Unprecedented reduction in air pollution and corresponding short-term premature mortality associated with COVID-19 lockdown in Delhi, India

ABSTRACT Countries around the world introduced strict restrictions on movement and activities known as ‘lockdowns’ to restrict the spread of the novel coronavirus disease (COVID-19) from the end of 2019. A sudden improvement in air quality was observed globally as a result of these lockdowns. To provide insight into the changes in air pollution levels in response to the COVID-19 restrictions we have compared surface air quality data in Delhi during four phases of lockdown and the first phase of the restriction easing period (25 March to 30 June 2020) with data from a baseline period (2018–2019). Simultaneously, short-term exposure of PM2.5 and O3 attributed premature mortality were calculated to understand the health benefit of the change in air quality. Ground–level observations in Delhi showed that concentrations of PM10, PM2.5 and NO2 dropped substantially in 2020 during the overall study period compared with the same period in previous years, with average reductions of ~49%, ~39%, and ~39%, respectively. An overall lower reduction in O3 of ~19% was observed for Delhi. A slight increase in O3 was found in Delhi’s industrial and traffic regions. The highest peak of the diurnal variation decreased substantially for all the pollutants at every phase. The decrease in PM2.5 and O3 concentrations in 2020, prevented 904 total premature deaths, a 60% improvement when compared to the figures for 2018–2019. The restrictions on human activities during the lockdown have reduced anthropogenic emissions and subsequently improved air quality and human health in one of the most polluted cities in the world. Implications: I am submitting herewith the manuscript entitled “Unprecedented Reduction in Air Pollution and Corresponding Short-term Premature Mortality Associated with COVID-19 Forced Confinement in Delhi, India” for potential publishing in your journal. The novelty of this research lies in: (1) we utilized ground-level air quality data in Delhi during four phases of lockdown and the first phase of unlocking period (25th March to 30th June) for 2020 as well as data from the baseline period (2018–2019) to provide an early insight into the changes in air pollution levels in response to the COVID-19 pandemic, (2) Chatarize the change of diurnal variation of the pollutants and (3) we assess the health risk due to PM2.5 and O3. Results from ground-level observations in Delhi showed that concentrations of PM10, PM2.5 and NO2 substantially dropped in 2020 during the overall study period compared to the similar period in previous years, with an average reduction of ~49%, ~39%, and ~39%, respectively. In the case of O3, the overall reduction was observed as ~19% in Delhi, while a slight increase was found in industrial and traffic regions. And consequently, the highest peak of the diurnal variation decreased substantially for all the pollutants. The health impact assessment of the changes in air quality indicated that 904 short-term premature deaths (~60%) were prevented due to the decline in PM2.5 and O3 concentrations in the study period. The restrictions on human activities during the lockdown have reduced the anthropogenic emissions and subsequently improved air quality and human health in one of the most polluted cities in the world.


Introduction
The novel coronavirus disease COVID-19 caused by SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2), was first identified in Wuhan province in mainland China in December 2019. By March 2020, the disease had spread rapidly to other countries and the outbreak was declared a global pandemic on 12 March 2020 (WHO 2020). The COVID-19 disease has affected 110 million people and caused more than 2.45 million deaths worldwide as of 18 February 2021. The United States of America (USA), Brazil, Mexico, India, and the United Kingdom (UK) have experienced the greatest impact to date (18 February 2021), with death tolls of around 505,000, 243,000, 178,000, 156,000and 119,000, respectively (https://coronavirus. jhu.edu/).
On 30 January 2020, India reported its first COVID-19 case in Kerala, which rose to three cases by 3 February 2020, all were students returning from pandemic zone Wuhan, China. No significant rise in transmissions was observed in India in February. After observing the COVID-19 rising cases in Europe and the USA, the Indian government announced "Janta (People) Curfew" on March 22, 2020, a 14 -h selfquarantine curfew to maintain social distance (PIB 2020). Three days later, on 25 March 2020, the Indian government announced a strict lockdown for the entire nation with 1.3 billion citizens. After 68 days of lockdown, the first phase of the unlock process started on 1 June 2020 with some restrictions still in place. The long-term lockdown was not able to prevent the spread of COVID-19 in India. As of 18 February 2021, the total number of positive cases in India stands at 10.96 million, with 10.66 million recoveries and 156,000 deaths. Maharashtra, Tamil Nadu, Karnataka and Delhi reported 51,100,12,400,12,200 and 10,900 thousand deaths from COVID-19 respectively (https://www.covi d19india.org/).
