Global fine-scale changes in ambient NO2 during COVID-19 lockdowns

Nitrogen dioxide (NO2) is an important contributor to air pollution and can adversely affect human health1–9. A decrease in NO2 concentrations has been reported as a result of lockdown measures to reduce the spread of COVID-1910–20. Questions remain, however, regarding the relationship of satellite-derived atmospheric column NO2 data with health-relevant ambient ground-level concentrations, and the representativeness of limited ground-based monitoring data for global assessment. Here we derive spatially resolved, global ground-level NO2 concentrations from NO2 column densities observed by the TROPOMI satellite instrument at sufficiently fine resolution (approximately one kilometre) to allow assessment of individual cities during COVID-19 lockdowns in 2020 compared to 2019. We apply these estimates to quantify NO2 changes in more than 200 cities, including 65 cities without available ground monitoring, largely in lower-income regions. Mean country-level population-weighted NO2 concentrations are 29% ± 3% lower in countries with strict lockdown conditions than in those without. Relative to long-term trends, NO2 decreases during COVID-19 lockdowns exceed recent Ozone Monitoring Instrument (OMI)-derived year-to-year decreases from emission controls, comparable to 15 ± 4 years of reductions globally. Our case studies indicate that the sensitivity of NO2 to lockdowns varies by country and emissions sector, demonstrating the critical need for spatially resolved observational information provided by these satellite-derived surface concentration estimates.

Here we derive spatially resolved, global ground-level NO 2 concentrations from NO 2 column densities observed by the TROPOMI satellite instrument at sufficiently fine resolution (approximately one kilometre) to allow assessment of individual cities during COVID-19 lockdowns in 2020 compared to 2019. We apply these estimates to quantify NO 2 changes in more than 200 cities, including 65 cities without available ground monitoring, largely in lower-income regions. Mean country-level population-weighted NO 2 concentrations are 29% ± 3% lower in countries with strict lockdown conditions than in those without. Relative to long-term trends, NO 2 decreases during COVID-19 lockdowns exceed recent Ozone Monitoring Instrument (OMI)-derived year-to-year decreases from emission controls, comparable to 15 ± 4 years of reductions globally. Our case studies indicate that the sensitivity of NO 2 to lockdowns varies by country and emissions sector, demonstrating the critical need for spatially resolved observational information provided by these satellite-derived surface concentration estimates.
Nitrogen dioxide (NO 2 ) is an important contributor to air pollution as a primary pollutant and as a precursor to ozone and fine particulate matter production. Human exposure to elevated NO 2 concentrations is associated with a range of adverse outcomes such as respiratory infections [2][3][4] , increases in asthma incidence 5,6 , lung cancer 7 and overall mortality 8,9 . NO 2 observations indicate air quality relationships with combustion sources of pollution such as transportation 6,21 . Initial investigations found substantial decreases in the atmospheric NO 2 column from satellite observations [10][11][12][13][14][15][16] and in ambient NO 2 concentrations from ground-based monitoring [17][18][19][20] during lockdowns enacted to reduce the spread of COVID-19. However, questions remain about the relationship of atmospheric columns with health-and policy-relevant ambient ground-level concentrations, and about the representativeness of sparse ground-based monitoring for broad assessment. Thus, there is need to relate satellite observations of NO 2 columns to ground-level concentrations. It is also important to consider the effect of meteorology on recent NO 2 changes 22 and to quantify NO 2 changes due to COVID-19 interventions in the context of longer-term trends 23 . Furthermore, air quality monitoring sites tend to be preferentially located in higher-income regions, raising questions about how NO 2 changed in lower-income regions where larger numbers of potentially susceptible people reside. Estimates of changes in ground-level NO 2 concentrations derived from satellite remote sensing would fill gaps between ground-based monitors, offer valuable information in regions with sparse monitoring, and more clearly connect satellite observations with ground-level ambient air quality.
Previous satellite-derived estimates of ground-level NO 2 used information on the vertical distribution of NO 2 from a chemical transport model to relate satellite NO 2 column densities to ground-level concentrations [24][25][26] . Recent work improved upon this technique by allowing the satellite column densities to constrain the vertical profile shape, allowing for more accurate representation of sub-model-grid variability, reducing the sensitivity to model resolution and simulated profile shape errors, and improving agreement between the satellite-derived ground-level concentrations and in situ monitoring data 27 . Applying this technique to examine changes in NO 2 during lockdowns bridges the gap between previous studies focusing on either ground monitors or satellite column densities, thus providing a more complete and reliable picture of the changes in exposure.
