Working Group I: The Scientific Basis

Other reports in this collection

8.10 Sources of Uncertainty and Levels of Confidence in Coupled Models 8.10.1 Uncertainties in Evaluating Coupled Models

Our attempts to evaluate coupled models have been limited by the lack of a more comprehensive and systematic approach to the collection and analysis of model output from well co-ordinated and well designed experiments. Important gaps still remain in our ability to evaluate the natural variability of models over the last several centuries. There are gaps in the specification of the radiative forcing (especially the vertical profile) as well as gaps in proxy palaeo-data necessary for the production of long time series of important variables such as surface air temperature and precipitation.

In order to assist future coupled model evaluation exercises, we would strongly encourage substantially expanded international programmes of systematic evaluation and intercomparison of coupled models under standardised experimental conditions. Such programmes should include a much more comprehensive and systematic system of model analysis and diagnosis, and a Monte Carlo approach to model uncertainties associated with parametrizations and initial conditions. The computing power now available to most major modelling centres is such that an ambitious programme that explores the differing direct responses of parametrizations (as well as some indirect effects) is now quite feasible.

Further systematic and co-ordinated intercomparison of the impact of physical parametrizations both on the ability to simulate the present climate (and its variability) and on the transient climate response (and its variability) is urgently needed.

The systematic analysis of extremes in coupled models remains considerably underdeveloped. Use of systematic analysis techniques would greatly assist future assessments.

It is important that in future model intercomparison projects the experimental design and data management takes heed of the detailed requirements of diagnosticians and the impacts community to ensure the widest possible participation in analysing the performance of coupled models.

8.10.2 Levels of Confidence

We have chosen to use the following process in assigning confidence to our assessment statements; the level of confidence we place in a particular finding reflects both the degree of consensus amongst modellers and the quantity of evidence that is available to support the finding. We prefer to use a qualitative three-level classification system following a proposal by Moss and Schneider (1999), where a finding can be considered:

“well established” - nearly all models behave the same way; observations are consistent with nearly all models; systematic experiments conducted with many models support the finding;
“evolving” - some models support the finding; different models account for different aspects of the observations; different aspects of key processes can be invoked to support the finding; limited experiments with some models support the finding; parametrizations supporting the finding are incompletely tested;
“speculative” - conceptually plausible idea that has only been tried in one model or has very large uncertainties associated with it.

8.10.3 Assessment

In this chapter, we have evaluated a number of climate models of the types used in Chapter 9. The information we have collected gives an indication of the capability of coupled models in general and some details of how individual coupled models have performed.

We regard the following as “well established”:

  • Incremental improvements in the performance of coupled models have occurred since the SAR, resulting from advances in the modelling of the oceans, atmosphere and land surface, as well as improvements in the coupling of these components.
  • Coupled models can provide credible simulations of both the annual mean climate and the climatological seasonal cycle over broad continental scales for most variables of interest for climate change. Clouds and humidity remain sources of significant uncertainty but there have been incremental improvements in simulations of these quantities.
  • Some non-flux adjusted models are now able to maintain stable climatologies of comparable quality to flux adjusted models.
  • There is no systematic difference between flux adjusted and non-flux adjusted models in the simulation of internal climate variability. This supports the use of both types of model in detection and attribution of climate change.
  • Several coupled models are able to reproduce the major trend in surface air temperature, when driven by radiative forcing scenarios corresponding to the 20th century. However, in these studies only idealised scenarios of only sulphate radiative forcing have been used.
  • Many atmospheric models are able to simulate an increase of the African summer monsoon in response to insolation forcing for the Holocene but they all underestimate this increase if vegetation feedbacks are ignored.

We regard the following as “evolving”:

  • Coupled model simulation of phenomena such as monsoons and the NAO has improved since the SAR.
  • Analysis of, and confidence in, extreme events simulated within climate models is emerging, particularly for storm tracks and storm frequency.
  • The performance of coupled models in simulating ENSO has improved; however, the region of maximum SST variability is displaced westward and its strength is generally underestimated. When suitably initialised, some coupled models have had a degree of success in predicting ENSO events.
  • Models tend to underestimate natural climate variability derived from proxy data over the last few centuries. This may be due to missing forcings, but this needs to be explored more systematically, with a wider range of more recent models.
  • A reasonable simulation of a limited set of past climate states (over the past 20,000 years) has been achieved using a range of climate models, enhancing our confidence in using models to simulate climates different from the present day.
  • Our ability to increase confidence in the simulation of land surface quantities in coupled models is limited by the need for significant advances in the simulation of snow, liquid and frozen soil moisture (and their associated water and energy fluxes).
  • Coupled model simulations of the palaeo-monsoons produce better agreement with proxy palaeo-data when vegetation feedbacks are taken into account; this suggests that vegetation changes, both natural and anthropogenic, may need to be incorporated into coupled models used for climate projections.
  • Models have some skill in simulating ocean ventilation rates, which are important in transient ocean heat uptake. However these processes are sensitive to choice of ocean mixing parametrizations.
  • Some coupled models now include improved sea-ice components, but they do not yield systematic improvements in the sea-ice distributions. This may reflect the impact of errors in the simulated near surface wind fields, which offsets any improvement due to including sea-ice motion.
  • Some coupled models produce good simulations of the large-scale heat transport in the coupled atmosphere-ocean system. This appears to be an important factor in achieving good model climatology without flux adjustment.
  • The relative importance of increased resolution in coupled models remains to be evaluated systematically but many models show benefits from increased resolution.
  • Our ability to make firmer statements regarding the minimum resolution (both horizontal and vertical) required in the components of coupled models is limited by the lack of systematic modelling studies.

We regard the following as “speculative”:

  • Tropical vortices with some of the characteristics of “tropical cyclones” may be simulated in high resolution atmospheric models but not yet in coupled climate models. Considerable debate remains over their detailed interpretation and behaviour.
  • Some modelling studies suggest that adding forcings such as solar variability and volcanic aerosols to greenhouse gases and the direct sulphate aerosol effect improves the simulation of climate variability of the 20th century.
  • Emerging modelling studies that add the indirect effect of aerosols and of ozone changes to greenhouse gases and the direct sulphate aerosol effect suggest that the direct aerosol effect may previously have been overestimated.
  • Lack of knowledge of the vertical distribution of radiative forcing (especially aerosol and ozone) is contributing to the discrepancies between models and observations of the surface-troposphere temperature record.

Our overall assessment
Coupled models have evolved and improved significantly since the SAR. In general, they provide credible simulations of climate, at least down to sub-continental scales and over temporal scales from seasonal to decadal. The varying sets of strengths and weaknesses that models display lead us to conclude that no single model can be considered “best” and it is important to utilise results from a range of coupled models. We consider coupled models, as a class, to be suitable tools to provide useful projections of future climates.

Other reports in this collection

IPCC Homepage