Markers of economic activity in satellite aerosol optical depth data

This study investigated the impact of COVID-19 lockdowns on satellite aerosol optical depth (AOD), to explore the hypothesis that if changes in economic activity are seen in emissions of NO2, an aerosol precursor, then AOD should change commensurably. We developed a technique to filter AOD data to isolate changes associated with anthropogenic emissions. Overall, in 37 of the 43 cities that were identified as top oxides of nitrogen (NO x ) emitters from their transportation sectors, AODs decreased by 21.2% ± 7.8%, 18.9% ± 11.7%, 27% ± 12.4%, 22.9% ± 7.6% in the United States, India, western Europe, and China, respectively—an average of 22.4% ± 7.4%. In contrast, AODs increased on average by 11.7% ± 8.4% in Taiwan, where economic stimulus was used as a strategy during the pandemic. This analysis implies NO x and volatile organic compounds emissions reductions from the transportation sector can be targeted, and by transitioning 6 million light duty vehicles from gasoline to electricity, the US can achieve 21% improvement in AOD.


Introduction
The impact of sudden/short-term economic activity changes on anthropogenic emissions and associated air pollution during the shutdown of traffic during major public events such as the 2008 Beijing Olympics has been reported by many studies (Cermak and Knuff 2009, Tan et al 2009, Witte et al 2009, Wang et al 2010, 2017, Hao et al 2011, Guo et al 2013, Ding et al 2015, Sun et al 2016, Tong et al 2016, Li et al 2019. Some of these studies used satellite observations of Ozone Monitoring Instrument (OMI) tropospheric column nitrogen dioxide (tropNO 2 ). As reported by the meta-analysis of Gkatzelis et al (2021) more recently, Tropospheric Monitoring Instrument (TROPOMI) tropNO 2 data were used to document the impact of COVID-19 pandemic related lockdown on its changes and attributed those reductions to improvements in air quality in general. Though NO 2 by itself is a harmful pollutant that leads to human health impacts, its role in the photochemical production of ozone and fine particulate matter (particles < 2.5 µm, PM 2.5 ) is the most important as these are the two pollutants with harmful human health impacts that are monitored at the surface to determine if air quality is good or bad. In countries like the US, the national NO 2 standard is weak and no location is out of compliance (Kroll et al 2020). Ozone and PM 2.5 are photochemically formed from precursor gaseous pollutants that include nitrogen oxide and nitrogen dioxide (NO + NO 2 = NO x ), sulfur dioxide (SO 2 ), ammonia (NH 3 ), and volatile organic compounds (VOCs), hereafter referred to simply as precursor gaseous emissions (Kroll et al 2020).
PM 2.5 is harmful to human health and is regulated across the globe. While particles are directly emitted into the atmosphere from the use of fossil fuels, they are also photochemically produced by oxidation of precursor gaseous emissions. Particle concentrations can be directly measured or inferred from optical measurements (column integrated light extinction in unitless quantity called aerosol optical depth, AOD) due to their absorption and scattering properties. Therefore, any changes in economic activity that lead to changes in primary and precursor emissions could lead to associated changes in AOD.
This study investigated the impact of COVID-19 lockdowns on satellite AOD, to explore the hypothesis that if changes in economic activity are seen in emissions of NO 2 , an aerosol precursor, then AOD should change commensurably. It is shown that during the COVID-19 lockdowns in the US, when human activity decreased, NO x emissions from traffic (passenger vehicles) dropped dramatically . Carbon monoxide, VOCs and NO x are the dominant emissions from vehicular (cars and trucks) combustion exhaust in addition to greenhouse gases, but of these, VOCs and NO x lead to secondary ozone and PM 2.5 formation. Primary PM 2.5 emissions are fifty times lower than NO x emissions from cars and trucks according to the Environmental Protection Agency emissions inventories in the US and thus are not a significant source for AOD (www. bts.gov/content/estimated-national-average-vehicleemissions-rates-vehicle-vehicle-type-using-gasolineand).
