ABSTRACT:
For many practical planning purposes, the only really useful climate forecasts are those which rule out certain changes as improbable rather than simply ruling them in as possible. And the changes most people are really interested in are not changes in decadal average temperature or rainfall, but changing risks of relatively extreme, and hence infrequent, weather events. Together, these statements raise far-reaching questions about the way we approach climate modelling, both in our priorities for model development and in our use of computing resources.

The need for some kind of estimates of uncertainty in climate forecasts has been recognised for some time. Early approaches (e.g. Allen et al, 2000; Forest et al, 2001; Stott and Kettleborough, 2002) were inspired directly by the “detection and attribution” literature which takes a relatively cautious, data-driven approach to uncertainty analysis (Hasselmann, 1993). More recently, perhaps born out of frustration with the large uncertainty ranges that the data-driven approach seems to generate, there has been a surge of interest in much more subjective approaches that explicitly combine the constraints provided by data with the prior expectations of the climate forecasters (e.g. Murphy et al, 2004). I will argue that this is a very dangerous development and that we abandon a staid reliance on data at our peril. If we take the explicitly subjective approach, then the authors of the IPCC Fifth Assessment will either have to relax the standards of proof they require for attribution statements (and endure the ensuing uproar) or risk appearing more confident about the future than they are about the past.

I will begin by discussing these questions in the context of very simple climate models, and proceed to explore their implications in results from the climateprediction.net slab (Stainforth et al, 2005; Piani et al, 2006) and coupled (Frame et al, in prep) ensembles, the latter also known as the BBC Climate Change Experiment. The focus will be on large-scale global and regional changes in average temperature, which although interesting scientifically have relatively little practical relevance. Hence in the third part of the talk, I will present some preliminary results from the climateprediction.net seasonal attribution project (Pall, 2007), which combines information provided by relatively coarse-resolution global climate models with a very large ensemble of simulations with a seasonal forecast resolution model to quantify the role of increased greenhouse gas levels on the risk of recent flooding events in the UK.

In the final part of the talk, I will talk about some of the implications for climate modelling. If we are to avoid relying on crude scaling arguments, systematically quantifying uncertainty in climate simulation and prediction appears to require ensembles of, potentially, thousands of climate models. Clever design might reduce the required ensemble size, but incorporating more sources of uncertainty would increase it again. We do not design most climate models with hundred-member, never mind thousand-member, ensembles in mind. Improving our “probabilistic resolution” need not necessitate sacrificing spatial resolution if we take a more flexible approach to the design of climate modelling experiments and our use of computing resources.

Biography and suggested readings will appear on the website shortly.

Please let me know if you have any queries- we hope to see you there.

This email has been sent to the WUN earth systems mailing list. If you wish to be removed or replaced on this list please let me know.

Thanks and best wishes,
Elisa

---------------------------------
Dr Elisa Lawson
[log in to unmask]

WUN Development Manager
International Office
University of Southampton
+44 (0)23 80592423 (Ext. 22423)