COVID-19 has shown higher disease severity and death risk in patients with co-existing illnesses (comorbidities) (Cai 2020;Guan et al. 2020). In Italy, 60% of COVID-19 deaths occurred in people with hypertension (69%), type-2 diabetes (32%), chronic renal failure (21%) and ischemic heart disease (27%) (Michelozzi et al. 2020). The excess in mortality was higher among men than among women, with an increasing trend by age (Michelozzi et al. 2020). A meta-analysis of data from China, Italy, Spain, UK, and New York State by Bonanad et al. (2020), reported that the number of COVID-19 deaths among the infected population aged ≥60 years old was 12.6 times higher than for those aged <60 years. The number of COVID-19 cases and death rates tend to be higher in both high population density and high particulate matter (PM) exposure areas. The viral genome (SARS-COV-2 RNA) has been found on particulate matter (PM 10 and PM 2.5 ) which may have increased the transmission and spread of COVID-19 (Setti et al. 2020). PM 10 and PM 2.5 also suppress innate anti-viral immunity and enhance influenza virus replication via metabolic pathway gene modulation, amplifying the number of deaths from respiratory and cardiovascular disease in COVID-19 patients . For example, Zhu et al. (2020) reported a 0.22% increase in COVID-19 positive cases for each 1 μg/m 3 increase in PM 2.5 concentrations in China. Cole, Ozgen, and Strobl (2020) found that a 1 μg/m 3 increase in PM 2.5 was associated with an increase of 13.0-21.4% in the COVID-19 death rate in the Netherlands, larger than the 8% increase in death rate reported for the same PM 2.5 increase in the USA (Wu et al. 2020).
The COVID-19 forced restriction in all public activities, except for essential sectors, throughout the world. This resulted in lowered anthropogenic emissions and consequent appreciable reductions in gaseous and PM concentrations across cities worldwide (Bauwens et al., 2020;Adams 2020;Berman and Ebisu 2020;Menut et al. 2020), except for O 3 concentrations which increased in some places in the UK (Semple and Moore 2020; Zoran et al. 2020). Studies in India show an appreciable reduction in air pollutants associated with the COVID-19 lockdown restrictions, although most of the studies focus on the first phase of lockdown (25 March -14 April 2020) (Jain and Sharma 2020;Mahato, Pal, and Ghosh 2020;Singh and Chauhan 2020;Srivastava et al. 2020). A significant reduction was noted for PM 10 , PM 2.5 and NO x level by 44%, 21% and 51%, respectively, in Delhi during the "Janta Curfew" on March 22-23, 2020 compared to the previous day (Table  1). This sudden decrease in air pollution was brought about by the combination of reduced vehicles on the road, functioning of only essential commercial units and weather conditions.
The lockdown began in March which is early spring in northern India. Human activities generate the majority of aerosols in this area. Motor vehicles, coal-fired power plants, and other industrial sources around urban areas produce nitrates and sulfates, and coal combustion produces soot and other carbon-rich particles. Rural areas add smoke rich in black carbon and organic carbon from cooking and heating stoves, as well as smoke from the burning of crop stubble on farmland (though farming fires occur more often in late September and October each year) (NASA 2020).
This study aimed to investigate the impact of COVID-19 restrictions on air quality, during four phases of lockdown and the first phase of unlocking in Delhi, with respect to (a) the characteristics of PM 10 , PM 2.5 , NO 2 , and O 3 based on daily data in 2020 and compared with the similar periods in 2018-2019; (b) the diurnal variation of identified pollutants in from 2018-2020, and (c) quantification of health benefits due to the reduction of pollutants concentrations in Delhi.

Data description and methodology
Delhi is the second-largest megacity in the world and the largest urban agglomeration in India with an estimated population of 19.3 million in 2020 (https://www.cen sus2011.co.in/census/state/delhi.html). The present study on air quality during COVID-19 restrictions has focused on Delhi which is the administrative capital and the second major financial city of India. The lockdown was renewed four times [Lockdown Phase 1 (LDP-1): 25 March-14 April 2020; LDP-2: 15 April 2020-3 May 2020; LDP-3 (4 May 2020-17 May 2020; LDP-3: 18 May 2020 -31 May 2020; Unlock Phase 1 (ULP-1): 1 June 2020 -30 June 2020] after observing the number of cases at the end of each phase. Each lockdown phase had a different level of restriction on public activity (more details are reported in Table 2).

