by Filippo Giorgi*
It is clear by now that the issue of anthropogenic climate change and its impact on human societies and natural ecosystems is among the most important environmental and scientific challenges of this century. The development of suitable policies to adapt to climate change and to stabilize greenhouse gas (GHG) concentrations below the “danger” threshold hinges upon the availability of climate information at scales from regional to country and even local level. Such information, along with the uncertainty associated with it, needs to be clearly communicated to end-users and policy-makers, so that adequate policy decisions to respond to climate change can be taken in a fully informed way.
At scales from regional to global, the climate-change signal is affected by two types of process: changes in large-scale circulations which affect the sequence of weather events that characterize the climate of a region (e.g. the location of storm tracks); and effects of regional to local forcings that modulate the broad-scale climate-change signal (e.g. complex topography, coastlines and land use). In addition, as the spatial scale decreases, climate variability increases, making the identification of anthropogenic signals from the underlying natural variability increasingly difficult. Further, the climate of a given region can be affected by processes occurring far away through teleconnection patterns. As a result of all these factors, projections of regional to local climate changes are extremely difficult and are generally characterized by a high level of uncertainty.
Coupled Atmosphere-Ocean General Circulation Models (AOGCMs) are the primary tools available today to simulate climate change. With the recent development of increasingly powerful computing platforms, the horizontal resolution of AOGCMs has increased. However, most AOGCMs used to produce climate change projections (e.g. for the recently completed CMIP3 ensemble, Meehl et al., 2007) are still run at horizontal resolutions of ~100-300 km, which is too coarse to produce fine-scale climate-change information for use in most impact assessment studies. This resolution also precludes the accurate simulation of extreme weather events, which is fundamental to assess many impacts of climate change. Therefore, since the late 1980s and early 1990s, different “regionalization” techniques have been developed to spatially refine the information produced by AOGCMs and provide data usable for impact assessment studies (Giorgi et al., 2001).
|Figure 1 — Schematic depiction of the Regional Climate Model nesting approach|
Regionalization tools have been increasingly applied to a wide range of climate-change problems, proving to be important resources for climate change research. In applying them, however, it is necessary to fully understand the key assumptions underlying their use along with their potential and limitations. This is especially important in view of the fact that, being computationally and technically more accessible than AOGCMs, regionalization techniques can achieve a widespread use throughout the scientific community and often represent the last step interfacing the climate information with the impact and policy-making application.
Starting from these considerations, this paper first presents an overview of the different regionalization techniques available today, focusing on their underlying assumptions, major recent developments, potential and limitations. This is followed by a description of the application of such techniques to provide climate change information for impact assessment studies. Future prospects and needs to enhance the applicability and reliability of regionalization tools are finally presented.
Regionalization techniques: basic underlying assumptions, recent developments, potential and limitations
Broadly speaking, four regionalization tools are currently available to refine (or downscale) the climate information produced by AOGCMs. They are traditionally referred to as:
- High-resolution “time-slice” Atmosphere General Circulation Models (AGCMs);
- Variable resolution AOGCMs (VarGCMs);
- Nested Regional Climate Models (RCMs);
- Statistical downscaling (SD) methods.
The AGCM approach (e.g. Cubasch et al., 1995) consists of simulating with an atmosphere-only global model some given time periods (or “time slices”) of a transient AOGCM simulation, say one for present-day (e.g. 1960-1990) and one for future (e.g. 2071-2100) climate conditions. The sea-surface temperature (SST) necessary for these simulations is provided by the AOGCM. Because an atmosphere-only model is run for a period of limited duration, the AGCM can attain relatively Latest issues. In fact, recent AGCM time-slice experiments have reached resolutions of a few tens of kilometres, clearly showing an improvement in the model performance with increasing resolution.
The main conceptual assumption underlying the use of time-slice AGCMs is that the SST forcing obtained from the AOGCM is consistent with the climatology of the high-resolution AGCM. Since this may not always be the case, such inconsistencies should be evaluated in the analysis of the results. The main advantage of time-slice AGCMs is their global coverage and their ability to simulate teleconnection patterns across remote regions. On the other hand, AOGCMs are the most expensive regionalization tools and thus need to be run on large computing platforms in order to achieve Latest issue.
