We diagnose forcing and climate feedbacks in benchmark sensitivity experiments with the new Met Office Hadley Centre Earth system climate model HadGEM2-ES. To identify the impact of newly-included biogeophysical and chemical processes, results are compared to a parallel set of experiments performed with these processes switched off, and different couplings with the biogeochemistry. In abrupt carbon dioxide quadrupling experiments we find that the inclusion of these processes does not alter the global climate sensitivity of the model. However, when the change in carbon dioxide is uncoupled from the vegetation, or when the model is forced with a non-carbon dioxide forcing – an increase in solar constant – new feedbacks emerge that make the climate system less sensitive to external perturbations. We identify a strong negative dust-vegetation feedback on climate change that is small in standard carbon dioxide sensitivity experiments due to the physiological/fertilization effects of carbon dioxide on plants in this model.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The inclusion of biogeophysical and chemical processes in climate models increases their ability to simulate potentially important climate impacts, such as an Amazon dieback [e.g., Cox et al., 2004]. It also allows carbon cycle feedbacks to be explored [e.g., Gregory et al., 2009]. However, it is unclear whether including such processes – and the potential feedbacks that may result – changes the climate system's sensitivity to external forcing in benchmark concentration-driven Coupled Model Intercomparison Project (CMIP) experiments that have long been used to quantify and compare model responses to external forcing.
 Here we aim to understand and quantify the influence of including these so-called ‘Earth system processes’ on forcing and feedbacks in experiments that are core to the CMIP5 experiment design [Taylor et al., 2012], such as an instantaneous quadrupling of atmospheric CO2 concentration (abrupt 4 × CO2). Although highly idealized, these experiments allow a straightforward separation of forcing and feedback processes, and can even be used to construct more policy-relevant scenarios [e.g.,Good et al., 2011]. They also provide information on the sources of uncertainty in multi-model climate projections.
 The Met Office Hadley Centre has recently developed a new state-of-the-art Earth System climate model, HadGEM2-ES [Martin et al., 2011; Collins et al., 2011]. The precise definition of an Earth system climate model is somewhat arbitrary: we refer to a model that includes both terrestrial and oceanic ecosystems, as well as interactive tropospheric chemistry, and the various couplings between them [Collins et al., 2011]. The inclusion of Earth system processes raises the possibility that new feedbacks will emerge. For example, HadGEM2-ES includes a mineral dust aerosol scheme whose emissions are coupled to a dynamic vegetation model. The dust affects climate through its influence on radiation and ocean biogeochemistry.
 One motivation for this study is the finding that the global temperature response in an ensemble of HadGEM2-ES simulations forced by a 1% per year compound increase in atmospheric CO2 is significantly larger than its predecessor, HadGEM1 (Figure 1a), which had few interactive Earth system processes. While repeating the experiment with HadGEM2-ES but with the new processes switched off (HadGEM2-AO inFigure 1a), appears to rule out the hypothesis that the increased sensitivity comes about due to their inclusion, we are still motivated to probe Earth system feedbacks further. This is because, i) there could be compensating Earth system responses, and ii), the above comparison does not isolate the impact of ‘CO2 physiological forcing’ – an important process which has been included in the last three generations of Met Office Hadley Centre models (see section 2.1). Resolving the forcing and feedback processes that determine the climate response will help to answer these questions.
2. Experimental Design
2.1. Climate Model Description
 We use two configurations of the new Met Office Unified Model, HadGEM2 [Martin et al., 2011]:
 1. HadGEM2-AO includes atmospheric, land surface and hydrology, aerosol, ocean and sea-ice processes. It is based on a previous configuration, HadGEM1, which was used in CMIP3, but with improvements designed to address specific systematic errors [Martin et al., 2011]. The model has 38 levels in the vertical and a horizontal resolution of 1.25° × 1.875° in latitude and longitude. It includes a mineral dust scheme which adds an important new natural aerosol species that affects climate through its interaction with radiation.