Materials used
Hourly air quality data from 34 continuous ambient air quality monitoring stations covering different regions of Delhi have been taken into consideration to assess the air quality during the four phases of the lockdown period and the first month of the unlocking period. The organizations responsible for these air quality monitoring stations include CPCB (Central Pollution Control Board); DPCC (Delhi Pollution Control Committee) and SAFAR (System of Air Quality and Weather Forecasting and Research) and IITM Pune (Indian Institute of Tropical Meteorology). Hourly time series data of four air pollutants, PM 2.5 , PM 10 , nitrogen dioxide (NO 2 ), and ozone (O 3 ) from 25 March to 30 June in 2018-2020, were downloaded from the CPCB online portal Central Control Room for Air Quality Management -Delhi NCR (https://app.cpcbccr.com/ccr/#/caaqm-dashboard/ caaqm-landing/data). Rigorous protocols for sampling, analysis and calibration are followed by CPCB to provide appropriate data quality assurance and quality control (QA/QC). Additional data criteria required for inclusion in this study were as follows (i) ≥ 80% hourly data capture for the period 25 March to 30 June; (ii) >12 h of available data in a day; (iii) spurious outliers in the data with z-scores exceeding an absolute value of 4 were removed and (iv) data below the detection limit of the measurement instruments were removed (Table S1). We observed that >93% of the hourly data at 11 monitoring stations was available for further calculation after our criteria and quality checks (Table 3 and Figure 1).
The daily average meteorological information was collected from OGIMET (www.ogimet.com), which uses freely available data from the National Oceanic and Atmospheric Administration (NOAA). Delhi typically has four seasons: winter (December-February), summer (March-May), monsoon (June-August) and post-monsoon (September-November) .

Methodology for quantifying air pollution changes
A differential approach was used to quantify air pollution changes coincident with COVID-19 during 2020 in comparison with the similar period in 2018-2019. For the March-June differential, we calculated daily average pollutant levels for March-June each year from 2018 to 2020. The differential was defined as the difference between 2020 values and the average of those for a 2year baseline (2018-2019). We searched the central and Delhi state government websites to identify the dates for different phases of the lockdown and the unlocking periods in Delhi. Table 2 reports the details of lockdown and unlock phases and their different levels of restrictions. A paired t-test was used to determine whether, on average, there was a change in average pollutant levels during each of the identified subperiods. We considered a p-value higher than α = 0.05, as not statistically significant (Table S4). The percentage variation of the average pollutant concentrations throughout the subperiods were examined. We also examined the role of different meteorological variables on the changes in air quality in Delhi during the period investigated.

Health impact assessment
We estimated short-term all-causes, cardiovascular diseases (CVD), ischemic heart disease (IHD), respiratory disease (RD) and stroke-related mortalities attributable to ambient PM 2.5 and O 3 -exposure at Indian capital from 2018-2020, using the loglinear exposure-response function, described in past studies (Seltzer, Shindell, and Malley 2018;Stanaway et al. 2018) as: • Power plants, household, diesel generator, sea salt and dust from trans-border.
• Goods vehicles, as all goods traffic was allowed to ply.
Lockdown-2 (15 April 2020 -3 May 2020 (19 days) (LDP-2) Except for the emergency sector, all sector remains closed and with a conditional relaxation for certain businesses. The possible source for air pollution: • Power plants, household, diesel generator, sea salt and dust from trans-border.
• Goods vehicles, as all goods traffic was allowed to ply.
• Industry, as some manufacturing units in industrial estates and industrial township, were allowed to operate with a 30-40% workforce.
• Brick kilns in rural areas started to operate. Brick kilns in surrounding Delhi are responsible for air pollution in Delhi city.
• Open burning from agriculture (maybe), like all agricultural and horticultural activity, was fully functional.
• Dust from construction activity where workers are available on site.
Lockdown-3 (4 May 2020 -17 May 2020 (14 days) (LDP-3) National lockdown with some relaxation in the day time. The movements of individuals, for all nonessential activity, was strictly prohibited between 7 pm to 7 am. The possible source for air pollution: • Power plants, household, diesel generator, sea salt and dust from trans-border.
