The global impacts of COVID-19 lockdowns on urban air pollution: A critical review and recommendations

The global spread of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in early 2020 was an unprecedented, highly disruptive event. Lockdowns instituted to control the subsequent coronavirus disease 2019 (COVID-19) pandemic led to rapid, unforeseen decreases in economic and social activity and associated emissions of air pollutants and greenhouse gases worldwide. It has been suggested in a number of perspective articles and comments that this episode provides a unique scientific opportunity to detect, attribute, and understand the impacts of anthropogenic emissions on the Earth’s atmosphere at all spatial scales, from regional to global (Forster et al., 2020; He et al., 2020; Kroll et al., 2020; Le Quéré et al., 2020; Liu et al., 2020d), and on the Earth System and climate generally (Diffenbaugh et al., 2020; Phillips et al., 2020; Raymond et al., 2020). Of particular interest have been shifts in regional air quality that have been documented by ground-level monitoring networks and spaceborne remote sensing instruments. Such changes, occurring to a varying extent on every continent except Antarctica, have been the subject of intense interest among the general public and within the scientific and regulatory communities charged with understanding the air quality impacts of anthropogenic emissions. These transient shifts within particular emissions sectors have the potential to test the efficacy of air pollution control strategies and may even provide a preview of the future state of the atmosphere in a world with more permanent reductions in emissions from certain sectors.

The concept of air quality acknowledges the health burden attributable to atmospheric pollutants (World Health Organization [WHO], 2019). The WHO assesses that air pollution is the number one environmental health risk globally, causing 7.1 million premature deaths per year, of which 4.2 million are attributable to outdoor air pollutants. The WHO defines guideline values for key air pollutants (see Table 1), yet national regulatory limit values vary widely and are often less stringent than the WHO guideline values. The air quality index (AQI) is a common term used by government agencies to define standards for the simultaneous presence of multiple pollutants. Individual pollutant concentrations are combined to derive the AQI and determine air quality levels. However, no agreed upon definition for AQI exists, with AQI determined in different ways for each country (Bishoi et al., 2009; Fareed et al., 2020).

In addition to emission and deposition processes, both sources and sinks of air quality relevant trace compounds are determined by atmospheric chemistry. Species that are emitted directly to the atmosphere are considered primary, whereas species formed through atmospheric chemical processes are referred to as secondary. The main species of concern for human health are particulate matter (PM) and tropospheric ozone (O₃; Gakidou et al., 2017). PM has both primary and secondary sources, while ozone is formed almost exclusively through atmospheric chemistry, that is, it is secondary in nature. Major pollutants that serve as precursors to O₃ and secondary PM include nitrogen oxides (NO_x = NO + nitrogen dioxide [NO₂]), volatile organic compounds (VOCs), sulfur dioxide (SO₂), carbon monoxide (CO), and ammonia (NH₃; see Figure 1).

Observational and laboratory approaches to understand relevant atmospheric chemical processes are complemented by modeling approaches to determine atmospheric composition on regional and global scales. Atmospheric chemical transport models (CTMs) account for (1) emissions from anthropogenic and natural sources, (2) atmospheric chemistry, and (3) transport, dilution, and deposition processes. The ability of CTMs to correctly simulate atmospheric composition is traditionally verified through comparisons of model and observational outputs. Extreme events, such as volcanic eruptions (Kristiansen et al., 2016; Wilkins et al., 2016; Beckett et al., 2020), wildfires (Liu et al., 2010), and heatwaves (Churkina et al., 2017; Zhao et al., 2019), play a particularly important role in this regard, as such events can expose model biases or missing processes.

The various national, statewide, and municipal lockdowns and implementations of social distancing for pandemic control of COVID-19 offer an “extreme” real-world experiment in which various anthropogenic sector-specific emissions of air pollutants have been suddenly and significantly reduced. This link of changes in human behavior and reduced anthropogenic emissions is expected (Beirle et al., 2003). In particular, during the pandemic, emissions from the transportation sector were reduced as a consequence of stay-at-home orders, as revealed by mobility data sets (Forster et al., 2020; Venter et al., 2020), for example. Early reports of observed decreases in NO_x and PM in various regions of the world are now complemented by data sets showing varied responses in the secondary pollutants O₃ and PM resulting from the nonlinear interactions involved in atmospheric chemistry (Seinfeld, 2006). The COVID-19 lockdowns, therefore, offer a unique opportunity to (1) verify emission inventories and (2) explore the sensitivity of secondary pollutants to emission changes. Several review articles have already been published as of the writing of this article. Shakil et al. (2020) used 23 publications through May 2020 to highlight the effects of lockdowns and environmental factors on air quality and recommended that future analyses include meteorological corrections. Srivastava et al. (2020) focused on the link between PM pollution and the positive correlation to COVID-19 cases as well as the impact of weather on pollutant concentrations that affect morbidity and mortality. Kumar et al. (2020) highlighted the key findings of 28 publications on the effects of lockdowns on pollutant concentrations. Finally, Le et al. (2020b) discussed 16 publications related to PM concentration reductions during the pandemic.

In this review, we summarize the available literature through September 30, 2020, comprising more than 200 publications, and the approaches used to quantify changes in atmospheric pollutant levels. We focus on species that are of relevance as air pollutants and short-lived climate forcers, namely, NO₂, PM_2.5, O₃, NH₃, SO₂, black carbon (BC), VOCs, and CO. We further provide an outlook on the tools and analyses required to expand from individual case studies to a global framework of readily comparable results. To enhance the readability of the text, we present the references in tables, which allows for structured overviews of all references relevant to respective methods, regions, or compounds. With the pandemic, and hence lockdowns, ongoing as of this writing, this review intends to serve as a milestone in identifying and quantifying the overall impacts of emission reductions to air quality.

Analysis of ground- and satellite-based observations of pollutants has received intense scientific focus in 2020. During the 7 months following the onset of the pandemic (March–September 2020), more than 200 manuscripts were accepted for publication in peer-reviewed journals. There are undoubtedly many others that are in preparation and review at the time of this writing or that have been published after October 2020. Subsequent reviews will be required to fully assess the breadth of this literature. The goal of this review is to provide an initial synthesis of this rapidly developing literature, as well as to provide some critical assessment of the state of the initial literature that may be useful for authors of manuscripts that follow.

To generate the database of peer-reviewed scientific articles used in this study, we utilized Google Scholar (Google, 2020) and searched the websites of prominent publishers of environmental scientific journals to find as many relevant and newly accepted papers as possible. We used the following search terms to query subject matter content: “COVID* AND air AND pollution” or “COVID* AND air AND quality.” The wildcard “*” accounted for common iterations such as “COVID-19” and “COVID2019,” while the Boolean operator AND was used to limit the results to studies related to air pollution or air quality topics. The first search was conducted in September 2020 and updated biweekly through October 30, 2020. We further limited the search results to papers that had undergone peer review, were accepted by September 30, 2020, and were published in English.

Each of the 219 papers that met the above criteria was examined by at least one coauthor to determine its overall relevance to the goals of this study. The papers were then added to our database and all pertinent information was manually cataloged. This included the author list, journal name, dates of submission and acceptance, the region and time frames studied, the type of data set used (ground-based, satellite, or both), and whether the authors accounted for the effects of seasonality/meteorology and the year-to-year variability in atmospheric concentrations.

Furthermore, we manually digitized the findings from 150 papers relating to the observed percentage change and/or concentrations of the pollutants discussed in each study (see Table S1 for the nondigitized yet reviewed papers). Figure 2 shows the cumulative number of papers by sample type, methodology, and region of study. This subset of our database comprises manuscripts published in 37 different scientific journals with a median submission to acceptance peer-review period of 35 days. There were 10 different geographic study regions, led by East Asia (N = 54 papers) and South Asia (N = 28 papers). These two regions were dominated by air quality studies in China (N = 46 papers) and India (N = 27 papers), respectively. The first COVID-related air quality manuscript was accepted on March 5, 2020 (Wang et al., 2020c) before the WHO declared COVID-19 a global pandemic on March 11, 2020.

Portions of our analysis rely on the stringency index (SI), a metric used to quantitatively compare lockdown measures for each country over time (Hale et al., 2020). The SI ranges from 0 (no lockdown) to 100 (strictest lockdown) based on a variety of measures meant to slow the spread of COVID-19 (see Section 3.2). We have not made explicit use of other common metrics of economic activity changes found in other papers, such as sector-specific mobility indices provided by Google or Apple (Forster et al., 2020) or traffic counts. The SI is convenient for the purpose of this article since its focus is appropriate for the continental and regional scales considered in the data synthesis presented here.

All data digitized for analysis in this review are available on the website https://covid-aqs.fz-juelich.de. This includes the observed percentage change in species concentration for NO₂, NO_x, CO, PM_2.5, PM₁₀, O₃, SO₂, NH₃, speciated nonmethane volatile organic compound (NMVOCs), aerosol optical depth (AOD), BC, and the AQI. Also, the absolute concentrations of NO₂, PM_2.5, O₃, and CO during the lockdown and reference periods are provided. Each data set is linked with the digital object identifier of the original publication, information on the corresponding author, region, country, city (where applicable), and the observational start and end times. This website is designed as a living version of this review, that is, as new literature emerges, authors of published papers are encouraged to upload their data to the database, thus complementing the data coverage in space, time, and compound dimensions. The data sets from the website are provided with free and unrestricted access for scientific (noncommercial) use including the option to generate targeted reference lists. Users of the database are requested to acknowledge the data source and reference this review in publications utilizing the data set.

Figure 2 shows that ground-based measurements comprise the largest fraction of the data used in the analysis of COVID-19 lockdowns to date. These data normally come from local, regional, or national air quality monitoring networks in various regions, as discussed in Section S1.1. Air quality monitoring networks include the U.S. Environmental Protection Agency (2020), the European Environment Agency EAA together with the European Monitoring and Evaluation Programme (2020), the China National Environmental Monitoring Center established by the Ministry of Ecology and Environment of China (Chu et al., 2021), and the Central Pollution Control Board in India managed by the Ministry of Environment, Forests, and Climate Change (Pant et al., 2020).

All these networks or infrastructures such as the Aerosols, Clouds and Trace gases Research Infrastructure and In-service Aircraft for a Global Observing System (Petzold et al., 2015) provide preliminary data in near real time, with final, quality-assured data updated either quarterly or biannually. Other data sources such as the OPEN-AQ data source (https://openaq.org/) compile network data into readily accessible, larger databases. However, the data quality assurance process is not always made clear in a given publication. For example, the OPEN-AQ platform explicitly makes no guarantee of quality assurance or assessment of accuracy. Data are uploaded in real time and not necessarily updated when quality assured (final) data are made available from a given air quality network. Papers published to date on COVID-19 lockdown effects using ground-based monitors generally specify the source of their data but commonly do not specify whether those data are preliminary or final. Given the speed with which these manuscripts were prepared, it is possible that many are based on data with no final quality control.

Roughly one-third of the publications discussed in this review make use of satellite observations. A large number of satellite data sets have been used, including:

• Sentinel-5P TROPOspheric Monitoring Instrument (TROPOMI) NO₂, CO, SO₂, and HCHO;

• AURA-OMI (Ozone Monitoring Instrument) NO₂, SO₂, and AOD;

• Terra and Aqua MODIS AOD, PM, and fire products; and

• Terra MOPITT CO and Aqua AIRS CO.

By far, the most used data set is the TROPOMI NO₂ tropospheric column product of Sentinel-5P, used in 41% of cases (calculated as the number of papers using TROPOMI NO₂ divided by the total number of satellite data sets used in the papers). The second most used data set is AURA-OMI NO₂, used in 27% of cases, followed by MODIS AOD, used in 14% of cases. All other data sets have been used sporadically (1–3 times). Of the papers reviewed herein that use satellite data, 61% used TROPOMI NO₂, 40% used OMI NO₂, and 21% used MODIS AOD (note that several papers used multiple satellite data sets for their analysis, on average 1.5 satellite data sets per paper). Taking the NO₂ data sets from OMI and TROPOMI together, 68% of the published satellite results on the COVID-19 impact on air quality were generated using these two data sets.

Note that satellite instruments like TROPOMI and OMI measure at one given overpass time (e.g., 13:30 local). As the diurnal profile of the emissions may have changed during the lockdowns, observed changes at a given overpass time may not be fully representative of the total changes. Also, TROPOMI and OMI tropospheric column NO₂ retrieval products contain detailed uncertainty estimates for each observation separately, typically ranging between 20% and 60% for polluted scenes. The use of averaging kernels in the data products is advised to remove the dependency on the retrieval a priori and reduce the associated uncertainties (see Section S1.2).

The atmospheric abundance of trace compounds is determined through the interplay of emissions, atmospheric chemistry, transport, and loss processes. To quantify the effect of changes in any of these, an analysis must isolate the influence of confounding parameters. The main focus of the literature reviewed here is the effect of emission changes on ambient mixing ratios of criteria pollutants. In general, three types of approaches are used: a comparison of observed concentrations to a reference period during which “business as usual” emissions prevailed (see Section 2.3.1), an analysis of observed concentrations when accounting for meteorological influences or atmospheric chemistry (e.g., photolysis frequencies, humidity, and temperature dependencies; Section 2.3.2), and a comparison of observed concentrations with the output of CTMs run to derive “business as usual” expected values (Section 2.3.3).

Nearly two-thirds of the studies summarized here were a direct comparison of lockdown periods to a reference measurement period (Table 2). Two main approaches were used: (1) a comparison of pollutant concentrations directly before and/or after a lockdown, that is, data sets covering a relatively short time period or (2) a comparison of pollutant concentrations from seasonally similar time periods, that is, data sets that included 2019, and often several other previous years, for the same period of time as the 2020 lockdown. The main advantage of these approaches is the simplicity in identifying relative changes. For the first approach, uncertainties arise due to the unquantified effects of seasonality, meteorology, and atmospheric chemistry. Although the second approach generally covers meteorological effects, uncertainty may still arise from other processes that affect the abundance of atmospheric trace compounds, such as climatological variability and exceptional events. It is not possible to conclude generally whether the use of direct comparisons to reference periods bias the derived changes low or high, as this will be determined by the specific conditions prevailing in each studied region. Unambiguous quantification of emission changes is, therefore, not possible, although the correlation of observed changes with indicators of emission activity (e.g., traffic counts, fuel sales, mobility, electric power consumption) can be explored. Various studies included in this work highlight the importance of identifying the effects of meteorology, atmospheric chemistry, and emission trends in the observed percentage emission changes and are discussed in the following section.

Meteorological factors have an important effect on atmospheric pollution levels (Shenfeld, 1970). Wind velocity, stability, and turbulence affect the dilution, transport, and dispersion of pollutants. Sunshine triggers the photochemical production of oxidants that form smog, whereas rainfall has a scavenging effect that washes out particles and some gases from the atmosphere. Furthermore, concentrations of various atmospheric pollutants can change due to decreasing trends of emissions in urban environments around the world (e.g., Warneke et al., 2012; Sun et al., 2018; Zheng et al., 2018). Changing pollutant concentrations can influence atmospheric chemistry by affecting the pollutant’s chemical sources and sinks and therefore its lifetime (e.g., Shah et al., 2020). With atmospheric chemistry and pollutant distribution changing with season and location (e.g., summer vs. winter, urban vs. remote locations), all the above highlight the need to quantify the effects of meteorology, atmospheric chemistry, and emission trends on atmospheric pollutant concentrations when describing pollutant changes during the pandemic.

Several studies quantified the effects of meteorology and emission trends on the observed pollutant changes, as summarized in Table 3. Of the 32 studies listed, 16 studies focused on East Asia, six on Europe, six on North America, three on South Asia, two on South America, and two were global studies. Over 98% of the measurements presented in publications that were included in this review were from urban environments. Different statistical approaches were used to account for the above effects, which are summarized below.

Two approaches were utilized to apportion pollutant concentrations to different sectors and to elucidate the role of atmospheric chemistry and/or meteorology. One commonly used approach was positive matrix factorization (PMF), a widely used receptor model to resolve pollution sources and quantify the source contributions. Studies using PMF focused on the PM_2.5 chemical composition, and the sources of organic particulate pollution, in Beijing (Cui et al., 2020), Wuhan (Zheng et al., 2020), and Tianjin (Dai et al., 2020), China. Conventional PMF analysis may suffer from information loss due to nonlinear dilution variations. Dai et al. (2020) incorporated the ventilation coefficient into their dispersion-normalized PMF, which reduced the dilution effect. The advantages of using PMF were highlighted in all studies, and their findings supported the substantial contribution of secondary sources, as well as the influence of local primary sources, to PM pollution. Finally, a hierarchical cluster analysis and principal component analysis were used in one study in India to investigate the impact of changing temperatures on pollutant concentrations (Bera et al., 2020). However, although these approaches will more reliably quantify observed changes in the atmospheric abundance of pollutants as a response to emission changes, the effects of meteorology and atmospheric chemistry are not always fully disentangled.

Several approaches were used to reduce the effects of meteorology on the interpretation of air quality. One approach is to examine tracer-tracer ratios (Homan et al., 2010; Borbon et al., 2013), for example, normalizing pollutants relative to a relatively long-lived species like CO. These ratios provide a simple way to account for dilution and are typically used to isolate the effects of secondary chemistry. A confounding factor is that many of the commonly used tracers in the denominator (e.g., CO) also changed significantly due to emission reductions related to COVID-19. Other studies performed dilution corrections by normalizing to meteorological variables such as planetary boundary layer height (Su et al., 2020) or satellite column data with solar zenith angle, wind speed, and wind direction (Goldberg et al., 2020). Another approach is to benchmark periods of similar meteorology in past years with meteorology experienced during lockdown periods (Tanzer-Gruener et al., 2020). Methods to deseasonalize lockdown periods with prelockdown periods or past years were also employed (Adams, 2020; Diamond and Wood, 2020; Metya et al., 2020). Finally, other approaches identified metrics to assess synoptic meteorological conditions conducive to air pollution episodes (Wang and Zhang, 2020) or performed back trajectory analysis, such as with the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model, to assess the origin of pollutants and long-range transport (Chang et al., 2020; Cui et al., 2020; Siciliano et al., 2020b; Zhao et al., 2020a; Shen et al., 2021).

A variety of more complex statistical approaches were also used to quantify the effects of meteorology, atmospheric chemistry, and emission trends. This included the following:

1. multivariate regression analysis methods, where two main data sets were used: the dependent/outcome variables describing the pollutant concentrations and the independent/exposure variables that adjusted for weather conditions (Bao and Zhang, 2020; Cameletti, 2020; Connerton et al., 2020; Jia et al., 2020a; Lei et al., 2020; Venter et al., 2020; Xiang et al., 2020);

2. machine-learning methods, where algorithms were trained on measurements of pollutants and meteorological parameters from previous years to predict the “business as usual” emission estimates for 2020 (Petetin et al., 2020; Wang et al., 2020e; Wyche et al., 2020; Zheng et al., 2020);

3. difference-in-difference methods, where the impact of lockdown measures on air quality were quantified through a fixed-effects ordinary least squares (OLS) approach with the key explanatory variable being the lockdown measures and weather variables used as vectors (Navinya et al., 2020; Liu et al., 2021a); and

4. generalized additive models that accounted for the additive effect of meteorology on the pollutant concentrations and their nonlinear relationships using the meteorological parameters as a model predictor input to derive the pollutant concentration (Ordóñez et al., 2020; Ropkins and Tate, 2020).

The majority of these studies included data sets from multiple years, thereby accounting not only for meteorological effects but also emission trends. Although each of these statistical tools has uncertainties associated with the representativeness of the input data sets, they constitute the best up-to-date published methods to quantify the effects of meteorology, and/or atmospheric chemistry, and/or long-range transport on pollutant concentrations.

Chemical transport modeling provides a means for disentangling the effects of changes in emissions, chemistry, and meteorology on observed changes in air quality due to changing emissions. Table 4 provides a summary of air quality or climate modeling studies published in the literature assessing the impacts of COVID-19. Of the 16 modeling studies listed, 14 are regional modeling studies: 12 of East Asia, one of Europe, and one of Europe and East Asia and two are global climate modeling studies. These modeling studies focused on lockdown measures in China and Europe, and the time period of study is limited to the winter of 2020. Studies of North America, South America, and Africa are notably missing, although it is anticipated that modeling studies will be published in the future for these regions.

Most of the modeling studies listed in Table 4 used a traditional forward Eulerian CTM, such as the Weather Research Forecasting and Community Multiscale Air Quality (Wong et al., 2012), Weather Research Forecasting with Chemistry (Grell et al., 2005), or Goddard Earth Observing System with Chemistry (Henze et al., 2007) models. All of the studies simulate business-as-usual conditions with a baseline emissions inventory. A variety of baseline inventories were used (Table 4), the most common being the global Emissions Database for Global Atmospheric Research (EDGAR) inventory (Crippa et al., 2020) and the Multi-resolution Emission Inventory for China (He, 2012). Many of the modeling studies adjusted their input emission inventories by scaling all emission sectors relative to changes in ambient or satellite observations (Griffith et al., 2020; Le et al., 2020a; Qiu et al., 2020; Zhang et al., 2020c; Zhang et al., 2021). Others have taken a sector-by-sector approach to scaling emission inventories (Forster et al., 2020; Huang et al., 2020; Le Quéré et al., 2020; Li et al., 2020a; Liu et al., 2020b; Menut et al., 2020; Wang et al., 2020d; Yang et al., 2020). For example, Forster et al. (2020) scaled mobile source emissions based on mobility and traffic count data, Huang et al. (2020) scaled industrial emissions based on economic and industrial activity data, and Le Quéré et al. (2020) scaled power generation emissions using energy statistics. Forster et al. have made publicly available a lockdown-adjusted daily inventory based on EDGAR emissions of CO₂, CH₄, N₂O, SO₂, BC, OC, CO, NMVOC, NH₃, and NO_x for each country throughout the COVID-19 lockdown period.

In general, by modeling both baseline and COVID-19-perturbed emissions scenarios, the effects of meteorology can be isolated from those related to changes in emissions and can then be used to quantitatively assess the impacts of emission changes on the formation of secondary pollutants, such as O₃ and PM_2.5. Other studies have modeled constant or prepandemic emissions during the lockdown period to quantify the expected changes in atmospheric concentrations due to meteorology alone and thereby deduce the fraction of observed changes in air quality that are due to emission changes. Additionally, TROPOMI NO₂ vertical column densities were used to derive top-down scaling factors of NO_x emission inventories (Zhang et al., 2020c; Zhang et al., 2021), which were then used to assess impacts on O₃ and PM_2.5 formation (Zhang et al., 2021). Ding et al. (2020) use an inverse modeling algorithm to derive top-down NO_x emissions in China. Finally, climate models were used in some studies to assess COVID-19 perturbations in bottom-up emission inventories and their impacts on global radiative forcing (Forster et al., 2020; Yang et al., 2020). Although the number of modeling studies comprises <10% of the total number of studies analyzed here (Figure 2), they provide an explicit means by which to control the effects of meteorology on observed changes in primary and secondary pollutants.

Worldwide lockdown measures strongly impacted the transportation sector (Forster et al., 2020; Le Quéré et al., 2020). To assess the impact that the transportation sector typically has on pollutant emissions for each country, a “business as usual” emission scenario was investigated using the 2015 EDGAR v5 (Crippa et al., 2020), which is the most recent year for which data are publicly available. Figure 3 shows a world map colored by the largest source of NO_x emissions for each country as well as characteristic examples of the contribution of different sectors to the NO_x, CO, and PM_2.5 emissions for various countries around the world. Emission sectors are separated into transportation, energy and manufacturing, industrial and other processes, building and miscellaneous, and agriculture by lumping IPCC emission categories (see Table S2). Following the IPCC guidelines, energy and manufacturing (IPCC 1.A.1, 1.A.2, 1.B.1, 1.B.2) is classified as fuel combustion activities associated with energy production and industry. All other industrial emissions are included under industrial and other processes. In the following, the contribution of the different sectors to NO_x, PM_2.5, and CO emissions is discussed and detailed differences for countries around the world are provided in Section S2. Pie charts in Figure 3 are used to highlight differences in the contribution of the various pollutant sectors for countries representative of different regions of the world, with an emphasis on the countries listed in Tables 5–10. The category “building emissions” includes residential, commercial, and institutional combustion as well as other combustion sources, whereas “miscellaneous emissions” apply to all remaining emissions from fuel combustion that are not specified elsewhere. Note that agricultural and land-use change emissions, for example, of NO_x from soil are not included in the EDGAR emission inventory, which likely results in an underestimation of the agricultural NO_x emissions. Global annual NO_x emissions based on the 2015 EDGAR inventory were 40 Tg of nitrogen with agriculture accounting for less than 1%. However, global soil NO_x emissions are estimated to be around 5 Tg/year (Yan et al., 2005).

The global EDGAR inventory provides context for expected changes in air pollutant species due to the COVID-19 pandemic, especially those related to the transportation, energy, manufacturing, and industrial sectors. Globally, the median transportation contribution was 36% (15%–51%), 8% (3%–19%), and 30% (5%–70%) for the NO_x, primary PM_2.5, and CO emissions, respectively.

Countries were further divided into developed (Annex I) and developing (Annex II) categories based on the United Nations (2020) Climate Change framework to examine the contribution differences of the various emission sectors. The median transportation contribution for Annex I countries was 44% (36%–56%), 14% (8%–19%), and 25% (17%–44%) for the NO_x, primary PM_2.5, and CO emissions, respectively, whereas for Annex II countries, it was 29% (5%–49%), 9% (2%–42%), and 35% (1%–75%). Although the contribution of transportation emissions varied for the above pollutants, it is evident that transportation reduction measures during lockdowns are expected to consistently have a greater impact on CO and NO_x emissions than for primary PM_2.5 worldwide.

Next, we discuss the impact of policy actions designed to mitigate the COVID-19 pandemic and relate the stringency of lockdown measures with observed changes in the atmosphere. The onset and temporal evolution of SARS-CoV-2 infection rates have varied globally, as have the respective lockdown periods, resulting in emission reductions that are distributed over time and space. Although the onset of lockdown measures is well-defined on national or state levels, the transition to the “new normal” after the initial containment of the disease still implies ongoing changes to anthropogenic emission sectors such as transportation. Emissions thus dropped rapidly at the beginning of the lockdown, but the increases after the initial lockdown are often much slower, and emissions may not return to their prepandemic levels, for example, due to changes in corporate policies for telecommuting, reduced business travel, and so on.

To better compare observations from different regions worldwide and at different times and stages of the pandemic, the government SI (Cameron-Blake et al., 2020; Petherick et al., 2020) is used. This index varies from 0 to 100 and takes into account available information on ordinal indicators of government responses to limit the spread of COVID-19. The index is available on national scales at a 1-day time resolution. The index provides a comparative measure only and is not designed to evaluate the effectiveness of a country’s response. Categories that are included in the index are (1) the implementation and extent of school closures, (2) implementation and extent of workplace closures, (3) restrictions on public events, (4) restrictions on gatherings, (5) closure of public transport, (6) method of public information campaigns, for example, public officials urging caution or coordinated campaigns across traditional and social media, (7) extent of measures to enforce the lockdown, (8) restrictions on internal movement, (9) restrictions on international travel, (10) COVID-19 testing policy, and (11) contact tracing. As such, the index includes both measures that impact emissions and measures with no obvious consequence for emissions. Here, we test the use of the government SI as an indicator for atmospheric composition change, with the data set (Cameron-Blake et al., 2020) as downloaded from the SI web page (Stringency Index, 2020). Figure 4 shows the country-based SI averaged over April as a representative month for the most stringent conditions globally. China is an exception where lockdown measures were implemented in February–March and relaxed in April. Also shown is the April difference in NO₂ column concentrations based on TROPOMI measurements for 2020 compared to 2019. The high spatial resolution of TROPOMI is highlighted for the United States, Canada, Europe, India, and East Asia. Since China was the first country to undergo a lockdown at a time that coincided with the celebration of the Chinese New Year, the 3 post-Chinese New Year weeks in 2020 are compared to 2019 for the high spatial resolution map of China. Results in Figure 4 are generated based on analysis performed as part of this work. The TROPOMI comparisons are only used qualitatively to show the effects of lockdowns in urban and industrialized environments around the world, and detailed analysis of emission reduction comparisons to stringency indices are the focus of future studies. The Copernicus Atmosphere Monitoring Service (CAMS) reanalysis results (Inness et al., 2019) were used to correct for variability of meteorology between the months of April 2019 and April 2020. The emissions used in the CAMS reanalysis were based on “business as usual” scenarios, unaffected by COVID-19 reductions. The CAMS 3-D NO₂ fields were interpolated to the location and time of all the individual TROPOMI observations used to construct the monthly mean. The averaging kernels were applied to obtain CAMS simulations of the TROPOMI observations. These data were averaged over the month of April, and the ratio 2020/2019 was applied to the TROPOMI monthly mean to correct for the expected meteorological impact on NO₂ between the 2 years.

Overall, densely populated regions around the world experienced NO₂ reductions, suggesting that the lockdowns and their consequent reduction in transportation and industrial activities influenced global NO₂ emissions. Specifically, various megacities shown in Figure 4 had detectable NO₂ reductions including New Delhi, India; Beijing, China; New York City and Los Angeles, United States; Paris, France; and Sao Paulo, Brazil. A notable example highlighting the effect of lockdowns on emission reductions is India, the country with the most severe restrictions during April (SI = 98.6), which experienced NO₂ column concentration reductions for urban, industrial, and even remote regions across the country. Less densely populated regions around the world had no change, or sporadic increases, in NO₂ column densities (up to 10¹⁵ molecules cm^–2). Although measurements at remote sites are reported in this review, they represent the minority of the collected literature values (fewer than 5% of the reviewed data sets) because most studies focused on measurements observed predominantly in urban environments where emissions reductions were more evident.

