828. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. US Environmental Protection Agency . Air Quality System Data Mart [internet database]. https://aqs.epa.gov/aqsweb/airdata/download_files.html (accessed May 18, 2021).
  28. Watkins N.; Baldauf R.. Near-Road NO2 Monitoring Technical Assistance Document; EPA-454/B-12-002; United States Environmental Protection Agency: Research Triangle Park, North Carolina, 2021; p 135. accessed 2022-12-17. https://www.epa.gov/sites/default/files/2020-09/documents/nearroadtad.pdf.
  29. Dunlea E. J.; Herndon S. C.; Nelson D. D.; Volkamer R. M.; San Martini F.; Sheehy P. M.; Zahniser M. S.; Shorter J. H.; Wormhoudt J. C.; Lamb B. K.; Allwine E. J.; Gaffney J. S.; Marley N. A.; Grutter M.; Marquez C.; Blanco S.; Cardenas B.; Retama A.; Ramos Villegas C. R.; Kolb C. E.; Molina L. T.; Molina M. J. Evaluation of Nitrogen Dioxide Chemiluminescence Monitors in a Polluted Urban Environment. Atmos. Chem. Phys. 2007, 7, 2691–2704. 10.5194/acp-7-2691-2007. [DOI] [Google Scholar]
  30. Steinbacher M.; Zellweger C.; Schwarzenbach B.; Bugmann S.; Buchmann B.; Ordóñez C.; Prevot A. S. H.; Hueglin C. Nitrogen Oxide Measurements at Rural Sites in Switzerland: Bias of Conventional Measurement Techniques. J. Geophys. Res. 2007, 112 (D11), D11307. 10.1029/2006JD007971. [DOI] [Google Scholar]
  31. Lamsal L. N.; Duncan B. N.; Yoshida Y.; Krotkov N. A.; Pickering K. E.; Streets D. G.; Lu Z. U.S. NO2 Trends (2005–2013): EPA Air Quality System (AQS) Data versus Improved Observations from the Ozone Monitoring Instrument (OMI). Atmos. Environ. 2015, 110, 130–143. 10.1016/j.atmosenv.2015.03.055. [DOI] [Google Scholar]
  32. Larkin A.; Geddes J. A.; Martin R. V.; Xiao Q.; Liu Y.; Marshall J. D.; Brauer M.; Hystad P. Global Land Use Regression Model for Nitrogen Dioxide Air Pollution. Environ. Sci. Technol. 2017, 51 (12), 6957–6964. 10.1021/acs.est.7b01148. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Sicard P.; Agathokleous E.; Anenberg S. C.; De Marco A.; Paoletti E.; Calatayud V. Trends in Urban Air Pollution over the Last Two Decades: A Global Perspective. Sci. Total Environ. 2023, 858, 160064. 10.1016/j.scitotenv.2022.160064. [DOI] [PubMed] [Google Scholar]
  34. Song J.; Wang Y.; Zhang Q.; Qin W.; Pan R.; Yi W.; Xu Z.; Cheng J.; Su H. Premature Mortality Attributable to NO2 Exposure in Cities and the Role of Built Environment: A Global Analysis. Sci. Total Environ. 2023, 866, 161395. 10.1016/j.scitotenv.2023.161395. [DOI] [PubMed] [Google Scholar]
  35. Cheeseman M. J.; Ford B.; Anenberg S. C.; Cooper M. J.; Fischer E. V.; Hammer M. S.; Magzamen S.; Martin R. V.; Van Donkelaar A.; Volckens J.; Pierce J. R. Disparities in Air Pollutants Across Racial, Ethnic, and Poverty Groups at US Public Schools. GeoHealth 2022, 6 (12), e2022GH000672 10.1029/2022GH000672. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. van Geffen J.; Eskes H.; Compernolle S.; Pinardi G.; Verhoelst T.; Lambert J.-C.; Sneep M.; ter Linden M.; Ludewig A.; Boersma K. F.; Veefkind J. P. Sentinel-5P TROPOMI NO2 Retrieval: Impact of Version v2.2 Improvements and Comparisons with OMI and Ground-Based Data. Atmos. Meas. Technol. 2022, 15 (7), 2037–2060. 10.5194/amt-15-2037-2022. [DOI] [Google Scholar]
  37. Eskes H.; van Geffen J.; Sneep M.; Veefkind P.; Niemeijer S.; Zehner C.. S5P Nitrogen Dioxide V02.03.01 Intermediate Reprocessing on the S5P-PAL System: Readme File; ESA, 2021; (1.0), p 12. https://data-portal.s5p-pal.com/product-docs/no2/PAL_reprocessing_NO2_v02.03.01_20211215.pdf (accessed 14 Nov, 2022). [Google Scholar]
