Comment on article by Sansó et al. [MR2383247

Abstract

This paper represents a very welcome combination of Statistics and Climate Science. I am sure that no-one who has studied the paper is in any doubt about how demanding this type of collaboration is: it is splendid that statisticians and climate scientists are working together to understand better uncertainty in future climate. As a statistician developing methods for computer experiments, I like climate science precisely because it is so challenging. In particular, the models are still quite poor on the scales for which we would like to use them (transient and regional behaviour). That is to say they have large structural errors : errors that cannot be removed simply by tuning the model parameters. They are also some of the most expensive models in the world to evaluate. Typical performance is about three model-years of output per day at the main research centres. Tony O’Hagan (2006) has termed the consequence of this paucity of evaluations ‘code uncertainty’. In ∗Address: Department of Mathematics, University Walk, Bristol, BS8 1TW, UK. Email j.c.rougier@bristol.ac.uk.