Figure 5 shows the relative changes in pollutant concentrations during the lockdown compared to reference periods for different continents and regions of the world. Pollutants include NO₂, NO_x, and CO, which have the largest expected contribution from transportation (see Section 3.1); PM_2.5 and O₃, secondary pollutants and the two most important pollutants for health impacts (Anderson et al., 2004); SO₂, NH₃, and NMVOCs, which are mostly related to primary gas-phase emissions; and PM₁₀, AOD, BC, and the AQI. For each region, ground-based measurements, satellite measurements, or modeling studies were performed for multiple countries, and often multiple cities within each country, using the different approaches discussed in the Methods section to determine the lockdown effects on pollutant concentrations. All results from these studies are combined in Figure 5 to determine the broader impacts of lockdown measures and establish the variability of changes in atmospheric composition. Numbers in parentheses show the number of publications and the number of data sets considered to produce the respective distributions. A higher spatial resolution analysis is presented in the following sections. An overview of the literature associated with the respective compounds and regions is provided in Tables 5–10, and relative changes on a national level are further discussed in the Supplement (Section S3). All data are downloadable from the database (see Section 2.1).

NO₂ decreased for all continents and regions during lockdowns. The median reductions ranged from 20% to 54% (see Figure 5), except for Africa, where a 70% reduction was found based on two studies in Morocco (see Table 5, Section S3.1). The median reduction in NO_x ranged from 26% to 67% (see Table 10 and Section S3.2). Note that the set of studies reporting NO_x is considerably smaller than the literature on NO₂. The median reduction in CO ranged from 16% to 49%. Within one region, India had the largest variability of reported CO changes, ranging from decreases of 80% to increases of 60% (see Table 8 and Section S3.3). Median reductions in PM_2.5 and PM₁₀ for all continents and regions ranged from 10% to 40% and 8% to 40%, respectively (see Tables 6 and 9 and Sections S3.4 and S3.5). PM_2.5 measurements were widely used, whereas PM₁₀ measurements were limited to fewer studies. The median change in O₃ ranged from a decrease of 15% to an increase of 18% (see Table 7 and Section S3.6). O₃ was the only pollutant that increased on a regional scale during the lockdowns, with a positive median change of 6.4% (±11%). The response of O₃ is complex and varies by season and region, as described further in Section 3.3.5. The median reduction in SO₂ for all continents and regions ranged from 5% to 49% (see Table 10 and Section S3.7). For other pollutants, including AOD, NMVOCs, NH₃, BC, and the AQI, a much smaller number of publications for only a few regions were reported (see Table 10 and Section S3.8).

The literature summarized in this section lacks consistency in the analysis methodology or degree of meteorological normalization, which can confound the attribution of changes in ambient pollutant concentration changes to emissions reductions associated with COVID-19 lockdowns. Here, we compare reported changes, sorted by those that either do or do not correct for meteorology, to the SI in order to assess whether the changes in pollutant concentration correlate with metrics of lockdown intensity across a global scale. Figure 6 shows box-and-whisker plots of NO₂, PM_2.5, and O₃ when combining data from all the countries around the globe and grouping them into different SI bins ranging from 20–40, 40–60, 60–80, and 80–100 (all bin ranges herein are defined as ≥ the lower number and < the higher number). Measurements were further separated into direct comparisons of lockdown to reference periods as discussed in Section 2.3.1 and comparisons that were quantified and corrected for meteorological effects (see Sections 2.3.2 and 2.3.3).

For studies that performed a direct comparison of lockdown to reference periods without a meteorological correction, no significant trend in the median with increasing SI was found for NO₂, PM_2.5, or O₃. Rather, the changes were similar across SI bins, with average pollutant changes (±standard deviation) of –36% (±5%), –20% (±7%), and +6% (±1), respectively (Figure 6). Conversely, the binned SI did correlate with the change in NO₂ and PM_2.5 for studies that accounted for the effects of meteorology. The median change in NO₂ decreased from –13% to −48% and in PM_2.5 decreased from –10% to –33%, whereas the median change in O₃ increased from 0% to 4% with increasing SI. Studies performed in the 40–100 SI range were statistically significant for all pollutants, and for both methods, with 19 or more data points per SI bin. Measurements were sparse for the 20–40 SI bin and statistically significant only for NO₂ (14). The same analysis was done for compounds less studied in the literature: CO, SO₂, and PM₁₀ (Figure S2). The change of these three pollutants did not correlate with the SI when at least three bins were populated, although dependencies on SI may become apparent for CO, SO₂, and PM₁₀ as more studies are published. Although there were only 44, 33, and 33 data sets in total that accounted for meteorology when reporting a change in concentration for CO, SO₂ and PM₁₀, respectively, stronger reductions were evident for all pollutants with increasing SI.

With the emissions of primary pollutants expected to decrease as the lockdown measures become stricter, these results highlight the importance of accounting and quantifying the effects of meteorology in order to quantitatively link changes in atmospheric abundance with changes in emissions. Although a direct comparison of reference to lockdown periods is valuable for identifying air-quality exceedances, its representativeness depends on the similarity of the meteorology during the reference period to the lockdown period. Furthermore, pollutants such as CO and NO_x arise from direct emissions, while PM_2.5 is often largely from secondary processes and O₃ has both local secondary production and destruction superimposed on a large background. Meteorology influences both the dilution and deposition of primary emissions, as well as the production and destruction of secondary species through the availability of oxidants and the rates of atmospheric chemical processes (see Figure 1). It should be stressed that data sets included in this work were from northern hemisphere springtime (Figure S1). Although an O₃ increase with other pollutant reductions was evident for this period, such increases can have different NO_x-VOC sensitivities than do summertime O₃ changes. Although an O₃ increase is evident in the existing literature, more analysis is required as more papers are published throughout the year to assess the effects of emission reductions on summer O₃ formation.

In the following, each pollutant will be further investigated on a per country basis using all available data sets including studies that do and do not correct for the effects of meteorology. Although this introduces higher uncertainties, it improves the global data coverage and provides better statistics for comparisons to emission inventories. The distribution of studies that makes direct comparisons and those that correct for meteorological effects will be discussed for each pollutant, and a comparison to emission inventories will be performed when available.

Figure 7 shows the median decrease in NO₂ concentration (circles) during lockdowns for each country colored by the SI. Included in this calculation are studies using both the direct comparison approach (67%) and studies that correct for meteorological effects (33%). An overview of the measurements grouped by observation type as ground-based only (48%), satellite only (12%), or both (41%), is also provided. Also plotted is the adapted EDGAR inventory median decrease in emissions during lockdowns based on Forster et al. (2020; star squares). It should be noted here that Forster et al. (2020) implement the reductions in the emission inventory by scaling individual emission sectors. Studies mostly report data from lockdowns when stringency indices are greater than 50. Both observations and inventory-based reductions are reported as percentage difference.

Figure 8 plots the observed median decrease of NO₂ for each studied country against that country’s inventory decrease (Forster et al., 2020). The observed decrease for each country is further binned by percentage decrease (<20%, 20%–30%, 30%–40%, 40%–50%, and >50%), and the median inventory decreases for each observation-based bin are calculated. The bin ranges were chosen arbitrarily to ensure more than five data points per bin. These binned data are then colored by the median SI. For most countries, the observations and emission inventory agree within a factor of 2 (shaded area in Figure 8). The NO₂ decrease is driven for both atmospheric observations and the emission inventory by the stringency of the lockdown measures, with larger NO₂ decreases observed for higher stringency indices. Overall, despite the NO₂ observation-based uncertainties associated with instrument limitations (Section 2.2.1), satellite measurement uncertainties (Section 2.2.2), meteorological dependencies in determining the effects of shutdown (Section 2.3), and the uncertainties associated with inventory estimation reductions (Forster et al., 2020), the two approaches result in consistent emissions decreases, in line with the SI. This suggests that the stringency of lockdown measures has a strong influence on emissions from transportation, as exhibited by mobility data sets used to adjust global emission inventories (Forster et al., 2020). The similarity between changes in the emissions inventory and changes in atmospheric observations due to lockdown measures further confirms the importance of traffic as a source of NO_x in cities around the world. A more detailed analysis of the differences between the two approaches is beyond the scope of this review.

