Distinct Regimes of O 3 Response to COVID-19 Lockdown in China

: Restrictions on human activities remarkably reduced emissions of air pollutants in China during the COVID-19 lockdown periods. However, distinct responses of O 3 concentrations were observed across China. In the Beijing–Tianjin–Hebei (BTH) and Yangtze River Delta (YRD) regions, O 3 concentrations were enhanced by 90.21 and 71.79% from pre-lockdown to lockdown periods in 2020, signiﬁcantly greater than the equivalent concentrations for the same periods over 2015–2019 (69.99 and 43.62%, p < 0.001). In contrast, a decline was detected ( − 1.1%) in the Pearl River Delta (PRD) region. To better understand the underlying causes for these inconsistent responses across China, we adopted the least absolute shrinkage and selection operator (Lasso) and ordinary linear squares (OLS) methods in this study. Statistical analysis indicated that a sharp decline in nitrogen dioxide (NO 2 ) was the major driver of enhanced O 3 in the BTH region as it is a NO x -saturated region. In the YRD region, season-shift induced changes in the temperature/shortwave radiative ﬂux, while lockdown induced declines in NO 2 , attributable to the rise in O 3 . In the PRD region, the slight drop in O 3 is attributed to the decreased intensity of radiation. The distinct regimes of the O 3 response to the COVID-19 lockdown in China offer important insights into different O 3 control strategies across China.


Introduction
Tropospheric O 3 is formed through complex reactions involving volatile organic compounds (VOCs) and nitrogen oxides (NOx), along with influences of meteorological conditions (such as radiation, temperature, wind, relative humidity, daily precipitation Atmosphere 2021, 12, 184 2 of 10 amount, and surface pressure). O 3 concentrations usually exhibit a nonlinear response to source emissions of precursors, in such a way that the O 3 response to emission control of one precursor (e.g., NO x ) also depends on emissions of other precursors (e.g., VOCs) due to their complex interactions [1]. Accordingly, it is challenging to identify the driving precursors and meteorological processes. There have been considerable attempts made to examine relationships between O 3 and influencing factors, such as VOCs, NO 2 , sunshine hours, temperature, wind speed, relative humidity, daily precipitation amount, surface pressure and geopotential height [1][2][3][4][5][6]. It was demonstrated that the O 3 sensitivity regimes and leading influencing meteorological factors vary across regions and time periods [1][2][3][4][5][6].
In December 2019, the coronavirus disease, named later as COVID-19 by the World Health Organization (WHO) [7], emerged in Wuhan and it has spread worldwide quickly since then. The entire metro network along with all other public transport in Wuhan were shut down on 23 January 2020 to prevent the spread of infection. Subsequently, public gatherings, planned events, etc. were canceled, and education was suspended in the rest of China. These restrictions remarkably reduced emissions of air pollutants in China [8]. TROPOspheric Monitoring Instrument (TROPOMI) and Ozone Monitoring Instrument (OMI) revealed the unprecedented declines in NO 2 column density over China [9]. Evident declines in particulate matter and CO were also identified, whereas observations suggested an unexpected uptrend in O 3 [10][11][12]. Liu et al. [13], Le et al. [14] and Zhang et al. [15] further declared that reductions in NO x emissions resulted in the enhancement of O 3 , which would have increased atmospheric oxidation and promoted the formation of secondary aerosols. Due to the uncertain sensitivity of O 3 to precursors in air quality models and large uncertainties in emission inventories [16], these findings need more careful investigation along with observations. Additionally, the roles of meteorological conditions in the enhancement of O 3 levels during the COVID-19 lockdown period are not well understood. The COVID-19 lockdown provides a terrific evaluation testbed of emission control policy in mitigating O 3 pollution in China.
In this study, we used statistical methods, e.g., Lasso (the least absolute shrinkage and selection operator) and ordinary linear squares (OLS), to select the driving factors for the enhancement of O 3 in China during the epidemic lockdown period. The selected factors are compared against those for previous years to represent the unusual situation of 2020. The results will advance our understanding of the relative importance of meteorological conditions and O 3 precursors in the formation of O 3 in China under a low-emission scenario.

