Perfecting Imperfect Models

I was at the Royal Met Soc wednesday meeting today. The abstracts are here: page 1 and page 2.

So... just for once I'm reviewing people I know, in some cases, so let me please stress that these are my personal reactions to the talks, not reflections of their true worth. These notes were written on friday, so I've forgotten some stuff. The meeting was quite full: about 170 people they said, once the late arrivals were in, which meant a very long coffee queue.

Anton Beljaars - in principle, this seemed like a nice idea - NWP and climate models are very similar (in code) so you can study errors in the climate version by looking at how the NWP version works, which is nice because then you are in a more controlled state. However, I didn't get much from his talk.

Mike Davey - this was the one in which I got Weather out and started reading about global dimming... sorry. The bit I remember was that it might be possible to improve seasonal forecasts by correcting the known model drift. Which is fair enough if you're interested in seasonal forecasts, but I'm not.

Steve Woolnough - a good talk. Largely about the MJO (which I have finally sorted out from the *QBO* (please don't laugh at me)). Anyway, model deficiencies in simulating the MJO, etc. Then (and this was the bit that interested me) the idea that maybe this is because of the failure of SST response in hadcm3, and maybe this was because of the once-a-day ocean/atmos coupling. But if you switch to hourly coupling, it makes little difference (aside: I've tried this too. I too noticed that it made little difference, though I was mostly interested in the Antarctic ice. There was alwaysa slight worry that the model wasn't behaving right, since no-one uses anything other than once-a-day). But then they tried altering the thickness of the model top level from 10m to 1m, and this makes a difference, because it makes the layer thin enough to allow a noticeable diurnal cycle, which affects the mean SST. So I liked that. As far as I remember, though, he didn't then go back and say, aha, with this the MJO is better. And he didn't say how it improved the overall SST sim (poss made it worse, since the SST gets bumped up by several oC during ?active? MJO periods).

Dave Sexton Stuff from climateprediction.net. I've always been a teensy bit dubious about this (see recent sci.env for some of my provably invalid objections :-) but the talk was very good. So, they have a big perturbed-physics ensemble (see-also QUMP) - 2000 members??? - which ?does/doesn't? include those rejected at the spin-up end as failing the "fits basic climatology" test. And then you get a spew of future projections, with a noticeable tendency to bunch at what looked about consistent with what we all think anyway. Someone in the audience pointed out that the fit-to-clim tests were very primitive, and that many of the passed members would have failed more stringent tests. I think DS accepted this point. I think that he said that this would be appearing in Nature soon.

[My wife says: Stoat stoat stoatety stoat stoat stoat (ho ho, that will teach her to edit my posts while I'm not looking, I've left it in...)]

Then it was tea break, and I listened to Alistair McD patiently explain basic radiation theory to William Ingram :-)

Tim Palmer TP, superstar, and he did give a good talk. This was about using cellular automata as a possible alternative subgrid scale parametrisation. I haven't heard this before; JCK says he heard it recently at a conf; this pdf contains quite a lot of the stuff in the talk (and more: there were no buttterflies in the talk; the CA stuff starts at p29); and the pdf references Palmer, 1997. I think the idea of the CA is that its very cheap, but can nonetheless allow communication between nieghbouring points in a way that tradiational params can't.

Alan Thorpe At NCAS, but newly appointed Chief Exec of NERC (story here) so he pays my wages... so I'd better be polite. I didn't think it was desperately impressive. He seemed to have been seduced by the Earth Simulator and showed a truely beautiful movie of 1km-scale clouds it had generated. It must have been beautiful (it was) because its pushed everything else he said out of my mind :-(

After that there was a Panel discussion at which ones heart sinks, because these are the sort of things that sound like a good idea at the start but come 5pm people are thinking of running off to their trains. But, in fact, it was good. It was sort-of lead by AT, because he put up some points for discussion, the first of which was "we don't need params, because in 10-15 years we'll be running globally at 1km in climate mode". Errr... well no-one was falling for that. Probably it was just to stimulate talk, fair enough. And what people said was: no way do we get to 1lm in just 10-15 years; and even at 1km you need params, just different ones. Paul Mason called ensemble forecasting a dead-end; when pressed (by an anguished TP) as to whether he really meant this he said yes, though possibly a necessary dead end.

Then I speed-walked back to Kings Cross via Mornington Crescent, which is fun, *and* I got a seat on the train home. And I read the latest Viz - and I must say, its better than it used to be.

[nb: this post is the winner of todays vote-a-post contest, so gets its timestamp changed to push it to the top of the stack]


James Annan said...

I'd rather you commented on this meeting which I was unable to attend for obvvious reasons :-)

James Annan said...


BTW I don't think David Sexton is directly linked to cpdn - he's is in the QUMP group at the Hadley Centre instead.