Modelling regional climate change
Obtaining reliable projections of climatic changes at the regional scale
is one of the most central issues within the global change debate. In order to
assess the social and environmental impacts of climate change and to develop
suitable policies to respond to such impacts, information about climate change
is needed by policy-makers not only at a national level, but on a regional and
local scale as well.
Despite the central importance of this issue, however, current uncertainties remain very high in the projection of regional climatic changes for the next several decades. This is due to the complexity of the processes that determine regional climate change, and the need for more comprehensive modelling tools and research strategies to address this problem.
Global climate modelling has undergone a steady development during the last three to four decades. However, only in the last decade or so have research efforts been directed towards developing models and approaches aimed at improving information about climate change on the regional scale. As a result, this area of research is still in its early stages and it is undergoing very rapid development.
This article starts by reviewing briefly the main processes that affect regional climate change. It then discusses the techniques that have been developed to enhance climate information at a regional and local scale. The main uncertainties underlying the current state of regional climate change projections are finally discussed. A substantial amount of literature on these topics has accumulated in recent years, and a number of general references are reported at the end of the article.
The processes of regional climate change
It must be first understood what is meant by ‘regional scale’. Indeed, different definitions are often implied in different contexts. For example, definitions of regional scale can be based on geographical, political and physiographic considerations, or considerations of climate homogeneity.
Because of this difficulty, in this article we adopt a broad definition, so that the term ‘regional scale’ covers the broad range of 1000 to 10,000,000 square kilometres. This range includes relatively small regions within a country, such as the Po river valley in Italy, to large regions within a continent, such as the Mediterranean Basin or the Sahel.
Smaller scales than this range are referred to as ‘local scale’, while larger scales are referred to as ‘large scale’. Given these definitions, the problem of regional climate change is essentially a problem of interactions of processes across scales.
The general circulation of the atmosphere — the wind patterns that characterise the global atmosphere — is regulated by large-scale processes and ‘forcings’, the term used to describe those factors that the Earth’s climate responds to. Examples of large-scale forcings are the flow of radiation from the Sun, the concentration of greenhouse gases in the atmosphere, the ocean temperature, the large-scale distribution of continents, oceans and ice, and large topographical systems (for example the Tibetan Plateau).
The general circulation of the atmosphere determines the sequence of weather events and wind patterns that characterize the climate of a region. Embedded within the large-scale circulation systems, regional and local forcings and wind circulations modulate the spatial and temporal structure of the regional climate.
Regional and local scale forcings can, for example, result from complex topography, land use characteristics, inland bodies of water, complex coastlines, short-lived radiatively active gases and aerosols, regional snow and sea ice distributions. Each of these can have a strong influence on climate fluctuations at the regional or local level.
Additional factors need also to be considered. First, the atmosphere is a highly non-linear system, and as a result processes occurring in one region can significantly influence the climate of distant regions (a process referred to as ‘teleconnection’).
Typical examples of this effect are the El Niño Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO) phenomena, in which regional anomalies in sea surface temperature and sea level pressure in the Tropical Pacific Ocean and North Atlantic, respectively, can affect weather regimes in areas far from the source of the anomaly.
Long-lived large-scale wind circulation patterns due to the chaotic behaviour of the atmosphere can also affect regional climates.
Another factor that needs to be recognized is that climate is determined not only by atmospheric processes, but also by complex interactions among different components of the ‘climate system’, namely the atmosphere, the oceans and inland water bodies, the ice, the biosphere and the chemical constituents of atmosphere and oceans. Interactions across the climate system take place over a wide range of temporal and spatial scales, and can be highly non-linear.
The difficulty in producing reliable regional climate change projections therefore results from the need to describe properly the effects of non-linear climate system interactions, forcings and circulations at the global, regional and local scale, along with the teleconnection effects of regional forcing anomalies.
This presents a formidable cross-disciplinary and multi-scaled scientific task that cannot be fully achieved by any of the individual modelling tools that are presently available. Rather, it requires combining different modelling techniques.