Meteorological Data and Observations of Air Pollutants
In our study, gridded meteorological variables with a spatial resolution of 0.1 • × 0.1 • were derived from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis dataset (https://cds.climate.copernicus.eu/cdsapp#!/home) [17]. We considered near surface temperature (T2), wind speeds (WS10), relative humidity (RH2), and mean surface net shortwave radiation flux (SWR) in this study. Hourly surface concentrations of O 3 , PM 2.5 , and NO 2 were obtained from the China National Environmental Monitoring Center (CNEMC) network (http://106.37.208.233:20035/, last access: 9 August 2020) [18]. As VOCs are not monitored by the CNEMC network, we used formaldehyde (HCHO) tropospheric vertical column concentrations (VCDs) retrieved from the Ozone Mapping and Profiling Suite Nadir Mapper (OMPS-NM) [19], and surface HCHO concentrations from Ground based Multi-Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) measurements in this study [20]. OMPS-NM measures UV radiation at wavelengths ranging from 300 to 380 nm using a single grating and a charge-coupled device (CCD) detector. It has a 110 • cross-track field of view (FOV), similarly to OMI (Ozone Monitoring Instrument). OMPS-NM products are provided at 50 km × 50 km spatial grids by combining measurements into 35 cross-track bins in OMPS-NM standard Earth science mode. Ground based MAX-DOAS observes tropospheric aerosols and trace gases based on scattered sunlight under different viewing angles [21]. Additionally, tropospheric vertical column concentrations of HCHO and NO 2 retrieved from TROPOspheric Monitoring Instrument (TROPOMI) [22][23][24], a satellite instrument on board of the Copernicus Sentinel-5 Precursor satellite, were used to distinguish O 3 sensitivity regimes.
Previously, satellite HCHO measurements were used to constrain VOCs emissions in Asia, Fu et al. [25] and Su et al. [19] also demonstrated that tropospheric HCHO is mostly concentrated below 1 km. To examine the reliability of representing surface concentrations of HCHO with satellite observed HCHO column, we compared OMPS HCHO VCDs against surface HCHO observations from MAX-DOAS at multiple stations across China, and found that they are significantly correlated ( Figure 1).
(CCD) detector. It has a 110°cross-track field of view (FOV), similarly to OMI (Ozone Monitoring Instrument). OMPS-NM products are provided at 50 km × 50 km spatial grids by combining measurements into 35 cross-track bins in OMPS-NM standard Earth science mode. Ground based MAX-DOAS observes tropospheric aerosols and trace gases based on scattered sunlight under different viewing angles [21]. Additionally, tropospheric vertical column concentrations of HCHO and NO2 retrieved from TROPOspheric Monitoring Instrument (TROPOMI) [22][23][24], a satellite instrument on board of the Copernicus Sentinel-5 Precursor satellite, were used to distinguish O3 sensitivity regimes.
Previously, satellite HCHO measurements were used to constrain VOCs emissions in Asia, Fu et al. [25] and Su et al. [19] also demonstrated that tropospheric HCHO is mostly concentrated below 1 km. To examine the reliability of representing surface concentrations of HCHO with satellite observed HCHO column, we compared OMPS HCHO VCDs against surface HCHO observations from MAX-DOAS at multiple stations across China, and found that they are significantly correlated (Figure 1).

Statistical Analysis
To investigate the influence of the COVID-19 lockdown on the association between the features (O3 precursors and meteorological conditions) and O3, we first applied the least absolute shrinkage and selection operator (Lasso) for relevant feature selection and then used ordinary least square (OLS) regression to obtain the statistical significance of the selected features. Lasso, proposed by Tibshirani [26], is featured both in feature selection and model interpretability. Lasso excels over other feature selection methods, such as stepwise selection, with consideration of a penalty term, which can shrink the regression coefficients towards zero to prevent overfitting. We can regard it as a variant of OLS by imposing sparse constraint to the regression model as follows: where i y and i j x are the regression target and the value of th j feature for the t h i sample, respectively.

Statistical Analysis
To investigate the influence of the COVID-19 lockdown on the association between the features (O 3 precursors and meteorological conditions) and O 3 , we first applied the least absolute shrinkage and selection operator (Lasso) for relevant feature selection and then used ordinary least square (OLS) regression to obtain the statistical significance of the selected features. Lasso, proposed by Tibshirani [26], is featured both in feature selection and model interpretability. Lasso excels over other feature selection methods, such as stepwise selection, with consideration of a penalty term, which can shrink the regression coefficients towards zero to prevent overfitting. We can regard it as a variant of OLS by imposing sparse constraint to the regression model as follows: where y i and x i j are the regression target and the value of j th feature for the i th sample, respectively. β j p j=1 are feature weights to be estimated from the model, and λ is a hyperparameter controlling the strength of the sparse constraint. Here, we choose the optimal value of λ via cross validation. The optimization algorithm to iteratively infer β j p j=1 is least angle regression (LARs) [27]. Considering the features are of different scales, instead of directly feeding them into the Lasso model, they are first normalized to have a zero mean and standard deviation of one. The output of Lasso is a set of features with non-zero weight values. In order to statistically assess their importance (e.g., sensitivity of O3 concentration to gaseous precursors and meteorological parameters), we adopted OLS to generate p values of t-tests for input features.

Evolution of O 3 from Pre-Lockdown to Lockdown Periods over 2015-2020
Since 23 January 2020 when Wuhan announced the lockdown, most of the economic activities and public transportation in China were restricted [28]. We define pre-lockdown and lockdown periods in this study as the five weeks before and after 23 January, respectively. Although lockdown occurred only in 2020, we use pre-lockdown and lockdown periods in this study to denote the same time periods in previous years as well. Human activities during the spring festival holidays are usually distinctively different from those during working days, including widespread usage of fireworks, reduction in traffic flow, shutdown of factories, etc. [29,30]. To avoid such influence, we exclude data over the spring festival holidays of 2015-2020. As emission levels and meteorological conditions vary greatly across seasons and years, we use the percentage changes of O 3 concentrations from pre-lockdown to lockdown periods to investigate the driving factors. As listed in Table 1, compared to the pre-lockdown period, O 3 concentrations in China increased by 64.27% in 2020, significantly higher than any previous year over 2015-2019 and the mean percentage change of 41.63% over 2015-2019 (p < 0.001). However, the changes of O 3 from before and after January 23 exhibit a heterogeneous spatial distribution (Figure 2). Pronounced growth of O 3 is observed in the Beijing-Tianjin-Hebei (BTH) region, especially in 2016, 2017 and 2020 (Figure 2b,c,f). In 2020, O 3 concentrations in the BTH region increased by 90.21% from pre-lockdown to lockdown periods, higher than the percentage changes for other years (Table 1). Similar enhancements of O 3 concentrations are also observed in the Yangtze River Delta (YRD) region ( Figure 2) over 2015-2020. In 2020, O 3 concentrations in the YRD region rose by 71.79% from prelockdown to lockdown periods, significantly higher than that for any other year and the mean of previous years (Table 1). In contrast, an opposite trend of O 3 from pre-lockdown to lockdown (−1.1%) periods in 2020 was identified for the Pearl River Delta (PRD) region jin-Hebei (BTH) region, especially in 2016, 2017 and 2020 (Figure 2b,c,f). In 2020, O3 concentrations in the BTH region increased by 90.21% from pre-lockdown to lockdown periods, higher than the percentage changes for other years (Table 1). Similar enhancements of O3 concentrations are also observed in the Yangtze River Delta (YRD) region ( Figure 2) over 2015-2020. In 2020, O3 concentrations in the YRD region rose by 71.79% from prelockdown to lockdown periods, significantly higher than that for any other year and the mean of previous years (Table 1). In contrast, an opposite trend of O3 from pre-lockdown to lockdown (−1.1%) periods in 2020 was identified for the Pearl River Delta (PRD) region (Table 1), which is significantly different from those for previous years. Over 2015-2019, the mean O3 concentrations increased by 14.83% from five weeks before January 23 to five weeks after January 23 in the PRD (Table 1). These inconsistent responses of O3 to pandemic lockdown across China indicate distinct regimes of O3 control.

Potential Driving Factors for Different Responses of O3 to Pandemic Lockdown in China
The concentration of O3 is affected by multiple factors, including the abundance of gaseous precursors (i.e., VOCs and NOx), intensity of radiation, winds, etc. We consider, in this study, meteorological parameters including temperature (T2), wind speed (WS10), relative humidity (RH2), and mean surface net shortwave radiation flux (SWR). Stronger temperature and solar radiation enhance emissions of O3 precursors and accelerate photochemical O3 production under high precursor concentrations [31]. Given the spatial distribution of natural emissions of VOCs across China, the effects of accelerated photochemical O3 production would be more important in north China, while both effects could be important in south China [31]. For the influence of related chemical species, we include

Potential Driving Factors for Different Responses of O 3 to Pandemic Lockdown in China
The concentration of O 3 is affected by multiple factors, including the abundance of gaseous precursors (i.e., VOCs and NO x ), intensity of radiation, winds, etc. We consider, in this study, meteorological parameters including temperature (T2), wind speed (WS10), relative humidity (RH2), and mean surface net shortwave radiation flux (SWR). Stronger temperature and solar radiation enhance emissions of O 3 precursors and accelerate photochemical O 3 production under high precursor concentrations [31]. Given the spatial distribution of natural emissions of VOCs across China, the effects of accelerated photochemical O 3 production would be more important in north China, while both effects could be important in south China [31]. For the influence of related chemical species, we include concentrations of HCHO, PM 2.5 and NO 2 . PM 2.5 concentrations are considered based on the recent reports on the interactions between O 3 formation and particles in China [32]. Similarly to the analysis of O 3 , we calculated the percentage changes of these parameters for each observation site. We then used Lasso and OLS methods to explore the potential triggering factors for the changes in O 3 during the COVID-19 lockdown period in different regions in China, as the relationship between O 3 formation and influencing factors would vary across regions [33,34]. The results for the unusual year of 2020 are also compared against those over 2015-2019.  Figure 3 are zero, suggesting the negligible roles of these variables. The importance of each selected parameters is statistically evaluated and summarized in Table S1. In the BTH region, intensified T and SWR from pre-lockdown to lockdown periods are important factors (positive effects) for enhanced O 3 over previous years (2015-2019), due to enhanced emissions of biogenic VOCs and photochemical O 3 production [35,36]. Additionally, declines in PM 2.5 might have also played a role (Figure 3a), due to the effects of higher actinic flux and reduced sink of hydroperoxyl [32]. A negative correlation between O 3 and PM 2.5 was also revealed by Chu et al. [37] in the winter in north China when the PM 2.5 concentration was above 50 ug/m 3 . In 2020, the enormous increase in O 3 was provoked mainly by the sharp declines in NO 2 , while reduced concentrations of PM 2.5 might also have played a role (Figure 3b). Different from previous years, changes in meteorological conditions exhibit negligible contributions in 2020 in the BTH region (Figure 3b). Previous model sensitivity analysis by Gao et al. [18] suggests that the BTH region is VOC-limited in winter and the same sensitivity is observed during the pre-lockdown period (Figure 4) in the BTH region, where declines in NO x would be likely to enhance O 3 concentrations [38][39][40]. Accordingly, enhanced O 3 concentrations in the BTH region during the 2020 pandemic lockdown was mainly driven by sharp declines in NO 2 (Table S1).   In the YRD region, for both previous years and 2020, O3 is enhanced by augmented T and SWR, while being negatively affected by changes in NO2 (Figure 3c). In 2020, the influences of these factors are stronger than those over 2015-2019. The response of O3 to declines in NO2 in the YRD region is in line with that in the BTH region. This is also consistent with the findings of sensitivity of O3 to emission sectors in the YRD by Gao et al. [18] and the sensitivity indicated by satellites (Figure 4). In addition, a positive relationship between changes in HCHO and response of O3 is identified in 2020 in the YRD. Although HCHO levels declined in the YRD region in 2020 (−18.64%, Table S3), its suppression of O3 formation might have been downplayed by stronger declines in NO2 (−56.63%, Table S3) and intensified T/SWR (Table S3). Thus, meteorological conditions (i.e., T and SWR) and changes in gaseous precursors (mainly declines in NO2) work together to augment O3 in the YRD.
In the PRD region, the enhancement of O3 from pre-lockdown to lockdown periods over 2015-2019 are well explained by the intensified SWR due to shift of season ( Figure  3e). Additionally, increases in NO2 would also slightly promote the formation of O3 (Figure 3e). As indicated in Gao et al. [18] and Figure 4, the PRD region exhibits a different sensitivity regime from the BTH and YRD regions, and the increase in NOx emissions is likely to promote the formation of O3 in the PRD. From pre-lockdown to lockdown periods in 2020, both T and SWR declined (Table S4). Lower T and SWR would suppress the photochemical reaction rates and reduce emissions of precursors to lower levels of O3. As indicated in Figure 3f, the dominant non-meteorological factor for the response of O3 in the PRD in 2020 is HCHO. As observed by satellite and surface monitors, HCHO abundance in the PRD increased from 9.83 to 9.97 10 15 molec/cm 2 , while NO2 concentrations declined by 55.47% (Table S4). In the PRD region, changes of O3 positively correlate with both changes in HCHO and NO2 (Figure 5b), suggesting that O3 formation in the PRD region is likely in the transition regime [5]. The promotion of O3 by enhanced HCHO In the YRD region, for both previous years and 2020, O 3 is enhanced by augmented T and SWR, while being negatively affected by changes in NO 2 (Figure 3c). In 2020, the influences of these factors are stronger than those over 2015-2019. The response of O 3 to declines in NO 2 in the YRD region is in line with that in the BTH region. This is also consistent with the findings of sensitivity of O 3 to emission sectors in the YRD by Gao et al. [18] and the sensitivity indicated by satellites ( Figure 4). In addition, a positive relationship between changes in HCHO and response of O 3 is identified in 2020 in the YRD. Although HCHO levels declined in the YRD region in 2020 (−18.64%, Table S3), its suppression of O 3 formation might have been downplayed by stronger declines in NO 2 (−56.63%, Table S3) and intensified T/SWR (Table S3). Thus, meteorological conditions (i.e., T and SWR) and changes in gaseous precursors (mainly declines in NO 2 ) work together to augment O 3 in the YRD.
In the PRD region, the enhancement of O 3 from pre-lockdown to lockdown periods over 2015-2019 are well explained by the intensified SWR due to shift of season (Figure 3e). Additionally, increases in NO 2 would also slightly promote the formation of O 3 (Figure 3e). As indicated in Gao et al. [18] and Figure 4, the PRD region exhibits a different sensitivity regime from the BTH and YRD regions, and the increase in NOx emissions is likely to promote the formation of O 3 in the PRD. From pre-lockdown to lockdown periods in 2020, both T and SWR declined (Table S4). Lower T and SWR would suppress the photochemical reaction rates and reduce emissions of precursors to lower levels of O 3 . As indicated in Figure 3f, the dominant non-meteorological factor for the response of O 3 in the PRD in 2020 is HCHO. As observed by satellite and surface monitors, HCHO abundance in the PRD increased from 9.83 to 9.97 × 10 15 molec/cm 2 , while NO 2 concentrations declined by 55.47% (Table S4). In the PRD region, changes of O 3 positively correlate with both changes in HCHO and NO 2 (Figure 5b), suggesting that O 3 formation in the PRD region is likely in the transition regime [5]. The promotion of O 3 by enhanced HCHO might have been downplayed by reduced intensity of SWR. Thus, decreased SWR from pre-lockdown to lockdown periods in 2020 served as the major driver of slightly declined O 3 in the PRD region.
ere 2021, 12, x FOR PEER REVIEW 9 of 12 might have been downplayed by reduced intensity of SWR. Thus, decreased SWR from pre-lockdown to lockdown periods in 2020 served as the major driver of slightly declined O3 in the PRD region.