The VarGCM approach consists of running the same type of simulations as in the AGCM one, but using a global model with gradually increasing horizontal resolution towards a given region of interest (e.g. Deque and Piedelievre, 1995). Similarly to the case of time-slice AGCMs, the potential inconsistency with the driving SST fields is a problem, although this can be circumvented by relaxing the VarGCM fields towards those of the AOGCM, providing the SST away from the high-resolution area of interest. Another important caveat is that the physical parameterizations used by VarGCMs need to operate at a wide range of spatial scales, which, in some cases, may break the limits of the applicability of the schemes. Today, several VarGCMs are available for climate simulation at refined regional resolutions reaching a few tens of kilometres and a VarGCM intercomparison project has recently been initiated (Fox-Rabinowitz et al., 2006).
The RCM approach consists of running the same type of experiments as in the VarGCM one, but with a limited-area RCM “nested” over the region of interest (Giorgi and Mearns, 1999). Because the model covers only a limited area, it can reach very high horizontal resolutions. In order to be run, the RCM needs meteorological lateral boundary conditions (LBC). In the nesting procedure, these are provided by corresponding AOGCM simulations or, alternatively, by fields from global analyses of observations. Most RCM studies to date have used the one-way nesting method, by which the RCM information does not feed back into the GCM. Recently, however, some two-way nested experiments have been completed with very encouraging results (Lorenz and Jacob, 2005).
RCM nesting is probably the most widely used dynamical downscaling method. The basic underlying assumption of this approach is that the AOGCM simulates the response of the global circulation to large-scale forcings (e.g. GHG radiative forcing) and the nested RCM simulates the effect of sub-GCM scale regional forcings (e.g. topography). It is important to stress that, when used in one-way mode, nested RCMs are not expected to correct large errors in the GCM forcing fields, but mostly to add fine-scale regional information to the large-scale climate signal. It is, therefore, critical to always first analyse the global model fields used for lateral boundary conditions before proceeding to an RCM experiment.
A few tens of RCM systems have been developed at laboratories worldwide and various intercomparison projects are underway for different areas (e.g. Takle et al., 2007). These projects have allowed us to better understand a number of technical issues related to the use of RCMs: choice of domain and physics parameterizations; use of different techniques for assimilating the forcing lateral boundary conditions; effect of internal model variability vs lateral boundary forcing; optimal resolution gap between forcing fields and model solution; and transferability of the models across regions. RCMs have been shown to improve the simulation of climatic spatial detail (e.g. as forced by complex topography and coastlines, see Figure 2) and extreme events compared to the driving global models (Giorgi, 2006). In addition, multi-decadal to centennial RCM simulations have been carried out at grid intervals of a few tens of kilometres or less, showing how fine- scale forcings can substantially affect the climate change signal (e.g. Gao et al., 2006). Numerous research efforts are underway to couple atmospheric RCMs to models of other components of the climate system (e.g. regional ocean/sea ice, chemistry/aerosol and land biosphere models) and a recent new RCM application is their use for seasonal prediction studies (Wang et al., 2004). A number of review papers are available on RCM development and application (Giorgi and Mearns, 1991, 1999; McGregor, 1997; Leung et al., 2003; Wang et al., 2004; Giorgi, 2006).
|Figure 2 — Observed and simulated monsoon precipitation over China (May-September) in an RCM and the driving GCM. Precipitation is obtained as the mean over a 30-year present day simulation and the grid spacing of the RCM is 20 km. Units are mm/day. The figure is an example of improvement attained by the RCM compared to the driving GCM (from Gao et al., 2008).|
In the statistical downscaling technique (e.g. Hewitson and Crane, 1996), the basic strategy is to develop statistical relationships between predictands of interest (e.g. precipitation at a certain location) and predictors that can be obtained from global model simulations (e.g. 500 hPa height). These relationships are constructed using observations and are then applied to the output of AOGCM simulations of future climate to obtain local climate-change information. Although this is the basic philosophy underlying statistical downscaling, many different variants are available (e.g. Giorgi et al., 2001; Christensen et al., 2007).