Reductions in emissions of precursor gases and PM 2.5 from traffic and industrial sources led to low aerosol concentrations and thus low AODs during the COVID-19 pandemic related lockdowns (Kumar et al 2020, Kumari et al 2020, Venter et al 2020, Hammer et al 2021, Khan et al 2021, Straka III et al 2021. Studies that investigated the impact of reduced emissions on climate found that though AOD decreased, the decrease was not large enough to detect changes in clouds, radiation, surface temperature etc. due to natural variability (Gettelman et al 2021, Jones et al 2021. Interpreting changes in PM 2.5 is complicated due to complex photochemistry related to secondary aerosol formation as well as emissions of primary particulates from natural and anthropogenic sources (Kroll et al 2020, Venter et al 2020, Hammer et al 2021. Even though Hammer et al (2021) used a global chemistry and transport model to separate the effects of meteorology and assumed transportation sector emissions changes of 50% due to lockdowns on PM 2.5 , they did not include the contributions of smoke from fires and the differences in fire activity between 2020 and 2019. Additionally, Hammer et al (2021) did not trace the exact reduction in secondary aerosol formation due to reduction in NO x emissions compared to the reduction in primary particulate emissions. Similarly, Venter et al (2020) attributed observed PM 2.5 increases in 2020, despite the lockdowns, to transported aerosol and cited historical burning patterns in areas where PM 2.5 increases were observed. In addition to weather, variability in emissions from natural sources and the long-term trends in PM 2.5 must be considered when assessing the impact of lockdowns on PM 2.5 and AOD.
Indeed, in many parts of urban areas dominated by NO x and VOC emissions from the transportation and industrial sectors, interpreting changes in PM 2.5 or AOD can be difficult due to regional transport of smoke or dust impacting local air quality. Additionally, the true change in AOD due to emissions changes in 2020 can only be identified after removing the long-term trend in AOD due to various clean air emissions controls in place across the world, especially in China as reported by Sogacheva et al (2018).
The main focus of this study is to develop a method to use satellite data to detect AOD changes due to changes in anthropogenic NO x emissions when there are changes in economic activity that last for a month or longer using the impacts of COVID-19 lockdowns on AOD as a case study. We used Suomi National Polar-orbiting Partnership satellite Visible Infrared Imaging Radiometer Suite (Suomi NPP VIIRS) AOD retrievals, along with Sentinel-5 Precursor satellite TROPOMI tropNO 2 observations as markers for urban/industrial pollution to isolate the impact of anthropogenic emissions changes on AOD. In this study, SO 2 emissions were not a focus because traffic emissions are not a significant source for SO 2 (McDuffie et al 2020). Globally, SO 2 emissions have fallen by 32% between 2004 and 2017 primarily due to the implementation of stricter emission standards for the energy and industry sectors after 2010 in China (Zheng et al 2018).

Suomi NPP VIIRS AOD
The Suomi NPP satellite VIIRS has been providing a pixel-level (∼750 m) AOD product for various user applications (e.g., Huff et al 2021, Li et al 2021. In order to put into perspective, the observed VIIRS AOD changes due to changes in economic activity measures, we wanted to ensure that the AOD record is consistent and devoid of any artificial trends. Therefore, we reprocessed 9 years of VIIRS AOD with an improved sensor calibration as well as an enhanced algorithm that provides retrievals over both dark vegetated and bright surfaces , Zhang et al 2016, Laszlo 2018. The recalibrated radiance data have improved long-term stability (better than 0.3%) and accuracy with uncertainty < 2% (Uprety et al 2020, Cao et al 2021. It also incorporated a bias correction for the reflective solar bands, which addressed known calibration biases for VIIRS bands M5 (0.672 nm) and M7 (0.865 nm). Other improvements to the calibration included terrain correction, straylight correction, and corrections to anomalies in the thermal bands. The reprocessed VIIRS AODs were compared to a global network of ground-based Sun photometers using match-up criteria defined by Liu et al (2014).
For the validation of the reprocessed AOD data, we stratified the AODs into three categories: AODs lower than 0.1, between 0.1 and 0.8, and greater than 0.8. For these three AOD ranges, the global mean biases/ root mean square errors were found to be 0.02/0.048, 0.008/0.06, and −0.121/0.39, respectively.

Sentinel-5 Precursor TROPOMI tropospheric NO 2
The Sentinel-5 Precursor satellite, containing the TROPOMI sensor, orbits the Earth in formation with Suomi NPP, and as a result, VIIRS AOD and TROPOMI tropNO 2 are considered observations of the same atmospheric column. The NO 2 algorithm retrieves total column NO 2 and separates the stratosphere from troposphere using the chemical transport model predicted stratospheric NO 2 analysis fields (van Geffen et al 2019). TROPOMI's tropNO 2 product (3.5 km × 5.6 km) with retrieval quality flag greater than 0.75 is used in this study.
Sources of error (∼25%) in tropNO 2 include altitude dependent air mass factors, stratospheretroposphere separation of NO 2 , a priori NO 2 profile and shape, surface albedo climatology, and calibration errors as a function of view angle (van Geffen

NO 2 filter for AOD (NO 2 F4AOD) to screen non-anthropogenic sources of aerosols
We developed a method to screen VIIRS AOD data using tropNO 2 data to isolate changes in AOD associated with changes in anthropogenic emissions. The rationale we took to develop this method is that NO 2 is short-lived and observed near source regions whereas aerosols have a longer lifetime and can be transported long distances. For example, emissions of NO x from fires are 8.3 times lower than PM 2.5 (Andraea 2019). Accordingly, when a transported smoke plume reaches an urban/industrial area, tropNO 2 concentrations are small but AOD values are high (Veefkind et al 2011). Given this scenario, when anthropogenic sources are dominant, both NO 2 and AOD values are elevated but when biomass burning sources are dominant, AOD values are higher and NO 2 values are lower.
In this NO 2 F4AOD method, we looked at tropNO 2 and AOD changes to see if they co-increased or co-decreased between 2020 and 2019 and filtered AOD data based on absolute changes in tropNO 2 greater than 8 µmoles m −2 , which is the expected retrieval error (van Geffen et al 2019). The NO 2 F4AOD method shown in figure 1 illustrates how the filtering of AOD is carried out. The top left panel shows the mean AOD difference between 2020 and 2019 for 10 February to 25 February. This period corresponds to the first COVID lockdown in China. AODs decreased in the central part of China corresponding to Hubei province, but there are increases in AODs in the southwestern and northeastern part of China; many studies reported similar findings indicating that warm and humid conditions were favorable for the production of sulfate aerosol from SO 2 emissions in northeastern China (Le et al 2020, Anderson et al 2021, Loeb et al 2021, Kong et al 2023. The tropNO 2 image for the same period, in the bottom left panel, shows a decrease in tropNO 2 in the same regions where AOD decreased, but no noticeable or significant tropNO 2 increases corresponding to the increase in AODs in southwestern and northeastern China are present. Prior to applying our screening method, we calculated background rural tropNO 2 using TROPOMI data and found it to be ∼16 µmoles m −2 ; this background value was subtracted from daily TROPOMI tropNO 2 data to isolate clusters of elevated tropNO 2 values associated with urban/industrial sources. Once we pass the AOD data through the filter (background tropNO 2 > 16 µmoles m −2 and ∆tropNO 2 > 8 µmoles m −2 , ∆AOD co-increase or co-decrease along with ∆tropNO 2 ), the AOD increases in southwestern China disappear, but some increase in AOD remains in northeastern China. Similarly, the increases in AOD over the ocean also disappear after application of the filter.
The differences in tropNO 2 and in AOD between 2020 and 2019 due to meteorology were minimized by averaging the satellite data for the duration of the lockdown period (approximately a month or longer). When looking for signatures of emissions changes in satellite observations between different years, averaging the data over 30 d is expected to minimize the influence of meteorology and differences in sampling in satellite data due to clouds , Liu et al 2021. However, the influence of meteorology can remain especially when change in NO x emissions is small (Wang et al 2021). This technique was applied to track changes in AOD between 2020 and 2019, in order to see the magnitude of changes in AOD, but ideally a reference climatology should be used because of interannual variability in meteorology that is not fully accounted for when differences between 2 years are used. The NO 2 F4AOD method could not be applied over longer time periods because tropNO 2 data are available only from 2018. To circumvent this problem, we identified the top 5% NO x emitting urban/industrial regions from the 2019 Community Emissions Data System (CEDS) emissions inventory and conducted an AOD time series analysis to account for the long-term trend present in AOD observations. Instead of using the traditional least-squares fit analysis to identify the longterm trend in the AOD time series data, we used the Theil-Sen slope method and determined the significance of the derived trends using the Mann-Kendall test (Thiel 1950, Sen 1968). This method is insensitive to outliers and is shown to be significantly more accurate than the least squares method. We also confirmed that these top 5% NO x emitting urban/industrial regions are under the influence of high NO x emissions from the transportation sector by analyzing the 2019 CEDS (McDuffie et al 2020). Finally, in reporting AOD changes due to the COVID lockdowns in various cities across the globe in 2020 compared to climatology, we accounted for the long-term AOD trend by adding the trend value to 2020 AODs.

Results
Figure 2 shows VIIRS AOD changes (∆AOD 2020-2019 ) identified due to changes in emissions between 2020 and 2019 for China, India, western Europe, and the US. The panels in the left column (figures 2(a)-(d)) are unfiltered data, whereas the panels in the right column (figures 2(e)-(h)) are filtered using NO 2 F4AOD. The 2020 lockdown periods are different for these four regions: China from 23 January to 8 April, Europe from 9 March to 4 July, US from 19 March to 15 April, and India from 25 May to 7 June.
Of the four regions investigated, the unfiltered VIIRS AOD changes in every region showed a combination of decreases (areas in blue) and increases (areas in red) during the lockdowns. However, the ∆AOD 2020-2019 from the NO 2 F4AOD method removed the increases associated with nonanthropogenic sources (e.g., smoke from fires, blowing dust). The largest reductions in AOD were observed in China because that is where AOD levels associated with urban/industrial pollution are the highest in the globe. Figure S1 shows 2019 annual mean AOD for US and China showing that pollution in China is four times greater than in the US. The lockdown areas of Hubei province, which includes Wuhan and other cities, resulted in an ∆AOD 2020-2019 change up to 0.2 over 2 months, but during the most stringent portion of the lockdown, from 10 February to 25 February, the ∆AOD 2020-2019 values decreased by up to 0.5. For a typical AOD value of unity corresponding to pollution, this is a 50% decrease.
There are some areas in China with increases in ∆AOD 2020-2019 due to smoke from fires; analysis of Suomi NPP VIIRS fire detection and fire radiative power (FRP) products shows distinct correlation with elevated levels of fire intensity in areas where an increase in ∆AOD 2020-2019 was observed. Figure S2 shows the difference in VIIRS smoke fraction between 2020 and 2019 in Asia. Fires in some regions in 2020 were quite strong, and smoke from these fires offset the reductions in AODs from lockdowns, especially in urban areas closer to the fires. Though Chinese cities such as Chengdu, Chongqing, Guiyang, Kunming, and Nanning are among the top 5% NO x emitting cities, ∆AOD 2020-2019 values increased in these cities because they were under the influence of locally transported smoke due to their proximity to intense burning in Southeast Asia. These local/regional transport of smoke from fires is a well understood phenomenon (Zhu et al 2017(Zhu et al , 2022. Note that over long distances, tropNO 2 from fires is not detected, but over short distances, it is still observed in smoke plumes; distance traveled by smoke plumes depends on wind speed and varies from one fire event to another. These increases are not due to unaccounted meteorological differences between 2020 and 2019 in our analysis; consistent with our analysis, Hammer et al (2021) showed no such AOD increases due to meteorology in their model simulations. The increase in ∆AOD 2020-2019 . Additionally, in downwind regions of China such as Seto Island Sea, AOD changes were minimal because SO 2 emissions from ships that are a source of sulfate aerosol along with industrial emissions did not change much between 2020 and 2019 though they have been trending downward (Itahashi et al 2021).
In India, emissions from the transportation sector, power plants, and industry dropped by up to 75% during the lockdown (Tibrewal and Venkataraman 2022). This led to lower values of AOD as observed by satellite data; in the unfiltered AOD map ( figure 2(b)), the entire Indo-Gangetic plain and southern India showed reductions in AOD, whereas central India showed increases in AODs. The increase in ∆AOD 2020-2019 in central India is attributed to intense fires in 2020 compared to 2019 (figure S2); AOD filtered using NO 2 F4AOD shows only decreases in AOD (figure 2(f)). Mishra and Rathore (2021) reported a 30% increase in fire activity in India in 2020 compared to 2019, which is consistent with the VIIRS smoke fraction difference between 2020 and 2019 shown in figure S2. It is notable though that the smoke fraction was higher in 2020 compared to 2019 in most parts of India including the Indo-Gangetic plain region. Despite the increased fire activity in 2020, AODs decreased overall during the lockdown. This is either due to this region being dominated by anthropogenic sources that lead to very high AODs and may even be higher or comparable to AODs due to smoke from fires or unusually high precipitation observed in 2020 (Sathe et al 2021).