Discussion
Due to the uncertainties in chemical transport modeling, we use statistical methods in this study, i.e., Lasso and OLS, to understand the driving factors for diverse responses of O3 concentration in China during the COVID-19 lockdown. We report that O3 exhibits distinct responses across China. In the BTH region, enhanced O3 concentrations during lockdown were mainly driven by sharp reductions in NOx emissions, which is related to the restrictions on transportation and economic activities. In the YRD region, both meteorological conditions (i.e., T and SWR) and changes in gaseous precursors (mainly declines in NO2) work together to augment O3 in the YRD. However, O3 declined in the PRD region during the pandemic lockdown, mainly due to decreased intensity of SWR. These results implicate that controlling sources of VOCs would be more efficient to reduce wintertime O3 levels in both BTH and YRD regions, while controlling either NOx or VOCs would work for the PRD region.
These results highly depend on the quality of the used observation datasets. The accuracy of CNEMC data has been validated and demonstrated extensively with independent surface observations in previous studies [41]. However, uncertainties in the adopted HCHO satellite column and its representation of near surface HCHO would bring about uncertainties in the conclusion. As O3 concentrations in the BTH region are mainly affected by NO2, and meteorological factors played a more important role in the other two regions, the adverse effects of the uncertainties in HCHO observations are limited in this study. Yet, this further emphasizes the need for densely distributed ground-based observations of HCHO and other VOC species. Our results also emphasize that O3 control strategies should be carefully designed with consideration of differences among regions.
Supplementary Materials: The following are available online at www.mdpi.com/xxx/s1, Table S1: The relationship between percentage changes of O3 and associated factors in the BTH, YRD and P RD regions (coefficients listed are significant at 0.05 level); Table S2: The mean values of considered parameters during pre-lockdown, lockdown periods, and percentage changes in the BTH region; Table S3: The mean values of considered parameters during pre-lockdown, lockdown periods, and percentage changes in the YRD region; Table S4: The mean values of considered parameters during pre-lockdown, lockdown periods, and percentage changes in the PRD region.

Discussion
Due to the uncertainties in chemical transport modeling, we use statistical methods in this study, i.e., Lasso and OLS, to understand the driving factors for diverse responses of O 3 concentration in China during the COVID-19 lockdown. We report that O 3 exhibits distinct responses across China. In the BTH region, enhanced O 3 concentrations during lockdown were mainly driven by sharp reductions in NOx emissions, which is related to the restrictions on transportation and economic activities. In the YRD region, both meteorological conditions (i.e., T and SWR) and changes in gaseous precursors (mainly declines in NO 2 ) work together to augment O 3 in the YRD. However, O 3 declined in the PRD region during the pandemic lockdown, mainly due to decreased intensity of SWR. These results implicate that controlling sources of VOCs would be more efficient to reduce wintertime O 3 levels in both BTH and YRD regions, while controlling either NO x or VOCs would work for the PRD region.
These results highly depend on the quality of the used observation datasets. The accuracy of CNEMC data has been validated and demonstrated extensively with independent surface observations in previous studies [41]. However, uncertainties in the adopted HCHO satellite column and its representation of near surface HCHO would bring about uncertainties in the conclusion. As O 3 concentrations in the BTH region are mainly affected by NO 2 , and meteorological factors played a more important role in the other two regions, the adverse effects of the uncertainties in HCHO observations are limited in this study. Yet, this further emphasizes the need for densely distributed ground-based observations of HCHO and other VOC species. Our results also emphasize that O 3 control strategies should be carefully designed with consideration of differences among regions.