10.4 Changes Associated with Biogeochemical Feedbacks and Ocean Acidification

10.4.1 Carbon Cycle/Vegetation Feedbacks

As a parallel activity to the standard IPCC AR4 climate projection simulations described in this chapter, the Coupled Climate-Carbon Cycle Model Intercomparison Project (C⁴MIP) supported by WCRP and the International Geosphere-Biosphere Programme (IGBP) was initiated. Eleven climate models with a representation of the land and ocean carbon cycle (see Chapter 7) performed simulations where the model was driven by an anthropogenic CO₂ emissions scenario for the 1860 to 2100 time period (instead of an atmospheric CO₂ concentration scenario as in the standard IPCC AR4 simulations). Each C⁴MIP model performed two simulations, a ‘coupled’ simulation where the growth of atmospheric CO₂ induces a climate change which affects the carbon cycle, and an ‘uncoupled’ simulation, where atmospheric CO₂ radiative forcing is held fixed at pre-industrial levels, in order to estimate the atmospheric CO₂ growth rate that would occur if the carbon cycle was unperturbed by the climate. Emissions were taken from the observations for the historical period (Houghton and Hackler, 2000; Marland et al., 2005) and from the SRES A2 scenario for the future (Leemans et al., 1998).

Chapter 7 describes the major results of the C⁴MIP models in terms of climate impact on the carbon cycle. This section starts from these impacts to infer the feedback effect on atmospheric CO₂ and therefore on the climate system. There is unanimous agreement among the models that future climate change will reduce the efficiency of the land and ocean carbon cycle to absorb anthropogenic CO₂, essentially owing to a reduction in land carbon uptake. The latter is driven by a combination of reduced net primary productivity and increased soil respiration of CO₂ under a warmer climate. As a result, a larger fraction of anthropogenic CO₂ will stay airborne if climate change controls the carbon cycle. By the end of the 21st century, this additional CO₂ varies between 20 and 220 ppm for the two extreme models, with most of the models lying between 50 and 100 ppm (Friedlingstein et al., 2006). This additional CO₂ leads to an additional radiative forcing of between 0.1 and 1.3 W m^–2 and hence an additional warming of between 0.1°C and 1.5°C.

All of the C⁴MIP models simulate a higher atmospheric CO₂ growth rate in the coupled runs than in the uncoupled runs. For the A2 emission scenario, this positive feedback leads to a greater atmospheric CO₂ concentration (Friedlingstein et al., 2006) as noted above, which is in addition to the concentrations in the standard coupled models assessed in the AR4 (e.g., Meehl et al., 2005b). By 2100, atmospheric CO₂ varies between 730 and 1,020 ppm for the C⁴MIP models, compared with 836 ppm for the standard SRES A2 concentration in the multi-model data set (e.g., Meehl et al., 2005b). This uncertainty due to future changes in the carbon cycle is illustrated in Figure 10.20a where the CO₂ concentration envelope of the C⁴MIP uncoupled simulations is centred on the standard SRES A2 concentration value. The range reflects the uncertainty in the carbon cycle. It should be noted that the standard SRES A2 concentration value of 836 ppm was calculated in the TAR with the Bern carbon cycle-climate model (BERN-CC; Joos et al., 2001) that accounted for the climate-carbon cycle feedback. Parameter sensitivity studies were performed with the BERN-CC model at that time and gave a range of 735 ppm to 1,080 ppm, comparable to the range of the C⁴MIP study. The effects of climate feedback uncertainties on the carbon cycle have also been considered probabilistically by Wigley and Raper (2001). A later paper (Wigley, 2004) considers individual emissions scenarios, accounting for carbon cycle feedbacks in the same way as Wigley and Raper (2001). The results of these studies are consistent with the more recent C⁴MIP results. For the A2 scenario considered in C⁴MIP, the CO₂ concentration range in 2100 using the Wigley and Raper model is 769 to 1,088 ppm, compared with 730 to 1,020 ppm in the C⁴MIP study (which ignored the additional warming effect due to non-CO₂ gases). Similarly, using neural networks, Knutti et al. (2003) show that the climate-carbon cycle feedback leads to an increase of about 0.6°C over the central estimate for the SRES A2 scenario and an increase of about 1.5°C for the upper bound of the uncertainty range.

Figure 10.20. (a) 21st-century atmospheric CO₂ concentration as simulated by the 11 C⁴MIP models for the SRES A2 emission scenario (red) compared with the standard atmospheric CO₂ concentration used as a forcing for many IPCC AR4 climate models (black). The standard CO₂ concentration values were calculated by the BERN-CC model and are identical to those used in the TAR. For some IPCC-AR4 models, different carbon cycle models were used to convert carbon emissions to atmospheric concentrations. (b) Globally averaged surface temperature change (relative to 2000) simulated by the C⁴MIP models forced by CO₂ emissions (red) compared to global warming simulated by the IPCC AR4 models forced by CO₂ concentration (black). The C⁴MIP global temperature change has been corrected to account for the non-CO₂ radiative forcing used by the standard IPCC AR4 climate models.