  38. Goldberg D. L.; Anenberg S. C.; Kerr G. H.; Mohegh A.; Lu Z.; Streets D. G. TROPOMI NO2 in the United States: A Detailed Look at the Annual Averages, Weekly Cycles, Effects of Temperature, and Correlation With Surface NO2 Concentrations. Earth’s Future 2021, 9 (4), e2020EF001665 10.1029/2020EF001665. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Goldberg D. L.; Tao M.; Kerr G. H.; Ma S.; Tong D. Q.; Fiore A. M.; Dickens A.; Adelman Z.; Anenberg S. C. Evaluating the Spatial Patterns of NOx Emissions in Polluted Areas with TROPOMI NO2. EarthArXiv 2023, 10.31223/X53T0M. [DOI] [Google Scholar]
  40. De Foy B.; Schauer J. J. An Improved Understanding of NOx Emissions in South Asian Megacities Using TROPOMI NO 2 Retrievals. Environ. Res. Lett. 2022, 17 (2), 024006. 10.1088/1748-9326/ac48b4. [DOI] [Google Scholar]
  41. Lee H. J.; Liu Y.; Chatfield R. B. Neighborhood-Scale Ambient NO2 Concentrations Using TROPOMI NO2 Data: Applications for Spatially Comprehensive Exposure Assessment. Sci. Total Environ. 2023, 857, 159342. 10.1016/j.scitotenv.2022.159342. [DOI] [PubMed] [Google Scholar]
  42. US Environmental Protection Agency . EQUATESv1.0: Emissions, WRF/MCIP, CMAQv5.3.2 Data—2002–2019 US_12km and NHEMI_108km, 2021. [Google Scholar]
  43. Foley K. M.; Pouliot G. A.; Eyth A.; Aldridge M. F.; Allen C.; Appel K. W.; Bash J. O.; Beardsley M.; Beidler J.; Choi D.; Farkas C.; Gilliam R. C.; Godfrey J.; Henderson B. H.; Hogrefe C.; Koplitz S. N.; Mason R.; Mathur R.; Misenis C.; Possiel N.; Pye H. O. T.; Reynolds L.; Roark M.; Roberts S.; Schwede D. B.; Seltzer K. M.; Sonntag D.; Talgo K.; Toro C.; Vukovich J.; Xing J.; Adams E. 2002–2017 Anthropogenic Emissions Data for Air Quality Modeling over the United States. Data Brief 2023, 47, 109022. 10.1016/j.dib.2023.109022. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Benish S. E.; Bash J. O.; Foley K. M.; Appel K. W.; Hogrefe C.; Gilliam R.; Pouliot G. Long-Term Regional Trends of Nitrogen and Sulfur Deposition in the United States from 2002 to 2017. Atmos. Chem. Phys. 2022, 22 (19), 12749–12767. 10.5194/acp-22-12749-2022. [DOI] [Google Scholar]
  45. Wilson B.; Pope M.; Porter J. R.; Kearns E.; Shu E.; Bauer M.; Freeman N.; Amodeo M.; Melecio-Vazquez D.; Hsieh H.; Tarasovitch M. Characterizing Changes in Extreme Ozone Levels under 2050s Climate Conditions: An Extreme-Value Analysis in California. Atmos. Environ.: X 2022, 16, 100195. 10.1016/j.aeaoa.2022.100195. [DOI] [Google Scholar]
  46. Hohsfield K.; Rowan C.; D’Souza R.; Ebelt S.; Chang H.; Crooks J. Evaluating Data Product Exposure Metrics for Use in Epidemiologic Studies of Dust Storms. GeoHealth 2023, 7 (8), e2023GH000824 10.1029/2023GH000824. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Manson S.; Schroeder J.; Van Riper D.; Kugler T.; Ruggles S.. IPUMS National Historical Geographic Information System: Version 14.0 [Database]; IPUMS, 2020. [Google Scholar]
  48. Clark L. P.; Harris M. H.; Apte J. S.; Marshall J. D. National and Intraurban Air Pollution Exposure Disparity Estimates in the United States: Impact of Data-Aggregation Spatial Scale. Environ. Sci. Technol. Lett. 2022, 9, 786–791. 10.1021/acs.estlett.2c00403. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. U.S. Census Bureau . 2019 TIGER/Line Shapefiles (machine-readable data files). https://www.census.gov/cgi-bin/geo/shapefiles/index.php (accessed June 03, 2021).