Statistical downscaling methods are based on the fundamental assumption that the statistical relationships developed using present day climate information are valid also under different climate conditions (Hewitson and Crane, 1996). This assumption is difficult to verify since conditions under increased GHG forcing may be expected to be substantially different from those seen in the historical records used to develop the statistical downscaling models. A second major assumption is that the predictors used in these models are fully representative of the climate change signal, which may require the combined use of a range of predictors based on the specific target of a particular study. Statistical downscaling methods are computationally inexpensive, which allows their ready application to the output of different GCMs. Another advantage of these models is that they can provide local information or information tailored for specific impact applications which may not be available from numerical models. A key issue is the availability of observational datasets of sufficient quality and length to develop robust statistical relationships.
Today, a large number of statistical downscaling approaches exist, including regression-type models, weather generator and weather classification schemes, neural networks, analogue and pattern scaling methods (Giorgi et al., 2001; Christensen et al., 2007). The availability of such a wide range of statistical downscaling methods makes their assessment an especially difficult task, since the models are often tied to specific applications. However, methods have matured to the point that they are today used for an extensive and varied range of impact studies at regional to local scales (Christensen et al., 2007). Wilby et al. (2004) provide a comprehensive discussion of issues pertaining to applications.
Use of regionalization tools to provide climate information needed for impact assessment and adaptation studies.
All regionalization techniques have undergone a tremendous development and an increasing use for a wide range of applications, from process studies to climate change and paleoclimate simulation.
Figure 3 shows the sequence of steps necessary to produce a “regionalized” climate-change scenario for use in impact assessment studies (Giorgi 2005). First, greenhouse-gas emission and concentration scenarios are produced based on assumptions of socio-economic development or target stabilization scenarios. These are then fed into coupled AOGCMs to produce transient climate-change simulations for the 21st century and possibly beyond. The fields from these simulations are fed into regionalization tools, which then produce the information that goes into impact models and, eventually, adaptation planning.
|Figure 3 — Schematic depiction of the steps involved in the production of climate change information usable for impact assessment work via regionalization methods|
Each step of the process is characterized by a certain level of uncertainty that aggregates in a cascade process from one step to the next. The largest sources of uncertainty derive from the use of different greenhouse-gas emission scenarios, different global model configurations and different regionalization methods and models (Giorgi, 2005). As a result of this cascade, the uncertainty related to regional projections is very high and over a limited number of regions only are we today confident in the climate-change projections produced by models (Giorgi et al., 2001; Christensen et al., 2007).
Two critical issues confronting users of regionalization techniques are the added value of using a regionalization tool, and the choice of regionalization method. Concerning the first: for a given application it needs to be carefully evaluated whether the use of a downscaling tool will provide additional useful information. This is clearly the case, for example, in areas of complex topography or in the study of extreme events. It is possible, however, that, for some applications, the AOGCM information can be used directly.
Concerning the second issue, different regionalization techniques have different advantages and limitations and the choice of the approach depends on the specific application and the availability of resources. The limited available intercomparisons across methods indicate that dynamical and statistical downscaling models show comparable performance in reproducing present day climate. The simulated climate change signal can be quite different, however, largely because of the use of specific predictors and assumptions in the statistical downscaling methods.
What an end user can expect from the model-based regionalization methods are time series down to sub-daily time scales of the basic climate variables (e.g. temperature, precipitation, wind) at the respective model resolutions for the entire 21st century or selected periods of it. Currently, the typical resolution obtainable from physical regionalization models is of the order of a few tens of kilometres. If higher resolution is needed, a statistical downscaling tool is required to downscale the climate model information to the local scale and produce time series of relevant climatic variables.
From these considerations, it can be argued that an optimal approach would be the combined use of different downscaling techniques. For example, a time-slice AGCM could be used to provide intermediate resolution information as a linkage from an AOGCM to a VarGCM or RCM, which can then produce finer-scale climate fields. These, in turn, can provide more detailed and internally consistent predictors for statistical downscaling models aimed at providing tailored information for impact-assessment studies.
