We obtain nationwide hourly ozone observations from the China National Environmental Monitoring Center (CNEMC) (http://www.cnemc.cn/en/, last access: 22 June 2019). The observational network became operational in 2013, and by 2017 it was monitoring six surface air pollutants (including ozone and particulate matter with an aerodynamic diameter less than or equal to 2.5 µm, PM2.5) in 454 cities (in total 1597 non-rural sites) in mainland China (Fig. 1). The datasets have been widely used to examine urban air quality issues over China in recent studies (Hu et al., 2017; Li et al., 2017b; Gao et al., 2018; Shen et al., 2019; Li et al., 2019). Quality controls to remove unreliable hourly observations are applied following our previous work (Lu et al., 2018a).

The Chinese Ministry of Environmental Protection (now the Chinese Ministry of Ecology and Environment) categorized a total of 74 major cities (provincial capital cities and major cities in the North China Plain (NCP), Yangtze River Delta (YRD), and Pearl River Delta (PRD); these are three well-developed city clusters in China with severe air pollution) as State I key cities for air pollution monitoring in 2013. The number of key cities for air pollution monitoring has expanded to 169 since July 2018 (CNEMC, 2018). The list not only includes more cities in the NCP (now 55 cities), YRD (41 cities), and PRD (9 cities), but also covers regions where air pollution is emerging. As shown in Table 1 and Fig. 1, they are the Fenwei Plain city cluster (FWP, 11 cities), Chengdu–Chongqing city cluster (or Sichuan Basin, SCB, 16 cities), central Yangtze River city cluster (CYR, 22 cities), and other provincial capital cities in northeastern China (NECH, 4 cities), western China (WCH, 7 cities), and southern China (SCH, 4 cities). We focus the analyses on these 169 cities grouped into 9 city clusters (Table 1) in this study. Each city contains several monitoring sites, and we average them hourly to represent air quality at the city level.

We use the GEOS-Chem chemical transport model (CTM) (v11-02rc; http://geos-chem.org, last access: 22 June 2019) to interpret surface ozone measurements in China. The model is driven by assimilated meteorological data obtained from the Goddard Earth Observing System (GEOS) of the NASA Global Modeling and Assimilation Office (GMAO). The GEOS-FP datasets are available at a native horizontal resolution of 0.25∘ latitude by 0.3125∘ longitude, a temporal resolution of 1 h for surface variables, and boundary layer height and 3 h for others. We use the nested-grid version of GEOS-Chem that has 0.25∘×0.3125∘ horizontal resolution over East Asia (70–140∘ E, 15–55∘ N) (Chen et al., 2009; Zhang et al., 2016; Shao et al., 2019), with boundary conditions archived from the global simulation at 2∘ latitude × 2.5∘ longitude resolution.

The model includes a detailed mechanism of tropospheric HOx–NOx–VOC–ozone–halogen–aerosol chemistry (Wang et al., 1998; Bey et al., 2001; Park et al., 2004; Mao et al., 2013). The chemical kinetics are obtained from the Jet Propulsion Laboratory (JPL) and International Union of Pure and Applied Chemistry (IUPAC) (Sander et al., 2011; IUPAC, 2013). Photolysis rates are calculated using the Fast-JX scheme (Bian and Prather, 2002). The linearized ozone parameterization (LINOZ) is used to simulate stratospheric ozone as described by McLinden et al. (2000). Other species in the stratosphere are calculated in the model based on archived monthly mean production and loss rates provided by the Global Modeling Initiative (GMI; Murray et al., 2013). Dry deposition of both gas and aerosols are calculated online based on the resistance-in-series algorithm (Wesely, 1989; Zhang et al., 2001). The wet deposition for water-soluble aerosols and gas in GEOS-Chem is described by Liu et al. (2001) and Amos et al. (2012). The non-local scheme for the boundary layer mixing process and the relaxed Arakawa–Schubert scheme for cloud convection are described in Lin and McElroy (2010) and Moorthi and Suarez (1992), respectively.

Emissions in GEOS-Chem are processed through the Harvard–NASA Emission Component (HEMCO) (Keller et al., 2014). Global anthropogenic emissions in this study are from the Community Emissions Data System (CEDS; the latest 2014 condition is used for the model simulation) (Hosely et al., 2018) overwritten by regional emission inventories over the US (National Emission Inventory, NEI), Canada (Canadian Criteria Air Contaminant), Mexico (Kuhns et al., 2005), Europe (European Monitoring and Evaluation Program, EMEP), Africa (DICE-Africa inventory) (Marais and Wiedinmyer, 2016), and East Asia and South Asia (MIX inventory) (Li et al., 2017a). In particular, we apply the latest Chinese anthropogenic emissions for 2016 and 2017 from the Multi-resolution Emission Inventory for China (MEIC; http://www.meicmodel.org, last access: 22 June 2019) (Zheng et al., 2018). The MEIC is a bottom-up emission inventory with particular improvements in the accuracy of unit-based power plant emission estimates (Liu et al., 2015), vehicle emission modeling (Zheng et al., 2014), and the NMVOC speciation method (Li et al., 2014). The bimonthly spatiotemporal distributions of anthropogenic NOx, CO, and non-methane VOC (NMVOC) emissions averaged for 2016–2017 are shown in Fig. S1 in the Supplement. The highest emissions are in the populated city clusters in central eastern China (typically includes NCP, FWP, YRD, CYR, and adjacent regions) with little seasonal variation. Total annual Chinese anthropogenic emissions of NOx, NMVOC, and CO are, respectively, 22.5, 28.4, and 141.9 Tg in 2016 and 22.0, 28.6, and 136.2 Tg in 2017 (Zheng et al., 2018). The annual Chinese NOx and CO emissions both decrease by approximately 20 % in 2017 compared to 2013 when the Chinese State Council initiated the Action Plan on Air Pollution Prevention and Control, but NMVOC emissions show a slight increase of 2 % (Zheng et al., 2018). All anthropogenic emissions are emitted at the lowest layer in the model with a thickness of 120 m. A recent study found that uncertainties in industrial emission injection heights could affect surface SO2 concentrations by about 80 % (Yang et al., 2019); this may further affect surface ozone concentrations and requires further study.

GEOS-Chem has implemented a number of natural emissions, some of which are calculated online in the model. Lightning NOx emissions are parameterized as a function of cloud-top height (Price and Rind, 1992), vertically distributed following Ott et al. (2010), and spatially constrained by climatological observations of lightning flash rates from the Lightning Imaging Sensor (LIS) and the Optical Transient Detector (OTD) satellite instruments (Murray et al., 2012). Following previous studies (Hudman et al., 2007; Zhang et al. 2014), the amount of NO released per flash is 500 moles for the lightning north of 35∘ N in Eurasia and 23∘ N in North America; it is 260 moles for the rest of the world. Soil NOx emissions are calculated based on nitrogen (N) availability in soil and edaphic conditions such as soil temperature and moisture as described in Hudman et al. (2010, 2012). Biogenic VOC (BVOC) emissions are calculated following the Model of Emissions of Gases and Aerosols from Nature (MEGAN version v2.1) algorithm (Guenther et al., 2012). Monthly mean biomass burning emissions for the year 2014 from the Global Fire Emissions Database version 4 (GFED4) are used in the simulation (van der Werf et al., 2017). Bimonthly mean total emissions of these natural processes averaged over China are summarized in Table 2. Mixing ratios of methane are prescribed in the model based on spatially interpolated monthly mean surface methane observations from the NOAA Global Monitoring Division for 1983–2016 and are extended to 2020 using the linear extrapolation of local 2011–2016 trends (Murray, 2016). We find that it leads to an approximate 1 % increase in methane concentration over eastern China in 2017 relative to 2016.

Model configurations are summarized in Table 3. The standard simulation (BASE) includes all anthropogenic and natural emissions as described above. We then conduct sensitivity simulations (1) with all anthropogenic emissions (all emitted pollutants except methane, which is prescribed in all simulations here) turned off over China (noCH) and (2) with all anthropogenic emissions turned off globally (noGLOBE). Ozone concentrations in noCH therefore represent total background ozone in China (hereafter called background ozone), and those in noGLOBE represent natural background ozone (hereafter called natural ozone). The differences in ozone concentration between BASE and noCH denote the domestic anthropogenic ozone enhancements (CH anthropogenic ozone).

