New Insights for Tracking Global and Local Trends in Exposure to Air Pollutants.

Over six million people die prematurely each year from exposure to air pollution. Current air quality metrics insufficiently monitor exposure to air pollutants. This gap hinders the ability of decisionmakers to address the public health impacts of air pollution. To spur new emissions control policies and ensure implemented solutions realize meaningful gains in environmental health, we develop a framework of public-health-focused air quality indicators that quantifies over 200 countries' trends in exposure to particulate matter, ozone, nitrogen oxides, sulfur dioxide, carbon monoxide, and volatile organic compounds. We couple population density to ground-level pollutant concentrations to derive population-weighted exposure metrics that quantify the pollutant levels experienced by the average resident in each country. Our analyses demonstrate that most residents in 171 countries experience pollutant levels exceeding international health guidelines. In addition, we find a negative correlation between temporal trends in ozone and nitrogen oxide concentrations, which─when qualitatively interpreted with a simple atmospheric chemistry box model─can help describe the apparent tradeoff between the mitigation of these two pollutants on local scales. These novel indicators and their applications enable regulators to identify their most critical pollutant exposure trends and allow countries to track the performance of their emission control policies over time.

Introduction

Poor ambient air quality remains one of the most critical public health issues in countries around the globe. Exposure to air pollution causes over 6.6 million premature deaths worldwide annually, constituting nearly 8% of the global burden of disease.[1−3] Nearly 92% of the global population lives in areas with unsafe concentrations of ambient air pollutants.[4] Air pollution is one of the greatest environmental health threats and is the fourth leading cause of early death behind high blood pressure, tobacco use, and malnutrition.[2,5]

The health effects of air pollutant exposure reach far beyond respiratory illness. Poor air quality is linked to cardiovascular disease,[6−8] diabetes,[9−11] cancer,[12−14] and other noncommunicable diseases.[15] Recent research has also suggested a link between chronic exposure to air pollutants and coronavirus disease 2019 (COVID-19) health outcomes.[16,17] While acute air pollution events (e.g., smog) have demonstrable impacts on human health,[18−22] chronic exposure to air pollution is the predominant contributor to morbidity and mortality.[23,24]

Although air quality is a global sustainability issue, its impacts and severity are highly variable both between and within countries.[25,26] Pollutant levels continue to diverge between industrialized and rapidly industrializing countries. Nearly 90% of deaths attributable to air pollution occur in low- and middle-income countries.[4] Southeast Asian countries are especially impacted, accounting for approximately 60% of these deaths. In contrast, many North American and European countries have implemented successful air pollution control policies.[27−29] East Asia and South Asia have seen a 21 to 85% increase in particulate matter (PM)-related mortality over recent decades, whereas Europe and North America have, respectively, seen a 58 and 67% decrease.[30] Despite progress, poor air quality remains a problem within industrialized countries. For instance, nearly 100 million people in the United States live in areas exceeding health-based air quality standards.[31] Inequities in air pollution at the subnational level result in disparate health impacts on different communities.[32−35]

As policymakers implement and refine air pollution policies, quantitative indicators that highlight worrying trends and track the efficacy of regulations are important tools to identify the most critical air quality issues. An indicator can stand alone—tracking a single issue—or be aggregated into composite indices that provide a holistic evaluation of air quality.[36] Several air quality indicators have helped link scientific data to environmental policymaking.[37,38] These efforts normally quantify ambient concentrations of air pollutants, emissions, or disease burden.[39−42] For instance, the State of Global Air Report[3] provides a yearly update on the health impacts of exposure to air pollution. The air quality-life index[43] similarly tracks life years lost from PM exposure above thresholds established by the World Health Organization and other environmental agencies.

Ideal indicators measure outcomes of interest to stakeholders.[44] For example, to quantify the health burden of air pollution, measures of health outcomes attributable to air pollutants (e.g., mortality rates and disability-adjusted life years) are preferable to ambient concentrations of air pollutants. However, constraints on data and scientific evidence preclude researchers from quantifying the health burden of many pollutants, which results in composite air quality indices that often compare disparate units of exposure.[3,38] In other instances, indicators have been based on data sets with limited spatial and temporal scope.[45,46] These limitations have made it difficult to identify leaders and laggards in air quality and prevent regulators from measuring the efficacy of implemented solutions over time. Indicators imperfectly represent complex systems and provide summaries of information for which the underlying data would be too cumbersome and complex to process in a reasonable timeframe. As indicators vary in their design and intended use, they can be misinterpreted, and different indicators can result in different conclusions.[47] For example, a study of environmental health indicators found that the ranking of 16 Latin American cities was highly dependent on which indicator was used, with a city ranked best under one indicator ranked worst under another.[48]