In Europe, emissions from the transportation sector decreased by 89% (Spain), 86% (Italy), 82% (France), 47% (Germany), and 70% (United Kingdom) during the lockdowns (Acharya et al 2021). Similar to China and India, domestic energy consumption in Europe increased due to work-fromhome activities yet reductions in retail and recreation activities led to ∆AOD 2020-2019 decreases.
In the US, after filtering for AOD changes not associated with the lockdown, small decreases in AOD were observed in the Interstate-95 corridor in the Northeast US (major transportation pathway), Los Angeles, San Francisco, and the Southeast US. The increased AODs over the Gulf of Mexico in the unfiltered data (figure 2(d)) are due to the monthlong Saharan dust transport event that occurred in June 2020 (Asutosh et al 2022); our NO 2 F4AOD filtered out those high AODs (figure 2(h)). Our findings for AOD changes in the US are consistent with Hammer et al (2021), Naeger and Murphy (2020), and Acharya et al (2021). Analysis of NOAA fuel-based emissions estimates shows that on-road PM 2.5 emissions decreased by 40% in Los Angeles and 55% in New York compared to a pre-COVID scenario. These reductions in emissions during the lockdown period are substantial but the magnitude of primary PM 2.5 emissions is small; decreases in AOD are the benefit of reductions in NO x and VOC emissions. Acharya et al (2021) report that some of the ∆AOD 2020-2019 increases seen in the US are attributable to a 30% increase in SO 2 emissions combined with high relative humidity and low wind speed that led to secondary sulfate aerosol formation. These increased AODs are filtered out in our analysis as we screen AOD using the tropNO 2 filter and SO 2 emissions are negligible from the transportation sector.
To quantify the changes in ∆AOD 2020-2019 in major urban/industrial regions dominated by emissions from the transportation sector, we identified 43 cities encompassing the top 5% NO x emitting regions across the globe and calculated the percent change in AOD (figures 3(a), (c), (e) and (g)). Barring cities such as Chengdu, Chongqing, Guiyang, and Nanning that were under the influence of smoke transport, most regions in China showed an overall AOD decrease of 22.9%± 7.6%. Similarly, ∆AOD 2020-2019 decreased by 21.2%± 7.8% in the US, 27%± 12.4% in Europe, 18.9%± 11.7% in India.
In figures 3(b), (d), (f) and (h), we show analysis of AOD changes in the 43 cities using climatology as a reference. Instead of comparing tropNO 2 filtered AOD data between 2020 and 2019, we used AOD data from 2012 to 2019 to conduct a time series analysis using the Theil-Sen method (Theil 1950, Sen 1968). This method accounts for the long-term trend in AOD and isolates AOD changes strictly due to lockdown-related emissions changes. The results are consistent with those obtained from the NO 2 F4AOD method (figures 3(a), (c), (e) and (g)). Mean AOD decrease, including cities that showed AOD increases using NO 2 F4AOD method is 16.5% and with the analysis using climatology as a reference, the mean AOD decrease is 18.5%.
The impact of the lockdowns on AOD changes was found to be marginally sensitive to the way the analysis is carried out, especially for cities that were in close proximity to fire activity. When considering the cities in the analyses reported in figure 3, we used the city perimeter based on population density polygons. For cities in the US, the metro area polygons were taken from the 1:500 K TIGER line/shape files downloaded from www.census.gov/ geographies/mapping-files/time-series/geo/tigerline-file.2019.html. For other countries, the city polygons were downloaded from https://gadm.org/data. html. The Global Administrative Areas (GADM) are free licensed worldwide high-resolution data available to the public. In our analysis, for most cities, we used administrative level 2 data or manually combined a group of nearby level 2 and level 3 cities.
