22

inexpensive to run, so that multiple diagnostic tests can be executed. They are useful mainly for exploring global scale questions. The upwelling-diffusion model, for example, has been used to investigate the role of the oceans in delaying the climatic response to increasing greenhouse gas concentrations and the role of ocean mixing-climate feedbacks in modifying the transient response (e.g., Hoffert, et al., 1980, Harvey and Schneider, 1985; Morantine and Watts, 1990), in exploring the importance of natural variability in observed global mean temperature variations during the past century (Wigley and Raper, 1990, Schlesinger and Ramankutty, 1995), in setting constraints on the magnitude of the global mean aerosol cooling effect (Wigley, 1989), and in assessing the relative roles of greenhouse gases, aerosols, and solar variability in explaining global mean temperature variations during the past century (Kelly and Wigley, 1992; Schlesinger and Ramankutty, 1992). The climate sensitivity in simple models is a prescribed parameter and is held constant for a given simulation. In complex models, the climate sensitivity is determined as a consequence of the explicitly computed processes and sub-grid scale parametrizations in the model, and is free to vary as the climate itself changes.

The sub-components of simple models can be constrained to replicate the overall behaviour of the more complex model subcomponents. For example, the climate sensitivity of simple models can be made to equal that of any particular AGCM or AOGCM by altering a single model parameter whose value implicitly accounts for the net, global mean effect of all the fast feedback processes which influence heat loss to space (on the other hand they cannot say, a priori, what that value should be). Similarly, the vertical diffusion coefficient and the upwelling velocity can be readily altered such that the oceanic uptake of heat (and associated sea level rise) closely matches that of any given OGCM. Globally-aggregated biosphere models can be adjusted to replicate the sensitivity to atmospheric CO2 and temperature changes obtained by regionally-distributed models. This allows the simple models to emulate the behaviour of the more detailed, regionally-resolved models.

Another consequence of the different computational demands of simple and complex models relates to initialization. Ideally, one should begin a simulation with anthropogenic forcing starting from a steady-state (or “balanced") climate, so that the simulated changes are due to the applied perturbation and not a consequence of the starting state. However, since the spin-up of coupled AOGCMs to a steady state requires thousands of simulated years, some anthropogenic forcing experiments using coupled AOGCMs have been started with the model in a nonequilibrium state. In such cases, "control" run projections with no imposed forcing yield a slowly changing or drifting climate. In order to determine the anthropogenic component of future change in such cases, one procedure is to subtract the control run climatic change from that of the perturbed run, on the assumption that the drift in the perturbed run is the same as in the control run, and that the clin.atic change and drift add linearly. This problem clearly complicates the experimental

design and could also affect the occurrence of abrupt ocean circulation changes. Simple one- and two-dimensional models, in contrast, can always be spun up to a steady state prior to applying an anthropogenic perturbation because of the low computational cost involved. In reality, natural variability exists in the atmosphere-ocean system, so that an exact equilibrium has never existed. However, the magnitude of such variability at the century time-scale is expected to be much smaller than human-induced climatic change over the next century.

One-dimensional models are clearly incapable of anticipating climatic "surprises", resulting from major changes in ocean circulation for example, although they can be used to assess the implications of such surprises. Complex AOGCMs have the potential to project such major changes in ocean circulation, although they are not yet crafted sufficiently well to do this reliably. Multi-basin versions of two-dimensional ocean models (ie., Stocker and Wright, 1991) which have been calibrated to simulate the observed climate and ocean circulation can also provide insights into the conditions under which major ocean circulation changes could occur.

