Tuesday, December 20, 2011


From: Tim Osborn <t.osborn@uea.ac.uk>
To: simon.tett@metoffice.gov.uk, philip.brohan@metoffice.gov.uk, Eduardo Zorita <Eduardo.Zorita@gkss.de>, Gerd B�rger <gerd.buerger@met.fu-berlin.de>
Subject: report back from PAGES/CLIVAR Wengen meeting
Date: Fri Jun 23 16:35:28 2006
Cc: Keith Briffa <k.briffa@uea.ac.uk>

Hi Simon, Philip, Eduardo & Gerd (cc Keith),

I thought you might be interested in a brief report back from the recent Wengen meeting, specifically about how SO&P-funded work on pseudo-proxies was covered and related hockey-stick issues.

**Please don't circulate this further, because it is just my personal viewpoint**

Thanks for letting me show some of your material. I skipped over some graphs I took from Philip's regression presentation at the SO&P meeting because Francis Zwiers covered forward/inverse/total least squares before me. I did show some results from Eduardo, including pseudo-proxy results from Erik-II. And I showed a figure from Gerd's "many flavours" pseudo-proxy paper.

The meeting included fairly intensive discussions about many issues, and this included some discussion of von Storch et al. (2004, 2006), Wahl et al. (2006), Mann et al. (2005), Burger and Cubasch (2005) and Burger et al. (2006). Generally the discussion was quite open, with only a few disdainful remarks made about the work of people not there -- certainly not enough to distract from useful discussions.

In general, most people accepted that the MBH method could, in some situations, result in biased reconstructions with too little low-frequency. I'm not sure how much Mike Mann accepted this, but it was reinforced by findings shown by Eugene Wahl that indicated some bias in their CSM pseudo-proxy studies, and particularly by Francis Zwiers who looked to have almost completely replicated the von Storch et al. results with respect to the MBH method (though he emphasised the preliminary nature of his work and he may not have implemented the MBH method correctly... we'll have to wait and see).

Mike showed many detailed psuedo-proxy tests of the RegEM method and these seemed quite convincing in showing little problem with that method... it does assume equal error in both instrumental and proxies, so it should show less bias than other methods that wrongly put all the error in the instrumental record (i.e., "typical" regression).

So... there was some confusion about how the MBH method can be biased but the RegEM not be biased (in pseudo-proxy tests) yet they give the same results for the real proxies.

Mike thought it might be the ECHO-G vs CSM differences, but I argued against this and was supported by Caspar Ammann and Eugene Wahl who did not think that the character of the model runs was a big factor in explaining different results.

There was limited discussion of trend/detrend and white/red noise pseudo-proxy issues. Many seemed to think that if pseudo-proxy studies showed that detrending definitely caused a problem, then this was a reason not to detrend. The alternative of finding a method that worked with detrended data was not really discussed.

The discussion was fairly constructive and for the most part friendly. Eugene Wahl in particular seemed keen to "build bridges" within the community.

I should also mention two of the workshop outcomes.

The first is that a paper is being planned based on the things discussed at the workshop and covering many issues from proxy data, forcings, model simulations and reconstructions. I hope that the authorship of this might be wider than just the participants of the workshop, but we will have to wait and see who else is asked to contribute.

The second is that we should set up a "climate reconstruction challenge". The idea would be to use a simulation (*not* of the last 1000 years, so none of us know the expected answer) and provide some data from a "calibration period" and some "pseudo-proxies" from the full period and make these public so that anyone could attempt to make a reconstruction using their favoured method(s). The true model NH temperature series would be kept secret for 6 months or so. Thus it would be a "blind" test and after attempts had been submitted they would be evaluated against the true result to assess which methods were most successful.

Caspar Ammann will probably provide the simulation, so he wouldn't take part in making any reconstructions. He would keep the details secret from all others so that any one, including MBH, you and us, could enter the challenge.

Finally, it was asked whether the model runs that have so far been used for pseudo-proxy studies (NCAR CSM, ECHO-G Erik-I, HadCM3, maybe ECHO-G Erik-II?) might be made publicly available for shared use, so results are less model dependent. This would just be the surface air temperature fields from the runs, not all the other variables. What do you think, Simon and Eduardo? If you are happy with this then they could get them from the SO&P website, so no need for data extraction on your part.

Hope you find this summary interesting. It's just my opinions. I've cc'd this to Keith in case he wants to say anything different!



No comments:

Post a Comment