4.2.1 Carbon dioxide (CO₂)

Atmospheric CO₂ concentrations have risen from 284.3 to 397.5 ppm between 1850 and 2014 (Meinshausen et al., 2017) resulting in an ERF of 1.89 ± 0.04 W m⁻² (Table 3) and is the largest individual contribution to the total historical forcing. Myhre et al. (2013a) report a SARF of 1.82 ± 0.19 W m⁻² for the period from 1750 to 2011, which, coincidentally, has a near identical rise in CO₂ of 113 ppm to that assessed here, although CO₂ forcing has a logarithmic dependency on concentration. An updated assessment based on line-by-line calculations (Etminan et al., 2016) increased SARF for 2015 relative to 1750 to 1.95 W m⁻² but for a larger CO₂ rise of 121 ppm. Applying the CO₂ SARF formula from Etiminan et al. (2016) to our case reveals a SARF of 1.80 W m⁻². Our ERF estimate for 2014 is, therefore, larger than the SARF by 0.09 W m⁻². As expected for a GHG, the ERF is dominated by the LWcs′ component (1.61 ± 0.02 W m⁻²) with an additional contribution from the SWcs′ component (0.09 ± 0.02 W m⁻²) coming from direct effects and an RA in snow cover across the northern latitudes. There is also a NET CRE^′ contribution (0.19 ± 0.02 W m⁻²) from a reduction in cloud cover. The SW direct effect occurs in the near-infrared and is small, equivalent to around 2 % of the LW forcing (Pincus et al., 2020). On that assumption, the SWcs′ component is dominated by the rapid albedo adjustment over the direct effect.

Rising atmospheric CO₂ also exerts an indirect forcing through rapid changes in plant stomatal conductance (Doutriaux-Boucher et al., 2009; Richardson et al., 2018) enhancing plant water use efficiency and reducing evapotranspiration leading to an increase in sensible heating at the surface and corresponding drying of the boundary layer and reduction in low clouds. This mechanism is known as a physiological forcing. In UKESM1 (piClim-CO2phys; Table 3), this forcing is small at PD and 4× CO₂ levels (0.03 ± 0.04 and 0.13 ± 0.03 W m⁻²), and scales in line with the total CO₂ ERF. It results from a balance between a negative LWcs′ component (−0.07 ± 0.03; −0.30 ± 0.02 W m⁻²) associated with surface warming and a positive SW CRE^′ component (0.16 ± 0.02; 0.53 ± 0.02 W m⁻²) associated with a reduction in low-level clouds and from smaller terms including a negative SWcs′ component from reduced water vapour. Previous Hadley Centre models at 4 × CO₂ found a similar but larger effect: 1.1 W m⁻² in HadCM3LC (Doutriaux-Boucher et al., 2009) and 0.25 W m⁻² in HadGEM2-ES (Andrews et al., 2012b). We find that UKESM1 has a similar LWcs′ component compared to HadGEM2-ES implying that UKESM1 has a similar surface warming adjustment associated with the physiological effect but offset by a weaker SW CRE^′ component. The inclusion of the physiological effect and rapid albedo adjustment in the CO₂ ERF acts to increase the forcing slightly. These additional adjustments likely account for our slightly higher ERF relative to the SARF from line-by-line estimates (Etminan et al., 2016).

4.2.2 Nitrous oxide (N₂O)

The ERF due to changes in N₂O concentration from the PI period (273 ppbv in 1850) to the present day (327 ppbv in 2014) is calculated as 0.25 ± 0.04 W m⁻² (Table 3), following Eq. (8) described above. The predominant contribution to the N₂O ERF is the LWcs′ component (0.28 ± 0.03 W m⁻²), with a small offset by the SWcs′ component (−0.04 ± 0.02 W m⁻²); the NETcs′ component sums up to 0.25 ± 0.03 W m⁻². The NET CRE^′ component is insignificant (0.01 ± 0.03 W m⁻²), with SW CRE^′ and LW CRE^′ contributions of 0.09 ± 0.03 W m⁻² and −0.08 ± 0.02 W m⁻², respectively. In comparison, the net ERF calculated here (0.25 ± 0.04 W m⁻²) is higher than the SARF values of 0.17 ± 0.03 W m⁻² from AR5 for 2011 (Myhre et al., 2013a) and of 0.18 W m⁻² for 2014 based on the updated expression from Etminan et al. (2016). This is likely to be due to the effect of adjustments associated with changing N₂O that were not considered as part of the SARF in AR5 (Myhre et al., 2013a) or Etminan et al. (2016), including O₃ depletion and fast cloud adjustments. The UKESM1 estimate agrees well with the AerChemMIP multi-model mean of 0.23 ± 0.05 W m⁻² (Thornhill et al., 2020). Previously, Hansen et al. (2005) calculated an IRF and a SARF of 0.15 W m⁻² due to the change in N₂O from 278 to 316 ppbv.

Figure 3Geographical and zonal mean distributions of the present-day (PD; year 2014) ERFs relative to the pre-industrial (PI; year 1850) period in (a) and (b) for GHGs, in (c) and (d) for aerosols, in (e) and (f) for O₃ precursors, in (g) and (h) for LU change, and in (i) and (j) for total anthropogenic, respectively. In the global distribution plots, global mean ERFs are included and areas are stippled where the ERF is not statistically significant at the 95 % confidence interval. In the zonal mean plots, the grey shading shows the ± 1 standard deviation in the zonal mean ERF. Units in W m⁻².

The global distribution of the individual components contributing to the N₂O ERF (Fig. 4) shows that the SWcs′ component is negligible. The LWcs′ component is predominantly positive, due mainly to N₂O direct forcing. The SW CRE^′ and LW CRE^′ components of the ERF are largely noise (at 95 % confidence level; Fig. 4), with a net contribution to the ERF of close to zero.

Figure 4Global distributions of the PD N₂O ERF components relative to the PI period, i.e. piClim-N2O minus piClim-control; (a) SWcs′, (b) SW CRE^′, (c) LWcs′, (d) LW CRE^′, (e) NETcs′, and (f) NET CRE^′, based on Eq. (8). Global mean values are shown in brackets. Regions where the ERF components are outside the 95 % confidence level are stippled. Units in W m⁻².

