key: cord-0811772-i8ewy33w
authors: Vadrevu, Krishna Prasad; Eaturu, Aditya; Biswas, Sumalika; Lasko, Kristofer; Sahu, Saroj; Garg, J. K.; Justice, Chris
title: Spatial and temporal variations of air pollution over 41 cities of India during the COVID-19 lockdown period
date: 2020-10-06
journal: Sci Rep
DOI: 10.1038/s41598-020-72271-5
sha: 704e6b72d28d476934be8bd55c78433ff9624e49
doc_id: 811772
cord_uid: i8ewy33w

In this study, we characterize the impacts of COVID-19 on air pollution using NO(2) and Aerosol Optical Depth (AOD) from TROPOMI and MODIS satellite datasets for 41 cities in India. Specifically, our results suggested a 13% NO(2) reduction during the lockdown (March 25–May 3rd, 2020) compared to the pre-lockdown (January 1st–March 24th, 2020) period. Also, a 19% reduction in NO(2) was observed during the 2020-lockdown as compared to the same period during 2019. The top cities where NO(2) reduction occurred were New Delhi (61.74%), Delhi (60.37%), Bangalore (48.25%), Ahmedabad (46.20%), Nagpur (46.13%), Gandhinagar (45.64) and Mumbai (43.08%) with less reduction in coastal cities. The temporal analysis revealed a progressive decrease in NO(2) for all seven cities during the 2020 lockdown period. Results also suggested spatial differences, i.e., as the distance from the city center increased, the NO(2) levels decreased exponentially. In contrast, to the decreased NO(2) observed for most of the cities, we observed an increase in NO(2) for cities in Northeast India during the 2020 lockdown period and attribute it to vegetation fires. The NO(2) temporal patterns matched the AOD signal; however, the correlations were poor. Overall, our results highlight COVID-19 impacts on NO(2), and the results can inform pollution mitigation efforts across different cities of India.

It is well known that air pollution in several regions of the world is due largely to human activities, such as from fossil fuel combustion from motor vehicles, industries, power plants, etc. With the COVID-19 pandemic, a reduction in pollution has been reported by several researchers in different regions of the world such as Italy, the USA and Spain [2] [3] [4] [5] . As a result of the COVID-19 pandemic, during March 25th-May 3rd, 2020 entire India was lockdown. As a result, there was reduction in pollution in Indian cities too 6, 7 . However, the specific amount of pollution decrease is not well-documented covering multiple cities in India, hence the focus of this study. Some of the metropolitan cities such as New Delhi, Bangalore, Mumbai in India are renowned for its air pollution. Since, cities are hotspots of air pollution, we focused on 41 cities based on their population size and analyzed how the air pollution varied during the lockdown period as compared to the previous year as well as the pre-lockdown period. We addressed the following questions: (a) How much was NO 2 pollution reduced during Phase-1 and 2 of the COVID-19 full country lockdown (March 25-May 3rd, denoted here as 2020-lockdown)? (b) Specifically, how did NO 2 in the 2020-lockdown compare to the same period in 2019, when there was no lockdown (denoted as 2019-no lockdown)? (c) How did NO 2 levels during the 2020-lockdown compare with January-March 24th 2020 (denoted here as 2020-pre-lockdown)? (d) Were the differences in NO 2 pollution reduction consistent across 41 cities? (e) Which cities had the highest and least reduction in NO 2 ? (f) Are there scaling effects in NO 2 levels in cities, i.e., based on the spatial distance to the city center? (g) What was the overall reduction in NO 2 for major cities across India and are the differences statistically significant? We addressed these questions using the remote sensing derived TROPOMI-NO 2 datasets and the MODIS Aerosol Optical Depth (AOD) data covering different cities in India. We focused on satellite-derived NO 2 only since the measurement algorithm is relative matured ones compared to the other gases. Also, adverse health effects of NO 2 include acute respiratory illness, decreased pulmonary function, asthma, lung cancer and cardiopulmonary mortality 8 ; thus, it is important to address spatial and temporal variations in NO 2 useful for pollution management and mitigation purposes.

