Key points are not available for this paper at this time.
How can we best apply our science to predicting risks of extreme behaviour in a system as complex as the climate? It would be desirable to be able to represent all of our knowledge about the risks so that it can be applied to enable effective decision-making. Risk assessments often consider only the range of behaviour displayed by climate models, but a substantial part of the risk seems likely to be due to the possibility of the real world veering outside this range. It will be illustrated how implicitly ignoring this component would lead to risks being systematically underestimated, and how multi-model and initial condition large ensembles can be misleading. Recent work on storyline methods has illustrated potential ways to think beyond numerical model simulations, but downplays the quantification of event risks. But since we generally lack clear bounds on how intense extreme events can be, this seems to leave open the question of just how intense should the events be that are considered in analyses. It also does not seem to satisfy decision analyses that seek to quantitatively trade off protection against extremes against other benefits. This presentation considers how we can go beyond counting events in simulations, using tools such as climate models to inform our future projections without being constrained to ignore possible outcomes that they cannot simulate, whilst also retaining as much quantitative knowledge about event risks as possible and acknowledging when ambiguities become very large. Frameworks from philosophy and decision analysis will be surveyed and it will be discussed how these may help to show a way forward in our climate prediction predicament. It will be suggested that climate science should aim to be pluralistic in the knowledge frameworks it considers, to be of use to the broadest possible range of decision making.
P.A. Watson (Fri,) studied this question.