4.2.3 Ozone (O₃)-depleting substances (ODSs)

The ERF from the PD–PI change in ODSs is quantified, using piClim-HC relative to piClim-control, as −0.18 ± 0.04 W m⁻², which is dominated by the NET CRE^′ component (−0.18 ± 0.02 W m⁻²; Table 3). Figure 5 shows global distributions of the SWcs′, SW CRE^′, LWcs′, LW CRE^′, NETcs′, and NET CRE^′ components, respectively. For CS conditions, the SW component (SWcs′) is characterized by negative values over the southern high latitudes (to a lesser extent in the northern high latitudes), which is linked to pronounced Antarctic O₃ depletion and some decreases in Arctic O₃ (not shown). The LWcs′ component is predominantly positive, reflecting the direct effect of ODSs acting as GHGs in the piClim-HC simulation. The positive LWcs′ component at high latitudes contains an offset caused by O₃ depletion. Overall, the global mean NETcs′ component is negligible (0.00 ± 0.02 W m⁻²; Table 3). Under cloudy conditions, the negative SW CRE^′ component and the positive LW CRE^′ component of the ERFs are anti-correlated, especially over the Southern Ocean, summing up to a net contribution to the ERF of −0.18 ± 0.02 W m⁻².

As mentioned previously, the ERF includes the direct effect of the ODSs acting as GHGs, the indirect effect of the O₃ depletion that they cause, and any resulting TOA changes due to other RAs. Quantifying the historical evolution of the total O₃ RF alone from the CMIP6 models, Skeie et al. (2020) found that UKESM1 was the only model with both tropospheric and stratospheric chemistry that had a negative total O₃ RF at the present day. This is likely due to UKESM1 having a global O₃ decline during the period of increasing ODSs that is stronger than other models and observations by at least a factor of 1.4 (Table 2 in Keeble et al., 2020). Likewise, in the AerChemMIP multi-model ODS ERF assessment, Morgenstern et al. (2020) used observed O₃ trends as a constraint on the modelled ODS ERF; they found that the ERF is likely to be between −0.05 and 0.13 W m⁻², with the UKESM1 estimate of −0.18 ± 0.04 W m⁻² outside of this range. Although Keeble et al. (2020), Skeie et al. (2020) and Morgenstern et al. (2020) suggest that the negative contribution from O₃ depletion to the ODS ERF is too strong in UKESM1, there is an additional negative offset to the direct radiative effect of the ODSs through the NET CRE^′ component (Fig. 5). Furthermore, the ODSs will have an impact on tropospheric O₃; their contribution to the tropospheric O₃ RF will be quantified in Sect. 4.4.2.

Figure 5Global distributions of the PD ODS ERF components relative to the PI period, i.e. piClim-HC minus piClim-control; (a) SWcs′, (b) SW CRE^′, (c) LWcs′, (d) LW CRE^′, (e) NETcs′, and (f) NET CRE^′, based on Eq. (8). Global mean values are shown in brackets. Regions where the ERF components are outside the 95 % confidence level are stippled. Units in W m⁻².

4.2.4 Methane (CH₄)

The global mean CH₄ concentration changed from 808.3 ppbv in the PI (year 1850) period to 1831.5 ppbv in the PD (year 2014) period, resulting in an ERF of 0.97 ± 0.04 W m⁻² (Table 3; Fig. 6). Most of the ERF (0.74 ± 0.02 W m⁻²; Table 3) is due to the LWcs′ component, with an additional positive contribution (0.11 ± 0.02 W m⁻²) from the SWcs′ component, which is consistent with the growing recognition of the importance of the SW absorption bands in CH₄ forcing (Collins et al., 2006; Li et al., 2010; Etminan et al., 2016). There are additional SW and LW CRE^′ components but these partly cancel out, leading to a small NET CRE^′ (0.12 ± 0.02 W m⁻²) in addition to the NETcs′ component (0.85 ± 0.03 W m⁻²). Estimates of the direct CH₄ ERF from HadGEM2 model simulations (Andrews, 2014) and the updated RF expression for CH₄ based on line-by-line calculations (Etminan et al., 2016) are on the order of 0.50–0.56 W m⁻², but the ERF calculated here is higher by more than 0.4 W m⁻². However, it is consistent with other studies (e.g. Hansen et al., 2005; Shindell et al., 2009; Myhre et al., 2013a), which concluded that the total climate forcing by CH₄ is almost double that of the direct forcing and is due to indirect effects. The UKESM1 estimate is also larger than the 0.69 W m⁻² radiative impact of an increase in CH₄ concentration of 1800 ppbv above PD levels quantified by Winterstein et al. (2019) with the ECHAM/MESSy Atmospheric Chemistry (EMAC) coupled model. Although the Winterstein et al. (2019) estimate included indirect forcings from O₃ and stratospheric water vapour (SWV), their direct CH₄ forcing in the LW is low relative to other models (Lohmann et al., 2010).

Figure 6Global distributions of the PD CH₄ ERF components relative to the PI period, i.e. piClim-CH4 minus piClim-control; (a) SWcs′, (b) SW CRE^′, (c) LWcs′, (d) LW CRE^′, (e) NETcs′, and (f) NET CRE^′, based on Eq. (8). Global mean values are shown in brackets. Regions where the ERF components are outside the 95 % confidence level are stippled. Units in W m⁻².

The UKESM1 ERF quantified here is at the upper end of estimates from the recent study of AerChemMIP multi-model ERFs by Thornhill et al. (2020). They found that the multi-model mean ERF was 0.70 W m⁻², with a standard deviation of 0.22 W m⁻². They attributed part of the inter-model spread to different complexities in the representation of interactive chemistry in the respective models, i.e. some models only captured the direct radiative effect of CH₄ (e.g. NorESM2) while others (e.g. UKESM1) also included indirect contributions from CH₄-driven changes in O₃ and SWV. However, the contribution to the ERF from tropospheric adjustments differed in both magnitude and sign between the models, with UKESM1 being the only model with a positive contribution to the ERF from tropospheric RAs. The relative contributions of the direct and indirect contributions to the total CH₄ ERF quantified here and the mechanism behind the positive tropospheric RA can be found in O'Connor et al. (2019).

