Ensembles of climate simulations

Broadly speaking, an ensemble is “a collection of comparable datasets that reflect variations within the bounds of one or more sources of uncertainty and that, when averaged, can provide a more robust estimate of underlying behaviour. Ensemble techniques are used by the observational, reanalysis and modelling communities” (IPCC, 2021: Annex VII – Glossary). 

In the context of projections (simulations of the future) from climate models, each member of the same ensemble must have been produced in compliance with a series of specifications (e.g. those of the CMIP6 or CORDEX ensembles) to minimize errors in interpreting the results. The sampling of the different sources of uncertainty (e.g. number of scenarios, number of models) varies depending on the type of ensemble and the scientific questions to be answered, as well as the intended application. Factors to consider include:

• The type of hazard or phenomenon studied

• The time horizon 

• The spatial scale 

• The dominant sources of uncertainty

• The level of confidence sought in the projected changes 

In general, the reliability of the uncertainty estimate increases with the number of members in the ensemble, up to a point. 

For the rest of this page we will describe the most common types of ensembles used in climate science.

Multi-model ensembles

Multi-model ensembles of climate projections are the best known, as they are used to assess the climate change signal and its robustness for a given GHG scenario. Many Earth system models have contributed to the CMIP multi-model ensembles on which much of the IPCC’s work is based. Insofar as the models involved are reasonable representations of the climate system, this type of ensemble aims to assess the uncertainty linked to the climate response to a radiative forcing defined by the GHG scenario. Since the internal structure and representation of the various processes differ from one model to another, we often speak of model uncertainty and the imperfections of models, and sometimes of structural uncertainty. By including a wide range of climate models, we can capture a large part of this source of uncertainty. 

There are also multi-model ensembles made up of simulations from regional climate models (RCMs). The best-known of these were developed as part of the CORDEX program, with much the same objectives as an ensemble using Earth system models. The value of RCMs lies in their ability to better capture certain processes involving finer spatial scales than those properly taken into account by ESMs. Because RCMs need simulations produced by ESMs to provide them with the conditions at the boundaries of the regional grid, the protocols generally ensure that each RCM uses several ESMs and, conversely, that a maximum number of ESMs provide the conditions at the boundaries of several RCMs. This approach makes it possible to separate the RCMs’ contribution to model uncertainty from that of the ESMs. Since producing a regional climate simulation requires greater resources than simulating an ESM covering the whole globe, it should be borne in mind that multi-model ensembles based on RCMs generally contain far fewer members than those based on ESMs.

Perturbed parameter ensembles

Perturbed parameter ensembles provide additional information for estimating model uncertainty, deepening the contribution made by the physical parameterizations of climate models. This type of ensemble uses a single climate model and a single GHG scenario. Each member of the ensemble corresponds to a simulation in which the value of one of the coefficients used in the formulation of one or other of the processes included in the physical parameterizations is modified. Perturbed parameter ensembles generally contain many members so that many processes of importance to the climate system can be varied. 

Although the above-mentioned ensembles are designed to sample model uncertainty, all climate simulations are subject to internal variability, an irreducible source of uncertainty that can never be reduced by model improvements or more precise GHG scenarios. In other words, each member of a multi-model ensemble or a perturbed parameter ensemble contributes to sampling the internal variability, but unfortunately these types of ensemble don’t allow us to isolate the contribution of this source of uncertainty.  

Initial-condition large ensembles

For this reason, initial-condition large ensembles are specifically designed to estimate the amplitude and change in internal variability at various temporal and spatial scales. These ensembles are based on a single ESM and a single GHG scenario. Most contain at least fifty members, for which the climate model is left as is. The different members of this type of ensemble are distinguished by the modifications made to the initial conditions used to start each of the simulations. In this way, they help separate the effect of the non-linear nature of the climate system from the effect of increasing GHGs on changes in climate variables and hazards. Furthermore, these ensembles are particularly useful for studying extremes, as their large number of members enables them to provide larger samples of these events, which are by definition rare in the observations or in other types of ensemble. They are also very useful for studying the major modes of natural climate variability, such as El Niño/La Niña, the North Atlantic Oscillation, and so on. However, it should be noted that the conclusions drawn from initial-condition large ensembles only come from one climate model, so it’s important to check whether other large ensembles produce similar results.

References

GIEC, 2021: Annexe VII – Glossaire [Publié sous la direction de Matthews, J.B.R., V. Möller, R. van Diemen, J.S. Fuglestvedt, V. Masson-Delmotte, C. Méndez, S. Semenov, A. Reisinger]. In Changements climatiques 2021 : Les éléments scientifiques. Contribution du Groupe de travail I au sixième Rapport d’évaluation du Groupe d’experts intergouvernemental sur l’évolution du climat [Publié sous la direction de Masson-Delmotte, V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M.I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J.B.R. Matthews, T.K. Maycock, T. Waterfield, O. Yelekçi, R. Yu et B. Zhou], Cambridge University Press.

Chen, D., M. Rojas, B.H. Samset, K. Cobb, A. Diongue Niang, P. Edwards, S. Emori, S.H. Faria, E. Hawkins, P. Hope, P. Huybrechts, M. Meinshausen, S.K. Mustafa, G.-K. Plattner, and A.-M. Tréguier, 2021: Framing, Context, and Methods. In Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Masson-Delmotte, V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M.I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J.B.R. Matthews, T.K. Maycock, T. Waterfield, O. Yelekçi, R. Yu, and B. Zhou (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 147–286, doi:10.1017/9781009157896.003.

Lee, J.-Y., J. Marotzke, G. Bala, L. Cao, S. Corti, J.P. Dunne, F. Engelbrecht, E. Fischer, J.C. Fyfe, C. Jones, A. Maycock, J. Mutemi, O. Ndiaye, S. Panickal, and T. Zhou, 2021: Future Global Climate: Scenario-Based Projections and Near-Term Information. In Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the
Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Masson-Delmotte, V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M.I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J.B.R. Matthews, T.K. Maycock, T. Waterfield, O. Yelekçi, R. Yu, and B. Zhou (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 553–672, doi:10.1017/9781009157896.006.