In our continuing quest for a North American upper treeline chronology that exemplifies the IPCC AR4 claim that additional data since TAR shows “coherent behavior” across multiple indicators, we turn now to the upper treeline proxies of Wilson and Luckman, 2002, 2003. They collected 20 Engelmann spruce sites in British Columbia in 1998, which together with a site from Washington archived at ITRDB, were discussed in two articles, Wilson and Luckman (Dendrochronologia 2002, Holocene 2003). WL03 said that: “All sites were at or within 100—200 m of upper treeline”.
Of the 21 sites, 7 were analyzed for MXD. None of the data has been archived, but it’s only 10 years since it was collected. [Update: This data was archived in August 2007, several months after this post.]
In both articles, subsets of the network varying in size from 5 chronologies to 19 chronologies were selected for analysis, yielding different reconstruction periods varying in start from 1600 to 1847 as shown below. WL03 reported that: “Five short chronologies were excluded from these analyses”. I counted 6 exclusions in their Table 1, one of which commenced in 1702. As Rob Wilson observed below, the six chronologies were excluded because they did not meet an EPS>0.85 standard by 1847.
Two of the three reconstructions (WL03 REG (short) and WL02) start with the calculation of varimax principal components. In WL02, the PCs are calculated from a combined RW-MXD network, while in WL03 the PCs are calculated separately for the RW and MXD networks. The left panel below shows the WL02 PCs, while the right panel shows the WL03 PCs, which are obviously very similar, partly because, as noted in WL03, the RW and MXD chronologies turn out to be “naturally orthogonal”. The weights of the WL02 PC1 (left) are predominantly MXD sites, while the PC2 and PC3 in each case can be interpreted as different weighted averages of RW chronologies. I would have preferred the presentation of simple (or defined weight) averages, as also encouraged by the NAS Panel. This was done for the third WL03 IBC (long) reconstruction in which 3 RW and 2 MXD chronologies were separately averaged. Unfortunately, only the PC series were illustrated (and not the averages).
However, the weights for the PC1 and PC2 are shown and provide evidence that an average of the RW chronologies (from the evidence of the PC2 and PC3) would have a decadal peak in the mid-20th century with late 20th century values below the long-term average, while an average of the MXD chronologies will show somewhat elevated late 20th century values, but no higher than corresponding values in the 1930s. No hockey sticks here. So in terms of our main search, the Engelmann spruce chronologies do not support the IPCC AR4 implication of “coherent behavior” by upper treeline proxies to the extent that this is supposed to support the HS.
Left – from Wilson and Luckman 2002; right, from Wilson and Luckman 2003.
While these two articles have not helped us in our quest for HS-shaped upper treeline reconstructions with values into the 1990s and 2000s, they are instructive in other ways. The traditional dendroclimatic rule of thumb was that lower border sites are precipitation limited and upper border sites are temperature limited. (This seems questionable to me in arid areas but that’s a topic for another day.)
Recently we observed that Gou et al 2007 observed that the situation in arid and semi-arid western China was “complex” and they used a lower border chronology for a reconstruction of minimum temperatures without specifically rejecting the upper border rule of thumb. Gou et al had argued, no doubt with conviction equal to that of Wilson and Luckman, that their chronologies enabled them to reconstruct past minimum temperatures, that warmer winter temperatures led to less frost in the ground and an earlier growing season.
Wilson and Luckman 2002, 2003 likewise eschew reconstruction of mean temperature, but, in their case, they propose a reconstruction of maximum temperature. WL02 provided the following graph showing different trends in maximum, mean and minimum temperature, proposing that the failure of ring widths to increase might be explained by increased nighttime respiration due to higher minimums, without commensurate photosynthesis increases due to no trend in daytime maximum temperatures.
Both papers report a search through climatological variables for the highest correlation. In a passing comment elsewhere, I mentioned that such results seemed somewhat “opportunistic” – in that they resulted from a search through climatological variables – and that it would be nice to see the results repeated on out-of-sample data. (This avoids potential problems of the Texas Sharpshooter type).