We also conduct several additional sensitivity simulations by turning off individual sources to estimate their ozone enhancements in the presence of all other sources. The ozone contribution from each source can therefore be estimated as the ozone difference between the BASE simulation and each sensitivity simulation. This difference quantifies the source contributions to present-day surface ozone, which includes the interaction of each specific source with all other sources rather than the pure impact from the specific source alone (Li et al., 2018a). Four such sensitivity simulations are conducted by turning off (1) BVOC emissions (noBVOC), (2) lightning NOx emissions (noLIGHT), (3) soil NOx emissions (noSOIL), and (4) biomass burning emissions (noBB).

Due to the long lifetime of ozone in the lower stratosphere (approximately years; Wang et al., 1998), the stratospheric contribution to tropospheric ozone is typically quantified using the tagged ozone simulation (TagO3) (Wang et al., 1998; Zhang et al., 2014) instead of perturbing the stratospheric ozone chemistry, which requires a long spin-up time. The tagged ozone simulation labels ozone produced above the tropopause from the photolysis of molecular oxygen as stratospheric ozone and then simulates its transport and chemical loss in the troposphere (Lu et al., 2019). This tagged stratospheric ozone is calculated based on ozone production and loss frequency archived from the BASE simulation.

For all simulations (except for the TagO3 simulation, which was spun up for 3 years), the global simulation (2∘ latitude × 2.5∘ longitude) was first conducted from April 2015 to November 2017. The results on 1 February 2016 were then interpolated to high resolution (0.25∘ latitude × 0.3125∘ longitude) over the nested domain and were used to initialize the nested model simulation. Results from the nested model for March–October in both 2016 and 2017 are analyzed.

Figure 2 compares the spatial distributions of observed and simulated bimonthly mean surface MDA8 ozone concentrations at the 169 cities (Fig. 1) averaged for the 2 years (2016–2017). Figures S2–S3 further compare daily MDA8 values in the nine city clusters and in individual cities. We use MDA8 throughout the analysis as it is the ozone air quality standard in China and also an important metric of human health exposure (Turner et al., 2016).

The model reproduces the spatial distribution of observed surface MDA8 ozone in the warm season (May–August) with a high spatial correlation coefficient (r) of 0.81–0.82 and a relatively small positive mean bias of 4.5–7.4 ppbv (7 %–13 %) in the 169 cities. In particular, the model captures the ozone hot spots over eastern China (including the NCP, YRD, FWP, and CYR city clusters) in May–June when both observations and model results show MDA8 ozone >60 ppbv in a number of cities (Fig. 2). In spring (March–April) and autumn (September–October), model results have larger positive biases (5.3–9.2 ppbv, 12 %–20 %) while still capturing the higher surface ozone in eastern China (r=0.32–0.46 for all 169 cities). The simulated daily MDA8 values are in good agreement with the observed values, with r ranging from 0.53 to 0.75 in the nine city clusters (Fig. S2). The model captures 57 % of the days with observed MDA8 >75 ppbv (approximately the grade II Chinese national air quality standard for the protection of residential areas) for all nine city clusters (and more than 60 % for NCP, YRD, and FWP cities). The model also reproduces the seasonal variation of surface ozone over mainland China, i.e., the peak in summertime and lower values in other months (Fig. 2). The seasonality of surface ozone typically reflects the dominant role of temperature and solar radiation in enhancing natural emissions of ozone precursors and accelerating photochemical ozone production under high precursor concentrations, with modulations by the arrival of the summer monsoon (Ding et al., 2013; Li et al., 2018b; Lu et al., 2018a).

Surface ozone levels at Chinese urban sites typically show large day-to-day variations driven by intense regional anthropogenic emissions, rapid chemical conversion, and shifts of meteorological patterns (Wang et al., 2006; Lu et al., 2010; Tan et al., 2018b). We find in Fig. S3 that observed MDA8 ozone concentrations in many cities can shift by more than 60 ppbv in 2 d. Such variations, however, are challenging for Eulerian chemical models to capture due to numerical diffusion, even at 10 km grid resolution, and representative issues when comparing gridded simulated results to observations at urban sites (Young et al., 2018). Artificial mixing of ozone precursors in model grids may cause higher ozone production efficiency and therefore positive ozone biases (Wild and Prather, 2006; Yu et al., 2016; Young et al, 2018). In addition, models have limited skills in presenting local emissions and meteorological conditions, particularly over complex terrain (e.g., in WCH and SCB cities) (Zhang et al., 2011; Ni et al., 2018). These limitations largely explain the ozone overestimates at Chinese urban sites and the deficiencies in capturing extremely high ozone values in our model, as well as many other modeling studies using GEOS-Chem at coarser resolutions (e.g., Wang et al., 2011; Ni et al., 2018) and regional models such as the Community Multi-scale Air Quality (CMAQ) model (Chen et al., 2018; Liu et al., 2018). We find that by increasing the resolution from 2∘×2.5∘ to 0.25∘×0.3125∘, model results show significant improvements in capturing the ozone daily variability and summertime high ozone episodes (model vs. observation correlation coefficients increase from 0.41–0.85 to 0.64–0.88 in individual cities), yet they are still not sufficient to reduce the high biases in spring and autumn (Fig. S3).

Figure 3 explores the contributions of anthropogenic, background, and natural sources to the bimonthly mean surface MDA8 ozone concentration in China averaged for 2016–2017. For the national average (averages of the model grids over the terrestrial land of China), we find that background ozone is the dominant source of present-day surface MDA8 ozone. Background contributions to surface ozone over China are as high as 90 % (49.4 ppbv) in March–April, followed by 84 % (43.2 ppbv) in September–October, 84 % (49.7 ppbv) in May–June, and 76 % (39.2 ppbv) in July–August (Fig. 3b). Natural sources contribute on average 80 % (44.1 ppbv) of the surface MDA8 ozone in March–April and 71 % (39.6 ppbv) in May–August (Fig. 3c). Higher background ozone concentrations are distributed in western and northern China where both natural ozone and ozone imported from foreign anthropogenic sources have larger influences than in eastern and southern China. The peak background contribution in spring reflects a mixed effect of larger transboundary transport from foreign anthropogenic sources driven by strong westerly winds (Zhang et al., 2008; HTAP, 2010; Ni et al., 2018) and larger transport from the stratosphere, as will be discussed in Sect. 3.3.

In contrast, ozone contributions from domestic anthropogenic sources are much smaller when averaged over China, with a peak contribution of about 24 % (12.1 ppbv) in July–August and less than 15 % (5–9 ppbv) in other months (Fig. 3d). The largest anthropogenic ozone enhancements are spatially centered in central eastern China following the hot spots of anthropogenic emissions (Fig. S1) and population density; they seasonally peak in summer due to active photochemistry at higher temperature and solar radiation conditions (Fig. S4). There is over 30 ppbv of ozone enhancement by domestic anthropogenic sources in central eastern China during July–August, which is comparable to background ozone there. Figure 3e shows the spatial distribution of foreign anthropogenic ozone enhancements estimated as the difference between the background ozone (Fig. 3b) and natural ozone (Fig. 3c). Larger contributions are found near the national borders and in the springtime. The foreign anthropogenic contributions we estimate here (∼5 ppbv for all seasons averaged over China), however, may underestimate their true contribution, since they are derived in the absence of Chinese domestic anthropogenic emissions and thus do not consider possible interactions with domestic emissions, e.g., ozone produced by foreign precursor enhancements reacted with domestic anthropogenic emissions.

Wang et al. (2011) estimated the contribution of background and anthropogenic sources to surface ozone over China in 2006 with an earlier version of GEOS-Chem (v8-01-01) at 0.5∘×0.667∘ resolution. They used daily mean ozone as a metric and found that background ozone accounted for about 93 % of total ozone in March–May and 80 % in June–August. In good agreement with Wang et al. (2011), our model results when converting to daily mean ozone as a metric estimate background ozone contributions of 95 % in March–April and 80 % in July–August. These are slightly higher than values derived from the MDA8 metric, reflecting higher domestic anthropogenic contributions to surface ozone at higher ozone levels as will be discussed later.

We now illustrate in Fig. 4 the impacts of individual natural sources on surface MDA8 ozone concentrations in China, including ozone enhancements from BVOC emissions, lightning NOx emissions, soil NOx emissions, biomass burning emissions, and ozone transported from the stratosphere. Here we focus on the source contributions to bimonthly means averaged over China and will present the daily variations at the city level in Sect. 3.4.

BVOC emissions enhance surface MDA8 ozone by 5.1 ppbv in July–August and 2.7–4.1 ppbv in other months averaged over China (Fig. 4a). The enhancements are particularly large over central eastern China where BVOCs enhance surface MDA8 ozone by more than 10 ppbv in July–August. These values are considered to be the actual contributions of BVOCs to ozone, which include the pure contributions, i.e., contributions of BVOCs to ozone in the absence of all other sources and the interactional effects of BVOCs with other sources. Li et al. (2018a) decomposed interactional and pure contributions and found that the interactional contributions of BVOCs and anthropogenic emissions enhanced summertime (August 2011) surface ozone in urban Xi'an by 14.3 ppbv, which is comparable with our results in this city, while the pure BVOC contributions alone are only about 2.6 ppbv. The seasonal and spatial variations of BVOC ozone enhancements (Fig. 4a) are mainly driven by the exponential dependency of BVOC emissions on temperature. As shown in Figs. S4 and S5, high biogenic isoprene emissions occur over central eastern China in July–August when daily maximum 2 m air temperature (TMAX) generally exceeds 30 ∘C. Previous studies have shown that estimates of BVOC ozone enhancements can be influenced by the model simulation of yields and fates of organic nitrates (RONO2), a by-product from the oxidation of isoprene and its carbonyl (Ito et al., 2009; Fu et al., 2015). Here the GEOS-Chem model assumes a 9 % yield of RONO2 from the reaction of isoprene peroxy radicals with NO, following recent laboratory observations (Paulot et al., 2012; Lee et al., 2014).

Lightning NOx emissions increase bimonthly mean surface MDA8 ozone by 6.5–9.9 ppbv averaged over China, with the largest contributions (typically more than 12 ppbv) found over the Tibetan Plateau (Fig. 4b). The large lightning ozone enhancements over western China (annual mean >7 ppbv) were also simulated by Murray (2016), and these values were higher than those over the western US (annual mean of 3–5 ppbv). However, the model may overestimate springtime lightning ozone enhancements over the Tibetan Plateau as the model shows high surface ozone biases of more than 15 ppbv in spring over this region (Fig. 2). In central eastern China, lightning activities also enhance surface MDA8 ozone by about 2–8 ppbv for all seasons. The spatiotemporal patterns of lightning ozone enhancements, however, do not follow the lightning NOx emission patterns that typically peak in summer over central eastern China in the middle troposphere (600–400 hPa) driven by strong convection (Figs. S5–S6). The large lightning ozone enhancements for all seasons over the Tibetan Plateau and the Altai Mountains (in western Mongolia) are mainly due to their high elevations. Although lightning NOx emissions are larger in July–August, the shorter ozone lifetime and stronger upward transport over central eastern China in these months can suppress downward mixing of lightning ozone enhancements to the surface (Fig. S4), resulting in their minimum influences there in the period. This also partly explains the decreases in surface ozone from spring to summer in western China as shown in Fig. 2.

Stratosphere–troposphere exchange (STE) is typically active in the boreal spring at northern midlatitudes associated with synoptic-scale and mesoscale processes, such as tropopause folds, gravity wave breaking, and deep convection (Stohl et al., 2003). We show in Fig. 4c that stratospheric ozone contributions at the surface are 8.7 ppbv in March–April and 6.7 ppbv in May–June averaged for China. Western China, particularly the Tibetan Plateau, is strongly influenced by deep stratospheric intrusions with more than 20 ppbv of surface MDA8 ozone originating from the stratosphere, consistent with the estimate of Xu et al. (2018) using a different CTM. We also find important stratospheric influences (6–10 ppbv) on surface ozone over the NCP and FWP regions in March–June. Stratospheric contributions to surface ozone are at a minimum in summer when STE is weakest and upward transport is strongest, limiting downward mixing of stratospheric ozone.

Soil NOx emissions increase national mean surface MDA8 ozone by 1.6 to 3.1 ppbv (Fig. 4d). Figure S5 shows that the model simulates the highest soil NOx emissions in May–August over central eastern China and northern Asia, consistent with those derived from satellite observations (Vinken et al., 2014). Large emissions in these regions can be explained by the combined effect of high temperature and frequent pulsing emissions of soil NOx after rainfall (Hudman et al., 2012). The largest soil ozone enhancements are not colocated with the emission hot spots but in regions such as northern China (2–6 ppbv in May–August) where surface ozone production is more sensitive to NOx as indicated by the modeled H2O2/HNO3 values (Fig. S7). The lower bimonthly mean soil NOx ozone contributions in central eastern China compared to other regions are due to averaging some negative daily values, as will be discussed in Sect. 3.4.