Decisionmakers therefore need better indicators to more consistently identify leaders and laggards in air quality and pinpoint policies that successfully reduce the public health burdens of air pollution. Here, we describe a new framework of air quality indicators that better enable policymakers to compare the efficacy of emissions control policies between countries and across multiple pollutants. Our analyses report population-weighted exposure derived from population density and ground-level concentrations of respirable particulate matter (PM2.5), ozone (O3); nitrogen oxides (NOx); sulfur dioxide (SO2); carbon monoxide (CO); and volatile organic compounds (VOCs) in over 200 countries around the world. These six pollutants cause much of air pollution’s ecological and health threats and are often the focus of emissions control policies. For instance, many of these pollutants exacerbate cardiovascular or respiratory diseases like asthma;[49,50] others can be carcinogenic.[51] We utilize pollutant data sets that are robust, global in coverage, and will be updated frequently through the foreseeable future.[52] This creates a reliable framework for quantifying trends in air quality over time and identifying the most critical pollutants that each country should strive to address.

Our approach expands the potential for air quality indicators to inform effective policies and mitigate the public health impacts of air pollutants. Although our indicators focus on the health impacts of air pollution, we note that poor air quality has other detrimental impacts on agriculture and ecosystem vitality by reducing crop growth and causing eutrophication.[53,54] We design our new indicators to track exposure trends in the same units of measurement across different pollutants. This better enables researchers and regulators to explore relationships among pollutants and facilitates a fuller understanding of how emission interventions targeting one pollutant can affect the exposure levels of another. By weighting pollutant concentration by population density, our metrics further facilitate the comparisons between different countries or subpopulations. Finally, we demonstrate how to couple our indicators with a simple atmospheric chemistry model in order to explore potential tradeoffs in air quality policies and design more effective regulations. While this box model is best-suited to qualitatively exploring feedbacks in a generic atmospheric chemistry regime, it may aid policymakers in their understanding of trends in air pollutant concentrations. Our indicators are therefore useful to policymakers striving to explore drivers of air pollution, anticipate trends in air quality, and quantify the efficacy of policies to mitigate the public health impacts of exposure to toxic air pollutants.

Methods and Calculations

Pollutant Concentration Data

We derive pollutant concentration data from the European Centre for Medium-Range Weather Forecast’s Atmospheric Composition Reanalysis 4 (EAC4) data sets (Figure a). These data are freely available from the Copernicus Atmospheric Monitoring Services’ Atmospheric Data Store (ads.atmosphere.copernicus.eu) and through a supported web-API (Figure a) and described previously.[52] The reanalysis data are derived from emission inventories and model output nudged to remotely-sensed observations. We note these data may have higher uncertainty in regions with less extensive monitoring networks or inventories. The chemical mechanisms used in the model are an extended version of the Carbon Bond 2005 (CB05) mechanism implemented in the CTM Transport model 5 (TM5).[55,56]

Figure 1

Maps indicate the (a) ground-level NOx concentration, (b) population density, and (c) exposure derived from both concentration and population density data. A country’s overall pollutant exposure value is the concentration that the average resident within a country experiences.

Ambient, ground-level (model-level 60) concentrations for PM2.5, O3, NOx, SO2, CO, and VOCs are freely available from the Atmospheric Data Store as monthly averaged globally gridded raster files at a spatial resolution of 0.7° × 0.7° (approximately 80 km × 80 km). Users can also download linearly interpolated data at a finer grid resolution using the Copernicus Atmosphere Monitoring Service (CAMS)-supported web-API. In these analyses, we use data downloaded at 0.125° by 0.125° (approximately 12 km × 12 km). This may result in less accurate concentration values for small countries like island nations or city-states whose geographic extent is less than 80 km. We note that the CAMS reanalysis ozone data can exhibit a high bias at high latitude grid cells, particularly the Arctic and Antarctic.[57,58] We estimate VOCs as the summed concentration of ethane, propane, formaldehyde, and isoprene. Our VOC indicator is a proxy for exposure to both industrial and natural organic gases. Monthly averages are derived from daily averages, which in turn are calculated by averaging pollutant concentration values at every 3 h model timestep between 0:00 and 21:00 UTC. This method of deriving monthly averages does not capture 8 h daily maximum concentrations that some air quality regulations target but rather the average concentration over the entire diurnal cycle. These monthly values were averaged to create annual values for each pollutant. Given the strong correlation between ozone concentration, temperature, and solar intensity, many air quality and health studies focusing on the ozone track maximum daily or seasonal pollutant concentrations rather than annual averages.[59−62] Our indicators can be modified to use a subset of the annual data, allowing researchers and policymakers to explore seasonality or diurnal trends. EAC4 data sets extend from 2003 to near-real time. However, as recent data do not yet reflect the effects of the global pandemic caused by COVID-19, we have truncated the data at December of 2018.