We selected the 41 cities in India, based on 7 different categories ranked by population (Fig. 1) . Rank-1 cities have the highest population of 5.0 million or greater, and rank-7 cities have less than 50,000 people. A map of the 41 cities selected for the study is shown in Fig. 1 . The results of our analysis of the spatial and temporal variation in the pollution levels are presented for: (a) individual cities; (b) averaged results based on the city's population ranking; (c) the top-seven highest polluted cities; (e) cities in northeast India and (f) coastal cities.

We used the TROPOspheric Monitoring Instrument (TROPOMI) onboard the Sentinel 5 precursor (S5P), operated by the European Space Agency (ESA) 9 to assess the tropospheric NO 2 background levels. Sentinel-5 Precursor (Sentinel-5 P), launched on 13th October 2017, was the first Copernicus mission satellite and can measure several trace gases such as NO 2 , ozone, formaldehyde, SO 2 , methane, carbon monoxide, and aerosols. The resolution is for all gases with 3.5 × 7 km 2 , except for CO and CH 4 , which is 7 × 7 km 2 . The TROPOMI instrument contains three spectrometers that cover the ultraviolet-near infrared region with two spectral bands at 270-500 nm and 675-775 nm and one spectrometer that covers the shortwave infrared band. Relatively, TROPOMI has a higher resolution compared to its predecessor, OMI which has a ground resolution of 13 km × 24 km at nadir. The TROPOMI NO 2 retrieval algorithm utilizes the bands of the ultraviolet-near-infrared spectrometer 10 . The retrievals are based on the NO 2 DOMINO system which was previously used for OMI spectra 11 with additional improvements 10 . The NO 2 slant column density is retrieved using the differential optical absorption spectroscopy www.nature.com/scientificreports/ (DOAS) method and separated into stratospheric and tropospheric components using the information from the data assimilation system and separation based on the altitude dependent air mass factors based on the lookup table approach. The final product provides the tropospheric vertical column densities, which describes the vertically integrated number of NO 2 molecules per unit area from the surface to tropopause. The data can be accessed either through the Near-Real-Time (NRTI) stream, the Offline stream (OFFL), or the Reprocessing (RPRO) stream. NRTI data are available within three hours after data acquisition, whereas OFFL and RPROdata are available within a few days after acquisition 10 . In this study, we used the TROPOMI, near-real-time operational product 9 obtained via the Copernicus open data access hub (https ://s5phu b.coper nicus .eu). Independent validation by the S5P Mission Performance Center (MPC) and S5P validation team concluded that OFFL level 2 NO 2 data are in overall agreement with reference measurements collected from global ground-based networks [12] [13] [14] . In addition to TROPOMI NO 2 , we also used the MODIS product MCD19A2.006: Terra and Aqua Multi-angle Implementation of Atmospheric Correction (MAIAC) Land Aerosol Optical Depth (AOD) gridded Level 2 product, specifically, the blue band (0.17um) 1-km daily data over land 14, 15 for our study. All processing was done using the QGIS software (3.10) "QGIS.org (2020). QGIS Geographic Information System. Open Source Geospatial Foundation Project https ://qgis.org/".

To generate time series of NO 2 columns over 41 different cities, we first selected pixels from an overpass area, defined by a different buffer radius (30, 45, 60, 75, 90 in km from the city center). We used data for which the quality assurance value is higher than 0.5 and the cloud fraction within the NO 2 retrieval window is below 40% 6 . The averaged tropospheric NO 2 column for each city within the buffer radius is calculated as, paired t-test. We used the paired t-test 16, 17 to compare the mean differences between NO 2 pollution levels during different months for the previous (2019) and the current year (2020). The t-test follows a Student's t-distribution under the null hypothesis of H0 that the means are equal, H 0 : µ 1 = µ 2 with the alternative hypothesis that H a : µ 1 ≠ µ 2 . The p-value is used to reject or accept the null hypothesis. The H 0 hypothesis was discarded when the p-value was less than 0.05 (significance level of 5% in this study) and the H a hypothesis is accepted 18 .

