Climate Audit

AS both David H’s observe, the Stern Review is expected next month. Their science views are summarized here. I may post some headnotes at a later time, but it’s an interesting browse and some of you may wish to comment on it specifically.

The U.K. Royal Society has recently sent a letter to ExxonUK which has attracted commentary about whether it is an interference with free speech (see Roger Pielke and the discussion there) . I’m interested in a different aspect of this letter – their reliance on IPCC detection and attribution discussion. The Royal Society takes exception to the following comment in Exxon reporting:

I think that there’s considerable justification for saying that IPCC conclusions rest on "expert judgement" rather than "objective, reproducible statistical methods". I don’t think that there’s anything necessarily wrong with people making decisions based on "expert judgement" – this is done all the time. Indeed the NAS Panel based its impressionistic assessment of temperature history on expert judgement rather than confidence interval estimation – a point made clearly at the NAS press conference. However, the Royal Society took great umbrage at the above characterization. They went on as follows:

In my canvassing of the Hockey Team results, I would say the opposite: Team results are not based on "objective, reproducible statistical methods"; however the Royal Society does not seem to have Hockey Team reconstructions in mind (although U.K. official opinion seems to be that the HS has not been dented); their idea of "objective, reproducible statistical methods" is the Detection and Attribution chapter of IPCC, chapter 12 – note the Appendix, in which they report:

Appendix 12.1: Optimal Detection is Regression The detection technique that has been used in most “optimal detection” studies performed to date has several equivalent representations (Hegerl and North, 1997; Zwiers, 1999). It has recently been recognised that it can be cast as a multiple regression problem with respect to generalised least squares (Allen and Tett, 1999; see also Hasselmann, 1993, 1997) ….

The "detection and attribution" literature uses terms like "optimal fingerprinting", which seems to be high-falutin term for multiple regression (or "multiregression" as it is sometimes referred to in this literature.) Prominent authors in this vein are Hegerl, Stott, Tett and Myles Allen of the climateprediction,et 11.5 deg C press release. I’ve browsed this literature and been put off by the opaque and inflated language, which sometimes makes Mann seem like Hemingway in clarity and purpose. However, it’s probably time to start grasping this particular nettle and see what actually lies underneath these "detection and attribution studies". I’m going to try to pin down which studies are "seminal" in this field. I would be interested in contributions from anyone who is successful in translating any of this turgid prose into conventional statistical concepts – ideally we would start with about 6-8 notes on specific studies that have been cited in the field.