• Vehicles: although some public transport was remain shut, private vehicles -four-wheelers and two-wheelers were ply. With the condition that public bus with 50% capacity and a maximum of 20 passengers, four-wheelers can carry two people and a driver and only one person on twowheelers.
• Industry, as all manufacturing industry in the urban area, was allowed to operate with a 30-40% workforce.
• Brick kilns in rural areas started to operate.
• Open burning from agriculture (maybe), like all agricultural and horticultural activity, was fully functional.
• Dust from all construction activity where workers are available on site.
Lockdown-4 (18 May 2020 -31 May 2020 (14 days) (LDP-4) National lockdown with some relaxation in the day time. The movements of individuals, for all nonessential activity, was strictly prohibited between 7 pm to 7 am, except the essential activity. The possible source for air pollution: • Same as Lockdown 3.
Unlock-1 (only for containment zones): 1 June 2020 -30 June 2020 (30 days) (ULP-1) All activity was normally operated. Although, the movements of individuals, for all non-essential activity, was strictly prohibited between 9 pm to 5 am, except the essential activity. Metro operations were remain suspended. The possible source for air pollution: • Power plants, vehicles, industry, household, brick kilns, open burning, diesel generator sets, sea salt and dust (from construction, soil, re-suspended and trans-border) where TMREL represents the theoretical minimum risk exposure level and ΔC is PM 2.5 or O 3 -exposure relative to TMREL. β is the exposure-response coefficient derived from the reported relative risk (RR), which links incremental changes in PM 2.5 or O 3 -exposure ΔX (10 μg/m 3 in average PM 2.5 concentrations or DMA8-h O 3 ). D 0 is the cause-specific death rate, obtained from   Table 2.
the GHDx database (http://ghdx.healthdata.org/gbdresults-tool). EP is the exposed population age ≥ 25 years and ΔMort is the estimated number of causespecific mortalities in a city. In this study, we used the average daily value of PM 2.5 and the average DMA8-h O 3 metric for short-term health risk analysis. We used TMRELs of 10 μg/m 3 for PM 2.5 (based on WHO guidelines) and 70 μg/m 3 for O 3 (as recommended in the HRAPIE project) (WHO 2013). In our PM 2.5attributed short-term mortality, β was estimated using adopted relative risk (RR) from a Chinese epidemiological study as no cohort study was available for India Yin et al. 2017) (S1.1).

Results and discussions
Overview of daily mean air quality levels Figure 2 and  (Figure 3). In Delhi, the significant contributors to annual PM 2.5 emissions include road dust (38%), vehicle exhaust emissions (20%), domestic fuel burning (12%), and industries (11%) (Sharma and Dikshit 2016). Previous studies of the Delhi megacity have reported that vehicular emissions (exhaust and non-exhaust) provide a significant contribution (up to 60%) to Delhi's PM 2.5 load . Only essential service (like police, hospital, army, and other emergency transport) vehicles were permitted during lockdowns phase 1 and 2, all commercial vehicular movements were restricted. This restriction resulted in a significant reduction in both exhaust and non-exhaust emissions, both significant contributors to the total PM load. Table 4 presents the mean, standard deviation and median of PM 2.5 concentrations at 11 monitoring sites during the four lockdown phases (i.e., LDP-1, LDP-2, LDP-3, LDP-4) and first-phase of unlocking (i.e., ULP-1), and for the same periods in 2018 and 2019. The details of PM 10 , NO 2 and O 3 concentrations in five phases in 2018-2019 and 2020 are reported in Table S4. Comparing data for the same seasonal period minimized the influence of confounding factors such as meteorological conditions. Pollutant concentrations (as % reduction) in 2020 compared with that of the previous two years in 11 monitoring stations in five phases are shown in Figure 4 to further demonstrate the impact of COVID-19 lockdown restrictions.