4.2.5 Total greenhouse gases (GHGs)

The major drivers of anthropogenic climate change are GHGs, whose forcing is offset by aerosols (Myhre et al., 2013a). Therefore, the total GHG ERF and the aerosol ERF are key values in understanding observed and modelled changes in the climate system since the PI period. As a result, a separate time slice simulation with all GHG concentrations (piClim-GHG; Table 1) at PD levels was conducted following the RFMIP protocol (Pincus et al., 2016).

The UKESM1 piClim-GHG experiment leads to an ERF of 2.92 ± 0.04 W m⁻² (Table 3), which is dominated by a positive LWcs′ component (3.08 ± 0.02 W m⁻²) that is partially offset by a negative SWcs′ component (−0.18 ± 0.02 W m⁻²). There are significant positive and negative contributions from the SW CRE^′ (0.65 ± 0.03 W m⁻²) and the LW CRE^′ (−0.63 ± 0.02 W m⁻²) components, but these largely cancel out and contribute little (0.02 ± 0.03 W m⁻²) to the total ERF (Table 3). This GHG ERF is lower than the ERF of 3.09 W m⁻² estimated from the physical model HadGEM3-GC3.1 (Andrews et al., 2019). Part of this discrepancy may be due to the inclusion in UKESM1 of indirect forcings by O₃ and/or aerosols from ODSs (Morgenstern et al., 2020) and CH₄ (O'Connor et al., 2020), for example. The UKESM1 estimate, however, is consistent with the AR5 year-2011 SARF estimate of 2.82 W m⁻². The latter estimate of 2.82 W m⁻² has been adjusted to an 1850 baseline from 1750, and taking stratospheric O₃ depletion, CH₄-driven SWV, and half of the tropospheric O₃ forcing (based on the attribution by Stevenson et al., 2013) into account, although some GHG concentrations (e.g. CH₄, CO₂) have increased between 2011 and 2014 (Nisbet et al., 2016) while others (e.g. ODSs) have declined (Engel et al., 2018).

The simulation piClim-GHG perturbs non-O₃-depleting HCs but piClim-HC does not. Nevertheless, the small positive RF from non-O₃-depleting HCs of 0.02 W m⁻² in 2011 (Myhre et al., 2013a) allows one to use the combination of GHG simulations to test for linearity. The sum of the ERFs from piClim-CO2, piClim-CH4, piClim-N2O, and piClim-HC is 2.93 ± 0.08 W m⁻², indicating that the ERFs add linearly and agree with the ERF from piClim-GHG (2.92 ± 0.04 W m⁻²).

4.4.1 Tropospheric ozone (O₃) changes

The tropospheric O₃ column for the piClim-control simulation, and tropospheric O₃ column differences between piClim-control and the piClim-O3, piClim-NOx, and piClim-VOC simulations are shown in Fig. 9. O₃ levels in the troposphere are the result of competing production and loss processes. Production occurs in the presence of NO_x and VOC ∕ CO, with most regions of the troposphere being NO_x-limited. PI tropospheric column O₃ values (Fig. 9a) show maxima over central Africa and the eastern tropical Pacific, the result of relatively large emissions of NO_x from soil, biomass burning, and lightning in this region, and minima over regions remote from NO_x sources, such as the western equatorial Pacific (where O₃ loss is efficient), consistent with Young et al. (2018). Tropospheric column O₃ is also low over regions of high surface elevation where the atmospheric column is shallower.

Figure 9Global distributions of tropospheric O₃ column (a) in piClim-control. Differences with respect to piClim-control from piClim-O3, piClim-NOx and piClim-VOC can be seen in panels (b), (c), and (d), respectively. Units in Dobson units (DU).

Tropospheric O₃ column values increase when the model is perturbed with the larger PD O₃ precursor emissions in piClim-O3, with the largest increases in the NH, particularly in southern and eastern Asia. The tropospheric O₃ burden increases from 280.9 Tg in piClim-control to 355.5 Tg in piClim-O3. As can be seen in Fig. 9, the dominant driver of these changes are increased NO_x emissions, and there is a similar pattern of changes in the piClim-O3 and piClim-NOx simulations, although the change in O₃ is smaller in the latter; the burden in piClim-NOx is 337.5 Tg. Small increases in O₃ are modelled in piClim-VOC, with some hotspots in regions such as South East Asia and the O₃ burden increases to 296.8 Tg.

The tropospheric O₃ burden difference between piClim-O3 and piClim-control of 74.6 Tg is slightly larger than the sum of the individual O₃ burden changes in piClim-NOx (56.6 Tg) and piClim-VOC (15.9 Tg), which total 72.5 Tg. However, these differences due to the non-linear nature of tropospheric chemistry are small, being on the order of <5 %. Similar behaviour is seen in the patterns of the tropospheric O₃ column difference between piClim-control and other experiments.

4.4.2 Tropospheric O₃ stratospherically adjusted radiative forcing (SARF)

As seen in earlier sections, it can be difficult to compare ERFs estimated from these experiments against estimates of SARF (e.g. Stevenson et al., 2013; Myhre et al., 2013a), due to the inclusion of indirect forcings and/or RAs other than the stratospheric temperature adjustment, although Shindell et al. (2013b) and Skeie et al. (2020) noted that for O₃, SARF and ERF estimates are comparable. Nevertheless, in order to compare against Stevenson et al. (2013), we estimate the tropospheric O₃ SARF from the piClim-CH4, piClim-VOC, and the piClim-NOx experiments by adopting a radiative kernel approach (e.g. Soden et al., 2008). This involves applying the tropospheric O₃ radiative kernel from Rap et al. (2015) to the diagnosed change in tropospheric O₃ (using the 150 ppbv O₃ isoline in piClim-control as a tropospheric mask, as used in Young et al., 2013; Stevenson et al., 2013; Rap et al., 2015) to calculate a SARF. While the ERF captures changes in radiative fluxes at the TOA due to whole-atmosphere responses, here we mask off the stratosphere and focus solely on the tropospheric O₃ response. In this way, we can directly compare against the best estimate of the 1850–2010 tropospheric O₃ SARF of 364 mW m⁻² by Stevenson et al. (2013) and quantify the contribution of different O₃ precursors (including CH₄) to the change in tropospheric O₃ and its SARF. We can also compare with more recent estimates from Checa-Garcia et al. (2018) and Yeung et al. (2019).

