Online climate change projections report 3.2.6 Combining uncertainties
The Earth System ensembles described in Section 3.2.5 are not large enough to provide a basis for training an emulator capable of estimating the model response at any point in the parameter space of ocean, sulphur cycle or processes (Section 3.2.3). This prevents us from including the relevant uncertainties via a formal application of Bayes' Theorem in an integration over the model parameter space (Section 3.2.7). However, we do include uncertainty estimates obtained from these ensembles in a simpler manner, by generalising the timescaling technique described in Section 3.2.4. This is done by configuring the simple climate model used in timescaling to include sulphate aerosol forcing, and simple globally averaged parameterisations of processes associated with the effects of terrestrial carbon cycle feedbacks on the atmospheric CO2 concentration. When running the simple model to estimate the transient climate response for some specified set of surface and atmospheric HadCM3 parameters, we sample the effects of additional Earth system processes by selecting from a distribution of possible values for the simple model parameters controlling global mean ocean heat uptake, sulphate forcing or CO2 concentration. For heat uptake, this is done by calculating values of ocean diffusivity for each of the 17 members of our ocean perturbed physics ensemble (Section 3.2.5), and also from 20 alternative simulations derived from the multi-model ensemble of coupled ocean-atmosphere models submitted to the IPCC AR4. The multi-model ensemble values were taken from the 23 models listed in Table 8.1 of Randall et al. ( 2007), omitting two models because data required for the calculation were not available, and one because the wrong climate change forcing was applied in the relevant experiment. Inclusion of the multi-model ensemble results enabled us to account in a simple way for structural uncertainties in ocean transport processes not sampled in our perturbed ocean ensemble. Values are then selected from these 37 possible values, assuming each to be equally plausible.
Including sulphate aerosol forcing uncertainties in timescaled projections
For sulphate aerosol forcing the approach is somewhat more complicated, because variations in physical atmospheric parameters (particularly those associated with cloud) are found to exert a significant influence on the forcing, in addition to variations in parameters directly associated with the sulphur cycle. Furthermore, a significant relationship between global mean aerosol forcing and climate sensitivity was found in our PPE_A1B_NOGHG ensemble (low sensitivity model variants tend to simulate high levels of low cloud, and therefore simulate larger changes in forcing in response to aerosols). We accounted for these factors by developing a regression relationship between a transformed function of aerosol forcing, and global climate feedback (the reciprocal of climate sensitivity). The distribution of forcing values is in the transformed units. Variations in transformed aerosol forcing, diagnosed from the 16-member perturbed sulphur cycle ensemble, were assumed independent of atmospheric perturbations and added to each member of our PPE_A1B_NOGHG ensemble, thus forming a dataset for regression which sampled uncertainty arising from both atmospheric and sulphur cycle processes. When running the simple climate model for a given location in parameter space (and hence a given climate sensitivity), we then sampled alternative aerosol forcing values from the error statistics of the regression relationship. This method gives a distribution of aerosol forcing values for present day climate (relative to pre-industrial conditions) similar to that given in the IPCC AR4 (see Figure 2.20 of Forster et al. 2007), based on the statistical assessment of the uncertainty of radiative forcing mechanisms documented by Haywood and Schulz (2007).
Including carbon cycle feedback uncertainties in timescaled projections
Given that carbon cycle uncertainties provide a leading order contribution to the uncertainty in global mean changes, and recognising that our perturbed physics ensemble does not sample uncertainties associated with structural carbon cycle assumptions in HadCM3C, we also include results from the C4MIP multi-model simulations in our sampling of possible feedbacks. We performed a pre-screening exercise in which the historical simulations of global carbon budget components (fraction of anthropogenic emissions stored in atmosphere, land and ocean) were compared with an observational constraint based on records of atmospheric CO2 increase, estimates of total emissions (fossil fuel plus land use emissions) and the oceanic uptake of anthropogenic CO2 (Sabine et al. 2004). Two of the perturbed physics simulations and one of the C4MIP simulations were found to be inconsistent with the spread of plausible values implied by estimates of observational uncertainty, so these were excluded. We also excluded results of the HadCM3LC model contributed to C4MIP, as this model is strongly related to that used for our perturbed physics simulations. This left 9 members of the C4MIP ensemble and 15 members of the perturbed physics ensembles, whose simulated global mean feedbacks were sampled in the timescaling procedure, assuming all 24 estimates to be equally plausible.
The parameterization of carbon cycle feedbacks in the simple climate model contains explicit temperature dependences, allowing the (significant) effect of variations in the global temperature response on the global mean carbon cycle response to be captured (e.g. Andreae et al. 2005). This is achieved using globally averaged calculations of changes to the vegetation and soil carbon stores consistent with the main features of the corresponding calculations used in the terrestrial ecosystem module of HadCM3 (Jones et al. 2003), which contains temperature-dependent parameterisations of photosynthesis and plant and soil respiration. With the exception of this carbon cycle-temperature relationship, and the aerosol forcing-climate sensitivity relationship described above, our timescaling method does not account for non-linear interactions between the global feedbacks in different Earth System modules. This is because time and resource limitations prevented us from running HadCM3 ensemble simulations in which parameters in all component modules were varied simultaneously. The UKCP09 projections are conditional on the assumption that additional non-linear interactions are likely to be small compared with the two significant known relationships referred to above. This issue is a subject of current research.
Potential contributions of ocean, sulphur cycle and carbon cycle processes to uncertainties in regional climate changes (beyond the effects directly attributable to uncertainties in global mean surface temperature) are not accounted for in the generalised timescaling technique. This is because results from the relevant ensembles indicate that such contributions would be relatively minor for changes over the UK (Section 3.2.5), and also because quantification of the impacts of non-linear interactions is beyond the scope of the experimental design for UKCP09 (see above). In some regions neglect of such regional effects would not be realistic, a good example being Amazonia where carbon release from forest dieback is dependent on regional changes in precipitation (Betts et al. 2004). The extent to which the UKCP methodology could be applied in other parts of the world will therefore depend upon careful evaluation of the potential impacts of regional effects not covered by our timescaling procedure, in addition to the validity of further assumptions required by our technique, such as the use of a linear scaling to global mean temperature changes (see Section 3.2.4).