A while ago, I discussed the very interesting study by Naurzbaev et al  (co-author Hughes), which calculated growth curves at 34 larch sites in a meridional transect from 55 to 72 N (at a longitude of about 90-100E) and 23 larch sites along an altitudinal transect from 1120 to 2350 m around Tuva (~ 51N, 95E). This appeared to me to be a nicely systematic approach to trying to extract climatic information from tree rings, as opposed to Mann’s peculiar data mining methods. They calculated growth curves by age for various sites and asserted that the parameters were related to latitude and elevation. This represented the simple point that ring widths at higher latitudes/elevations were thinner than for lower latitudes/elevations – which is the premise of using ring widths as a temperature proxy.
Hughes and Funkhouser [Climatic Change, 2003] studied 4 bristlecone/foxtail paired sites at upper and lower elevations. The elevation differences ranges from 200 meters (Timber Gap) to 600 meters (White Mountains). I thought that it would be interesting to do some simple comparisons of ring widths at these paired sites. In order to use bristlecone ring widths directly as a temperature proxy, one would presume that the ring widths at the (cooler) higher sites would be narrower than the ring widths at the (warmer) lower sites.
Figure 1 below shows top panel (mean ring widths) and bottom panel (age-adjusted "chronologies") for upper and lower sites in the Spring Mtns NV (Charleston Peak). Strikingly, the average ring widths for the upper site and consistently and significantly wider than the lower site. This is going the wrong way for the ring widths to operate as a temperature proxy – the warmer site has narrower ring widths. ( The age composition of the upper and lower sites is pretty similar and has nothing to do with the effect.) In bristlecone literature, the lower sites are said to be "precipitation limited" while the upper sites are said to be "temperature limited". But there’s obviously more going on than this simple dichotomy. First, the two series are correlated (0.29 here and the other pairs are higher). Additionally, it begs the question about transitions – the difference in height is only 400 m here. Do the ring widths increase at an intermediate elevation – say 3200 m. The diagram also demonstrates an important loss of information in the "age-adjusted" chronologies: the dimensionless units end up removing the interesting information about the relative ring widths.
Figure 1. Spring Mountains NV. Top panel – Mean ring width for upper site (Spring Mt Upper 3420 m) and lower site (Spring Mt Lower 3000 m); bottom panel – archived chronologies (age-adjusted) for Charleston Peak (upper) and Spring Mt Lower
Secondly, I show the same information for the White Mountains CA, where the same phenomenon occurs – wider ring widths at 3400/3475 m as compared to 2805 m. These are the most famous bristlecone sites: Sheep Mountain and Campito Mountain are the upper sites (these are well-known to readers of this site as the most highly weighted proxies in the MBH North American PC1 and the most influential proxies in the NH temperature reconstruction). The lower site is Methuselah Walk, a famous bristlecone site. The top panel of Figure 2 below shows the simple mean ring width (without age adjustment) by year. The correlation between the two series is 0.38. The bottom panel shows the archived chronologies for Methuselah Walk and Sheep Mountain. The age-adjustment here, as elsewhere, has the effect here of decreasing MWP levels relative to modern levels. But once again, in the simplest terms, the higher cooler site has wider ring widths.
Figure 2. White Mountains CA Top panel – Mean ring width for upper sites (Sheep Mt 3475m; Campito Mt 3400 m) and lower site (Methuselah Walk 2805 m); bottom panel – archived chronology (age-adjusted) for Sheep Mt and Methuselah Walk. Horisontal scale is from 0 to 2000.
The third Hughes and Funkhouser  site is Pearl Peak NV. The same phenomenon occurs once again. The higher site (3170 m this time) has wider widths than the lower site (2810 m). The correlation here (0.51) is quite high.
Figure 3. Pearl Peak NV. Top panel – Mean ring width for upper site (Pearl Peak Upper 3170 m) and lower site (Pearl Peak Lower 2810 m); bottom panel – archived chronology (age-adjusted) for Pearl Peak Upper. Horizontal scale is inconsistent between panels. It is correct for lower panel; upper panel goes from 700 to 2000.
Finally, here is the comparable information from Timber Gap CA, a foxtail pine site, previously discussed with pictures here . This has the least difference between upper (3216 m) and lower (3017 m) sites. The correlation is 0.49. The effects of age-adjustment on the chronologies is quite marked here in affecting the relative MWP and modern levels. (Given the potency of the issue of MWP and modern levels, it is remarkable how much hangs on nuances of age adjustment methods.) A very striking aspect of the top panel is the reversal of levels between lower and upper sites around 1900. What accounts for this? You’d think that such an interesting phenomenon would have provoked some commentary in the literature prior to using the series as key temperature proxies.
Figure 4. Timber Gap CA. Top panel – Mean ring width for upper site (Timber Gap Upper 3216 m) and lower site (Timber Gap Lower 3017 m); bottom panel – archived chronologies (age-adjusted) for Timber Gap Upper (upper) and Timber Gap Lower.
In the past, I’ve pointed out many problematic aspects to bristlecones as a temperature proxy – see especially our E&E article. The CO2 fertilization hypothesis of Graybill and Idso is an important consideration, but we did not limit ourselves to this particular hypothesis. Among other points, we observed that, since bristlecone is competing with sagebrush, it is obviously stressed by precipitation, a point which is made in the specialist literature. – let alone THE critical proxy in NH temperature reconstruction. In any event, there are obviously some complicated gradients affecting bristlecone growth and the sites do NOT display the simple temperature gradient that is essential to a good temperature proxy. Problems with these proxies have an impact on MBH, Crowley and Lowery , Esper et al  and Moberg et al .