Autoregressive moving average model with intervention. We used the univariate autoregressive moving-average (ARMA) analysis 19, 20 with the intervention 21, 22 to quantify the impacts of COVID-19 on the pollution levels. Specifically, ARMA models are developed as linear functions of NO 2 values with the random shocks or errors based on the lockdown dates. The main difference between the ARMA model and ARIMA model is the integral part of the latter, i.e., a measure of how many nonseasonal difference values are needed to obtain stationarity. Thus, if no differencing is involved, then the model becomes ARMA. In this study, we implemented the ARMA modeling framework in three important steps 8 (a) identification of the model; (b) estimation of the coefficients and (c) verification of the model. All these steps are implemented in an iterative process, resulting in a number of tentative models. First or second-order differencing (nonseasonal and/or seasonal) is useful for the non-stationary means. The identification of the number of terms to be included in the ARMA model is based on the analysis of the autocorrelation (ACF) and partial autocorrelation (PACF) functions of the differenced time series data. The model coefficients were estimated by means of the maximum likelihood method. Also, the verification of the model is performed through diagnostic checks of residuals through the normal probability plots and standardized residuals. Finally, Akaike's Information Criteria (AIC) and Log-likelihood criterion were used to establish the model fit.

The intervention analysis helps to determine whether an event affects a timeseries of data, the known source and timing of intervention due to COVID-19, and the datasets in our case are is 2020-pre lockdown (January 1st-March 24th, 2020) versus 2020 lockdown period (March 25th-May 8th, 2020). The basic ARIMA model is given as (Eq. 1), when the intervention-free time series Zt follows the ARIMA (p = autoregressive parameter or the number of lag observations included in the model, also called the lag-order; d = the number of times the raw observations are differenced or degree of differencing and q = size of the moving average window) × (P,D,Q)s (pre-intervention) model with the seasonal period of S, an external shock, mt, has an additive impact. Zt is the time series before the COVID outbreak and mt is the function indicating the impact of the outbreak.

In the above equation, Y t includes the intervention,

is the seasonal MA polynomial, and a t is the white noise WN (0, σ 2 ). As mentioned, we used only ARMA model in our study.

Further, following the Box and Jenkins 19 , the intervention effect m t (due to COVID-19 in our case) can be calculated as either with the pulse function P The pulse function is generally used when a certain event happens at time T, and its effect is limited (Eq. 2), whereas step function is used when the event is continuous after T (Eq. 3). In the above calculation, an indicator function either a unit step or a unit pulse 20 are transformed by an AR(1) process with a parameter delta, and then scaled by a magnitude which is the coefficient on the transformed indicator function 21, 22 . Thus, the model can www.nature.com/scientificreports/ represent changes that are abrupt and permanent (step function with delta = 0, or pulse with delta = 1), abrupt and non-permanent (pulse with delta < 1), or gradual and permanent (step with delta < 0). The algorithm is based on the ARMA transformation and linear regression to find the magnitude. We tried a step function, as our data fits such a context (with COVID-19 impacts on NO 2 reduction, which continued from March 25th to April 3rd) to arrive at the smallest standard error on the magnitude 22 with the ARMA intervention analysis 23 .

The ARMA results are reported for before and after the intervention for seven dominant cities where NO2 pollution was most evident. Fig. 2 and 2020 (January 1st-March 24th) pre-lock down versus 2020 COVID lockdown period, is shown in Fig. 3 . Also, details for each city for the mean tropospheric NO 2 variations for 2020 pre and post-lockdown periods is provided in Supplementary Materials. To infer the data quality, we used the violin plots ( Fig. 4 ) that combine the basic summary statistics of a box plot with a kernel density plot. In the violin plots, the thick black bar in the center represents the interquartile range, the central white dot represents the median value, and the whiskers show a 1.5 × interquartile range (IQR) in the rest of the data. On each side of the black line is a kernel density estimation to show the shape of the data distribution. The wider sections of the violin plot represent a higher density of observations, and the skinnier sections represent lower density. Thus, for example, for the 2020 lockdown data, the violin is thicker in the center, suggesting that most of the values had consistently higher frequency around the median. In contrast, 2019-January and 2019-February data had relatively higher tapering ends with elongated distribution of the data compared to the other plot. Further, a clear decrease in the median NO 2 value can be seen, i.e., in general, the NO 2 pollution during the 2020 pre-lockdown period was considerably less for all 41 cities compared to 2019. The IQR is a measure of variability; thus, for 2019-January data, it is higher, suggesting the NO 2 values are more spread out from the median value compared to the 2020 lockdown period (Fig. 4) . While we don't intend to quantify drivers of these variations, there are many complex interacting factors like transportation, industry, biomass burning, etc., that might have affected these values.

The paired t-test was quite useful to infer the statistical significance between the two datasets for different months of 2019-no lockdown versus 2020 lockdown. For example, the results from January-2019 versus January-2020 mean tropospheric NO 2 for all 41 cities suggested an overall reduction by 11%, and the results from the paired t-test were statistically different (January 2019, M = 3.05e+15, SD = 2.53e+15) versus (January 2020, M = 2.57e+15, SD = 1.90e+15); t(40) = (4.21), p = 0.0001. Since the p-value is less than 0.05 (significant level of 95%), we rejected the null hypothesis and accepted the alternate hypothesis that mean differences between the two independent data exists, suggesting a decrease in pollution.

The results from February-2019 NO 2 versus February-2020 NO 2 suggested an overall reduction of 8%, and the results from the paired t-test were statistically different (February 2019, M = 2.75e+15, SD = 2.13e+15) versus (February 2020, M = 2.43e+15, SD = 1.63e+15); t(40) = (2.992), p = 0.0047. Since the p-value is less than 0.05 (significant level of 95%), we rejected the null hypothesis and accepted the alternate hypothesis that mean differences between the two independent data exists, suggesting a decrease in pollution. , p = 0.000. Since the p-value is less than 0.05 (significant level of 95%), we rejected the null hypothesis and accepted the alternate hypothesis that mean differences between the two independent data exists, suggesting a decrease in pollution.

Analysis of data between March 25th to May 2019 (no-lockdown period) versus March 25th to May 3rd 2020 (COVID lockdown period) suggested an overall NO 2 reduction of 19% and the results are statistically different (2019 no lockdown, M = 2.45e+15, SD = 1.38e+15) versus (2020 lockdown, M = 1.74e+15, SD = 5.74e+14); t(40) = (4.393)p = 0.0001. Since the p-value is less than 0.05, we rejected the null hypothesis and accepted the alternate hypothesis that mean differences between the two independent data exists. In summary, these results clearly suggest a statistically significant reduction in NO 2 pollution during the 2020 for different months and the lockdown period. (Fig. 5) . The mean reduction of NO 2 in these seven cities during the 2020 lockdown period is 50.27%. Further, we also calculated the NO 2 variations during the 2020-pre lockdown versus 2020 lockdown and found an almost 50.47% reduction during the 2020 lockdown period. We also did a buffer analysis to infer the spatial scaling effects on the reduction in NO2 for different cities. From the city center based on the latitude and longitude, the mean tropospheric NO 2 were analyzed at varying Fig. 6 . Of the different cities, New Delhi and Delhi had the most NO2 reduction (61.6% and 60.2%), followed by Bangalore (48.2%), Ahmedabad (46.70%), Nagpur (46.20%), Gandhi Nagar (45.5%) and Mumbai (43.1%) respectively. Further, for all the cities, as the distance from the city center increased, the NO2 pollution decreased exponentially. Further, of all cities, the highest decrease was noted for Ahmedabad, followed by Gandhi Nagar, Mumbai, etc. (Fig. 6) . We attribute the differences to land cover variations within the city including local meteorology impacting pollution in these cities.