3.6.2 Data Limitations of Biosphere Models

Spatially-detailed terrestrial biosphere models are highly dependent on data-sets of land cover, land use, terrain, climate and soil characteristics. The quality of the existing data sets is currently low due to classification problems, data availability and poor temporal and spatial coverage (SAR WGII: Section 2.5.3). The marine biosphere has, in some ways, a less complex spatial heterogeneity than the terrestrial biosphere and is therefore simpler to model. Nevertheless, the available data on spatial heterogeneity in the biosphere limits the use of spatially-explicit models and adds to their uncertainty in both input variables, parameter settings and results. Although some of the spatially explicit models are included in climate models (e.g., Goldewijk, et al., 1994), they are still mainly research tools to assess responses of the biosphere more comprehensively. Simple, globally aggregated models of the terrestrial and marine biosphere are currently more frequently used tools for the analyses of alternative scenarios involving the biosphere. These models have been calibrated against global scale observations but cannot simulate the detailed response of the biosphere. In the long run, regionally resolved models will have to be used.

each of a wide variety of emission scenarios. For these reasons, simple models were used extensively in the SAR WGI to explore the impact of alternative emissions scenarios of CO2 and other gases on global mean surface temperature change and sea level rise (SAR WGI: Sections 6.3, 7.5.2, and 7.5.3).

Relatively simple climate and carbon cycle models have also been used as one of the core components of Integrated Assessment Models (IAMs), which are based on the integration of models that simulate the most critical processes of the climate system (human emissions, biosphere, oceans and atmosphere), and are used to explore the impacts of diverse emissions scenarios generated by alternative energy sources, different land-use changes, pollution control, and population policies. Although the climate component of such models is globally- (e.g., Wigley and Raper, 1995) or zonally-aggregated (as in de Haan, et al., 1994), they have been linked to a number of regionally resolved sub-models spanning a wide range of human activities and impacts. One of the more advanced IAM is the IMAGE 2 model, which is described in Alcamo (1994). This model calculates emissions of different greenhouse gases

23

from energy and land use; computes atmospheric concentrations by accounting for atmospheric chemistry and carbon uptake by the oceans and biosphere; and computes changes in climate and sea level as well as impacts on ecosystems and agriculture. These calculations allow for a transient determination of driving forces (including changed policies), climatic change, and its impacts. The policy relevance of such models lies in the comprehensiveness of simulations of many components in the climate system (see Figure 3).

The premise behind using simple models for policy analysis, with their focus on global scale changes, is that preventative responses to the risk of climatic change might be a collective response based on global scale aggregated impacts and risks, rather than on the local impacts and risks for any given nation undertaking a response. On the other hand, regionally-resolved models are needed, in conjunction with sector- and regionspecific impact assessment tools, in order to translate global scale changes into specific impacts and hence to determine the globally-aggregated risk associated with a given magnitude and distribution of global scale change.

In this section, we provide details concerning the specific SCMs and modules, and the associated assumptions, used in the SAR WGI. We begin with the computation of radiative forcing from emission scenarios, followed by the projection of global mean temperature change, and finally, the projection of future changes in sea level (as illustrated in Figure 4).

4.1 The Biogeochemical Component of a Simple Climate Model: Turning Emissions into Radiative Forcing

The following subsections describe the methods used in the SCM simulations described in the SAR WGI to compute the perturbation in greenhouse gas concentrations (SAR WGI: Sections 2.1 and 6.3), and the radiative forcings associated with perturbations in greenhouse gas and aerosol amounts (SAR WGI: Section 6.3). The quantitative relationships used are summarized in Appendices 1 and 2.

4.1.1 Treatment of Well-Mixed Gases with Well-Defined Lifetimes

The rate of removal from the atmosphere of nitrous oxide (N2O) and the halocarbons is, to a first approximation, linearly proportional to the amount of gas in the atmosphere. That is, a fixed fraction of the amount of gas present at the start of a given year is removed per year, so that if the concentration of the gas doubles, for example, the mass removal rate doubles. These gases also have long lifetimes in the atmosphere relative to the time required for complete wind mixing to occur, so they are of relatively uniform concentration. As a result, the atmosphere can be regarded as a single, well-mixed box. The most important parameter is the average lifespan of a molecule of gas in the

atmosphere, t, which provides the link between concentration and rate of removal. Figure 6 illustrates the treatment of these gases. The numerical values of t as adopted in the SAR WGI (Section 6.3) are summarized in Appendix 1; since the main removal process for most gases occurs through chemical reactions in the atmosphere, we use the term atm in Appendix 1.