Since 2005, the gold standard for satellite NO 2 observations has been the Ozone Monitoring Instrument (OMI) on board NASA's Earth Observing System Aura satellite 28,29 . The newest remote sensing spectrometer, the European Space Agency's TROPOspheric Monitoring Instrument (TROPOMI) 30 on the Copernicus Sentinel 5p satellite, has been providing NO 2 observations with finer spatial resolution and higher instrument sensitivity since 2018. These attributes allow the generation of TROPOMI NO 2 maps at 100 times finer resolution (approximately 1 × 1 km 2 ) with a one-month averaging period 31,32 , an improvement over the spatial and temporal averaging needed for accurate OMI maps (typically approximately 10 × 10 km 2 over one year) 24 . Concurrently, the excellent stability of the OMI instrument over the last 15 years provides an ideal dataset for long-term trend analysis 28,33 that offers context for recent TROPOMI data.
Lockdown restrictions act as an experiment about the efficacy of activity reductions on mitigating air pollution. The Oxford COVID-19 Government Response Tracker (OxCGRT, https://www.bsg.ox.ac.uk/ research/research-projects/coronavirus-government-response-tracke r#data) has been monitoring government-imposed restrictions, and studies have indicated that NO 2 decreases were larger for cities in countries with strict lockdowns 34 . However, there is limited information on lockdown stringency on sub-national levels or on how various emission sectors respond to lockdowns. An observation-based metric for lockdown intensity could provide useful information for examining lockdowns on city-level scales or for examining the effects on air quality that are associated with lockdowns in different emission sectors.
Here we leverage the high spatial resolution of TROPOMI to infer global ground-level NO 2 estimates at, to our knowledge, an unprecedented spatial resolution sufficient to assess individual cities worldwide, and to examine changes in ground-level NO 2 occurring during COVID-19 lockdowns from January-June 2020. Case studies presented here demonstrate how the satellite-based estimates provide The colour intensity represents the statistical significance of the trend. c-e, Population-weighted mean NO 2 from ground monitors and from satellite-derived NO 2 sampled at groundmonitor locations in China (c), Europe (d) and North America (e), normalized by the mean concentration during the period where ground-monitor data are available. The black (ground-derived) and red (satellite-derived) values give the trends for the period where ground-monitor data are available. Only monitors with data available over the entire time period are included. Error bars represent population-weighted standard deviations. f, Population-weighted mean satellite-inferred ground-level NO 2 concentrations in South America, Africa and the Middle East, and Oceania. Trends during the given time periods are given at top. Time periods were chosen to reflect the most recent years where a consistent trend is observed. Error bars represent uncertainties in populationweighted means using a bootstrapping method.

Article
information on important spatial variability in lockdown-driven NO 2 changes, and in the NO 2 response to lockdowns in various emissions sectors. We also use TROPOMI to provide fine-scale structure to the long-term record of OMI observations (2005-2019), which provides an opportunity to examine trends in ground-level NO 2 over the last 15 years to provide context for the recent changes.

Global NO 2 concentrations and trends
Global annual mean TROPOMI-derived ground-level NO 2 concentrations for 2019 provide an initial baseline (Fig. 1). The excellent resolution (~1 × 1 km 2 ) of ground-level NO 2 concentrations reveal pronounced heterogeneity . NO 2 enhancements are apparent over urban and industrial regions. TROPOMI-derived ground-level concentrations exhibit consistency with in situ observations (r = 0.71, N = 3,977, in situ versus satellite slope = 0.97 ± 0.02), as shown in Supplementary Fig. 8. Neglecting the spatial and temporal variability in the NO 2 column-to-surface relationship degrades the consistency with ground monitors (slope = 0.78 ± 0.01), demonstrating the importance of relating satellite columns to surface concentrations for exposure assessment.
Examination of long-term changes in air pollution offers context for changes during COVID-19 lockdowns (Fig 1, Supplementary Figs. 1-7). Satellite-derived NO 2 concentrations decreased from 2005-2019 in urban areas across most of the USA and Europe, eastern China, Japan, and near Johannesburg, South Africa, largely reflecting emission controls on vehicles and power generation. NO 2 increases are observed in Mexico, the Alberta oil sands region in northern Canada, and throughout the Balkan peninsula, central and northern China, India and the Middle East, broadly consistent with reported trends in ground-monitor data [35][36][37]  that agree well with trends in the ground-monitor data (within 0.7% yr −1 in North America, 0.3% yr −1 in Europe, and 1.2% yr −1 in China). Figure 2 shows the April 2020 to April 2019 difference between mean ground-level NO 2 concentrations derived from TROPOMI observations. NO 2 concentrations are lower in most regions in 2020 than in 2019, particularly over urban areas, with global population-weighted mean concentrations decreasing by 16% in 2020 relative to 2019. Fig. 3 shows regional maps focusing on the month with the largest change in population-weighted regional mean concentration for each region, with an additional period included for China, as lockdown restrictions occurred earlier than in other countries. Regional population-weighted mean concentrations decreased by 17-43%. The largest decreases occur in China in February with concentration decreases exceeding 10 parts per billion by volume (ppbv) and substantial decreases persisting in eastern urban areas through April. Thus these lockdown measures temporarily bolstered the decreasing trends across North America 42 and Europe 25 over the last two decades and in China since 2012 43 , owing to technological advances in vehicles and power generation, while temporarily buffering changes from increasing energy demands in India and the Middle East 40,44,45 . NO 2 increases in April 2020 in central China (Chengdu and Chongqing) because lockdowns began lifting during this time.  Figure 3 shows maps of long-term NO 2 trends for context. In most regions, the observed changes during COVID-19 restrictions exceed the expected year-to-year differences observed in the long-term trends (Table 1). 2020-2019 population-weighted mean concentration changes are lower than long-term trends by factors of 17 ± 7 in North America, 19 ± 2 Europe, of 2.9 ± 0.6 in Africa and the Middle East, of 3.6 ± 0.6 in Asia, 8 ± 7 in South America, and 2 ± 2 in Oceania.

Article
Meteorological differences are calculated with the GEOS-Chem chemical transport model using emission inventories that do not include changes that occurred owing to COVID-19 lockdown policies but do reflect meteorological changes. Supplementary Fig. 10 shows TROPOMI-derived changes at 2.0° × 2.5° resolution for comparisons with simulated values at the same resolution. Population-weighted NO 2 concentration changes due to meteorology in Asia, Europe, South America, Africa and the Middle East are a factor of 2-6 smaller than observed; thus, meteorology alone cannot explain the observed decreases. Concentration increases in the central USA, as noted in other studies 10 , do not appear to be meteorologically driven and may be due to changes in biogenic NO x sources.  Countries with largest populations and annual mean population-weighted NO 2 concentrations greater than 1 ppbv are shown for months with the greatest 2020-2019 difference and strict lockdown conditions (stringency index >20), sorted by population. Regional and global data also shown.

City-and country-level NO 2 changes
The fine resolution of our satellite-derived ground-level NO 2 dataset enables the assessment of larger changes in NO 2 concentrations from 2020-2019 evident at the city level. We calculate changes in TROPOMI-observed monthly mean ground-level NO 2 from 2020-2019 over 215 major cities (the ten most populous cities in each country with a population greater than 1 million) for the month with the greatest monthly mean lockdown stringency index, compared with expected changes due to meteorology and long-term trends (Supplementary Table 1 Consistent analysis of individual cities as enabled by this dataset reveals a mean observed decrease of 32 ± 2% for these 215 cities. The mean expected meteorologically driven change was −1 ± 1% and the mean expected change owing to long-term trends was a decrease of 1.4 ± 0.4%. Supplementary Fig. 12 shows these reductions to be consistent with those found in 381 ground-monitor values from 79 studies 34 (32 ± 2%). Of the 215 cities included here, 65 are in countries that did not have ground-monitoring data available for previous studies. Notably, the 65 cities without monitors are largely in lower-income countries of Africa and southeast Asia. The average gross national income per capita for unmonitored countries is US$7,100, compared to US$25,000 for monitored countries, illustrating the potential of satellite-derived ground-level concentrations for providing information about lower-income regions. In summary, the observed decreases in NO 2 across more than 200 cities worldwide were generally driven by COVID-19 lockdowns, with locally varying modulation by meteorology and business-as-usual changes. Table 1 shows monthly mean country-level population-weighted NO 2 concentrations, changes during COVID-19 lockdown restrictions, meteorological effects and long-term trends for the month with the greatest 2020-2019 change. Meteorological effects were generally minor at the national and regional scale. Multi-year trends provide context for the scale of the changes observed during COVID-19 lockdowns. The decrease in March NO 2 concentrations in the USA from 2019 to 2020 was equivalent to four years of long-term NO 2 reductions. Similarly, changes in NO 2 during COVID-19 lockdowns were equivalent to greater than three years of reductions in China, and up to 23 years in Germany. Globally, the April 2020 population-weighted NO 2 concentration was 0.53 ± 0.06 ppbv lower than in April 2019, equivalent to 15 ± 4 years of global NO 2 reductions.

