Towards Measures to Establish the Relevance of Climate Model Output for Decision Support

How exactly can decision-support and policy making benefit from the use of multiple climate model experiments in terms of coping with the uncertainties on climate change projections? Climate modelling faces challenges beyond those of weather forecasting or even seasonal forecasting, as with climate we are now (and will probably always be) required to extrapolate to regimes in which we have no relevant forecast-verification archive. This suggests a very different approach from traditional methods of mixing models and skill based weighting to gain profitable probabilistic information when a large forecast-verification archive is in hand. In the case of climate, it may prove more rational to search for agreement between our models (in distribution), the aim being to determine the space and timescales on which, given our current understanding, the details of the simulation models are unimportant. This suggestion and others from Smith (2002, Proc. National Acad. Sci. USA 4 (99): 2487-2492) are interpreted in the light of recent advances. Climate models are large nonlinear dynamical systems which insightfully but imperfectly reflect the evolving weather patterns of the Earth. Their use in policy making and decision support assumes both that they contain sufficient information regarding reality to inform the decision, and that this information can be effectively communicated to the decision makers. There is nothing unique about climate modeling and these constraints, they apply in all cases where scientific modeling is applied to real-word actions (other than, perhaps, the action of improving our models). Starting with the issue of communication, figures from the 2007 IPCC Summary for Policy Makers will be constructively criticized from the perspective of decision makers, specifically those of the energy sector and the insurance/reinsurance sector. More information on basic questions of reliability and robustness would be of significant value when determining how heavily to weight climate model output in the decision process; one obvious example is the question of over what spatial and time averages modelers expect information in current climate distributions to be robust. The IPCC itself suggests continental/seasonal, while distributions over 10's of kilometers/hourly is on offer. Our aim here is not to resolve this discrepancy, but to develop methods with which it can be addressed. This is illustrated in the context of using another physically based, imperfect model setting: using Newton's laws in an actual case of NASA hazard evaluation. Our aim is to develop transparent standards of good practice managing expectations, which will allow model improvements over the next decades to be seen as progress by the users of climate science.