Figure 10 shows the global distribution of the tropospheric O₃ SARF from piClim-CH4, piClim-NOx, and piClim-VOC experiments and their sum using the kernel approach. It shows that the tropospheric O₃ SARF is strongest over the NH sub-tropics and weakest over the SH high latitudes. The strongest SARF occurs in regions of warm surface temperatures and high albedo, coinciding with the largest tropospheric O₃ change (Shindell et al., 2013a). As was the case in Stevenson et al. (2013), the SARF is weaker over regions of high altitude (e.g. the Tibetan Plateau) due to there being less O₃ column aloft to absorb in the LW. The tropospheric O₃ SARF from the kernel approach is 414 mW m⁻². However, the increase in tropospheric O₃ (and its resulting SARF) are offset by decreases due to ODSs (e.g. Søvde et al., 2011, 2012; Shindell et al., 2013a). Applying the kernel method to the diagnosed decrease in tropospheric O₃ from the piClim-HC experiment (Sect. 4.2.3), we find a SARF offset of −101 mW m⁻². Although larger in magnitude by nearly a factor of 2 than the estimates from Søvde et al. (2011, 2012) and Shindell et al. (2013a) due to the strong O₃ depletion in UKESM1 (Keeble et al., 2020; Skeie et al., 2020; Morgenstern et al., 2020), it reduces our original estimate to 313 mW m⁻². This revised estimate is within the 30 % uncertainty of the Stevenson et al. (2013) estimate, albeit lower than their central estimate by 13 %. It is also consistent with a number of other estimates: the CMIP6 historical O₃ dataset (312 mW m⁻² from Checa-Garcia et al., 2018), a recent study in which observational isotopic data was used as a constraint on historical increases in tropospheric O₃ (330 mW m⁻² derived from the GEOSChem model in Yeung et al., 2019), and a parametric model based on multi-model source–receptor relationships (290 ± 3 mW m⁻² from Turnock et al., 2019). However, in masking off the stratosphere, the UKESM1 estimate for the O₃ SARF attributable to O₃ precursors is likely to be underestimated. For example, Søvde et al. (2011, 2012) estimate that approximately 15 % of the O₃ response from changes in CH₄ and other O₃ precursors may be in the stratosphere and hence not considered here.

Table 5Contribution to the tropospheric O₃ SARF from the different perturbation experiments (piClim-NOx, piClim-VOC, and piClim-CH4) relative to the pre-industrial control (piClim-control). Also shown is the absolute difference between the equilibrium and prescribed CH₄ concentrations in the different experiments and the resulting additional contribution to the total tropospheric O₃ SARF from the CH₄-driven response in O₃.

n/a: not applicable.

Download Print Version | Download XLSX

In considering the attribution of the tropospheric O₃ SARF to its precursors, initial estimates (Fig. 10; Table 5) indicate that it is predominantly NO_x driven (49 %), followed by CH₄ (34 %), with the smallest contribution from VOCs and CO (17 %). This is qualitatively consistent with Stevenson et al. (2013). However, as outlined in detail in that paper, these estimates do not account for potential CH₄ (and O₃) changes that would occur if these experiments were driven by CH₄ emissions rather than concentrations (e.g. Shindell et al., 2005). Taking the same approach as Stevenson et al. (2013), we calculate an equilibrium CH₄ concentration for the perturbation experiments using the total CH₄ lifetime relative to piClim-control based on the following (Fiore et al., 2009):

where CH4piClim-X is the global mean equilibrium CH₄ concentration in the piClim-X experiments, CH4piClim-control is the prescribed global mean CH₄ concentration in piClim-control, and τ_{piClim−control} and τ_piClim−X are the whole-atmosphere CH₄ lifetimes in the piClim-control and piClim-X perturbation experiments, respectively. The CH₄–OH feedback factor (Prather, 1996) is denoted by f and is defined as

where s, called the sensitivity coefficient, is calculated from the following:

Using the whole-atmosphere CH₄ burden and its removal by OH, the whole-atmosphere CH₄ lifetime is 8.1 and 9.8 years in piClim-control and piClim-CH4, respectively, when adjusted for stratospheric removal (120 years lifetime) and soil uptake (160 years lifetime). From Eqs. (10) and (11), the UKESM1 feedback factor f is 1.28, consistent with the range of other estimates (Prather et al., 2001; Shindell et al., 2005; Fiore et al., 2009; Stevenson et al., 2013; Voulgarakis et al., 2013; Turnock et al., 2018) and within 5 % of the observationally constrained best estimate of 1.34 (Holmes et al., 2013). Then, using the equilibrium minus prescribed difference in surface CH₄ concentrations (Table 5), we calculate the additional tropospheric O₃ SARF in the piClim-CH4, piClim-NOx, and piClim-VOC experiments by applying the tropospheric O₃ radiative kernel (Rap et al., 2015) to the O₃ response derived from scaling the (piClim-CH4 minus piClim-control) O₃ response based on the relationship in Turnock et al. (2018); these additional contributions to the tropospheric O₃ SARF are shown in Table 5.

Figure 10Global distribution of tropospheric O₃ SARF diagnosed from (a) piClim-CH4, (b) piClim-NOx, (c) piClim-VOC, and (d) their sum relative to piClim-control, based on the diagnosed change in tropospheric O₃ and the tropospheric O₃ radiative kernel of Rap et al. (2015). Global mean values are included. Units in mW m⁻².