 2. HadGEM2-ES is a configuration which builds on HadGEM2-AO by including terrestrial and oceanic ecosystems and tropospheric chemistry [seeCollins et al., 2011]. The terrestrial ecosystem includes the TRIFFID dynamic vegetation model [Cox, 2001]. These ecosystems provide new couplings between dust and vegetation (dust emission depends on bare soil fraction), and between dust and the ocean biogeochemisty (dust deposition provides iron for plankton growth), which could provide feedbacks with climate change. The model also includes interactive di-methyl sulphide (DMS) emissions from phytoplankton, providing an important source of sulphate aerosol that may also vary with climate. Finally, the model includes an interactive tropospheric chemistry scheme, which allows methane, ozone and sulphur oxidants to vary with meteorology and climate.
 Both configurations include the response of plant physiology to changes in atmospheric CO2 concentrations. For example, under increased CO2 plant stomata do not open as widely, reducing the evapotranspiration flux to the atmosphere, a process referred to as ‘CO2 physiological forcing’ [e.g., Doutriaux-Boucher et al., 2009]. The altered surface energy balance can rapidly influence moisture availability, near surface cloud cover, temperature and precipitation, amongst other things [e.g., Andrews et al., 2011]. In addition to this, increased CO2 enhances plant growth, potentially changing vegetation distribution and surface albedo [e.g., O'Ishi et al., 2009]. Changes in vegetation distribution will occur on a longer timescale than the rapid physiological forcing processes. Both effects are simulated by HadGEM2-ES, but changes in vegetation distribution are excluded from HadGEM2-AO because the vegetation cover is fixed.
2.2. Climate Model Experiments
 From fully coupled pre-industrial (year 1860 conditions) control simulations we perform four step change forcing experiments.
 1. 4 × CO2HadGEM2-ES: Atmospheric CO2 concentration is instantaneously quadrupled.
 2. 4 × CO2HadGEM2-AO: Atmospheric CO2concentration is instantaneously quadrupled using HadGEM2-AO. Note that this simulation includes CO2 physiological forcing, but not changes in vegetation distribution.
 3. 4 × CO2RAD HadGEM2-ES: Atmospheric CO2 concentration is instantaneously quadrupled, but plants continue to see control levels of CO2 (the radiation code and ocean biogeochemistry see 4 × CO2). Differencing this and experiment (1) allows us to isolate the effects of CO2 physiological/fertilization forcing and feedback processes.
 4. Solar HadGEM2-ES: Solar constant is instantaneously increased by ∼3.3%, from 1365 to 1410.7 Wm−2. This magnitude of forcing is designed to give a similar sized radiative forcing to 4 × CO2.
 The simulations ran for a common length of 65 model years. The climate response – i.e. the global-annual-mean surface-air-temperature change, ΔT (units K) – of each of these experiments is summarized in Figure 1b.
2.3. Radiative Forcing Calculations
 We diagnose the ‘adjusted radiative forcing’, F (units Wm−2), by repeating these experiments with fixed sea-surface temperatures (SSTs).F is then determined from the change in global energy balance [Shine et al., 2003; Hansen et al., 2005]. We assume 4 × CO2Fin HadGEM2-ES and HadGEM2-AO to be the same (see below).
 In practice, we implement this design by replacing the ocean and sea-ice model with climatological SST and sea-ice distributions based on 1860 conditions (derived from the fully coupled pre-industrial simulations). Similarly, we provide an ocean–atmosphere DMS flux climatology. Following the CMIP5 experimental design [Taylor et al., 2012], we fix the vegetation distribution and run the model for 30 years. Thus, under this framework, any change in vegetation cover is regarded as a feedback rather than a forcing.
 We assume the 4 × CO2forcing in HadGEM2-ES and HadGEM2-AO to be the same because the only difference between a HadGEM2-ES and HadGEM2-AO fixed-SST experiment would be the inclusion of the interactive tropospheric chemistry scheme. While potentially important for determining aerosol forcing, we do not believe it will significantly impact forcings derived for CO2. In addition, tests showed that uncoupling the interactive CH4 and O3 from the radiation scheme impacted the diagnosed 4 × CO2F by less than 3%.
