Two interesting new proxy studies out recently, one with a meticulous archive, one without.
Kinnard et al 2011 (Nature) here is a proxy reconstruction using 69 proxies to reconstruct Arctic sea ice. It contains a comprehensive archive: all proxies as used as archived; all code is archived. Three of the coauthors are Canadians, including David Fisher, who’s had a commendable history of archiving and distributing data even before web-based solutions. (Fisher’s CDs of data were the core of the ice core collections in MBH98 and Mann et al 2008).
Christiansen and Ljungqvist (Clim Past Disc) here is a reconstruction using 91 proxies (all plotted). Although Ljungqvist has recently made two substantial archives collating recent proxies, for some reason, the present study lacks such an archive. Like Kinnard, it uses a procedure said by the authors to be innovative (but without code.) The combination of no archive and no source code detracts from the ability to efficiently analyse the article. I’ve written to Ljungqvist hoping that they will remedy the situation (and I’m hopeful that they will). Unless they do, I don’t plan to consider this article.
Back to Kinnard. The article uses a complicated multivariate method (Partial Least Squares). The authors provide information on their procedures, but overlook one of the most important aspects: at the end of the day, Partial Least Squares, like other methods, results in a vector of weights for the various proxies. While the authors present maps showing loadings for different PCs, unfortunately they didn’t connect the dots to carry out the linear algebra to extract the weights. It will take a while to analyse.
The D’Arrigo Proxies
In 2005, I tried to get the component chronologies (and measurement data) for D’Arrigo et al 2006 (of which Rob Wilson was a co-author, responsible for much if not most of the analysis, but not in control of archiving decisions.) Unfortunately, six years later, the D’Arrigo-Wilson chronologies remained unarchived.
Kinnard used 11 tree ring chronologies, of which 9 were attributed to D’Arrigo et al 2011 (one to Grudd; one to a third party.) The attribution to Grudd is incorrect: this series also comes from D’Arrigo et al. To my knowledge, this is the first time that these chronologies from D’Arrigo et al 2006 have been archived. (Definite progress here – CA readers will recall that Nature required Moberg to archive third party data sets even if the originating author hadn’t archived the data. Nice to see this happening without a complaint being required.)
Unfortunately, the authors seem to have jumbled 7 of the series, so that the wrong location is attached. (The sites are transposed in the style of the incorrect location of the MBH98 instrumental precipitation data used as temperature proxies.) I noticed this when I plotted up their Tornetrask version which looks as follows. This doesn’t look like the Grudd Tornetrask series at all – as any knowledgeable reviewer would have known. It looks like a Yamal version.
As CA readers are aware, the Polar Urals series of D’Arrigo et al 2006 was actually Yamal (though the core counts illustrated in their figure came from Polar Urals.).
Figure 2. “Polar Urals” series from D’Arrigo et al 2006. (Actually Yamal with Polar Urals core counts.)
Here is the complete transposition (as archived. It’s possible that it’s an archiving error rather than a substantive error):
Tornetrask – relocated to Central Northwest Territories;
Forgfjorddalen, Norway – relocated to Yakutia, Siberia
Central Northwest Territories, Canada – relocated to Taymir, Siberia
Yakutia, Siberia – relocated to Polar Urals, west Siberia
Taymir, Siberia – relocated to Yukon, Canada
Yamal, Siberia – relocated to Tornetrask, Sweden
Yukon, Canada – relocated to Forgfjorddalen, Norway
If the error in their archive exists in their data as used, these erroneous locations will obviously affect spatial maps of loadings and weights, to the extent that these proxies are used. This sort of error should have been observable almost immediately to anyone familiar with the proxies.
Over and above the transposition error, Kinnard et al have incorrectly used the Yamal chronology as the “Polar Urals” chronology (with Polar Urals core counts). In fairness to Kinnard et al, D’Arrigo et al incorrectly labeled the Yamal chronology as Polar Urals and then refused to issue a Corrigendum acknowledging the error.
Kinnard et al do not discuss the discrepancy between the divergence problem and the chronologies selected in D’Arrigo et al. Senior author D’Arrigo told the NAS panel that you have to pick cherries if you want to make cherry pie. The large population Briffa et al 1998 showed declining ring widths in a very large population of high-latitude sites (the divergence problem). The tree ring sites selected by Kinnard are also high-latitude sites, but on balance go up. The inconsistency between the decline in the large population and the rise in the small subset suggests biased selection at some point in the process – an issue not addressed by Kinnard. By blending the tree ring results with other proxies, the information from proxy class separately is not shown. I’ll try to extract this on another occasion.