Further uncertainties regarding carbon uptake were addressed with a 14-member multi-model ensemble using the CMIP2 models to quantify contributions to uncertainty from inter-model variability as opposed to internal variability (Berthelot et al., 2002). They found that the AOGCMs with the largest climate sensitivity also had the largest drying of soils in the tropics and thus the largest reduction in carbon uptake.

The C⁴MIP protocol did not account for the evolution of non-CO₂ greenhouse gases and aerosols. In order to compare the C⁴MIP simulated warming with the IPCC AR4 climate models, the SRES A2 radiative forcings of CO₂ alone and total forcing (CO₂ plus non-CO₂ greenhouse gases and aerosols) as given in Appendix II of the TAR were used. Using these numbers and knowing the climate sensitivity of each C⁴MIP model, the warming that would have been simulated by the C⁴MIP models if they had included the non-CO₂ greenhouse gases and aerosols can be estimated. For the SRES A2 scenario, these estimates show that the C⁴MIP range of global temperature increase by the end of the 21st century would be 2.4°C to 5.6°C, compared with 2.6°C to 4.1°C for standard IPCC-AR4 climate models (Figure 10.20b). As a result of a much larger CO₂ concentration by 2100 in most of the C⁴MIP models, the upper estimate of the global warming by 2100 is up to 1.5°C higher than for the standard SRES A2 simulations.

The C⁴MIP results highlight the importance of coupling the climate system and the carbon cycle in order to simulate, for a given scenario of CO₂ emissions, a climate change that takes into account the dynamic evolution of the Earth’s capacity to absorb the CO₂ perturbation.

Conversely, the climate-carbon cycle feedback will have an impact on the estimate of the projected CO₂ emissions leading to stabilisation of atmospheric CO₂ at a given level. The TAR showed the range of future emissions for the Wigley, Richels and Edmonds (WRE; Wigley et al., 1996) stabilisation concentration scenarios, using different model parametrizations (including the climate-carbon feedback, Joos et al., 2001; Kheshgi and Jain, 2003). However, the emission reduction due to this feedback was not quantified. Similar to the C⁴MIP protocol, coupled and uncoupled simulations have been recently performed in order to specifically evaluate the impact of climate change on the future CO₂ emissions required to achieve stabilisation (Matthews, 2005; Jones et al., 2006). Figure 10.21 shows the emissions required to achieve CO₂ stabilisation for the stabilisation profiles SP450, SP550, SP750 and SP1000 (SP450 refers to stabilisation at a CO₂ concentration of 450 ppm, etc.) as simulated by three climate-carbon cycle models. As detailed above, the climate-carbon cycle feedback reduces the land and ocean uptake of CO₂, leading to a reduction in the emissions compatible with a given atmospheric CO₂ stabilisation pathway. The higher the stabilisation scenario, the larger the climate change, the larger the impact on the carbon cycle, and hence the larger the emission reduction relative to the case without climate-carbon cycle feedback. For example, stabilising atmospheric CO₂ at 450 ppm, which will likely result in a global equilibrium warming of 1.4°C to 3.1°C, with a best guess of about 2.1°C, would require a reduction of current annual greenhouse gas emissions by 52 to 90% by 2100. Positive carbon cycle feedbacks (i.e., reduced ocean and terrestrial carbon uptake caused by the warming) reduce the total (cumulative) emissions over the 21st century compatible with a stabilisation of CO₂ concentration at 450 ppm by 105 to 300 GtC relative to a hypothetical case where the carbon cycle does not respond to temperature. The uncertainty regarding the strength of the climate-carbon cycle feedback highlighted in the C⁴MIP analysis is also evident in Figure 10.21. For higher stabilisation scenarios such as SP550, SP750 and SP1000, the larger warming (2.9°C, 4.3°C and 5.5°C, respectively) requires an increasingly larger reduction (130 to 425 GtC, 160 to 500 GtC and 165 to 510 GtC, respectively) in the cumulated compatible emissions.

Figure 10.21. (a) Atmospheric CO₂ stabilisation scenarios SP1000 (red), SP750 (blue), SP550 (green) and SP450 (black). (b) Compatible annual emissions calculated by three models, the Hadley simple model (Jones et al., 2006; solid), the UVic EMIC (Matthews, 2005; dashed) and the BERN2.5CC EMIC (Joos et al., 2001; Plattner et al., 2001; triangles) for the three stabilisation scenarios without accounting for the impact of climate on the carbon cycle (see Table 8.3 for details of the latter two models). (c) As for (b) but with the climate impact on the carbon cycle accounted for. (d) The difference between (b) and (c) showing the impact of the climate-carbon cycle feedback on the calculation of compatible emissions.

The current uncertainty involving processes driving the land and ocean carbon uptake will translate into an uncertainty in the future emissions of CO₂ required to achieve stabilisation. In Figure 10.22, the carbon-cycle related uncertainty is addressed using the BERN2.5CC carbon cycle EMIC (Joos et al., 2001; Plattner et al., 2001; see Table 8.3 for model details) and the series of S450 to SP1000 CO₂ stabilisation scenarios. The range of emission uncertainty was derived using identical assumptions as made in the TAR, varying ocean transport parameters and parametrizations describing the cycling of carbon through the terrestrial biosphere. Results are thus very closely comparable, and the small differences can be largely explained by the different CO₂ trajectories and the use of a dynamic ocean model here compared to the TAR.