Methane (CH4) is somewhat more complicated in that t depends on the concentration itself. Nevertheless, the atmosphere can still be treated as a single well-mixed box as far as CH, is concerned. and concentration changes can be computed if the CH1 lifetime is updated during the course of the computations. Thus, Figure 6 can also be applied to CH, as long as it is understood that the lifetime varies with the concentration itself, so that the removal rate now varies non-linearly with the concentration. As noted in Section 3.3, the dependence of the CH4 lifetime on CH, concentration is affected by the concurrent concentrations of NO. CO and VOCs in the atmosphere, which vary significantly between regions. Emissions of these gases are also likely to change significantly over time, but, for purposes of computing changes in CH removal rate time in SAR WGI (Section 6.3), these emissions were assumed to be constant. This feedback is based on calculations using three-dimensional models, as discussed by Osborn and Wigley (1994). The currently estimated CH, lifetime is given in Appendix 1.

Unlike the gases discussed in the preceding section, CO2 does not have a well-defined lifetime. This is due to the multiplicity and complexity of processes involved in the removal of CO2 from the atmosphere (as discussed in Section 3.2). Figure 7 illustrates the carbon cycle components and flows that have been included in the simple carbon cycle models used in SAR WGI (Sections 2.1 and 6.3). In two of the simple models used in the SAR WGI those of Jain, et al., (1995) and Siegenthaler and Joos (1992) — ocean chemistry and vertical mixing processes are explicitly computed using the one dimensional upwelling-diffusion model or a variant of it. In the third model used in the SAR WGI — that of Wigley (1991) — a reasonably accurate mathematical representation of the uptake of carbon by an OGCM, which was first employed by Harvey (1988), is used.

These three carbon cycle models are such that, when driven by anthropogenic fossil fuel emissions, the simulated build-up of

26

atmospheric CO2 since the industrial revolution is close to that observed. Hence, when net emissions due to deforestation and forest regrowth are added (estimated to be 1.1±1.0 GtC/yr averaged over the 1980s; see SAR WGI: Table 2.1), additional sinks are required in order to avoid too large a CO2 build-up. One way to create such a sink, which is the method used in the SAR WGI calculations (Sections 2.1 and 6.3), is to specify a CO2 fertilization effect on the terrestrial biosphere. The larger the assumed past land-use emissions, the greater the required fertilization effect. If this effect is then extrapolated in some way (not necessarily linearly) into the future, the projected future CO2 concentration will be lower the greater the assumed past and present land-use emissions (given that land-use emissions will eventually fall). The long-term validity of this extrapolation is highly uncertain (SAR WGI: Sections 2.1.1 and 9.2.3.2: and SAR WGII: Section A.2.3)

As noted in SAR WGI (Section 2.1) and in IPCC94 (Chapter 1). there are other mechanisms besides a CO2 fertilization effect through which the carbon cycle could be balanced in the presence of net land-use emissions. For example, nitrogen fertilization of portions of the terrestrial biosphere as a by-product of NO, emissions could be causing an extra 0.2-1.0 GtC/yr of carbon to be taken up (SAR WGI: Sections 2.1.1 and 9.2.3.4). Climatic changes during recent decades could also be causing the terrestrial biosphere to absorb a significant amount of carbon (SAR WGI: Sections 2.1.1 and 9.2.3.1). To the extent that these mechanisms have been operative, the CO2 fertilization effect is weaker, to the extent that they do not increase as quickly as a CO2 fertilization effect, extrapolation of an overestimated CO2 fertilization effect will lead to projected atmospheric CO2 concentrations that are too small.

