The two-way coupled system (Yan et al., 2014; Yan et al., 2016) is built upon version 9-02 of GEOS-Chem (http://wiki.seas.harvard.edu/geos-chem/index.php/Main_Page, last access: 4 April 2017). Here we couple the global GEOS-Chem model (at 2.5∘ longitude × 2∘ latitude) with its nested model covering Asia (70∘ E–150∘ E, 11∘ S–55∘ N, at 0.667∘ longitude × 0.5∘ latitude). Through the PeKing University CouPLer (PKUCPL) for two-way coupling, for every 3 h the global model provides lateral boundary conditions for the nested model, while the nested model results replace the global model results within the nested domain (Yan et al., 2014, 2016). Both models are driven by the GEOS-5 assimilated meteorological fields at their respective horizontal resolutions from the NASA Global Modeling and Assimilation Office. There are 47 vertical layers for both models, and the lowest 10 layers are about 130 m thick each.

Both the global and nested GEOS-Chem models include the full gaseous HOx–Ox–NOx–CO–NMVOC chemistry (Mao et al., 2013) and online aerosol calculations, with further updates detailed in J.-T. Lin et al. (2012) and Yan et al. (2016). As aromatics are not explicitly represented in the model, following J. T.Lin et al. (2012), we approximate the ozone production of aromatics by increasing anthropogenic emissions of propene by a factor of 4, based on their reactivity differences, their similarity in emission spatial variability, and recently estimated emission amounts of aromatics (Liu et al., 2010). We use the Linoz scheme for ozone production in the stratosphere (McLinden et al., 2000). We adjust the stratospheric production rate in the nested model to ensure that the stratosphere–troposphere exchange (STE) of ozone in the nested model matches the STE in the global model over the same nested domain (Yan et al., 2016). Vertical mixing in the planetary boundary layer (PBL) is parameterized by a nonlocal scheme (Holtslag and Boville, 1993; Lin and McElroy, 2010), and convection in the model employs the relaxed Arakawa–Schubert scheme (Moorthi and Suarez, 1992).

Table 1 lists the emission inventories used here. Global anthropogenic emissions of NOx and CO in 2008 are from the Emission Database for Global Atmospheric Research (EDGAR v4.2). Anthropogenic NMVOC emissions are from the REanalysis of TROpospheric chemical composition (RETRO) inventory for 2000. Anthropogenic emissions over China, the rest of Asia, the United States, Canada, Mexico, and Europe are replaced by regional inventories MEIC (for 2008), INTEX-B (for 2006), NEI2005 (for 2005), CAC (for 2008), BRAVO (for 1999), and EMEP (for 2007), respectively. Emissions of CO and NOx are scaled to 2008 in the United States and to 2006 in Mexico (http://wiki.seas.harvard.edu/geos-chem/index.php/Scale_factors_for_anthropogenic_emissions, last access: 27 April 2017). We use daily biomass burning emissions from Global Fire Emission Database version 3 (GFED3) (van der Werf et al., 2010). Biogenic emissions of NMVOCs are calculated online based on the MEGAN v2.1 scheme (Guenther et al., 2012). For lightning NOx emissions, flash rates are calculated based on the cloud-top height and constrained by climatological satellite observations (Murray et al., 2012), and the vertical profile of emitted NOx follows Ott et al. (2010). Online calculation of soil NOx emissions follows Hudman et al. (2012).

Table 2 presents 10 full-chemistry simulations to quantify Chinese and foreign anthropogenic contributions to springtime ozone over China in 2008. A base simulation (CTL) includes all emissions. The second simulation excludes anthropogenic NOx, CO, and NMVOC emissions worldwide to determine the natural ozone (xANTH). Eight additional simulations exclude anthropogenic emissions over China (xCH), Japan and Korea (xJAKO), Southeast Asia (xSEA), South Asia (xSA), the rest of Asia (xROA), Europe (xEU), North America (xNA), and the rest of the world (xROW), respectively (see regional definitions in Fig. 1). All simulations cover November 2007 through May 2008, with the first 4 months used for spin-up, except for additional CTL simulations in other years for model evaluation purposes.

Table 2 also shows 10 tagged simulations (denoted as T_CTL, T_xANTH, etc.) with respect to CTL and the other eight zero-out sensitivity simulations. Each tagged simulation includes 10 tracers to track ozone produced within the troposphere of eight source regions, produced within the troposphere of the oceanic regions, or transported from the stratosphere. Considering the time for STE of air, all tagged ozone simulations are spun up for 10 years.

Ozone production is nonlinearly dependent on its precursors, adding uncertainties to the source attribution calculated by emission perturbation methods (Wu et al., 2009). To account for this issue, we use a linear weighting method to adjust all ozone attribution results, unless stated otherwise. Below is an example to determine the contribution from Chinese anthropogenic emissions (here Ci represents the sensitivity simulation for one of the eight emission source regions). The adjustment is performed for each grid cell over China. Equation (1) calculates the fractional Chinese contribution (α) to the sum of ozone from individual anthropogenic source regions and from natural sources; the simulations involved are all full-chemistry runs (CTL, xCH, xEU, … , xANTH). Equation (2) applies the fractional contribution α to the total ozone in CTL to obtain the final adjusted Chinese contribution. α=ConCTL-ConxCH∑i=18ConCTL-ConCi+ConxANTH CCH=α×ConCTL=ConCTL-ConxCH∑i=18ConCTL-ConCi+ConxANTH×ConCTL Figure 2a shows the spatial distribution of the ratio of total surface ozone in CTL to the pre-linear-weighting-adjustment sum of natural ozone, domestic anthropogenic ozone, and foreign anthropogenic ozone. The ratio is close to unity over central and western China. Over most of the eastern regions, the ratio is between 1.05 and 1.10, although it can reach 1.30 at a few locations. Figure 2b further compares the vertical profile of China average total ozone in CTL and the profile of the pre-linear-weighting-adjustment sum of natural ozone, domestic anthropogenic ozone, and foreign anthropogenic ozone. The difference between the two profiles is rather small. These results suggest relative small effects of chemical nonlinearity. And the linear weighting adjustment further removes these effects.

A similar approach was used by B. G. Li et al. (2016) to estimate the contribution of China to global radiative forcing, although in their study 20 % (instead of 100 %) of emissions over individual emission source regions are removed in the sensitivity simulations.

This study presents model evaluation over China and its neighboring countries in spring. We also evaluate the simulation of CO, a relatively long-lived transport tracer. Figure 3 shows the suite of ground, aircraft, and ozonesonde measurements.

Figure 4 compares the springtime time series of modeled (solid red line) and observed (solid black line) maximum daily average 8 h (MDA8) ozone concentrations at 10 sites with daily measurements. Model data are sampled at times and locations coincident with valid observations.

Figure 4a–b evaluate the model results at Gucheng and Longfengshan. To compare to observations in spring 2007 at these two sites, we conduct an additional full-chemistry simulation for 2007. At these sites, the model captures the observed MDA8 ozone, with a normalized mean bias (NMB) of 3 % at Gucheng and 5 % at Longfengshan. The respective correlation coefficients (R) for day-to-day variability are 0.51 and 0.59; the modest correlation is primarily because the model does not capture a few short-term spikes.

At Lin'an (Fig. 4c), the modeled spring average MDA8 ozone matches the observed value (68.9 ppb versus 65.1 ppb, R=0.64). The model cannot reproduce the observed extreme low values on several days. This deficiency is likely due to representative errors of model meteorology. Located in a hilly area, this site often receives rains and fogs in spring, which is not captured by the model meteorology at a resolution of 0.667∘ longitude × 0.5∘ latitude. We find that the extremely low observed ozone values normally occur on days with high relative humidity (black dashed line, reflecting rainy or foggy days), when the model underestimates RH (red dashed line) and overestimates ozone.