Figure 9 shows the median decrease in SO₂ concentration during the lockdown for each country colored by the SI (circles). Studies examining changes in SO₂ mostly used the direct comparison approach (80%), but 20% of studies corrected for meteorological effects. Also shown is the adapted EDGAR inventory median percentage drop in SO₂ emissions during the lockdown based on Forster et al. (2020; star squares). The pie charts show the breakdown by study measurement type, that is, ground-based, satellite, or both, for China (91%, 5%, and 5%, respectively), India (80%, 0%, and 20%, respectively), and for all other countries (76%, 8%, and 16%, respectively). All studies reported measurements from lockdown periods with an SI greater than 50. The majority of the studies were performed in China (25) and India (17), while three or fewer studies were performed for the remaining 18 countries.

Qualitatively, the inventory SO₂ emissions decreased with increasing SI, but the observed SO₂ changes were poorly correlated with either (Figure 10). Countries expected to account for the majority of SO₂ emissions globally based on the 2015 EDGAR inventory were China, India, and the United States, plus international shipping emissions (Figure S3). Observed SO₂ decreases and the Forster inventory reduction estimates showed discrepancies for China (16% vs. 26%), India (14% vs. 41%), and the United States (7% vs. 21%), suggesting that discrepancies on global scale emission estimates may also be expected. All but two studies were done in urban environments, and the average time period per study was greater than 50 days, resulting in urban-dominated SO₂ statistics for a long time period. The inventory SO₂ emission reductions are greater for the energy and manufacturing sources than from transportation. A lack of consistency in predicted versus observed SO₂ reductions (Figure 10) may therefore point toward uncertainties in the SO₂ inventory. NO₂, by contrast, arises primarily from transportation and shows better agreement between observation and inventory reduction estimates (Figure 8). However, there are fewer SO₂ observations and substantially fewer with meteorological normalization. There are also larger uncertainties associated with its measurement from ground-based and satellite-borne instruments. Further assessment of SO₂, a major precursor for PM_2.5, is an important topic for further study as more high-quality results from the COVID-19 period become available.

Figure 11 shows the median decrease in PM_2.5 concentration during lockdowns for each country colored by the SI (circles). Included in this calculation are studies using the direct comparison approach (i.e., no meteorological correction, 70%) and studies that correct for meteorological effects (30%). Effectively all analysis available to date considers PM_2.5 mass, and there is only limited information available on the changes in aerosol composition in response to lockdown measures. Also shown is the inventory median decrease in PM_2.5 emissions during the lockdowns approximated by both the organic and BC emission reductions based on the adjusted EDGAR inventory (Forster et al., 2020; star squares). We note that global inventories typically do not include speciated fractions of PM_2.5, which can be significant and reported by national-scale inventories (e.g., road dust, brake wear, tire wear). An inventory prediction of PM_2.5, which has both primary and secondary sources, is complicated by secondary PM formation, as discussed further below. PM_2.5 studies mostly report data from lockdowns when stringency indices are greater than 50.

Figure 12 further highlights the challenges associated with the comparison of PM_2.5 observations to the adapted EDGAR inventory based on Forster et al. (2020). The observed decreases for each country are further binned into percentage decrease ranges as in Figure 8. As the SI increased, the observed PM_2.5 median decreased; however, the inventory PM_2.5 emission reductions were poorly correlated. PM_2.5 can either be directly emitted or formed via secondary chemistry from a wide variety of other primary emissions (NO_x, SO₂, NH₃, NMVOCs, etc.). Therefore, a direct comparison of emission inventories and observations is challenging if the primary and secondary sources are not disentangled.

By the literature cutoff time of this review (September 30, 2020), only a few studies had been published that investigated the effects of secondary chemistry and local primary emissions on PM levels and composition in China (Chang et al., 2020; Chen et al., 2020a; Cui et al., 2020; Dai et al., 2020; Li et al., 2020b; Sun et al., 2020; Zheng et al., 2020), Bangladesh (Masum and Pal, 2020), and South Africa (Williams et al., 2020). More studies will be essential to understand the complexity of PM_2.5 pollution. For example, studies that address the possible effects of long-range transport that affect background PM levels are needed. Furthermore, changing atmospheric chemistry regimes can change secondary PM production rates. Characteristic examples that highlight this complexity are (1) the effects of NO_x reductions on organic peroxy radical (RO₂) chemistry affecting dimer formation and highly oxygenated molecules (e.g., McFiggans et al., 2019), (2) changes in organic and inorganic equilibrium partitioning, due to changes in particle acidity and the associated shifts between nitrate and sulfate formation in the particle-phase (e.g., Guo et al., 2016; Wang et al., 2016; Wang et al., 2020d), (3) changes in production rates of organic and inorganic pollutants due to increased availability of oxidants when emissions are reduced (e.g., Nault et al., 2018; Shah et al., 2018; Laughner and Cohen, 2019; Womack et al., 2019), and (4) changes in the rate of nighttime chemical processes (e.g., Kiendler-Scharr et al., 2016). PM composition measurements in addition to PM mass will be essential to assess these and other effects in order to elucidate changes in PM pollution arising from the COVID-19 emission reductions.

Figure 13 shows the observed median change in O₃ concentration from ground-based measurements during the lockdown for each country colored by the SI (circles). Included in this calculation are studies using the direct comparison approach without meteorological normalization (69%) and studies that correct for meteorological effects (31%). A violin plot shows the overall distribution of O₃ changes. Studies for most countries had SI values above 50 and either showed minor median O₃ decreases or increases in the 5%–20% range. O₃ increases greater than 50% were observed for Milan, Italy (reflected by the high 75th percentile values; Collivignarelli et al., 2020), as well as studies in Peru (Venter et al., 2020), Ecuador (Parra and Espinoza, 2020; Zambrano-Monserrate and Ruano, 2020), and Iraq (Hashim et al., 2020). To assess whether the changes in O₃ were driven by changes in emissions, the change in observed O₃ was plotted against the SI for each country together with the medians binned as done for Figures 8 and 12 (Figure 13, right panel). As the lockdown measures became more stringent, the percentage change in O₃ increased, suggesting that significant changes in O₃ formation were driven by emission reductions.

O₃ is a secondary pollutant whose formation results from the interplay of NO_x, VOC emissions, and meteorology (Sillman, 1999). Most regions have a significant background O₃ concentration, and local emissions may either deplete O₃ from this background or produce it photochemically. Although NO_x emission reductions were evident during the lockdown, changes in VOC concentrations and composition have not been well investigated (see Figure 5). Furthermore, the literature covered in this review was predominantly focused on February, March, and April (Figure S1). Studies were also weighted toward the northern hemisphere, representing late winter and early spring. During these months, O₃ concentrations are expected to be low due to reduced wintertime photochemistry (Khoder, 2009). For the studied periods, it is therefore expected that an increase in O₃ could be more sensitive to NO emission reductions that would reduce O₃ titration. However, summertime measurements of O₃, NO_x, and VOCs are essential to investigate how changes in emissions affect O₃ formation when photochemistry is at its peak. Greater biogenic VOC and wildfire biomass burning emissions during the summer significantly alter VOC speciation and abundance. It is therefore evident that although reduced emissions increased O₃ concentrations in late winter and early spring, more studies will be necessary to address COVID-19-related shifts in summertime O₃, which is sensitive both to the local chemical environment and broad-scale changes in the ozone background.