Changing from a concentration-based perspective to an emissions-based view increases the tropospheric O₃ SARF from 414 to 469 mW m⁻² (an increase of 13 %), which agrees better with the central estimate from Stevenson et al. (2013) once the offset by ODSs is accounted for (−101 mW m⁻²). It also changes the relative contributions of the different O₃ precursors. The contribution of CH₄ now dominates (45 %), with NO_x playing a smaller role (37 %), while the contribution from VOC ∕ CO emission increases is relatively unchanged (18 %). These emissions-based contributions are well within the spread of estimates from Stevenson et al. (2013), who quantified contributions from six of the ACCMIP models: CH₄ (44 ± 12 %), NO_x (31 ± 9 %), and VOC ∕ CO (25 ± 3 %) although there are some differences with Shindell et al. (2005) and Shindell et al. (2009). In both Shindell et al. studies, the contribution from NO_x is lower at 15 ± 8 and 11 %. The difference may be due to the strong sensitivity of the CH₄ lifetime to NO_x in the GISS model compared with other models (Wild et al., 2020) and/or could be due to differences in VOC chemistry; Archibald et al. (2010) showed that the response of OH to increasing NO_x strongly depends on the treatment of VOC chemistry. Nevertheless, this approach demonstrates the importance of an emissions-based view of climate forcing and is more directly relevant to policy makers than a concentration-based view (Shindell et al., 2005, 2009).

4.4.3 ERF: role of other oxidants and aerosols

In addition to O₃ being a GHG, it is also important for secondary aerosol formation along with other oxidants: the OH radical, hydrogen peroxide (H₂O₂), and the nitrate (NO₃) radical. The OH radical in UKESM1 (Sellar et al., 2019) is involved in aerosol nucleation of gas-phase sulfuric acid (H₂SO₄) via the reaction with sulfur dioxide (SO₂), leading to new particle formation (Mulcahy et al., 2020). O₃ and H₂O₂ are important for SO₂ oxidation in cloud and aerosol droplets, creating sulfate aerosol mass but not number. Likewise, oxidation of monoterpenes by O₃, OH, and NO₃ determines the rate of formation of secondary organic aerosol (SOA) in UKESM1 (Kelly et al., 2018; Mulcahy et al., 2020) although it does not lead to new particle formation. Thus, sulfate aerosol alone, through changes in O₃, OH, and H₂O₂, has potentially important impacts on cloud and aerosol radiative properties. Indeed, Karset et al. (2018) found that oxidant changes in the CAM5.3-Oslo model alter the relative importance of different chemical reactions, leading to changes in aerosol size distribution, cloud condensation nuclei (CCN), and the aerosol ERF.

Figure 11 (bottom row) shows a global distribution of OH at 1 km altitude in piClim-control and changes in the perturbation experiments relative to piClim-control. The piClim-control experiment shows an OH maximum in the equatorial humid regions, where photolytic production of OH from excited oxygen atoms (O¹D) and water vapour (H₂O) is at a maximum. In piClim-O3, OH increases throughout the NH due to increases in O₃, which is the precursor of O¹D. These increases in OH are driven largely by increases in NO_x. However, the piClim-VOC experiment shows the opposite behaviour. While VOC and CO emission increases serve to increase O₃, they also remove OH via direct reaction with OH. This latter effect outweighs the small increase in O₃ in piClim-VOC and there are decreases in OH throughout the troposphere.

When OH is lower, we anticipate a decrease in the number of CCN and a decrease in CDNC, leading to larger cloud droplets (Twomey, 1977) and an increase in Reff. The middle rows of Fig. 11 show these effects at work. In piClim-control, the distribution of CDNC shows large values in equatorial regions, regions of continental outflow, and regions of deep convection. Large increases in CDNC are seen in piClim-NOx with large decreases in piClim-VOC. These results reflect the changes in OH in these experiments. Increases in OH lead to increases in CDNC, and vice versa, but it should be noted that the effect of OH on CDNC is seen over a larger region downwind, particularly in East Asia and over the North Atlantic. The impact of NO_x emissions appears to dominate, given the similarity between piClim-O3 and piClim-NOx.

Figure 11Global distributions of AOD at 550 nm (row a; left column), cloud droplet number concentration (CDNC; cm⁻³) at 1 km altitude (row b; left column), cloud droplet effective radius (Reff; µm) at 1 km altitude (row c; left column), and OH number concentration (cm⁻³) at 1 km altitude (row d; left column) in piClim-control. Differences with respect to piClim-control from piClim-O3, piClim-NOx, and piClim-VOC can be seen in the second, third, and right-hand columns, respectively.

While tropospheric column O₃ increases in both piClim-NOx and piClim-VOC, leading to positive contributions to the tropospheric O₃ SARF (Sect. 4.4.2), the ERFs are very different between the two simulations (0.33 ± 0.04 W m⁻² for piClim-VOC and 0.03 ± 0.04 W m⁻² for piClim-NOx relative to piClim-control). Further, the spatial pattern of the ERF does not match those regions of largest O₃ changes. Instead, the dominant driver of the ERF differences are changes to OH and the subsequent impacts on aerosol particle formation and clouds. In particular, the positive tropospheric O₃ SARF in piClim-NOx (0.2 W m⁻², Table 5) appears to be nearly completely offset by a negative aerosol forcing, largely from ACI driven by changes in oxidants and aerosol nucleation; the contribution to the global mean ERF from ARI is only −0.03 ± 0.01 W m⁻², despite the strong regional changes in AOD at 550 nm (Fig. 11; top row). Similarly, the aerosol IRF in piClim-VOC is negligible (less than 0.01 ± 0.01 W m⁻²). However, the forcing in piClim-VOC due to ACI, particularly from the SW ΔCRE^′ component (Table 3), enhances the positive tropospheric O₃ SARF (0.07 W m⁻²), leading to an ERF of 0.33 ± 0.04 W m⁻².