Regional climate modelling
The primary tools today available to simulate long-term climate change are known as coupled Atmosphere-Ocean General Circulation Models (AOGCMs). These are three-dimensional mathematical representations of the global atmosphere-ocean-sea ice system. Advanced AOGCMs also include representation of land biosphere, atmospheric chemistry and atmospheric aerosols.
In a climate simulation, an AOGCM describes the evolution over time of the atmosphere and ocean. These simulations are carried out using powerful supercomputers. A ‘transient’ climate change experiment consists of the description with an AOGCM of the evolution of climate through an historical time period up to a future time period. Typically transient climate change experiments cover the period of 1860-2100.
During the historical time period, climatic forcings (for example greenhouse gas concentration) are obtained from available observations. During the future time period, the climatic forcings are derived from different estimates, or ‘scenarios’ of future population, economic and technological development.
Limitations in computer power force the horizontal grid interval (in other words the space between the points in which model data can be obtained) of present day AOGCMs to be about a few hundred kilometres. As a result, processes and atmospheric circulations occurring at smaller scales cannot be explicitly described.
This represents a fundamental drawback to the use of AOGCMs for projections of regional climate change. In practice, in most regions of the world significant differences in climate can occur over distances of a few tens of kilometres or even less.
In addition, the coarse resolution of AOGCMs prevents them from properly describing extreme atmospheric events, such as hurricanes and tropical storms, that are important elements of the climate of many regions of the world.
Although computing power is rapidly increasing, it will still be several years before AOGCM simulations become feasible at high resolution. Different ‘regionalisation’ techniques have therefore been developed over the last decade or so to improve the regional information provided by coupled AOGCMs, and to provide climate information at a fine scale (Giorgi and Mearns 1991; Giorgi et al. 2001a).
Such techniques can be classified into three categories:
1. High and variable resolution ‘time-slice’ Atmosphere General Circulation Model (AGCM) experiments;
2. ‘Nested’ limited area (or regional) climate models (RCMs);
3. Statistical downscaling methods.
There is a common strategy behind the use of these regionalisation techniques. The AOGCM is used to simulate the response of the global atmospheric circulations to changes in external forcings (for example in the concentration or greenhouse gases or in the solar energy input to the atmosphere). This response accounts for the contribution of the interactions within the climate system that are described in most AOGCMs, for the natural variability of the atmosphere and for the non-linear teleconnection patterns produced by regional forcing anomalies.
In this respect, our current AOGCMs have performed relatively well in reproducing many basic characteristics of the general circulation, such as major belts of precipitation in the tropics or the seasonal migration of mid-latitude storm tracks. These climate models have also shown some success in describing the El Niño Southern Oscillation and North Atlantic Oscillation phenomena and related teleconnection patterns, although significant improvements are still needed here (McAvaney et al. 2001).
Once the AOGCM simulations are completed, the regionalisation techniques can be used to enhance the AOGCM information at a regional level by providing an account of the effects of forcings and processes that take place at the sub-AOGCM grid scale.
It should be pointed out that up to the present day, most climate impact studies have used information on climate change directly obtained from simulations with AOGCMs, without using any ‘regionalisation’ tool. This is primarily because of the ready availability of such information and the fact that techniques for simulating climate at a regional level have only recently been developed.
For some applications, the regional information provided by AOGCMs may be sufficient — for example in cases where the climate variables of interest do not vary greatly over the scales of interest. In addition, AOGCMs have had some success in reproducing surface climate characteristics on a subcontinental scale (Kittel et al. 1998; Giorgi and Francisco 2000; Giorgi et al. 2001a).
However, where the climate shows considerable variation at a sub-AOGCM scale, the use of a regionalisation technique is essential to enhance the AOGCM information at the regional and local level.
The ‘added value’ provided by regionalisation techniques depends on the spatial and temporal scales of interest. It is important, however, to stress that AOGCMs remain the starting point for all regionalisation techniques that are currently used.
