All those exercises refer to the AMIP intercomparison of WGNE, which plays an important role for the validation of the atmospheric part of coupled models participating in CLIVAR.

Modelling and the specific CLIVAR objectives

In essence, the CLIVAR objectives are to understand and to predict - to the extent possible - climate variability and climate change, including human influences.

Predicting . . .

Prediction of anthropogenic climate change was discussed in chapter 6. An accepted strategy is to use the most sophisticated models that can be run, given the available computer resources. This requires a careful spin-up followed by a long (typically a hundred year) forward integration. The forcing deserves particular attention. Several such simulations are required to allow the evaluation of different stabilisation scenarios. The results of these predictions are used in impact studies. It is important to assess the reliability of climate change predictions. Information on this can be obtained by comparing predictions from different models. Confidence in model predictions is derived from their ability to describe present and past climate. This explains the need for the simulation of past climate and for a detailed comparison of model results with observations.

Another important topic is the predictability of natural climate variations on decadal to centennial time scales. Studies of natural variability in long integrations suggest that global mean climate is more predictable than regional climate. The (political) desire for reliable climate predictions at regional scales challenges scientists to provide a meaningful answer.

Several centres make operational climate forecasts on seasonal time scales with coupled atmosphere/ocean models. The real-time availability of observations and their assimilation into model fields is essential for initialisation. The predictions are validated against observations. The quality of the models can also be assessed by studying their ability to simulate the mean climate. Predictability is also an important aspect here. There are indications that the use of a combination of different models may result in better probabilistic forecasts than those that can be inferred from a single model.

. . . and Understanding

Understanding can be achieved at different levels. A first step towards understanding is rather descriptive: How can one best characterise climate variability? What are the dominant modes? Models can help answer these questions. A second step can be made by studying sensitivities. In this approach one varies details of the models and one then studies the effect in terms of the climate response. Numerous examples have been given in the preceding sections, for example: the effect of cloud parameterisation on the climate sensitivity, the influence of SST on the atmospheric circulation and the identification of key land regions affecting monsoon variability. In a third step one tries to identify the cause and effect chains (which may occur as feedback loops) that cause climate variations. Typical questions are: What caused the 1997 El Niño? What is the cause of the spectacular increase of the NAO index between 1960 and 1990? What causes the observed global warming? What caused the drought in the Sahel in the 1960s? What caused the global warming in the 1920s? What is the probability for rapid change in the global thermohaline circulation? What caused the desertification of Mesopotamia and northern Africa in historic times? Why is the Holocene climate so stable? What caused the rapid climate variations during the last glacial? These questions are pertinent to an embarrassingly large range of space and time-scales. Therefore, there is a general need to numerically simulate the fully coupled climate system as well as its various subsystems, with complex and less complex models.

B7.2.2 Modelling Strategy

In practice, these points are not as unrelated as they seem, because the application of existing models often provides useful information concerning possible improvements. Nevertheless, a strategy should address each of these issues separately.

The use of existing models

Models are widely applied in pursuit of the CLIVAR objectives. This implicitly shows that present models are felt to be sufficiently useful for this purpose. One may wonder then whether one should perhaps concentrate on one of these models. The answer is no, because a comparison of results obtained with different models is essential for obtaining information about the reliability of model simulations and the sensitivity for particular parameterisations. Another need for model diversity is the existence of different research questions, which can best be addressed with different models of different complexity. Here follows a short overview of the different types of model, and their use.

Complex coupled GCMs

Very complex climate models, such as a global coupled atmosphere/ocean/sea-ice General Circulation Models (GCMs) are used for global change prediction and for the study of natural variability. Nested regional versions are used for regional studies. The use of complex coupled GCMs is limited by their enormous computer requirements.

Energy balance / Upwelling diffusion models

"Energy Balance / Upwelling Diffusion" models are the other extreme. These highly parameterised models are used for the study of climate change at multicentennial time scales. They are box models that give a very simple description of the climate system.

