Ambient NO2 Air Pollution and Public Schools in the United States: Relationships with Urbanicity, Race-Ethnicity, and Income.

Schools may have important impacts on children's exposure to ambient air pollution, yet ambient air quality at schools is not consistently tracked. We characterize ambient air quality at home and school locations in the United States using satellite-based empirical model (i.e., land use regression) estimates of outdoor annual nitrogen dioxide (NO2). We report disparities by race-ethnicity and impoverishment status, and investigate differences by level of urbanicity. Average NO2 levels at home and school for racial-ethnic minoritized students are 18-22% higher than average (and 37-39% higher than for non-Hispanic, white students). Minoritized students are less likely than their white peers to live (0.55 times) and attend school (0.58 times) in areas below the World Health Organization's NO2 guideline. Predominantly minoritized schools (i.e., >50% minoritized students) are less likely than predominantly white schools (0.43 times) to be in locations below the guideline. Income and race-ethnicity impacts are intertwined, yet in large cities, racial disparities persist after controlling for income.

Disparities in Air Pollution Exposure in the United States by Race/Ethnicity and Income, 1990-2010.

BACKGROUND: Few studies have investigated air pollution exposure disparities by race/ethnicity and income across criteria air pollutants, locations, or time. OBJECTIVE: The objective of this study was to quantify exposure disparities by race/ethnicity and income throughout the contiguous United States for six criteria air pollutants, during the period 1990 to 2010. METHODS: We quantified exposure disparities among racial/ethnic groups (non-Hispanic White, non-Hispanic Black, Hispanic (any race), non-Hispanic Asian) and by income for multiple spatial units (contiguous United States, states, urban vs. rural areas) and years (1990, 2000, 2010) for carbon monoxide (CO), nitrogen dioxide (NO2), ozone (O3), particulate matter with aerodynamic diameter ≤2.5µm (PM2.5; excluding year-1990), particulate matter with aerodynamic diameter ≤10µm (PM10), and sulfur dioxide (SO2). We used census data for demographic information and a national empirical model for ambient air pollution levels. RESULTS: For all years and pollutants, the racial/ethnic group with the highest national average exposure was a racial/ethnic minority group. In 2010, the disparity between the racial/ethnic group with the highest vs. lowest national-average exposure was largest for NO2 [54% (4.6 ppb)], smallest for O3 [3.6% (1.6 ppb)], and intermediate for the remaining pollutants (13%-19%). The disparities varied by U.S. state; for example, for PM2.5 in 2010, exposures were at least 5% higher than average in 63% of states for non-Hispanic Black populations; in 33% and 26% of states for Hispanic and for non-Hispanic Asian populations, respectively; and in no states for non-Hispanic White populations. Absolute exposure disparities were larger among racial/ethnic groups than among income categories (range among pollutants: between 1.1 and 21 times larger). Over the period studied, national absolute racial/ethnic exposure disparities declined by between 35% (0.66µg/m3; PM2.5) and 88% (0.35 ppm; CO); relative disparities declined to between 0.99× (PM2.5; i.e., nearly zero change) and 0.71× (CO; i.e., a ∼29% reduction). DISCUSSION: As air pollution concentrations declined during the period 1990 to 2010, absolute (and to a lesser extent, relative) racial/ethnic exposure disparities also declined. However, in 2010, racial/ethnic exposure disparities remained across income levels, in urban and rural areas, and in all states, for multiple pollutants. https://doi.org/10.1289/EHP8584.

National Empirical Models of Air Pollution Using Microscale Measures of the Urban Environment.

National-scale empirical models of air pollution (e.g., Land Use Regression) rely on predictor variables (e.g., population density, land cover) at different geographic scales. These models typically lack microscale variables (e.g., street level), which may improve prediction with fine-spatial gradients. We developed microscale variables of the urban environment including Point of Interest (POI) data, Google Street View (GSV) imagery, and satellite-based measures of urban form. We developed United States national models for six criteria pollutants (NO2, PM2.5, O3, CO, PM10, SO2) using various modeling approaches: Stepwise Regression + kriging (SW-K), Partial Least Squares + kriging (PLS-K), and Machine Learning + kriging (ML-K). We compared predictor variables (e.g., traditional vs microscale) and emerging modeling approaches (ML-K) to well-established approaches (i.e., traditional variables in a PLS-K or SW-K framework). We found that combined predictor variables (traditional + microscale) in the ML-K models outperformed the well-established approaches (10-fold spatial cross-validation (CV) R2 increased 0.02-0.42 [average: 0.19] among six criteria pollutants). Comparing all model types using microscale variables to models with traditional variables, the performance is similar (average difference of 10-fold spatial CV R2 = 0.05) suggesting microscale variables are a suitable substitute for traditional variables. ML-K and microscale variables show promise for improving national empirical models.

Particulate matter air pollution and national and county life expectancy loss in the USA: A spatiotemporal analysis.

BACKGROUND: Exposure to fine particulate matter pollution (PM2.5) is hazardous to health. Our aim was to directly estimate the health and longevity impacts of current PM2.5 concentrations and the benefits of reductions from 1999 to 2015, nationally and at county level, for the entire contemporary population of the contiguous United States. METHODS AND FINDINGS: We used vital registration and population data with information on sex, age, cause of death, and county of residence. We used four Bayesian spatiotemporal models, with different adjustments for other determinants of mortality, to directly estimate mortality and life expectancy loss due to current PM2.5 pollution and the benefits of reductions since 1999, nationally and by county. The covariates included in the adjusted models were per capita income; percentage of population whose family income is below the poverty threshold, who are of Black or African American race, who have graduated from high school, who live in urban areas, and who are unemployed; cumulative smoking; and mean temperature and relative humidity. In the main model, which adjusted for these covariates and for unobserved county characteristics through the use of county-specific random intercepts, PM2.5 pollution in excess of the lowest observed concentration (2.8 µg/m3) was responsible for an estimated 15,612 deaths (95% credible interval 13,248-17,945) in females and 14,757 deaths (12,617-16,919) in males. These deaths would lower national life expectancy by an estimated 0.15 years (0.13-0.17) for women and 0.13 years (0.11-0.15) for men. The life expectancy loss due to PM2.5 was largest around Los Angeles and in some southern states such as Arkansas, Oklahoma, and Alabama. At any PM2.5 concentration, life expectancy loss was, on average, larger in counties with lower income and higher poverty rate than in wealthier counties. Reductions in PM2.5 since 1999 have lowered mortality in all but 14 counties where PM2.5 increased slightly. The main limitation of our study, similar to other observational studies, is that it is not guaranteed for the observed associations to be causal. We did not have annual county-level data on other important determinants of mortality, such as healthcare access and quality and diet, but these factors were adjusted for with use of county-specific random intercepts. CONCLUSIONS: According to our estimates, recent reductions in particulate matter pollution in the USA have resulted in public health benefits. Nonetheless, we estimate that current concentrations are associated with mortality impacts and loss of life expectancy, with larger impacts in counties with lower income and higher poverty rate.

Land Use Regression models for 60 volatile organic compounds: Comparing Google Point of Interest (POI) and city permit data.

Land Use Regression (LUR) models of Volatile Organic Compounds (VOC) normally focus on land use (e.g., industrial area) or transportation facilities (e.g., roadway); here, we incorporate area sources (e.g., gas stations) from city permitting data and Google Point of Interest (POI) data to compare model performance. We used measurements from 50 community-based sampling locations (2013-2015) in Minneapolis, MN, USA to develop LUR models for 60 VOCs. We used three sets of independent variables: (1) base-case models with land use and transportation variables, (2) models that add area source variables from local business permit data, and (3) models that use Google POI data for area sources. The models with Google POI data performed best; for example, the total VOC (TVOC) model has better goodness-of-fit (adj-R2: 0.56; Root Mean Square Error [RMSE]: 0.32 µg/m3) as compared to the permit data model (0.42; 0.37) and the base-case model (0.26; 0.41). Area source variables were selected in over two thirds of models among the 60 VOCs at small-scale buffer sizes (e.g., 25 m-500 m). Our work suggests that VOC LUR models can be developed using community-based sampling and that models improve by including area sources as measured by business permit and Google POI data.

National PM2.5 and NO2 exposure models for China based on land use regression, satellite measurements, and universal kriging.

Outdoor air pollution is a major killer worldwide and the fourth largest contributor to the burden of disease in China. China is the most populous country in the world and also has the largest number of air pollution deaths per year, yet the spatial resolution of existing national air pollution estimates for China is generally relatively low. We address this knowledge gap by developing and evaluating national empirical models for China incorporating land-use regression (LUR), satellite measurements, and universal kriging (UK). Land use, traffic and meteorological variables were included for model building. We tested the resulting models in several ways, including (1) comparing models developed using forward variable selection vs. partial least squares (PLS) variable reduction, (2) comparing models developed with and without satellite measurements, and with and without UK, and (3) 10-fold cross-validation (CV), Leave-One-Province-Out CV (LOPO-CV), and Leave-One-City-Out CV (LOCO-CV). Satellite data and kriging are complementary in making predictions more accurate: kriging improved the models in well-sampled areas; satellite data substantially improved performance at locations far away from monitors. Variable-selection models performed similarly to PLS models in 10-fold CV, but better in LOPO-CV. Our best models employed forward variable selection and UK, with 10-fold CV R2 of 0.89 (for both 2014 and 2015) for PM2.5 and of 0.73 (year-2014) and 0.78 (year-2015) for NO2. Population-weighted concentrations during 2014-2015 decreased for PM2.5 (58.7 µg/m3 to 52.3 µg/m3) and NO2 (29.6 µg/m3 to 26.8 µg/m3). We produced the first high resolution national LUR models for annual-average concentrations in China. Models were applied on 1 km grid to support future research. In 2015, >80% of the Chinese population lived in areas that exceeded the Chinese national PM2.5 standard, 35 µg/m3. Results here will be publicly available and may be useful for epidemiology, risk assessment, and environmental justice research.

Satellite-Based Land-Use Regression for Continental-Scale Long-Term Ambient PM2.5 Exposure Assessment in Australia.

Australia has relatively diverse sources and low concentrations of ambient fine particulate matter (<2.5 µm, PM2.5). Few comparable regions are available to evaluate the utility of continental-scale land-use regression (LUR) models including global geophysical estimates of PM2.5, derived by relating satellite-observed aerosol optical depth to ground-level PM2.5 ("SAT-PM2.5"). We aimed to determine the validity of such satellite-based LUR models for PM2.5 in Australia. We used global SAT-PM2.5 estimates (∼10 km grid) and local land-use predictors to develop four LUR models for year-2015 (two satellite-based, two nonsatellite-based). We evaluated model performance at 51 independent monitoring sites not used for model development. An LUR model that included the SAT-PM2.5 predictor variable (and six others) explained the most spatial variability in PM2.5 (adjusted R2 = 0.63, RMSE (µg/m3 [%]): 0.96 [14%]). Performance decreased modestly when evaluated (evaluation R2 = 0.52, RMSE: 1.15 [16%]). The evaluation R2 of the SAT-PM2.5 estimate alone was 0.26 (RMSE: 3.97 [56%]). SAT-PM2.5 estimates improved LUR model performance, while local land-use predictors increased the utility of global SAT-PM2.5 estimates, including enhanced characterization of within-city gradients. Our findings support the validity of continental-scale satellite-based LUR modeling for PM2.5 exposure assessment in Australia.

Long-term nitrogen dioxide exposure assessment using back-extrapolation of satellite-based land-use regression models for Australia.

Assessing historical exposure to air pollution in epidemiological studies is often problematic because of limited spatial and temporal measurement coverage. Several methods for modelling historical exposures have been described, including land-use regression (LUR). Satellite-based LUR is a recent technique that seeks to improve predictive ability and spatial coverage of traditional LUR models by using satellite observations of pollutants as inputs to LUR. Few studies have explored its validity for assessing historical exposures, reflecting the absence of historical observations from popular satellite platforms like Aura (launched mid-2004). We investigated whether contemporary satellite-based LUR models for Australia, developed longitudinally for 2006-2011, could capture nitrogen dioxide (NO2) concentrations during 1990-2005 at 89 sites around the country. We assessed three methods to back-extrapolate year-2006 NO2 predictions: (1) 'do nothing' (i.e., use the year-2006 estimates directly, for prior years); (2) change the independent variable 'year' in our LUR models to match the years of interest (i.e., assume a linear trend prior to year-2006, following national average patterns in 2006-2011), and; (3) adjust year-2006 predictions using selected historical measurements. We evaluated prediction error and bias, and the correlation and absolute agreement of measurements and predictions using R2 and mean-square error R2 (MSE-R2), respectively. We found that changing the year variable led to best performance; predictions captured between 41% (1991; MSE-R2 = 31%) and 80% (2003; MSE-R2 = 78%) of spatial variability in NO2 in a given year, and 76% (MSE-R2 = 72%) averaged over 1990-2005. We conclude that simple methods for back-extrapolating prior to year-2006 yield valid historical NO2 estimates for Australia during 1990-2005. These results suggest that for the time scales considered here, satellite-based LUR has a potential role to play in long-term exposure assessment, even in the absence of historical predictor data.

Does Urban Form Affect Urban NO2? Satellite-Based Evidence for More than 1200 Cities.

Modifying urban form may be a strategy to mitigate urban air pollution. For example, evidence suggests that urban form can affect motor vehicle usage, a major contributor to urban air pollution. We use satellite-based measurements of urban form and nitrogen dioxide (NO2) to explore relationships between urban form and air pollution for a global data  set of 1274 cities. Three of the urban form metrics studied (contiguity, circularity, and vegetation) have a statistically significant relationship with urban NO2; their combined effect could be substantial. As illustration, if findings presented here are causal, that would suggest that if Christchurch, New Zealand (a city at the 75th percentile for all three urban-form metrics, and with a network of buses, trams, and bicycle facilities) was transformed to match the urban form of Indio - Cathedral City, California, United States (a city at the 25th percentile for those same metrics, and exhibiting sprawl-like suburban development), our models suggest that Christchurch's NO2 concentrations would be ∼60% higher than its current level. We also find that the combined effect of urban form on NO2 is larger for small cities (ß × IQR = -0.46 for cities < ∼300 000 people, versus -0.22 for all cities), an important finding given that cities less than 500 000 people contain a majority of the urban population and are where much of the future urban growth is expected to occur. This work highlights the need for future study of how changes in urban form and related land use and transportation policies impact urban air pollution, especially for small cities.

Independent Validation of National Satellite-Based Land-Use Regression Models for Nitrogen Dioxide Using Passive Samplers.

Including satellite observations of nitrogen dioxide (NO2) in land-use regression (LUR) models can improve their predictive ability, but requires rigorous evaluation. We used 123 passive NO2 samplers sited to capture within-city and near-road variability in two Australian cities (Sydney and Perth) to assess the validity of annual mean NO2 estimates from existing national satellite-based LUR models (developed with 68 regulatory monitors). The samplers spanned roadside, urban near traffic (≤100 m to a major road), and urban background (>100 m to a major road) locations. We evaluated model performance using R2 (predicted NO2 regressed on independent measurements of NO2), mean-square-error R2 (MSE-R2), RMSE, and bias. Our models captured up to 69% of spatial variability in NO2 at urban near-traffic and urban background locations, and up to 58% of variability at all validation sites, including roadside locations. The absolute agreement of measurements and predictions (measured by MSE-R2) was similar to their correlation (measured by R2). Few previous studies have performed independent evaluations of national satellite-based LUR models, and there is little information on the performance of models developed with a small number of NO2 monitors. We have demonstrated that such models are a valid approach for estimating NO2 exposures in Australian cities.

Development of West-European PM2.5 and NO2 land use regression models incorporating satellite-derived and chemical transport modelling data.

Satellite-derived (SAT) and chemical transport model (CTM) estimates of PM2.5 and NO2 are increasingly used in combination with Land Use Regression (LUR) models. We aimed to compare the contribution of SAT and CTM data to the performance of LUR PM2.5 and NO2 models for Europe. Four sets of models, all including local traffic and land use variables, were compared (LUR without SAT or CTM, with SAT only, with CTM only, and with both SAT and CTM). LUR models were developed using two monitoring data sets: PM2.5 and NO2 ground level measurements from the European Study of Cohorts for Air Pollution Effects (ESCAPE) and from the European AIRBASE network. LUR PM2.5 models including SAT and SAT+CTM explained ~60% of spatial variation in measured PM2.5 concentrations, substantially more than the LUR model without SAT and CTM (adjR2: 0.33-0.38). For NO2 CTM improved prediction modestly (adjR2: 0.58) compared to models without SAT and CTM (adjR2: 0.47-0.51). Both monitoring networks are capable of producing models explaining the spatial variance over a large study area. SAT and CTM estimates of PM2.5 and NO2 significantly improved the performance of high spatial resolution LUR models at the European scale for use in large epidemiological studies.

Satellite-Based NO2 and Model Validation in a National Prediction Model Based on Universal Kriging and Land-Use Regression.

Epidemiological studies increasingly rely on exposure prediction models. Predictive performance of satellite data has not been evaluated in a combined land-use regression/spatial smoothing context. We performed regionalized national land-use regression with and without universal kriging on annual average NO2 measurements (1990-2012, contiguous U.S. EPA sites). Regression covariates were dimension-reduced components of 418 geographic variables including distance to roadway. We estimated model performance with two cross-validation approaches: using randomly selected groups and, in order to assess predictions to unmonitored areas, spatially clustered cross-validation groups. Ground-level NO2 was estimated from satellite-derived NO2 and was assessed as an additional regression covariate. Kriging models performed consistently better than nonkriging models. Among kriging models, conventional cross-validated R(2) (R(2)cv) averaged over all years was 0.85 for the satellite data models and 0.84 for the models without satellite data. Average spatially clustered R(2)cv was 0.74 for the satellite data models and 0.64 for the models without satellite data. The addition of either kriging or satellite data to a well-specified NO2 land-use regression model each improves prediction. Adding the satellite variable to a kriging model only marginally improves predictions in well-sampled areas (conventional cross-validation) but substantially improves predictions for points far from monitoring locations (clustered cross-validation).

National Spatiotemporal Exposure Surface for NO2: Monthly Scaling of a Satellite-Derived Land-Use Regression, 2000-2010.

Land-use regression (LUR) is widely used for estimating within-urban variability in air pollution. While LUR has recently been extended to national and continental scales, these models are typically for long-term averages. Here we present NO2 surfaces for the continental United States with excellent spatial resolution (∼100 m) and monthly average concentrations for one decade. We investigate multiple potential data sources (e.g., satellite column and surface estimates, high- and standard-resolution satellite data, and a mechanistic model [WRF-Chem]), approaches to model building (e.g., one model for the whole country versus having separate models for urban and rural areas, monthly LURs versus temporal scaling of a spatial LUR), and spatial interpolation methods for temporal scaling factors (e.g., kriging versus inverse distance weighted). Our core approach uses NO2 measurements from U.S. EPA monitors (2000-2010) to build a spatial LUR and to calculate spatially varying temporal scaling factors. The model captures 82% of the spatial and 76% of the temporal variability (population-weighted average) of monthly mean NO2 concentrations from U.S. EPA monitors with low average bias (21%) and error (2.4 ppb). Model performance in absolute terms is similar near versus far from monitors, and in urban, suburban, and rural locations (mean absolute error 2-3 ppb); since low-density locations generally experience lower concentrations, model performance in relative terms is better near monitors than far from monitors (mean bias 3% versus 40%) and is better for urban and suburban locations (1-6%) than for rural locations (78%, reflecting the relatively clean conditions in many rural areas). During 2000-2010, population-weighted mean NO2 exposure decreased 42% (1.0 ppb [∼5.2%] per year), from 23.2 ppb (year 2000) to 13.5 ppb (year 2010). We apply our approach to all U.S. Census blocks in the contiguous United States to provide 132 months of publicly available, high-resolution NO2 concentration estimates.

A national satellite-based land-use regression model for air pollution exposure assessment in Australia.

Land-use regression (LUR) is a technique that can improve the accuracy of air pollution exposure assessment in epidemiological studies. Most LUR models are developed for single cities, which places limitations on their applicability to other locations. We sought to develop a model to predict nitrogen dioxide (NO2) concentrations with national coverage of Australia by using satellite observations of tropospheric NO2 columns combined with other predictor variables. We used a generalised estimating equation (GEE) model to predict annual and monthly average ambient NO2 concentrations measured by a national monitoring network from 2006 through 2011. The best annual model explained 81% of spatial variation in NO2 (absolute RMS error=1.4 ppb), while the best monthly model explained 76% (absolute RMS error=1.9 ppb). We applied our models to predict NO2 concentrations at the ~350,000 census mesh blocks across the country (a mesh block is the smallest spatial unit in the Australian census). National population-weighted average concentrations ranged from 7.3 ppb (2006) to 6.3 ppb (2011). We found that a simple approach using tropospheric NO2 column data yielded models with slightly better predictive ability than those produced using a more involved approach that required simulation of surface-to-column ratios. The models were capable of capturing within-urban variability in NO2, and offer the ability to estimate ambient NO2 concentrations at monthly and annual time scales across Australia from 2006-2011. We are making our model predictions freely available for research.

Western European land use regression incorporating satellite- and ground-based measurements of NO2 and PM10.

Land use regression (LUR) models typically investigate within-urban variability in air pollution. Recent improvements in data quality and availability, including satellite-derived pollutant measurements, support fine-scale LUR modeling for larger areas. Here, we describe NO2 and PM10 LUR models for Western Europe (years: 2005-2007) based on >1500 EuroAirnet monitoring sites covering background, industrial, and traffic environments. Predictor variables include land use characteristics, population density, and length of major and minor roads in zones from 0.1 km to 10 km, altitude, and distance to sea. We explore models with and without satellite-based NO2 and PM2.5 as predictor variables, and we compare two available land cover data sets (global; European). Model performance (adjusted R(2)) is 0.48-0.58 for NO2 and 0.22-0.50 for PM10. Inclusion of satellite data improved model performance (adjusted R(2)) by, on average, 0.05 for NO2 and 0.11 for PM10. Models were applied on a 100 m grid across Western Europe; to support future research, these data sets are publicly available.

Effects of income and urban form on urban NO2: global evidence from satellites.

Urban air pollution is among the top 15 causes of death and disease worldwide, and a problem of growing importance with a majority of the global population living in cities. A important question for sustainable development is to what extent urban design can improve or degrade the environment and public health. We investigate relationships between satellite-derived estimates of nitrogen dioxide concentration (NO(2), a key component of urban air pollution) and urban form for 83 cities globally. We find a parsimonious yet powerful relationship (model R(2) = 0.63), using as predictors population, income, urban contiguity, and meteorology. Cities with highly contiguous built-up areas have, on average, lower urban NO(2) concentrations (a one standard deviation increase in contiguity is associated with a 24% decrease in average NO(2) concentration). More-populous cities tend to have worse air quality, but the increase in NO(2) associated with a population increase of 10% may be offset by a moderate increase (4%) in urban contiguity. Urban circularity ("compactness") is not a statistically significant predictor of NO(2) concentration. Although many factors contribute to urban air pollution, our findings suggest that antileapfrogging policies may improve air quality. We find that urban NO(2) levels vary nonlinearly with income (Gross Domestic Product), following an "environmental Kuznets curve"; we estimate that if high-income countries followed urban pollution-per-income trends observed for low-income countries, NO(2) concentrations in high-income cities would be ∼10× larger than observed levels.

National satellite-based land-use regression: NO2 in the United States.

Land-use regression models (LUR) estimate outdoor air pollution at high spatial resolution. Previous LURs have generally focused on individual cities. Here, we present an LUR for year-2006 ground-level NO(2) concentrations throughout the contiguous United States. Our approach employs ground- and satellite-based NO(2) measurements, and geographic characteristics such as population density, land-use (based on satellite data), and distance to major and minor roads. The results provide reliable estimates of ambient NO(2) air pollution as measured by the U.S. EPA (R(2) = 0.78; bias = 22%) at a spatial resolution (∼ 30 m) that is capable of capturing within-urban and near-roadway gradients in NO(2). We explore several aspects of temporal (time-of-day; day-of-week; season) and spatial (urban versus rural; U.S. region) variability in the model. Results are robust to spatial autocorrelation, to selection of an alternative input data set, and to minor perturbations in input data (using 90% of the data to predict the remaining 10%). The modeled population-weighted (unweighted) mean outdoor concentration in the United States is 10.7 (4.8) ppb. Our approach could be implemented in other areas of the world given sufficient road network and pollutant monitoring data. To facilitate future use and evaluation of the results, concentration estimates for the ∼ 8 million U.S. Census blocks in the contiguous United States are publicly available via the Supporting Information.