At Shangri-La, Waliguan, and Issyk-Kul (Fig. 4d–f), with high altitudes (1640–3816 m) and few local anthropogenic sources, the model overestimates the MDA8 ozone by 7–8 ppb (12 %–14 %). At Everest-Pyramid in Nepal (Fig. 4g, at 5079 m altitude), the overestimate reaches 13 ppb (19 %). These positive biases are due to overestimated transport from the free troposphere and stratosphere. The model captures the temporal variability in MDA8 ozone quite well (R=0.72–0.78) at the three Japanese sites (Yonaguni-jima, Tsukuba, and Ryori; Fig. 4h–j). The model's NMB is within 3 % at Yonaguni-jima and Ryori. There is an overestimate at Tsukuba (NMB =19%), mostly reflecting the large positive biases on a few days.

Table 4 shows model comparisons with monthly mean EANET ozone data. These data represent daily mean rather than MDA8 values, based on the availability of observations. At seven sites, the model results exceed the observations with a mean difference of 7 ppb (16 %). At the other eight sites, the model results are smaller than the observations with a mean difference of 7 ppb (11 %). These differences reflect model biases as well as a sampling bias due to lack of knowledge on which days contain valid observations.

Table 4 further compares the modeled monthly mean daily mean ozone in spring 2008 to the observations in various years collected from the literature. Again, the comparison is affected by a sampling bias. Although not our primary focus, this extended comparison gives a sense of how model ozone is situated in the general ozone pollution phenomena in China. The model reproduces the average magnitude of ozone at the three mountainous sites (Mts. Tai, Hua, and Huang) with a mean bias below 5 ppb (9 %). The model has a large overestimate by 48 % at the Hok Tsui coastal rural site in Hong Kong (36.0 versus 53.4 ppb), although the times are different (2008 versus 1994–2007). Wang et al. (2009) shows that the springtime ozone concentration at this site increased from 1994 to 2007 at a rate of 0.41 ppb yr-1, partly explaining this difference. The remaining difference may reflect that the model resolution is not able to represent the complex local terrain and land–sea contrast at this site. The model overestimates ozone at an urban site in Nanjing by 16 %, although the observations were made in 2000–2002 when Chinese anthropogenic emissions of NOx were only about half of those in 2008 (Xia et al., 2016).

We also evaluate the modeled daily average CO at six sites within and outside China with available hourly observations (Fig. 5). Overall, the model captures the day-to-day variability in daily mean CO fairly well (R=0.40 at Lin'an, 0.60 at Shangri-La, 0.56 at Ryori, and 0.73–0.82 at the other three sites). It has a small mean bias (within 4 %) at Bukit Koto Tabang, and Ryori, although with negative biases (by 13 %–33 %) at the other four sites. Such an underestimate is typical in global simulations (Young et al., 2013), and it may be related to excessive OH (Young et al., 2013; Yan et al., 2014, 2016) and/or underestimated emissions (Kopacz et al., 2010; Wang et al., 2011). Compared to the coarse-resolution global model alone, our two-way coupling results in less CO underestimation (Yan et al., 2014), although it does not eliminate the bias.

Figure 6a–c compare modeled ozone in 2008 to MOZAIC data over 2000–2005 at the airports of Beijing, Shanghai, and Hong Kong. Although model and MOZAIC data are in different years, to achieve the best sampling consistency, we sample the model results at times of day when the commercial aircrafts take off or land with available MOZAIC data. The timing information is shown in Fig. 6. GEOS-Chem reproduces the vertical gradient of MOZAIC ozone in general. The model underestimates MOZAIC ozone in the PBL over the Beijing airport mainly due to inconsistent temporal sampling, as further comparison with GPSO3 ozonesonde data (Bian et al., 2007; Wang et al., 2012), in which model results are sampled at times coincident with the observations, show little model bias (within 4 %, Fig. 6g). Over Hong Kong, the model captures the weak vertical gradient between 2 and 11 km, although it has a positive bias below 2 km due to its inability to capture the complex terrains and local pollution source characteristics around the airport. The model overestimates ozone in the middle and upper troposphere over Shanghai, with larger biases at higher altitudes, likely indicating too strong STE. Other causes may include differences in meteorology and growth in emissions between 2000–2005 and 2008, as discussed for the surface ozone in Sect. 3.1.

Figure 7 compares the modeled ozone profiles to WOUDC data at six sites. Here model results are sampled at ozonesonde launch times, and ozonesonde data are regridded to match the model vertical resolution. Overall, GEOS-Chem captures the vertical gradient of ozone fairly well. The model reproduces the overall weak vertical gradients at Hanoi, Hong Kong, Sepang, and Naha. It also reproduces the rapid increases above 8 km at Sapporo and Tateno, although it has positive biases at 10–20 ppb. GEOS-Chem reproduces the observed middle and upper tropospheric ozone at Hong Kong and Sepang, although it has an overestimate in the lower troposphere, consistent with the bias shown in Fig. 6c.

Figure 6d–f also compare the modeled CO with the MOZAIC data. Similar to the evaluation results for surface CO, GEOS-Chem generally underestimates the MOZAIC CO at most heights above the three airports, although it captures the vertical shape fairly well.

Our simulation has a small NMB for surface ozone, about 10 % averaged over 10 sites with hourly data (Fig. 4 and Table 3) and about 15 % averaged over 21 sites with monthly data from EANET and the literature (Table 4). The model also captures the general vertical distribution of ozone at 10 locations over China and nearby regions, with a tropospheric mean bias at 12 %. These agreements allow using the model for source attribution studies in the next sections. Conversely, with a horizontal resolution of about 50 km over Asia, the model often fails to simulate the complex terrains, local meteorological conditions, and/or local emission characteristics at several hilly or airport sites. The model also tends to overestimate the STE influences over Asia. Addressing these issues warrants future research with improved model resolutions and STE representation.

GEOS-Chem tends to underestimate CO over Asia (by 20 % on average), similar to many other models (Kopacz et al., 2010; Young et al., 2013). We conduct a sensitivity simulation by doubling Chinese anthropogenic CO emissions, which results in a slight increase in surface ozone by 0.1–0.4 and 2–3 ppb over clean and polluted areas of China, respectively. The low sensitivity of ozone to CO emissions was also found by Z. Jiang et al. (2015). We thus conclude that our ozone simulations over China are influenced insignificantly by the underestimate in CO.

Figure 8a shows the modeled spatial distribution of near-surface daily mean ozone in spring 2008 over China from all natural and anthropogenic sources, i.e., the CTL case. Ozone concentrations reach 75–80 ppb over the southern Tibetan Plateau, and they are minimum (25–40 ppb) over the North China Plain and many populous cities across eastern China. Ozone is about 45–60 ppb over the vast southeast, northwest, and northeast.

The simulated natural ozone (i.e., without anthropogenic emissions worldwide, the xANTH case) shows a strong gradient from the southern Tibetan Plateau (65–75 ppb) to the northwest (35–40 ppb) and the east (20–35 ppb) (Fig. 8c). Wang et al. (2011) show similar gradients of natural ozone in 2006. Natural ozone contributes 80 %–90 % of total surface ozone over Tibet and the northwest with low local anthropogenic emissions. The large natural ozone concentrations over Tibet are a result of vertical transport from the free troposphere and stratosphere due to Tibet's high altitudes and hilly terrains (that are conducive to vertical exchange) (Ding and Wang, 2006; Lin et al., 2015; Xu et al., 2017). They pose potential threats for public health and ecosystems there.

The simulated background ozone (i.e., without Chinese anthropogenic emissions, the xCH case) is shown in Fig. 8b. The background ozone is higher than the natural ozone by 2–11 ppb over most Chinese regions (Fig. 9b). This indicates large influences of foreign anthropogenic emissions through atmospheric transport of ozone and its precursors, as discussed in detail below.

Figure 9a shows the spatial distribution of domestic anthropogenic contributions to daily mean surface ozone over China (difference between the control run and the sensitivity simulation, CTL–xCH, followed by a linear weighting adjustment). Over most of the west and northeast, Chinese anthropogenic emissions are relatively low, and they result in ozone concentrations of 0–4 ppb. In contrast, domestic contributions reach 16–25 ppb over the south due to more emissions and favorable conditions for photochemistry. Over the North China Plain and many populous cities, Chinese anthropogenic emissions lead to reductions (instead of enhancements) of surface ozone. This is because of a weak ozone production efficiency and a strong titration effect by excessive domestic NOx emissions. Figure 9d–f show that when Ox (=O3+NO2) is considered, Chinese anthropogenic contributions vary from 2–4 ppb over the west to 6–12 ppb over the North China Plain and 20–35 ppb over the southeast (Fig. 9d).

Figure 9b shows the simulated contributions to Chinese surface ozone by all foreign anthropogenic emissions. Foreign contributions reach 7–11 ppb along much of the Chinese borders and they exceed 6 ppb over the vast northern regions. The foreign contribution reduces from the border to the inner areas, with a minimum (2–3 ppb) over the Sichuan Basin where the air is more isolated. In terms of anthropogenic ozone, foreign contributions account for up to 90 % over most of western and northeastern China (Fig. 9c), consistent with the findings by Li et al. (2015) for western China in 2000. Foreign anthropogenic contributions to Ox over China are similar to their contributions to ozone (Fig. 9e), except at places with strong Chinese NOx emissions that lead to titration of ozone.

Figure 10 further shows the contributions to Chinese surface ozone by anthropogenic emissions in seven individual foreign regions. The pattern of influence differs among these source regions due to differences in the location of source region, emission magnitude, pollutant lifetimes, and transport pathways. Anthropogenic emissions in Japan and Korea result in 0.6–2.1 ppb of ozone enhancement along the Chinese coast. The tagged ozone simulation with NAQPMS by J. Li et al. (2016) also showed that about 0.5–3.0 ppb of ozone over northeastern China in spring 2010 was produced over the Korean Peninsula, although there is a difference between ozone produced over a region and ozone produced from that region's emissions. Emissions from Southeast Asia contribute 1–5 ppb over much of the southern provinces. Emissions from South Asia mostly affect southwestern China and Tibet (by up to 5–10 ppb over the border), due to effective transport by strong southwesterly winds associated with the Indian monsoon. The rest of Asia consists of many countries to the west of China, whose total contributions are about 2–5 ppb over much of northwestern China.

European anthropogenic emissions contribute 2.1–3.0 ppb of ozone along the northern border of China. The contributions decrease southward and are above 1 ppb over half of Chinese land areas. The Model for Ozone and Related chemical Tracers (MOZART) simulations by Li et al. (2015) also showed a European contribution of 2 ppb to surface ozone over northern China in 2000. North American anthropogenic emissions increase ozone by 1.8–2.7 ppb over much of western China, by 1.5–2.1 ppb over the populous North China Plain, and by less than 0.9 ppb over the south. The contributions are smaller than springtime Asian anthropogenic influences on western North America (e.g., 1–5 ppb averaged over 2001–2005; Brown-Steiner and Hess, 2011b), although the affected population is larger by roughly an order of magnitude.

Influences from the rest of the world are about 0.6–1.2 ppb over Tibet and smaller over other Chinese land territory. The larger values over Tibet reflect its higher altitude and greater sensitivity to long-range transport via the free troposphere.

Figure 11a shows whether domestic or foreign anthropogenic contributions are higher at individual locations. Domestic anthropogenic contributions are higher than foreign contributions over southern China and parts of northern China. However, foreign anthropogenic contributions exceed domestic contributions over western China and most of the north, including the populated North China Plain. Over western China, foreign emissions contribute 70 %–90 % of the total anthropogenic ozone (Fig. 9c).

Figure 11b further highlights the largest foreign contributor to surface anthropogenic ozone at each location of China. North America is the largest foreign contributor over about half of Chinese land territory, including the populated North China Plain. Europe is the largest foreign contributor for the vast northeastern region, the rest of Asia for the western border region, South Asia for southwestern China, Southeast Asia for southern China, and Japan and Korea for the eastern coast of China.

The HTAP and several other studies have used 20 % perturbation simulations (i.e., reducing anthropogenic emissions in each source region by 20 %) to study the transboundary ozone problem. Such studies are source–receptor analyses that are more relevant to the question of how much a modest cut in foreign emissions would reduce ozone pollution over a targeted receptor region. To compare with such a method, here we ran one more set of full-chemistry simulations by decreasing 20 % anthropogenic emissions over each of the eight emission source regions (similar to the zero-out method in Table 1). We also applied the linear weighting method to account for the nonlinearity of ozone chemistry. Figures 9a and 12a compare the Chinese anthropogenic contributed ozone calculated from zero-out and from 20 % perturbation simulations. Compared to the zero-out method, the 20 % perturbation method leads to less Chinese contributed ozone, with negative values over more regions and smaller positive values over southern China. This result confirms our general finding that in spring 2008, the excessive domestic NOx emissions lead to relatively weak ozone production and/or strong ozone titration. Comparing with the zero-out method, the absolute foreign anthropogenic ozone obtained from 20 % perturbation simulations is smaller by 2–3 ppb over the northern border of China (comparing Figs. 9b and 12b), whereas the percentage of foreign contributions increase from 10 %–20 % to 20 %–40 % over southeastern China (comparing Figs. 9c and 12c). Nonetheless, the spatial patterns are similar between the two methods for both the absolute and the relative foreign contributions.

As peak ozone is a critical problem for human health, here we show the domestic versus foreign contributions to modeled extreme ozone values in spring 2008 (defined as the average of the top 5 % hourly ozone concentrations) (Fig. 12d–f). As expected, the Chinese domestic contribution is larger for extreme ozone than for mean ozone; the negative values also disappear over the North China Plain and northeastern China (comparing Figs. 9a and 12d). The absolute foreign contribution (in parts per billion) is also enhanced across China (comparing Figs. 9b and 12e). The percentage foreign contribution is within 10 % over southern China, about 10 %–50 % over the north, and above 70 % over the west. Nevertheless, these results for extreme ozone should be interpreted with more caution, as the model cannot simulate the dates of extreme ozone very well (Fig. 4).

Previous studies have shown notable interannual variability in surface ozone over China driven by changes in precursor emissions and meteorology (Xu et al., 2008; Jin et al., 2015; Wang et al., 2017). To test how the interannual variability in meteorology and emissions would affect our source attribution findings, we have repeated all zero-out runs for spring 2012, the latest year when the GEOS-5 meteorological fields are available. Emissions for 2012 were adopted from the Community Emissions Data System (CEDS) inventory (Hoesly et al., 2018); 2012 is also the latest year the CEDS emissions for China are adjusted by the MEIC inventory. Table 5 shows the anthropogenic emissions in the two years. All zero-out simulation results in 2012 underwent the same linear weighting adjustment as for those in 2008. Figure 12g–i show the results for domestic versus foreign contributed ozone in spring 2012, compared to the results for spring 2008 (Fig. 9a–c). In absolute terms, Chinese contributed ozone is similar between 2008 and 2012 (comparing Figs. 12g and 9a), reflecting the slight changes in domestic precursor emissions (Table 5). From 2008 to 2012, the absolute foreign contributed ozone increased along the southern boarder due to much enhanced emissions in Southeast Asia and South Asia. The absolute foreign contributions decrease over the north and south, reflecting the net effect of changes in European and North American emissions (within 20 % for both NOx and NMVOC), increased emissions in the rest of Asia, and changes in meteorology. In relative terms (Figs. 9c and 12i), the percentage of foreign anthropogenic contributions to total anthropogenic ozone decrease from 2008 to 2012 over southern China. Nonetheless, in both years the percentage of foreign contributions exceeds 50 % over western China and is 5 %–40 % over southern China. Therefore our general finding that both foreign and domestic contributions to Chinese anthropogenic ozone are important holds true for these two years.