Intermediate complexity models

Intermediate complexity models are also needed. One reason to use these is that the time scale over which the thermohaline ocean circulation adjusts into a thermodynamic equilibrium is several thousand years, which presents a major challenge in understanding its parameter dependencies, in predicting its future fate and for establishing the equilibrium states of coupled models. Even relatively short-term simulation experiments over some decades may depend strongly on the thermodynamic balance existing in the initial state (e.g. the proximity to critical stability thresholds) and this can only be investigated by long integrations. Other reasons for using intermediate complexity models are the need for a large number of experiments, and the need to investigate the carbon cycle and feedbacks with the biosphere etc

Component models (atmosphere, ocean, sea-ice, . .) and regional models

In addition to the component models used in coupled model simulation (atmospheric GCMs, oceanic models, and ice models) one needs separate stand-alone special-purpose versions, to investigate processes in these subsystems, and to test new parameterisations. These models are needed both in global and in regional implementations. Regional models are very useful tools for testing physical parameterisations.

Both the existence of different research objectives and the need to estimate model uncertainty from model intercomparisons strongly support the idea of a multi-model approach. This very approach is at the basis of the European strength.

Model improvement

Despite the usefulness of existing models, there is also an urgent need for improvement, because all of these models suffer from severe errors, both in their simulation of the mean climate and in the simulation of its variability. These errors seriously affect the reliability of the assessment of climate change due to anthropogenic causes and the full deployment of prediction systems at shorter time scales. They also limit our ability to study the causes of climate variability. There is no widely agreed approach on how to reduce uncertainty in modelling climate variability and climate change. Therefore, the strategy for the next decade is a complex one, with developments in several directions.

Complex coupled GCMs

Here a mixed strategy applies. Better numerics and higher resolution may bring some benefit, but much of the uncertainty lies in the problem of representing sub grid scale processes. Significant improvements are necessary in the physics of each component of the climate system. In particular improved parameterisations are needed in the atmosphere, the ocean, the land-surface and the ice models. New processes missing in present climate models, need to be included

(geochemistry, ecosystem, biology).

Intermediate complexity coupled models

What is particularly required is models which adequately resolve the oceanic circulation with its various time scales crucial for DecCen variability, but which can still be integrated easily for thousands of years. A crucial link in the model hierarchy will therefore have to be ocean general circulation models coupled to highly efficient atmospheric models, which are much faster than atmospheric GCMs but more capable than energy balance models (which fail to capture the hydrological cycle properly, among other problems). The lack of such models is one of the main obstacles to progress in understanding many aspects of the ocean's role in climate.

Regional climate models

Recent studies have shown that high-resolution or limited-area regional climate models have often the same systematic errors as their larger scale counterparts. Regional climate modelling thus deserves specific attention. It will be useful to have a diversity of very-high-resolution climate models to address scientific issues linked to climate change over Europe. Nesting procedures have to be carefully tested and validated.

Atmosphere-only models

Atmospheric models need to be improved, for example with better cloud parameterisations, better convective parameterisations and better land-surface schemes.

Ocean-only models

Ocean modelling poses a twofold challenge. On the one hand, in order to better understand the dynamics of circulation changes, we need models of higher resolution, which can resolve crucial small-scale processes known to influence the large-scale circulation. Examples of such processes which are particularly important for decadal changes are narrow boundary currents, overflows and flows through straits, watermass formation and ocean eddies. On the other hand, as high-resolution models are generally too time-consuming to be used in climate simulations, coarse-resolution models are also required which need however substantially improved parameterisations for all unresolved processes, in order to give the correct large-scale transports.

Recommendations concerning specific modelling objectives

Predictions

• Experimental operational forecasting on seasonal to interannual time scales in a (carefully-selected) multi-model ensemble approach should continue with existing models and with models becoming available in the next few years. This will require improved (re-)analysis of the upper ocean.

Past climate

Past climates need to be modelled both to improve our understanding of the mechanisms of natural variability and to evaluate climate models against available past data. We recommend:

• to focus the evaluation of models on key periods : the last hundreds to thousand years, the mid-Holocene and the last glacial maximum, avoiding the use of special-purpose models and investigating the use of coupled models, especially for the Holocene. The periods of special interest have been discussed in more detail in chapter 7 in relation to the availability of observational data.

 

Study mechanisms

• The study of mechanisms requires additional dedicated studies (in addition to the simulations of past climate) with a variety of coupled and component models of different complexity. This should lead to a further understanding of feedback mechanisms including mechanism that might lead to abrupt change.