A number of other processes that could influence future atmospheric CO2 concentrations have also been neglected in projections of global mean temperature and sea level (SAR WGI: Section 6.3) and in the CO2 stabilization calculations (SAR WGI: Section 2.1). In particular, no account has been taken of the potential for accelerated respiration of biomass and soil carbon due to warmer temperatures (leading to a potentially large release of CO2), release of carbon to the atmosphere due to die back of forests if climatic zones shift too rapidly, or the impact of warmer ocean temperatures and changes in ocean circulation on the oceanic uptake of CO2 (potentially leading to either a small release or additional absorption of CO2). Until the relative importance of alternative mechanisms for absorbing anthropogenic CO2 is better known, quantification of the uncertainties in future atmospheric CO2 projections will remain difficult (see SAR WGI: Chapters 9 and 10 for a discussion of the potential impact of these processes on the carbon cycle).

4.1.3

Treatment of Gases not Directly Emitted Tropospheric ozone is produced indirectly through chemical reactions involving CH4, CO, NO,, and VOCs, which have both natural and anthropogenic sources. Proper computation of tropospheric ozone build-up requires three-dimensional atmospheric chemistry/transport models. Since the relationships between CO, NO,, VOCs and tropospheric O3 build-up are uncertain, and the adequacy of current three-dimensional models is questionable, only the increase in tropospheric O3 associated with increasing CH4 concentration has been included in the SAR WGI (Section 6.3) projections of global mean temperature and sea level beyond 1990. This forcing is assumed to be directly proportional to the increase in methane concentration, with a value of 0.08 W m2 in 1990. Up to 1990, tropospheric ozone radiative forcing associated with emissions other than CH, is also included. This forcing is assumed to have been proportional to fossil fuel emissions and to have reached a value of 0.32 W m22 by 1990, and is then held constant. The total forcing in 1990 due to changes in tropospheric ozone has an uncertainty of at least £50 per cent (see IPCC94: Section 4.3.6).

Problems also remain with regard to stratospheric models, which still cannot fully explain observed stratospheric O3 losses. In the SAR WGI projections of global mean temperature and sea level (Section 6.3), stratospheric O3 loss is assumed to vary with the tropospheric chlorine loading to the power 1.7, plus a bromine loading term weighted relative to chlorine by a factor of about 40 at present. The forcing associated with stratospheric ozone loss is then assumed to be directly proportional to the ozone loss, leading to the relationship between forcing and chlorine and bromine loading given in Appendix 2. This relationship was calibrated by comparing the computed global mean forcing due to stratospheric ozone changes with detailed radiative transfer calculations based on the observed ozone loss over the period 1979 to 1990 (Ramaswamy, et al., 1992). The total direct halocarbon forcing in 1990 calculated using the expression in Appendix 2 is 0.27 W m2, and 0.1 W m2 when stratospheric 03

The global mean concentrations of three kinds of aerosols have increased through human activity by a sufficiently large amount to have important effects on climate: sulphate (SO4) aerosols, which are produced from the oxidation of sulphur-containing precursors and which are emitted through the combustion of coal and oil and from smelting of certain metals; soot (black carbon) aerosols, directly released from combustion of coal, oil, and biomass; and organic aerosols (other than soot), released from the combustion of biomass or produced from chemical transformation of VOCs (IPCC94: Chapter 3). Dust aerosols from land-surface changes might also have noticeable climatic impacts (SAR WGI: Sections 2.3 and 2.4)

As discussed in Section 3.3, the processes determining the amount, distribution, and properties of aerosols in the atmosphere can be simulated, and the global mean forcing computed, only by using three-dimensional AGCMs. When using SCMs, one must therefore use results from AGCMs to establish a quantitative link directly between present global emissions and present global mean forcing. Because the atmospheric aerosol burden responds essentially instantaneously to changes in emissions, specification of an emission scenario amounts to specifying a concentration scenario. In the SAR WGI (Section 6.3), the relationship between emissions and atmospheric aerosol loading is assumed to be linear. Although this is not exactly true, the error so introduced is overwhelmed by uncertainties in the link between atmospheric aerosol loading and global mean radiative forcing. In practice, atmospheric aerosol loading is not explicitly computed; rather, global emissions are directly linked to global mean forcing using the results of AGCMs (as discussed below in Section 4.1.5).