Compared to other natural sources, biomass burning emissions lead to lower ozone enhancements averaged over China (less than 2 ppbv). This is because biomass burning emissions are generally small over China (less than 2 % of the total NOx emissions over China) (Wang et al., 2007; Lin, 2012). Larger influences are found during boreal spring (March–April) in southwestern China than other seasons and regions in China. The enhancements are particularly large in the southern part of Yunnan Province with over 15 ppbv of ozone enhancement driven by intense fires in Southeast Asia (Fig. S5d). Previous studies have shown that Eulerian chemical models such as GEOS-Chem tend to overestimate ozone enhancements from biomass burning emissions in regions adjacent to emission sources (i.e., in southwestern China) but underestimate them in downwind regions (Zhang et al., 2014; Lu et al., 2016). This can reflect a number of limitations, including inadequate representation of wildfire emission and chemistry (e.g., missed short-lived VOCs, fewer PAN emissions, and improper emission height), as well as the stretched-flow numerical diffusion of narrow plumes in the model even at fine resolutions (Lu et al., 2016; Eastham et al., 2017).

It should be noted that the total contributions of the above individual natural sources (sum of Fig. 4) together account for about half (from 47.4 % in September–October to 53.4 % in July–August) of natural ozone levels (Fig. 3c). The missing rest can be largely attributed to contributions from global methane. Fiore et al. (2008) found that 1 Tg of methane emissions would lead to an increase in global mean surface ozone by about 15 pptv (∼9 ppbv with present-day annual emissions of 600 Tg), and such sensitivity doubled at high NOx levels. Further analyses are therefore needed to quantify the contributions of global methane levels to background ozone in China. The discrepancies may also be due to nonlinear interactional effects between different sources that are not captured by sensitivity simulations with individual emissions turned off.

It is of particular importance to assess sources contributing to surface MDA8 ozone in major Chinese cities. Figure 5 shows the ozone contributions from background, anthropogenic, and natural sources in nine city clusters (Fig. 1, Table 1) averaged for March–October (Fig. 5a) and for May–August (Fig. 5b) 2016–2017. For city-cluster averages in both time periods, background sources (and dominantly natural sources) have larger contributions than domestic anthropogenic sources to surface ozone. Averaged for March–October, domestic anthropogenic contributions are 23 %, 34 %, and 27 % for NCP, YRD, FWP city clusters, respectively. The smallest domestic anthropogenic contributions are found over northeastern China (NECH, 10 %) and western China (WCH, 18 %), where anthropogenic emissions are low and natural source influences from the stratosphere, lightning, and soil are relatively high. For the city clusters in central (CYR, SCB) and southern China (PRD, SCH), Chinese anthropogenic contributions account for 35 %–42 % in March–October. In May–August (Fig. 5b), the domestic anthropogenic contribution fractions increase by about 6 %–9 % in all city clusters except PRD and reach 31 % in the NCP, 43 % in the YRD, and 32 % in the FWP, reflecting the increasing importance of domestic anthropogenic contributions to surface ozone in the peak ozone season. The domestic anthropogenic contribution fractions in PRD remain the same for the March–October and May–August averages as surface ozone there typically peaks in September–October (Fig. 2).

In Fig. 6, we present the source contributions to daily MDA8 surface ozone in 2017 categorized by different surface ozone levels, which further illustrates the higher contributions of domestic anthropogenic sources to higher daily MDA8 ozone values in Chinese cities. For all city clusters, domestic anthropogenic contributions are much smaller (typically less than 15 % of the total simulated ozone) when simulated MDA8 ozone concentrations are lower than 50 ppbv. These lower contributions are more prevalent during non-summer seasons when ozone is dominantly controlled by background ozone, except for PRD and SCH. Figure 6 also shows that domestic anthropogenic emissions can cause ozone decreases over central eastern China and northern China due to the NOx-saturated chemical regime in spring and fall (Fig. S7), leading to conditions with background ozone higher than simulated total ozone.