Model improvement and model development

For future application models should be improved. This involves:

• A hierarchy of models from one-dimensional column models to mesoscale oceanic and atmospheric models have to be used, in connection with field experiments, to develop new parameterisations for (O)GCMs.

In the long term the simulation of the climate system will have to be completed by introducing several processes currently not represented, but of great relevance to the assessment of climate change effects. They include a complete geochemistry, and the modelling and simulation of interactive marine and terrestrial ecosystems. It will be important to devise a timetable for model development and application in Europe. Some of the improvements currently under way will find their application in prediction systems within the next few years. In contrast it may take a decade to develop a full bio-geochemical model. Even so, one should work now already on the development of such a complex model.

Model intercomparisons

Model intercomparisons have been and will continue to be useful, provided they are carefully selected and aim at well-defined scientific goals. They can contribute to our understanding of sensitivities and mechanisms, they allow for better selection of ensembles and they may suggest ways of improving the models

Although there is a need to add further processes to coupled climate models (e.g. biology, chemistry), the value of this is severely limited if the underlying control climate is poor, and while there remains the threefold uncertainty in climate sensitivity due to poor representation of physical processes. Therefore, we believe that the support for new high profile projects should not be at the expense of more fundamental work on reducing uncertainty. Model intercomparisons provide one of the best opportunities for increasing understanding, provided they are carefully designed and have well-defined scientific goals. They are most likely to be successful if the number of models is limited, and sufficient resources are available to enable an in-depth analysis and supporting sensitivity studies.

B7.2.3 Integration of the European effort

Europe has a strong position in climate modelling. The framework which has enabled this is first of all a strong and long term commitment from national agencies to support modelling centres. In addition, the EU has played an important role in enhancing multinational collaboration. Central EU support will also be essential in the future. Without this Europe would be in danger of losing its lead. In view of the scientific need for a multi-model approach it cannot be a sensible objective to go for one single Euromodel. However, there are strong arguments in favour of a more integrated approach to climate modelling in Europe. Our suggested strategy is a threefold one, namely

• to optimise the existing human potential through collaborations between different groups;

European modelling collaboration

(Multi)national collaboration is already well-developed in Europe. In view of the complexity of the climate problem these collaborations should be further stimulated. We make a number of recommendations, which would allow an efficient approach towards the modelling objectives spelled out in section 8.2. For their implementation flexible and effective mechanisms should be sought. These mechanisms should increase the collaboration, also outside joint projects, and they should help optimise the embedding of a European CLIVAR contribution within the international CLIVAR activities.

Model development consortia

Model development must build on existing national efforts. The reason is simple: the development of a model typically takes four or five years from initiation to useful experimentation, and considerable human and computing resources.

There is also a considerable maintenance cost, testing that the model still gives the same results as operating and compiling systems are updated, and one has to tailor input and output routines for the particular system the model is being run on. Therefore, the development of climate models requires substantial funding over an extended period of time. Unless the EU is able to offer a substantial long-term commitment to supporting modelling, it has to build on what already exists. This does however not mean that uncoordinated development of different models at different places is necessarily the best way.

Perhaps the most effective organisation for modellers is to form consortia that group around a handful of community models. The development of coupled climate models and sophisticated component models requires a high level of specialisation. One needs people working on different parameterisations, such as clouds, radiation, land surface processes, orography, etc. Others, the software engineers, need to put the code together. Still others have to provide appropriate forcings, collect data for validation, perform the model runs, make model results available and analyse the results. Model development has to be done with utmost care in view of the potential economic impact of model predictions. Therefore, one should consider assigning the different modelling tasks to groups of people with shared responsibility. It has been suggested that it might be useful in this context to investigate the possibility of further specialisation of different institutions. Europe does have an enormous potential.

Model comparisons

Focused comparisons between different (component) models have been strongly recommended. This will provide an excellent opportunity for European groups to collaborate. Europe is in a unique position to lead the way in this respect as it has three or four differently constructed generic climate models which are state-of-the-art but produce diverse predictions of climate change. We can make substantial progress in removing what has proved one of the more stubborn sources of uncertainty in model predictions.

