Online climate change projections report Annex 3.3
The current generation of climate models can capture the broad-scale features of present day climate (Figures A3.2 and A3.3) and historical climate change (Figure A3.1). This is particularly true for surface variables such as temperature and mean sea-level pressure and for those three-dimensional fields which capture the large-scale structure of winds and temperatures throughout the atmosphere. Even for fields such as mean precipitation, the models are able to reproduce many of the large-scales features with some fidelity. These features are generated by the dynamical and physical processes in the model and are not prescribed.
Nevertheless, models are certainly not perfect even on large-scales, as evident but the panels in Figures A3.2 and A3.3 which show differences between the model ensemble mean fields and the observations. For example, the ensemble mean of the HadCM3 ensemble with perturbations to atmosphere-component parameters (PPE_A1B — see Chapter 3) shows a clear warm bias in summer Northern Hemisphere continental regions (which we discuss latter). In addition, there are biases which are common to both the perturbed physics and multi-model ensembles. Models tend to produce a double ITCZ (Intertropical Convergence Zone) in the Pacific whereby zonally-oriented large-scale rain bands appear in both hemispheres, where in reality, the southern hemisphere rain band is oriented NW-SE. In addition, variables such as convective (shower) precipitation can be highly localised so are harder to model, as are fields such as surface winds. When regional factors are important — for example in highly mountainous regions — global models may find it hard to capture the small-scale details of the present day climate. Hence there is plenty of room for improvement in climate models and this is an extensive field of research, both within the Met Office Hadley Centre and internationally. (Further discussion of model evaluation is presented below and can also be found in, for example, Chapter 8 of IPCC AR4. Discussion of the mean climates of the regional model versions can be found in Chapter 5 of this report).
A critical issue for prediction is how these model errors and biases affect the pattern and magnitude of climate change. The main drivers of climate change are global in nature in terms of their radiative forcing and there is a significant degree of commonality between models in terms of their large-scale projections of mean future change (Figure A3.4). The commonality is stronger in the case of temperature, but there are also similar patterns of response in terms of the mean precipitation in models. Different models all show greater warming over land compared to over the ocean and greater warming at high-latitudes in comparison with the tropics in the winter hemisphere. The latter may be understood in terms of simple physical reasoning: in this case, albedo feedbacks whereby snow or ice covered regions become exposed as the planet warms and, as a result, more sunlight is absorbed by the underlying surface. Other important feedbacks include the positive water-vapour feedback; water vapour (a potent natural greenhouse gas) will increase as air temperature increases. The directions of such feedbacks are relatively well understood but their absolute magnitude is still under investigation. Feedbacks from clouds represent a significant source of uncertainty in total global feedbacks and these may also drive variations in local climate changes (clouds remain one of the most-complex and most-studied of feedbacks under climate change). Because of these global-scale uncertainties, the PDFs presented in this report are (a) constructed from a relatively large number of ensemble members which explore uncertainties in large-scale feedbacks and (b) constrained by a number of observed large-scale fields; the relative likelihood of each model version in its ability to simulate the large-scale nature of climate and historical climate change is taken into account (see Chapter 3).
Looking more locally, we see similar patterns of warming in both summer and winter in region of the UK and NW Europe, with the multi-model ensemble mean showing a slightly greater ensemble mean warming than in the case of the perturbed physics ensemble mean. Perhaps more surprising is the similarity of the patterns of precipitation change in the two different ensembles, with increased precipitation during the winter over much of NW Europe and a drying in the Mediterranean region in summer. This indicates common physical mechanisms for the change between different models. Nevertheless, those physical mechanisms may act in subtly different geographical areas and with different strengths in different models. In the summer case, the perturbed physics ensemble drying extends more into the north and over the UK, whereas in the multi-model ensemble the line of zero mean change cuts the UK. This is why it is so important to include information from other climate models in UKCP09.
For some variables the response to climate change may be quite different in different perturbed physics or multi-model members and the resulting PDFs of change quite wide. We should not necessarily assume that the use of the multi-model ensemble in generating the PDFs provides some kind of upper-bound uncertainty in the predictions. The existence of common errors in multi-model and perturbed physics ensembles may, for example, impact the pattern or magnitude of the climate change response seen in all ensembles. There may be other possible formulations of models which could give rather different responses that could affect the level of uncertainty in the PDFs. Nevertheless, without any evidence of the possibility of very different climate change, the most defensible approach is to look to the multi-model ensembles to provide evidence for a discrepancy in PDFs generated from the perturbed physics ensembles (see Chapter 3 and Annex 2 for more details). The impact of model formulation (e.g. horizontal and vertical resolution) on the magnitudes and patterns of climate change is a very active area of research.
In general, regional aspects of climate change may be influenced by local regional processes such as the enhancement of rainfall on the windward-side of mountainous regions. Hence the use of the ensemble of regional-model simulations and statistical downscaling techniques in generating the PDFs presented here. Importantly, the regional models are driven by output from the global models that represent the large-scale pattern of climate change. Hence there is an internal consistency in the information which is derived completely from model output.