Widespread and persistent ozone pollution in eastern China during the non-winter season of 2015: observations and source attributions

Rapid growth of industrialization, transportation, and urbanization has caused increasing emissions of ozone (O3) precursors recently, enhancing the O3 formation in eastern China. We show here that eastern China has experienced widespread and persistent O3 pollution from April to September 2015 based on the O3 observations in 223 cities. The observed maximum 1 h O3 concentrations exceed 200 μg m−3 in almost all the cities, 400 μg m−3 in more than 25 % of the cities, and even 800 μg m−3 in six cities in eastern China. The average daily maximum 1 h O3 concentrations are more than 160 μg m−3 in 45 % of the cities, and the 1 h O3 concentrations of 200 μg m−3 have been exceeded on over 10 % of days from April to September in 129 cities. Analyses of pollutant observations from 2013 to 2015 have shown that the concentrations of CO, SO2, NO2, and PM2.5 from April to September in eastern China have considerably decreased, but the O3 concentrations have increased by 9.9 %. A widespread and severe O3 pollution episode from 22 to 28 May 2015 in eastern China has been simulated using the Weather Research and Forecasting model coupled to chemistry (WRF-CHEM) to evaluate the O3 contribution of biogenic and various anthropogenic sources. The model generally performs reasonably well in simulating the temporal variations and spatial distributions of near-surface O3 concentrations. Using the factor separation approach, sensitivity studies have indicated that the industry source plays the most important role in the O3 formation and constitutes the culprit of the severe O3 pollution in eastern China. The transportation source contributes considerably to the O3 formation, and the O3 contribution of the residential source is not significant generally. The biogenic source provides a background O3 source, and also plays an important role in the south of eastern China. Further model studies are needed to comprehensively investigate O3 formation for supporting the design and implementation of O3 control strategies, considering rapid changes of emission inventories and photolysis caused by the Atmospheric Pollution Prevention and Control Action Plan released by the Chinese State Council in 2013.


Introduction
In the urban planetary boundary layer (PBL), ozone (O 3 ) is formed as a result of photochemical reactions involving volatile organic compounds (VOCs) and nitrogen oxide (NO x ) in the presence of sunlight (Brasseur et al., 1999): NO 2 + hυ → NO + O 3 P (290 nm < λ < 420 nm) where hυ represents the energy of a photon; O 3 P and O 1 D represent the ground state and electronically excited oxygen atoms, respectively; RO 2 , RO, and OH denote peroxy, oxy-, and hydroxyl radicals, respectively.High O 3 concentrations ([O 3 ]) are of major environmental concern due to its deleterious impacts on ecosystems (e.g., National Research Council, 1991) and human health (Lippmann, 1993;Weinhold, 2008).
The emissions of O 3 precursors, VOCs and NO x , have been significantly increased recently in China due to rapid industrialization and urbanization, and increasing transportation activity (e.g., Zhang et al., 2009;Kurokawa et al., 2013;Yang et al., 2015).Satellite measurements have demonstrated that NO x emissions have been increased by a factor of 2 in central and east China from 2000 to 2006 (Richter et al., 2005).Zhang et al. (2009) have also shown an increasing trend of NO x emissions with an enhancement of 55 % in China from 2001 to 2006.NO x emissions have still continued to increase since 2006, due to increasing power plants and vehicles (S.W. Wang et al., 2012;Y. Wang et al., 2013;Yang et al., 2015).In addition, the agriculture has been proposed to have a large potential to produce NO x (Oikawa et al., 2015).VOC emissions have been estimated to increase by 29 % during 2001-2006 in China (Zhang et al., 2009), and predicted to increase by 49 % by 2020 relative to 2005 levels (Xing et al., 2011).Additionally, modeling studies have been performed to investigate the O 3 pollution in eastern China (Wang et al., 2010;Liu et al., 2012;Situ et al., 2013;Huang et al., 2015).For example, Tie et al. (2013) have analyzed the characteristics of regional O 3 formation to explain the O 3 pollution in Shanghai and its surrounding area using the Weather Research and Forecasting model coupled to chemistry (WRF-CHEM).Using the observation-based chemical model, Xue et al. (2014) have provided insights into the ozone pollution in Beijing, Shanghai, and Guangzhou by analyzing the O 3 precursors and the potential impacts of heterogeneous chemistry.
Increasing O 3 precursor emissions have caused O 3 to be one of the most serious air pollutants of concern during summertime, particularly in eastern China, including the North China Plain (NCP), Yangtze River Delta (YRD), and Pearl River Delta (PRD) (e.g., Xu et al., 2011;Tie et al., 2013;Li et al., 2013;Feng et al., 2016).For example, a maximum O 3 concentration of 286 ppb has been observed in urban plumes from Beijing (Wang et al., 2006).Chen et al. (2015) have reported that the average maximum daily [O 3 ] exceed 150 µg m −3 in the summer of 2015 at most of monitoring sites in Beijing.Wu et al. (2017) have also shown that, during summertime in 2015 in Beijing, the average O 3 concentration in the afternoon was 163.2 µg m −3 , and the frequency of the O 3 exceedance with hourly [O 3 ] exceeding 200 µg m −3 was 31.8 %.In addition, Cheng et al. (2016) have demonstrated an increasing trend of daily maximum 1 h [O 3 ] from 2004 to 2015 in Beijing, and Ma et al. (2016) have reported a significant increase of surface O 3 at a rural site in NCP.In the PRD region, the annual average near-surface O 3 level has been reported to increase from 24 ppbv in 2006 to 29 ppbv in  2 Model and methodology

WRF-CHEM model and configurations
In the present study, we use a specific version of the WRF-CHEM model (Grell et al., 2005) to investigate the O 3 formation in eastern China.The model is developed by Li et al. (2010Li et al. ( , 2011aLi et al. ( , b, 2012) ) at the Molina Center for Energy and the Environment, including a new, flexible gas-phase chemical module and the Models-3 community multiscale air quality (CMAQ) aerosol module developed by the US EPA (Binkowski and Roselle, 2003).The wet deposition of chemical species is calculated using the method in the CMAQ module and the dry deposition parameterization follows Wesely (1989).The fast radiation transfer model (FTUV) is used to calculate photolysis rates (Tie et al., 2003;Li et al., 2005), considering the impacts of aerosols and clouds on the photochemistry (Li et al., 2011b).The ISORROPIA version 1.7 is used to calculate the inorganic aerosols (Nenes et al., 1998).
The secondary organic aerosol (SOA) is predicted using a non-traditional SOA module, including the volatility basis set (VBS) modeling approach and SOA contributions from glyoxal and methylglyoxal.Detailed information about the WRF-CHEM model can be found in Li et al. (2010Li et al. ( , 2011aLi et al. ( , b, 2012)).
A high O 3 pollution episode from 22 to 28 May 2015 in eastern China is simulated using the WRF-CHEM model.The WRF-CHEM model adopts one grid with horizontal resolution of 10 km and 35 sigma levels in the vertical direction, and the grid cells used for the domain are 350 × 350 (Fig. 1).The physical parameterizations include the microphysics scheme of Hong and Lim (2006), the Mellor-Yamada-Janjic (MYJ) turbulent kinetic energy (TKE) planetary boundary layer scheme (Janjić, 2002), the unified Noah land-surface model (Chen and Dudhia, 2001), the Goddard longwave (Chou and Suarez, 2001) and shortwave parameterization (Chou and Suarez, 1999).The National Centers for Environmental Prediction (NCEP) 1 • × 1 • reanalysis data are used to obtain the meteorological initial and boundary conditions, and the meteorological simulations are not nudged in the study.The chemical initial and boundary conditions are interpolated from the 6 h output of MOZART (Horowitz et al., 2003).The spin-up time of the WRF-CHEM model is 28 h, which is generally long enough for simulations considering that the initial and boundary conditions are adopted from MOZART, a global chemical transport model.The SAPRC-99 chemical mechanism is used in the present study.The anthropogenic emissions are developed by Zhang et al. (2009) and Li et al. (2017), including contributions from agriculture, industry, power generation, residential, and transportation sources.The biogenic emissions are calculated online using the MEGAN (Model of Emissions of Gases and Aerosol from Nature) model developed by Guenther et al. (2006).Detailed model configurations are given in Table 1.The simulation domain is shown in Fig. 1.
For discussion convenience, eastern China is divided into four sections: (1) northeast China (including Heilongjiang, Jilin, Liaoning, and the east part of Inner Mongolia, hereafter referred to as NEC), (2) the North China Plain and surrounding areas (including Beijing, Tianjin, Hebei, Shandong, Henan, Shanxi, and the north part of Jiangsu and Anhui, hereafter referred to as NCPs), (3) the YRD and surrounding areas (including the south part of Jiangsu and Anhui, Shanghai, Zhejiang, and Hubei, hereafter referred to as YRDs), and (4) the PRD and surrounding areas (including Fujian, Jiangxi, Hunan, Guangxi, and Guangdong, hereafter referred to as PRDs) (shown in the Supplement, Fig. S1).

Statistical methods for comparisons
We use the mean bias (MB) and the index of agreement (IOA) to assess the WRF-CHEM model performance in simulating air pollutants against measurements.
where P i and O i are the calculated and observed pollutant concentrations, respectively.N is the total number of the predictions used for comparisons, and O represents the average of the prediction and observation, respectively.The IOA ranges from 0 to 1, with 1 showing perfect agreement of the prediction with the observation.

Air pollutants measurements
The hourly near-surface CO, NO 2 , SO 2 , and PM 2.5 mass concentrations from April to September 2015 in eastern China are released by China MEP and can be downloaded from the website http://www.aqistudy.cn/.China MEP releases the pollutant observations using the mass concentration (µg m −3 or mg m −3 ) as the unit.Therefore, in order to keep consistent with the observations, the mass concentration is used in the paper, although the mixing ratio (such as ppbv) is a more common unit used in the literature for air pollutants.
3 Results and discussions

O 3 pollution in eastern China
Continuous deterioration of air quality in China has engendered the implementation of the There are several possible reasons for the O 3 pollution deterioration in eastern China since implementation of the AP-PCAP.Firstly, if the O 3 production regime in eastern China is VOC sensitive, the decrease of NO x due to implementation of the APPCAP likely enhances the O 3 formation.Secondly, mitigation of PM 2.5 or aerosols directly or indirectly increases the photolysis rates and expedites the O 3 formation.Thirdly, increasing transportation activities enhances the emissions of VOCs and semi-VOCs, facilitating the O 3 formation.In addition, variability of meteorological situations also leads to the [O 3 ] fluctuation (Calkins et al., 2016).Hence, implementation of the APPCAP does not help mitigate [O 3 ], and unfortunately, severe O 3 pollution has been looming in eastern China.
In      Generally, the simulated O 3 spatial patterns are consistent with observations, but the model underestimation or overestimation still exists.For example, the model remarkably overestimates the observed [O 3 ] on 24 May, and also cannot well reproduce the high [O 3 ] on 25 May in PRD.There are several reasons for the model biases in simulating [O 3 ] distribution.Firstly, the meteorological situations play a key role in air pollution simulations (Bei et al., 2010(Bei et al., , 2012)), determining the formation, transformation, diffusion, transport, and removal of the air pollutants.Therefore, uncertainties in meteorological field simulations significantly influence the air pollutant simulations.On 24 May, the model fails to predict the rainy or overcast weather, leading to remarkable overestimation of [O 3 ] in PRD.Secondly, the 10 km horizontal resolution is used in simulations, which cannot resolve cumulus clouds well.The model overestimates the [O 3 ] observed in some cities with [O 3 ] much lower than their surrounding cities, which is primarily caused by the model failure in resolving convections.Thirdly, the fast changes in emissions are not reflected in the emission inventories used in the present study.
Figure 7 provides the diurnal profiles of calculated and observed near-surface [O 3 ] averaged over the ambient monitoring sites in provinces and municipalities in eastern China during the episode.The model reasonably well reproduces the temporal variations of surface [O 3 ] compared to observations, e.g., peak [O 3 ] in the afternoon due to active photochemistry and low [O 3 ] during nighttime caused by the NO x titration.Three provinces in NEC (Jilin, Liaoning, and Inner Mongolia) are apparently impacted by the transboundary transport from NCPs when the south winds are prevailing (Fig. 6).Thus, the uncertainties of wind field simulations constitute one of the most important reasons for the model biases in modeling [O 3 ] in these three provinces.The model underestimates considerably the observed [O 3 ] in the three provinces (Fig. 7a, c, d), with MBs exceeding 19 µg m −3 .The model generally exhibits good performance in simulating [O 3 ] variations in the provinces of NCPs (Fig. 7e-l) with IOAs exceeding 0.90, but is subject to underestimating the observations, particularly in Beijing which is also significantly influenced by the transboundary transport (Wu et al., 2017).In YRDs, the model cannot well predict the observed [O 3 ] in Shanghai, which is affected by the sea breeze when the large-scale wind fields are weak.In general, however, current numerical weather prediction models, even in research mode, still have difficulties in producing the location, timing, depth, and intensity of the sea-breeze front (Banta et al., 2005;J. Wang et al., 2013).The model reasonably predicts the [O 3 ] variations compared to measurements in PRDs (Fig. 7p-t) with IOAs more than 0.7, but overestimates the observed [O 3 ] with MBs varying from 3.8 to 16.7 µg m −3 , showing model biases in modeling precipitation processes.
The comparisons of simulated versus observed distributions and temporal variations of NO 2 mass concentrations ([NO 2 ]) are shown in the Supplement (Figs.S3 and S4).The simulated high near-surface [NO 2 ] are mainly concentrated in NCP, YRD, and PRD, which is generally consistent with the measurements.The model also reasonably yields temporal variations of [NO 2 ] compared to measurements, but the simulations of [NO 2 ] are not as good as those of [O 3 ], and the IOAs in Liaoning, Tianjin, and Shanghai are lower than 0.5.The difference between simulations and observations is frequently rather large during nighttime, which is perhaps caused by the model biases in modeling nighttime PBL or the complexity of nighttime chemistry.Another possible reason for NO x biases in simulations is lack of consideration of the NO x emissions in the agricultural region, which has been proposed to generate high NO x emissions under hightemperature conditions (Oikawa et al., 2015).In general, the calculated distributions and variations of [O 3 ] and [NO 2 ] are consistent with the corresponding observations, showing that the simulations of meteorological fields and emission inventories are reasonable, providing the base for sensitivity studies.

Sensitivity studies
O 3 formation in the PBL is a complicated nonlinear process, depending on its precursors of NO x and VOCs from biogenic and various anthropogenic sources.It is imperative to evaluate the O 3 contribution from various sources for devising  the O 3 control strategy.Rapid growth of industries, transportation, and urbanization has caused increasing emissions of NO x and VOCs in eastern China (e.g., Zhang et al., 2009;Huang et al., 2011;Z. Wang et al., 2012;Y. Wang et al., 2013;Yang et al., 2015).Numerous studies have also demonstrated that biogenic VOCs, such as isoprene and monoterpenes, play a considerable role in the O 3 formation in the PBL (e.g., Chameides et al., 1988;Tao et al., 2003;Li et al., 2007Li et al., , 2014)).Therefore, sensitivity studies are used to evaluate the O 3 contributions of biogenic, industry, residential, and transportation sources in eastern China, respectively.It is worth to note that emissions of power plants are directly associated with residential living and industrial activities.Thus, in the study, 75 % of emissions from power plants are assigned to the industry source and the rest are assigned to the residential source according to the ratio of the power consumption used in industrial activities to residential living (Wang et al., 2012).The factor separation approach (FSA) is used to evaluate the contribution of some emission source to the O 3 concentration by differentiating two model simulations: one with all emission sources and the other without some emission source.Therefore, except the control simulations with all emissions, an additional four sensitivity simulations are performed in which the biogenic, industry, residential, and transportation emissions are excluded, respectively, to assess their corresponding contributions to the O 3 formation in eastern China.S1 in the Supplement further presents the emission rates of major O 3 precursors from different emission sources in the model domain during the study episode.The industrial source dominates the VOC and NO x emissions, playing a key role in the O 3 formation.The transportation source emits more NO x and active VOCs, such as olefins and aromatics, than the residential source, contributing considerably to the O 3 formation.
Another three sensitivity studies are conducted to further explore the high O 3 formation in eastern China, in which only the industry, residential, and transportation sources are considered, respectively.It is worth to note that biogenic emissions are included in all the three sensitivity simulations considering that the biogenic emissions provide natural O 3 precursors and cannot be anthropogenically controlled.Figure 10 presents the O 3 contributions from individual anthropogenic sources averaged in the afternoon dur-

Summary and conclusions
In the present study, air pollutant observations, released by China MEP, have been analyzed to explore the O 3 pollution situation in eastern China.Analyses of air pollutant observations in 66 cities from 2013 to 2015 have shown that, although implementation of the APPCAP has considerably decreased the CO, SO 2 , NO 2 , and PM 2.5 mass concentrations from April to September in eastern China, the [O 3 ] have in-  Widespread and persistent O 3 pollution poses adverse impacts on ecosystems and human health.Considering the key role of the industry source in the high O 3 formation, mitigation of the industry source becomes the top choice to improve the O 3 pollution in eastern China, particularly with regard to the VOC emissions that are still not fully considered in the current air pollutant control strategy.Rapid increase of vehicles also enhances the VOC and NO x emissions and the transportation source plays an increasingly important role in the O 3 pollution.In addition, the rapid decrease of PM 2.5 due to implementation of the APPCAP reduces the aerosol and cloud optical depth, which is subject to enhance the O 3 formation by increasing the photolysis.Hence, stringent control strategies of VOCs and NO x need to be designed comprehensively and implemented to avoid the looming severe O 3 pollution in eastern China.
Although the model performs generally well in simulating O 3 and NO 2 during a 7-day O 3 pollution episode in eastern China, uncertainties from meteorological field simulations and emission inventory still cause model biases.Meteorological conditions play a key role in the formation of air pollution, determining the formation, transformation, diffusion, transport, and removal of the air pollutants in the atmosphere (Bei et al., 2010(Bei et al., , 2012)).A nudging of wind and temperature fields using observations generally improves the simulation of meteorological fields, reducing the model biases in reproducing the O 3 temporal variation and spatial distribution.Thus, future studies are needed to improve the meteorological fields using the data assimilation, such as the fourdimension data assimilation (FDDA).Taking into consideration the complexity of the O 3 formation and rapid changes of emission inventories, further model studies need to be performed to investigate the O 3 formation for supporting the

Figure 1 .
Figure 1.WRF-CHEM simulation domain with topography.The filled circles represent centers of cities with ambient monitoring sites and the size of circles denotes the number of ambient monitoring sites of cities.The red and blue filled circles show the cities with air pollutant observations since 2013 and 2015, respectively.
2015, O 3 observations had been performed in 223 cities with 1064 monitoring sites in eastern China, which were used to analyze the O 3 pollution situation from April to September.For comparisons, Fig. 2 shows the distribution of observed maximum 1 h [O 3 ] in mainland China from April to September in 2015.The cities with the maximum 1 h [O 3 ] exceeding 300 µg m −3 are mainly concentrated in NCPs, YRDs, and PRD.In eastern China, there are only two cities with the maximum 1 h [O 3 ] less than 200 µg m −3 .About 28 % of cities have observed more than 400 µg m −3 [O 3 ] (about 200 ppb), showing widespread O 3 pollution in

Figure 2 .
Figure 2. Distribution of observed maximum 1 h [O 3 ] in mainland China from April to September 2015.

Figure 3 .
Figure 3. Distribution of average daily maximum 1 h [O 3 ] in mainland China from April to September 2015.
Figure 3 presents the distribution of average daily maximum 1 h [O 3 ] in mainland China from April to September 2015.The average daily maximum 1 h [O 3 ] are more than 120 µg m −3 in more than 95 % of the cities, and 160 µg m −3 in 46 % of the cities in eastern China.Particularly, there are seven cities with the average daily maximum 1 h [O 3 ] exceeding 200 µg m −3 during 6 months.Figures 4 and 5 show the distributions of exceedance days with the maximum 1 h [O 3 ] exceeding 160 and 200 µg m −3 in mainland China from April to September 2015, respectively.There are more than

Figure 4 .
Figure 4. Distribution of days with the maximum 1 h [O 3 ] exceeding 160 µg m −3 in mainland China from April to September 2015.