3. Adjusted Radiative Forcing
 The net 4 × CO2HadGEM2-ES/AOF is 7.0 Wm−2. Decomposing Finto longwave clear-sky (LWCS), shortwave clear-sky (SWCS) and net cloud radiative effect (CRE) fluxes (defined as the difference between net downward all-sky and clear-sky fluxes), so thatF = FLWCS+ FSWCS+ FCRE, reveals that Fis dominated by processes acting on the LW clear-sky fluxes (FLWCS = 7.2 Wm−2), as expected for a greenhouse gas. F diagnosed from the solar forcing experiment is 7.1 Wm−2. This is mostly composed of a SW clear-sky term (FSWCS = 9.5 Wm−2), partly offset by negative CRE term (FCRE = −1.8 Wm−2), which principally comes about due to clouds reflecting more solar radiation back out to space under enhanced solar luminosity. The geographical distributions of all the forcing components are given in Figure S1 in the auxiliary material.
 The geographical distribution of 4 × CO2Ffor HadGEM2-ES/AO is shown inFigure 2a. Maxima (F > 14 Wm−2) coincide with land regions where we expect CO2 physiological forcing processes to occur, such as the Amazon and central African forests. Doutriaux-Boucher et al.  showed that reduced stomatal conductance due to elevated CO2levels immediately reduces the surface latent heat flux to the atmosphere, reducing low-level clouds. These rapid cloud adjustments alter the top-of-atmosphere radiation balance and are included in our definition of adjusted radiative forcing. This is seen inFigure 2b, which shows the difference between 4 × CO2F and 4 × CO2RAD F. There are clear increases in the adjusted radiative forcing over land due to CO2 physiological effects, especially in the regions mentioned above, as well as the boreal and temperate forests. Consistent with a rapid reduction in cloudiness over land, the 4 × CO2 physiological forcing term is dominated by a net CRE component (Figure S1).
 Globally, the CO2physiological effect only increases the HadGEM2-ES/AO forcing by ∼0.25 Wm−2. This is somewhat smaller than the 1.1 Wm−2 reported by Doutriaux-Boucher et al.  who used an older model, HadCM3LC. Given the small global difference in F between 4 × CO2 and 4 × CO2RAD, we do not believe differences in F alone can account for the additional ∼1 K warming seen in the 4 × CO2HadGEM2-ES experiment (Figure 1b), compared to the 4 × CO2RAD HadGEM2-ES experiment. This suggests that when the vegetation does not see the change in CO2 there exist different feedbacks (next section).
4. Climate Response
 In response to a given F, ΔT (Figure 1b) is determined by the heat balance of the climate system,
where N (units Wm−2) is the change in net heat flux into the climate system and −α (units Wm−2 K−1) is the climate feedback parameter [e.g., Gregory et al., 2004].
 We calculate α in our fully coupled abrupt forcing experiments using averages of the last 30 years of N and ΔT, along with F from section 3. Using an alternative linear regression technique [Gregory et al., 2004], while giving slightly different values for the various feedback terms, leads to the same conclusions outlined below. Note that N and F can also be evaluated at any grid point and for any component (e.g. SW and LW), allowing the geographical distribution of the feedback terms to be examined. The global mean feedback terms are presented in Table 1, the maps for all terms are presented in Figure S2.
Table 1. Climate Feedback Parameter −α, And Its Componentsa
4 × CO2
4 × CO2
4 × CO2RAD
The components are LW clear-sky, SW clear-sky and cloud radiative effect (CRE) (all in units of Wm−2K−1).
 Differences in αbetween HadGEM2-ES and HadGEM2-AO diagnosed from the 4 × CO2 experiments are small (Table 1), hence the similarly evolving ΔT (given similar ocean heat uptakes). This is consistent with the TCR results of section 1.
 When the plants do not see the increased CO2 (the 4 × CO2RAD HadGEM2-ES experiment) the feedbacks behave somewhat differently. −α is more negative in the 4 × CO2RAD HadGEM2-ES experiment, compared to the fully coupled 4 × CO2 experiment, meaning that it is more stabilizing to radiative forcing (as seen clearly in Figure 1b). This difference arises predominantly due to a ∼40% reduction in the positive SW clear-sky feedback term (Table 1).