It is therefore very important that AOGCMs perform well in simulating the large-scale atmospheric features that affect regional climates. In other words, regionalisation tools are not alternative to AOGCMs. Rather they are complimentary tools that increase the effectiveness of climate simulation at the regional scale.
High resolution and variable resolution experiments
This technique consists of identifying periods of interest (or ‘time-slices’) within AOGCM transient climate simulations, and modelling these at a higher resolution, or with a variable resolution atmospheric global model to provide additional spatial detail (e.g. Bengtsson et. al., 1995; Cubasch et al., 1995; Deque and Piedelievre 1995). The variable resolution AGCM would have its maximum resolution at the region of interest, and its resolution would gradually decrease as one moved away from this region.
Because only an atmospheric model is used, data for sea surface temperature and sea ice distribution (in addition to the greenhouse gas and aerosol concentration) are needed to perform an AGCM climate simulation. These data are obtained from the corresponding periods selected from a coupled AOGCM experiment in which both the evolution of the ocean and of the ice are simulated. Typically, time slices covering the ‘present day’ (for example 1960-1990) and a future period (for example 2070-2100) are simulated in order to calculate changes in relevant climatic variables.
The philosophy behind the use of time-slice AGCM simulations is that, given the climate forcing obtained from the corresponding AOGCM simulation, relatively high resolution information can be obtained either globally or regionally without having to carry out the whole transient simulation using high resolution climate models.
Uniform resolution time-slice AGCM experiments have already been run at grid intervals of about 100 km covering the globe, while variable resolution AGCM experiments have reached grid intervals of about 50 km over specific regions.
The use of both techniques is based on the assumption that the large-scale circulation patterns in both the coarse and high resolution GCMs are not very different from each other. If this were not the case, it would throw doubt on the consistency of the high-resolution AGCM climate model, and the forcing from the sea surface temperature and sea ice in the coarse resolution AOGCM would be questionable.
The main advantage of this approach is that the resulting simulations are globally consistent and capture the feedback from the regional high-resolution atmospheric circulations on the global climate. However the relatively low level of experience with high-resolution AGCMs means that although increasing the resolution of a global model improves some aspects of the model’s climatology, it can degrade others.
For example, model parameters often need to be ‘re-tuned’ at the higher resolutions, and the descriptions of physical processes used by the model need to apply over a wide range of scales. The use of both high-resolution and variable-resolution global models is computationally very demanding, placing limits on the extent to which resolution can be increased using this method.
Nested regional climate models
Regional or limited area climate models are mathematical representations of the atmosphere limited to specific regions of interest, rather than at a global scale. Because regional models cover limited areas, with the available computing resources they can use grid intervals much smaller than those used by AOGCMs.
They can therefore describe processes that occur at a much finer spatial scale. In order to carry out a regional model simulation, meteorological data are needed at the lateral boundaries of the domain (or area) covered by the model.
The so-called ‘nested’ regional climate modelling technique consists of using the meteorological fields obtained from a global model simulation to provide the meteorological lateral boundary conditions needed to perform high-resolution regional model simulations (e.g. Dickinson et al. 1989; Giorgi 1990; McGregor 1997; Giorgi et al. 1994; Jones et al. 1995, 1997; Giorgi and Mearns 1999).
The regional model simulations typically cover selected time periods within the global model simulation. The global model that provides the boundary conditions for the regional model is usually referred to as ‘driving’ model.
Also needed for the regional model simulation is the forcing from sea surface temperature, sea ice, greenhouse gases and aerosols. These are also provided by the driving AOGCM.
In summary, the nested regional model can be visualized as providing a high-resolution zoom in effect over a selected region. Up to now, this technique has only been used in one direction, that is with no feedback from the regional climate model to the global climate model.
The objective of the regional model simulation is to account for forcings that take place over a region on a scale finer than the grid interval of the global climate model, and to enhance the simulation of atmospheric circulations and climate variables (for example temperature and precipitation) at fine spatial scales.