For sulphur, two emission scenarios were considered in the SAR WGI (Section 6.3): one in which anthropogenic emissions are held constant at the 1990 level after 1990, and one in which the emissions of SO2 are as specified in the IS92a scenario (IPCC, 1992: Table A3.12). In the latter case, total anthropogenic sulphur emissions will increase from 75 TgS in 1990 to 147 TgS in 2100. Dust aerosols are neglected in the SAR WGI (Section 6.3) projections of global mean temperature and sea level, while the radiative forcing associated with organic aerosols from biomass burning is assumed to scale with gross deforestation up to 1990 (when the forcing is assumed to have been -0.2 W m2), then is held constant.

Calculating Radiative Forcing From Concentrations

27

Given the concentrations of globally uniform greenhouse gases, the direct radiative forcing can be computed by using simple formulae which provide a close fit to the results of detailed radiative transfer calculations. In the case of CH, indirect forcings also arise through the formation of stratospheric water vapour from oxidation of CH4, and through effects on tropospheric O3. In the SAR WGI (Section 6.3), the stratospheric water vapour forcing is assumed to vary directly with the CH forcing, while the tropospheric O3 forcing due to CH, emission is assumed to vary linearly with the increase in CH, concentration (see Appendix 2).

The forcing associated with both stratospheric and tropospheric O, changes varies substantially regionally, since the 03 changes themselves exhibit strong regional variation (IPCC94: Section 2.6; SAR WGI: Section 2.2). It is assumed in the SAR WGI (Section 6.3) that the global mean climatic response is proportional to the global mean forcing, which in turn is assumed to be directly related to the change in global mean concentration. As noted in the SAR WGI (Section 2.2), changes in stratospheric O, provoke further radiative forcings through induced changes in tropospheric chemistry, and this indirect forcing could be two to three times the direct forcing. Due to uncertainties in the magnitudes of these potential effects, they have been neglected in the SAR WGI projections of global mean temperature and sea level. As noted in Section 2.3.4, the assumption that the relationship between global mean temperature response and global mean forcing is the same for O3 as for CO2 might introduce further error. However, this error is at present overwhelmed by the large (factor of two to three) uncertainty in the forcings due to both tropospheric and stratospheric O3 changes.

As discussed in Section 4.1.4, the global mean aerosol forcing in the models used in the SAR WGI (Section 6.3) is based on the ratio of present-day global emissions to present-day forcing, as computed from an AGCM for a limited number of aerosol distributions. Since atmospheric aerosol concentrations vary directly and immediately with emissions, this contains an implicit relationship between concentration and forcing. The direct component of the forcing is assumed to vary linearly with concentration and hence with emissions, while the indirect forcing is assumed to increase more slowly than emissions, based on our understanding of the key physical mechanisms involved. Both the direct and indirect global mean forcings by sulphate aerosols are highly uncertain (SAR WGI: Section 2.4.2 and 6.3.2); in the SAR WGI projections of global mean temper ature and sea level, these forcings are assumed to have been -0.3 W m2 (out of an uncertainty range of -0.2 to -0.8 W m22) and -0.8 W m2 (out of an uncertainty range of 0.0 to -1.5 Wm2), respectively, with the indirect forcing varying with the logarithm of concentration and thus of emission (see Appendix 3). Thus, as sulphate aerosol loading increases the indirect forcing becomes smaller relative to the direct forcing.