Figure 3(top row) shows the geographical distribution of the SW clear-sky feedback in all of the experiments. This is most positive in the regions affected by sea-ice and land snow cover, which retreat with global warming, reducing the surface albedo. There is a marked difference between the SW clear-sky feedback maps derived from the HadGEM2-ES 4 × CO2 and 4 × CO2RAD experiments. In the 4 × CO2RAD case, there exists a large negative feedback that is centered on Australia, but spreads out into the Indian and Southern Pacific oceans. This is not present in the 4 × CO2 experiment.
 One explanation for this difference in response is a change in emission of Australian dust, with the dust then being transported over the Southern Pacific and Indian oceans. This is confirmed by analyzing the change in dust optical depth (Figure 3, middle row). In the 4 × CO2RAD experiment there is an increase in dust optical depth in these regions, consistent with the negative SW clear-sky feedback, as additional dust in the atmosphere will reflect more SW radiation back out to space. The enhanced dust emission is associated with a change in vegetation distribution, namely, an increase in bare soil fraction (Figure 3, bottom row), which may exacerbate the feedback through a consequent drying of the surface and increase in wind speed [Collins et al., 2011]. There are also smaller changes in bare soil fraction and dust optical depth around West Africa and the Horn of Africa, which have correspondingly smaller effects on the radiation budget.
 Coupling the change in CO2 to the vegetation (the standard 4 × CO2HadGEM2-ES experiment) prevents the increase in bare soil fraction and the large negative SW clear-sky feedback from arising (Figure 3). Consistent with this, there are little/no changes in dust optical depth centered on Australia. Put another way, CO2physiological/fertilization effects encourage plant growth, preventing an increase in bare soil fraction over Australia that would otherwise occur under global warming in this model. This compensation between Earth system processes is why we did not simulate a large dust-vegetation feedback in our standard 4 × CO2experiment, and why the feedbacks in HadGEM2-ES closely resembled those in HadGEM2-AO.
 What is not clear is whether this compensation should continue in the long term. The CO2 physiological/fertilization component is driven by the change in CO2 concentration, with which the vegetation will presumably equilibrate on some timescale. This may, however, be different to the timescale of the vegetation response to climate change.
 We have extended the standard 4 × CO2HadGEM2-ES experiment to 270 years and re-computed the feedbacks using the last 30 years of this simulation. The feedback components are similar to those determined using the shorter simulation, suggesting that even on this timescale the compensation is still present. We also have 100 years of the HadGEM2-AO simulation, during which the ΔTremains similar to that in HadGEM2-ES throughout.
 We test the robustness of the vegetation-dust feedback observed in the 4 × CO2RAD HadGEM2-ES experiment by analyzing the feedbacks in an independent forcing scenario, the solar experiment. The feedbacks are similar to those in the 4 × CO2RAD experiment (Table 1), including the reduced SW clear-sky feedback due to the dust-vegetation response described above (Figure 3). Changes in dust optical depth and bare soil fraction confirm this (Figure 3). This suggests that when CO2changes are uncoupled from the vegetation, or when the model is forced with a non-CO2forcing, the vegetation-dust feedback is a robust response to climate change in HadGEM2-ES.
 Is the large dust-vegetation response plausible? Averaged over the last decade of the HadGEM2-ES 4 × CO2RAD simulation, the annual-mean surface temperature increases by ∼5 K over Australia. In addition to this large warming, annual-mean precipitation reduces from 1.9 mm day−1 to 1.43 mm day−1, a 25% reduction in an already arid region. It is perhaps not surprising therefore, to see substantial increases in bare soil fraction (the simulated change was from 58% and 76%). It is also worth noting that once the bare soil fraction reaches 100%, then the vegetation-dust response could not respond any further. This is analogous to the sea-ice feedback once the sea-ice disappears altogether.