An example of such an approach (not including the last, SD step) is given in Figure 4 (from Diffenbaugh et al., 2007). This figure presents the change in the occurrence of events of high-danger heat index (a measure of heat stress based on temperature and relative humidity) over the Mediterranean basin for the period 2071-2100, compared to 1961-1990 under the A2 emission scenario of IPCC (2000). It was obtained by first using a time-slice AGCM with SSTs from a coupled AOGCM. The AGCM fields were then used as lateral boundary conditions for a nested RCM simulation at 50-km grid scaling. Fields from this simulation were in turn used to provide lateral boundary conditions for a further nested RCM run at 20 km grid interval. The key aspect of Figure 4 is the ability of this multiple downscaling approach to provide very fine information on heat stress (note, for example, the sharp coastal signal), which can then be used to devise suitable adaptation measures.
|Figure 4 — Simulated increase in the occurrence of days with high-danger heat index with a multiple nested RCM system (see text) (from Diffenbaugh et al., 2007)|
Another example of a downscaling application is shown in Figure 5, in which a statistical downscaling model is used to produce changes in June-July-August precipitation over Africa from different AOGCM outputs. This example, in particular, shows the ability of SD models to be readily applied to the output of different global model simulations.
|Figure 5 — Simulated change in June-July-August precipitation over Africa as obtained from a statistical downscaling (SD) model applied to output from different global model simulations (B. Hewitson, personal communication)|
Prospects and needs for the future use of regionalization techniques in climate change studies
Regionalization tools are today an essential and established aspect of climate change research. Most regionalization techniques can now be implemented on relatively inexpensive computing platforms (e.g. PCs or PC clusters), and this is greatly expanding the user base. On the one hand, this “proliferation” process helps in better understanding and assessing the applicability of the models but, on the other, it requires increased care in their proper application.
What are the most pressing needs in relation to this increased use of regionalization techniques? One is certainly the availability of improved observational datasets for the validation and calibration of the dynamical and statistical models. Present day RCMs can be run at grid scales of 10 km or less, and statistical downscaling models can reach the local scale. In addition, the need is increasing for the application of these models to all regions of the world, including remote and mountainous areas. This will require high-quality, fine-scale observing datasets, possibly of global coverage, to use for dynamical model validation and testing and SD model calibration. In this regard, for example, GCOS can provide a fundamental framework to improve the quality of current observing datasets.
The target horizontal resolution of physical models is rapidly approaching 10 km or less. At this resolution, many current systems need to be upgraded, both in their dynamical and physical components. For instance, many AGCMs, VarGCMs and RCMs use the hydrostatic assumption and employ convection schemes that are based on a clear scale separation between the cloud scale and the model grid scale. Neither of these assumptions holds as the horizontal resolution approaches values of 10 km or less. Boundary layer schemes may also need to be similarly upgraded. The next few years will thus necessarily see marked efforts in model development.
Understanding the sources of uncertainty associated with the projection of regional climate change is an essential piece of information in order to quantify the impacts of climate change (Giorgi, 2005). As mentioned, a key source of uncertainty is the use of different models or regionalization tools. Assessing such uncertainty requires the coordinated completion of ensembles of simulations for different climate-change scenarios, models and methods. This coordination is not easy within a regional research context, because different partners are often interested in different issues and regions. In this regard, WMO can play a central and critical role in facilitating interactions of different modelling communities to build common frameworks for the intercomparison and combined use of models and methods.
Finally, regionalization tools can play the fundamental role of directly involving scientists from developing countries in the climate-change modelling arena (Huntingford and Gash, 2005). These countries are likely to be the most vulnerable to climate change and are, therefore, in great need of adaptation policies. It is thus essential that they build internal know-how concerning the development of regional-to-local climate change information of direct relevance to their specific needs. This will require a strong effort of education and networking in the understanding and use of all modelling tools described here, as well as in the use of model output information available from international projects. This can be achieved through regular programmes of training workshops, exchange visits and south-south/south-north collaborative research projects. Although some such efforts are currently under way (Pal et al., 2007), the role of WMO in this regard will be essential.