Population Data

Population data are derived from the Gridded Population of the World (GPWv4.11)[63] data set (Figure b). In constructing GPWv4.11, population data from the 2010 Population and Housing Censuses, including data between 2005 and 2014, are extrapolated to produce population estimates for five-year intervals between 2000 and 2020.[64] In our analyses, we use an optional extension that adjusts historic and future population data to reflect predictions from the United Nations World Population Prospects Report[65] and as a result have greater consistency across countries for global analyses. Global rasters were downloaded in 2.5 arc-minute spatial resolution (approximately 4.5 km). We linearly interpolate populations to derive annual data between the given five-year intervals.

Population-Weighted Exposure

We couple ambient ground-level pollutant concentration data to population data in order to derive population-weighted exposure (Figures c and S1). Population weighting ensures that the country average concentrations are reflective of population distribution in relation to air quality, reducing the bias inherent in spatial averages that include extensive rural areas with typically lower air pollution levels. We regridded pollutant data by linearly interpolating ground-level concentrations to the same spatial resolution as the population density data. We then derive the fraction of the country’s total population living within each grid cell. Each country’s population-weighted exposure (E) is defined as the sum over the country’s grid cells (i) of the pollutant concentration (c) times the fraction of the total country population living within the grid cell (P) (eq )

The population-weighted exposure (E) indicates the pollutant concentration that an average person within a country is exposed to. The fractional population value, P, can be considered as a normalized population density, which measures the number of people per unit of square area. The population-weighted exposure, E, thereby captures heterogeneous patterns in population that are indicated by population density data. These estimated pollution levels better reflect population-level exposure than do a simple spatial averaging as higher weight is given to areas with higher population density. We use administrative unit boundaries specified by the 2015 Global Administrative Unit Layers (GAUL) Level 0 shapefiles, available from the United Nations Food and Agricultural Administration (http://www.fao.org).

Air Quality Model

We develop and apply a simple atmospheric chemistry box model to interpret correlations between trends in O3, NOx, and VOCs. Our model is based on one derived by Seinfeld and Pandis, 1998.[66] The model’s simplified approach promotes clarity and accessibility at the expense of analytical rigor. Policymakers can easily adapt this model to quickly, but qualitatively, explore tradeoffs between different regulatory regimes, such as the potential increases in O3 exposure, resulting from decreased NOx emissions. More complex and analytically rigorous models, such as the U.S. Environmental Protection Agency’s empirical kinetic modeling approach (EKMA) model,[67] are often customized to the chemical conditions in specific cities or regions.[68−70] We refer the reader to the extensive body of the literature on O3 chemistry modeling for further information.[71,72]Table summarizes our model’s reactions and representative rate constants. Pollutant concentrations indicate the ambient mixing ratios over a 10 h period in a closed system, approximating a stagnant atmosphere over an urban environment. Additional discussion of the model development and its limitations are found in the manuscript’s Supporting Information section.

Table 1

Simplified VOC + NOx + O3 Mechanism

reaction	rate constant (298 K)
1.	1.09 × 10–12a
2.	7.7 × 10–12b
3.	8.1 × 10–12
4.	1.1 × 10–11
5.	2.9 × 10–12
6.	5.2 × 10–12c
7.	0.015d
8.	1.9 × 10–14

Rate constant for propane.[93] Propane is selected as the model VOC due to its relative importance in reactions in the urban atmosphere.[94,95]

Rate constant for CH3O2 + NO.[66]

Rate constant for CH3O2 + HO2.[66]

A typical NO2 photolysis rate constant (45° N Summer, Solar Zenith Angle = 50°). Values vary depending on light intensity.[66] See the Supporting Information section for further discussion.

