date: Tue Jun 21 15:17:51 2005 from: Keith Briffa subject: Re: Moberg paper to: "Brohan, Philip" Philip I do agree with your sentiments here - and if it proves possible , would like to be involved , along I am sure with Tim. We have as you know , been thinking along the same lines (in terms of the apparent differences in many reconstructions being due to the specific choice of regression method - rather than fundamental different signal in the selected proxy predictors). We intend ,still to do some work to illustrate this , largely stimulated by your contributions in a couple of the SOAP meetings. It now transpires that Gabbi Hegerl and Tom Crowley (along with Myles Allen and Henry Pollack and Eduardo ) have yet another NH reconstruction (using many data in common with various existing reconstructions , and heavily biased towards tree-ring data) , calibrated using a total least squares approach applied to regional , decadally-smoothed predictors, and subsequently seemingly coinciding with the Pollack Ground Surface temperature reconstruction (Pollack and Smerdon ) and the ECHO-G simulation- suggesting a much larger amplitude of temperature change (warming) over the last 500 years than many other reconstructions. In case you have not seen this , I attach it (please treat Hegerl et al. paper as confidential). Obviously this approach (smoothing and TLS) gives larger amplitude change than simple least squares on interannual data - and re-calibrating other reconstructions in the same way would demonstrate this. We have been working for a while trying to recreate the Esper et al series , which originally used many of the data Hegerl/Crowley use, smoothed and simply adjusted to match the mean and standard deviation of the Mann et al series originally , and then later re-calibrated against smoothed temperature data by Cook et al. Our purpose was to discuss a likely flaw in the logic of the tree-ring standardisation in the first instance and then show the potential for the final reconstruction to be biased through time, according to the changing constituent series in the single mean tree-ring predictor. It turns out that the method they actuallly used does not corespond to the description in the papers - but they include a re-normalisation step before producing the mean predictor , that mitigates much of the problem. I will not go on about this here though. The conclusion of this work though , is that the final reconstruction amplitude is extremely sensitive to the scaling method and period used. As we discussed in Reading , it is becoming the accepted practise to sacrifice any formal regression validation (and any chance to realistically gage the validity of the regression estimates ) for the sake of getting large low-frequency variance in the reconstruction - Moberg's work exemplifies this. There is a special PAGES/CLIVAR workshop being organised to address these exact same issues - as well as the approaches to testing reconstruction methodology using model simulations. This will be held in Switzerland next year and we (you and us etc) could work towards this as a medium-term goal , as well as a more immediate response to Moberg. Best wishes Keith At 11:52 20/06/2005, you wrote: Keith That's not good news, and I'm surprised as well as disappointed to hear it. The errors in the Moberg paper were evident to quite a few different people on first reading and I thought the Mann et al. comment explained them pretty well. The problems with the MOBERG paper stem from incorrect use of statistical regression. It is not the only paper to use regression incorrectly. I now believe that many of the differences in paleo-temperature reconstructions come from differences in regression methods and are due to different (implicit) assumptions about the errors in the paleo and instrumental temperature series. And also that by making those assumptions explicit we can show that some reconstructions are better than others. So rather than write a specific critique of Moberg et al, I'd like to write a more general paper on methodological differences. Perhaps a specific comment on the MOBERG method could be built into that. Philip On Fri, 2005-06-17 at 10:12, Keith Briffa wrote: > Philip et al. > don't know if you know , but the Comment by Mann et al. , dicussing the > Moberg methodology will not be published now. Nature considered it > unwarranted to include the simulation evidence - but would perhaps of > allowed some other brief remarks which Mike and the rest of us decided did > not justify publication - apparently he is including some of the content in > a revised paper (J.Climate) which was originally a response to the von > Storch work. I still consider it important that something from a wider > group is published giving an indication at least that there are problems > with Moberg's approach - note also the inconsistency in the timing of the > (premature) warming in the 20th century in his record (apparently > associated with the rigidity of the wavelet approach they used). Their > reconstruction is the only one giving apparent evidence of a warmer > Medieval period , and may also be highlighted in a DEFRA document that > Geoff Jenkins is the contact for! > Could/should you consider resurrecting the comment , or should we consider > more comparisons of his method using the ECH-G data and writing a paper? > Keith > > > > -- > Professor Keith Briffa, > Climatic Research Unit > University of East Anglia > Norwich, NR4 7TJ, U.K. > > Phone: +44-1603-593909 > Fax: +44-1603-507784 > > [1]http://www.cru.uea.ac.uk/cru/people/briffa/ -- Philip Brohan, Climate Scientist Met Office Hadley Centre for Climate Prediction and Research Tel: +44 (0)1392 884574 Fax: +44 (0)1392 885681 Global climate data sets are available from [2]http://www.hadobs.org -- Professor Keith Briffa, Climatic Research Unit University of East Anglia Norwich, NR4 7TJ, U.K. Phone: +44-1603-593909 Fax: +44-1603-507784 [3]http://www.cru.uea.ac.uk/cru/people/briffa/