Time series analysis. The time series plots for 2020 data from January 1st-May 3rd, 2020, for all the seven cities are shown in Fig. 7 . In the figures, data are represented as black lines with the standard errors in the shaded area along with the regression line in red color and 95% confidence bands in orange. In addition, trend characteristics of the slope, intercept, and Pearson's R is also given for each plot. Thus, for example, both New Delhi and Delhi showed the highest Pearson's R of − 0.72, followed by Mumbai (− 0.68), Ahmedabad (− 0.61), etc., and least for Nagpur (r = − 0.32). The slope was larger for New Delhi, followed by Delhi, Mumbai, Ahmedabad, etc., and least for Bangalore (Fig. 7) . The ARMA modeling was based on the before COVID (pre-intervention) from January 1, 2020-March 24th, 2020 versus during COVID (post-intervention) from March 25th-May 3rd, 2020. Dividing the entire time series data into two sets helped us to assess the magnitude of differences in NO2 pollution separately. The time series plots for the seven cities before and during COVID intervention is shown in Fig. 8a,b. In the figures, the blue line represents the data, the red line represents the residuals, and the green line represents the intervention. The various AR models fitted for different cities before and after COVID-19 are shown in Tables 1 and 2. An AR(1) www.nature.com/scientificreports/ autoregressive process is one in which the current value is based on the immediately preceding value, while an AR(2) process is one in which the current value is based on the previous two values. An AR(0) process is used for white noise and has no dependence between the terms. Results suggested a clear decline in NO 2 pollution (green line in the plots) due to COVID-19. For all cities, the pre-intervention data, too, showed a reduction in NO 2 ; however, the reduction was much higher during post-intervention as reflected in the ARMA coefficients. Thus, in all our AR models, the AR coefficients were negative for both pre-and-post intervention COVID data. In particular, for the post-intervention COVID dataset, the coefficients were highly negative and are below 1, suggesting that the NO2 reduction is highly persistent. The size of the moving average window for different cities varied from 0 to 1 for post-intervention COVID data and 1 to 2 for pre-COVID data. Specific to the model performance or measure of goodness of fit, either log-likelihood or AIC can be used. We used both the indicators to assess the consistency in the model performance. The higher the value of Log-likelihood, the better the fit of model coefficients. Thus, for example, post-intervention COVID data consistently had higher values compared to pre-intervention COVID datasets. In contrast to the Log-likelihood estimator, the lesser the AIC value, the better the model performance. Thus, a closer examination of Table ( 1) AIC values suggests that for most of the postintervention COVID data, the AIC values are much lower than the pre-intervention COVID data, suggesting higher performance. Both the Log-likelihood and AIC criterion suggested relatively higher model performances for the post-intervention COVID data. Further, for both for the pre-and-post intervention COVID data, both the Log-likelihood and AIC values showed consistency in the order of model performance for different cities. For example, for the pre-intervention COVID data, the Log-likelihood values were higher for Nagpur, followed by New Delhi, Delhi, Bangalore, Gandhinagar and Ahmedabad (Table 2) ; the AIC followed a similar order with lower values. For the post-intervention COVID data (Table 2) , the Log-likelihood ratio estimator showed higher values for Ahmedabad, Delhi, Nagpur, Gandhinagar, Bangalore, Mumbai, and New Delhi and AIC followed a similar order with lower values. The intervention effect can also be assessed in terms of magnitude for both preand post-intervention COVID datasets. For both the pre -and post-intervention datasets, the magnitude was negative, suggesting a decrease in pollution as time progressed; however, the values were more negative for the post-intervention COVID data compared to the pre-intervention COVID data. Thus, for the pre-intervention COVID data, a higher reduction in NO 2 pollution can be seen for New Delhi, followed by Delhi, Ahmedabad, Mumbai, Gandhinagar, Nagpur, and Bangalore. For the post-intervention COVID data, a higher reduction in NO 2 pollution can be seen for New Delhi, Delhi, Mumbai, Ahmedabad, Gandhinagar, Nagpur, and Bangalore. Further, except for Bangalore, where the NO 2 reduction was relatively higher for pre-intervention COVID data, for all the other cities, the post-intervention COVID NO 2 reduction was higher than the pre-intervention COVID datasets. In summary, the ARMA with intervention analysis helped to assess the data in a much more robust way for assessing the pre-and-post intervention COVID related NO 2 reduction. www.nature.com/scientificreports/ whereas the NO 2 data represents the data for only the troposphere. Despite these differences, both the datasets showed overall decreasing mean concentrations during the 2020 lockdown period, and the temporal patterns matched for the specific dates.

Variations in no 2 based on the population rank. Results from the 2019 no-lockdown period versus 2020 lockdown period for various cities based on the population ranks are shown in Table 3 . Various population rank categories are as follows: Rank-1 (5.0 million and greater); Rank-2 (1.0 to 4.9 million); Rank-3 (500,000-999,999); Rank-4 (250,000-499,000); Rank-5 (100,000-249,000); Rank-6 (500,00-99,999); Rank-7 (< 50,000). Thus, for Rank-1 population cities, the mean reduction in NO 2 was 51%, Rank-2-30%, etc. In the case of Rank-5 and Rank-6 cities, there was an increase in pollution of 3% and 22%, respectively. We attribute the differences to the geographical location; for example, most of the cities (not all) in these two ranks are coastal with dominant wind and sea breeze influences, compared to the other cities.

