Polar Urals #1

The Polar Urals site is a staple of multi-proxy studies.

The Briffa et al. [1995] version is used in Jones et al [1998], MBH98, MBH99, Briffa et al. [2001] and most recently Jones and Mann (2004). An earlier version by Graybill and Shiyatov was used in Bradley and Jones [1993], Hughes and Diaz [1994] and anomalously again in Crowley and Lowery (2000) [incorrectly citing the Briffa version]. The site is used in Esper et al [2002], whose data is unavailable and the version is not stated.The Briffa reconstruction is shown in Figure 1 below. Briffa et al. [1995] stated that 1032 was the "coldest year" of the millennium (over 6 deg C below normal) – the 1032 downspike is noticeable in the graph. The Briffa reconstruction also shows cold not simply in 1032, but through the 11th and 12th centuries, during the famous period of Viking exploration. Briffa et al. argued that these results showed that the Viking warmth was regional at best and to disparage the idea of a MWP. These conclusions were dramatic, especially at the time.

Polar 1

FIGURE 1. Briffa Temperature Reconstruction from Polar Urals.

The Briffa reconstruction was vital to the Jones et al. reconstruction: it is one of only three Jones et al. [1998] proxies in the early 11th century and the only one with negative values. (I’ve already pointed to strange goings-on with the Tornetrask proxy, which is one of the three. I’ll show some impact of the Polar Urals proxy below.

But first, one would assume that a study with such a profound impact would have careful quality control in its most controversial period. I’ll examine three interlocking quality issues: adequate number of cores, core quality and crossdating. I’ll only discuss core numbers today.

The number of available series cores (assuming the Briffa crossdating for now) drops sharply before the 12th century, falling from 6 in 1100 to 5 before 1075, 4 before 1047 and only 2-3 before 1047. There are only 3 series in 1032 (which have crossdating problems to be discussed.) Thus, the proof that 1032 is the "coldest year" of the millennium in the Polar Urals (and in Jones et al [1998]) depends on the maximum density of 3 (poorly-dated) cores from the Polar Urals!

The criterion usually used by Briffa to determine the adequacy of a "signal" is a Subsample Signal Strength [Wigley, Briffa and Jones, 1984] of at least 0.85. Subsample Signal Strength is a function of the average mean inter-series correlation (rbar)and the number of series. Figure 2 below shows the mean inter-series correlation (by century) and SSS for Polar Urals. In well-replicated modern samples, the rbar is nearly 0.6 and the SSS is nearly 1. The mean interseries correlation (rbar) is lower in the periods with low replication. The SSS falls below Briffa’s standard criterion of 0.85 at almost exactly 1100. In the “coldest” year of the millennium (1032), the Subsample Signal Strength is far below Briffa’s standard. Briffa et al. [1995] provide no reason for disregarding their usual policies.

Urals 2
Figure 2. Solid — SSS; dashed — rbar (by century). All calculations done on MXD using spline standardization (COFECHA 32-year period).

I have a calculation handy showing the impact on the Jones et al[1998] reconstruction of excluding the Polar Urals record prior to 1050 as shown in Figure 3 below (this should be done to 1100, and I’ll update this graphic to show this). Obviously the 11th century is no longer colder than the 20th century. This graphic merely deals with 3 Polar Urals trees and does not consider the additional impact of the Tornetrask "adjustment" on the Jones, Briffa et al. [1998] multi-proxy study. It is surely remarkable that results from densities of 3 tree cores can affect 11th century climate estimates by over 2 degrees in some years. There are further strange features to the Polar Urals dataset which I’ll report on over the next few days.


Figure 3. Jones et al. [1998] NH Temperature reconstructions. Top – As reported; middle; without the Polar Urals series prior to 1050; bottom – difference between the two calculations.


  1. TCO
    Posted Sep 18, 2005 at 11:21 AM | Permalink

    While 3 cores may be a bit outre in the world of statistical surveying, do you really think the answer would change with a largery survey? Surely the odds are well over 50%, that they would not change. Why don’t you advocate a fuller sample. However, if it validates Briffa, you need to be up front about that. This should be about assembling peices of truth…whichever direction they support.

  2. Steve McIntyre
    Posted Sep 18, 2005 at 1:55 PM | Permalink

    You might see my op ed – Bring the Proxies Up to Date! If I were writing it now, I’d emphasize that there are a lot more proxies around than are archived – so it would be bring them up to data and report them!

    As to the Urals, it’s not just that there are only 3. This might be OK if there were long cores. The bigger problem is that they are almost certainly misdated – see Crossdating post.

  3. TCO
    Posted Sep 23, 2005 at 10:25 PM | Permalink

    yeah…I read that one. But don’t be shifting the issues like a slippery Mann worm. number of samples is an independant issue from misdating.

%d bloggers like this: