The second image below is, in my opinion, one of the most disquieting images ever presented at Climate Audit.
Two posts ago, I observed that the number of cores used in the most recent portion of the Yamal archive at CRU was implausibly low. There were only 10 cores in 1990 versus 65 cores in 1990 in the Polar Urals archive and 110 cores in the Avam-Taimyr archive. These cores were picked from a larger population – measurements from the larger population remain unavailable.
One post ago, I observed that Briffa had supplemented the Taimyr data set (which had a pronounced 20th century divergence problem) not just with the Sidorova et al 2007 data from Avam referenced in Briffa et al 2008, but with a Schweingruber data set from Balschaya Kamenka (russ124w), also located over 400 km from Taimyr.
Given this precedent, I examined the ITRDB data set for potential measurement data from Yamal that could be used to supplement the obviously deficient recent portion of the CRU archive (along the lines of Brifffa’s supplementing the Taimyr data set.) Hantemirov and Shiyatov 2002 describe the Yamal location as follows:
The systematic collection of subfossil wood samples was begun, in 1982, in the basins of the Khadytayakha, Yadayakhodyyakha and Tanlovayakha rivers in southern Yamal in the region located between 67°00 and 67°50 N and 68°30 and 71°00 E (Figure 1). These rivers flow from the north to the south; hence, no driftwood can be brought from the adjacent southern territories At the present time, the upper reaches of these rivers are devoid of trees; larch and spruce-birch-larch thin forests are located mainly in valley bottoms in the middle and lower reaches.
Sure enough, there was a Schweingruber series that fell squarely within the Yamal area – indeed on the first named Khadyta River – russ035w located at 67 12N 69 50Eurl . This data set had 34 cores, nearly 3 times more than the 12 cores selected into the CRU archive. Regardless of the principles for the selection of the 12 CRU cores, one would certainly hope to obtain a similar-looking RCS chronology using the Schweingruber population for living trees in lieu of the selection by CRU (or whoever).
As a sensitivity test, I constructed a variation on the CRU data set, removing the 12 selected cores and replacing them with the 34 cores from the Schweingruber Yamal sample. As shown below, this resulted in a substantial expansion of the data set in the 19th and 20th centuries and a modest decline in the 18th century. (Hantemirov and Shiyatov 2002 had reported a selection of long cores of 200-400 years; while the CRU archive does not appear to be the precisely the same as the unavailable Hantemirov and Shiyatov 2002 archive, it does appear to be related. This pattern of change indicates that the age of the CRU cores is systematically higher than the age of the Schweingruber cores.)
Figure 1. Comparison of core count. Black – variation with Schweingruber instead of CRU; red- archived version with 12 picked cores.
The next graphic compares the RCS chronologies from the two slightly different data sets: red – the RCS chronology calculated from the CRU archive (with the 12 picked cores); black – the RCS chronology calculated using the Schweingruber Yamal sample of living trees instead of the 12 picked trees used in the CRU archive [leaving the rest of the data set unchanged i.e. all the subfossil data prior to the 19th century]. The difference is breathtaking.
Figure 2. A comparison of Yamal RCS chronologies. red – as archived with 12 picked cores; black – including Schweingruber’s Khadyta River, Yamal (russ035w) archive and excluding 12 picked cores. Both smoothed with 21-year gaussian smooth. y-axis is in dimensionless chronology units centered on 1 (as are subsequent graphs (but represent age-adjusted ring width). [Amended Sep 28 6 pm. Replaces url]
Finally, here is another graphic showing the same two RCS chronologies, but adding in an RCS chronology on the merged data set obtained by appending the Schweingruber population to the CRU archive – this time retaining the 12 cores. Unsurprisingly this is in between the other two versions, but most importantly it has no HS.
Figure 3. Also showing merged version up to 1990. (After 1990, there is only the few CRU cores and it tracks the CRU version.) [Amended Sep 28 6 pm. Replaces url ]
I hardly know where to begin in terms of commentary on this difference.
