The Impact of Future Emission Policies on Tropospheric Ozone using a Parameterised Approach

20 This study quantifies future changes in tropospheri c ozone (O3) using a simple parameterisation of source-recepto r relationships based on simulations from a range of m dels participating in the Task Force on Hemispher ic T ansport of Air Pollutants (TF-HTAP) experiments. Surface and tropo spheric O3 changes are calculated globally and across 16 regi ons from perturbations in precursor emissions (NO X, CO, VOCs) and methane (CH 4) abundance. A source attribution is provided for each source region along with an estimate of uncert ainty based on the spread of the results from the m odels. Tests against 25 model simulations using HadGEM2-ES confirm that the approaches used within the parameterisation are va lid. The O3 response to changes in CH 4 abundance is slightly larger in TF-HTAP Phase 2 th an in the TF-HTAP Phase 1 assessment (2010) and provides further evidence that controlling CH 4 is important for limiting future O3 concentrations. Different treatments of chemistry and meteorology in models remains one of the largest uncertainties in calculating the O 3 response to perturbations in CH4 abundance and precursor emissions, particularly ov er the Middle East and South Asian regions. Emissio n changes for 30 the future ECLIPSE scenarios and a subset of prelim inary Shared Socio-economic Pathways (SSPs) indicat e th t surface O 3 concentrations will increase by 1 to 8 ppbv in 2050 across different regions. Source attribution analy sis highlights the growing importance of CH4 in the future under current legislation. A global tropospheric O3 radiative forcing of +0.07 W m -2 from 2010 to 2050 is predicted using the ECLIPSE scenari os nd SSPs, based solely on changes in CH 4 abundance and tropospheric O3 precursor emissions and neglecting any influence o f climate change. Current legislation is shown to b e inadequate in 35 limiting the future degradation of surface ozone ai r quality and enhancement of near-term climate warm ing. More stringent future emission controls provide a large reduction n both surface O3 concentrations and O 3 radiative forcing. The parameterisation provides a simple tool to highligh t the different impacts and associated uncertaintie s of local and hemispheric emission control strategies on both surface air qua lity and the near-term climate forcing by troposphe ric O3. 40 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2017-1220 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 24 January 2018 c © Author(s) 2018. CC BY 4.0 License.


Introduction
Tropospheric ozone (O3) is an air pollutant at both regional and global scales. It is harmful to human health (Brunekreef and Holgate, 2002;Jerrett et al., 2009;Turner et al., 2016;Malley et al., 2017), whilst also affecting climate  and causing damage to natural and managed ecosystems (Fowler et al., 2009; United Nations Economic Commission for Europe (UNECE), 2016). Long-range transport of air pollutants and their precursors can degrade air quality at locations remote 5 from their source region (Fiore et al., 2009). Predicting source-receptor relationships for O3 is complex due to large natural background sources, formation of O3 from local emissions, non-linear chemistry and inter-continental transport processes (TF-HTAP, 2010). In particular, it is uncertain how the interaction of local and regional emission controls with global changes (e.g. of methane and climate) could affect O3 concentrations in the near-term future (2050s) (Jacob and Winner, 2009;Fiore et al., 2012;von Schneidemesser et al., 2015). This is evident from the wide range of modelled O3 responses in future emission and 10 climate scenarios (Kawase et al., 2011;Young et al., 2013;Kim et al., 2015). The setting and achieving of effective future emission control policies is therefore difficult, as a substantial proportion of O3 comes from outside individual countries and regions.
Phase 1 of the Task Force on Hemispheric Transport of Air Pollutants (TF-HTAP1) (TF-HTAP, 2010) coordinated several 15 sets of experiments using multiple models to study the source-receptor relationships from the intercontinental transport of O3 and its precursors. It found that at least 30% of the total change in surface ozone concentration within a particular source region can be attributed to emission changes of similar magnitude that are external to the source region (TF-HTAP, 2010). This highlights the importance of source contributions outside the control of local/regional air pollutant policies, including those of stratospheric origin, natural sources and intercontinental transport. Changes in global methane (CH4) concentrations are also 20 an important contributor to baseline O3 concentrations and are shown to be as important as changes in local source region emissions (TF-HTAP, 2010). Improving our understanding of the impact of anthropogenic emission changes on the sourcereceptor relationships arising from the intercontinental transport of tropospheric O3 and its precursors will ultimately reduce the uncertainty in the impact of O3 on air quality and climate, improving future predictions.

25
To predict how O3 concentrations might respond to future changes in emissions, a simple parameterisation was developed based upon the surface O3 response in different chemistry models contributing to TF-HTAP1 Wild et al., (2012). The surface O3 response in these models was calculated from simulations with reductions in tropospheric O3 precursor emissions across the four major northern hemisphere emission regions (Europe, North America, East Asia, and South Asia). The parameterisation using these results provided a fast and simple tool to predict future surface O3 concentrations for the 30 Intergovernmental Panel on Climate Change (IPCC) Representative Concentration Pathways (RCPs), highlighting the importance of future changes in emissions and CH4 abundance for surface O3 concentrations and quantifying the associated uncertainty.
A second phase of model experiments, TF-HTAP2, was initiated to extend the work from TF-HTAP1 and further consider the 35 source-receptor relationships between regional emission reductions and air pollutants. Major advances in TF-HTAP2 include more policy-relevant source-receptor regions aligned to geo-political borders, a larger variety of idealised 20% emission reduction experiments, more recent (2008)(2009)(2010) emission inventories that are consistent across all models and the use of new and updated models (Galmarini et al., 2017). 40 Here we improve and extend the parameterisation of Wild et al., (2012) by including additional information from TF-HTAP2 to refine the source-receptor relationships arising from emission changes, long range transport and surface O3 formation. The parameterisation provides the contribution from local, remote and methane sources to the total surface O3 response in each Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2017-1220 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 24 January 2018 c Author(s) 2018. CC BY 4.0 License. emission scenario. The range of responses from the models contributing to the parameterisation provides an estimate of the uncertainty involved. The parameterisation is extended to estimate changes in tropospheric O3 burden and its impact on O3 radiative forcing. It is then used with the latest emission scenarios from ECLIPSE V5a (Klimont et al., 2017;Klimont et al., in prep.) and the 6 th Coupled Model Intercomparison Project (CMIP6) (Rao et al., 2017) to explore how source-receptor relationships change in the future, informing the future direction of emission control policies. These predictions of changes in 5 surface and tropospheric O3 are based solely on changes in precursor emissions, as the parameterisation does not represent any impact from future changes in climate.
Section 2 of this paper describes the parameterisation and the updates from TF-HTAP1 to TF-HTAP2, including the extension from surface O3 to global tropospheric O3 and its radiative forcing. Section 3 outlines the testing and validation of the 10 parameterisation. A comparison is made to results from TF-HTAP1, highlighting changes in the O3 response to changes in methane abundance. In section 4, the parameterisation is applied to the ECLIPSE V5a and CMIP6 emission scenarios to predict future surface O3 concentrations over the period 2010 to 2050. Section 5 O3 uses the same future emission scenarios to predict future tropospheric O3 burden and radiative forcing. We conclude by suggesting how this approach could be used to inform future emission policy in relation to O3 concentrations. 15

Original Ozone Parameterisation
The parameterisation developed in this study is based on an earlier version developed for the TF-HTAP1 experiments by Wild et al., (2012). This simple parameterisation enabled the regional response in surface O3 concentrations to be estimated based on changes in precursor emissions and CH4 abundance. The input for this parameterisation came from 14 different models that 20 contributed to TF-HTAP1. All the models ran the same emission perturbation experiments (20% reduction in emissions of oxides of nitrogen (NOX), carbon monoxide (CO), non-methane volatile organic compounds (NMVOCs) individually and all together) over the four major northern hemisphere source regions of Europe, North America, East Asia and South Asia.
Additional experiments included global perturbations of emission precursors, as well as a 20% reduction in global CH4 abundance. The multi-model responses from the 20% emission perturbation experiments are then scaled by the fractional 25 emission changes from a given emission scenario over each source region. The monthly mean O3 response (∆O3) is the sum over each receptor region (k) of the scaled O3 response from each model to the individual precursor species (i -CO, NOX and NMVOCs) in each of the five source regions (j -Europe, North America, East Asia, South Asia and rest of the world), including the response from the change in global CH4 abundance (Eq. 1, reproduced from Wild et al., 2012): The emission scale factor fij for each emission scenario is defined as the ratio of the fractional emission change (∆ 0.2 × ⁄ ) to the 20% emission reduction in the TF-HTAP1 simulations (Eq. 2). A similar scale factor for methane (fm) is based on the ratio of the change in the global abundance of CH4 to that from the 20% reduced CH4 simulation (∆ 0.2 × ⁄ ). 35 Perturbations to emissions of CO, NOx and NMVOCs induce a long-term (decadal) change in tropospheric O3 from the change in the oxidising capacity of the atmosphere (OH) and the CH4 lifetime (Wild and Akimoto, 2001;Collins et al., 2002;Stevenson et al., 2004). The long-term impacts from 20% global emission reductions can reduce the O3 response by 6-14% from NOx emission changes and increase the O3 response by 16-21% from CO changes (West et al., 2007). This long-term response is not accounted for in the simulations used here as CH4 abundances are fixed. 40 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2017-1220 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 24 January 2018 c Author(s) 2018. CC BY 4.0 License. Wild et al., (2012) found that this simple linear scaling relationship between emissions and surface O3 was sufficient for small emissions perturbations, but that the relationship started to exhibit larger non-linear behaviour for larger perturbations, particularly for NOx. The linear scaling factor is sufficient for the surface O3 response from emission perturbations of CO and NMVOCs as non-linear behaviour from these precursors is small . To account for non-linear behaviour of surface O3 to NOx emission changes, the scale factor, f, is replaced with , which has a quadratic dependency on f (Eq. 2), and 5 is based on additional simulations of surface O3 response to larger emission perturbations undertaken in Wild et al., (2012).
For titration regimes, where a reduction in NOX emissions may lead to an increase in O3, the surface O3 response is limited (by a factor of 2 − for emission reductions and for emission increases the linear scaling factor is used (Eq. 2). The spatial extent and magnitude of titration regimes is assumed constant as it is based on model simulations from a single meteorological 10 year.
The surface O3 response to changes in global CH4 abundances is also non-linear and showed similar behaviour to that due to NOx emissions. Therefore, in Eq. 2 is also used to represent the O3 response to increases and decreases in CH4 abundances.

15
In summary, the surface O3 response to CO and NMVOC emission perturbations is represented by fij and to changes in CH4 abundances by (Eq. 2). The representation of the surface O3 response to NOx emissions is determined by the conditions specified in Eq. 2.

Phase 2 of TF-HTAP
A second phase of simulations has been undertaken as part of TF-HTAP to further study the transport of air pollutants and their impacts and to assess potential mitigation options (Galmarini et al., 2017). Phase 2 (TF-HTAP2) involved experiments using new and/or updated models that conducted idealised 20% perturbation simulations of O3 emission precursors for different source regions and source sectors over the years 2008 to 2010. A 20% emission perturbation was chosen to generate a sizeable 25 response, whilst still being small enough to minimise non-linear chemistry effects. To determine the O3 response to CH4 changes, simulations increasing methane to 2121 ppbv (18%) and decreasing to 1562 ppbv (-13%) from a baseline of 1798 ppbv were undertaken in TF-HTAP-2. This range in CH4 abundances was selected to encompass the uncertainty in CH4 changes in 2030 from the 5 th Coupled Model Intercomparison Project (CMIP5) scenarios of RCP8.5 and RCP2.6 (Galmarini et al., 2017). 30 The source and receptor regions were updated to represent 16 new receptor regions (14 of which are also sources), aligned on geo-political and land/sea boundaries (Figure 1). Emission inventories (consistent across all models (Janssens-Maenhout et al., 2015)) and meteorology (driving data specific to individual models) were updated to consider the years 2008 to 2010 (the focus of TF-HTAP1 was 2001). The Global Fire Emission Database version 3 (GFED3 -http://globalfiredata.org/) biomass 35 burning (grassland and forest fires) emissions were recommended for TF-HTAP2 experiments, although some models selected other inventories. Individual modelling groups used their own information for other natural emission sources (e.g. biogenic VOCs, lightning NOx), as many of these are based on internal model calculations and not externally prescribed datasets.
Priority in TF-HTAP-2 was placed on conducting a baseline simulation, a simulation with increased CH4 concentrations and seven regional simulations involving 20% reductions of all precursor emissions across the globe, North America, Europe, East 40 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2017-1220 Manuscript under review for journal Atmos. Chem. Phys.

Improvements to the surface Ozone Parametric Model for TF-HTAP2
Differences in the experimental setup in TF-HTAP1 and TF-HTAP2 means that it is not straight forward to replace the simulations underpinning the parameterisation of Wild et al., (2012) with those from TF-HTAP2. The larger number of 10 simulations and fewer models involved preclude the development of a robust parameterisation based solely on TF-HTAP2 simulations. We therefore extend the existing parameterisation by including additional information from the new simulations in TF-HTAP2. To maintain a robust response over the major source regions of Europe, North America, East Asia and South Asia, results from the 14 models contributing to TF-HTAP1 over these regions were retained in the parameterisation. Results from the models contributing to TF-HTAP2 were then incorporated, accounting for the different baseline year for emissions 15 (2010 rather than 2001) and the change in size and number of source/receptor regions.

New Baseline Year
The baseline year used in the parameterisation was first adjusted from 2001 (TF-HTAP1) to 2010 (TF-HTAP2), to reflect changes in anthropogenic emissions between these years. It should be noted that the emission inventories used in TF-HTAP1 were not consistent between models, particularly for NMVOCs, and this partially contributed to the different O3 responses 20 (Fiore et al., 2009). In TF-HTAP2, the same anthropogenic emission inventory was used in all models to prevent uncertainty in anthropogenic emissions dominating the variability across models.
The parameterisation of Wild et al., (2012)  that the range of responses over the models dominates the uncertainty in O3 concentrations and is much greater than differences due to the subset of models contributing to each study or from changing emissions over the period 2000 to 2010.

Source Region Adjustment
The original parameterisation was based on the continental-scale emission source regions defined in TF-HTAP1. To continue using these results in an improved parameterisation, the O3 response fields were adjusted to represent the equivalent source 5 regions in TF-HTAP2. The different regional definitions used within TF-HTAP1 and TF-HTAP2 experiments are shown in Figure 1 and are particularly large for Europe, where the TF-HTAP1 source region covers parts of five TF-HTAP2 source regions (Europe, Ocean, North Africa, Middle East and Russia Belarus and Ukraine). O3 response fields from TF-HTAP1 models that formed the basis of the original parameterisation were adjusted to be more representative of the equivalent TF-HTAP2 source region. 10 No single model contributed experiments in both TF-HTAP1 and TF-HTAP2 to inform the adjustment of source regions. Therefore, 20% emission perturbation simulations were conducted with HadGEM2-ES (Collins et al., 2011;Martin et al., 2011), which contributed to TF-HTAP2 experiments, for the TF-HTAP1 source regions of Europe, North America, East Asia and South Asia. The ratio of the O3 responses between the simulations using TF-HTAP1 and TF-HTAP2 source regions was 15 then applied to the O3 response fields from each of the TF-HTAP1 models used in the parameterisation of Wild et al., (2012).
We assume that each model behaves in a similar way as HadGEM2-ES when the source regions are adjusted in this way. This generates an O3 response field from emission perturbations within the equivalent TF-HTAP2 source regions of Europe, North America, East Asia and South Asia. The resulting O3 parameterisation is based on a larger number of models (14 adjusted TF-HTAP1 models) than would have been available from using TF-HTAP2 simulations alone (7 TF-HTAP2 models), allowing 20 for a larger diversity of model responses to represent the four major emission source regions.

Additions from TF-HTAP2
The O3 responses from emission perturbations for the other ten TF-HTAP2 source regions were then used to augment the source region adjusted O3 response fields from TF-HTAP-1. This extends the parameterisation to cover a much larger range of source regions (14 in total) than was previously possible. Table 3 lists the number of model simulations available for the 25 TF-HTAP2 source regions over and above the four main source regions of Europe, North America, South Asia and East Asia, highlighting the sparseness of results for some of the TF-HTAP2 regions.
The monthly O3 response fields from the additional ten TF-HTAP2 emission source regions were converted onto the same standard grid (1° x 1° in the horizontal, with 21 vertical levels based on regular pressure intervals from the surface at 1000 hPa 30 to an upper level of 10 hPa) as used for the four source regions from the adjusted TF-HTAP1 models. In addition, the fields from the TF-HTAP1 models are based on the O3 response to the individual emissions perturbations of NOx, CO and NMVOCs, whereas the regional emission perturbation simulations for TF-HTAP2 are based on all emission precursors together (due to the limited availability of results from regional individual precursor emission simulations in TF-HTAP2). To maintain consistency with the TF-HTAP1 parameterisation, the O3 response for each TF-HTAP2 emission perturbation simulation is 35 divided up to represent the response from individual emission precursors, as Wild et al., (2012) and Fiore et al., (2009) previously showed that O3 responses from individual emission perturbations matched closely to that from combined emissions changes (within 2-7%). Therefore, the fractional contribution from individual emissions to the total O3 response in the multimodel mean of TF-HTAP1 models is used to apportion the contribution from individual emissions in TF-HTAP2 simulations to the total O3 response. 40 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2017-1220 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 24 January 2018 c Author(s) 2018. CC BY 4.0 License.
The CH4 perturbation experiments in TF-HTAP2 were based on global changes of -13% and +18% to reflect the expected atmospheric abundance in RCP2.6 and RCP8.5, respectively. These were adjusted to the 20% reduction used in TF-HTAP1 using the parameterisation, allowing O3 responses to CH4 from the original 14 TF-HTAP1 models and the five TF-HTAP2 models that provided sufficient results to be combined.

Extension to Tropospheric Ozone 5
The parameterisation has been extended from the surface through the depth of the troposphere, enabling the calculation of the tropospheric O3 burden. The three-dimensional monthly O3 fields from the model simulations are interpolated onto 21 vertical levels at regularly spaced mid-level pressure intervals from 1000 hPa to 10 hPa. These O3 fields were then used with the parameterisation to generate global and regional tropospheric O3 burdens for each scenario, with the tropopause defined as an O3 concentration of 150 ppbv (Prather et al., 2001). An O3 radiative forcing is derived by using the tropospheric O3 burden 10 from the parameterisation and the relationship between radiative forcing and tropospheric column O3 change based on multimodel ensemble mean results from the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) (Stevenson et al., 2013). This relationship is provided as a two-dimensional global map, enabling regional and global O3 radiative forcing to be calculated from the parameterisation.

Testing and Validation 15
The original parameterisation developed by Wild et al., (2012) was based on the surface O3 response to 20% regional emission perturbations from TF-HTAP1 for 2001. We have updated this to reflect conditions in 2010, and have made significant improvements based on results from TF-HTAP2. The parameterisation has been extended to include the 14 sources regions in TF-HTAP2. Output is provided on a standard grid to facilitate the calculation of O3 responses over any selected receptor regions. The parameterisation has been extended to generate three-dimensional O3 fields that permit the calculation of 20 tropospheric O3 burden and O3 radiative forcing for any scenario. To test and verify the improved parameterisation, additional simulations have been conducted with HadGEM2-ES, which are discussed in the following sections

Scaling Factors
We conducted experiments with HadGEM2-ES where all O3 anthropogenic precursor emissions were reduced by 50% and 75% over Europe, to complement the existing 20% emission reduction scenarios performed as part of TF-HTAP2. Figure 2  25 shows a comparison of the annual and monthly surface O3 response from the 20%, 50% and 75% European emission reduction simulations across Europe (a local receptor) and North America (remote receptor), using HadGEM2-ES and the parameterisation based on the O3 response fields from HadGEM2-ES alone (a self-consistent test of the parameterisation). The largest errors of <1 ppbv occur over the source region ( Fig. 2c), with smaller errors of <0.1 ppbv for the remote receptor region (Fig 2d). This small internal error between the parameterisation based on HadGEM2-ES and HadGEM2-ES simulations 30 indicates that the parameterisation of O3 is working well for emission changes at least as great as 50%. This is similar to the results of Wild et al., (2012) and indicates that the parameterisation performs well. Figure S1 compares the output of HadGEM2-ES simulations with the parameterisation based on O3 response fields from multiple models. The magnitude of error is larger at ~2.0 ppbv over Europe and ~0.3 ppbv over North America for a 75% reduction. This highlights that the uncertainty in the parameterised O3 response is dominated by the large spread in O3 responses over the different models rather 35 than by errors in the parameterisation.

Global Emission Perturbation
To further test the parameterisation, we compare the surface O3 response from the parameterisation to a HadGEM2-ES model  Klimont et al., (2017) and Klimont et al., (in prep.)). The ECLIPSE V5a emission scenarios provide future greenhouse gas and air pollutant emissions based on 5 assumptions of energy use, economic growth and emission control policies for different anthropogenic emission sectors from the International Energy Authority (IEA). Three scenarios from ECLIPSE V5a are used in this study: Current Legislation (CLE) assumes future implementation of existing environmental legislation, Current Legislation with Climate policies (CLIM) is an energy and climate scenario targeting 2°C of climate warming in which air pollutants and CH4 are reduced and Maximum Technical Feasible Reduction (MTFR) is the introduction of maximum feasible available technology assuming no economic 10 or technological constraints. Emissions of O3 precursor species and CH4 are available at decadal increments over the period 2010 to 2050 for each ECLIPSE scenario (MTFR is only available for 2030 and 2050). The CH4 abundance was derived from the CH4 emissions at decadal increments by using a simple box model that accounts for the sources and sinks of CH4 and the feedbacks on its chemical lifetime following Holmes et al., (2013). Table 4 shows changes in annual CH4 abundances and NOx emissions from the ECLIPSE scenario, with changes in CO and NMVOCs shown in Table S1 and S2 respectively. An 15 ECLIPSE CLE 2030 scenario was generated by scaling the anthropogenic emissions in the TF-HTAP2 BASE scenario by the fractional emission changes in NOx, CO and NMVOCs in CLE. A HadGEM2-ES simulation was performed using the change in emissions based on CLE for comparison to the parameterisation.

30
For South Asia, the parameterisation based on HadGEM2-ES and on the multi-model responses agrees well but differs substantially (in sign and magnitude) from the HadGEM2-ES simulated O3 changes. The largest difference in surface O3 concentrations of 5 ppbv between the model and the parameterisation occurs in the winter months (December, January, February), with differences in summer being much smaller (0.5 to 1 ppbv). Over the South Asian region the ECLIPSE CLE emission scenario predicts a ~70% increase in NOX emissions by 2030 (Table 4). This large increase could lead to errors in 35 the parameterised response due to the transition from O3 production to titration, which the parameterisation is unable to represent well. However, it is able to represent the O3 responses in the TF-HTAP2 models for a smaller emission change of 20% over South Asia ( Table 5). The boundary layer mixing in HadGEM2-ES over South Asia (a region with challenging topography) has been shown to be insufficient, particularly in winter (Hayman et al., 2014;O'Connor et al., 2014), and a large increase in NOx emissions could lead to a transition to O3 titration over this region, accounting for the discrepancy in surface 40 O3 responses.
As the parameterisation of Wild et al., (2012) did not show a similar discrepancy over South Asia for large emission perturbations, a comparison has been made between the monthly surface O3 response from HadGEM2-ES and the parameterisation across both the TF-HTAP1 and TF-HTAP2 definitions of South Asia in January and July (Figure 4). This shows that continental O3 titration in January is less evident in the HadGEM2-ES simulation over the larger TF-HTAP1 South Asia region, as it includes a large area of ocean. The TF-HTAP2 South Asia region is only continental and HadGEM2-ES 5 shows the larger impact of O3 titration over the continental region in January. The parameterisation and HadGEM2-ES O3 responses agree much better over South Asia in July when there is less evidence of O3 titration effects. The parameterisation, using only HadGEM2-ES as input, is not able to represent the O3 response in HadGEM2-ES over TF-HTAP2 South Asia as it is based on a 20% emission reduction simulation of HadGEM-ES, where the extent of O3 titration over the continental area is small. Additional model simulations conducted with large emission increases over South Asia would be able valuable to further 10 explore this issue, although none are currently available.
These results highlight that caution is needed when applying the parameterisation with emission changes larger than 50-60%, as noted previously in Wild et al., (2012). In particular, the shift into O3 chemical titration regimes cannot be represented easily in a simple parameterisation. For smaller emission changes, the parameterisation is shown to be relatively robust at 15 representing monthly surface O3 changes.

CMIP5 Scenarios
We now use the improved parameterisation described above to explore how future predictions of regional surface O3 for the RCPs used in CMIP5 have changed since TF-HTAP1. The four RCPs assume different amounts of climate mitigation to reach 20 a target anthropogenic radiative forcing in 2100: RCP2.6, RCP4.5, RCP6.0 and RCP8. 5 (van Vuuren et al., 2011). Emissions of O3 precursor species and CH4 are available at decadal increments over the period 2010 to 2050 for each RCP. CH4 emissions are converted to CH4 abundances in each RCP using the MAGICC model which takes into account feedbacks on the CH4 lifetime . The parameterisation only accounts for the impact from changes in anthropogenic emissions over the period 2010 to 2050 and does not account for changes in climate, but on this near-term timescale changes 25 in O3 are dominated by emission changes rather than climate effects (Fiore et al., 2012). There are large differences in global CH4 abundances in the four scenarios, and this strongly influences the O3 responses.  (Table S3). The results in Fig 5. across Europe, North America, South Asia, East Asia and globally are similar to those based on TF-HTAP1 in Wild et al., (2012) (Fig. 5 and Table 6) but differ slightly in magnitude due to the change in the spatial extent 35 of the individual source regions from TF-HTAP1 to TF-HTAP2. Additionally, the improved parameterisation provides O3 changes for other regions that were not previously available, including the Middle East and Africa. This provides useful additional information on surface O3 over these important regions under future emission change.

Sensitivity of Ozone to Methane
The importance of controlling CH4 to achieve future reductions in O3 has been highlighted in earlier studies, along with the provides an opportunity to assess whether the sensitivity of O3 to CH4 identified in TF-HTAP1 remain the same. Experiments with both increased (CH4INC) and decreased (CH4DEC) global abundance of CH4 were conducted in TF-HTAP2. However, these experiments used an increase of 18% and a reduction of 13% to align with 2010 to 2030 changes in global CH4 abundance under RCP8.5 and RCP2.6, in contrast to the 20% reduction used in TF-HTAP1. 5 Wild et al., (2012) found that a 20% increase in CH4 abundance yielded an 11.4% smaller surface O3 response than that from a 20% decrease in CH4. For simplicity the parameterisation used the same scaling factor as for NOx emissions (Eq. 2 i.e. = 0.95 + 0.05 1 ), which represents a 10% smaller response for successive emission increases. The two TF-HTAP2 models that contributed results to both CH4DEC and CH4INC simulations allow us to check the expression used here. We find a slightly larger sensitivity, with both models yielding a 12.6% smaller surface O3 response for an increase in CH4 than a decrease 10 (Eq. 3). Since this O3 response to CH4 in TF-HTAP2 is comparable to that from TF-HTAP1, for simplicity and consistency we chose to retain the same scaling factor for both NOx and CH4 (Eq. 2).
= 0.937 + 0.063 1 15 To enable a direct comparison with TF-HTAP1 results, the O3 response from the CH4DEC and CH4INC experiments in TF-HTAP2 are scaled to represent the response from a 20% reduction in CH4 abundances, using Eq. 3. An adjustment factor is calculated based on the global mean difference between the TF-HTAP2 O3 response in each experiment and that of an equivalent 20% reduction in CH4 abundance (calculated using Eq. 3), resulting in a factor of 1.557 for CH4DEC and -1.256 for CH4INC. The global mean O3 responses from CH4DEC (-0.69 ± 0.01 ppbv, 2 models) and CH4INC (0.81 ± 0.14 ppbv, 7 20 models) are adjusted to generate the equivalent O3 responses to a 20% reduction in CH4 abundance, which are used in the parameterisation (-1.05 ± 0.12 ppbv). This response is ~14% larger globally than that in TF-HTAP1 (-0.90 ± 0.14 ppbv, 14 models), highlighting a slightly increased sensitivity of O3 to CH4.
To explore the differences between TF-HTAP1 and TF-HTAP2 models the CH4 lifetime and feedback factor for each TF-25 HTAP2 model (where data is available) can be calculated in accordance with Fiore et al., (2009). The feedback factor is the ratio of the atmospheric response (or perturbation) time to global atmospheric lifetime and describes how the atmospheric CH4 abundance responds to a perturbation in CH4 emissions e.g. a feedback factor of 1.25 means that a 1% increase in emissions would ultimately generate a 1.25% increase in CH4 concentrations (Fiore et al., 2009). The feedback factors can be used in conjunction with CH4 emission changes for a region, to relate the O3 response from the reduction in CH4 abundance in TF-30 HTAP scenarios to that equivalent from emissions, taking into account both the long-term and short response of emissions on O3 (Fiore et al., 2009). Table 7 summarises the calculated CH4 lifetime and feedback factors for the two TF-HTAP2 models that have provided the appropriate fields. These two models show slightly shorter methane lifetimes and a higher feedback factor (F) than the TF-HTAP1 mean values. This suggests that the sensitivity of O3 to changes in CH4 in the two TF-HTAP2 models is slightly larger than the TF-HTAP1 multi-model mean. The increased feedback factor also indicates that a slightly 35 larger reduction in methane emissions is required to achieve a comparable reduction in O3 concentrations.
Overall, the sensitivity of O3 to a change in CH4 abundance is slightly larger in the two TF-HTAP2 models considered here than in TF-HTAP1 models, but still within the range of the TF-HTAP1 multi-model ensemble. The results from TF-HTAP2 will not significantly change any conclusions from TF-HTAP1 but suggests that the previous O3 changes estimated from TF-40 HTAP1 are conservative. The O3 response to CH4 remains one of the most important processes to understand for controlling future O3 concentrations.

Surface Ozone under ECLIPSEv5a Emissions
The parameterised approach is used with the ECLIPSE v5a emission scenarios described above to determine regional changes in future surface O3 concentrations. Surface O3 concentrations for the CLE (current legislation) scenario are predicted to increase from 2010 to 2050 across all regions ( Figure 6). Annual mean surface O3 concentrations increase by 4 to 8 ppbv 5 across the South Asia and Middle East regions due to the large increases expected in NOx emissions (Table 4), although there is substantial uncertainty in the parameterisation over these regions. Surface O3 concentrations over Europe and North America in 2050 are similar to those in 2010, even though their regional NOx emissions decrease by ~50%. The contributions of different sources to the total surface O3 change has been analysed for each source region ( Figures S2 to S13). Results for Europe ( Figure. 7) and South Asia ( Figure. 8) are shown here, as these regions experience contrasting changes in surface O3. Across Europe, 10 surface O3 from local and remote (mainly North American) sources is reduced in response to emission decreases, and the contribution from CH4 increases by 1.6 ppbv in 2050 (Fig. 7). The increase in global CH4 abundance in the CLE scenario increases surface O3 over Europe, offsetting the reduction in O3 from local and remote sources. This contrasts strongly with South Asia where local sources dominate the total O3 response. This demonstrates how different local and hemispheric emission control strategies are needed in different regions. 15 For the CLIM (climate policies on current legislation) scenario, annual mean surface O3 concentrations in 2050 decrease slightly or stay at 2010 concentrations due to reductions in anthropogenic emissions and control of CH4 emissions leading to a decrease in its abundance (Table 4). The source contribution analysis for Europe (Fig. 7) and South Asia (Fig. 8) shows that CH4 contributes much less to the total surface O3 change under this scenario than CLE. For South Asia, there is also a reduction 20 in the contribution from local sources to surface O3. Under CLIM, remote sources start to dominate the contribution to European surface O3 changes in 2050, increasing to -1.3 ppbv. However, across South Asia the contribution from local sources (+3.2 ppbv) is greater than from remote sources (-1.4 ppbv) in 2050, reflecting the importance of local emissions in this region.
The contribution of CH4 sources to the total surface O3 response is smaller in CLIM due to the targeting of CH4 for climate mitigation purposes. The implementation of these climate policy measures shifts the dominant factor driving future O3 changes 25 within a receptor region towards extra-regional sources.
The MTFR scenario (maximum technically feasible reduction) considers large reductions in emissions (Table 4)

Surface Ozone under CMIP6 Emissions
We provide an initial assessment of surface O3 changes from a subset of the preliminary emission scenarios developed for the 35 future air pollutant emission pathways are mapped onto the SSPs and represent differing targets for pollution control, the speed at which developing countries implement strict controls and the pathways to control technologies (Rao et al., 2017).
Increasingly stringent air pollutant emission controls are assumed to occur with rising income levels because of the increased focus on human health effects and the declining costs of control technology. SSP2 is a medium pollution control scenario that follows current trajectories of increasing levels of regulation. SSP1 and SSP5 are strong control scenarios where pollution 5 targets become increasingly strict. A weak pollution control scenario is adopted in SSP3 and SSP4 where the implementation of future controls are delayed (Rao et al., 2017).
We select three preliminary SSPs to represent scenarios of business as usual (SSP3 BASE), middle of the road (SSP2 60) and enhanced mitigation (SSP1 26). The SSP2 60 and SSP1 26 scenarios have climate mitigation targets of 6.0 and 2.6 W m -2 in 10 2100 applied to them. Currently, air pollutant emissions for each SSP are available globally and across five world regions from https://tntcat.iiasa.ac.at/SspDb/. The air pollutant emissions for each region have been mapped onto the equivalent TF-HTAP2 source regions and the grouping of regions is shown in Table 8 along with the percentage change in global CH4 abundance and NOx emissions over the period 2010 to 2050. The relative changes in CO and NMVOCS emissions are shown in Table S4 and S5. Gridded versions of these emission scenarios will be made available in due course (K. Riahi, personal communication, 15 2017), which will allow a more accurate evaluation of the impacts arising from these scenarios.
Surface O3 concentrations increase across all regions in 2050 for the SSP3 BASE scenario (Figure 9). Europe, North America and East Asia show an increase in surface O3 of 1 to 3 ppbv, a larger response than in the ECLIPSE CLE scenario. Smaller increases in surface O3 are predicted over the Middle East (~3 ppbv) and South Asia (~5 ppbv) compared to CLE. Methane 20 dominates the total surface O3 response over Europe in SSP3 BASE, with small contributions from local and remote emission sources over the period 2010 to 2050 (Fig. 10). Local emissions are the main contribution to O3 changes over South Asia, with a slightly larger influence from CH4 than in CLE (Fig. 11).
For SSP2 60 (middle of the road scenario), surface O3 concentrations reduce slightly by 2050 and to a greater extent than in 25 ECLIPSE CLIM due to the larger reductions in NOx emissions and global CH4 abundances (Table 4 and 9). Over South Asia, NOx emissions in SSP2 60 decrease by 22% from 2010 to 2050, with a corresponding O3 change of -3 ppbv, compared to CLIM where NOx emissions increase by 66% and the corresponding O3 change is +3 ppbv. However, this difference could arise from using preliminary SSP emissions based on five large world regions, where emission changes in South Asia and nearby regions such as East Asia are combined together. For Europe and South Asia the source contributions for each region 30 ( Fig. 10 and 11) are similar to those in CLIM. Remote sources are more important under this intermediate climate mitigation scenario, with local emissions sources becoming more important by 2050 over South Asia.
Large reductions in surface O3 concentrations are predicted across all regions in the strong mitigation scenario (SSP1 26) (Fig.   9). The improvements in O3 concentrations are less than predicted under the ECLIPSE MTFR due to a smaller reduction in 35 NOx emissions. Northern mid-latitude regions show reductions in surface O3 concentrations of up to 6 ppbv under SSP1 26, similar to MTFR. Over South Asia, surface O3 is predicted to be reduced by up to 7 ppbv, which is less than under MTFR.
The source contributions for both Europe (Fig. 10) and South Asia (Fig. 11) are similar to MTFR with the importance of remote sources and the increasing importance of CH4 by 2050 evident over Europe. Over South Asia, the increasing importance of local and CH4 sources is clear by 2050. 40 This analysis of preliminary CMIP6 emission scenarios highlights the large range of future regional surface O3 responses that are possible depending on the climate and air pollutant policies applied. The assumptions within each of the future SSPs, Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2017-1220 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 24 January 2018 c Author(s) 2018. CC BY 4.0 License. particularly for CH4, results in different sources dominating the contribution to the total surface O3 response. Uncertainties in the assumed growth rate of CH4 under the two current legislation scenarios (CLE and SSP3 BASE) result in a 1 ppbv difference in surface O3 over Europe and North America, highlighting the importance for future air quality of reducing CH4 on a global scale. The CMIP6 scenarios allow a larger range of pathways to be explored than were available in ECLIPSE or the CMIP5 RCPs, including those of strong, medium and weak policies on air pollutants and climate change. The parameterisation can be 5 used to provide a rapid assessment of the impact of differing policy measures on surface O3 concentrations across different regions, along with a clear source attribution. This can ultimately inform selection of policies that are most beneficial to future air quality.

Future Tropospheric Ozone Burden and Radiative Forcing
As discussed in section 2.3, the parameterisation has been extended to generate three-dimensional O3 distributions throughout 10 the troposphere, using a tropopause defined as a O3 concentration of 150 ppbv (Prather et al., 2001). Tropospheric O3 column burdens are calculated in each grid cell for each emission scenario. These are used to infer changes in O3 radiative forcing by using the relationship between radiative forcing and tropospheric column O3 (W m -2 DU -1 ) and its spatial variation with latitude and longitude from the ACCMIP multi-model ensemble (Stevenson et al., 2013). Tropospheric O3 burdens and O3 radiative forcings are calculated for the CMIP5 RCPs to evaluate the parameterisation against values from the ACCMIP multi-model 15 study (Stevenson et al., 2013). Additionally, future projections of O3 radiative forcing are made for the ECLIPSE and CMIP6 SSPs.
The change in tropospheric O3 burden for the ECLIPSE CLE scenario in 2030 from the parameterisation was evaluated against the change from the equivalent HadGEM2-ES simulation, a self-consistent test. The parameterisation produced a change in 20 global tropospheric O3 burden of -0.9 Tg which compares well with the -1.0 Tg change from HadGEM2-ES simulations. In comparison to the ACCMIP multi-model mean (Stevenson et al., 2013), the parameterisation produces changes in the global O3 burden and O3 radiative forcing from 2000 to 2030 for RCP2.6 and RCP6.0 (Table 9)  to the ACCMIP multi-model mean. The parameterised response compares reasonably well for smaller emission changes but is unable to reproduce fully the larger changes in the high emission (low mitigation) scenario (RCP8.5), as it does not include natural emissions or represent impacts on O3 from changes in climate that are contained within the ACCMIP models. The net impact of climate change on global tropospheric O3 radiative forcing was estimated from the ACCMIP multi-model ensemble to be between -20 to -30 mW m -2 (a negative feedback) (Stevenson et al., 2013). The parameterisation generates gridded changes in the tropospheric O3 column burden and radiative forcing, which can be used to calculate changes over different regions. Figures 13 and 14 show that the largest relative changes in O3 burden for the ECLIPSE scenarios occur over the Middle East, South Asia and South East Asia (> 10%), with a corresponding larger impact on O3 radiative forcing (-0.3 W m -2 in MTFR). Smaller relative changes in the tropospheric O3 burden are found for CLE over Europe and North America. For MTFR a 15% reduction in O3 burden is predicted over Europe and North America, similar to 5 that over South Asia, but the change in O3 radiative forcing is not as large (-0.2 W m -2 compared to -0.3 W m -2 over South Asia). The parameterisation allows the regional near-term climate implications (in terms of O3 radiative forcing) from future emissions changes to be explored under different air quality and climate policy scenarios. It also highlights the wide range of near-term climate forcing that is possible over particular regions from future emission policies.

Conclusions 10
In this study, we describe improvements and extensions to a simple parameterisation of regional surface O3 responses to changes in precursor emissions and CH4 abundances based on multiple models. We incorporate results from phase 2 of the Hemispheric Transport of Air Pollutants project to create an enhanced parameterisation that includes new models, a greater number of source regions, a new baseline of 2010 and an extension to three dimensions to represent O3 changes throughout the troposphere. These improvements allow impacts on surface O3 concentrations and the near-term O3 radiative forcing to be 15 calculated from different emission scenarios. Model simulations using HadGEM2-ES confirm the validity of the parameterisation and adjustments made here. There is a slight increase in the response of O3 to CH4 for the TF-HTAP2 models, resulting in a slightly higher sensitivity of O3 to CH4 changes. The extent of the difference varies on a regional basis, but is within the range of model responses in TF-HTAP1. A global O3 radiative forcing of +0.07 W m -2 is predicted by 2050 (relative to 2010) under the current legislation scenarios of the SSPs and ECLIPSE. There is a large and diverse regional response in O3 radiative forcing with some regions e.g. Middle 5 East and South Asia more sensitive to changes in emissions than others, and these show a large positive O3 radiative forcing under current legislation. However, application of aggressive emission mitigation measures leads to large reductions in O3 radiative forcing (-0.10 W m -2 ), lessening the near-term impact on climate.
The new parameterisation provides a valuable assessment tool to evaluate the impact of future emission policies on both 10 surface air quality and near-term climate forcing from O3. It also provides a full source attribution along with a simple measure of uncertainty, given by the spread of the multi-model responses that reflect different transport and chemistry processes in models. Whilst not replacing full chemistry simulations it provides a quick way of assessing where to target future modelling efforts. However, these O3 responses are based on changes to anthropogenic emissions only, with no account taken of the impact on O3 and/or its natural precursor emissions due to future changes in chemistry or climate. The parameterisation could 15 be extended further by including a feedback factor to take some account of the impact of future climate change on O3.
Additional improvements could include coupling the output to an offline radiation model to enable improved calculation of O3 radiative forcing, using O3 fields from the parameterisation within a land surface model to assess the impacts of O3 on vegetation and the carbon cycle or with O3 dose-response functions to calculate impacts on human health.

5
2 Total atmospheric CH4 lifetime (years) defined as the reciprocal mean of τOH and assuming a lifetime in the stratosphere and soils of 120 years and 160 years respectively (Prather et al., 2001). 3 The feedback factor is the ratio of the atmospheric response (or perturbation) time to the global atmospheric lifetime. It is defined as 5 = 6 6 − 7 ⁄ where S is determined from the BASE and CH4 perturbation simulations and defined as 7 = 89 :; < = 9 >? @AB ⁄ and CH4 abundances for TF-HTAP2 are 1798 ppbv in BASE, 1562 ppbv in CH4DEC and 2121 ppbv for CH4INC (Prather et al., 2001         Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2017-1220 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 24 January 2018 c Author(s) 2018. CC BY 4.0 License. Figure 9: Annual mean change in regional surface O3 concentrations between 2010 and 2050 from the parameterisation for the CMIP6 emissions scenarios of SSP3 baseline (red), SSP2 with a radiative forcing target of 6.0 W m -2 (purple) and SSP1 with a radiative forcing target of 2.6 W m -2 (green).