Wednesday, April 11, 2007

EGU poster

It's just the same old same old for next week's EGU meeting in Vienna. If I'd managed to get it published anywhere I'd have moved on to something else, but although this seems to have been dragging on pretty much for ever, only a handful of people have actually seen it so far (mostly at last summer's workshop) and it's far from clear how widely accepted it is. So I'll give it one last airing and see if it generates any reaction...

Jules tells me I should mention her poster, and also this one, which was originally submitted as a rather speculative offer to describe some not-yet-done work, and which has basically degenerated into a glorified advert (see also here) for someone to come and do it - anyone want to visit to Japan for a few years?

Please don't find any typos. Or if you do, don't tell me :-)

8 comments:

Anonymous said...

Acceptance? Vindication? : http://ams.allenpress.com/perlserv/?request=get-abstract&doi=10.1175%2FJCLI4064.1
I don't subscribe, but the abstract suggest that your ideas are filtering through. What do you think of the paper?
Work in Japan? If only I had the qualifications, I'd bite my own hand off for the chance (assuming being one-handed wasn't a problem).

James Annan said...

Fergus,

Thanks for the ref. The paper is here - google is very good at finding publicly-available versions!

The idea of "robust Bayes" analysis is that instead of using a single set of prior assumptions, a range is used, producing a range of posteriors. This is an attempt to get around the problem of having to make highly detailed decisions about things when one is really rather ignorant. However, you still need to choose a plausible range of priors, so the problem does not really go away. By simply turning Bayes' Theorem around, it is easy to see that any posterior can be generated simply be choosing the prior to be proportional to the wished-for posterior divided by the likelihood. So the universe of all possible priors gives the vacuous result of all possible posteriors.

Unfortunately they still make the mistake of claiming that a uniform prior in S is "uninformative", and include such a prior in the range that they use. The submission date on this work predates my saying anything on the matter so I can't claim any credit for that :-)

Hopefully some of the authors will be at the EGU. I know at least one planned to be there.

Anonymous said...

James,
Michael is interested in your relationship with Gabi: http://initforthegold.blogspot.com/

I was going to say something, but perhaps you'd like to comment...
Regards,

Michael Tobis said...

Yes, I thought I had posted here to that effect. Sorry if there was a glitch. James, I am very interested in your opinion of my understanding of the debate.

James Annan said...

Well, as you saw, it's just the same old stuff I've been going on about for some time now.

Dave Frame previously complained (and not entirely unreasonably) that I shouldn't pick on him alone as he is not the only one advocating uniform priors as representing ignorance, There are indeed plenty of examples in the literature, and Fergus just added one more to the list. Hey, I might even have written something like that myself somewhere before I leant better :-)

As for Gabi Hegerl, there's no particular "feud" as far as I'm concerned, but as IPCC lead author she's supposed to be representing the views of climate scientists. Therefore, it seems entirely appropriate that I should clearly identify the disagreements I have with that aspect of her work (and in fact two different people have independently told me that she was directly reponsible for that section of the AR4). If she is now openly acknowledging that these widely-publicised pdfs are not credible (as your post seems to imply) then I would consider that to be a significant step forward.

I do disagree with part of your characterisation of the issue. It is actually Hegerl et al who arbitrarily cut off high sensitivites with P(S>10) = 0 in their prior, contrasting with Frame et al's P(S>10)=50%. My "extended tails" prior has P(S>10)=5%. If you can explain how both 0 and 50% can both represent "ignorance" on this issue whereas 5% is some dodgy question-begging then I'm all ears.

As for Lubos, well I hope no further comment is required...

Anonymous said...

I use this to find interesting stuff now and again: http://www.ukcip.org.uk/news_releases/38.pdf

Items 8 & 9 refer to papers by Rougier and Hall et. al. on issues related to probabilistic assessment from Climatic change vol. 81.

The Rougier in particular looks interesting
Regards,

Anonymous said...

James,

I'm curious what you think of the following:

http://aerosols.lanl.gov/conf2006/talks/files/Volz.pdf

Does this positive feedback from wind induced change in seawater effective emissivity reduce calculated climate sensitivity?

Anonymous said...

Apparently the url got cutoff:

http://aerosols.lanl.gov/
conf2006/talks/files/Volz.pdf