A negative SARF attributable to NO_x emissions has been found in other studies (Shindell et al., 2009; Collins et al., 2010) from a balance between the direct O₃ response (positive SARF), the NO_x-driven CH₄ response (negative SARF), and the subsequent O₃ response to CH₄ changes (negative SARF). Inclusion of ARI and ACI from sulfate and nitrate aerosol further increases the magnitude of the net negative forcing or cooling (Shindell et al., 2009; Collins et al., 2010). However, a study by Fry et al. (2012) found that the chemistry response is sensitive to the location of the emissions, with so large an uncertainty that it is difficult to determine whether NO_x emissions cause a warming or cooling. Indeed, other indirect effects such as NO_x deposition to the terrestrial biosphere leading to fertilization (Collins et al., 2010) and/or NO_x-driven O₃ damage (Sitch et al., 2007; Collins et al., 2010) increase the uncertainty further through changes in CO₂. Thornhill et al. (2020) show that the ERF from changes in NO_x emissions among the AerChemMIP models differ in both sign and magnitude. Here, the longer timescale CH₄ response to NO_x emissions (Collins et al., 2010) is constrained, nitrate aerosol is neglected (Sellar et al., 2019; Mulcahy et al., 2020), and CO₂ is concentration-driven (Sellar et al., 2019). Nevertheless, in UKESM1, the negative forcing due to ACI from sulfate aerosol offsets the positive NO_x-driven O₃ SARF, leading to a negligible ERF overall.

Previous work has also found a positive SARF from VOC and CO emissions, due to the combined indirect forcings by O₃ and CH₄ (Shindell et al., 2005; Forster et al., 2007); they estimate a global mean SARF of 0.21 ± 0.10 W m⁻² at the PD (year 1998) relative to the PI (year 1750) period. Via the same mechanisms as was the case with NO_x, the magnitude of this SARF increases (to 0.25 ± 0.04 W m⁻² for the year 2000) when additional indirect forcings from sulfate, nitrate, and CO₂ are included (Shindell et al., 2009). More recently, Stevenson et al. (2013) found the SARF from VOC ∕ CO emissions to be marginally higher, at 0.29 W m⁻², excluding aerosols, with contributions of 0.09, 0.08, and 0.12 W m⁻² from O₃, CH₄, and CO₂, respectively. The SARF contribution from O₃ alone quantified here (87 mW m⁻²; Table 5) is consistent with the Stevenson et al. (2013) estimate. Despite excluding the longer-term CH₄ and CO₂ responses, the ERF from VOC ∕ CO emissions of 0.33 ± 0.03 W m⁻² is higher than previous estimates of SARF due to the additional positive contribution from ACI driven by OH changes. However, Fry et al. (2014) found that the SARF from VOCs is sensitive to the location of emissions and could influence the strength of the contribution from aerosols. Interestingly, other AerChemMIP models show a negative ERF from VOC and CO emissions (Thornhill et al., 2020); these differences in sign of the ERF warrant further investigation.

As a result of the very different chemical response between NO_x and VOC ∕ CO emissions, both in terms of the magnitude of the O₃ changes and the different impacts on OH, aerosols, and clouds, a comparison of the ERFs (Table 3) indicates that the ERF from piClim-O3 is not a linear combination of that from piClim-NOx and piClim-VOC for the NET, CS, and CRE components. In particular, there are differences in the LWcs′ and SW ΔCRE^′ components (Table 3). These results clearly suggest that Earth system (ES) interactions, particularly chemistry–aerosol coupling, can strongly affect estimates of climate forcing. Here, these interactions alter the ERF from O₃ precursor emissions, while other studies (e.g. Shindell et al., 2009; Karset et al., 2018) show that they also affect estimates of anthropogenic aerosol forcing.

4.5.1 Non-methane near-term climate forcers (NTCFs)

The anthropogenic ERF due to CH₄, aerosols, and O₃ abundances was identified as the main source of uncertainty in the total anthropogenic ERF since PI times (Myhre et al., 2013a). This is due to the uncertainty in the individual forcings (e.g. Bellouin et al., 2020), but the interaction between individual forcings as well as the non-linear response of climate feedbacks due to ACI (Feichter et al., 2004; Deng et al., 2016; Collins et al., 2017; Shim et al., 2019) may play a role. In this section, three experiments related to non-CH₄ NTCFs are discussed: (1) the combined simulation (piClim-NTCF), which is identical to piClim-control except that aerosol and O₃ precursor emissions are set to PD (year 2014) levels, (2) piClim-aer, which changes aerosol and aerosol precursor emissions only , and (3) piClim-O3 which changes O₃ precursor (VOC, CO, NO_x) emissions only from PI to PD levels. More details are described in Table 1.

The ERF of non-CH₄ NTCFs is −1.03 ± 0.04 W m⁻² (Table 3; Fig. 12). The negative ERF results from the combination of a weak negative NETcs′ component (−0.03 ± 0.03 W m⁻²) and a strong contribution due to the NET CRE^′ component (−1.00 ± 0.02 W m⁻²). The weak negative contribution in the CS (NETcs′) is due to the negative SWcs′ component (−0.26 ± 0.02 W m⁻²) being largely offset by the positive LWcs′ component (0.23 ± 0.03 W m⁻²). The negative SWcs′ component is correlated with changes in AOD at 550 nm (spatial correlation coefficient of −0.44) and is predominantly due to the aerosol IRF (−0.29 ± 0.01 W m⁻²). The LWcs′ component is positive (0.23 ± 0.03 W m⁻²) due to both aerosols and O_3, but with the aerosol IRF only contributing 0.04 ± 0.01 W m⁻². The spatial variations in the global distribution of the LWcs′ component are closely related to land surface temperature (T_s) changes; the correlation coefficient is −0.78 with a statistical significance well over 99 %. Considering this good correlation, the increased LWcs′ component is in response to the T_s change due to NTCFs. The negative SW ΔCRE^′ component (−0.92 ± 0.03 W m⁻²), however, dominates the ERF and is largely correlated with changes in cloudiness, with the spatial pattern correlation of −0.59 between the SW ΔCRE^′ component and cloud fraction.

Figure 12Global distributions of the PD NTCF ERF components at the top of atmosphere (TOA) relative to the PI period, i.e. piClim-NTCF minus piClim-control, based on Eq. (8): (a) SWcs′, (b) SW CRE^′, (c) LWcs′, (d) LW CRE^′, (e) NETcs′, and (f) NET CRE^′. Global mean values are shown in brackets. Regions where the ERF components are outside the 95 % confidence level are stippled. Units in W m⁻².