Figure 10.22. Projected CO₂ emissions leading to stabilisation of atmospheric CO₂ concentrations at different levels and the effect of uncertainty in carbon cycle processes on calculated emissions. Panel (a) shows the assumed trajectories of CO₂ concentration (SP scenarios)(Knutti et al., 2005); (b) and (c) show the implied CO₂ emissions, as projected with the Bern2.5CC EMIC (Joos et al., 2001; Plattner et al., 2001). The ranges given in (b) for each of the SP scenarios represent effects of different model parametrizations and assumptions illustrated for scenario SP550 in panel (c) (range for ‘CO₂ + climate’). The upper and lower bounds in (b) are indicated by the top and bottom of the shaded areas. Alternatively, the lower bound (where hidden) is indicated by a dashed line. Panel (c) illustrates emission ranges and sensitivities for scenario SP550.

The model results confirm that for stabilisation of atmospheric CO₂, emissions need to be reduced well below year 2000 values in all scenarios. This is true for the full range of simulations covering carbon cycle uncertainty, even including the upper bound, which is based on rather extreme assumptions of terrestrial carbon cycle processes.

Cumulative emissions for the period from 2000 to 2100 (to 2300) range between 596 GtC (933 GtC) for SP450, and 1,236 GtC (3,052 GtC) for SP1000. The emission uncertainty varies between –26 and +28% about the reference cases in year 2100 and between –26 and +34% in year 2300, increasing with time. The range of uncertainty thus depends on the magnitude of the CO₂ stabilisation level and the induced climate change. The additional uncertainty in projected emissions due to uncertainty in climate sensitivity is illustrated by two additional simulations with 1.5°C and 4.5°C climate sensitivities (see Box 10.2). The resulting emissions for this range of climate sensitivities lie within the range covered by the uncertainty in processes driving the carbon cycle.

Both the standard IPCC-AR4 and the C⁴MIP models ignore the effect of land cover change in future projections. However, as described in Chapters 2 and 7, past and future changes in land cover may affect the climate through several processes. First, they may change surface characteristics such as albedo. Second, they may affect the ratio of latent to sensible heat and therefore affect surface temperature. Third, they may induce additional CO₂ emissions from the land. Fourth, they can affect the capacity of the land to take up atmospheric CO₂. So far, no comprehensive coupled AOGCM has addressed these four components all together. Using AGCMs, DeFries et al. (2004) studied the impact of future land cover change on the climate, while Maynard and Royer (2004) performed a similar experiment on Africa only. DeFries et al. (2002) forced the Colorado State University GCM (Randall et al., 1996) with Atmospheric Model Intercomparison Project (AMIP) climatological sea surface temperatures and with either the present-day vegetation cover or a 2050 vegetation map adapted from a low-growth scenario of the Integrated Model to Assess the Global Environment (IMAGE-2; Leemans et al., 1998). The study finds that in the tropics and subtropics, replacement of forests by grassland or cropland leads to a reduction in carbon assimilation, and therefore in latent heat flux. The latter reduction leads to a surface warming of up to 1.5°C in deforested tropical regions. Using the ARPEGE-Climat AGCM (Déqué et al., 1994) with a higher resolution over Africa, Maynard et al. (2002) performed two experiments, one simulation with 2 × atmospheric CO₂ SSTs taken from a previous ARPEGE transient SRES B2 simulation and present-day vegetation, and one with the same SSTs but the vegetation taken from a SRES B2 simulation of the IMAGE-2 model (Leemans et al., 1998). Similar to DeFries et al. (2002), they find that future deforestation in tropical Africa leads to a redistribution of latent and sensible heat that leads to a warming of the surface. However, this warming is relatively small (0.4°C) and represents about 20% of the warming due to the atmospheric CO₂ doubling.

Two recent studies further investigated the relative roles of future changes in greenhouse gases compared with future changes in land cover. Using a similar model design as Maynard and Royer (2004), Voldoire (2006) compared the climate change simulated under a 2050 SRES B2 greenhouse gases scenario to the one under a 2050 SRES B2 land cover change scenario. They show that the relative impact of vegetation change compared to greenhouse gas concentration increase is of the order of 10%, and can reach 30% over localised tropical regions. In a more comprehensive study, Feddema et al. (2005) applied the same methodology for the SRES A2 and B1 scenario over the 2000 to 2100 period. Similarly, they find no significant effect at the global scale, but a potentially large effect at the regional scale, such as a warming of 2°C by 2100 over the Amazon for the A2 land cover change scenario, associated with a reduction in the DTR. The general finding of these studies is that the climate change due to land cover changes may be important relative to greenhouse gases at the regional level, where intense land cover change occurs. Globally, the impact of greenhouse gas concentrations dominates over the impact of land cover change.