Meteorological conditions contributed to changes in dominant patterns of summer ozone pollution in Eastern China

Ground-level O3 pollution has become one of the most consequential air quality problems in China. Many previous studies have addressed the increasing trend of surface O3 concentrations in Eastern China. In this study, a new feature, i.e. the change in the dominant patterns of surface O3, was revealed, and the associated physical mechanisms were analyzed. The impacts of meteorological conditions and anthropogenic emissions were separated, and the change in the O3 dominant pattern was found to be mainly due to the variability in the meteorological conditions. From 2017 to 2019, the stable confrontation of the western Pacific subtropical high (WPSH) and East Asian deep trough (EADT) was closely related to the south-north covariant pattern of O3, because the variability in the meteorological conditions centered on the North China and Huanghuai regions. In the period of 2015–2016, the joint movements of the WPSH and EADT modulated the meteorological anomalies, creating a dipole mode in Eastern China that contributed to out-of-phase variations in O3 in North China and the Yangtze River Delta.


Introduction
At the beginning of 2013, the central government of China enacted and executed many measures to improve air quality (Xue et al 2019). As expected, the concentrations of fine particulate matter (PM 2.5 ) substantially declined (Wei et al 2017, which brought large health and economic benefits , Xue et al 2019. However, the concentrations of surface ozone showed contrasting trends and resulted in severe photochemical pollution in summer (Tang et al 2018. Surface O 3 levels that exceeded the standard damaged human health (Felzer 2020, Stenke 2020) and weakened net primary productivity (Yue et al 2017) in Eastern China. In addition to the increasing trend and related negative impacts, the dominant patterns of daily ground-level O 3 also changed in Eastern China in recent years (Yin et al 2019a). From 2014 to 2016, summer ozone pollution showed a south-north differential (D SN ) pattern with two centers in North China (NC) and Yangtze River Delta (YRD). However, the dominant patterns shifted to a south-north covariant (C SN ) pattern and was centered in North China and the Huanghuai region (NCH) in 2017 and 2018. This feature was preliminarily revealed, but the reasons driving this transition of the dominant patterns are still unclear.
The variability in surface O 3 was determined by the joint effects of emissions and meteorological conditions. As revealed by many previous studies, anthropogenic emissions have played important roles in the long-term trend of ozone pollution (Zhang et al 2014, Li et al 2019a. Numerical experiments using an up-to-date regional chemical transport model showed that the causes of increasing O 3 due to changes in anthropogenic emissions varied geographically (Liu and Wang 2020). After removing the components regressed by meteorological variables, Zhu et al (2019) found that the increasing trend of O 3 in NC was largely explained by decreases in PM 2.5 , which slowed down the sink of hydroperoxy radicals. The increase in surface O 3 in YRD could be attributed to a reduction in nitrogen oxide (NOx) emissions and an increase in the release of volatile organic compounds (VOCs) (Yu et al 2019). Two kinds of the most important precursors, NOx and VOCs, were largely emitted in the economically developed regions from NC to the YRD (figures S1(a) and (b), which are available online at https://stacks.iop.org/ERL/15/124062/mmedia). Because the economic pattern in Eastern China has been relatively stable, the map of emitted VOCs and NOx did not significantly change in the recent 10 years, indicated by high spatial correlation coefficients for 4080 grids cells (figure S1(c)). Therefore, the variations in anthropogenic emissions mainly influenced the long-term temporal trend in the surface O 3 and the perennial locations of severe air pollution but had difficulty determining the changes in daily O 3 -polluted modes.
Meteorological conditions play important roles in variations in ground-level O 3 from daily to interannual time scales. As a major teleconnection pattern in the Northern Hemisphere, the positive phase of Eurasia pattern can modulate the local meteorological conditions to enhance the photochemical reactions, thereby increase the ozone concentration in NC (Yin et al 2019b). The East Asian summer monsoon were also significantly influenced the interannual variations in surface O 3 pollution in China (Yang et al 2014). On a daily time scale, severe heat waves in summer frequently resulted in severe O 3 pollution in the YRD (Pu et al 2017). The western Pacific subtropical high (WPSH) stronger than its summer climate mean is closely related to the presence of environmental conditions that are not conducive to ozone formation, such as higher humidity, more cloud cover, less solar radiation and lower air temperature in the YRD (Zhao and Wang 2017). Photochemical production was shown to be enhanced between 800 hPa and 900 hPa above NC on dry and sunny days, and the formed O 3 was transported to the surface by downward air flow related to anticyclonic circulations at 500 hPa Liao 2019, Dong et al 2020). The meteorological conditions associated with two dominant patterns (i.e. C SN and D SN patterns) were also modulated by large-scale atmospheric circulations, such as the WPSH and East Asian deep trough (EADT).
Although the changes in dominant patterns of daily O 3 pollution were preliminarily uncovered, the robustness of these transitions and the reasons have not yet been analyzed, to the best of our knowledge. In this study, we aimed to verify the roles of daily meteorological conditions in the change from D SN to C SN and the associated physical mechanisms. The remainder of this paper is organized as follows. The data and methods are described in section 2. Section 3.1 confirms the change in the dominant O 3 patterns and the roles of meteorological conditions through observational and numerical approaches. Then, the possible physical mechanisms are examined in section 3.2. The main conclusions and necessary discussions are included in section 4.

Data descriptions
The hourly ozone concentration data since May 2014 are publicly available at https://www.aqistudy.cn/ historydata/ (last access: 12 April 2020). The number of observation sites in 2014 was much fewer than that at later dates; thus, we used the ozone data from 2015 to 2019 to analyze the dominant patterns of the maximum daily 8 h average ozone concentrations (MDA8 O 3 ). The sites with more than 5% of the data missed were eliminated, and then 660 sites in Eastern China were employed in this study. The NOx and VOCs data were emissions data from the Multi-resolution Emission Inventory for China, which can be downloaded from the website of www.meicmodel.org.
The daily atmospheric data employed in the stepwise multiple linear regression (MLR) are listed in table S1; the variables included surface air temperature (SAT) and relative humidity (RH), 10 m zonal and meridional wind (u10m and v10m), boundary layer height, precipitation, low and medium cloud cover (mlcc), 850 hPa meridional wind, and sea level pressure. These reanalysis data, with a resolution of 0.125 • × 0.125 • , were provided by the fifth generation European Center for Medium-Range Weather Forecasts reanalysis dataset (Copernicus Climate Change Service 2017). Furthermore, the zonal and meridional wind, geopotential height, RH, and vertical velocity at different vertical levels and downward solar radiation at the surface were also used to analyze the associated atmospheric circulations.

Quantifying contributions of meteorological conditions according to MLR
A number of previous studies have examined the meteorological influences on air pollution in China (Li et al 2019a, Yin et al 2019b and decomposed the impacts of meteorological conditions and anthropogenic emissions with a MLR approach (Zhai et al 2019). In this study, we also constructed a stepwise MLR model with the highly correlated meteorological variables at each measured site. That is, the regression values were identified as the impacts of meteorological conditions on surface ozone (MDA8 O 3 M), and the residual differentials were the effects of anthropogenic emissions (MDA8 O 3 R, equation (1)). Nine of the significantly correlated meteorological variables were considered candidates (table S1) and then were severally selected by the stepwise MLR at each station (equation (2)). To focus on synoptic-scale variability, all data (X k and MDA8 O 3 ) were deseasonalized by subtracting the 30 d moving averages from the original data for each year (Tai et al 2012), and the linear trend was synchronously removed (Zhai et al 2019).
is the regressed ozone time series for station i, and X i,k (t) is the corresponding time series for the deseasonalized and detrended meteorological variables. The MDA8 O 3 R component cannot be explained by the MLR model and includes variability mainly attributed to changes in anthropogenic emissions (Pei et al 2020) and possible noise due to limitations of MLR. In order to examine the possible multicollinearity problem, the variance inflation factor of each developed MLR model were calculated (Kutner et al 2004). The largest variance inflation factor is less than 4.5, therefore, the multicollinearity problem is insignificant. To avoid overfitting, we also tried to limit the largest number of input factors in the stepwise MLR model to three (Leung et al 2018, Li et al 2019a, and the results are consistent with the former.