As was the case for other ERFs (e.g. GHGs, aerosols, and O₃ precursor gases), the range of perturbation simulations carried out with UKESM1 enables the role of non-linear interactions to be investigated. The total GHG ERF was found to be equal to the sum of the individual GHG ERFs (Sect. 4.2.5) but non-linearities were evident for the aerosol (Sect. 4.3) and the O₃ precursor (Sect. 4.4) ERFs. As a result, this study also attempts to estimate the effects of the non-linear interactions between chemistry and aerosols on the combined aerosol and O₃ precursor ERF. When combined (aerosol and aerosol precursor, and O₃ precursor emissions), their interaction may induce an effect that differs from the sum of the individual single ERFs. The ERFs do not add linearly, particularly in the SWcs′ and NET CRE' components (Table 3). Firstly, we calculate the aerosol IRFs using Eq. (7). In piClim-NTCF, the net (SW + LW) aerosol IRF is −0.24 ± 0.01 W m⁻², which is more negative than the sum of the aerosol IRF in piClim-aer (−0.15 ± 0.01 W m⁻²) and piClim-O3 (−0.02 ± 0.01 W m⁻²). This is due to the sulfate aerosol loading being higher in piClim-NTCF relative to piClim-aer by up to 3.4 (27) % globally (regionally), and driven by changes in oxidants due to the PD levels of O₃ precursors. Secondly, the NET CRE^′ component contributes to the non-linearity in the ERFs. The NET CRE^′ component is more negative in piClim-NTCF (−1.00 ± 0.02 W m⁻²) than in the sum of piClim-aer (−1.0 ± 0.02 W m⁻²) and piClim-O3 (0.08 ± 0.02 W m⁻²). This is primarily the result of differences in the SW ΔCRE^′ component, again driven by the higher sulfate loading in piClim-NTCF relative to piClim-aer.

Although there is no direct coupling between aerosols and photolysis in UKESM1 (Archibald et al., 2020), aerosol-mediated cloud adjustments result in SW reduction and surface cooling. These changes impact thermal and photochemical reactions leading to reduced photolysis rates (O₃+ hν→ O(¹D) + O₂) in the lower troposphere, while enhancing the photolysis rate of O₃ in the upper troposphere (black and blue dotted lines in Fig. 13), leading to a reduction in surface O₃ in piClim-NTCF relative to the sum. However, despite differences in O₃ (not shown), the LWcs′ components of the ERFs appear to add linearly.

Figure 13Vertical distribution of changes in j(O¹D) photolysis rates (10⁻⁷ s⁻¹) in piClim-NTCF (black solid line), the sum (blue dotted line) of piClim-aer and piClim-O3 relative to piClim-control, and the difference between them (red dashed line).

Download

4.5.2 Land use

LU change causes RF primarily by changing surface albedo; croplands and pastures have higher albedos than forests and are less able to mask the high albedo of snow cover. Increased albedo leads to a negative SWcs′ component (−0.30 ± 0.02 W m⁻²; Table 3) that is damped by the SW CRE^′ (0.09 ± 0.03 W m⁻²; Table 3), and the inclusion of small positive LWcs′ (0.02 ± 0.03 W m⁻²; Table 3) and LW CRE^′ (0.03 ± 0.01 W m⁻²; Table 3) terms leads to a net ERF of −0.17 ± 0.04 W m⁻². The LU ERF of UKESM1 is close to being within the “very likely” range (−0.15 to −0.10 W m⁻²) of AR5 (Myhre et al., 2013a). However, the AR5 estimate is valid for LU change since 1700, not 1850 as in our estimate. Andrews et al. (2017) used HadGEM2-ES, UKESM1's predecessor, to calculate a 1700–1860 LU ERF of −0.1 W m⁻². If we apply this adjustment to our estimate, we find an ERF of −0.27 W m⁻², which is well outside of the AR5 range. The UKESM1 LU ERF (−0.17 ± 0.04 W m⁻²) is reduced relative to HadGEM2-ES, which produced an 1860–2005 LU ERF of −0.40 W m⁻² (Andrews et al., 2017). Robertson (2019) showed that, even in the absence of snow cover, the albedo response to LU change in HadGEM2-ES is stronger than observed; at the location of deforestation, the surface SW response to total deforestation was generally 5 W m⁻² too large. It is likely that the bias found in HadGEM2-ES still exists in UKESM1.

While the global mean LU ERF is small, regionally it can be the dominant source of the anthropogenic ERF. Figure 14 shows the distribution of the LU ERF and its cloud-free and aerosol-free component, which emphasizes the surface flux contribution. The ERF is mostly confined to regions of LU change, with deforestation in North America and western Eurasia causing a negative ERF and increased tree cover in central Europe causing a positive ERF. Alongside the distribution of LU change itself, the magnitude of the ERF is larger in the mid-latitudes than the tropics, because the masking of snow cover by trees greatly increases the albedo response to changes in tree cover. The large regions of negative ERF in the northern mid-latitudes are consistent with previous model studies (e.g. de Noblet-Ducoudre et al., 2012); they are caused by agricultural expansion and the forcing is expected to have gradually increased over the period 1850–1970 in North America and over the period 1900–1980 in Eurasia. The seasonality of leafiness, snow cover and insolation causes the LU ERF to be largest in NH spring (March to May mean of −0.28 ± 0.11 W m⁻²) and smallest in autumn (September to November mean of −0.08 ± 0.09 W m⁻²).

Figure 14Global distribution of the present-day (PD; year 2014) ERF relative to the pre-industrial (PI; year 1850) period of (a) LU ERF and (b) the contribution from the cloud-free and aerosol-free component. Global mean values are included and areas are stippled where the ERF and/or its component are not statistically significant from zero at the 95 % confidence interval. Units in W m⁻².