  50. U.S. Bureau of Labor Statistics . QCEW County-MSA-CSA Crosswalk. https://www.bls.gov/cew/classifications/areas/county-msa-csa-crosswalk.htm (accessed Aug 24, 2022).
  51. U.S. Census Bureau . Urban Area Relationship Files. https://www.census.gov/geographies/reference-files/2010/geo/relationship-files.html#par_textimage_470670252 (accessed April 18, 2023).
  52. U.S. Environmental Protection Agency . U.S. EPA 2018 Air Trends Report Methodologies, 2018. https://gispub.epa.gov/air/trendsreport/2021/documentation/Air_Trends_Report_Methodologies.docx (accessed Nov 14, 2022).
  53. Yu S.; Eder B.; Dennis R.; Chu S.-H.; Schwartz S. E. New Unbiased Symmetric Metrics for Evaluation of Air Quality Models. Atmos. Sci. Lett. 2006, 7 (1), 26–34. 10.1002/asl.125. [DOI] [Google Scholar]
  54. Beckerman B.; Jerrett M.; Brook J. R.; Verma D. K.; Arain M. A.; Finkelstein M. M. Correlation of Nitrogen Dioxide with Other Traffic Pollutants near a Major Expressway. Atmos. Environ. 2008, 42 (2), 275–290. 10.1016/j.atmosenv.2007.09.042. [DOI] [Google Scholar]
  55. Tessum M. W.; Larson T.; Gould T. R.; Simpson C. D.; Yost M. G.; Vedal S. Mobile and Fixed-Site Measurements To Identify Spatial Distributions of Traffic-Related Pollution Sources in Los Angeles. Environ. Sci. Technol. 2018, 52 (5), 2844–2853. 10.1021/acs.est.7b04889. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Miller D. J.; Actkinson B.; Padilla L.; Griffin R. J.; Moore K.; Lewis P. G. T.; Gardner-Frolick R.; Craft E.; Portier C. J.; Hamburg S. P.; Alvarez R. A. Characterizing Elevated Urban Air Pollutant Spatial Patterns with Mobile Monitoring in Houston, Texas. Environ. Sci. Technol. 2020, 54 (4), 2133–2142. 10.1021/acs.est.9b05523. [DOI] [PubMed] [Google Scholar]
  57. Liu J.; Marshall J. D. Spatial Decomposition of Air Pollution Concentrations Highlights Historical Causes for Current Exposure Disparities in the United States. Environ. Sci. Technol. Lett. 2023, 10 (3), 280–286. 10.1021/acs.estlett.2c00826. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Chambliss S. E.; Pinon C. P. R.; Messier K. P.; LaFranchi B.; Upperman C. R.; Lunden M. M.; Robinson A. L.; Marshall J. D.; Apte J. S. Local- and Regional-Scale Racial and Ethnic Disparities in Air Pollution Determined by Long-Term Mobile Monitoring. Proc. Natl. Acad. Sci. U.S.A. 2021, 118 (37), e2109249118 10.1073/pnas.2109249118. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Larkin A.; Anenberg S.; Goldberg D. L.; Mohegh A.; Brauer M.; Hystad P. A Global Spatial-Temporal Land Use Regression Model for Nitrogen Dioxide Air Pollution. Front. Environ. Sci. 2023, 11, 1125979. 10.3389/fenvs.2023.1125979. [DOI] [Google Scholar]
  60. Miranda M. L.; Edwards S. E.; Keating M. H.; Paul C. J. Making the Environmental Justice Grade: The Relative Burden of Air Pollution Exposure in the United States. IJERPH 2011, 8 (6), 1755–1771. 10.3390/ijerph8061755. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. deSouza P.; Kinney P. L. On the Distribution of Low-Cost PM2.5 Sensors in the US: Demographic and Air Quality Associations. J. Expo. Sci. Environ. Epidemiol. 2021, 31, 514–524. 10.1038/s41370-021-00328-2. [DOI] [PubMed] [Google Scholar]
