Uncertain future of climate uncertainty

by Judith Curry

How believable are the IPCC’s continually increasing confidence levels?

Disconcerting learning on climate sensitivity and the uncertain future of uncertainty

Alexis Hannart, Michael Ghil, Jean-Louis Dufrwsne, Philippe Naveau

Abstract How will our estimates of climate uncertainty evolve in the coming years, as new learning is acquired and climate research makes further progress? As a tentative contribution to this question, we argue here that the future path of climate uncertainty may itself be quite uncertain, and that our uncertainty is actually prone to increase even though we learn more about the climate system. We term disconcerting learning this somewhat counter-intuitive process in which improved knowledge generates higher uncertainty. After recalling some definitions, this concept is connected with the related concept of negative learning. We illustrate disconcerting learning on several real-life examples and characterize mathematically certain general conditions for its occurrence. We show next that these conditions are met in the current state of our knowledge on climate sensitivity, and illustrate this situation based on an energy balance model of climate. We finally discuss the implications of these results on the development of adaptation and mitigation policy.

Published in Climatic Change, [link] to abstract.

The paper’s Discussion and Conclusions gives a good overview of their argument, excerpts:

Oppenheimer et al. (2008) introduced a probabilistic definition of learning in the context of scientific research on environmental problems. These authors showed that learning does not necessarily lead to truer beliefs, a situation they termed negative learning. We have extended this analysis here to show that learning does not necessarily lead to more certain beliefs either, a situation for which we introduced the term of disconcerting learning. Negative learning corresponds to an increase in PDF bias, disconcerting learning corresponds to an increase in PDF dispersion. We have shown that the latter differs from, and is not tied to, the occurrence of the former. In other words, learning may well result in a state of knowledge which is closer to the truth and yet more uncertain, cf. Fig. 1.

We have shown that this rather counter-intuitive situation typically arises when a surprising but inconclusive piece of evidence is found. We introduced  a probabilistic model based on reasonable assumptions about learning, and used it to confirm that disconcerting learning in general occurs as a result of surprising but inconclusive evidence at a particular step in the learning process. Furthermore, we narrowed in on this situation arising when the PDF that reflects the state of knowledge is asymmetric or has heavy tails. We have shown that the dispersion of the trajectories of uncertainty as learning occurs—i.e. the uncertainty on the uncertainty—is high when disconcerting learning is prone to happen.

Finally, because pronounced asymmetry appears to be a pervasive feature of the PDF of climate sensitivity in our current state of knowledge [AR4], climate uncertainty is thus prone to remain high or to increase—even if and as climate science makes steady progress—and thus its future trajectory is itself highly uncertain. 

At first, the news that substantial research efforts dedicated to improving our understanding of the climate system could potentially result in an increased uncertainty on the outcome of future climate change may sound rather discouraging. On the other hand, the present article also provides a rational justification for the fact that  constant or even increasing uncertainty is perfectly compatible with steady scientific progress and improved knowledge of the climate system. In other words, our results suggest that the uncertainty on climate sensitivity should not be considered as an appropriate metric to monitor progress in climate science, as has sometimes been suggested.

Indeed, the initial increase of uncertainty is caused by the inconclusive nature of the surprising evidence. As more reassuring evidence confirms what was at first a surprise, uncertainty will eventually decrease. 

JC comments:

The pause in global surface temperature anomalies for the past 15+ years is arguably an example of negative disconcerting learning.  Well the IPCC doesn’t seem to have learned anything yet from this and are not yet disconcerted since they expect the warming to resume imminently.  However outside the context of the IPCC, negative disconcerting learning is taking place, as scientists put forward alternative hypotheses for testing.  So the pause is arguably a good thing for climate science.

However, the IPCC feels compelled to continually increase confidence levels (now to 95% on the 20th century attribution), despite the reality of the situation whereby negative disconcerting learning is taking place and previous assumptions are being challenged.  Acknowledgement of negative learning, even disconcerting learning, seems important to avoid this senseless march towards increasing confidence as the model projections and observations diverge.

The paper also has a section on policy implications of negative and disconcerting learning, in context of optimal decision making.  In light of the perspective presented by this paper, Roger Pielke has a tweet today that sums it up perfectly:

An alternative to calling for accepting science as starting point would be to focus on policies robust to scientific debates.