When simulated MDA8 ozone levels are higher than 75 ppbv, domestic anthropogenic contributions significantly increase to 28.8±6.8 ppbv (35 %) in NCP, 27.8±7.1 ppbv (33 %) in FWP, 38.8±10.7 ppbv (46 %) in YRD, 37.7±6.7 ppbv (46 %) in SCB, and 54.7±11.8 ppbv (61 %) in PRD. For the extremely high ozone episodes (simulated MDA8 ozone >100 ppbv), their contributions are as high as 42.0±5.3 ppbv (41 %) in NCP, 39.1±4.8 ppbv (38 %) in FWP, 61.4±7.7 ppbv (58 %) in YRD, 52.4±10.7 ppbv (48 %) in SCB, and 73.2±7.0 ppbv (69 %) in PRD. We also find in Fig. 6 that there would be no days with MDA8 >80 ppbv if there were no domestic anthropogenic influences in all 169 major Chinese cities. Our results show that background ozone has a dominant role in the monthly mean surface ozone, on top of which domestic anthropogenic contributions are of particular significance in ozone pollution episodes. Reducing anthropogenic contributions can therefore be effective in alleviating surface ozone pollution. One exception is the WCH region, where background ozone contributions are dominant for all seasons and even lead to high MDA8 ozone concentrations approaching 80 ppbv.

Figures 7–9 further show the time series of observed vs. simulated daily MDA8 ozone and the source contributions in representative cities from the three most polluted city clusters (Beijing in NCP, Shanghai in YRD, and Xi'an in FWP) in March–October 2017. The model is able to capture the daily variability of MDA8 ozone as well as the ozone peaks in Shanghai and Xi'an (r=0.78 and 0.87, respectively). However, it fails to reproduce the high ozone episodes in Beijing during May–June 2017, although still with a high temporal correlation coefficient (r=0.78), reflecting some model limitations as discussed in Sect. 3.1.

Analyses of the daily MDA8 ozone concentrations in the three cities indicate that both background and domestic anthropogenic ozone enhancements are significantly higher during high ozone episodes than clean days, consistent with Fig. 6. This may reflect their common correlations with meteorological parameters such as temperature. As shown in Figs. 7–9, the deseasonalized summertime (May–August) daily TMAX values are strongly correlated with both observed and simulated ozone, with r (TMAX vs. observed ozone) of 0.72 in Beijing, 0.54 in Shanghai, and 0.75 in Xi'an. We find that such positive temperature–ozone correlations (r>0.60) are widespread in the cities over central eastern China in both observations and model results, as shown in Fig. 10. High summertime temperatures in China are typically associated with anticyclone systems (e.g., the West Pacific Subtropical High) that bring a clear sky with strong solar radiation and increased air stagnancy (Pu et al., 2017; Zhao et al., 2017). Such weather conditions tend to enhance ozone chemical production, weaken pollution ventilation, and favor natural emissions of ozone precursors. We find large temperature-driven daily variations of BVOC ozone enhancements during May–August in all three cities (Figs. 7–9). Taking Shanghai (Fig. 8) as an example, observed and simulated high ozone episodes are usually associated with large BVOC ozone enhancements of over 20 ppbv. Nevertheless, in the absence of anthropogenic emissions, the total MDA8 ozone would be below 70 ppbv despite high temperatures and background ozone in those cities. Implementing emission control measures under high temperature conditions can be effective to avoid ozone exceedance in Chinese cities.

Other natural processes such as lightning, soil emissions, and stratospheric transport also contribute to MDA8 ozone daily variations and can episodically lead to large ozone enhancements. For example, in Xi'an, a stratospheric intrusion enhanced surface ozone by over 25 ppbv in early May 2017 and pushed the observed MDA8 ozone to approach 80 ppbv (Fig. 9). Surface ozone enhancements by lightning are episodically higher than 10 ppbv and show a similar temporal evolution as stratospheric ozone enhancements, as downward air motion delivers both ozone enhancements originating from high altitudes. Ozone enhancements from soil NOx emissions become notable when there are pulsing soil emissions after rainfall (e.g., ∼10 ppbv in mid-August in Xi'an). They can be either positive or negative, reflecting the monthly or even daily shifts of the ozone chemical production regime, as also reported from satellite observations (Jin and Holloway, 2015) and several field campaigns in urban China (e.g., Lu et al., 2010). Biomass burning contributions to surface MDA8 ozone are generally small (less than 2 ppbv, not shown) in the three cities.