Disentangling the meaning of statements made using models about the real world can be difficult. Compare the following three key findings from the Summary for Policymakers (PDF link) of the most recent report from the Intergovernmental Panel on Climate Change. All of these key findings from the same short document refer to conclusions informed by output from the same models, and they are all presented as real-world statements which are relevant for policymakers. Yet they take quite different approaches to reporting model output.
1. Temperature projections give a 5-95% model range and downgrade the uncertainty assessment “after accounting for additional uncertainties or different levels of confidence in models”. This is a meaningful statement about the authors’ expectations of a real future climate (given the hypothetical changes in forcing):
Increase of global mean surface temperatures for 2081-2100 relative to 1986-2005 is projected to likely be in the ranges derived from the concentration-driven CMIP5 model simulations, that is, 0.3C to 1.7C (RCP2.6), 1.1C to 2.6C (RCP4.5), 1.4C to 3.1C (RCP6.0), 2.6C to 4.8C (RCP8.5). SPM E.1, page 20
Here, model output has been used to inform expert opinion, with expert judgement used to quantify the plausible chance of non-modelled outcomes occurring.
2. AMOC projections give a model spread and best estimate, attaching no likelihood to the numerical values reported from models but a very likely assessment of the sign of the change (not conditional on forcing). The multi-model mean is explicitly identified as a best estimate:
It is very likely that the Atlantic Meridional Overturning Circulation (AMOC) will weaken over the 21st century. Best estimates and ranges [footnote: The ranges in this paragraph indicate a CMIP5 model spread.] for the reduction are 11% (1 to 24%) in RCP2.6 and 34% (12 to 54%) in RCP8.5.” SPM E.4, page 24
Here, model output has been presented as a proxy for expert opinion. The unstated assumption is that the model average is the best estimate. It would be helpful for the authors to clarify whether the direct use of model output as a best estimate is because it is genuinely thought that the models capture all relevant processes, or because there is no extant information on which to base further expert judgement about the inadequacies of the models.
3. Arctic sea ice projections give a range from the lowest to highest RCP scenarios, reporting only the model outcomes but then attaching a medium confidence label – when surely, this should be extremely high confidence that the models give these reductions. There is an unstated implication that we have medium confidence of the real-world outcome lying between these bounds:
Year-round reductions in Arctic sea ice extent are projected by the end of the 21st century from multi-model averages. These reductions range from 43% for RCP2.6 to 94% for RCP8.5 in September and from 8% for RCP2.6 to 34% for RCP8.5 in February (medium confidence) SPM E.5, page 24
Here, model output has been presented as a proxy for expert opinion, but in addition a confidence assessment of only “medium” has been used to weaken the statement without giving any information, either qualitative or quantitative, about the degree to which it is weakened. There is no sense in which anyone could use this statement to generate a real-world expectation beyond the sign of the change.
These three ad-hoc methods of dealing with the difficult step of translating model-world values into something fit for a “Summary for Policy Makers” demonstrate the need for a more transparent guidance on the use and interpretation of model output. In my view the first is reasonably well-considered and documented, the second is disingenuous and requires explicit statement and clarification of the underlying assumption, and the third is positively misleading.
Read more
The Lorentz Principles for the Communication of Climate Information
Escape from Model Land
London Mathematical Laboratory
Or join the discussion on Twitter: https://twitter.com/H4wkm0th