Figure 14 shows the median change in pollutant concentrations for PM₁₀, NH₃, NO_x, AOD, BC, AQI, NMVOCs, and CO during the lockdown for each country colored by the SI. Note that for AQI, no unique definition exists, and it is used to assess the simultaneous presence of multiple pollutants. For example, in the United States, AQI is calculated based on the concentration of PM, O₃, SO₂, and CO (Bishoi et al., 2009), whereas in China, AQI is determined by the concentrations of the above four pollutants plus NO₂ (Fareed et al., 2020). Studies using the direct comparison approach accounted for 80%, 100%, 56%, 57%, 86%, 82%, 50%, and 72% of the data sets for PM₁₀, NH₃, NO_x, AOD, BC, AQI, NMVOCs, and CO, respectively. For the majority of countries, a decrease in pollutant concentrations was evident during the lockdowns compared to reference periods (see also Table 8). On average, decreases of 22% (±19%), 9% (±13%), 43% (±25%), 9% (±20%), 51% (±13%), 11% (±28%), 59% (±64%), and 27% (±18%) were observed for PM₁₀, NH₃, NO_x, AOD, BC, AQI, NMVOCs, and CO, respectively. More studies are needed to better understand the effects of the lockdowns on the above pollutant concentrations. With most of the current literature failing to account for the possible effects of meteorology, these results suggest a need for future studies in this area. Furthermore, NMVOCs and NH₃ can contribute to PM pollution through atmospheric chemical processes, and BC has direct impacts on climate forcing, which highlights the need to better monitor their concentrations.

An interesting question is to what degree emissions reductions during the COVID-19 pandemic brought regions into compliance with air quality standards. Table 1 lists the WHO exposure guidelines for a series of common air pollutants. Assessment of compliance with these standards requires comparison to absolute pollutant concentrations. Only a subset of the literature reviewed here reports data in absolute units, with the majority reporting relative changes without providing the underlying concentration values. This section therefore summarizes literature that reported the concentration of NO₂, PM_2.5, O₃, and CO using the direct comparison approach.

Figure 15 shows the mean concentrations of these four pollutants during lockdowns (circle markers) and reference periods (star markers) reported by 96 publications across different continents. Asia, the largest and most populous continent, is further separated into different geographical regions. Violin plots show the distribution of concentrations reported within each region. Circle markers are colored by the percentage change of the mean lockdown concentration with respect to mean reference period concentrations per region/continent. Also shown are the WHO guideline values for NO₂, PM_2.5 and O₃ for different exposure times, including 1 year, 24 h, and 8 h. Finally, the number of publications and the number of collected measurements (from different sites and/or times) are provided in parentheses per region/pollutant. Although the WHO guideline values are limited to multiple-hour or annually averaged exposure times for the different pollutants, observation-based concentration averages range from 1 week to 5 months. A direct comparison of observations to the WHO guideline values is therefore challenging. However, the WHO guideline values for hourly and annual means provide a range of concentrations that put the observed means into perspective. For example, if monthly measurements of a pollutant are greater than the hourly WHO guideline values, then exceedances by definition occurred during the studied period; if observations are greater than the annual guideline values, then exceedances could also occur if high concentrations were to persist beyond the observation period. In the following, each pollutant is examined separately, and the literature corresponding to each pollutant is provided in Tables 5–8. Concentration changes on a national level are further discussed in Section S4 and downloadable from the database (See Section 2.1).

The mean NO₂ concentration reported by all studies for lockdown conditions was 22 ± 18 µg m^–3 (1 standard deviation) and lower than the reported mean, 31 ± 15 µg m^–3, for the respective reference periods. The lockdown and reference concentrations from 100% ground-based measurements were below the WHO annual guideline values, and NO₂ concentrations decreased during the lockdown compared to the reference period for all regions except West Asia. Literature corresponding to these measurements is provided in Table 5.

The mean PM_2.5 concentration reported for lockdown conditions was 24 ± 14 µg m^–3 and lower than the reported mean, 32 ± 22 µg m^–3, for the respective reference periods. The lockdown and reference concentrations (80% ground-based, 2% satellite, 18% both) were above the WHO mean annual guideline value of 10 µg m^–3 for nearly all regions, and mean values in Asia often exceeded the 24-h guideline value of 25 µg m^–3. PM_2.5 decreased during the lockdown compared to the reference periods for all regions except West Asia. However, this decrease was not sufficient to reduce concentrations below the WHO guideline values during the lockdown, especially for regions in Asia. This variability in PM_2.5 concentrations may reflect the much wider variety of PM_2.5 sources with secondary PM_2.5 responding to the changes in NO_x, VOCs, SO₂, and NH₃ as well as many other sources. Literature corresponding to these measurements is provided in Table 6.

The mean O₃ concentration reported by all studies for lockdown conditions was 43 ± 21 µg m^–3 and higher than the reported mean, 36 ± 19 µg m^–3, for the respective reference periods. The lockdown and reference concentrations from 100% ground-based measurements were below the 8-h mean WHO guideline value (100 µg m^–3, approximately 50 ppb) for all regions. However, the ground-based mean values were calculated as averages for multiple weeks/months, including nighttime measurements, allowing for the possibility that midday O₃ might still have exceeded the 8-h guideline value. O₃ increased during the lockdowns compared to the reference periods for all regions except South Asia. An overview of the literature corresponding to these measurements is provided in Table 7 and is dominated by urban measurements (>98%).

The mean CO concentration reported by all studies for lockdown conditions was 580 ± 310 µg m^–3 and lower than the reported mean, 630 ± 440 µg m^–3, for the respective reference periods. There are currently no WHO guideline values for CO to compare to; however, the U.S., European, and Chinese guidelines shown in Table 1 are significantly greater than the observed CO concentrations. CO decreased during the lockdowns compared to the reference periods for all regions. An overview of the literature corresponding to these measurements is provided in Table 8.

Analysis of emissions changes and their resulting influence on air quality worldwide during the COVID-19 pandemic is a rapidly evolving topic of intense public and scientific interest. This review provides a summary of the current literature that has examined mainly the stringent, early lockdown periods in February–May 2020. Despite the short duration of the observational time period and the limited time since that period to this writing, the number of papers and the depth of the analysis is substantial. Already, there are several initial conclusions, recommendations for careful consideration and best use of the observations, and a list of suggestions for further analysis. This review synthesizes these reported changes in air pollutants during the COVID-19 lockdowns and further provides context for these changes using the SI, a unified, globally consistent measure of the policy response to confining the pandemic. All data digitized for analysis in this review are available on the website in https://covid-aqs.fz-juelich.de. This website is designed as a living version of this review, that is, as new literature emerges, authors of published papers are encouraged to upload their data to the database, thus complementing the data coverage in space, time, and compound dimensions.

Much of the COVID-19 air quality literature surveyed for this review does not explicitly account for the effect that the year-to-year variability, largely driven by meteorology, has on observed atmospheric concentration changes. The dependence of concentration changes on the SI are readily apparent in the meteorologically corrected data but not in the uncorrected data. We recommend that all future analyses take explicit account of meteorology and specify the method for doing so, for example, as outlined in Section 2.3.2, or perform chemical transport modeling (Section 2.3.3) since disregarding this consideration largely increases the associated uncertainties.

Two of the main species arising from primary emissions analyzed to date are NO₂ and SO₂, both of which have readily available ground-based monitoring networks and satellite remote sensing data sets. They also arise from different emission sectors, with NO_x (for which NO₂ is a proxy) having its largest contribution from transportation and SO₂ from power generation and certain industrial sources. For NO₂, the observed changes correspond within uncertainties with the estimated emission inventory reductions when accounting for COVID-19 lockdowns. The analysis of NO₂ also encompasses the largest number of publications and the largest number that explicitly account for meteorological effects. Analysis of SO₂, by contrast, shows distinct evidence for reductions during the lockdown periods, but those emissions reductions are not as clearly associated with predictions from inventories. This difference may be due to incomplete information of the impacts of COVID-19 on industrial activity and the more limited publication database of SO₂ changes during COVID-19, especially for papers that account for meteorology, or uncertainties in the measurement database for SO₂. We recommend further investigation of the SO₂ reductions, especially since this has the potential to inform inventories for this critical PM_2.5 precursor.