Statistical Analyses

To consider trends in air quality, we employ the Theil-Sen estimator to fit a model to annual data.[60]

However, simple linear trends computed across all years can miss trend reversals[73] and be biased by outliers, the Theil-Sen regression method more rigorously estimates the magnitude of trends.[74,75] As described below, we calculate the Theil-Sen slope for trends between 2003 and 2018 and rank countries based on that slope for each indicator.

To explore relationships between trends, we also calculate the proportional increase or decrease in population-weighted exposure for each of the six air pollutants, derived by dividing the difference between the average of the most recent five years’ data (2014–2018) and the earliest five years’ data (2003–2007) by the earliest five years’ data. These results indicate which countries improved or worsened their population-weighted mean exposures relative to their baseline average exposure between 2003 and 2007. We then derive the pairwise Pearson correlation coefficient of countries’ proportional change in exposure for different pollutants. We note that unlike the Theil-Sen estimator, the Pearson correlation coefficient is a non-robust statistical calculation.

Results and Discussion

Most countries’ residents are exposed to air pollutant concentrations exceeding public health standards or guidelines. For PM2.5 and NOx, 171 and 27 countries’ population-weighted mean exposures exceeded the World Health Organization’s annual average limits, 10 and 40 μg m–3, respectively, in 2018. The worst-performing (i.e., highest pollution level) countries vary by pollutant. Table lists the five countries with the highest average population-weighted exposure between 2014 and 2018 for each pollutant. As policymakers and public health officials are often interested in maximum O3 concentrations rather than annually averaged values, we also include peak values for O3 exposure, defined as the average of the daily maximum concentration in summer (June or December). Countries’ exposure values are generally higher than their spatially averaged concentration values (Table S1), which indicates worse air quality in more populated areas. China is among the worst five countries for four exposure indicators: PM2.5, NOx, SO2, and CO. This finding is consistent with the broad base of the literature investigating sources and concentrations of air pollutants in China.[76−78] Despite these high exposure rates, however, China exhibits a downward trend in the most recent years for these pollutants (Table ).[79,80] India is listed among the five countries with the worst air quality for PM2.5 and CO exposure as are Nepal, Pakistan, and Bangladesh. The United Arab Emirates (UAE) is listed twice, for O3 and NOx exposure. Israel is also listed twice, for NOx and SO2 exposure. Although our indicator rankings vary between pollutants, these results demonstrate that many countries’ populations are experiencing unsafe levels of multiple pollutants.

Table 2

Five Worst-Performing (Highest Air Pollution) Countries by Population-Weighted Pollutant Exposure, Average of 2014 to 2018

rank	PM2.5 (μg m–3)	O3 (ppb)	peak O3 (ppb)	NOx (ppb)	SO2 (ppb)	CO (ppb)	VOCs (ppb)
218	Pakistan	Qatar	Burundi	Chinaa	Mexicoa	Nepal	Central African Rep.a
	110	80.1	128	71.2	254	900	298
217	India	Bahrain	Central African Rep.	South Koreaa	Kuwaita	India	Dem. Rep. Congoa
	101	79.2	127	70.5	136	759	211
216	Bangladesh	U.A.E.	Eswatini	Kuwait	Rwandaa	Pakistan	Rep. Congoa
	95.9	79.1	119	63.8	123	757	180
215	Nepal	Kuwait	Chad	U.A.E.	Israela	Bangladesh	South Sudana
	92.3	73.4	114	58.0	92.2	736	168
214	Chinaa	Cyprus	Lesotho	Israela	Chinaa	Chinaa	Boliviaa
	92.2	70.5	113	54.5	78.3	693	141

Country’s average exposure has declined or plateaued over the past five years.

Table 3

Five Worst-Trending (Largest Increase in Air Pollution) Countries by Population-Weighted Pollutant Exposure: 2003 to 2018 Thiel-Sen Regression

rank	PM2.5 (μg m–3 yr–1)	O3 (ppb yr–1)	peak O3 (ppb yr–1)	NOx (ppb yr–1)	SO2 (ppb yr–1)	CO (ppb yr–1)	VOCs (ppb yr–1)
218	Bangladesh	Nauru	Nauru	Chinaa	India	South Africa	Rep. Congoa
	+1.49	+0.59	+0.53	+1.83	+0.76	+12.9	+7.30
217	South Africa	Netherlandsa	Netherlandsa	Iran	Pakistan	Pakistan	Dem. Rep. Congoa
	+1.46	+0.51	+0.48	+1.13	+0.73	+10.2	+6.03
216	India	U.K.a	Belgiuma	Israela	Bangladesh	Iran	Honduras
	+1.30	+1.10	+0.39	+0.93	+0.73	+8.62	+2.35
215	Pakistan	Japana	U.K.a	Iraq	Indonesiaa	India	El Salvadora
	+1.20	+0.46	+0.37	+0.92	+0.56	+8.23	+2.15
214	Nepal	South Korea	Denmark	U.A.E.	Iran	Bangladesh	Myanmar
	+1.01	+0.45	+0.36	+0.85	+0.54	+7.66	+1.79

Country’s average exposure has declined or plateaued over the past five years.

The benefit of population-weighted exposure metrics over spatially-averaged concentration metrics is illustrated by data for large countries with highly variable population densities. Tracking air quality by averaging pollutant concentration alone gives equal weight to areas with relatively few inhabitants. For instance, subpopulations in large proportions of the midwestern United States, northern Russia, and western China experience lower pollutant concentrations than the average resident in more populated areas (Figure ). While the spatially-averaged 2018 NOx concentration in these countries are 7.6, 3.0, and 22.2 parts per billion (ppb), the population-weighted NOx concentrations are 17.3, 20.0, and 71.2 ppb, respectively. Population-weighted and spatially-averaged concentrations are correlated across all countries, yet each indicator shows several countries where population-weighting produces markedly different results from spatial averaging (Figure S2). Population-weighted exposure indicators can therefore provide a more relevant proxy for the public health impacts of air quality, with respect to the number of people affected, than would indicators based on spatially-averaged concentrations alone. We note, however, that population-weighting can obscure air quality problems in rural areas by giving more emphasis to pollutant concentrations in urban environments. Our results, when aggregated to the country-level, are therefore limited in their capacity to describe heterogeneities in different subpopulations’ exposures. Policymakers could use lower-level administrative boundaries to perform subnational analyses.

Comparison of spatially-averaged and population-weighted values can provide insight into the patterns of air quality in relation to urbanicity. The results further indicate that densely populated areas do not invariably correspond to areas with high pollutant concentrations. For instance, NOx concentrations can vary by over 2 orders of magnitude between areas of a country with the same population density (Figure S3). Indicators tracking national-level exposure should therefore account for the heterogeneity in exposure patterns. Population weighting is one approach to incorporate such heterogeneity. This finding demonstrates the need for indicators to use data sets that provide complete spatial coverage of countries’ air quality. Indicators based on sparse ground-based observation networks may inadvertently misrepresent the true exposure values in large countries.

Our indicators can also be used to track how countries’ air quality evolves over time, as described by the slope of the Theil-Sen linear model (Tables and 4). For clarity, we also indicate whether a country’s exposure values have declined or plateaued in the past five years. This further indicates whether recent air quality policies have successfully reversed longer-term trends in pollutant exposure. Several countries with the worst absolute exposure values also exhibit the greatest trends over the entire time period, although we emphasize that these values do not reflect recent developments. For instance, China’s data show the largest average annual rate of change in NOx exposure (+1.8 ppb yr–1) between 2001 and 2018 while also exhibiting the highest absolute NOx exposure in recent years (71 ppb). Despite this long-term increase, our indicator suggests that China’s NOx exposure peaked around 2012 and has since plateaued or declined. This indicates that recent efforts in China to curb sources of NOx are yielding meaningful gains in air quality.[81,82] Analysis by Shah et al. (2020)[83] indicates that a steeper decline in NOx concentrations in central China over this period than our country-wide exposure metrics, suggesting that growing populations in more polluted coastal urban environments offset some other regions’ improvements in air quality. India and Bangladesh are also among the five worst countries in terms of absolute and trending PM2.5 exposure. Other countries are not among the worst performers in terms of highest absolute exposure but exhibit steep trends of deteriorating air quality. For instance, South Africa falls into the five worst trending countries for PM2.5 and CO exposure. Japan’s population has also experienced elevated O3 exposures in recent decades, with an average increase of 1.3 ppb yr–1. We note that our CAMS-derived O3 exposure values are in broad agreement with previous metrics, such as those derived by the Tropospheric Ozone Assessment Report (TOAR).[3,84] However, our ozone trends tend to be flatter than these observationally based metrics, indicating that further research is needed to resolve differences between CAMS reanalysis data and observations when constructing air quality indicators. Finally, we reiterate that our O3 metric was derived from annual average exposure. Ozone exhibits a strong seasonal pattern, with ground-level concentrations highest in warmer summer months (Figure S4). Researchers and policymakers can adopt the indicator framework used here to track trends in the maximum concentration levels to better gauge acute exposure events and public health impacts. For instance, the United States National Ambient Air Quality Standard for O3 tracks the highest daily maximum 8 h concentration, averaged over 3 years. Many areas have seen progress in this peak-based metric, indicating an inverse trend from the annual average O3 metric. For comparison, we include trends in peak O3 exposure in Tables and 4. These results indicate that China’s success in reducing annually-averaged O3 exposure are not replicated in terms of peak concentrations. While our peak-based O3 metric demonstrates a modest improvement when averaged over the entire country (−0.06 ppb yr–1), several other studies indicate rising peak ozone concentrations in China’s urban atmospheres.[85,86]

Table 4

Five Best-Trending (Largest Decrease in Air Pollution) Countries by Population-Weighted Pollutant Exposure: 2003 to 2018 Thiel-Sen Regression

rank	PM2.5 (μg m–3 yr–1)	O3 (ppb yr–1)	peak O3 (ppb yr–1)	NOx (ppb yr–1)	SO2 (ppb yr–1)	CO (ppb yr–1)	VOCs (ppb yr–1)
1	Indonesia	China	Israel	Japan	Serbia	Indonesia	Central African Rep.
	–1.18	–0.34	–0.20	–2.02	–3.51	–11.0	–8.87
2	South Korea	Israel	Myanmar	South Korea	Mexico	Italy	Paraguay
	–1.18	–0.29	–0.16	–1.43	–2.91	–7.91	–3.99
3	North Korea	Lebanon	Lebanon	U.S.A.	Bulgaria	South Korea	Bolivia
	–0.95	–0.17	–0.16	–1.34	–2.78	–7.25	–2.23
4	Japan	Myanmar	Cyprus	U.K.	N. Macedonia	North Korea	South Sudan
	–0.95	–0.15	–0.15	–1.21	–2.40	–6.39	–1.75
5	Belgium	Rep. Congo	Yemen	Netherlands	Hungary	Belgium	Chad
	–0.89	–0.14	–0.13	–1.14	–1.58	–6.08	–1.49

Many countries have trended toward cleaner air since 2003 (Table ). Belgium, Indonesia, South Korea, and North Korea place among the top five clean-trending countries for PM2.5 and CO exposure. Our trend analyses also indicate that some countries are making progress toward reducing their most critical air pollutants. For instance, Mexico has the highest absolute SO2 exposure (254 ppb) but also exhibits the second highest negative (meaning a decrease in pollution) exposure trend (−2.91 ppb SO2 yr–1). Although SO2 has natural sources, the majority of emissions come from anthropogenic activity like fossil fuel combustion.[87] Policy efforts near population centers like Mexico City to use low-sulfur fuels for electricity generating units may therefore be yielding improvements to air quality.[88,89] Our indicator captures the downward trend in SO2 exposure, allowing policymakers to quantify the success of strategies to mitigate air pollutant emissions. This exemplifies how decisionmakers can use our indicators to identify their country’s most serious air pollutants and track their implemented policies’ efficacy at reducing exposure levels.

Correlations between Pollutants

Comparing trends in multiple air quality indicators can demonstrate whether country performance across multiple pollutants is linked and reveal insights into the ways in which pollution levels are changing. Several pollutants’ exposure trends are correlated (Figure ). However, the above results (Tables and 4) report temporal trends by absolute levels (i.e., pollution increments). We also examine temporal trend by proportion (i.e., do the countries with the highest initial levels of air pollution have the highest or lowest proportional change?). Most exposure trends are positively correlated, indicating that nations have tended to reduce exposure to multiple pollutants simultaneously over our indicators’ observation period. The strongest correlations exist between CO and NOx (r = 0.75; p-value = 1.1 × 10–5) and CO and PM2.5 (r = 0.75; p-value = 3.0 × 10–10). Although these correlations are calculated from country-wide annual-average changes in pollutant exposure, researchers and policymakers can adapt the indicators to explore relationships at other geographic and temporal scales. Country-wide averages may overlook significant regional or local trends. The O3–NOx anticorrelation is likely to be more pronounced in VOC-limited urban areas or in winter months. We emphasize that these trends should not be taken to imply drivers of air quality. For instance, these trends and correlations may be impacted by transboundary air pollution—that is, pollution originating outside the jurisdiction of a country’s air quality policies. Furthermore, although VOCs and O3 are anticorrelated, there is no atmospheric chemical mechanism that would produce an O3 disbenefit from reduced VOC emissions.

