Attribution of ground-level ozone to anthropogenic and natural sources of NO x and reactive carbon in a global chemical transport model

. We perform a source attribution for tropospheric and ground-level ozone using a novel technique which accounts separately for the contributions of the two chemically distinct emitted precursors (reactive carbon and oxides of nitrogen) to the chemical production of ozone in the troposphere. By tagging anthropogenic emissions of these precursors according to the geographical region from which they are emitted, we determine source/receptor relationships for ground-level ozone. Our methodology reproduces earlier results obtained through other techniques for ozone source attribution, and also delivers 5 additional information about the modelled processes responsible for intercontinental transport of ozone, which is especially strong during the spring months. The current generation of chemical transport models used to support international negoti-ations aimed at reducing the intercontinental transport of ozone show especially strong inter-model differences in simulated springtime ozone. Current models also simulate a large range of different responses of surface ozone to methane, one of the major precursors of ground-level ozone. Using our novel source attribution technique, we show that emissions of NO x from 10 international shipping over the high seas play a disproportionately strong role in our model system to the hemispheric-scale response of surface ozone to changes in methane, as well as to the springtime maximum in intercontinental transport of ozone and its precursors. We recommend a renewed focus on improvement of the representation of the chemistry of ship NO x emissions in current-generation models. We demonstrate the utility of ozone source attribution as a powerful model diagnostic tool, and recommend that similar source attribution techniques become a standard part of future model inter-comparison studies. diagnostic information about the origin and budget of springtime ozone in our model, along with information about the springtime budget of peroxyacetyl nitrate (PAN), which is also associated with springtime 555 long-range transport and ozone production. We show that a substantial proportion of the free-tropospheric PAN simulated by our model in spring is not produced in the polluted boundary layer over the major anthropogenic source regions, but is rather produced in our model downwind of these regions through the interaction of transported anthropogenic reactive carbon and NO x emitted from international shipping. Reactive carbon of anthropogenic origin (and its oxidation products, including PAN) builds up in our model across the entire Northern Hemisphere during the winter months, and then contributes in our simulations 560 to a short burst of hemispheric-scale ozone production during spring. In all but the most polluted source regions, anthropogenic NMVOC do not make a signiﬁcant contribution to simulated ground level ozone in any other season but spring. We showed here that export of anthropogenic reactive carbon from East Asia may be playing a dominant role in contributing to the build up of reactive carbon in the Northern Hemisphere over winter, and thus to the hemispheric-scale production of

With the exception of surface-based anthropogenic emissions of NO x , CO, and NMVOC, the tag identities used in this study 160 are identical to those used in Butler et al. (2018). In this study, all surface-based anthropogenic emissions are tagged with a label representing the geographical location at which the emissions occur. This approach allows attribution of simulated ozone to anthropogenic precursor emissions from specific locations. Specifically, anthropogenic emissions of NO x and reactive carbon are tagged according to their Tier 1 Source Region as defined for the HTAP phase 2 multi-model ensemble experiment, which is described in more detail in (Galmarini et al., 2017). Due to computational constraints, not all of the HTAP Tier 1 regions 165 are tagged in this study. Since the primary focus of this study is on the attribution of ground-level ozone in the Northern Hemisphere, only the major anthropogenic Northern Hemisphere source regions are tagged, while other anthropogenic sources are tagged with the label "Rest of the World". A full list of the tags used in the NO x -and VOC-tagged runs is given in Table 1. The explicitly tagged source regions differ between the NO x -tagged and VOC-tagged runs because VOC tagging is computationally more expensive than the NO x -tagging (Butler et al., 2018). One important difference between this study 170 and Butler et al. (2018) is that anthropogenic emissions of CO for each source region are tagged together with emissions of NMVOC in this study in order to save computational resources.
In addition to the NO x -and VOC-tagged base runs described above, we also perform two additional runs in order to investigate the response of tropospheric ozone to a perturbation in the tropospheric burden of methane: one with NO x tagging; and another with VOC tagging. In each of these methane perturbation runs, the initial atmospheric methane burden and the 175 methane mixing ratio imposed at the surface as a boundary condition are reduced by 20%. This translates to a surface methane mixing ratio of 1410 ppb in these methane perturbation runs. In these methane perturbation runs, all other sources of NO x and reactive carbon are left unchanged. The methane perturbation runs also require two years of spinup for the model to arrive at steady state.
CAM4-chem in version 1.2.2 of the CESM has previously been evaluated by Tilmes et al. (2015), and the modified version 180 used in this study has also been discussed thoroughly by Butler et al. (2018). In Section 3, we describe the key differences in methane and tropospheric ozone between our base simulation and the CAM4-chem simulation reported by Tilmes et al. (2015), and compare our simulated surface ozone with observations from TOAR (Schultz et al., 2017) as well as with the ensemble of CTM simulations from the HTAP phase 2 multi-model study (Galmarini et al., 2017). The full set of CTMs participating in the HTAP phase 2 multi-model ensemble is given in Table 3 of Galmarini et al. (2017). In this study we compare surface soil NO x , and ozone input from the stratosphere all contribute additionally to modelled global background ozone. Emissions of NO x from shipping contribute significantly to ozone over the major northern hemisphere ocean basins, which is also transported over continental regions. South Asia stands out in comparison with the other major Northern Hemisphere source regions, in that ozone produced from NO x emitted in South Asia is relatively localised to the South Asian region itself, and not transported into the hemispheric background to the same extent as ozone produced from NO x emissions in the other major Northern 230 Hemisphere source regions. Table 2 shows that NO x emissions from lightning and aircraft are especially efficient at producing ozone in the free troposphere, consistent with previous work (eg. Beck et al., 1992;Jacob et al., 1996). Similarly, surface emissions of NO x from regions closer to the tropics (eg. South East Asia and Middle America) produce ozone more effectively due to rapid convective transport of emitted NO x into the free troposphere, consistent with Zhang et al. (2016). Of the major Northern Hemisphere 235 source regions, NO x emissions from South Asia are the most efficient at producing ozone, consistent with a stronger role of vertical transport over this region. In contrast, NO x emissions from the major anthropogenic source regions in the high northern latitudes (Europe, East Asia, and North America) are among the least productive of all global NO x emissions, consistent with a relatively small amount of convective transport, leading to higher rates of NO x removal. Despite their low ozone production efficiency, emissions of NO x in the high northern latitudes contribute significantly to surface ozone across the northern 240 hemisphere ( Figure 2 and Table 2).
Table 2 also shows that NO x emissions from shipping are also relatively efficient at producing ozone, which is also consistent with previous work (eg. Lawrence and Crutzen, 1999;Hoor et al., 2009). The high ozone production efficiency of ship emissions is due to their location in relatively pristine regions with few other sources of NO x . Due to the high ozone productivity of ship emissions, and being emitted at relatively high latitudes, they contribute significantly to the Northern Hemispheric 245 background ( Figure 2 and Table 2). As noted above, the ozone production from ship NO x is likely to be overestimated due to the artificial dilution of emissions into relatively coarse model grid cells. Mertens et al. (2018) report a contribution of shipping to the tropospheric ozone burden of 18 Tg(O 3 ) using their tagging technique, and based on a model simulation with ship NO x emissions of 6 Tg(N)yr − 1. In our study, we calculate a contribution of ship NO x to tropospheric ozone of 19.9 Tg(O 3 ) based on ship NO x emissions of 4.28 Tg(N)yr − 1, implying a much higher 250 ozone production efficiency for ship NO x in our study. Since the tagging technique used by Mertens et al. (2018) is based on the technique described by Grewe et al. (2017), which combines the effects of tagged NO x and reactive carbon precursors into a single tagged ozone molecule during ozone production, we do not expect our results to be directly comparable. Since shipping emits significantly more NO x than reactive carbon, we would expect the combinatorial tagging approach of Mertens et al. (2018) to attribute less ozone to shipping than our method, as the ozone produced from ship NO x would also be partially 255 attributed to the reactive carbon precursor involved in the ozone production. Indeed, Mertens et al. (2018) report maximum contributions of shipping to surface ozone of about 10 ppb in summer over major Northern Hemisphere ocean basins. In our study, surface ozone attributable to shipping over these regions can exceed 20 ppb (see the Supplementary Material).

Attribution to reactive carbon emissions
Methane and biogenic emissions clearly stand out as major reactive carbon precursors to tropospheric ozone, contributing 260 35% and 24% respectively to the tropospheric ozone burden in our simulation. Anthropogenic emissions of reactive carbon (excluding biomass burning) together contribute about 14 % to the tropospheric ozone burden. The relatively low influence of anthropogenic reactive carbon emissions on ground-level ozone has been noted elsewhere (eg. HTAP, 2010;Butler et al., 2018), but despite this low overall ozone productivity, anthropogenic reactive carbon emissions from source regions in higher northern latitudes still contribute disproportionately highly to surface ozone in the Northern Hemisphere (Table 3).

265
Due to the emissions of CO being tagged together with emitted VOC in this study, the contribution of each tagged source to the tropospheric ozone burden (and therefore also the ozone production efficiency of each tagged source) is a mixture of ozone production due to emitted CO and emitted NMVOC. The ozone attributed to methane oxidation in Table 3 is due do all stages of methane oxidation in the MOZART-4 chemical mechanism, including the final step in which CO from earlier stages of methane oxidation is itself oxidised to CO 2 . The oxidation of CO can produce at maximum one peroxy radical 270 (HO 2 ). The maximum ozone production potential of CO is therefore 1 mole of ozone per mole of emitted CO. VOC (including methane) can produce significantly more ozone per mole emitted carbon, when taking into account the subsequent oxidation of the initial oxidation products (Butler et al., 2011). Future studies using this tagging methodology should consider tagging CO emissions separately from NMVOC emissions if they aim to determine the ozone production efficiency of anthropogenic NMVOC emissions from different world regions. Butler et al. (2018) did tag NMVOC emissions separately from CO emissions, 275 but did not tag anthropogenic emissions separately according to their geographical region. We reexamined the output of the otherwise identical VOC-tagged run described by Butler et al. (2018) in order to determine the ozone production efficiency of NMVOC emissions from anthropogenic, biomass burning, and biogenic sources. Respectively, these are 0.0580, 0.0354, and 0.0268 (mol(O 3 )/mol(C)). The ozone production efficiency of biogenic NMVOC recalculated from Butler et al. (2018) is not significantly different from the value reported here in Table 3, reflecting the relative minor contribution of CO to the total 280 amount of emitted biogenic reactive carbon. For biomass burning and anthropogenic sources however, the ozone production efficiency of NMVOC emitted from these sources is greater than the corresponding value from Table 3, reflecting the fact that the numbers from Table 3 also include emissions of CO.
Interestingly, we note that methane has a higher ozone production efficiency (0.0689 mol mol −1 , Table 3) than any of the NMVOC in our runs. The low ozone production efficiency of biogenic NMVOC is consistent with large amounts of isoprene 285 being emitted in remote regions under low-NO x conditions, where loss of peroxy radicals through reaction with other peroxy radicals could be expected to dominate (Atkinson, 2000). It might however be expected that anthropogenic NMVOC would have a higher ozone production efficiency, due to their being co-emitted with anthropogenic NO x , favouring the conversion of NO to NO 2 through reaction with peroxy radicals, and thus the production of ozone. The relatively low production efficiency of anthropogenic NMVOC in our model runs could be due to the relatively simple chemistry of methane oxidation being 290 well-described in the version of the MOZART-4 chemical mechanism used here, in which the relatively complex chemistry of the higher NMVOC has been simplified. Coates and Butler (2015) noted that the ozone production potential of NMVOC https://doi.org/10.5194/acp-2020-436 Preprint. Discussion started: 20 May 2020 c Author(s) 2020. CC BY 4.0 License. in simplified chemical mechanisms tended to be lower than the more comprehensive Master Chemical Mechanism (Saunders et al., 2003). Utembe et al. (2010) previously noted increased tropospheric ozone in a CTM when using a more explicit oxidation mechanism for NMVOC. The extremely low ozone production efficiency of reactive carbon from oceanic sources 295 in Table 3 is due to the lack of any ozone forming pathways in the oxidation of dimethyl sulphide (DMS) in the MOZART-4 chemical mechanism as used in this study. DMS is the dominant source of reactive carbon over the oceans in our model simulations.
To our knowledge, the only other study to perform source attribution of global tropospheric ozone specifically to reactive carbon precursors is Butler et al. (2018), on which the present study builds. Here, we attribute 113 Tg of ozone to methane 300 oxidation (Table 3). Grewe et al. (2017) attribute 45 Tg of ozone to methane using their tagging approach, which combines the effects of tagged NO x and reactive carbon precursors into a single tagged ozone molecule during ozone production. Ozone production due to methane oxidation under their combinatorial tagging approach would be expected to also include attribution to the source of NO x involved in the ozone production. We would thus expect Grewe et al. (2017) (Table 2), except in South Asia and East Asia, where the annual average surface ozone mixing ratio is closer to 40 ppb. The difference in each case is primarily due to a larger source of ozone produced from locally emitted precursors. Transport from the stratosphere contributes approximately 2-4 ppb to annual average surface ozone depending on the receptor region, consistent with the 2.91 ppb contribution of stratospheric ozone to the annual average surface ozone in the Northern Hemisphere average surface ozone (Table 3). As already shown in the previous section, anthropogenic sources of 320 NO x dominate other NO x sources as ozone precursors, while the major reactive carbon precursors are methane and BVOC.
In each of the five regions shown in Figure 4, natural sources and long-range transport of ozone produced from extra-regional anthropogenic precursors contribute more to the annual average surface ozone than anthropogenic emissions within the region itself. In each region, the local anthropogenic NO x emissions produce more ozone than can be attributed to anthropogenic NO x emissions in any other Tier 1 regions, but with only one exception (South Asia), the combined contribution of external 325 anthropogenic NO x emissions to annual average surface ozone is greater than the local contribution. The importance of longrange transboundary transport of ozone has been noted elsewhere (HTAP, 2010).
While anthropogenic precursor emissions from South Asia contribute significantly to surface ozone within the South Asia region, they contribute relatively little to surface ozone in the other four regions shown in Figure 4. This is also consistent with the surface ozone maps in Figure 2, and the higher ozone productivity of NO x emissions from South Asia when compared with 330 the other major Northern Hemisphere source regions ( Table 2). Emissions of ozone precursors (particularly NO x ) from South Asia are transported efficiently into the free troposphere, where they contribute disproportionately to the global tropospheric ozone burden (as also noted by Zhang et al., 2016), but the contribution of South Asian emissions to surface ozone in other parts of the Northern Hemisphere is disproportionately smaller than emissions from the other HTAP Tier 1 regions. All three receptor regions show a seasonal cycle of ozone with a spring-summer ozone maximum superimposed on a year-345 round ozone baseline. The summertime maximum in ozone is clearly due to local photochemical production from the combination of locally-emitted anthropogenic NO x and biogenic VOC. The strong role of locally-emitted precursors in the production of ozone in summer is consistent with earlier work (eg. Reidmiller et al., 2009;Huang et al., 2017;Jonson et al., 2018;Han et al., 2019), while the importance of biogenic VOC emissions, especially isoprene, for ozone production in summer has also been noted elsewhere (eg. Chameides et al., 1992;Andersson and Engardt, 2010;Han et al., 2019). Biogenic emissions of NO x 350 (from soils) also contribute to this summertime maximum in local photochemical ozone production in all three of the regions shown in Figures 5, 6, and 7, but to a much smaller extent than anthropogenic NO x emissions.

Seasonal cycles of surface ozone
The year-round baseline ozone in our model simulations in all three receptor regions can be primarily explained by slower photochemistry involving methane as the reactive carbon precursor, in combination with extra-regional anthropogenic NO x ( Figures 5, 6, and 7). Both show a minimum contribution in winter of about 10 ppb in all three receptor regions. The contribution 355 of methane to surface ozone is slightly larger in summer, coinciding with the peak in local anthropogenic NO x emissions, consistent with local photochemical ozone production from enhanced local methane oxidation. The contribution of extraregional anthropogenic NO x to surface ozone is largest in spring, coinciding with the peak in the contribution of extra-regional anthropogenic reactive carbon, consistent with long-range transboundary transport of ozone produced elsewhere. (HTAP, 2010;Lin et al., 2012;Jonson et al., 2018;Ni et al., 2018). This transported ozone can be attributed to input from the stratosphere, as well as extra-regional anthropogenic emissions of NO x and reactive carbon. In our simulations, the contribution of stratospheric ozone peaks around March, while the contribution of extra-regional anthropogenic emissions tends to peak around April, when it contributes more strongly to monthly average surface ozone in each region than local anthropogenic NO x ( Figures 5, 6, and 7). In all regions, the springtime peak in the contribution of extra-regional anthropogenic reactive carbon is 365 smaller, but more pronounced than the corresponding springtime peak in the contribution of extra-regional anthropogenic NO x .
Previous work has identified uncertainties in the treatment of ozone production from NMVOC oxidation as a potential source of inter-model differences (eg. Emmerson and Evans, 2009;Utembe et al., 2010;Coates and Butler, 2015). The relatively large influence of anthropogenic NMVOC on springtime ozone (compared with its influence during other time of the year) could be a contributing factor to the large spread in springtime ozone simulated by current generation CTMs (Figure 1).

370
NO x from shipping is the largest single contributor to springtime transboundary ozone transport in all three receptor regions shown here. We note however that the coarse resolution of our model (2 degrees) would be expected to exaggerate the effects of ship NO x on ozone production due to rapid dilution of the emissions (von Glasow et al., 2003), as well as exaggerate the transport of NO x and ozone near coastlines due to unrealistically high diffusion between adjacent land and ocean grid cells. The shows a similar magnitude for the influence of ship NO x on summertime ozone in Europe as for springtime ozone. Lupaşcu and Butler (2019), using a regional model at 50×50 km resolution and a similar ozone tagging system as used in the present study, showed that the contribution of ship NO x to ozone in coastal regions of Europe reaches a maximum level in summer. Jonson et al. (2020), using a global model with a resolution of 0.5 × 0.5 degrees showed that shipping near coastal regions contributes significantly to ozone over North West Europe in both spring and summer, while NO x emissions from shipping on 385 the high seas makes a stronger contribution to European ozone in spring than in summer. In contrast, the study of Aksoyoglu et al. (2016), using a higher-resolution regional model (20 ×20 km) for Europe, does not indicate such a strong role for ship NO x on summertime ozone over Europe.
In other receptor regions, the influence of ship NO x emissions on surface ozone is largest in spring (Figures 6 and 7), suggesting a stronger influence of NO x emissions over the high seas on springtime ozone in our simulations in these regions.  (Table 3). Reactive carbon emissions from East Asia contribute 405 approximately equally to springtime ozone in North America as East Asian NO x emissions, and in Europe, East Asian reactive carbon contributes more to springtime ozone than East Asian NO x . Our model simulations with NO x -and VOC tagging provide a unique opportunity to examine the origin and fate of PAN as simulated in our model, since this allows simultaneous attribution of simulated PAN to both is NO x precursor and its reactive carbon precursor. Comparison of the bottom two panels in each of Figures 8, 9, 10 shows consistently that for any 420 given land-based HTAP Tier 1 source region, the anthropogenic NMVOC emissions contribute more to PAN formation than the anthropogenic NO x emissions from that region to the PAN modelled in all HTAP Tier 2 receptor regions. The balance of extra-regional PAN in all cases is due to NO x emissions from shipping. In our simulations, significant amounts of PAN  Figures 8, 9, and 10 also show that the reactive carbon component of PAN is generally more persistent than the NO x component. For example, the contribution of anthropogenic NMVOC from North America to springtime PAN over East Asia is only slightly lower than its contribution to springtime PAN over Europe, which is much closer to North America considering 430 the prevailing westerly winds (bottom right panels of Figures 8 and 9). In contrast, the contribution of anthropogenic NO x from North America to springtime PAN in East Asia is substantially less than its contribution to springtime PAN over Europe (bottom left panels of Figures 8 and 9).

Long-range transport of ozone precursors
We examine the Northern Hemisphere budget of reactive carbon in more detail in Figure 11. This figure shows the seasonal cycle of the Northern Hemisphere column-integrated total reactivity with respect to the OH radical of all reactive carbon 435 containing species in our simulation, attributed to their emission source. The total OH reactivity of reactive carbon species of an airmass is often linked to its ozone production potential (Chameides et al., 1992;Kleinman et al., 2002). The OH reactivities shown in Figure 11 include in each case the OH reactivity of the primary emitted species, as well as the OH reactivity of each carbon-containing oxidation product. These were calculated using monthly averaged output of the modelled concentration of each carbon-containing species (including its associated tags), and the temperature-and pressure-dependent rate coefficients 440 for their reaction with the OH radical, then averaged over all Northern Hemisphere grid cells, weighted by air density.
The total Northern Hemisphere OH reactivity of reactive carbon remains fairly constant year-round at about 0.6 -0.7 s −1 , but the seasonal cycles of the OH reactivity attributable to different reactive carbon sources show more variability. Methane (and its oxidation products) contribute about 0.2 -0.3 s −1 (almost half of the total hemispheric reactivity), with a slight maximum in the summer, consistent with enhanced oxidation (and thus enhanced availability of more reactive methane oxidation products) 445 due to higher OH in summer. The contributions of anthropogenic and biogenic reactive carbon sources to total hemispheric reactivity are similar, ranging between about 0.1 -0.3 s −1 , but with distinct seasonal cycles. The reactivity of biogenic carbon is highest in summer-autumn (consistent with the Northern Hemisphere growing season), while reactivity of anthropogenic carbon is highest in winter-spring (consistent with constant year-round anthropogenic emissions, and a build-up of reactive carbon over winter due to lower hemispheric OH). The build-up of anthropogenic reactive carbon throughout the Northern 450 Hemisphere over winter, combined with the resumption of OH chemistry in spring is consistent with the disproportionate effect of extra-regional anthropogenic reactive carbon on springtime ozone seen in Figures 5, 6, and 7. Uncertainties in the model chemical mechanisms associated with the oxidation of anthropogenic NMVOC (eg. Emmerson and Evans, 2009;Utembe et al., 2010;Coates and Butler, 2015) may thus also contribute to the large spread in simulated ozone seen in the HTAP ensemble during spring (Figure 1). 455 Figure 11 also shows the geographical origin of Northern Hemisphere anthropogenic carbon reactivity. Emissions of reactive carbon from East Asia stand out as the single major source of enhanced anthropogenic carbon reactivity in winter and spring in our simulations. This is consistent with the high emissions of reactive carbon from this region in 2010 noted earlier (Table 3). 2018) should be associated with increased oxidation of reactive carbon, and thus potentially less export of reactive carbon into the Northern Hemisphere background during summer. We expect, however, that increasing emissions of reactive carbon in East to increased East Asian contribution to extra-regional springtime ozone in other parts of the Northern Hemisphere.
Our tagging technique is currently the only one we know of which is capable of examining the budget of reactive carbon in 465 the level of detail presented in this study. The separate tracking of the carbon-containing and nitrogen-containing components of PAN is particularly informative, suggesting that significant amounts of PAN are formed downwind of source regions in our model, especially during winter and spring, due to a build-up of anthropogenic reactive carbon over winter when photochemistry is relatively slow. Given the large variety in model representations of NMVOC chemistry, including PAN formation and decomposition processes (Emmerson and Evans, 2009;Knote et al., 2015) and the large inter-model differences in simulated

Tropospheric ozone sensitivity to methane
We performed an additional set of model runs with both NO x -and VOC-tagging with the methane surface boundary condition reduced from 1760 ppb to 1410 ppb, a reduction of 350 ppb, or 20%. This perturbation can also be expressed as an increase of 25%. Here we interpret the methane perturbation run in terms of the atmospheric response to a 25% increase in the methane 480 surface mixing ratio at steady state.
In response to the 25% increase in the imposed surface mixing ratio of methane, the total tropospheric burden increased by 776 Tg(CH 4 ), an increase of 23%. The strength of the annual tropospheric chemical sink of methane due to OH increased by 72.5 Tg(CH 4 ), or 15.2%. The corresponding increase in the methane lifetime was 0.48 years, or 6.75%. The relatively small growth in the chemical methane sink compared with the magnitude of the perturbation in methane itself is consistent with the 485 feedback of methane on its own lifetime due to depletion of OH (Prather, 1996). Table 4  The relative increase in tropospheric ozone due to methane (13.0%) is comparable to, but slightly smaller than the increase in the magnitude of the chemical methane sink due to OH (15.2%), consistent with the troposphere as a whole becoming slightly more NO x -limited with increasing methane. The relative increase in the total ozone burden (2.98%) is, however, significantly lower than the increase in the ozone produced from methane oxidation. When the methane burden is increased, 495 the contribution of every other reactive carbon source to the tropospheric ozone burden decreases (each by approximately 1 NO x -limited atmosphere with increasing methane. In a future with an increased methane burden, control of NMVOC emissions could be expected to be less effective at large-scale reduction in ground-level ozone. Table 5 shows the change in the contributions of different NO x sources to tropospheric ozone in response to the 25% 500 increase in methane burden. As expected, all NO x sources become more productive when the total atmospheric burden of reactive carbon is increased (consistent with the troposphere as a whole becoming more NO x -limited). The increase in the productivity of the different NO x sources under an increased burden of methane is however not uniform. Ozone production due to NO x from shipping stands out as highly sensitive to the global methane burden in our simulations. Ship NO x accounts for almost 30% of the 1 ppb increase in Northern Hemisphere average surface ozone when the methane burden is increased by 505 25% (Table 5), despite being a much smaller percentage of total global NO x emissions ( Table 2).
The spatial distribution of the increase in annual average surface ozone from ship NO x in response to the 25% increase in methane is similar to the spatial distribution of surface ozone due to ship NO x in our base run ( average surface ozone from ship NO x in the three HTAP Tier 2 regions examined here in response to the 25% increase in methane is similar to the seasonal cycle of surface ozone due to ship NO x in our base run (Figures 5, 6, and 7). The maximum response of surface ozone from ship NO x to rising methane is simulated over the major Northern Hemisphere ocean basins in summer (which in our simulations influences surface ozone in North West Europe, Figure 5), while the influence of this response over most Northern Hemisphere continental regions is generally higher in winter-spring (as seen in North East China, 515 Figure 6).
Previous work (Lawrence and Crutzen, 1999) has noted the disproportionate influence of ship NO x on tropospheric ozone due to the diffuse and widespread nature of this source over regions which would otherwise have very low mixing ratios of NO x . Fiore et al. (2008) noted that the response of surface ozone to increased methane was especially strong in ship tracks. Myhre et al. (2011) also showed that ship NO x emissions reduce the global methane lifetime much more than terrestrial NO x 520 emissions. We note again that the contribution of ship NO x to ozone in our simulations (as in most current-generation CTMs) is likely to be an overestimate due to the unrealistic dilution of these emissions into coarse model grid cells (von Glasow et al., 2003), and the lack of explicit plume chemistry (Vinken et al., 2011). We do expect however, that the interaction between ship NO x and methane for ozone production would persist in our model even with a more realistic treatment of ship emissions, since this interaction is likely due to the location, rather than the magnitude of ship emissions. We are not aware of any previous 525 work linking the combined influence of these two sources to a potentially disproportionate influence on background ozone in the Northern Hemisphere, and on modelled surface ozone air quality in inhabited regions of the Northern Hemisphere, especially in spring. Given the current uncertainty in attribution of recent trends in methane (Turner et al., 2019) and the potential for future increases in methane emissions, combined with slower reductions in NO x emissions from international shipping than from other sectors (eg. the SSP5 future emission scenario Rao et al., 2017), we expect that model simulations 530 of future background ozone in the Northern Hemisphere, especially during spring, may come to be increasingly influenced by ozone produced through the interaction of methane and ship NO x . Future work should investigate the ozone production through interaction of these two sources in more detail.

Conclusions
We have performed a source attribution for tropospheric ozone in a chemical transport model using a novel technique which 535 separately accounts for the influence of both the emitted NO x and the emitted reactive carbon precursors on simulated tropospheric ozone. By tagging anthropogenic emissions of NO x and reactive carbon according to their geographical region we have calculated source/receptor relationships for the Northern Hemisphere. The results of our study are consistent with previous work, and provide a number of important new insights of relevance to both the mitigation of intercontinental transboundary air pollution and ongoing efforts to reduce the uncertainty in the current generation of chemical transport models.

540
Consistent with previous work, annual average ground-level ozone in all major Northern Hemisphere regions is primarily influenced by extra-regional emissions of both NO x and reactive carbon. In all cases, local anthropogenic emissions of ozone precursors have a smaller influence on annual average ozone than the combined effect of precursor emissions from the rest of the world. As a reactive carbon precursor, methane contributes 35% of the tropospheric ozone burden, and 41% of the Northern Hemisphere annual average surface mixing ratio, more than any other source of reactive carbon. Our novel tagging 545 methodology also reproduces the well-known dependence of summer ozone maxima on local emissions of anthropogenic NO x and biogenic reactive carbon, and the enhanced importance of intercontinental transport of ozone from remote anthropogenic sources in spring. Consistent with previous work, we find that emissions of NO x at low latitudes produce free-tropospheric ozone more effectively due to more efficient vertical transport. We show, however, that NO x sources at higher northern latitudes have a stronger influence on ground-level ozone, which has a lower radiative forcing but a higher influence human health and 550 ecosystems.
The current generation of chemical transport models has particular difficulty in simulating the intercontinental transport of ozone, as shown by the large spread in ensemble simulations of ground-level ozone during the spring months. We show that our tagging methodology can deliver detailed diagnostic information about the origin and budget of springtime ozone in our model, along with information about the springtime budget of peroxyacetyl nitrate (PAN), which is also associated with springtime 555 long-range transport and ozone production. We show that a substantial proportion of the free-tropospheric PAN simulated by our model in spring is not produced in the polluted boundary layer over the major anthropogenic source regions, but is rather produced in our model downwind of these regions through the interaction of transported anthropogenic reactive carbon and NO x emitted from international shipping. Reactive carbon of anthropogenic origin (and its oxidation products, including PAN) builds up in our model across the entire Northern Hemisphere during the winter months, and then contributes in our simulations 560 to a short burst of hemispheric-scale ozone production during spring. In all but the most polluted source regions, anthropogenic NMVOC do not make a significant contribution to simulated ground level ozone in any other season but spring.
We showed here that export of anthropogenic reactive carbon from East Asia may be playing a dominant role in contributing to the build up of reactive carbon in the Northern Hemisphere over winter, and thus to the hemispheric-scale production of