 There are also limitations in the underlying dynamic vegetation model, TRIFFID. In particular, competition between the plant functional types is based on the Lotka-Volterra equations, which tend not to favor coexistence of plant functional types [e.g.,Arora and Boer, 2006]. This can lead to TRIFFID simulating extreme values in fractional vegetation cover. Collins et al. showed that HadGEM2-ES simulates too much bare soil fraction and dust loads in some arid regions, including Australia. Deficiencies in the underlying climatology do not necessarily mean its sensitivity to external forcing is wrong, but it does raise a question about its fidelity. On the other hand, the dust-vegetation response could be underestimated if the underlying climatology has too little vegetation to lose.
 It should be remembered that this is an idealized experimental design resulting in large and rapid changes in climate, which in turn can lead to changes in bare soil and dust emission that may seem extreme. We believe the processes are still sufficiently well represented to explore potential feedbacks involving dust [Collins et al., 2011]. In addition, feedbacks in models need to be documented in order to constrain and better understand uncertainties in multi-model climate projections.
5. Summary and Discussion
 We have examined the sensitivity of a fully-coupled Earth system climate model, HadGEM2-ES, in benchmark 1% CO2 and abrupt 4 × CO2 CMIP experiments. By comparing results to those from a physical configuration of the model, we have quantified the influence of including Earth system processes in a physical climate model on standard climate sensitivity metrics such as the ‘transient climate response’ (TCR) and climate feedback parameter.
 Under abrupt CO2 quadrupling, inclusion of coupled Earth system processes has little influence on the net climate feedback parameter in our model. Similarly, the global temperature response under a 1% per year increase in CO2 is largely unaffected. However, we found that this was due to a compensation of Earth system responses that is specific to CO2. When the vegetation does not see the change in CO2, or when the model is forced with a non-CO2 forcing, an increase in bare soil fraction over Australia results in increased dust emission. The additional dust in the atmosphere then acts as a negative feedback on climate change through enhanced reflection of shortwave radiation back out to space, reducing the global sensitivity of the model. In standard 4 × CO2 experiments this feedback is small due to CO2 plant physiological/fertilization effects.
 Note that in our framework, changes in vegetation distribution are regarded as a feedback. However, unlike feedbacks involving clouds, water vapor etc., changes in vegetation do not necessarily operate on timescales closely tied to the global mean temperature change, ΔT. In this case, extremely long simulations in which the vegetation has equilibrated with ΔT would be required to be correctly expressed as a ΔT-dependent feedback.
 While our focus has been on processes that affect the global sensitivity of the model, we suggest further work on regional coupled Earth system processes is desirable. For example, changes in dust deposition over the ocean could affect ocean biogeochemistry, leading to possible changes in DMS emissions and carbon uptake.
 Biogeophysical and chemical processes are not as well understood or constrained as many other drivers of climate change [Collins et al., 2011]. They might also be relatively poorly represented in models. In particular, our model is known to have deficiencies in simulating the vegetation distribution in some arid regions [Collins et al., 2011]. While we believe the dust-vegetation feedback to be robust in our model (given that it arose similarly in two independent forcing experiments), this does raise questions about its fidelity. Similarly, the compensation of Earth system responses observed in the 4 × CO2experiments could be model dependent. Similar analysis by other modeling groups will help to identify robust results and/or potential sources of uncertainties in multi-model climate projections using the new generation of Earth System models.
 Finally, we suggest that idealized sensitivity experiments should not just focus on standard CO2 forcings. As we have shown, it would be useful if modeling groups routinely performed sensitivity experiments with a range of forcings, such as CO2 radiative effects only, changes in the solar constant, and possibly other greenhouse gases, aerosols, and combinations thereof.
 We thank C. Jones, J. Gregory, C. Senior, W. Ingram and M. Joshi for useful discussions and comments. P. Halloran provided 1%CO2 ensemble members. We acknowledge two anonymous reviewers for helping improve the clarity of the manuscript. Performing the MOHC CMIP5 simulations was supported by the European Commission's 7th Framework Program, under grant agreement 226520, COMBINE project. This work was supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101).