Tuesday, October 25, 2011


(You'll have to google your own links, due to computing/time/wifi limitations.)

Despite our best intentions ("Let's go for a dawn run along the Cherry Creek Trail!"), jetlag got the better of us and we only woke at 9am. Due to a prior appointment with a breakfast burrito, we missed most of the opening blurbs, but there was still plenty of acronym soup to go round. And I'm still not really much wiser about the actual purpose of it all...there's just enough politics mixed in with the science to be a little disconcerting, to me at least.

The plenaries have been mostly rather general, though there was an interesting (to me) talk on poorly-quantified aerosol effects. The parallel sessions were more detailed. Organisationally it has been a bit of a mixed bag - the setting seems very grand but there have been several presentation problems, and quite a few speakers are absent - including, to my amusement, someone who told one of their underlings on Friday that they were not attending, leaving said underling not only having to present, but even write, their 30 minute invited presentation over the next couple of days while at the conference.

I spent most of the day in the seasonal to decadal prediction session, which still remains rather a challenge. A few years ago, some meeting of the great and good decided that they were all going to do 30y forecasts, and unfortunately our institute took this suggestion rather more seriously than it deserved (in our opinion). Of course, skill beyond a couple of years (if that) has proved hard to find. The models do suggest a bit more potential predictability, but issues of model error and initialisation make this rather a challenge. Still, it's a good excuse for more funding.

Tim Palmer presented some good results from his stochastic parameterisation, but this work only extends to the seasonal time scale. Having started his talk with a complaint that all climate models were too similar, it was perhaps a little incongruous that he ended with an appeal for us all to club together to build a single climate model with stochastic parameterisations. I suppose you can't criticise one model for being too similar to itself :-) But it seems obvious to me that one model - even a stochastic one - cannot reasonably represent our uncertainties regarding the behaviour of the climate system over long time scales (and indeed if it did, then Jim Hansen would have got there 30 years ago.) There was a surprising amount of discussion throughout the session about the good performance of the multi-model mean, and I was relieved that another commenter saved me from having to suggest that an ensemble of (different) stochastically-parameterised models would outperform a single one.

Posters were enhanced by a reasonable quantity of free beer (that despite prior threats, was indeed "free as in beer"). But they weren't particularly on-topic for me, and I was flagging by then anyway.

No comments: