Changes in atmospheric ozone abundances (year 2100 vs. 2000) due to future perturbations in radiative and chemical drivers have been calculated using the UK Met Office's Unified Model containing the United Kingdom Chemistry and Aerosols sub-model (UM-UKCA). The model is a stratosphere-resolving (model lid ∼ 84 km) CCM that comprehensively describes both stratospheric and tropospheric chemistry (Morgenstern et al., 2009; O'Connor et al., 2014), with interactive ozone and water vapour. Further details of the model are provided in Banerjee et al. (2014, 2016).

Data from six time-slice experiments with fixed seasonally varying boundary conditions are used in this study and summarized in Table 1. All but the ΔCH4 experiment are described in detail by Banerjee et al. (2016). The control experiment (Base) represents the state of the atmosphere at year 2000. The remaining five experiments perturb selected boundary conditions to year 2100 levels. Owing to computational limitations, we have not explored all possible RCP scenarios for these perturbations but rather choose a subset that is commonly explored within the literature. Experiments ΔCC4.5 and ΔCC8.5 perturb the climate state (i.e. including atmospheric radiative effects of WMGHGs, plus changes in sea surface temperatures (SSTs) and sea ice) according to the medium-low (RCP4.5) and high (RCP8.5) future emissions scenarios respectively, without changing any chemical boundary conditions. Here, the WMGHGs considered are CO2, CH4, N2O, CFCs, HCFCs and HFCs. In contrast, experiments ΔODS, ΔO3pre and ΔCH4 leave climate boundary conditions unperturbed at year 2000 conditions, but instead perturb chemical boundary conditions, i.e. surface concentrations of ODSs, emissions of non-methane ozone precursors (from anthropogenic and biomass burning sources defined as in Lamarque et al., 2010) and the surface concentration of CH4 respectively. In this way, we distinguish the chemical and transport effects on ozone resulting from changes in the physical climate state from changes in abundance of reactive gases. All RCP scenarios project a common reduction in ODS and non-methane ozone precursor emissions, so we arbitrarily follow the RCP4.5 scenario in the ΔODS and ΔO3pre experiments. In the CH4 experiment, an increase in the CH4 surface concentration by more than a factor of 2 (from 1.75 to 3.75 ppmv) is imposed according to the RCP8.5 scenario to explore the impact of a very large increase in CH4. The initial atmospheric concentrations of ODSs and CH4 were also perturbed by the same factor in ΔODS and ΔCH4 respectively, in order to reduce spin-up time. In all simulations, including ΔO3pre, emissions from natural sources (e.g. isoprene emissions) are non-interactive and are held fixed at year 2000 levels. In the ΔODS run, by design, the direct radiative effect of ODSs and associated changes in physical climate state (WMO, 2014) are not captured since their concentrations are held fixed at year 2000 values within the radiation scheme. Similarly, the radiative effect of CH4 on climate is not captured by design in the ΔCH4 run.

There are some forcings and interactions that we do not consider in this study. Firstly, our focus lies on estimating the future ozone RF from emitted gases. We do not simulate any associated aerosol forcing, with aerosol precursor emissions and their oxidant fields being held fixed in all simulations (following the scheme of Bellouin et al., 2011). Secondly, the “snapshot” experiments of this study do not consider various transient interactions. For example, the background conditions of NOx and ODSs affect CH4 concentrations, but these couplings are not considered when perturbing NOx, ODSs and CH4 individually in the ΔO3pre, ΔODS and ΔCH4 experiments (potential consequences for the CH4-induced ozone RF are, however, discussed in Sect. 3.4).

Each experiment is spun up for 10 years and integrated for a further 10 years. It is confirmed that this spin-up period was long enough for stratospheric concentrations of perturbed gases to reach steady state. Using averages of the last 10 years, the monthly mean ozone field for each experiment is first interpolated onto the Base pressure levels. The differences in ozone between Base and each perturbation experiment are then used as input to the radiative calculations.

The differences in ozone abundances between year 2000 and 2100 calculated from the UM-UKCA experiments described in Sect. 2.1 are input to the Edwards and Slingo (1996) offline radiative transfer model (RTM) to diagnose the associated all-sky ozone RF. The model includes nine longwave (LW) and six shortwave (SW) spectral bands

The names of the spectral files used in the RTM are spec3a_lw_hadgem1_wz_spec and spec3a_sw_hgem1_ln6e_mean_spec for LW and SW respectively.

We calculate stratosphere-adjusted RFs using the fixed dynamical heating (FDH) method as described by Maycock et al. (2011). The calculations use monthly and zonally averaged climatologies of temperature, water vapour, ozone, WMGHGs, cloud properties and surface albedo from the UM-UKCA Base experiment. The monthly mean year 2100 changes in ozone from each experiment are added to this background climatology, and stratospheric temperatures are adjusted using an iterative method to re-establish radiative equilibrium under the assumption that the local dynamical contribution to the heating rates does not change (IPCC, 2007). Surface and tropospheric conditions remain fixed. The RF is then diagnosed as the change in net radiative flux (downward = positive) at the tropopause. The stratospheric temperature adjustment strongly affects the calculated LW (and hence total) RF for stratospheric ozone changes, with the adjustment being largest where the SW-driven temperature changes are largest (Forster and Shine, 1997).

The lapse-rate tropopause (WMO, 1957) from the Base experiment is used for the stratospheric adjustment and also to perform calculations to separate the RFs due to changes in tropospheric and stratospheric ozone abundances alone. In the climate change experiments, ΔCC4.5 and ΔCC8.5, the tropopause rises; the ramifications for employing a climate-consistent tropopause height for the ozone RF will be shown to be small (see Sect. 3.1). While the lapse rate tropopause is a standard measure for computing RF values, other tropopause definitions exist, including the level at which ozone equals 150 ppbv (Prather and Ehhalt, 2001). For the Base run, the climatological ozone tropopause lies very close to the thermal tropopause; for example, the tropospheric ozone burdens differ by only 2 % between the two definitions. Furthermore, Stevenson et al. (2013) find less than 10 % differences in the tropospheric ozone RF between 1850 and 2000 diagnosed in the ACCMIP models using these two tropopause definitions. Thus, for simplicity we adopt the standard lapse rate tropopause definition in this study.

Recent studies have quantified the so-called effective radiative forcing (ERF), which accounts for rapid tropospheric adjustments (e.g. in cloud properties) resulting from the introduction of a forcing agent, in addition to the standard stratospheric temperature adjustment. A common way to calculate ERFs is to perform fixed SST global model experiments. As such, estimates of ERF are subject to statistical uncertainties arising from internal atmospheric and climate variability. Forster et al. (2016) showed that the 5–95 % confidence intervals on an ERF estimated from a global climate model is around 0.1 W m-2 for a 10-year fixed SST integration. Since the UM-UKCA experiments performed in this study are 10 years long, this would mean that the uncertainties in the estimated ERFs would, in many cases, be larger than the signal being detected. Furthermore, the differences between RF and ERF for ozone have been found to be small in previous studies (Hansen et al., 2005; Shindell et al., 2013b) and so RF is still widely adopted to assess the climate forcing from ozone (Myhre et al., 2013). For these reasons, we utilize the standard stratosphere-adjusted methodology to diagnose ozone RFs.

The radiative effects due to changes in ozone can be considered as a climate forcing mechanism (i.e. they impart an RF on climate; Myhre et al., 2013), although in the case of the impact of changes in greenhouse gases, some part of the effect may be considered as a climate feedback mechanism (e.g. Nowack et al., 2014). However, this distinction is not central to this study, since the UM-UKCA simulations use prescribed SSTs and sea ice, and thus we wish only to quantify the net radiative effect of simulated future changes in ozone resulting from different drivers (see, e.g., Stevenson et al., 2013). For simplicity, we refer to the radiative impact of simulated changes in ozone as an RF throughout the manuscript.

The sign of the whole-atmosphere ozone RF under climate change depends on the WMGHG emissions scenario considered: a positive RF is calculated for ΔCC4.5 (+0.05 W m-2), but a negative RF for ΔCC8.5 (-0.07 W m-2) (Fig. 1, Table 2).

The difference between the two scenarios arises mainly from the stratospheric ozone RF, which is less negative in ΔCC4.5 (-0.04 W m-2) than in ΔCC8.5 (-0.15 W m-2) (Fig. 1, Table 2). Figure 2a further shows that this difference stems from the LW, rather than the SW, contribution to RF. As will be discussed in Sect. 4, the stratospheric LW contribution to RF in ΔCC8.5 is dominated by the effects of a reduction in ozone in the tropical lower stratosphere (Fig. 3b); this is driven by an increase in the upwelling mass flux by 27 %, with an additional contribution from a higher tropopause also being likely. Qualitatively similar conclusions have been drawn for larger 4×CO2 perturbation experiments (Nowack et al., 2014). In contrast, ΔCC4.5 shows a small positive stratospheric LW RF (Fig. 2a). This can partly be explained by more comparable changes in tropical lower stratospheric ozone (driven by an increase in the upwelling mass flux by 10 %) and upper stratospheric ozone (driven by cooling from CO2) (Fig. 3b). Indeed, in a related study focusing on tropical column ozone (Keeble et al., 2017), we find that the change in lower stratospheric ozone scales more strongly with GHG concentration than the change in upper stratospheric ozone: |0.03| vs. |0.02| DU ppmv-1 CO2-equivalent, where CO2-equivalent is the concentration of CO2 that would cause the same RF as the mixture of all WMGHGs.

Figure 1 highlights that the RF due to tropospheric ozone changes could also be an important component of the whole-atmosphere RF due to climate change, which models without comprehensive tropospheric chemistry are unlikely to capture properly. The total tropospheric RFs are positive for both ΔCC4.5 (0.10 W m-2, 0.040 W m-2 DU-1) and ΔCC8.5 (0.07 W m-2, 0.069 W m-2 DU-1) and are dominated by the LW forcing (Fig. 2b; see also Rap et al., 2015). The tropospheric ozone increase and its RF is smaller for the greater climate forcing (ΔCC8.5) due to the relatively stronger effects of tropospheric ozone reductions over ozone increases (the drivers of which are discussed below) than under a weaker climate forcing (ΔCC4.5). The tropospheric RFs outweigh (ΔCC4.5) or partly cancel (ΔCC8.5) the negative RF from stratospheric ozone changes. Consideration of CH4 adjustments reduces the positive tropospheric ozone RFs by 0.02 W m-2 (ΔCC4.5) and 0.05 W m-2 (ΔCC8.5) (see Supplement Table S1), but does not change the sign of the overall tropospheric or whole-atmosphere RFs. Note that the respective changes in CH4 abundance to steady state lead to direct RFs that are larger in magnitude: -0.10 and -0.22 W m-2 (Table S1).

A large driver of the tropospheric ozone RF is the increase in lightning NOx emissions (LNOx) under climate change. We use an additional simulation that fixes LNOx to Base values within the ΔCC8.5 experimental set-up (labelled ΔCC8.5(fLNOx); see Banerjee et al., 2014) to deduce that the increase in LNOx under climate change at RCP8.5 (global total 4.7 Tg (N) yr-1) leads to a tropospheric ozone RF of 0.24 W m-2 (compare rows for ΔCC8.5 and ΔCC8.5(fLNOx) in Table 2). The tropospheric ozone RF from LNOx is enhanced slightly by an increase in STE that is caused by a strengthened stratospheric circulation, but it is offset primarily by the effects of increased humidity-driven ozone loss (Banerjee et al., 2016). The smaller tropospheric ozone RF in ΔCC8.5 compared to ΔCC4.5 is likely a result of the humidity-driven ozone losses cancelling ozone increases in the extratropics (Fig. 3), as well as larger ozone reductions around the tropopause due to a higher tropopause (e.g. see orange line for ΔCC8.5 in Fig. 3c around 12 km).

Interestingly, the increase in LNOx is also associated with a stratospheric ozone RF of 0.04 W m-2 (compare rows for ΔCC8.5 and ΔCC8.5(fLNOx) in Table 2). This RF is consistent with increases in lower stratospheric ozone abundances following its transport from the upper troposphere (Banerjee et al., 2014). Overall, the whole-atmosphere RF is over 5 times larger in magnitude (-0.37 W m-2) when LNOx is held fixed than when it is allowed to vary with climate change in ΔCC8.5 (-0.07 W m-2), which points to a potentially important role of LNOx as a chemistry–climate feedback.

There is considerable inter-model spread in the tropospheric ozone response, and thus in the associated ozone RF, to climate change (Stevenson et al., 2013). The multi-model mean tropospheric ozone RF between 2000 and 2100 under RCP8.5 across eight CCMs is a negligible value of about -0.01 W m-2 (calculated from the final row of Table 12 in Stevenson et al., 2013, by taking the difference of the climate change-induced ozone RFs between 1850–2000 and 1850–2100). However, this reflects cancellations between larger magnitude positive and negative values for individual models: the inter-model range spans ± 0.07 W m-2. Our value of 0.07 W m-2 is on the upper end of the inter-model range and could reflect a particularly large sensitivity of LNOx to climate in our model: 0.96 Tg (N) yr-1 K-1 (Banerjee et al., 2014) compared to a multi-model mean of 0.37 ± 0.06 Tg (N) yr-1 K-1 for the same eight CCMs discussed above (calculated using Table S2 of Finney et al., 2016). Our results serve to show that reducing the inter-model uncertainty in tropospheric ozone projections, and not just in stratospheric projections, is crucial for constraining the future whole-atmosphere ozone RF. Moreover, we show that the whole-atmosphere RF can result from cancellations between stratospheric and tropospheric RFs that are individually larger in magnitude. Thus, it is important to comprehensively simulate effects from both the stratosphere and troposphere to capture the climate impacts of ozone.

Finally, we note that, in order to maintain consistency with previous studies (Nowack et al., 2014; Stevenson et al., 2013), the values of the ozone RF discussed thus far do not consider the effect of the increase in tropopause height under climate change. We calculate that employing climate-consistent tropopause heights causes only small differences (≤|0.02| W m-2) for the stratospheric, tropospheric and whole-atmosphere RFs (Table 2).

The whole-atmosphere ozone RF calculated for the ΔODS perturbation is +0.05 W m-2 (Fig. 1, Table 2). This offsets around one-quarter of the estimated direct RF of the ozone-depleting halocarbons between 2000 and 2100 under RCP4.5, which we estimate to be around -0.22 W m-2 as the difference between the total halocarbon forcing (-0.15 W m-2) (Meinshausen et al., 2011) and the non-ODS halocarbon (HFC) forcing (around +0.07 W m-2 from Fig. 1 of Xu et al., 2013). The future ozone RF due to ODSs is one-third of the estimated magnitude over the historical period (-0.15 W m-2 between 1750 and 2011, Myhre et al., 2013), since ODS concentrations have not returned to pre-1960 values by the end of the century; note there is a slight overlap of around a decade between our reference point (year 2000) and the historical period as defined in Myhre et al. (2013).

Despite large stratospheric ozone changes occurring in the ΔODS experiment (up to 7 DU km-1; Fig. 3), the stratospheric ozone RF is negligible. This arises from the almost complete cancellation between two larger terms: the LW RF (mainly due to ozone increases in the lower stratosphere) and SW RF (mainly due to ozone increases in the upper stratosphere) (Fig. 2a). Note that the degree of cancellation between the LW and SW RF, and hence the sign of the stratospheric ozone RF, appears to be model dependent (Arblaster et al., 2014). This is likely due to inter-model differences in the vertical structure of the ozone response and/or in the background climatology (and hence changes in the LW component following stratospheric temperature adjustments).

The importance of the stratosphere in this experiment is found instead in the enhancement of STE by virtue of there being more stratospheric ozone available for transport; this is the primary driver of changes in tropospheric ozone in the middle and high latitudes (Fig. 1; Banerjee et al., 2016). Consistently, we calculate a tropospheric ozone RF of +0.06 W m-2 (Fig. 1, Table 2) or 0.035 W m-2 DU-1, which is enhanced by 0.01 W m-2 when CH4 adjustments are considered (alongside a direct CH4 RF of 0.03 W m-2; Table S1). We further use a “stratospheric ozone tracer” (see Banerjee et al., 2016) to determine that ∼ 85 % of the tropospheric RF in the ΔODS experiment can be attributed to ozone of stratospheric origin, emphasizing the importance of STE for the climate effects of ozone.

The whole-atmosphere ozone RF in ΔO3pre is -0.09 W m-2 (Fig. 1, Table 2). This arises primarily through reductions in tropospheric ozone in the Northern Hemisphere (see Fig. 3b, d, f) and the associated RF (-0.10 W m-2 or 0.035 W m-2 DU-1). Consideration of the effects of changes in CH4 abundance to steady state results in an additional indirect ozone RF of +0.01 W m-2 and a direct CH4 RF of +0.03 W m-2 (Table S1). Nonetheless, the overall combined effect of ozone and CH4 changes still represents a climate co-benefit (-0.05 W m-2) from air pollution measures. As described previously by Banerjee et al. (2016), the changes in non-methane ozone precursor emissions do not affect stratospheric ozone abundances (see also Fig. 3). In contrast, Sect. 3.4 will show that CH4 is distinct from the non-methane ozone precursors in that it can affect stratospheric ozone and its RF.

The ozone-derived climate effects of changes in non-methane ozone precursor emissions and CH4 have often been compared (e.g. Stevenson et al., 2013; West et al., 2007). Indeed, we find in the next subsection that future increases in CH4 abundance would negate the climate benefits of reductions in non-methane ozone precursor emissions. However, we here emphasize that these benefits could also be negated by future reductions in ODSs, which has previously not been noted: the whole-atmosphere ozone RF in ΔODS is over half the magnitude of the RF in ΔO3pre (Fig. 1, Table 2) indicating that the combination of these perturbations would result in a smaller net ozone RF.

The ΔCH4 perturbation, in which the surface concentration of CH4 is increased from 1.75 to 3.75 ppmv following the RCP8.5 scenario, shows the largest whole-atmosphere ozone RF (0.18 W m-2) within the set of perturbations considered (Fig. 1, Table 2). Unsurprisingly, the bulk of this RF (0.15 W m-2, 0.036 W m-2 DU-1) is due to increases in tropospheric ozone, which occurs at all latitudes (Fig. 3). The ozone increase is 4.3 DU in the global and annual mean and corresponds to a sensitivity of 2.2 DU ppmv (CH4)-1, which falls within the range of other individual studies of 1.7–3.5 DU ppmv (CH4)-1 (Fiore et al., 2002; Kawase et al., 2011; Shindell et al., 2005; West et al., 2007).

A small fraction (∼ 15 %) of the whole-atmosphere RF is due to the stratospheric ozone RF (0.03 W m-2, Fig. 1), which is the same as the estimate in Portmann and Solomon (2007) for the same CH4 increase. As for the ΔODS experiment, the total stratospheric RF is the result of compensating LW and SW RFs (Fig. 2a), but with a slight dominance of the LW effect over the SW in ΔCH4. Correspondingly, the ΔCH4 perturbation exhibits a pattern of ozone response that is similar to that for ΔODS throughout most of the stratosphere; e.g. the perturbations to CH4 (dark blue line, Fig. 3) and ODSs (light blue line, Fig. 3) both show pronounced increases in high-latitude lower stratospheric ozone. The similarity arises through the common reduction in active (ozone-depleting) chlorine abundance. In ΔCH4, this occurs through an increase in the conversion of active chlorine to its reservoir, HCl, via the reaction CH4+ Cl → HCl + CH3. There are further drivers of stratospheric ozone changes in this experiment (although we do not quantify their separate effects on ozone or the stratospheric RF): increases in lower stratospheric ozone (and hence the LW forcing) occur through NOx-mediated production and transport of relatively high ozone amounts from the troposphere; increases in ozone through production of stratospheric water vapour and the consequent cooling; and reductions in ozone through greater HOx-catalysed loss (Fleming et al., 2011; Portmann and Solomon, 2007; Revell et al., 2012; Wayne, 1991). As in ΔODS, there might also be some contribution of stratospheric ozone changes to tropospheric changes through stratosphere to troposphere transport of air containing higher ozone amounts. Our estimate of the whole-atmosphere CH4-driven ozone RF (0.18 W m-2) is greater than the previous estimate of 0.13 W m-2 in Portmann and Solomon (2007) for the same CH4 increase. The difference is due to the larger tropospheric RF (0.15 versus 0.10 W m-2); note that they did not directly diagnose the tropospheric RF due to the simplicity of their tropospheric chemistry scheme, which could explain the difference.

There are several interactions due to time-varying emissions that are not considered in this “snapshot” experiment. Firstly, the increase in CH4 is imposed under year 2000 NOx conditions. If NOx emissions were to decrease in the future, the ozone production efficiency of CH4 would be reduced (Young et al., 2013), and the tropospheric ozone RF would be smaller. Secondly, the increase in CH4 is imposed under year 2000 ODS loadings. As ODS loadings decrease throughout the century, the importance of CH4 in converting Cl to HCl will decrease (Fleming et al., 2011) leading to smaller stratospheric ozone changes and RF.

Finally, we note that the normalized ozone RF (NRF) for tropospheric ozone varies between 0.035 and 0.069 W m-2 DU-1 for the set of perturbations considered (Table 2). Low NRFs (0.035–0.036 W m-2 DU-1) are calculated for the ΔODS, ΔO3pre and ΔCH4 experiments. Higher values are found for the climate change scenarios: 0.040 W m-2 DU-1 (ΔCC4.5) and 0.069 W m-2 DU-1 (ΔCC8.5). This is consistent with increases in LNOx driving ozone increases in the tropical upper troposphere where the LW radiative forcing is most sensitive to ozone changes (Rap et al., 2015). Indeed, without the increase in LNOx under climate change at RCP8.5 in the ΔCC8.5(fLNOx) experiment, the NRF is reduced to 0.045 W m-2 DU-1. Due to the dependence of the NRF on the vertical and latitudinal profile of the ozone change, we argue that it is inappropriate to scale the NRF (e.g. the commonly used multi-model value of 0.042 W m-2 DU-1; Myhre et al., 2013) to obtain the tropospheric ozone RF for different emissions scenarios and different models. Instead, we demonstrate that the NRF is a useful metric to compare the efficiency with which different perturbations (in a single model) affect climate through tropospheric ozone changes; likewise, the NRF could also be used to compare the effects of the same perturbation in different models.

The ozone RFs discussed thus far should be a good indicator of changes to the annual and global mean energy balance in response to ozone perturbations (IPCC, 2007). However, the spatially and temporally inhomogeneous nature of these changes leads to substantial variations in RF across latitudes and seasons; these are explored in the following section.