I’ve received an email notifying me that this comment constituted further “misinformation” – the Dendro Truth Squad remains alert to the slightest potential error at climateaudit. Notwithstanding this cryptic notice from the Truth Squad, I fail to see what’s wrong with this observation.
Gou et al and Luckman and Wilson are equally authoritative; both search through climatological variables and come up with different variables to reconstruct. That’s what seems opportunistic. Such concerns can easily be put to rest by, say, Luckman and Wilson showing the superiority of their hypothesis on the Gou data or vice versa. I have no horse in this race, other than a desire for replicable analysis.
In any event, we at climateaudit welcome this continued scrutiny from the Dendro Truth Squad, and, as mentioned before, we urge the Dendro Truth Squad to extend their vigorous inspection of even the most minute comment on a climateaudit thread to other ventures, such as the IPCC Fourth Assessment Report.
Wilson and Luckman also noted that there had been increasing cloud cover in the Canadian Cordillera and that this increased cloud cover could be associated with the phenomenon of increasing minimum temperatures without a commensurate increase in maximum temperature. They noted that trends for Tmax and Tmin may have varied in the past and that recent trends in DTR may not be unique, a conclusion that they described as having “potentially profound” consequences for temperature reconstructions in that they hypothesized that it might be “necessary to reconstruct all 3 parameters to establish past temperature variability.”
Wilson and Luckman reported a variety of statistics for their reconstruction, but, as far as I can tell, did not report the regression coefficients. As noted elsewhere, I’m inclined to view their regression methodology as leading to some overfitting in the 20th century in that there is an inverse OLS regression on two “naturally orthogonal” series. Wilson and Luckman argue that this orthogonality is a good thing in that it avoid multicollinearity; I’m inclined to say that it’s a bad thing if you’re trying to extract a “signal”. Wilson and Luckman did not report regression coefficients for the various reconstructions, although they report many verification statistics. Without the regression coefficients themselves, it’s impossible to determine the relative contribution of the regressands (or even whether they all have positive signs.)
As against concern about “natural orthogonality”, Wilson has argued here that RW and MXD target different seasons. WL02 shows response coefficients for their PC1 (~MXD) and PC2 (~RW) series, with the MXD series having very high correlation to August temperatures and a medium correlation to spring temperatures (but apparently not July temperatures); while the RW PC had medium correlations to July temperatures. Correlations are shown for max, min and mean temperatures. The slight difference in seasonal response would justify the orthogonality of the series only if August and June-July temperatures were themselves “naturally orthogonal”. I haven’t checked this situation in Canadian Rocky Mt data, but, if this is so, then this presents a further substantial obstacle to reconstructing a seasonal average, let alone an annual average.
I’m wondering whether the natural orthogonality is more like the orthogonality that one gets from different Legendre polynomials – if you have two integrals with both positive and negative correlations to a “signal”, then the two integral series can well be orthogonal. The inversion problem is that if RW and MXD chronologies are functioning similarly to Legendre polynomials with positive and negative coefficients on monthly temperature signals (let alone precipitation signals), then each of them may be orthogonal to the annual average (which itself is the first Legendre polynomial).
In any event, the evidence is quite clear that there is no HS shape in any simple averages of the site chronologies. I’ll try to re-visit the Wilson and Luckman Jasper/Athabaska reconstruction in the light of this and the related St George-Luckman reconstructions on a future occasion.
I’ve not covered all aspects of the papers as my interest at this time is in raw proxy series that show a strong positive response to recent warming. Rob performs interesting comparisons of his reconstructions to other reconstructions. These papers also originate his interest in the divergence problem. All of Rob Wilson’s papers are online at his website here and I’d encourage interested people to read the originals, since Rob does not always (or rarely) agrees with my characterization of the papers.