Christensen, J.H. et al., 2007: Regional Climate Projections. Chapter 11 of: Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (S. Solomon et al., eds). Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 847-940.
Cubasch, U., et al., 1995: Regional climate changes as simulated by time-slice experiments. Climatic Change, 31, 273-304.
Deque, M. and J.P. Piedelievre, 1995: Latest issue climate simulation over Europe. Climate Dynamics, 11, 321-339.
Diffenbaugh, N.S., J.S. Pal, F. Giorgi and X. Gao, 2007: Heat stress intensification in the Mediterranean climate change Hot-Spot. Geophysical Research Letters, 34, L11706.
Fox-Rabinowitz, M. et al., 2006: Variable resolution general circulation models: stretched grid model intercomparison project (SGMIP). Journal of Geophysical Research, 111, D16104.
Gao, X., J.S. Pal and F. Giorgi, 2006: Projected changes in mean and extreme precipitation over the Mediterranean region from Latest issue double nested RCM simulations. Geophysical Research Letters, 33, L03706.
Gao, X. et al., 2008: Reduction of future monsoon precipitation over China: comparison between a Latest issue RCM simulation and the driving GCM. Meteorology and Atmospheric Physics (in press).
Giorgi, F., 2005: Climate change prediction. Climatic Change, 73, 239-275.
Giorgi, F., 2006: Regional climate modeling: status and perspectives. Journal de Physique, IV, 139, 101-118.
Giorgi, F. and L.O. Mearns, 1991: Approaches to the simulation of regional climate change: a review. Reviews of Geophysics, 29, 191–216.
Giorgi, F., and L.O. Mearns, 1999: Introduction to special section: regional climate modeling revisited. Journal of Geophysical Research, 104, 6335- 6352.
Giorgi, F., et al., 2001: Regional climate information—evaluation and projections. Chapter 10 of: Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change (J.T. Houghton et al., eds), Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 583-638.
Hewitson, B.C. and R.G. Crane, 1996: Climate downscaling: techniques and application. Climate Research, 7, 85-95.
Huntingford, C. and J. Gash, 2005: Climate equity for all. Science, 309, 1789.
Intergovernmental Panel on Climate Change, 2000: Special Report on Emission Scenarios, N. Nakicenovic et al. (eds), Cambridge University Press, New York, 599 pp.
Leung, L.R., L.O. Mearns, F. Giorgi and R.L. Wilby, 2003: Regional climate research: needs and opportunities. Bulletin of the American Meteorological Society, 82, 89-95.
Lorenz, P. and D. Jacob, 2005: Influence of regional scale information on the global circulation: a two-way nesting climate simulation. Geophysical Research Letters, 32, L14826.
McGregor, J.L., 1997: Regional climate modeling. Meteorology and Atmospheric Physics, 63, 105-117.
Meehl, G.A., et al., 2007: The WCRP CMIP3 multimodel ensemble—a new era in climate change research. Bulletin of the American Meteorological Society, 88, 1383-1394.
Pal, J.S. et al., 2007: Regional climate modeling for the developing world: the ICTP RegCM3 and RegCNET. Bulletin of the American Meteorological Society, 88, 1395-1409.
Takle, E.S. et al., 2007: Transferability intercomparison—an opportunity for new insight on the global water cycel and energy budget. Bulletin of the American Meteorological Society, 88, 375-384.
Wang, Y. et al., 2004: Regional climate modeling: progress, challenges and prospects. Journal of the Meteorological Society of Japan, 82, 1599-1628.
Wilby, R.L. et al., 2004: Guidelines for Use of Climate Scenarios Developed from Statistical Downscaling Methods. IPCC Task Group on Data and Scenario Support for Impact and Climate Analysis (TGICA), http://ipcc-ddc.cru.uea.ac.uk/guidelines/StatDown_Guide.pdf.
I would like to thank Rupa Kumar Kolli for his editorial suggestions and E. Coppola for his technical help.
* Abdus Salam International Centre for Theoretical Physics, Trieste, Italy