Model applications

Priorities for different modelling applications have been listed in the preceding section. Part of these activities will be carried out by national groups. Other activities will require multinational collaborations. It has been suggested that these collaborations focus on a small number of projects which would keep Europe at the forefront of climate science, without requiring massive model development. Specifically an extended climate simulation with a credible high-resolution ocean model was suggested. But alternatives have also been suggested, such as simulations with more complex models, i.e. with better chemistry and biology or a long Holocene hindcast with a suitable coupled model. The advantage of selecting one of these possibilities as very advanced application over the other should be carefully argued.

A European climate computing facility

Despite its spectacular growth, available computer capacity continues to limit the advance of climate research. The following activities will require enormous resources

• Long simulations with complex models at high resolution,

A serious European contribution to CLIVAR, will require an estimated overall enhancement in computational resources by a factor of 10-50 over the next few years. This means that the requirements will increase to the order of a few hundred Gflops/s sustained overall performance, growing again by a factor of 5-10 over the next five years to Teraflop levels. Data handling resources must also be enhanced accordingly in order to match the computational facilities. The European climate modelling community needs this capacity, if it wants to contribute to the CLIVAR numerical experimentation programme in a convincing way, and to give sound guidance for the CLIVAR observational programmes. A lack of computing capacity would result in a serious delay of CLIVAR.

European climate modelling should make full use of capacity available at national centres and at ECMWF. Additional supercomputing facilities should be provided through the establishment of a European climate computing facility. Its primary tasks would be the acquisition of computer facilities exceeding the national possibilities. It should provide computer time as well as access software. The facility could be hosted by an existing national or international centre or by a network of such centres. This facility should also help facilitate the access to the models and computing centres, though training programmes.

The exchange of software and model output

Scientific studies which lead to an understanding of why different models give different climate responses should be strongly supported. There are several ways in which these studies could be further facilitated, though none of these ways are an end in themselves. First the exchange of software (component models, models, software) and model results should be encouraged. This would involve a better documentation of models, a better standardisation of model output to facilitate uniform diagnosis of mechanisms across different models, better accessibility to models and model output, and further development of standardised couplers, (i.e. interfaces based on physical principles). Unless such an approach is supported, progress in reducing uncertainty and improvement of regional predictions will stagnate.

There have been discussions about the possible development of a common model kernel, which would allow component models to be used in combination and should permit easy exchange of physics packages developed by national groups. This would be a very demanding job, and require considerable resources. In practice, more than one kernel might be needed to maintain the full diversity of dynamical schemes in both the ocean and atmosphere. There is doubt therefore whether such an enormous software project could be successful if only limited resources are available. Further discussion of this issue is recommended.

Impact studies of climate change imply the use of downscaling methods applied to the results of global or regional scale climate models. The actors of such studies come from a very wide spectrum of research fields with a very wide range of requirements. The access of the impact community to climate model output should thus be facilitated. This implies a support of actions aiming at developing a network of a few specialised data centres having the responsibility of the distribution of data on climate scenarios provided by climate modelling groups.

To overview these developments a European Climate Modelling Group should be established with representatives from all interested research centres. This group must build on (and should not duplicate) the activities of the international (CLIVAR) Numerical Experimentation Groups. It is desirable for such a group to have some responsibility with respect to the allocation of resources. This group should also foster the exchange of information about model developments and results between different centres. Obviously, there would be a need to set up a number of working groups on such subjects as standardisation and documentation, and also for the different types of model.

Summary of recommendations concerning integration of the European effort

Summarising, we have made specific recommendations for a better integration of the European modelling effort with respect to human potential, hardware and software. As part of each of these three recommendations a mechanism for its implementation is also recommended. When establishing such mechanisms attention should be paid to interrelationships and the possibility of giving combined tasks and resources to one body. The recommendations are the following:

• Collaboration between the climate modelling centres and other European climate research groups should be increased. Flexible and effective mechanisms should be sought to stimulate sharing of expertise and human potential, to increase collaboration also outside joint projects, and to optimise the embedding of a European CLIVAR contribution within the international CLIVAR activities. The formation of model development consortia around a number of community models should be encouraged. Studies which lead to an understanding of why different models give different climate responses should be strongly supported.