The Yamal chronology has always been an exception to the large-scale “Divergence Problem” that characterizes northern forests. However, using the Schweingruber population instead of the 12 picked cores, this chronology also has a “divergence problem” – not just between ring widths and temperature, but between the two versions.
Perhaps there’s some reason why Schweingruber’s Khadyta River, Yamal larch sample should not be included with the Yamal subfossil data. But given the use of a similar Schweingruber data set in combination with the Taimyr data (in a case where it’s much further away), it’s very hard to think up a valid reason for excluding Khadyta River, while including the Taimyr supplement.
Perhaps the difference between the two versions is related to different aging patterns in the Schweingruber population as compared to the CRU population. The CRU population consists, on average, of older trees than the Schweingruber population. It is highly possible and even probable that the CRU selection is derived from a prior selection of old trees described in Hantemirov and Shiyatov 2002 as follows:
In one approach to constructing a mean chronology, 224 individual series of subfossil larches were selected. These were the longest and most sensitive series, where sensitivity is measured by the magnitude of interannual variability. These data were supplemented by the addition of 17 ring-width series, from 200–400 year old living larches.
The subfossil collection does not have the same bias towards older trees. Perhaps the biased selection of older trees an unintentional bias, when combined with the RCS method. This bias would not have similarly affected the “corridor method” used by Hantemirov and Shiyatov themselves, since this method which did not preserve centennial-scale variability and Hantemirov and Shiyatov would not have been concerned about potential bias introduced by how their cores were selected on a RCS chronology method that they themselves were not using.
Briffa’s own caveats on RCS methodology warn against inhomogeneities, but, notwithstanding these warnings, his initial use of this subset in Briffa 2000 may well have been done without fully thinking through the very limited size and potential unrepresentativeness of the 12 cores. Briffa 2000 presented this chronology in passing and it was never properly published in any journal article. However, as CA readers know, the resulting Yamal chronology with its enormous HS blade was like crack cocaine for paleoclimatologists and got used in virtually every subsequent study, including, most recently, Kaufman et al 2009.
As CA readers also know, until recently, CRU staunchly refused to provide the measurement data used in Briffa’s Yamal reconstruction. Science(mag) acquiesced in this refusal in connection with Osborn and Briffa 2006. While the Yamal chronology was used in a Science article, it originated with Briffa 2000 and Science(mag) took the position that the previous journal (which had a different data policy) had jurisdiction. Briffa used the chronology Briffa et al (Phil Trans B, 2008) and the Phil Trans editors finally seized the nettle, requiring Briffa to archive the data. As noted before, Briffa asked for an extension and, when I checked earlier this year, the Yamal measurement data remained unarchived. A few days ago, I noticed that the Yamal data was finally placed online. With the information finally available, this analysis has only taken a few days. [Update: the Yamal measurement data used in Briffa 2000 proved to be identical to the measurement data used in Hantemirov and Shiyatov 2002, which I had obtained from Hantemirov in 2004, though this had not been stated anywhere. This was the very first look at the Taimyr dataset and it was, in fact, examining the Taimyr procedure that prompted this post.]
If the non-robustness observed here prove out (and I’ve provided a generating script), this will have an important impact on many multiproxy studies that have relied on this study. Studies illustrated in the IPCC AR4 spaghetti graph, Wikipedia spaghetti graph or NAS Panel spaghetti graph (consult them for bibliographic refs) that use the Yamal proxy include: Briffa 2000; Mann and Jones 2003; Jones and Mann 2004; Moberg et al 2005; D’Arrigo et al 2006; Osborn and Briffa 2006; Hegerl et al 2007, plus more recently Briffa et al 2008, Kaufman et al 2009. (Note that spaghetti graph studies not included in the above list all employ strip bark bristlecone pines – some use both.)
Update: Sep 30: Here’s a blow-up of Figure 3 above, from 1850 on. Legend as in Figure 3. The “combined” information is shown to 1990, since post-1990 is, as noted above, limited to the CRU version and, obviously, reverts back to the CRU.