Analysis of changes on primary emissions of other species, such as CO, NMVOCs, BC, and NH₃, is more limited. CO is covered in the literature by 67 publications, albeit predominantly without explicitly accounting for meteorological impacts on pollutant abundance. BC, NH₃, and NMVOCs are covered by seven, five, and 10 publications, respectively, far from providing a global overview at this stage. This reflects the frequency with which these compounds are typically measured by air quality monitoring networks. However, considering the important role of SO₂ and VOCs in secondary aerosol formation and of CO and VOCs in ozone chemistry, further analysis of available COVID-19 changes is needed for these species.

Both PM_2.5 and O₃ have large secondary sources arising from complex atmospheric chemical cycles, and together they are responsible for the majority of adverse health outcomes associated with air pollution. Total PM_2.5 mass and O₃ are covered comparatively widely in the literature, although there is little information on speciated PM_2.5 composition. The COVID-19 literature shows that PM_2.5 decreases with increasing SI, whereas O₃ increases with SI. Key uncertainties in the understanding of these changes should be addressed through the combined use of observational and atmospheric chemistry modeling approaches. With the chemistry leading to ozone and secondary aerosol (both organic and inorganic) formation being nonlinearly dependent on NO_x levels, the lockdown periods and seasonality of its effect on pollutants offers unique possibilities to assess model abilities in capturing changes on local to global scales. Further analysis of photochemically active periods with reduced emissions in the northern hemisphere in forthcoming literature may be particularly informative.

More use can be made of the high-resolution capabilities of the TROPOMI sensor, in particular for NO₂. Analyses of the data at high resolution may provide COVID-19 emission impact estimates of different sources and source sectors such as individual power plants, highways, shipping, urban areas, industrial complexes, and airports. The influence of the day-to-day changes in weather is, however, substantial on such local scales, and we recommend that such high-resolution satellite-based studies make use of high-quality weather analyses and chemical modeling. When using satellite NO₂ measurements, we advise that the averaging kernels remove the dependency on the retrieval a priori profile shape, which can always be done when three-dimensional CTMs are involved in the analysis. Satellite instruments like TROPOMI and OMI measure at one given overpass time (e.g., 13:30), but it should be considered that the diurnal profile of the emissions may have changed during the lockdowns. Satellite retrievals often suffer from systematic uncertainties. In the case of TROPOMI NO₂, we mention a negative bias compared to surface remote sensing observations, with an apparent linear scaling with the tropospheric column amount over polluted regions. This scaling suggests that relative changes, for example, the percentage reduction compared to a reference time period, are not so sensitive to the negative bias, and we recommend reporting such relative differences, which is done in most of the papers studied in this review.

As the atmospheric chemistry community makes continued efforts toward observational coverage of the pandemic influences on atmospheric composition, we anticipate that additional data sets will become available for further analysis. We recommend attention to the following issues, although this list is certainly far from comprehensive.

First, changes in O₃ associated with COVID-19 emissions reductions, particularly during photochemically active seasons, may be informative for assessing the sensitivity of this photochemistry to NO_x and VOCs in different regions. The northern hemisphere late winter and early spring period that dominates this review reflects O₃ data that are not particularly sensitive to photochemistry. However, careful comparisons of meteorologically normalized O₃ to detailed photochemical models may elucidate NO_x and VOC sensitivities that can inform regional O₃ mitigation strategies across the world.

Second, changes in PM_2.5 may enable similar sensitivity analyses to primary emissions. As PM_2.5 composition depends on a number of emission sources and chemical cycles, a broader analysis of chemically speciated PM_2.5 data, where available, will be especially informative.

Third, but related to both of the above, the seasonality of O₃ and PM_2.5 may be addressed if there are sufficient observations of emissions reductions across different hemispheres or times of year. Given the trajectory of the COVID-19 pandemic at the time of this writing, such an analysis may be feasible even within the northern hemisphere. Particularly for PM_2.5, there is a well-known seasonality, with severe effects arising from distinct cycles and emissions that occur in midlatitude summer and winter.

Fourth, expansion of the available analyses to include a larger number of species would help to constrain and inform emissions inventories. This review has provided an initial analysis of the difference between NO₂ and SO₂. Further analysis, to include detailed analysis of CO, BC, NH_3, and especially speciated NMVOCs, where available, would provide unprecedented tests of the current understanding of emissions inventories across an array of sectors.

Fifth, analysis of the radiative forcing associated with short-lived climate forcers is a priority. For example, regional emissions changes should lead to both local and hemispheric effects on O₃. The influence on this broader scale, or background O₃, needs to be evaluated through both modeling and observational efforts. Remote sensing O₃ products and vertical profiles from, for example, O₃ LIDAR networks will be particularly informative. Similarly, changes in PM_2.5 affect both regional air quality and global climate. Widespread global reductions in primary emissions and PM_2.5 precursors must similarly be evaluated in terms of their short-term climate forcing in 2020.

Finally, changes in the oxidative capacity of the global atmosphere arising from COVID-19 may also have occurred with changes in NO_x and other species, but those changes have yet to be evaluated. Such changes have the potential to influence the lifetime of methane, an important greenhouse gas. Model evaluations will be informative in this regard since we anticipate few, if any, observations of the influence lockdowns have on oxidants such as HO_x radicals.

We note that this review has been limited in scope to air pollutants that are of importance as short-lived climate forcers. However, to our knowledge, no information is currently available on short-lived climate forcers such as methane and halogenated compounds. N₂O and CO₂ are beyond the scope of this review with the latter evaluated elsewhere (Le Quéré et al., 2020). The developing analysis of the COVID-19 emission reductions will certainly address these topics.

The review compiles data published in the peer-reviewed literature. All data are accessible through https://covid-aqs.fz-juelich.de designed as a living version of this review. The data sets from the website are provided with free and unrestricted access for scientific (noncommercial) use including the option to generate targeted reference lists. Users of the database are requested to acknowledge the data source and reference this review in publications utilizing the data set. As new literature emerges, authors of published papers can upload their data to the database, thus complementing the data coverage in space, time, and compound dimensions.

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

Text S1–S4. Figure S1–S4. Table S1–S2. Docx.

The authors acknowledge discussions with Pieternel Levelt and Pepijn Veefkind on the satellite measurements, as well as the support from John Douros of Royal Netherlands Meteorological Institute in computing the Copernicus Atmosphere Monitoring Service (CAMS) global reanalysis simulations of the TROPOMI NO₂ observations. This article contains modified CAMS Information (2020). They also acknowledge Alina Zimmermann and Michael Decker for support on the database operation and Kenneth C. Aikin for IGOR programming support for the figures. Finally, they acknowledge and appreciate the fast response from the editor and reviewers for this timely publication.

The authors acknowledge institutional funding from the Earth and Environment research field of the Helmholtz Association (FZ Jülich), the NOAA Atmospheric Chemistry Carbon Cycle and Climate (AC4) Program, and the ESA Impacts of COVID-19 lockdown measures on Air quality and Climate project.

The authors declare no competing interests.

Performed the literature review and categorized the papers: GIG, JBG, SB, BCM, AKS.

Extracted data from the reviewed literature and performed analyses: GIG, JBG.

Provided Figure 1: CT.

Provided Figure 4: HE, ACL.

Designed the database and web page: ARG, AP.

All authors contributed to the interpretation of data, drafted and/or revised this article, and approved the final version for submission.