The LU ERF is calculated by modifying three land surface fields: land cover, LAI, and canopy height. The values of the modified fields are taken from a coupled UKESM1 simulation that is only subject to historical LU change. In the coupled configuration of UKESM1 (Sellar et al., 2019), the three land surface fields are prognostic fields calculated using a dynamic global vegetation model (DGVM), while in the atmosphere-only ERF configuration they are prescribed fields. This choice was made because the land surface fields can be very slow to respond to changes in land use and RF. In the coupled simulation of historical climate change including all forcings (historical), the land surface fields respond to changes in CO₂, climate, and land use, but if we use the land surface fields from this simulation, we find no substantial change in the LU ERF (−0.20 W m⁻²).

In addition to the ERF caused by albedo changes, LU change can alter climate via a number of other mechanisms. LU change alters the land carbon sink. In particular, deforestation emits CO₂ to the atmosphere, and so some fraction of the CO₂ ERF is attributable to LU change. LU change also affects surface climate via changes in roughness length and transpiration. These non-radiative mechanisms usually drive larger temperature changes than the albedo response and both the non-radiative mechanisms and the CO₂ forcings tend to oppose the albedo response.

4.5.3 Total anthropogenic ERF

As noted above, historical climate change has been driven by a wide range of anthropogenic activities that act together, alongside natural changes, to perturb the Earth's radiation balance. The total anthropogenic ERF is, therefore, a key metric in understanding observed and modelled changes in the climate system since the PI era. These various anthropogenic drivers are not necessarily independent of each other and it is therefore worthwhile to calculate the total anthropogenic ERF from a separate time slice simulation including all perturbations together (piClim-anthro; Pincus et al., 2016). We completed a dedicated time slice simulation with all concentrations, emissions, and land use set to 2014 levels, including all GHG concentrations (CO₂, N₂O, CH₄, HCs – both ODSs and non-ODSs), O₃ precursors (VOC, CO, NO_x), LU changes, and anthropogenic aerosol or aerosol precursor emissions (SO₂, OC, BC). This experiment was not proposed in AerChemMIP (Collins et al., 2017) but is included here as part of RFMIP (Pincus et al., 2016). The main difference is that atmospheric chemistry in UKESM1 is fully interactive whereas other models participating in RFMIP (e.g. Andrews et al., 2019) use the CMIP6 O₃ dataset to represent changes in tropospheric and stratospheric O₃ and prescribe oxidants for secondary aerosol formation.

The UKESM1 piClim-anthro experiment leads to an ERF of 1.76 ± 0.04 W m⁻² (Table 3), which is dominated by a positive LWcs′ component due to GHGs and partially offset by a negative SWcs′ component due to aerosols and negative contributions from the SW ΔCRE^′ and LW ΔCRE^′ components (Table 3). This ERF is a little lower than the equivalent estimate from HadGEM3-GA7.1, which was 1.81 W m⁻² (Andrews et al., 2019). The UKESM1 estimate is also lower than the median estimate from CMIP5 models assessed in AR5, which equates to approximately 1.9 W m⁻² after adjustment to the reference period of 1861–1880 to 2010–2011 (Andrews and Forster, 2020). AR5 also provided an overall central estimate of 2.2 W m⁻² and a 5 %–95 % confidence range of 1.0 to 3.2 W m⁻² (after adjustment to the same reference period as above in Andrews and Forster, 2020) taking into account multiple streams of evidence. Andrews and Forster (2020) re-evaluated this as 2.3 W m⁻² with a narrower range of 1.7–3.0 W m⁻² [5 %–95 % confidence] using a combination of atmospheric model outputs and observational constraints. The lower bound of this range was reduced to 1.5 W m⁻² if larger uncertainties were assumed for the climate feedback parameter or for the global-mean surface temperatures anomalies used to constrain the forcing. The UKESM1 estimate of 1.76 ± 0.04 W m⁻² is therefore within the original uncertainty range given by AR5 and just within the range proposed in Andrews and Forster (2020).

There are several factors that contribute to the relatively low estimate of anthropogenic ERF in UKESM1. Firstly, the anthropogenic aerosol ERF in UKESM1 (and HadGEM3-GC3.1) is −1.09 ± 0.04 W m⁻² (well within the uncertainty range of Bellouin et al., 2020) and offsets a major portion of the positive GHG ERF (2.92 ± 0.04 W m⁻²). Secondly, the ERF from piClim-HC is negative (−0.18 ± 0.04 W m⁻²) due to a strong O₃ response and connected aerosol-mediated cloud adjustments (Morgenstern et al., 2020). Thirdly, adjustments in vegetation lead to an appreciable negative ERF from LU changes (−0.17 ± 0.04 W m⁻²). The stronger negative ERF from piClim-HC balanced by a stronger positive ERF from piClim-CH4 (0.97 ± 0.04 W m⁻²) due to indirect effects and the positive tropospheric RA term (Thornhill et al., 2020) largely explain why the UKESM1 estimate is only ∼ 0.05 W m⁻² lower than HadGEM3-GA7.1. As shown in Fig. 3i, the negative contributions more than offset the positive GHG ERF in certain regions. For instance, the anthropogenic ERF is negative over large parts of North America and Asia (from a combination of LU change and aerosol ERFs; see Fig. 3c and g) and at southern high latitudes (from O₃ depletion due to ODSs, see Fig. 3a; Sect. 4.2.3). In contrast, the anthropogenic ERF is strongly positive over the tropics and SH sub-tropics (Fig. 3e) where the direct radiative effect of GHGs dominates.

The couplings between chemistry, aerosol, and land surface processes included in UKESM1 increase the possibilities for non-linear interactions among the various anthropogenic forcing agents. However, these apparently have little net overall effect on the total forcing (1.76 ± 0.04 W m⁻²), which is within the uncertainty of the sum of the forcings (1.87 ± 0.08 W m⁻²) from the four separate groups that it includes (GHGs, aerosol, O₃ precursors, and land use). The two estimates are not statistically different given that the standard error on these is around 0.03–0.04 W m⁻² (Table 3). This does not imply that the forcings act independently as it is possible that competing non-linear interactions cancel in this particular case. As shown in Sect. 4.5.1, the aerosol and O₃ precursor emission ERFs did not add linearly.