Variations in no 2 in northeast Indian cities. In contrast to other cities, northeast Indian cities had an almost 24% increase in NO 2 levels during the 2020 lockdown period compared to the 2019 no-lockdown period during similar dates. Also, a comparison of NO 2 levels for 2020 pre-lockdown versus post lockdown suggested an average NO 2 increase of 36% during the 2020 lockdown period for the cities in northeast India (Fig. 9a,b) . Our preliminary analysis of VIIRS active fire data suggests that an increase in NO 2 levels may be due to vegeta- www.nature.com/scientificreports/ tion fires, which increased during the 2020 lockdown period compared to 2019 during non-lockdown periods, especially in areas around the cities of Imphal, Dispur, Kohima, Shillong and Agartala (Fig. 10) . A detailed analysis of NO 2 increase in relation to vegetation fires using daily datasets is ongoing.

Variations in no 2 in coastal cities. Coastal cities had almost 22% NO 2 reduction during the 2020 lockdown period compared to the 2019 no-lockdown period with the highest in Mumbai and Kolkata (Fig. 11a) . Also, a comparison of NO 2 levels for 2020 pre-lockdown versus post lockdown suggested an average NO 2 reduction of 30% during the 2020 lockdown period with the highest reduction in Mumbai with a 43.08% reduction (Figs. 5 and 11b) . However, the overall NO 2 reduction during the 2020 lockdown period is relatively lower for the coastal cities compared to the top six non-coastal cities of New Delhi (61.74%), Delhi (60.37%), Bangalore (48.25%), Ahmedabad (46.20%), Gandhinagar (45.64%), and Nagpur (46.13%).

Ground-based measurements. We obtained the ground-based NO 2 measurement data (µg/m 3 ) for fifteen different cities of the total 41 cities of our current focus from the Central Pollution Control Board (CPCB) 24 , India. Additional data for the other cities that matched our currently studied cities including spatial and temporal data from the CPCB were not available. 68) . Further, a comparison of the March 2020 values for these cities suggested an 18% reduction due to COVID-19 lockdown. These results also match closely with the reduction in NO 2 reported for some of the cities using ground-based measurements 6, 7 . Also, correlating the TROPOMI tropospheric NO 2 data for the April lockdown period suggested a Pearson (r) of 0.33. The poor correlation can be attributed to the satellite data resolution aspects (3.5 × 7 km 2 ), compared to the ground station data footprint which might be much smaller than the satellite footprint. In addition, we infer that more ground station data at both spatial and temporal scales is required to validate the satellite data.

An overview of the results suggests significant differences and patterns in NO 2 and AOD which are briefly highlighted. India has four climatological seasons 25 , Winter (December-February), Summer or Pre-monsoon (March-May), Monsoon or rainy season (June to September) and Post-monsoon or autumn season (October-November). Thus, the 2020 lockdown period mostly occurred during the Summer or pre-monsoon season.

In general, most of the cities in the northern part of India see elevated pollution during the post-monsoon season due to the combined effect of anthropogenic and atmospheric factors. For example, in states of Punjab and Haryana, important sources of pollution include agricultural residue burning, industrial and vehicular emissions, dust storms, burning of solid fuels for heating, etc., which cause elevated pollution levels not only in these states but also the neighboring capital city, New Delhi 26 . In addition, during the post-monsoon season, due to the temperature inversion, there is less dispersion of pollutants resulting in smog events. In contrast, during the summer, the dispersion of pollutants is relatively higher compared to the post-monsoon season; the warmer air is lighter and rises upwards more easily carrying the pollutants away from the land surface and mixes the pollutants with the clear air in the upper layers of the atmosphere 27,28 , resulting in lesser concentrations. In addition to the summer effect, due to the COVID-19 lockdown, we found a significant reduction in pollution in major metropolitan cities. We found several variations, with some cities having more reduction in NO 2 than others in northeast India which experienced an increase in pollution due to the fires during the COVID-19 lockdown period. More thorough research is needed to understand the fire phenomenon, emissions and meteorology using the daily datasets. Also, in the coastal areas, the impact of sea and bay breezes on air quality, including air-mass transportation studies needs to be examined to address the spatial and temporal variations. We also infer the need to validate satellite measurements with the ground-based measurements. Our results on NO 2 reduction during the COVID-19 lockdown period match with the other studies conducted for some of the cities using the ground-based measurements and CPCB data from India.

Overall, this study focused on COVID-19 impacts on NO 2 pollution. Our results suggested a significant reduction in NO 2 during the lockdown period for most of the cities of India, except those located in Northeast India. The results from the study include variation in NO 2 based on geographical location, population ranks, distance from the city center, and robust statistical tests to determine the significance of a change in 41 different cities. Interestingly, we found notably higher vegetation fires during the lockdown period in Northeast Indian cities which warrant further investigation. The adverse effects of NO 2 pollution are well known in the literature. For example, higher doses of NO 2 can cause respiratory ailments 8 . Also, NO 2 and other oxides of nitrogen can react with water, oxygen and other chemicals to form acid rain which can be harmful to fish and other wildlife. The acid rain can washout nutrients and minerals from the soil damaging the crops and vegetation including damage to buildings and structures. Considering these detrimental effects, it is important to arrive at effective NO 2 pollution abatement strategies. Specific to the pollution abatement, the issue of spatial scale is increasingly being realized. Our results reveal greater variations in terms of NO 2 with some cities having the highest reduction compared to the other based on location and also variations based on the distance to the city center. Thus, policies to mitigate air pollution can be framed based on the local pollutant variations, needs and priorities. The spatial NO 2 variations highlighted in 41-different cities in our study can serve as a benchmark to address such variations and can help decision-makers to arrive at efficient air quality management plans involving local stakeholders. Although a temporary lockdown in emissions due to COVID-19 is a minor reduction in the overall pollution footprint, the current situation provides some useful insights on how policies like mandatory lockdown can have a measurable positive impact on the pollution control. As the economy reopens, the emissions will rebound, however, some of the policies such as working remotely could keep emissions under control post-COVID-19 situation. We also infer a strong need for a political will and social interventions to curb pollution beyond COVID-19 in India.

Role of the chronic air pollution levels in the Covid-19 outbreak risk in Italy

Correlation between environmental pollution indicators and COVID-19 pandemic: a brief study in Californian context

Lockdown for CoViD-2019 in Milan: what are the effects on air quality

Indirect effects of COVID-19 on the environment

Influence of air pollution on coronavirus (COVID-19): some evidences from studies at AUH, Gurugram

Effect of lockdown amid COVID-19 pandemic on air quality of the megacity Delhi

Synthesizing Qualitative and Quantitative Health Evidence, a Guide to Methods 330-331

TROPOMI on the ESA Sentinel-5 Precursor: a GMES mission for global observations of the atmospheric composition for climate, air quality and ozone layer applications

TROPOMI ATBD of the Total and Tropospheric NO 2 Data Products

An improved tropospheric NO 2 column retrieval algorithm for the ozone monitoring instrument

S5P Mission Performance Centre Nitrogen Dioxide

Quarterly Validation Report of the Copernicus Sentinel-5 Precursor Operational Data Products

Assessment of the quality of TROPOMI high-spatial-resolution NO 2 data products in the Greater Toronto Area

Angle Implementation of Atmospheric Correction (MAIAC) Data User's Guide

Essential Medical Statistics 2nd edn

Biostatistical Analysis 4th edn

Intervention analysis with applications to economic and environmental problems

Time series model specification in the presence of outliers

A fast algorithm for the exact likelihood of autoregressive-moving average models

Paleontological Data Analysis

Central Pollution Control Board (CPCB)

Vegetation fires, absorbing aerosols and smoke plume characteristics in diverse biomass burning regions of Asia

Radiative effects of pollutants in the atmospheric boundary layer

Nitrogen oxides in the troposphere: global and regional budgets

Authors are grateful to the Sentinel-5P and MODIS AOD product developers and for making them freely available. Authors thank Central Pollution Control Board, New Delhi, India, for providing the ground based NO 2 station data for various cities. The funding support received from the NASA Land Cover/Land Use Change Program for the South/Southeast Asia Research Initiative is greatly acknowledged.

K.P.V. conceived, analyzed data and wrote the manuscript. AE helped in parts of data processing and crosschecking. S.B. helped in manuscript formatting and editing. K.L., S.S., G.J.K. and C.J., provided important science suggestions to strengthen the manuscript, helped in refining and editing.

The authors declare no competing interests.

Supplementary information is available for this paper at https ://doi.org/10.1038/s4159 8-020-72271 -5.Correspondence and requests for materials should be addressed to K.P.V.Reprints and permissions information is available at www.nature.com/reprints.Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.