Figure 2

Correlation analyses between population and pollutants’ exposure trends. Oval shape qualitatively indicates the strength of each correlation, with more narrow ovals indicating stronger correlations. Numbers indicate the Pearson correlation coefficients (r) for the proportional change (Δ) in the population-weighted exposure level for a given pollutant between the most recent five years’ data (2014–2018) and the earliest five years’ data (2003–2007) relative to the 2003–2007 baseline. Statistically insignificant correlations (p > 0.05) are indicated in gray.

Population growth was correlated with increased exposure trends in four out of five of our air pollutant exposure indicators. Significant positive correlations (p-value ≤ 0.05) exist between a country’s proportional population change and its proportional change in population-weighted exposure for CO, NOx, PM2.5, SO2, and VOCs, although correlations are only moderate (0.27 ≤ r ≤ 0.51) (Figure ). Conversely, absolute population was only correlated with trends in NOx exposure. This indicates that a country’s total population is not associated with its residents’ exposure levels, other than for NOx. Rather, policy and regulatory choices, pollutant transport, and both anthropogenic and natural pollutant sources factor into a country’s average pollutant exposure. Countries can use our indicators to identify their better-performing peers, weigh policy options, and adopt similar solutions to the best-performing countries to mitigate air pollution in their own jurisdictions. However, we also note that factors beyond policies’ immediate control—such as transboundary pollutant transport and natural pollutant sources like wildfires—can impact air quality.

Interpreting Ozone Trends with an Air Quality Model

Trends in population-weighted O3 exposure are notably anticorrelated with trends in exposure to other pollutants (Figure ). These anticorrelations are strongest between NOx and O3 (r = −0.35; p-value = 3.1 × 10–5) and VOCs and O3 (r = −0.32; p-value = 9.3 × 10–2). Many countries that have successfully reduced exposure to NOx and VOCs have therefore exacerbated exposure to O3 and vice-versa—a counterintuitive finding that is well documented and characterized by previous studies and air quality models.[71] To facilitate a better understanding of this apparent tradeoff, we apply the EAC4 pollutant data to a simplified atmospheric chemistry model that describes the relationship between ambient NOx, VOC, and O3 concentrations. The simulation takes the form of a box model in which initial concentrations of NOx and VOC are set, and pollutant concentrations are allowed to evolve over time without any emissions, deposition, transport, or dilution. Our model’s simplicity makes it easily adaptable and implementable by policymakers seeking to explore the potential tradeoffs of different regulatory options. However, the model does not consider the additional regional effects of natural sources, variation in hydrocarbon regimes, chemical transport, and demographics. We detail several limitations of the model in the Supporting Information section. Our model is best used as a qualitative tool to help researchers and policymakers explore trends in, rather than predict absolute magnitudes of, O3 concentration’s response to changing ambient NOx and VOC concentrations.

We generate an O3 isopleth plot based on 500 box model simulations, with initial VOC concentrations ranging linearly between 0.1 and 200 ppb, and initial NOx concentrations ranging linearly between 0.05 and 150 ppb (Figures a and S5). Ambient O3 concentration is dependent on NOx concentration and the relative proportion of NOx to VOCs. Countries may achieve lower O3 concentrations by increasing ambient NOx concentrations, illustrating the inherent tradeoff between controlling pollutant exposure in the NOx + VOC + O3 system. To explore the potential significance of this phenomenon, we indicate pollutant trends in major urban centers in China, Japan, and the United States (Figure a). Arrows indicate the difference between the 2003–2007 and 2014–2018 ambient NOx and VOC concentrations in the spatial grid cell containing these urban locations. These trends indicate that China’s observed improvement in O3 exposure may in fact be due to worsening NOx concentrations, although our analyses do not account for the meteorological impacts on O3 trends observed in more sophisticated models.[85] Conversely, our model predicts that some major metropolitan areas in Japan and the United States have trended toward higher annual average O3 concentrations as their NOx and VOC concentrations have decreased since 2003. We emphasize that the indicators considered here track annual average concentrations. The way indicators are defined can impact the resulting trends they observe. For instance, O3 metrics tracking the daily maximum 8 h average ozone concentration have observed decreasing O3 trends over the same time period.[59] This dichotomy is reflected in previous statistical analyses of urban O3 concentrations in the United States, which have shown that 95th percentile concentrations have decreased by 0.5–2 ppb yr–1, whereas 5th to 50th percentile concentrations have increased by 0.1–1 ppb yr–1 over the same time period.[90]

