With hurricanes over Czech and Rain in New Mexico and the truely bizarre shuttle flying even though unsafe, is there any time or space for another round of hockey stick wars? No… don’t go away, its interesting, really it is!
There is an exchange of letters in the most recent Science, between Rahmstorf and von S (et al. (curiously enough I’ve just noticed that Simon Tett is one of the al.; presumably because of the HadCM3 use)). Sadly “exchange” isn’t quite right: just a letter and reply. This is a clear case of where it would be nice to see a more extended exchange to pick at the areas of disagreement (NRC panel anyone?). Anyway, a copy of the letters in question is here; if you have the right codes you can read it guilt-free here instead.
What is this about? In 2004, von Storch did something in principle very sensible: he tested the MBH reconstruction method with the output from a GCM run over the last 1000 years. Its a good idea: feed in the MBH locations and lengths, see how well it reconstructs the (known, cos this is a model) global or hemispheric mean. However… there were various problems with the way von S did this, which makes it unclear if the test was any use (RC).
He did this with his ECHO-G model, and with HadCM3, which was only represented in the online supplement and ref’d in the main text with the not entirely honest “Similar results are obtained with a simulation with the third Hadley Centre coupled model (HadCM3)”. Well, you can look at the figures in the recent letter (not quite sure why they only go back to 1750; might be because HadCM3 starts then) to see this just isn’t true: the HadCM3 results are consistent with the IPCC 2001 graph, and the error in the model reconstructions in that case is small. This appears to be at least in part due to the large climate drift in the ECHO-G simulations, driven by the wrong initial conditions and possibly over-exhuberant natural forcing. nb there is a version of ECHO-G, re-run without the drift problems, but I don’t know if any results from that have been published. It would be interesting to see if the idea of the drift in ECHO-G being important is right or not, and presumably fairly easy to test.
This is a devastating criticism and so von S et al. obviously have held their hands up and said “its a fair cop guv”… no, of course not: no one ever does in exchanges of this sort. Their main response seems to be that although the IPCC 2001 (ie, MBH99) and HadCM3 look like they fit, they don’t really, because the wrong (too big) errors bars are being used: the errors bars should be much smaller. This is where it gets tedious and technical and you go off to watch the world cup, but I’ll persevere anyway, as far as I can follow it. von S cites Gerber (2003) as an update on the errors bars on MBH98. Sounds plausible? But its wrong. Gerber use the ’98 error estimates, which were revised and increased for MBH99, or so I’m told. But the ones Rahstorf is using are the MBH99 ones, ie the ones used in IPCC 2001, ie the oes everyone is used too. And which the NRC have just told us should probably be bigger anyway. So it seems rather peverse of von S to claim they should be smaller. And the Gerber estimates (I’m told; sorry I haven’t read it :-) are apparently derived for the running means by sqrt(n)’ing, on the (unreasonable) assumption of no autocorrelation.
[Update: Stephan Rahmstorf's view]