key: cord-0846632-m0xib8a0
authors: Ling, Chaohao; Li, Yongfei
title: Substantial Changes of Gaseous Pollutants and Health Effects During the COVID‐19 Lockdown Period Across China
date: 2021-05-01
journal: Geohealth
DOI: 10.1029/2021gh000408
sha: bfd8bc9fb5a450e8de96309f0019f9cd8445f679
doc_id: 846632
cord_uid: m0xib8a0

The human movement and economic activities have been drastically reduced due to the Coronavirus Disease 2019 (COVID‐19) outbreak, leading to the sharp decreases of pollutant emissions and remarkable air quality improvement. Nevertheless, however, the changes of gaseous pollutant concentrations and health effects across China during the COVID‐19 lockdown period remained poorly understood. Here, a random forest model was applied to assess the impact of COVID‐19 lockdown on pollutant concentrations and potential health effects. The results suggested that estimated NO(2), SO(2), and CO concentrations in China during January 23–March 31, 2020 decreased by 13.68%, 25.71%, and 7.42%, respectively compared with the same periods in 2018–2019. Nonetheless, the predicted 8‐h O(3) concentrations across China suffered from 1.29% increases during this period. The avoided premature all‐cause, cardiovascular disease (CVD), respiratory disease (RD), and chronic obstructive pulmonary disease (COPD) mortalities induced by NO(2) decrease during COVID‐19 lockdown period reached 3,954 (3,076–4,832), 635 (468–801), 612 (459–765), and 920 (653–1,186) cases. However, the increases of all‐cause, CVD, RD, and COPD mortalities due to O(3) increase during COVID‐19 lockdown period achieved 462 (250–674), 79 (29–129), 40 (−25–105), and 52 (−34–138) cases. The natural experiment demonstrated the drastic emission reduction measures could significantly decrease the NO(2), SO(2), and CO concentrations, while they significantly elevated the O(3) concentration. It is highly imperative to propose more coordinated air pollution control strategies to control O(3) pollution.

and traffic inevitably resulted in the decreases of pollutant emissions, and might improve the local air quality (Baldasano, 2020; Kroll et al., 2020) . Although some previous studies have assessed the response of air quality improvement to emission reduction during the periods of APEC Blue and Parade Blue (Guo et al., 2016; Xu et al., 2017) , most of these events only focused on a urban or regional scale. In contrast, COV-ID-19 lockdown provided an unprecedented chance to estimate the short-term effects of economic activity counterfactual to "business as usual" at a national scale.

Recently, some researches have quantified the short-term trends of gaseous pollutants both from space and surface perspectives (He et al., 2020; Lian et al., 2020) . C. observed that both of NO 2 and CO columns in China displayed significant decreases during COVID-19 lockdown period based on satellite products. Later on, Shi and Brasseur (2020) also confirmed that the surface NO 2 and CO concentrations over China decreased by 55% and 23%, respectively. The substantial decreases of pollutant concentrations certainly resulted in the increases the health benefits. Bray et al., (2021) observed that global NO 2 column based on satellite (ozone monitoring instrument (OMI) on Aura) reduced by approximately 9.19% and 9.57% during March-April. Chen et al., (2020) applied the observation data to estimate that about 8,911 NO 2 -related deaths could be avoided during the COVID-19 outbreak period. Unfortunately, the use of satellite product or surface observation alone did not accurately reflect the effect of COVID-19 lockdown on the air quality alleviation. It was well known that the column concentrations generally represented the total concentrations of gaseous pollutants in the troposphere even the stratosphere (McLinden et al., 2014) , which were not entirely derived from surface anthropogenic emissions. Thus, some researchers used the ground-level observation data to assess the impact of COVID-19 on air quality. Mahato et al., (2020) found that both of NO 2 and CO in Delhi, India also showed considerable declines during lockdown. However, each isolated site only possessed limited spatial representative area (0.25-16.25 km 2 ) and the trend analysis based on these monitoring sites alone might overestimate the decrease trend because most of these sites were located in the urban areas and these areas were more sensitive to the emission reduction compared with the rural regions (Li, Cui, et al., 2020; Shi et al., 2018) . Moreover, the monitoring sites were unevenly distributed over China, and some key regions (e.g., Hubei province) showed scarce monitoring sites, which could significantly increase the probability of exposure misclassification and the uncertainty of assessment result (Li, Cui, et al., 2020; . Thus, it was highly imperative to combine the surface observation data and satellite product to develop an empirical model to fill the gaps lack of monitoring sites and then to accurately assess the short-term variations and health effects of gaseous pollutants during COVID-19 lockdown period across China.

Here, we employed the random forest (RF) model to predict the gridded NO 2 , SO 2 , CO, and 8-h O 3 concentrations across China during January 23, 2020-March 31 in 2018-2020. Then, the difference of pollutant concentrations during COVID-19 lockdown period and those during the same periods in 2018 and 2019 were quantified. Finally, the health benefits (costs) of NO 2 and O 3 during this period were determined.

The daily NO 2 , SO 2 , CO, and 8-h O 3 datasets during January 23, 2020-March 31, 2020 across China were downloaded from the website of Ministry of Ecology and Environment of the People's Republic of China (http://www.cnemc.cn/en/). Meanwhile, these gaseous pollutant datasets during the same periods in 2018 and 2019 were also obtained from the website to compare the annual variation and to assess the effect of COVID-19 lockdown. This period was selected to assess the impact of COVID-19 lockdown on air quality since most of the residents have been forced to stay at home. After March 31, many provinces formulated some policies to resume production though the epidemic was not over.

The ground-level observation network has expanded to 1,641 monitoring sites covering 336 cities in 31 provinces (autonomous region, municipalities) across China, all of which were depicted in Figures S1 and S2. All of these monitoring sites were designed as a mixture of urban, suburban, and background sites. These monitoring sites suffered from unevenly distributed across the entire China. Most of these sites focused on East China, while the West China possessed relatively scarce monitoring sites especially in the Tibetan Plateau. The data quality in all of the sites were assured on the basis of HJ 630-2011 specifications.

The tropospheric NO 2 column density, total SO 2 column, and total O 3 column (spatial resolution: 0.25°) were collected from ozone monitoring instrument (OMI) level-3 product onboard the Aura satellite to estimate the surface NO 2 , SO 2 , and 8-h O 3 concentrations, respectively. The retrievals of surface CO mixing ratios obtained from measurements of pollution in the troposphere were used as the key variable to predict the surface CO concentrations across China. The gaseous pollutant columns derived from OMI with cloud radiance fraction >0.5, terrain reflectivity >30%, and solar zenith angles >85 must be removed. In addition, the cross-track pixels frequently influenced by row anomaly should be deleted. The retrievals of CO mixing ratios were resampled to 0.25° grids using area-weighted average method.

Apart from these satellite products, some meteorological data and geographical covariates should be added into the model (Table S1 ). The meteorological data including 2 m dewpoint temperature (D 2m ), evaporation (E), mean boundary layer dissipation (Mbld), surface pressure, T 2m , total precipitation, 10 m U wind component (U 10 ), and 10 m V wind component (V 10 ) (spatial resolution: 0.25°) during 2018-2020 were obtained from European Centre for medium-range weather forecasts (ECMWF). The 30-m resolution elevation data set was collected from geographical and spatial data cloud. The data of population density (1 km resolution) were obtained from the China Resource and Environmental Science Data Center . Additionally, the land use data with 30 m resolution (e.g., waters, grassland, urban, forest, and agricultural land) were also incorporated into the model.

The RF approach produced a large amount of decision trees based on independent bootstrap samples. Each node of decision tree was split depending on the best result with the traversal of all the variables which were randomly selected at that node. At last, the lowest out-of-bag error was selected to assure the optimal model. The model has been widely applied to estimate the air pollutant concentrations and accurately captured nonlinear and high-order interactions between the predictors and dependent variables. The detailed algorithm of RF model is summarized as follows :

where (x i , y i ) denotes the sample for i = 1, 2, …, N in M regions (M 1 , M 2 , …, M z ), I represents the weight of the tree branch, L is the branch of each decision tree, c m represents the response to the model, Δ z c denotes the best value, m represents the feature variable, c 1 represents the average of left branch, while c 2 denotes the average of right branch. n is the split point.

In our study, the RF model was applied to estimate the daily concentrations of gaseous pollutants during January 23-March 31 in 2018, 2019, and 2020. To evaluate the modeling performance of RF approach, sample-based 10-fold cross-validation technique was utilized to test the predictive power. Besides, the by-year cross-validation method was applied to validate the generalization ability of the model. The determination coefficient (R 2 ), root mean square error (RMSE), and mean absolute error (MAE) were selected as the key statistical indicators to quantitatively assess the model performance.

The premature mortality due to excessive NO 2 and O 3 exposure was calculated based on the following equation :

where M denotes the premature mortality due to excessive NO 2 and O 3 exposures; ER represents the exposure-response coefficient (Tables S2 and S3); y 0 represents baseline mortality of a specific disease ( 

The satellite data, meteorological factors, elevation, land use types, and other geographical covariates were applied to estimate the gridded NO 2 , SO 2 , CO, and 8-h O 3 concentrations across China during January 23-March 31 in 2018, 2019, and 2020 using RF model. As shown in Figure The predictive accuracy of RF model exhibited significantly yearly difference. In general, the CV R 2 values for pollutant estimates in 2018 and 2019 were significantly higher than 2020. It was assumed that the response of satellite products (column concentrations) to sharp changes of surface pollutant concentrations during the COVID-19 lockdown period might be not very sensitive. Both of RMSE and MAE for most pollutants except O 3 showed the highest values in 2018, followed by 2019 and 2020, which might be attributable to relatively higher concentrations of gaseous pollutants in 2018. On the contrary, both of RMSE and MAE showed the highest values in 2020 because the surface O 3 concentrations still suffered from persistent increases across China in recent years .

Overall, the predictive performances for all of the pollutant estimation during 2018-2020 were robust, while the transferability of this model was still remained unknown. Therefore, the by-year CV was applied to test the model's transferability in order to ensure the robustness of this model. As shown in Figure 1 , the by-year R 2 values of NO 2 , SO 2 , CO, and 8-h O 3 estimates across China were 0.62, 0.57, 0.51, and 0.68, respectively. These R 2 values were only slightly lower than the CV R 2 values of training models, and both of RMSE and MAE for by-year CV results were in good agreement with the training models. All of these results confirmed that the RF model could be employed to analyze the temporal changes and health benefits caused by COV-ID-19 lockdown.

LING AND LI 10.1029/2021GH000408

As shown in Figure 2 , the estimated NO 2 , SO 2 , and CO concentrations in China during January 23-March 31, 2020 decreased by 13.68%, 25.71%, and 7.42%, respectively compared with the same periods in 2018-2019 ( Figure S7 -S12). However, the predicted 8-h O 3 concentrations across China suffered from 1.29% increases during the COVID-19 lockdown period ( Figure S13 ). The dramatic decreases of NO 2 and SO 2 concentrations in China during this period was attributable to the substantial emission reduction of NO x and SO 2 associated with the shutdown of industries and reduction of vehicular transportation and domestic flights (>70%) (Chang et al., 2020) . Miyazaki et al. (2020) also verified that both of the NO x and SO 2 emissions across China in 2020 decreased by 36% compared with 2015. Compared with NO 2 and SO 2 , the CO concentrations seems to show the slight variation during COVID-10 lockdown. It was assumed that CO was regarded as a product of residential combustion and power generation (H. Wang et al., 2019) , and the home quarantine enhanced residential burning (heating and cooking), which might offset the decrease of industrial emission. In contrast, the surface O 3 concentration across China displayed slight increase during this period. It was assumed that the aerosol decrease might promote the O 3 increase because the aerosols scavenge HO 2 and NO x radicals that otherwise would produce O 3 (Shi & Brasseur, 2020) . Tie et al., (2005) reported that the loss of the HO 2 radical on the surface of sulphate particles significantly prohibited the O 3 formation, which explained the inverse relationship between NO 2 and O 3 concentrations.

The concentration changes of these gaseous pollutants response to regions decreased by 16.63%, 15.53%, and 13.81%, respectively. The 8-h O 3 concentrations in BTH, YRD, PRD, and Wuhan increased by 0.89%, 2.86%,−3.61%, and 3.76%, respectively. Among all of these regions, the NO 2 , SO 2 , and CO concentrations in Wuhan exhibited the most striking decrease owing to the earliest and most drastic measures to reduce people's exposure to the COVID-19. Following Wuhan, both of BTH and YRD suffered from remarkable air pollution alleviation. YRD experienced more remarkable NO 2 decrease, while BTH exhibited more dramatic SO 2 decrease. It was supposed that more of the industrial points such as coal-fired power plants and cement industries were located on BTH (Qi et al., 2017) . The sudden outbreak of COVID-19 caused the shutdown of these industries, which facilitated the SO 2 decrease. Nevertheless, YRD suffered from frequent NO 3 − pollution events due to the high loadings of NO x emission, and thus the COVID-19 lockdown triggered the rapid decrease of NO 2 concentration Yao et al., 2019) . The O 3 changes in different regions were inversely related with the NO 2 variations. Monks et al., (2015) revealed that nitric oxide (NO) emitted into the atmosphere converted a large fraction of O 3 into NO 2 when NO emission was sufficient.

In order to further reveal the impact of COVID-19 lockdown on gaseous pollutant changes, the temporal variability of the difference between 2020 and 2018-2019 were shown in Figure 3 . We can find that the weekly variability of NO 2 concentration in some major regions (e.g., Wuhan) totally displayed the gradual increases during the COVID-19 lockdown period, while the weekly variability of NO 2 level across China LING AND LI 10.1029/2021GH000408 6 of 10 was not pronounced. It was supposed that some western provinces generally possessed less pollution emissions compared with the developed regions of East China (Azimi et al., 2018; Sun et al., 2018; van der A et al., 2017) , and thus the response of air quality improvement to emission reduction was not significant. Wuhan suffered from the sharp decrease of NO 2 concentration since the first week (−58.19%) because Chinese government first imposed a lockdown in Wuhan. After the lockdown in Wuhan, the lockdown policies were expanded to many megacities of China (C. , and thus the sharp decreases of NO 2 concentrations in BTH, YRD, and PRD were lagged behind about one week. After four weeks of COV-ID-19 outbreak, the decreases of NO 2 levels have been significantly shrunken because many cities began to resume production and the anthropogenic emissions began to increase (Chang et al., 2020) . In PRD, the NO 2 concentrations in late March 2020 returned to the same levels as 2018-2019. The weekly variability of SO 2 and CO displayed the similar characteristics to NO 2 , while the duration of CO decline was longer than NO 2 and SO 2 (Figures S11 and S12). In contrast to these pollutants, the 8-h O 3 concentration showed the decreasing trend during COVID-19 lockdown except the sporadic week (The fifth week in PRD) due to the unfavorable meteorological conditions. Based on the original data, the fifth week in PRD was characterized with the static weather including low wind speed (2.4 m/s), which caused the higher ozone concentration during this week.

The substantial changes of air pollutant concentrations during COVID-19 period generally plays an important role on the human health, which could be estimated based on population, baseline incidence rates for specific outcomes, and epidemiological exposure-response functions. Owing to the short window of the COVID-19 lockdown, we only estimated the short-term health benefits (costs) associated with NO 2 and 8-h O 3 exposure. In our study, we estimated the avoided premature mortalities and derived from CVD, RD, and COPD and the total mortalities during January 23-March 31, 2020 and the same periods during 2018-2019. The difference of mortalities were regarded as the health benefits (costs) during COVID-19 lockdown. As shown in Table 1 decrease, the all-cause mortality in this city was still lower than those in some megacities due to the relatively few population. Nonetheless, the mortalities derived from O 3 exposure showed the slight increases in most regions across China. The increases of all-cause, CVD, RD, and COPD mortalities due to O 3 increase during COVID-19 lockdown period reached 462 (250-674), 79 (29-129), 40 (−25-105), and 52 (−34-138) cases. The spatial characteristics of mortalities due to O 3 increase were in good agreement with those induced by NO 2 decrease. Both of YRD and BTH suffered from the higher health costs because of the O 3 increase. In PRD, the mortalities induced by O 3 exposure during COVID-19 lockdown period still showed the decreasing trend because the local NO 2 and PM 2.5 concentrations did not show significant decreases compared with BTH and YRD. Based on the estimates, COVID-19 lockdown saved 3,954 lives due to the NO 2 decrease, while it led to about 462 mortalities owing to the O 3 increase. Overall, the air pollution declines response to the COVID-19 lockdown might play an important role on the disease transmission and health care system. It should be noted that our estimates of health benefits (costs) still suffers from uncertainties. First of all, the exposure-response coefficient was obtained from previous references (Chen et al., 2018) , and the parameter might vary during different study periods. Besides, the estimates of pollutant concentrations also show some uncertainties, which increase the errors of health effect assessment.

The unprecedented steps performed to stop the transmission of COVID-19 plays an important role on the air pollution alleviation. −3 (−1-−5) −2 (-4-1) −2 (-5-1)

Notes. The positive value indicates the health benefits during COVID-19 lockdown, while the negative one suggests the health costs. Abbreviation: COPD, chronic obstructive pulmonary disease. The natural experiment shed light upon that the stringent lockdown measures significantly decreased the concentrations of NO 2 , SO 2 , and CO concentrations because the human movement and economic activities have been strictly restricted. However, the 8-h O 3 concentrations did not show remarkable decrease even displayed slight increases in most regions across China. The result indicated that the reduction of industrial emission and vehicle emission were beneficial the dramatic decrease of CO and NO 2 concentrations. Thus, the ultralow emission measures and oil quality improvement should be further implemented. In addition, the emissions of VOCs and carbonaceous aerosols should be also constrained in order to control the elevation of O 3 concentration. Moreover, the coordinated air pollution control strategies of PM 2.5 and O 3 are needed because excessive PM 2.5 emission reduction might promote O 3 production.

Air pollution inequality and its sources in SO 2 and NO X emissions among Chinese provinces from

COVID-19 lockdown effects on air quality by NO 2 in the cities of Barcelona and Madrid (Spain). The Science of the Total Environment

Impact of coronavirus outbreak on NO 2 pollution assessed using TROPOMI and OMI observations

Impact of lockdown during the COVID-19 outbreak on multi-scale air quality

Puzzling haze events in China during the coronavirus (COVID-19) shutdown

Air pollution reduction and mortality benefit during the COVID-19 outbreak in China. The Lancet Planetary Health

Associations between ambient nitrogen dioxide and daily cause-specific mortality: Evidence from 272 Chinese Cities

The impact of the control measures during the COV-ID-19 outbreak on air pollution in China

Atmospheric inverse estimates of CO emissions from Zhengzhou

Impact of various emission control schemes on air quality using WRF-Chem during APEC China

The short-term impacts of COVID-19 lockdown on urban air pollution in China

The complex chemical effects of COVID-19 shutdowns on air quality

Coronavirus disease 2019 (COVID-19): Current status and future perspectives

Satellite-based estimation of full-coverage ozone (O3) concentration and health effect assessment across Hainan Island

Developing a novel hybrid model for the estimation of surface 8 h ozone (O3) across the remote Tibetan Plateau during

Impact of city lockdown on the air quality of COVID-19-hit of Wuhan city. The Science of the Total Environment

Spatiotemporal distributions of surface ozone levels in China from 2005 to 2017: A machine learning approach

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

Improved satellite retrievals of NO2 and SO 2 over the Canadian oil sands and comparisons with surface measurements

Air quality response in China linked to the 2019 novel coronavirus (COVID-19) lockdown mitigation

Tropospheric ozone and its precursors from the urban to the global scale from air quality to short-lived climate forcer

A high-resolution air pollutants emission inventory in 2013 for the Beijing-Tianjin-Hebei region

The response in air quality to the reduction of Chinese economic activities during the COVID-19 outbreak

Spatial representativeness of PM 2.5 concentrations obtained using observations from network stations

Long-term trends of anthropogenic SO2 , NOx , CO, and NMVOCs emissions in China

Chemistry-triggered events of PM2.5 explosive growth during late autumn and winter in Shanghai

Cleaning up the air: Effectiveness of air quality policy for SO2 and NOx emissions in China

Assessing the environmental externalities for biomass-and coal-fired electricity generation in China: A supply chain perspective

Estimating 1-km-resolution PM2.5 concentrations across China using the space-time random forest approach. Remote Sensing of Environment, 231, ARTN 111221

Satellite-derived 1-km-resolution PM1 concentrations from

Air quality improvement in a megacity

Beijing Parade Blue pollution control actions

The effects of firework regulation on air quality and public health during the Chinese Spring Festival from 2013 to 2017 in a Chinese megacity

Coronavirus disease 2019 (COVID-19): A perspective from China

The authors declare that they have no competing interests.

The land use types and population data are provided by Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (RESDC) (https://www.aqistudy.cn/historydata/), in Chinese.

This work was supported National Natural Science Foundation of China (Nos. 91744205 and 42067061) and National Key Research and Development Program of China (2019YFA0607100).