The cat has finally dragged in Mann et al, Robustness of proxy-based climate field reconstruction methods, url đ Supplementary info. This article was first cited by the rather bilious đ Referee #2 for Burger and Cubasch , as though Burger should have been aware of its findings. The coauthors are the “independent” authors: Wahl, Ammann and Rutherford.
Perhaps responding in part to prior criticism, Mann has provided an extensive supplementary information, including code for many steps. (Whether the code works and whether it’s complete are different questions, but on the surface at least, it’s a big improvement.)
Jean S writes:
#99: It’s been already a while đ Supplementary info is available here.
Please, could someone check if I got this right:
Mann is reporting that his (in)famous North-American PC1 is orthogonal (r=0.011422767) to local temperature?
(MBHandMXDcorr.xls, MBH1980-sheet, row 89 and MBHHandMXDcorrNoInstr.xls, MBH1980-sheet, row 67)
Robustness paper:
UC writes:
Under the assumption of moderate or low signal-to noise ratios (e.g., lower than about SNR 0.5 or “80 % noise”), which holds for the MBH98 proxy network as noted earlier, the value of for the “noise” closely approximates that for the “proxy” (which represents a combination of signal and noise components).
Ah, that’s the way to estimate the redness of proxy noise. But isn’t this obvious, we have a model
and as is zero, we can estimate of n directly from the proxy data đ
I haven’t had time to read it yet, but will try to do so soon, but a few quick comments.
I noticed that Mann has continued to use his PC methodology without changing a comma, notwithstanding the strong statement of Wegman that his PC methodology was simply “wrong” and the statement by the NAS panel that it should be avoided (And North’s testimony at the House E&C hearing that he agreed with Wegman) In effect, Mann is saying that, using his PC1, he can “get” a hockeystick not only with the Partial Least Squares regression of MBH98, but with the variation of RegEM (and I recall US pointing out some odd de-centering of his RegEM method.)
It’s one thing for Mann to keep using his PC methodology in the face of criticism from the Wegman and North panels, but why did the JGR reviewers acquiesce in the continued use of Mannian PCs? Pretty pathetic. Actually, it’s not just the JGR reviewers – Mann’s PC1 has been used recently by Osborn and Briffa 2006, Hegerl et al 2006 and Juckes et al – it’s as though the Team is brazenly showing solidarity with Mann to spite Wegman and others.
The word “bristlecone” is not mentioned anywhere in Mann’s new paper. So it’s a strange sort of “robustness” that Mann is proving. It’s already been agreed that, if you take the bristlecones out of the network, you can’t get a HS. So the original claim that the reconstruction is “robust” to the presence/absence of dendroclimatic indicators is false, although you won’t see a hint of that in this paper. Again, what were the reviewers doing? This has been a topical issue – why didn’t they ask Mann to consider it?
I checked Jean S’ comment about the PC1 correlation or lack of correlation and Jean S is right. The correlation of the MBH98 PC1 to the gridcell chosen here is 0.01 – not an imposing correlation for the one series that is essential to the reconstruction.
Here’s something else that’s amusing and shows Mann’s ridiculously perverse stubbornness and the ineptness of climate science referees. It’s been known for over 4 years that Mann mis-located a Paris precipitation in a New England gridcell (and Toulouse precipitation in South Carolina) and that the “Bombay” precipitation series does not come from Bombay. The mislocation of the Paris precipitation series was not corrected in the 2004 Corrigendum and, in the new SI, Paris precipitation is still shown with a New England location (“The rain in Maine falls mainly in the Seine.”) Mann and the mini-Manns duly report that the mis-located precipitation series has a positive correlation to New England gridcell temperatures. You’d think that they try to fix this sort of stuff at some point, but nope, the rain in Maine still falls in the Seine.
Update: I downloaded the reported correlation from the Mannian SI and then re-calculated correlations between the proxies in the AD1820 network and HadCRU3. Their SI stated:
Correlations were calculated between all 112 proxy indicators and both (1) local temperatures (average over the 4 nearest 5 degree lon x lat temperature gridpoints) … during overlapping intervals.
I then did a scatterplot comparing the reported correlations to the ones calculated using HadCRU3 – I used the single gridcell in which the record was located. The high correlations in the top right corner are correlations of actual temperature data to gridcell temperature – something that doesn’t seen like much of an accomplishment. Mann reported that all the correlations were positive, but I got no fewer than 60 out of 112 with negative correlations. Some prominent series e.g. GaspĂ© ring widths had negative correlations with HAdCRU3 gridcell temperatures.
I presume, if a correlation was negative, that Mann just changed it to positive. For a few series, e.g. Quelccaya accumulation, there’s a plausible reason for this, but, in such cases, it would be better policy to invert the series ahead of time on a priori grounds. But there are some real puzzlers. For example, the GaspĂ© series (#53 – treeline11;St Anne) has a negative correlation (-0.11) with HadCRU3 gridcell, but Mann reports a positive correlation of 0.34. It’s hard to tell what’s going on – maybe Mann “inverted” negative correlations for reporting purposes, but they don’t invert the series for use in reconstructions.