  62. Mullen C.; Flores A.; Grineski S.; Collins T. Exploring the Distributional Environmental Justice Implications of an Air Quality Monitoring Network in Los Angeles County. Environ. Res. 2022, 206, 112612. 10.1016/j.envres.2021.112612. [DOI] [PubMed] [Google Scholar]
  63. Harkey M.; Holloway T.; Oberman J.; Scotty E. An Evaluation of CMAQ NO 2 Using Observed Chemistry-Meteorology Correlations: CMAQ EVALUATION WITH NO 2 -METOROLOGY CORRELATIONS. J. Geophys. Res. Atmos. 2015, 120 (22), 11775–11797. 10.1002/2015JD023316. [DOI] [Google Scholar]
  64. Gan C.-M.; Wu Y.; Madhavan B. L.; Gross B.; Moshary F. Application of Active Optical Sensors to Probe the Vertical Structure of the Urban Boundary Layer and Assess Anomalies in Air Quality Model PM2.5 Forecasts. Atmos. Environ. 2011, 45 (37), 6613–6621. 10.1016/j.atmosenv.2011.09.013. [DOI] [Google Scholar]
  65. Environmental Protection Agency . Our Nation’s Air, Trends through 2021, 2022. https://gispub.epa.gov/air/trendsreport/2021/(accessed Oct 15, 2022).
  66. Wei J.; Liu S.; Li Z.; Liu C.; Qin K.; Liu X.; Pinker R. T.; Dickerson R. R.; Lin J.; Boersma K. F.; Sun L.; Li R.; Xue W.; Cui Y.; Zhang C.; Wang J. Ground-Level NO 2 Surveillance from Space Across China for High Resolution Using Interpretable Spatiotemporally Weighted Artificial Intelligence. Environ. Sci. Technol. 2022, 56 (14), 9988–9998. 10.1021/acs.est.2c03834. [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Ghahremanloo M.; Lops Y.; Choi Y.; Mousavinezhad S.; Jung J. A Coupled Deep Learning Model for Estimating Surface NO 2 Levels From Remote Sensing Data: 15-Year Study Over the Contiguous United States. J. Geophys. Res.: Atmos. 2023, 128 (2), e2022JD037010 10.1029/2022JD037010. [DOI] [Google Scholar]
  68. Amini H.; Danesh-Yazdi M.; Di Q.; Requia W.; Wei Y.; Abu-Awad Y.; Shi L.; Franklin M.; Kang C.-M.; Wolfson J.; James P.; Habre R.; Zhu Q.; Apte J.; Andersen Z.; Xing X.; Hultquist C.; Kloog I.; Dominici F.; Koutrakis P.; Schwartz J.. Hyperlocal Super-learned PM2.5 Components across the Contiguous US; Research Square, 2022. preprint; In Review. [Google Scholar]
  69. Montgomery A.; Daepp M. I. G.; Abdin M. I.; Choudhury P.; Malvar S.; Counts S.; Horton D. E. Intraurban NO2 Hotspot Detection across Multiple Air Quality Products. Environ. Res. Lett. 2023, 18 (10), 104010. 10.1088/1748-9326/acf7d5. [DOI] [Google Scholar]
  70. Anenberg S.Nitrogen Dioxide Surface-Level Annual Average Concentrations V1 (SFC_NITROGEN_DIOXIDE_CONC); GES DISC, 2023. [Google Scholar]
  71. Goldberg D.Annual, Warm Season, and Monthly TROPOMI NO2 Averages over CONUS, 2021. [Google Scholar]
  72. The George Washington University High Performance Computing; Cluster. Building a Shared Resource HPC Center Across University Schools and Institutes: A Case Study, 2020. arXiv:2003.13629. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

es3c03999_si_001.pdf (2.8MB, pdf)

Articles from Environmental Science & Technology are provided here courtesy of American Chemical Society

RESOURCES