NO 2 as a lockdown indicator
The relationship between this satellite-derived ground-level NO 2 dataset and lockdown stringency provides supporting evidence for the impact of travel restrictions (Supplementary Fig. 13). The ratio of population-weighted mean observed NO 2 in 2020 to 2019 was calculated for each country and each month from January to June. The 2020/2019 NO 2 ratio in countries with the strictest lockdown (monthly minimum stringency indices greater than the 75th percentile) was 29 ± 3% lower than for countries with the weakest lockdowns (monthly median stringency indices less than the 25th percentile). Maximum and median ratios were also lower for countries with strict lockdowns. Both distributions have similar variability (standard deviations 0.02 and 0.03) which demonstrates similar interannual variability due to meteorology for both sets. When focusing on only the month with the strictest lockdown for each country, changes in population-weighted NO 2 are correlated with lockdown intensity, with changes in countries with strict lockdowns (average decrease 43% if lockdown index >80) more than three times as large as in those with weaker lockdowns (12% if lockdown index <40).
This relationship suggests that changes in satellite-derived NO 2 concentrations offer observational information on the spatial distribution of lockdown effects that is not available through policy-based stringency indices. For example, although the policy-based stringency index in most cases provides a single value for a country, city-level NO 2 concentration decreases in India are in the range 30-84%, reflecting variability in local mobility restrictions, emissions sources, and their sensitivity to lockdowns. Supplementary Fig. 14 explores the sensitivity of NO 2 concentrations to emissions from the transportation and electricity sectors in India, China and the USA by examining the distribution of changes in NO 2 concentration at the 20 largest population centres and 20 largest fossil fuel-burning power plants in each country. All countries have substantial NO 2 decreases in cities, but the sensitivities vary in areas associated with the electricity sector, with decreasing concentrations near power plants in India (mean change −35 ± 4%) and China (−28 ± 8%) but insignificant changes in the USA (−4 ± 8%). Observed NO 2 changes at these power plants exceed expected changes from meteorology alone (−8 ± 2%, −1 ± 4% and −1 ± 3% in India, China and the USA, respectively). Although variability between power plants reflects a mix of regionally varying factors, including meteorology, electricity demand, fuel type and plant-specific emission controls, as well as changes in nearby emissions from other sectors including transportation, these differences indicate a sensitivity of local air quality to activity restrictions affecting the energy sector.
Examining geographic differences in satellite-derived NO 2 concentrations within metropolitan regions is also informative. For example, variability between emission sources is apparent around the city of Atlanta, Georgia, USA ( Supplementary Fig. 15). The population-weighted NO 2 concentration in Atlanta and the surrounding region dropped by 28% from April 2019 to 2020, but with substantial spatial variability in the observed change. The greatest NO 2 decreases are found near a large coal-powered electricity plant to the southeast of the city, with significant changes near another plant to the northwest. Decreases were also larger near the Hartsfield-Jackson International Airport-reflecting the dramatic slowdown in air travel-and over suburban regions to the west and northeast of the city centre, than in the downtown core. Supplementary Fig. 15 also demonstrates the range of NO 2 changes experienced by the local population. Over 1.2 million people live in regions where NO 2 decreases exceeded 40%, whereas nearly 1 million people experienced decreases of 10% or less. Similar heterogeneity in population exposure exists in other major cities, as demonstrated by Supplementary Fig. 16. For example, a subset of over 1 million people in the Paris metropolitan area experienced NO 2 decreases of 75% (4.5 ppbv) or more (10th-percentile exposure), whereas another similar-sized subset experienced changes of 23% (0.6 ppbv) or less (90th-percentile exposure). Of the cities examined here, 68 had an interquartile range in population exposure change during lockdowns of 20 percentage points or larger, 22 of which were unmonitored cities. Studies have found that NO 2 changes during lockdowns varied among socioeconomic, ethnic and racial groups in US cities 46 , and thus the variability in other major cities observed here suggest similar disparities may occur elsewhere. The heterogeneity of NO 2 changes demonstrates the need for the finely resolved information on lockdown effects that is offered by satellite observations. We find that using this satellite-derived NO 2 dataset as an observational proxy for lockdown conditions is also useful for identifying links between lockdown-driven emission changes and secondary pollutants. For example, several studies have found little to no change in fine particulate matter (PM 2.5 ) during lockdowns as meteorology, long-range Article transport and nonlinear chemistry complicate the relationship between PM 2.5 and NO x emissions 47,48 . A challenge in these studies has been limited observational information on the local lockdown intensity. Recent work examining 2020-2019 changes in satellite-derived PM 2.5 concentrations found that lockdown-driven decreases in PM 2.5 concentration can be identified by separating the meteorological effects from emissions effects using chemical transport modelling and focusing on regions with the greatest sensitivity to emission reductions 49 . Here we examine that same satellite-derived PM 2.5 dataset using TROPOMI-derived ground-level NO 2 concentrations to identify the regions where PM 2.5 concentrations are most likely associated with lockdowns or sensitive to NO x emissions. Supplementary Fig. 17 shows the distribution of changes in monthly mean PM 2.5 concentrations from 2020-2019 for China in February and North America and Europe in April. Regions with the largest 2020-2019 NO 2 concentration decreases (90th percentile) are considered to be those with significant NO x emission reductions. Population-weighted mean PM 2.5 concentrations decreased overall; however, regions with the largest NO 2 decreases experienced greater local changes in PM 2.5 concentration in China and to a lesser extent in North America, indicating that the sensitivity of PM 2.5 to changing NO x emissions can be inferred. The year-to-year variability of PM 2.5 concentrations in Europe is similar regardless of changes in NO 2 , indicating a greater role of meteorology or transport on PM 2.5 in this region and period. These results are consistent with previous findings when using chemical transport modelling to identify regions where local emissions are important 49 . Thus, the observational proxy on lockdown conditions offered by these satellite-derived surface NO 2 concentrations offers spatially resolved information to identify where PM 2.5 and NO 2 (and by proxy, NO x emissions) are most strongly coupled.