We also did another set of analyses where we identified the city center and used a radius of 27.5 km to define the heart of the city. The results are consistent for most cities in magnitude and sign of AOD change. However, for Nanning, Guiyang, Beijing, and Kunming in China, we found major differences and attribute them to these regions being under the influence of smoke from nearby fires (table 1). For Nanning and Guiyang, the sign of AOD change did not flip but the magnitude of the impact was found to be greater when polygon shape was used to identify the city. For Beijing, when non-polygon shape was used to identify the city, there was no AOD change but when polygon shape was used, the AOD change was −8%. For Kunming, there was no AOD change using non-polygon shape, but the AOD change was 50% using polygon shape. This is expected because Kunming was under heavy influence of smoke and the wider the spatial area considered, the larger the influence of smoke from fires. For all other cities in Europe, India, and the US, no major differences were found in the AOD changes based on the two estimates.
The robustness of our AOD changes due to lockdowns were also cross-verified using Synthetic Control Methods (Abadie 2021). Using this method, we investigated AOD changes in Wuhan, Beijing, and Shanghai in China. In this purely statistical approach, we tested the effect of intervention (lockdown) by constructing a weighted combination of controls from similar cities outside of China that did not have lockdowns at the same time. Based on this analysis, AOD changes in 2020 due to the lockdown in Wuhan, Beijing, and Shanghai were −18.1%, 13%, and −12%. Though AODs decreased in 2020 for Wuhan and Shanghai, the magnitude (−12%) is smaller than the value estimated from the tropNO 2 filter method (−21%) and the Thiel-Sen slope method (−23%). For Beijing, the AOD change estimated using SCM is consistent with the other two methods.
To test if our methods work for the opposite scenario where economic activity increases, we applied our methods to Taiwan, which did not use lockdowns to curb the spread of the COVID-19 pandemic but instead used isolation from the rest of the world. During the pandemic, Taiwan invested nearly one trillion Taiwanese Dollars ($T) into its economy, which grew by 2% in the first quarter, and due to which its exports to China increased by 10.3% in May 2020 compared to May 2019 (www. brookings.edu/blog/order-from-chaos/2020/06/29/ taiwan-faces-a-changed-economic-outlook-in-asiafollowing-covid-19/). When AODs in the northern, central, and southern Taiwan regions were analyzed using NO 2 F4AOD method, we found that AODs in 2020 increased by 1.5%, 11.4%, and 12.1% for northern, central, and southern Taiwan respectively. The AOD changes estimated from the Thiel-Sen slope method also showed increased AODs but slightly smaller in magnitude: 1.5%, 2.3%, and 7.9% for northern, central, and southern Taiwan respectively. Figure S3 shows ∆AOD 2020-2019 for Taiwan for February with and without NO2F4AOD method that shows how AODs increased in 2020 due to economic activity stimulated by new investments.
The premise of this study is based on the assumption that tropNO 2 and AOD co-vary in polluted regions across the globe. This assumption has been tested two different ways. First, by analyzing relationship between tropNO 2 and AOD for 2019 and 2020 for 43 cities when pandemic related lockdowns were not a factor. Figure S4 shows that our assumptions hold with the relationship between tropNO 2 and AOD having a positive linear correlation with a smaller slope when airmasses are dominated by urban pollution whereas when smoke plumes are present, the linear relationship has a much steeper slope. Second, by showing a map of how tropNO 2 and AOD co-vary across China between 2020 and 2019 for 10 February to 25 February (figure S5). The tropNO 2 and AOD variability is consistent with known knowledge of regions with reduced NO x emissions from transportation sector. Similarly, we found the regions where AOD increased and tropNO 2 decreased to be those influenced by smoke from fires.