Figure 3

(a) Ozone isopleth diagram derived from our simplified NOx + VOC + O3 chemical model (Table ). The isopleth plot was generated from 500 box model runs with initial VOC concentrations ranging between 0.2 and 200 ppb, and initial NOx concentrations ranging between 0.2 and 150 ppb. Values indicate the final concentration of pollutants after integrating reactions over 10 h assuming a closed system. Arrows indicate the difference between the 2003–2007 and 2014–2018 average NOx and non-methane VOC concentrations in grid cells containing select major urban environments. (b) Comparison of CAMS-derived indicator data and modeled change in O3 exposure for the same urban environments.

Despite the simplicity of our chemical model, we replicate the direction and, to an extent, the magnitude of the O3 concentration trend in 12 cities between 2003 and 2018 (Figure b). The model output and indicator results discussed here agree with prior analyses exploring trends in urban O3 concentrations. Simon et al. (2015) observed that O3 concentrations in many U.S. urban environments have increased since 1998 despite falling emissions of NOx and VOCs.[90] Itano et al. (2007) further found that downward trends in NOx emissions were an important contributor to increased O3 concentrations in Osaka, Japan.[91] These counterintuitive relationships partly result from the complex relationships between precursor concentrations and ozone concentrations (Table ). Numerous studies also highlight the importance of meteorological changes, including regional temperature response to climate change, as an important driver of tropospheric O3 concentrations.[85,92] We reiterate that these results are based on trends in annually averaged O3 concentrations. Several cities exhibiting increasing annually averaged O3 concentrations have achieved lower peak O3 concentrations (Table S2).

These results join other studies’ findings, demonstrating that downward trends in O3 exposure may not necessarily indicate successful air quality policies. The uniformity between our indicators allows regulators to explore relationships between different pollutants’ concentrations, facilitating deeper insights into the drivers of poor air quality. Our indicators are therefore useful tools for policymakers seeking to understand the relationship between air pollutants and the drivers of pollutant exposure trends. However, we reiterate that while our indicators and chemistry model may qualitatively help explain O3 trends, air quality regulators should use more comprehensive models tuned to the local chemistry and meteorology of specific urban environments to quantify the potential impact of NOx and VOC control measures on actual ambient O3 concentrations.

Policy Implications

Air quality indicators help researchers and regulators mitigate the public health impacts of air pollutants by quantifying trends and identifying successful emissions control policies that countries can adopt from their top-performing peers. Our metrics improve on previous indicators by analyzing population-weighted exposure rather than spatially-averaged concentration trends. This approach more fairly quantifies the exposure of a country’s residents by more heavily weighting pollutant concentration data in more densely-populated areas. Exposure-based indicators therefore provide more insightful foundations for public health regulations by considering how both demographics and air pollutant concentrations evolve over time.

Despite some progress toward reducing ambient concentrations of, and exposure to, harmful air pollutants in the developed world,[3] our results demonstrate that many countries’ residents continue to breathe unsafe air. Nearly 170 countries’ annual-average exposure values exceed the World Health Organization’s limits for PM2.5, and over 20 countries also exceed NOx limits. Several countries’ annual-average exposure values for O3 and SO2 also exceed hourly or daily averaged World Health Organization limits, highlighting critical air quality issues in these countries.

Our indicators are also useful tools for policymakers seeking to understand the drivers of air quality trends. The simple chemical model we applied to O3, NOx, and VOC data can help illustrate that improvements in NOx exposure may exacerbate exposure to O3. Although this model’s simplicity prevents it from quantifying the precise effects of emission control policies, it can help policymakers qualitatively explore the tradeoffs inherent in regulating O3 and its precursors. Our framework of exposure-based air quality indicators therefore helps regulators pinpoint their country’s most urgent air quality issues, track the performance of implemented regulations, and understand the drivers of air quality trends over time.