Implications
The pronounced decreases in ground-level NO 2 found here for over 200 cities worldwide during COVID-19 lockdowns are a culmination of recent advancements in techniques for estimating ground-level NO 2 from satellite observations 27 alongside higher-resolution satellite observations from TROPOMI that allow for estimating high spatial resolution, short-term changes in NO 2 exposure. This method bridges the gap between monitor data (that measure ground-level air quality but have poor spatial representativeness) and satellite column data (that provide spatial distributions but are less representative of ground-level air quality). The ability to infer global ground-level NO 2 concentrations with sufficient resolution to assess individual cities and even within-city gradients is an important development in satellite remote-sensing instrumentation and algorithms. Additionally, these satellite-derived ground-level NO 2 concentrations offer information about unmonitored communities and populations that are underrepresented in studies focused on ground-monitor data. These cities are found to have different characteristics of NO 2 concentrations and changes during lockdowns that motivate the need for satellite observations in the absence of local ground monitoring. The changes in ground-level NO 2 due to COVID-19 lockdown restrictions, which exceed recent long-term trends and expected meteorologically driven changes, demonstrate the impact that policies that limit emissions can have on NO 2 exposure. This information has relevance to health impact assessment; for example, studies focused on ground-monitor data have indicated improvements in health outcomes related to improved air quality during lockdowns, including an estimated 780,000 fewer deaths and 1.6 million fewer paediatric asthma cases worldwide due to decreased NO 2 exposure 20 . Our study demonstrates considerable spatial variability in lockdown-driven ground level NO 2 changes that does not necessarily correlate with population density, demonstrating probable uncertainties arising from extrapolating changes observed by ground monitors to estimate broad changes in population NO 2 exposure. Satellite-based ground-level NO 2 estimates provide high-resolution information on the spatial distribution of NO 2 changes in 2020 that cannot be achieved through ground monitoring, particularly in regions without adequate ground monitoring, and should improve exposure estimates in future health studies. Additionally, ground-level concentrations from downscaled OMI observations provide the opportunity to contrast effects of past mitigation efforts on long-term NO 2 trends against the short-term changes resulting from more dramatic regulations, and a chance to improve studies of health outcomes related to long-term NO 2 exposure.
The strength of the links between observed changes in NO 2 concentration and lockdown stringency indicates that satellite-based ground-level NO 2 concentrations offer useful observational, spatially resolved information about lockdown conditions. This provides an observational metric for examining the efficacy of lockdown restrictions on restricting mobility for studies examining the spread of COVID-19. Here we exploited this information to illustrate the differing sensitivity of NO 2 concentrations to changes in various emission sources to lockdown restrictions. Future applications of these data could include examining socioeconomic drivers that impact this variability within and between countries. Comparisons between satellite-derived ground-level NO 2 and PM 2.5 also indicate the utility of these data as an observational proxy for identifying regions where secondary pollutants such as PM 2.5 or ozone are more likely to be sensitive to NO x emissions; these links are otherwise difficult to trace without the use of chemical transport models 50 .
These data offer information to improve NO 2 -exposure estimates, to examine exposure trends, and subsequently estimate changes in health burden. These developments provide an excellent opportunity for advances in air quality health assessment in relation to NO 2 and its combustion-related air pollutant mixture.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-021-04229-0.