The decrease of NO 2 concentrations has shown considerable variation between the four-phases of lockdown [NO 2 : 49.2 ± 34.4% to 39.9 ± 35.6%], although the overall decline is lower in the unlock-phase (ULP-1), about 23.7 ± 42.1%. In Delhi, nearly 52% of NO x emission is attributed to industrial point sources (largely from power plants) followed by vehicular sources (36%) (Nandi 2018). In LDP, the decline rate of NO x is mainly due to full restrictions on vehicles, whereas in ULP, NO x is still lower than the past year as there were comparatively fewer vehicles on the road. In LDI-1, the highest decline for NO 2 was observed at the traffic sites JNS and MM, 81.3 ± 8.8% and 63.8 ± 5.9% respectively. In ULP-1 these sites still had lower NO 2 levels compared to the average daily values in 2018-2019 for the same period, although at some industrial sites, like RH and VV, NO 2 concentrations increased by 39.1% and 8.2% in ULP-1 (Table S4)   Road traffic was strictly restricted in the first two phases of lockdown, however, this was relaxed to some extent in the next lockdown phases. In ULP-1, pollutants concentrations were still lower compared to the same period 2019 and even in LDP-4 in 2020, due to the fear of increasing rates of COVID-19 cases in Delhi during ULP-1, visits to workplaces reduced by 60%, while retail and recreation activities reduced by 84% (Shrangi and Pillai 2020). Three coal-based thermal power plants that continued to operate throughout the lockdowns operated can be expected to have a similar contribution during the study period in 2020 and the equivalent period in 2018-2019. Their previous contributions were estimated to be 9.0%, 10.9% and 7.2% of PM 10 , PM 2.5 and NO x load for Delhi (Ministry of Heavy Industry and Public Enterprises 2018) (Table S6). Switching-off of the main PM 10 and PM 2.5 sources like transport, industries, agricultural burning, road dust, construction, restaurant and airport could be responsible for 80.4% and 72.8% reduction of PM 10 and PM 2.5 (Table S6), although during LDP-1 the PM 10 and PM 2.5 were reduced by only 61.1% and 48.0%, possibly due to the transport of PM into the city from surrounding states Haryana and Uttar Pradesh (Purohit et al. 2019). Additionally, these sources also emit secondary aerosol which could have contributed to the reduction of PM during lockdown . In lockdown phases, biofuel burning from residential areas for cooking purposes, such as LPG, wood and coal has increased (IANS 2020), and could have contributed to overall air pollution.
The COVID-19 restrictions reduced urban anthropogenic emission activities across India by different amounts, due to the different anthropogenic sources of pollutants and different meteorological conditions. In the first phase of lockdown, Mahato, Pal, and Ghosh (2020) reported a 60%, 39% and 53% reduction of PM 10 , PM 2.5 and NO 2 , respectively, compared to 2019 in Delhi. Reductions in PM 2.5 concentrations, based on ground-level monitoring data, was 35% in Kolkata and 28% in Delhi during 22-31 March 2020 . Sharma et al. (2020) reported a 43% decline of PM 2.5 from 16 March to 14 April, using a WRF-AERMOD modeling system, when compared with a similar period in previous years. Dhaka et al. (2020) observed that in the first week of lockdown (25-31 March 2020), PM 2.5 showed large reductions (by 40-70%) compared to the pre-lockdown conditions over the Delhi-NCR region (Table 1). In UK cities, the lockdown (23 March 2020) caused a sharp drop in NO 2 pollution (~60% after two weeks), however, no consistent reduction was seen in PM 2.5 over the same period. In the UK, PM 2.5 levels were higher in many areas during the UK lockdown than at any other time in 2020 to date (https:// www.bbc.com/news/science-environment-52113695).

Meteorological conditions and change of pollution levels
The variations in meteorological parameters in 2020 during the study period are summarized in Table 6 (Table S2 for 2018-2019). The month of March marks the beginning phase of summer in Delhi and the month of May is the peak of the summer season, therefore, from LDP-1 to LDP-4, the temperature has increased by ~8.3°C. The start of the monsoon rain in June brings down the temperature, where the mean temperature during ULP-1 was ~1.0°C less than the temperature during LDP-4. In inland city Delhi, a continuous decrease of relative humidity (RH) was reported from March to May (LDP-1 to LDP-4) as temperatures increased, peaking in June, primarily due to the beginning of monsoon rain. The opposite pattern to RH was observed for wind speed (WS) with the highest reported WS of 10.4 ± 3.4 km/h in LDP-4 in 2020.