The nested regional modelling technique essentially originated from numerical techniques used in weather prediction. However, it is by now extensively used in various climate applications, from studies of palaeoclimates to those of anthropogenic climate change.
Over the last decade, regional climate models have proven to be flexible tools, capable of reaching high resolution (down to 10-20 km or less) and simulation times of several decades. They have also been able to describe successfully climate feedback mechanisms acting at the regional scale.
However, this approach has some theoretical limitations. For example, the regional model simulations are affected by systematic errors in the driving meteorological fields provided by global models, and two-way interaction between regional and global climate is not described. In addition, for each application, careful consideration needs to be given to the ways that the models are configured.
From the practical viewpoint, regional model simulations can be demanding on computational resources, both in terms of computation power and data storage. Despite such obstacles, however, simulations with current regional models have already used grid intervals of a few tens of kilometres and less. This still implies that processes occurring at finer scales are not explicitly described.
An alternative approach to regional modelling involves statistical downscaling. Under this approach, regional or local climate information is derived by first developing a statistical model which links large-scale climate variables (or ‘predictors’) to regional and local variables (or ‘predictands’). The large-scale output of an AOGCM simulation is then fed into this statistical model in order to estimate the corresponding local and regional climate characteristics.
A range of such statistical downscaling models have been developed for regions where sufficiently good datasets are available to allow the models to be properly calibrated. These techniques which, like regional modelling, also originate from weather prediction are currently used for a wide range of climate applications. (von Storch 1995; Hewitson and Crane 1996; Wilby and Wigley 1997; Murphy 1999; Zorita and von Storch 1997).
One of the main advantages of statistical downscaling techniques is that they do not require large amounts of computational resources, and they can therefore be easily applied to output from different AOGCM experiments.
Another advantage is that they can be used to provide information at specific locations, while regional models are still limited by their grid interval.
However, statistical downscaling methods are based on empirical models and not on models that explicitly describe the physical processes that affect climate and this may limit their applicability.
In addition, the major theoretical weakness of statistical downscaling methods is that the fundamental assumption on which they are based — that the statistical relationships developed for present-day climate also hold under the different forcing conditions of possible future climates — is often not verifiable.
Sources of uncertainties in modeling regional climate change
It is clear from the discussion above that there are several levels of uncertainty involved in modelling regional climate change.
Firstly, there is the uncertainty associated with the emission and corresponding concentration of forcing agents such as greenhouse gases and aerosols. This uncertainty is due to the possible range of future social, economical and technological developments as well as feedbacks within the global carbon cycle.
The next level of uncertainty results from the simulation of the transient climate response by coupled AOGCMs for a given emission scenario.
In general, different AOGCMs simulate different responses to the same forcing due to the different ways that the models represent the processes that affect climate.
This uncertainty has both global and regional aspects. For example, projected precipitation changes over the same region may vary substantially not only in magnitude but also in sign using different AOGCMs (Giorgi and Francisco 2000).
Finally, the third level of uncertainty is related to the regionalisation of the AOGCM simulations. Not only is the uncertainty in AOGCMs transmitted to the regionalisation tools, but also different regionalisation models generally provide different downscaled climates even under the same AOGCM forcing.
Sources of uncertainty are of different nature. On the modelling and statistical downscaling side, there are uncertainties associated with an imperfect knowledge and/or representation of physical processes, simplifications and assumptions in the models and/or approaches, and the inter-model or inter-method differences in the representation of the climate response to given forcings.
It is also important to recognise that regional climate observations can be characterized by a high level of uncertainty, especially in remote regions and in regions of complex topography.
Finally, the internal variability of the global and regional climate system adds a further level of uncertainty in the evaluation of a climate change simulation.
It is difficult to find unambiguous criteria to evaluate the uncertainty and reliability of a particular regional climate prediction. In general, a model’s ability to reproduce observed historical climate and climate variations suggests increased reliability of the climate change simulation.
One measure of uncertainty could therefore be related to the extent to which a model simulation deviates from observed climate.
Another measure of reliability is a model’s ability to reproduce known climate conditions that are different from those currently prevailing, such as palaeoclimatic conditions that existed thousands of years ago.
A third measure of reliability could be the extent to which simulations by different models (or methods) converge. Based on this criterion, a measure of uncertainty could be the spread of results across models (or methods).
Within this context, however, converging simulations might also indicate that models share common flaws, as fundamental modelling assumptions are shared by most models. Substantial research efforts are currently taking place to devise rigorous methods for assessing the level of uncertainty related to global and regional climate change projections.
As we have seen, the task of simulating regional climate change and evaluating the uncertainties associated with the simulations is an extremely difficult one. The 1996 IPCC Second Assessment Report (Kattenberg et al. 1996) presented essentially no conclusions on regional climate change projections, apart from those in broad latitudinal bands.
Since then, the availability of a greater number of AOGCM simulations with better quality models has allowed some emerging patterns of change to be recognized over specific regions of sub-continental scale (Giorgi et al. 2001a,b). So far, however, regionalisation methods, the development of which is still in its relatively early stages, have not been used comprehensively to produce reliable fine-scale projections.
On the other hand, the work being carried out with all the regionalisation techniques described above clearly indicates that fine scale structure of the regional climate change signal can occur in response to regional and local forcings, even though more work is needed to assess the statistical significance of this signal.
In particular, modelling evidence shows that both topography and the land surface conditions strongly affect the surface climate change signal at scales smaller than the grid interval of AOGCMs.
This implies that the information obtained from AOGCMs needs to be used cautiously in studies of the impacts of climate change, particularly in regions that are characterised by pronounced variability in forcings on fine scales. It also implies that suitable regionalisation techniques should be used to enhance the results obtained from AOGCMs over these regions.
As we have seen, different regionalisation techniques have different advantages and limitations, and a range of considerations may influence the choice of a particular technique in any situation. In practice, the joint use of different techniques may in some instances provide the most suitable approach.
For example, a high resolution AGCM simulation can be used as an intermediate step between coupled AOGCMs and regional climate models or statistical downscaling methods. Furthermore, the convergence of results from different approaches applied to the same problem can increase the confidence in the results, while differences between approaches can help us to understand the behaviour of the models.
At present, a number of regional modelling systems are available which are well tested, usable on different computing platforms (including powerful PCs and desktop workstations), and applicable to any region of the world. Similarly, a number of statistical downscaling models and methods have been developed for a wide range of uses.
Despite this availability, however, the potentials and limitations of each of the different techniques need to be well understood before they are applied to the construction of specific regional climate change scenarios.
We have only recently reached a point at which different regionalisation techniques are understood well enough to be applied to the production of climate change scenarios; indeed, the uncertainties related to these techniques are still relatively poorly known.
A coherent picture of regional climate change, achieved through available regionalisation techniques, will require more coordinated efforts to evaluate the different methodologies, compare methods and models to each other and apply these methods to climate change research in a comprehensive strategy that involves a range of AOGCM and regionalisation experiments.
The author works at the Abdus Salam International Centre for Theoretical Physics, Trieste, ITALY. E-mail: [email protected]
Bengtsson, L., Botzet, M. and Esch, M. Hurricane-type
vortices in a general circulation model. Tellus 47A, 175-196
Cubasch, U., Waszkewitz, J., Hegerl, G., and Perlwitz, J. Regional climate changes as simulated in time-slice experiments. Climatic Change 31, 273-304 (1995).
Deque, M., and Piedelievre, J. P. High resolution climate simulation over Europe. Climate Dynamics 11, 321-339 (1995).
Dickinson, R.E., Errico, R.M., Giorgi, F. and Bates, G.T. A regional climate model for the western United States. Climatic Change 15, 383-422 (1989).
Giorgi, F. On the simulation of regional climate using a limited area model nested in a general circulation model. Journal of Climate 3, 941-963 (1990).
Giorgi, F., and Mearns, L.O. Approaches to regional climate change simulation: A review. Reviews of Geophysics 29, 191-216 (1991).
Giorgi, F., and Mearns, L.O. Introduction to special section: Regional climate modeling revisited. Journal of Geophysical Research 104, 6335-6352 (1999).
Giorgi, F. and Francisco R. Evaluating uncertainties in the prediction of regional climate change. Geophysical Research Letters 27, 1295-1298 (2000).
Giorgi, F., Shields Brodeur, C., and Bates, G.T. Regional climate change scenarios over the United States produced with a nested regional climate model. Journal of Climate 7, 375-399 (1994).
Giorgi, F. et al. Regional Climate Information — Evaluation and Projections in Climate Change 2001; The Scientific Basis, Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change (IPCC), (eds. Houghton, J.T., Ding, Y., Griggs, D.J., Noguer, M., van der Linden, P.J. and Xiaoxu, D.) Chapter 10 (Cambridge University Press, Cambridge, 2001a) 583-638.
Giorgi, F., Whetton, P.W., Jones, R.G., Christensen, J.H., Mearns, L.O., Hewitson, B., von Storch, H., Francisco, R., and Jack, C. Emerging patterns of simulated regional climatic changes for the 21st century due to anthropogenic forcings. Geophysical Research Letters 28, 3317-3320 (2001b).
Hewitson, B.C. and Crane, R.G. Climate downscaling: Techniques and application, Climate Research 7, 85-95 (1996).
Jones, R.G., Murphy, J.M. and Noguer, M. Simulations of climate change over Europe using a nested regional climate model. I: Assessment of control climate, including sensitivity to location of lateral boundaries. Quarterly Journal of the Royal Meteorological Society 121, 1413-1449 (1995).
Jones, R.G., Murphy, J.M., Noguer, M. and Keen, A.B. Simulation of climate change over Europe using a nested regional climate model. II: Comparison of driving and regional model responses to a doubling of carbon dioxide. Quarterly Journal of the Royal Meteorological Society 123, 265-292 (1997).
Kattenberg, G.A.et al. Climate Models — Projections of Future Climate in Climate Change 1995; The Science of Climate Change, Contribution of Working Group I to the Second Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) (eds. Houghton, J.T., Meira Filho, L.G, Callander, B.A., Harris, N., Kattenberg, A. and Maskell, K.) Chapter 6 (Cambridge University Press, Cambridge, 1996) 285-357.
Kittel, T.G.F., Giorgi, F. and Meehl, G.A. Intercomparison of regional biases and doubled CO2 sensitivity of coupled atmosphere-ocean general circulation model experiments. Climate Dynamics 14, 1-15 (1998).
McAvaney, B.J. et al. Model Evaluation in Climate Change 2001; The Scientific Basis, Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change (IPCC), (eds. J.T. Houghton, Y. Ding, D.J. Griggs, M. Noguer, P.J. van der Linden and D. Xiaoxu) Chapter 8 (Cambridge University Press, Cambridge, 2001) 471-523.
McGregor, J.L. Regional climate modeling. Meteorology and Atmospheric Physics 63, 105-117 (1997)
Murphy, J.M. An evaluation of statistical and dynamical techniques for downscaling local climate. Journal of Climate 12, 2256-2284 (1999).
von Storch, H. Inconsistencies at the interface of climate impact studies and global climate research. Meteorologische Zeitschrift 4 NF, 72-80 (1995).
Wilby, R.L., and Wigley, T.M.L. Downscaling general circulation model output: a review of methods and limitations. Progress in Physical Geography 21, 530-548 (1997).
Zorita, E. and von Storch, H. A survey of statistical downscaling techniques. GKSS report 97/E/20 (1997).
The author works at the Abdus Salam International Centre for Theoretical Physics, Trieste, ITALY. E-mail: [email protected]