Because transportation sector emissions were the most impacted during the lockdowns across the globe, this analysis demonstrates that targeted lowering of NO x emissions could have a benefit on lowering particulate pollution in favorable meteorological conditions. This can potentially be achieved by transitioning gasoline-based vehicles to electricity, for example. Using the tropNO 2 to on-road NO x emissions relationship (∆tropNO2 = 0.33 * ∆NO x + 2.33) and the AOD to tropNO 2 (AOD = 0.0022 * tropNO 2 + 0.102) relationship for the Los Angeles area reported in Kondragunta et al (2021), we estimate that 12 million gasoline-based vehicles must be removed from the road on any given day to see the 21% improvement in AOD that was observed across the US during the lockdown. This analysis did not include VOC emissions reductions because we do not have data on the relationship between VOCs and AOD. However, the contribution of NO x and VOC emissions to PM 2.5 is 1:1.1 according to the EPA (www3.epa.gov/ttnchie1/ conference/ei13/mobile/hodan.pdf). Considering a targeted NO x and VOC emissions reductions from light duty vehicles, the number of vehicles that need to transition to electricity comes down to 6 million. There are currently only 2.6 million electric vehicles in the US compared to 109 million gasoline powered cars (www.anl.gov/esia/light-duty-electricdrive-vehicles-monthly-sales-updates; dot.gov) that emit ∼486 tons of NO x per day. Transitioning to electric vehicles to reduce particulate pollution appears to be a lofty target because despite the rise in electric car sales, the transition is not fast enough, and it is uneven across the US, with some states doing better than others in electric car sales (https://afdc. energy.gov/data/10962). Emissions of NO x from gasoline powered light duty vehicles (cars) are the lowest (∼0.157 g/mile) whereas diesel powered heavy duty vehicles' NO x emissions are the highest (∼3.518 g/mile). Targeting the transition of diesel operated vehicles to bio-diesel or electricity could be a better policy for effective particulate pollution reductions in the urban areas.

Conclusions
The COVID-19 pandemic provided an opportunity for the scientific community to study the influence of changes in economic activity on anthropogenic emissions that lead to harmful pollution and to determine if those changes can be detected in satellite data. In this study, we specifically focused on capturing changes in AOD using satellite data on monthly time scales. To account for AOD data that are influenced by non-anthropogenic sources such as biomass burning, we developed a NO 2 F4AOD technique. Using tropNO 2 data to filter AOD data, we isolated the changes in AOD due to changes in reduced emissions from the transportation sector caused by COVID-19 pandemic related lockdowns; overall, in 37 of the 43 cities that were identified as top NO x emitters from the transportation sector, AODs decreased by 21.2% ± 7.8%, 18.9% ± 11.7%, 27% ± 12.4%, 22.9% ± 7.6% in the United States (7 cities), India (8 cities), western Europe (6 cities), and China (16 cities), respectively-a mean decrease of 22.4% ± 9.4%; when the six cities with increased AOD are included, the mean decrease is 16.5%. These results are consistent with our hypothesis that when tropNO 2 decreases AOD decreases and vice versa except when gains of improved air quality due to reduced NO x emissions were offset by emissions of aerosols and aerosol precursors from fires; in cities such as Nanning, Guiyang, and Kunming in China that were downwind of major fires; transported smoke increased AODs by up to 23%.
The magnitude of AOD changes were found to be minimally sensitive to the spatial averaging domain or the reference data used. When AOD in 2020 was compared to reference climatology using time series analysis, the mean AOD decrease is found to be 18%, which is similar to the 22% decrease found using NO 2 F4AOD method.
Our findings are consistent with many other studies that either analyzed surface PM 2.5 or AOD data to understand the impact of lockdown on particulate pollution. Specifically, Hammer et al (2021) used global model simulations to study the same geographic regions that we studied. Unlike tropNO 2 that is easy to observe in urban/industrial areas near source regions due to the short lifetime of NO 2 , aerosols are long-lived and are transported long distances. Therefore, any analysis of AOD data should account for contributions from non-anthropogenic sources such as dust and smoke. Using the species-tospecies relationship between tropNO 2 and AOD, we were able to isolate true AOD changes due to lockdowns except in cities that were in close proximity to fires.
We demonstrated that the methods we developed are robust and have the capability to detect and track changes in AOD due to changes in economic activity. When we applied our NO 2 F4AOD filtering technique to study AOD changes in Taiwan, we observed a mean 11.7% ± 8.4% increase in AOD across northern, western, southern Taiwan where most of its population and economic activity are concentrated; these AOD increases were expected based on Taiwan's strategy of using economic stimulus and not a lockdown during the COVID-19 pandemic. Going forward, our methods can be successfully applied to track economic changes on aerosols, on global to regional scales, such as power grid and roadway shutdowns associated with hazardous weather conditions (e.g., hurricanes, ice storms, etc). Analysis of these current and past events will provide insights into how effective pollution control strategies can be developed. It is estimated that reducing 6 million light duty vehicles from the US roadways can lead to 21% improvement in particulate pollution.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: http:// noaa-jpss.s3.amazonaws.com/index.html#SNPP/ VIIRS/SNPP_VIIRS_Aerosol_Optical_Depth_EDR_ Reprocessed/.