Simulating surface O 3 with fixed emissions by GEOS-Chem
The hourly ozone concentrations were also simulated with the nested-grid version of the global 3D chemical transport model (GEOS-Chem) with detailed oxidant-aerosol chemistry and driven by MERRA-2 assimilated meteorological data (Gelaro et al 2017). This model used dry deposition velocity to compute the ozone concentration at a given altitude (such as 10 m) above the surface so that the simulated data could be compared with the data from the observational network (Travis and Jacob 2019). The emissions data for 2010 were downloaded from the website of http://geoschemdata.computecanada.ca/ExtData/HE MCO/AnnualScalar. The ozone simulations were configured with a horizontal resolution of 0.5 • × 0.625 • nested grid over East Asia and driven by changing meteorological fields from 2015 to 2019 but with fixed emissions in 2010. Thus, the annual differences of simulated O 3 solely resulted from the changes of meteorology-related processes rather than anthropogenic emissions. The ozone concentration is determined by several physical-chemical processes. We used non-local planetary boundary layer (PBL) mixing in the simulation, so the emissions and dry deposition trends within the PBL were applied within the mixing (Holtslag and Boville 1993). Wet deposition was not considered because of its small contribution to the O 3 mass flux (Liao et al 2006). Thus, in the analysis of the physical-chemical processes related to meteorological conditions, the chemistry, convection, PBL mixing, transport and their sum within the PBL were the focus.

Extracting the dominant patterns by EOF
The empirical orthogonal function (EOF) is to decompose the original data into orthogonal basis functions, including spatial and temporal coefficients and corresponding variance contribution (North et al 1982). In the spatial pattern of EOF, the sites with large spatial coefficient are the center of daily variability where the decomposed values frequently and largely varied from their mean state. In our research, the EOF analysis was applied to the daily observations of MDA8 O 3 in Eastern China from 2015 to 2019, and was also applied to the simulated MDA8 O 3 to compared with the results of the observational results. Because the O 3 variability was purely produced by changing meteorological conditions in the numerical simulations, the change in dominant patterns of simulated MDA8 O 3 was influenced by meteorological contributions. In this study, we used the test method from North et al (1982) to verify the significance of the separated EOF patterns. That is, if the eigenvalue (λ i ) of the ith mode satisfied the condition

Changes in dominant O 3 patterns
As illustrated by Yin et al (2019a), two dominant patterns of summer ozone pollution were determined based on the observations from 2015 to 2018. The first prominent pattern changed synergistically in Eastern China and the second pattern showed remarkable south-north differences. When the range of data was extended to 2019, these two patterns were also significant (passing the North test), with variance contributions of 25% and 12.9% for the first (C SN pattern) and second (D SN pattern) EOF modes, respectively (figures 1(a) and (b)). In the C SN pattern, the daily MDA8 O 3 in Eastern China covaried (figure 1(a)) and was centered on NCH. However, the variation in MDA8 O 3 showed features that varied between the south and north and was centered on NC and YRD in the D SN pattern ( figure 1(b)). These first two patterns were successfully reproduced by GEOS  coefficients (>0.6) confirmed that the most dominant mode in 2015 and 2016 was the D SN pattern, which then shifted to the C SN pattern (figure 1(c)). As mentioned above, the concentrations of O 3 were significantly correlated with the meteorological conditions, and the top meteorological predictors in the MLR varied geographically (Li et al 2019a). In this study, we performed stepwise regression on the deseasonalized and detrended MDA8 O 3 at each site using the meteorological variables in table S1. The coefficient of determination (R 2 , fraction of variance explained) was used to describe the ability of the optimized MLR to reproduce the variability in the observed MDA8 O 3 (figure S3). The 5 years averaged values of R 2 for the NC, YRD and NCH were 0.51, 0.45 and 0.46, respectively, indicating good performance. The fitted values were the O 3 components influenced by meteorological conditions and were denoted as MDA8 O 3 M. In Beijing (largest city in NC), the correlation coefficient between the observed MDA8 O 3 and MDA8 O 3 M was above 0.76 in each year, exceeding the 99% confidence level ( figure 2(a)). In the largest city of the YRD (Shanghai), this correlation coefficient was above 0.68 and exceeded the 99% confidence level ( figure 2(b)).
The    Can meteorological conditions influence the change in dominant patterns of surface ozone pollution in Eastern China? That is, whether the EOF results of MDA8 O 3 M also changed from the D SN pattern from 2015 to 2016 to the C SN pattern since 2017 should be examined. As expected, the first EOF mode shifted from the D SN to the C SN pattern in the 5 years (figure S5), which was further verified by the calculated spatial correlation coefficients, which are shown in figure 1(d). The correlation coefficient between the first EOF mode and D SN pattern was significant (above 0.6) from 2015 to 2016 but dramatically declined after that. However, the correlation coefficient with the C SN pattern increased to become significant starting in 2017 ( figure 1(d)). The  1(e)). In addition, the dominant pattern of MDA8 O 3 R resembled the distributions of the NOx and VOCs emissions (figure S1), indicating a close relationship between them.

Associated physical mechanisms
In the former section, by separating the contributions of anthropogenic emissions and meteorological conditions, we speculated that the change in the dominant patterns of surface ozone pollution in Eastern China was mainly driven by the variability in meteorological conditions. In this section, we attempted to analyze the associated physical mechanisms. For both the C SN M and D SN M patterns, the time series exceeding 1 (lower than −1) standard deviation were defined as their positive (negative) phase ( figure  S7). The associated large-scale atmospheric circulations and meteorological conditions were composited According to the positive (negative) phase of C SN M, the positive (negative) MDA8 O 3 M was concentrated in the NCH region (figures S8(a) and (b)). Anomalous anticyclonic circulations were located above the NCH region in the middle (figure S9(a)) and lower (figure 4(a)) troposphere and resulted in downward air flow from the stratosphere to the near surface ( figure 5(c)). The transport of O 3 from the stratosphere to the surface was an important way to elevate the concentrations of surface O 3 (Langford et al 2009). On the other hand, the significant descending air flows indicated efficient adiabatic heating (resulting in high SAT, figure 5(b)) and dry air below 300 hPa ( figure 5(c)). Furthermore, mlcc were significantly reduced, which allowed intense solar radiation to enhance the photochemical reactions ( figure 5(a)). The produced O 3 could be transported to the surface by downward air flow related to anticyclonic circulations (Gong and Liao 2019). The composite results from GEOS-Chem also verified the impacts of the meteorological conditions. The chemical reactions had large positive values (54.7 Tons d −1 ), and the sum of all physicalchemical processes was 35.1 Tons d −1 , resulting in more O 3 ( figure 5(d)). Therefore, the negative and positive phases of the abovementioned meteorological conditions significantly influenced the variability in the MDA8 O 3 M in the NCH regions.
The MDA8 O 3 M for the positive and negative phases of the D SN M pattern had similar centers (i.e. NC and YRD) but presented opposite signs (figures S8(c) and (d)). The adjacent cyclonic and anticyclonic anomalies were located over the YRD and NC in the middle ( figure S9(b)) and lower ( figure 4(b)) troposphere, inducing ascending and descending motions (figure 6(c)), respectively. The contrast between the meteorological conditions in NC and the YRD was stark, i.e. hot-dry in NC but cool-moist in the YRD (figures 6(b) and (c)). More importantly, the photochemical production of O 3 in NC was significantly enhanced but was restrained in the YRD by the variability in solar radiation ( figure 6(a)). The GEOS-Chem simulations also confirmed the stark contrast of the chemical reactions related to the meteorological conditions. The hot-dry environments in NC resulted in chemical reactions that were 64.8 Tons d −1 stronger than those in the cool-moist air in the YRD (figure 6(d)). Thus, the differences in meteorological conditions between the NC and YRD played important roles in the formation of the D SN M pattern.
As illustrated in figures 4(a) and (b), the configurations of WPSH and EADT were closely related to the distribution of MDA8 O 3 M in Eastern China. The WPSH occupied the east of China, and the EADT was stronger and extended to the Korean Peninsula for the positive phase of C SN M, and vice versa. These two large-scale atmospheric circulations stably confronted each other and resulted in persistent changing centers of meteorological conditions in NCH. The WPSH C index was defined as the area-averaged geopotential height at 850 hPa (Z850) in the Huai River region (29-36 • N, 112-122 • E; green box in figure 4(a)), and the EADT C index was calculated as the area-averaged Z850 in black box in figure 4(a)). The annual variances of WPSH C and EADT C from 2017 to 2019 were larger than those from 2015 to 2016 (figures 4(c) and (d)), indicating that these two key factors frequently fluctuated and effectively modulated the local metrological conditions. Therefore, the stable confrontation of WPSH and EADT showed that the variations in meteorological conditions presented features associated with the C SN M pattern of of the first EOF mode was almost twice that of the second pattern), were selected as two cases to examine the rationality of the above physical mechanisms. The latitude corresponding to the WPSH ridge line at 120 • E and the longitude corresponding to the EADT axis at 45 • N were extracted to quantify the daily variability in the positions of these two key factors. The standard deviation of the WPSH ridge position was 4.9 degrees in 2018 and was 22% stronger in 2016 (i.e. 6 degrees). Similarly, the standard deviation of the EADT axis position in 2016 was also stronger than that in 2018 (figures 7(a) and (b)). The positions of the WPSH and EADT varied slightly more in 2018 and resulted in a stable changing center of meteorological conditions in NCH. The observed MDA8 O 3 in NC and the YRD varied synchronously and had a significantly positive correlation coefficient (0.38, figure 7(c)). The first EOF mode of MDA8 O 3 simulated by GEOS-Chem also showed the features of the C SN M pattern (figure 7(e)). In 2016, the positions of the WPSH and EADT frequently varied, leading to a stark contrast in meteorological conditions in NC and the YRD and thus out-of-phase MDA8 O 3 between NC and the YRD (the correlation coefficient between them was −0.35, exceeding the 99% confidence level, figure 7(d)). The simulated MDA8 O 3 also appeared as a D SN M pattern (figure 7(f)).

Conclusions and discussion
In this study, we confirmed that the dominant pattern of summer ozone pollution in Eastern China changed. From 2015 to 2016, the daily ozone pollution showed differential pattern between the south and north. However, the ozone pollution in the south and north covaried after 2017 (i.e. it changed from a south-north differential pattern to a south-north covariant pattern). By separating the impacts of meteorological conditions and anthropogenic emissions, we found that the change in the O 3 dominant pattern was mainly due to the variability in the meteorological conditions. From 2017 to 2019, the WPSH and EADT stably confronted each other, and the fluctuation in their intensity resulted in variability in meteorological conditions centered on NCH. Anomalous anticyclonic circulations located above the NCH region resulted in hot-dry air and stronger photochemical reactions and transported O 3 to the surface and vice versa. These physical mechanisms dominated and forced the daily O 3 to covary in Eastern China. In the period of 2015-2016, the WPSH and EADT frequently and mutually advanced and retreated, modulating the meteorological anomalies to create a dipole mode in Eastern China. Cyclonic and anticyclonic anomalies were located over the YRD and NC and led to a stark contrast in meteorological conditions in NC and the YRD, i.e. hot-dry in NC but cool-moist in the YRD. The enhanced photochemistry in NC and restrained O 3 production in the YRD significantly contributed to the different levels of MDA8 O 3 in these two areas and vice versa. These physical mechanisms induced the daily O 3 in Eastern China to show out-of-phase variations.
O 3 observations with high resolution can be obtained from the observational network of the China Ministry of Ecology and Environment after 2014. Although there were fewer sites in 2014, Zhao and Wang (2017) reported that the O 3 pattern differed between the north and south. In total, these two dominant patterns are highlighted in the periods of 2014-2016 and 2017-2019. The positions of WPSH and EADT and their joint movements have interannual change, which contribute to the change in dominant patterns of surface ozone. The aforementioned interannual change of WPSH and EADT is also affected by some external forcings, such as sea surface temperature (Jeong et al 2018Yu and Sun 2020) and Arctic sea ice (Guo et al 2014, Wu et al 2015, which were worthy of future studies. The time range of the observations was short and limited to revealing a long-standing relationship. A possible path was to employ the simulated O 3 concentrations by GEOS-Chem for more than three decades because this model successfully recognized these two patterns from 2015 to 2019, and the external forcing factors that lead to changes in the location of WPSH and EADT might be explained in the future. Since the emission inventory has not been updated in a timely manner after 2017, we adopted statistical methods for this study and verified our conclusions by simulations with changing meteorological fields from 2015 to 2019 and fixed emissions in 2010. If the emission inventory is updated, the simulations with variable emissions but fixed meteorological fields will illustrate more information for studying the change in dominant patterns of MDA8 O 3 .