There are 3 different versions of the Alberta site that have been applied in multiproxy reconstructions: 1) Luckman et al  used in Jones et al ; Crowley and Lowery ; Briffa ; 2) Esper et al.  ; 3) the version used in Osborn and Briffa , presumably from Luckman and Wilson  and presumably the one used in D’Arrigo et al  (although details of the latter are lacking). Each version has replication issues with permutations based on data combinations and methodology variations.
A theme that’s emerging for me here is: on a macro-scale, the Hockey Team takes great comfort in what they regard as "similarity" in the hockey sticks in the spaghetti graphs. However, at the level of individual sites, it’s remarkable how little similarity there is between versions. I’ve commented on this for Polar Urals. I’ve noticed something similar for Greenland – remind me to post this up. Today I’ll nibble at the Alberta site, showing the 3 versions used in multiproxy studies and a trial emulation of the Esper et al. version. I’ll get to trial emulations of the Luckman versions on another occasion.
Figure 1 below shows the three multiproxy version in a format often used by the Hockey Team – the data has been smoothed with a gaussian 40-year filter and then scaled. As you see immediately, the two Luckman versions (the later one being labelled Osborn here since it was used in Osborn and Briffa) are fairly similar ( unsmoothed correlation is 0.72; smoothed correlation is 0.65), but the Luckman versions are quite different from the Esper version (0.26 for both smoothed and unsmoothed correlation between Esper version and later Luckman version.) The low-frequency variability in the Esper version is greater than the low-frequency variability in either Luckman version ( readers should recall that there’s been much pontificating about the causes of differing low-frequency variability with arm-waving about tropics versus extratropics, but none of the commentators seem to have bothered examining individual series in detail. Maybe the issues are different than people think.) The Luckman reconstructions are primarily based on the MXD reconstructions, Esper on the his RCS RW series.
We also see remarkable differences in the early 15th century values – these are at series lows for the Esper reconstruction and near series highs for the Luckman reconstructions. It seems to me that differences such as this should cause people to pause a little before concluding that they know final results to within 0.2 deg C. or so, as Mann claimed recently at the NAS panel. I would like it if authors reconciled site results: for example, it would be interesting to see how Esper et al  reconciled their highly divergent results to those of Luckman et al 1997, at least in their Supplementary Information; same with Luckman and Wilson  with respect to Esper et al. 2002.
Figure 1. Alberta ring widths from near Athabasca Glacier for Luckman et al , Esper et al  and as used in Osborn and Briffa .
I’ve done a fair bit of work trying to emulate each of the versions, and each emulation attempt leads into many by-ways. Today, I’ll report on the one where the data was most cleanly available (at least as of 7 days ago) – Esper et al. . Figure 2 below shows the Esper version (colored blue for consistency), my "emulation" of this using RCS on the data set ath.rwl provided by Science presumably from Esper; and thirdly, my emulation using RCS on all 4 Schweingruber data sets in the area (cana170w, cana171w, cana096 and cana097) – only the first two are used by Esper (but the first and third are used in L97 and all four plus others are used in LW05).
While there were important similarities, there were some important and puzzling differences. The correlation between my emulation and the archived Esper version was 0.85 (0.83 – smoothed), but the differences are striking. The Esper version does not contain the 20th century decline shown in my emulation. This difference probably relates back to Esper’s distinction between "linear" and "nonlinear" trees – a distinction that is not made in botanical texts on conifers to my knowledge. I have requested details on how this distinction was made for a very long time and it still remains as mysterious as ever. Also intriguing is the RCS result using all 4 available data sets in the area – in other cases (e.g. Mangazeja), Esper collected 4 data sets – why not here? Especially since cana096 was actually used in L97, which Esper cited.
LW05 has dated quite a few snags to the MWP which remain unarchived. Other than that, it appears to me that the 4 data sets used in the calculation are, at a minimum, a very large subset of the LW05 data and should yield an approximation to the larger LW05 data set in its later portion.
(Updated) The all-WDCP emulation shown below, based on one RCS curve for all 4 data sets, should be a reasonable approximation of a RCS RW version in Luckman and Wilson  periods using one regional curve. Such an RCS curve shows the remarkable phenomenon of the site chronology declining steadily from the late 18th century. LW05 show an STD chronology, which I’ll show in a forthcoming post. As to the RCS method applied to RW data, LW05 observed:
"Significantly more low-frequency information was captured using the MXD data (See Appendix) but no significant gain was observed by using the RCS method on the RW data (analysis not shown)".
In comments below, Rob Wilson pointed out that the RCS version used in their analysis contained startification levels. Obviously such stratification will tend to move results from a bulk RCS curve towards the STD curve. Obviously, there is more low-frequency information in the “bulk” RCS curve, but this may not be so in a “stratified” RCS curve. Hopefully, we ‘ll be able to get some information on the form of stratification and see how this works.
The difference between my RCS calculation using all four WDCP data sets and the Esper archived version using two archived data sets is also dramatic. Again I’m not saying that anything is right or wrong. Also startification will tend to reduce differenecs. At the NAS panel, Mann purported to be able to estimate 11th century temperature to within 0.2 deg C (a far lower confidence interval claimed in LW05 for example – see Rob Wilson comment below). It’s hard for me to see how Mann can reach such conclusions resulting from such unstable results.
Figure 2. RCS on Alberta site. Black – emulation using Esper rwl data; red – emulation using 4 WDCP data sets; blue – archived Esper version.
Note: I have other material which I intend to post up and will post up information on the STD RW reconstruction and on regressions (I’d intended to do so anyway, see “nibble” comment above, but see Rob Wilson comment below.)