Figure 5 .
Figure 5. Distribution of days with the maximum 1 h [O 3 ] exceeding 200 µg m −3 in mainland China from April to September 2015.

Figure 7 .
Figure 7.Comparison of measured (black dots) and predicted (blue line) diurnal profiles of near-surface O 3 averaged over all ambient monitoring stations in provinces of eastern China from 22 to 28 May 2015.

Figure 8 Figure 8 .
Figure 8 shows the contribution of near-surface [O 3 ] averaged in the afternoon during the whole episode from industry, residential, transportation, and biogenic emissions.The industry source plays a more important role in the O 3 formation than the other three sources, with the O 3 contribution of 10-50 µg m −3 in the afternoon in eastern China.In highly industrialized areas, such as Hebei, Tianjin, Shandong, Zhejiang, etc., the O 3 contribution of the industry source exceeds 30 µg m −3 .The residential source is not important in the O 3 formation and contributes about 2-15 µg m −3 O 3 generally.The transportation source plays a considerable role in the O 3 formation, accounting for about 5-30 µg m −3 O 3 in eastern China.The O 3 enhancement due to biogenic emissions

Figure 9 .
Figure 9. O 3 contributions of industry (red line), residential (brown line), transportation (blue line), and biogenic emissions (green line) in NEC, NCPs, YRDs, and PRDs, as a function of simulated [O 3 ] in the control case.

Figure 10 .
Figure 10.O 3 contributions when only the industry (red line), residential (brown line), and transportation emissions (blue line) are considered in NEC, NCPs, YRDs, and PRDs, as a function of simulated [O 3 ] in the control case.

Figure 11 .
Figure 11.Distributions of the average O 3 concentration during peak time with (a) all anthropogenic emissions, (b) industry emissions alone, (c) residential emissions alone, and (d) transportation emissions alone on May 2015.

Table 2 .
Observed hourly mass concentrations of pollutants averaged in the afternoon from April to September 2013 and 2015 in 65 cities of eastern China.PollutantsCO (mg m −3 ) SO 2 (µg m −3 ) NO 2 (µg m −3 ) O 3 (µg m −3 ) PM 2.5 (µg m −3 ) total of 65 cities, with 427 monitoring sites, have air pollutant observations from 2013 to 2015 during April to September in eastern China (Fig.1).Considering the occurrence of high[O 3] in the afternoon (12:00-18:00 Beijing time (BJT)), Table2provides the average concentrations of air pollutants in the afternoon from April to September in the 65 cites of eastern China in 2013 and 2015.Apparently, implementation of the APPCAP has decreased the mass concentrations of CO, SO 2 , NO 2 , and PM 2.5 in eastern China, particularly with regard to SO 2 , with a reduction of close to 40 % from 2013 to 2015.The [O 3 ] however exhibit an increasing trend, enhanced by 9.9 % from 2013 to 2015.Additionally, if the O 3 exceedance is defined as hourly [O 3 ] exceeding 200 µg m −3 (Shan et al., 2016)on Prevention and Control Action Plan (hereafter referred to as APPCAP), released by Chinese State Council in September 2013 to reduce PM 2.5 by up to 25 % by 2017 relative to 2012 levels.Therefore, variations of air pollutants from 2013 to 2015 demonstrate the mitigation effects of implementation of the APPCAP on the air quality to a considerable degree.A (the second grade of National Ambient Air Quality Standards in China), the O 3 exceedance frequency in the afternoon had increased from 5.2 % in 2013 to 6.8 % in 2015, enhanced by about 31.5 %.The ozone monitoring instrument (OMI) satellite observations have also shown that the annual O 3 concentration has increased by 1.6 % per year over central and eastern China from 2005 to 2014(Shan et al., 2016).