No systematic correlations were observed between pollutants and weather parameters from March to June ( Figure S1). This was due to the interplay between two different weather and air pollution emission scenarios from March to June. The end of March represents the LDP-1, where all the services were closed and there were no active emission sources. However, June represents the ULP-1, where essential services were opened gradually. These scenarios lead to different emission patterns, which led to the inconsistencies observed in the correlation coefficients. Worthy of note is the negative correlation coefficient between RH and air pollutants, and the positive correlation coefficient between temperature and air pollution is increased in 2020, compared to 2018-2019. The higher variation in coefficient values between PM 2.5 and O 3 , (2018: −0.11; 2019: 0.38 and 2020: 0.17) may be due to the sensitivity of O 3 formation in different PM 2.5 concentration levels ( Figure S1). The sensitivity of O 3 formation in the Delhi city region is dominated by local traffic emissions, and O 3 and traffic emissions are anticorrelated. The response surfaces show that a reduction in local traffic emissions alone of 50% could decrease Delhi PM 2.5 loading by 15%-20%, but this would also increase O 3 by 20%-25%. To prevent the side effect of increasing O 3 , controls on traffic emissions would be required to reduce only by 25-30%, which also permits a further reduction of PM 2.5 by 5%-10% (Chen et al. 2020).
The meteorological variables were divided into five equal intervals between minimum and maximum during the total study period from 2018-2020 and it was observed that the different range of meteorological conditions influenced the reduction of air pollution in 2020 differently, compared to 2018 and 2019. For example, in the average temperature of 22.8, 26.7, 30.6, and 34.5°C, the average percentage change of PM 2.5 decreased by 62.2, 42.6, 37.9 and 24.3% respectively. The further increase of the atmospheric temperature to 38.4°C, when the lockdown was in the fourth phase the PM 2.5 reduction level increased slightly (28.9%), due to the highest temperature reported in May. Similarly, RH and WS have significantly influenced the change of pollutants concentrations in COVID-19 lockdown phases in 2020, compared to 2019 ( Figure 5). More details are reported in Figure S2 (supplemental material).

Diurnal variation in pollutant concentrations
Figures 6a, b show the average diurnal variation of PM 10 and PM 2.5 concentrations in five-phases from 2018 to 2020 in Delhi. The diurnal variation of PM 10 and PM 2.5 concentrations were largely characterized by a "W" type double wave. The morning peak occurred around 07:00 to 10:00, and an afternoon valley between 15:00 to 17:00. The peak in the night appeared after 20:00 or midnight and then gradually decreased in the early hours of the morning in 2018 and 2019. The higher concentrations of PM is consistent with the morning and evening rushhour traffic pattern and the afternoon dip was mainly attributable to a higher atmospheric mixing layer, which enhanced air pollution diffusion (San Martini, Hasenkopf, and Roberts 2015). During COVID-19 pandemic restrictions, overall PM concentrations decreased and similar diurnal variation was observed in LDP-1 to LDP-3. The morning peak was followed by a gradual decrease in PM concentrations through the afternoon. This feature may be explained in part by the growth of the mixed layer depths and stronger atmospheric ventilation during the afternoon. The morning peak might be associated with the fumigation effect in the boundary layer, which brings aerosols from the nocturnal residual layer shortly after the sunrise. As the day advances, increased solar heating leads to increased turbulent effects and a deeper boundary layer, leading to faster dispersion of aerosols and hence dilution of PM concentrations occurs near to the surface after 15:00 (lateafternoon) (Tiwari et al. 2013). This suggests that meteorological processes such as vertical mixing, advection and transport are the dominant factors controlling PM in the daytime. In contrast, freshly emitted pollutants are trapped at night when the planetary boundary layer (PBL) is shallow, and concentrations are very sensitive to the emission flux so that the diurnal pattern of emissions is the dominant factor at night (Chen et al. 2020). In LDP-4 a different diurnal shape of PM was observed in 2020, PM 10 concentrations continuously increased from 04:00 to 11:00 and for PM 2.5 the increase was observed between 04:00 to 08:00 and then gradually decreased in the afternoon. In ULP in June 2020, PM 10 concentrations continuously increased from morning 07:00 to 21:00 and then gradually decreased, whereas PM 2.5 concentrations continuously decreased from midnight to late afternoon.
Late evening hour (20:00-22:00) peaks of PM 10 , PM 2.5 and NO 2 were observed in Delhi ( Figure 6). This could be linked to the decrease in night temperature and boundary level height (Ravindra et al. 2021). Figure 6d shows the hourly average diurnal variation of O 3 concentrations during 2018-2020, which was opposite that of the other air pollutants with "Ո" shape. O 3 concentrations reach a minimum value before sunrise. From 06:00 to 08:00, coinciding with the morning peak traffic, the NO x concentrations increase rapidly in Delhi (Nandi 2018) and solar radiation is still weak during this period, which leads to a greater depletion of O 3 due to titration with NO. Around 09:00, along with increases in solar radiation and temperature, the photochemical reactions become more active and O 3 concentrations begin to increase, and they peak around 11:00-16:00. As time progress from LDP-1 to LDP-3, the peak of the diurnal curve in O 3 was increased where NO concentrations were decreased from LDP-1 to LDP-3 in 2020. Between 20:00 and 23:00, the O 3 concentrations decrease rapidly due to the decrease in solar radiation and titration with NO during evening peak traffic. With the decrease in NO 2 and CO concentrations in the afternoon, O 3 concentrations increased and it was suggested that the maximum O 3 concentrations in the afternoon were mainly due to photochemical reaction under intense solar radiation conditions, leading to the consumption of NO 2 and CO emissions (Shi et al. 2019).

Health benefit
In this study, we estimated short-term PM 2.5 and O 3exposure associated health risk only, as other pollutants concentrations are below the threshold value. The selected risk factor estimates for the Chinese population may not be applicable in India as, race, education background, marital status, food habit, alcohol consumption rate, cigarette-smoking status, socioeconomic status, body mass index (BMI) may be different. However, the range of ambient PM 2.5 and O 3 concentrations in Delhi is quite similar to those in the selected epidemiological study. The short-term PM 2.5 and O 3 -exposure are associated with the increase of premature deaths among adults (≥25 years) from all-cause of nonaccidental, cardiovascular disease (CVD), ischemic heart disease (IHD), respiratory disease (RD) and stroke. Based on the log-linear model, the estimated PM 2.5 and O 3 -related premature mortalities along with 95% confidence interval (CI), when the daily PM 2.5 and DMA8-h O 3 concentrations meet the threshold value in 2018-2019 and 2020. The averages short-term all-cause PM 2.5attributed deaths were 1,186 (95% CI: 811-1,505) in which 33%, 21%, 13%, 8% and 26% were due to CVD, IHD, RD, stroke and 'other cause' of deaths during the study period in 2018-2019. The all-cause premature death decreased to 603 (95% CI: 412-767) in 2020 and CVD, IHD, RD, stroke and 'other cause' of deaths decline to 17%, 11%, 6%, 4% and 13%, respectively. The DMA8-h O 3 attributed all-cause premature mortality was 321 (95% CI: 174-467) in 2018-2019, in which CVD, IHD, RD, stroke and 'other cause' of deaths were contributed by 30%, 16%, 29%, 9% and 17% respectively, and in 2020, the O 3 exposure-related deaths were zero as O 3 concentrations were below the threshold. Overall avoided premature deaths due to the decline of PM 2.5 and O 3 were about 60% in Delhi during the overall study period (25 March to 30 June) as compared to similar periods of 2018 − 2019. Bherwani et al. (2020) reported that reduction of PM 2.5 in LDP-1 alone is responsible for a 35.5% reduction of long-term all-cause total premature mortality compared to 2019. Venter et al. (2020) estimated that PM 2.5 -related reductions in mortality (short-term) burden were 5,300 (1,000-11,700) for India during the first two weeks of lockdown as compared to similar periods of 2017 − 2019. A similar study in Delhi by Kumar et al. (2020) reported a 49% reduction of PM 2.5 -attributed allcause death (short-term) during the first 47 days of the lockdown.
While there were reports of clear pollution-free skies and long-distance views to Himalayan mountains as a result of COVID-19 lockdowns, it has been suggested that indoor air pollution might have increased during the same period as people spent more time indoors and burnt more fuel for cooking and heating. About 78% of India's 1.3 billion population uses solid fuels for their primary and secondary needs. It has been estimated that when solid biomass fuels are burnt for cooking or heating, PM 2.5 concentrations are in the range of 163-600 µg/m 3 , about 6-23 times the safe level of daily air pollution exposure of 25 µg/m 3 recommended by the World Health Organization (WHO) (Tripathi 2020). Around 50% of the population that is usually out of the home during peak cooking hours (WHO 2018) were confined indoors during the lockdown periods, therefore the number of people affected (i.e., asthma, premature death) by indoor air pollution is likely to have increased significantly during the lockdown.
During the COVID-19 lockdown period, the present study and other recent studies highlighted the impact of restricted anthropogenic activities on air quality across Indian megacities and other megacities around the world (Table 1). In India, PM 2.5 , PM 10 and NO 2 show a significant reduction of up to 40-50% in many of its megacities during the lockdown period as highlighted by Mahato, Pal, and Ghosh (2020), Singh and Chauhan (2020), Srivastava et al. (2020) and Jain and Sharma (2020). The PM concentrations in Delhi were still above WHO guidelines, therefore a higher number of people will be affected by long-term exposure. Generally, air quality in megacities is affected by emission sources, atmospheric reactivity, and meteorology. In India, significant reduction of air pollution is primarily due to the decrease in major anthropogenic activities such as vehicles, industries, and other fugitive sources such as household cooking, emissions from local industries, street food vendors, semi-open cooking in restaurants (using coal for cooking), and other non-exhaust emissions. Meteorology also played an important role in emission reduction during the lockdown as intermittent rain events were also observed during the lockdown in some parts of India. Sharma et al. (2020) reported that meteorology was favorable during the lockdown, otherwise, the predicted PM 2.5 levels could be around 33% higher than levels reported during the lockdown.
In agreement with the current study and all the above studies also shed light on the factors leading to air quality improvement, where air pollution levels remain relatively high and attaining the standard norms is a challenge. Our study results indicate that temporary lockdown could be considered to mitigate environmental and public health damage to some extent. The lockdown gave us a rare opportunity to establish the baseline pollution level of air pollutants in Delhi, key information to set appropriate target limits. Learnings from this natural experiment will help us and develop effective management policies for achieving better air quality in urban centers like Delhi. In the past, measures like oddeven traffic restrictions and air pollution emergencies were implemented in megacities such as New Delhi, however, this failed to make a measurable impact on air pollution levels (Kumar et al. 2017). For locationspecific policy development, source apportionment studies need to compare during the lockdown and nonlockdown periods to provide a better understanding of source contribution in each location (Ravindra et al. 2021).

Conclusion
The COVID-19 restrictions reduced the anthropogenic emissions during the lockdown and unlock phases in the megacity Delhi. This study provides evidence of the reductions using air quality monitoring data in 2020 (25 March to 30 June 2020) compared with the same period in the proceeding years 2018-2019. This study also analyzed the influencing factors including meteorology, the diurnal variation and the short-term health impact of the decrease of pollutants concentrations. The conclusions can be drawn as: • Relatively high reductions of PM 10 , PM 2.5 , and O 3 (61%, 48%, and 42% respectively) in lockdown phase 1, whereas the highest reduction of NO 2 (49%) was observed in the lockdown phase 2. The restrictions on vehicular, industry, road dust and construction activities are mainly responsible for the overall reduction, although other sources like biomass burning for cooking, coal power plants, diesel generating sets, and waste incinerators can not be ignored as PM 10 and PM 2.5 still exceeded the NAAQS. • Even though the maximum restrictions were withdrawn in unlock phase 1, the levels of PM 10 , PM 2.5 , NO 2 , and O 3 still declined by 40%, 30%, 24% and 14% respectively, as compared with 2018-19 levels, thus demonstrating the residual effect of restrictions in altering peoples behavior and activity patterns. • Except for O 3 , the highest decline of pollutant concentrations was observed at traffic and industrial monitoring sites. At some monitoring sites, like Delhi Technological University, Mandir Marg and Nehru Nagar, O 3 continuously increased during the study period perhaps due to the decrease in the levels of PM 2.5 and NO. • Pollutants concentrations displayed a gradual decrease in the first three phases, however, after that pollutants concentrations gradually increased as day by day the restrictions were removed. Analysis of diurnal variation revealed that the implementation of lockdown helped to suppress PM 10 and PM 2.5 peaks during the daytime. And the diurnal variation of O 3 showed generally increased concentrations with the highest peak during the first to the third phase of lockdown in 2020. • An appreciable reduction of PM 2.5 and O 3 concentrations during COVID-19 restriction led to a decrease of 903 premature deaths, which is about 60% lower compared to similar periods of 2018 − 2019. This demonstrates that if main emission generating activities are controlled systematically, significant pollution reduction and health gains can be achieved. This study has suggested that lockdowns have acted as a useful lens to identify the primary and secondary sources of air pollution which will help to develop future air quality management policy.