Climate Audit

A few days ago, I became aware that the long-sought Yamal measurement data url had materialized at Briffa’s website – after many years of effort on my part and nearly 10 years after its original use in Briffa (2000).

I am very grateful to the editors of Phil Trans B (Roy Soc) – at long last, a journal editor stood up to CRU, requiring Briffa to archive supporting data. They actually asked Briffa to archive the data last year. He asked for further time. When I looked earlier this year, it was still unarchived. However, when I looked again a few days ago, it had finally been archived (without anyone at CRU having the courtesy to inform me that they had rectified the situation.)

I’m assuming that CA readers are aware that, once the Yamal series got on the street in 2000, it got used like crack cocaine by paleoclimatologists, and of its critical role in many spaghetti graph reconstructions, including, most recently, a critical role in the Kaufman reconstruction.

I’ll also assume that readers are familiar with the difference between the Yamal chronology and the Polar Urals “Update” reconstruction and of the execrable story behind the non-reporting and abandonment of the Polar Urals “Update”. (See various CA posts and the quick review in my 2008 Erice presentation.) The graphic below shows both the similarities and differences: while the two series are similar on a decadal basis, the medieval-modern relationships are reversed with a huge 20th century pulse at Yamal and an imposing MWP at Polar Urals.

Figure 1. Comparison of the (smoothed) Yamal reconstruction (Briffa 2000, Briffa et al 2008, Kaufman et al 2009 and many others) to the Polar Urals Update (Esper et al 2002).

Reconciliation of the disparate Yamal-Polar Urals Update (and other such regional reconciliations) seems to me like the first order of business if multiproxy reconstructions are to advance. As an analyst in Toronto, I can comment on the differences, but am not in a position to resolve them. Such reconciliations are properly the obligation and responsibility of the field scientists involved. Unfortunately, to date, people in the field have not honored this responsibility and, to an outside observer, seem to have done no more than pick the version (Yamal) that suits their bias.

Until now, Briffa’s refusal to archive Yamal measurement data and the acquiescence of journals (including Science) in this obstruction has made it impossible to get even a foothold on the factors governing the differences between the two series. The archive is still seriously inadequate for full statistical analysis, but now I can at least get a foothold and will begin commentary on the issues in a series of posts.

First, to clear a little underbrush. There is one other version of these series that readers may encounter: Hantemirov and Shiyatov archived a Yamal reconstruction at NCDC that has no hockey stick blade whatever. This version was promoted by a commenter (Lucy Skywalker) at Jeff Id’s as being a priori more valid than Briffa’s. Although the Hantemirov and Briffa chronologies have a very different visual appearance (especially the non_HSness of the Hantemirov version), there is an extremely high correlation between the very different looking Hantemirov-Shiyatov and Briffa Yamal chronologies. (If you regress the Briffa recon against the Hantemirov recon for the pre-1800 version, you get a huge r^2 of 0.81). The two series clearly have the same raw material.

However, in my opinion, the issue is considerably more nuanced than simply preferring the Hantemirov chronology. The Hantemirov and Shiyatov chronology adjusts for age (“standardization”) through a “corridor” method, whereas the Briffa chronology uses a “RCS” method to standardize for age. In other studies involving relatively short-lived trees (such as as Yamal), the corridor method has been found to yield very similar results to “conventional” standardization; such methods are also known to remove any centennial-scale variability from the reconstruction. As a result, no conclusions should be drawn with respect to centennial-scale variability from the Hantemirov chronology. No adverse conclusions should be found against the Briffa chronology merely because it differs from the Hantemirov chronology. There are other reasons to be concerned about the Briffa chronology, but these have to be presented and supported.

Yamal Counts

The main focus of the program at Yamal described in Hantemirov and Shiyatov 2002 was the collection of subfossils from river beds. The original article considers cores from throughout the Holocene (this aspect is reminiscent of the long Finnish chronology), observing that the treeline in the Holocene Optimum was well to the north of the present line.

Today, I’m not going to review these findings (I will on another occasion – interested readers can go straight to the original article.)

First, I’m going to review the inventory of fossil and subfossil trees reported in the original article in order to crosscheck this information against the CRU archive.

Hantemirov and Shiyatov report three classes of subfossil tree: surface – 30; alluvial – 1945 and peatland – 196. They report that these yielded 2171 sawn samples [this adds up exactly]. H and S:

Remains of dead trees can be found lying on the surface and tend to be up to a maximum of 750 years old. Within the frame of this research, some 30 of these dead trees have been collected…

by far the most significant source of subfossil wood remains, often trunks in a near-complete state, with bark, roots and large branches, is the material found in alluvial deposits… 1945 samples have been collected from alluvial deposits at the time of writing…

The second important source of subfossil wood is peat deposits…To date, 196 samples
have been recovered from such peat deposits…

a total of 2171 sawn wood samples has been collected.

Later they report the selection of a subsample of these subfossil samples, together with core results from 17 living trees:

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.

They go to observe (and I’ll return to this point in another post):

The data were treated separately for each valley system and tree position transformed into anomalies from their own present-day limit, and, because no river valley could, as yet, supply sufficient samples to cover the whole period, the data were then combined to form a single regional indication of tree-line shifts. This should be considered as a preliminary result because there is some subjectivity in the way the different valley data were expressed.

I mention this point, because the Briffa chronology appears to take a different approach i.e. not considering each river valley separately. More on this on another occasion.

Rob Wilson could have collected 17 cores from living trees in a morning. I presume that these 17 ring-width series from living larches are a sample from a larger program on living larches. This graphic shows the use of over 30 cores from about 600 to 1500 and the use of about 17 cores in the 19th and 20th century (presumably the 17 cores from living trees.) The following graphic (H and S Figure 6) shows the core count by year for the 241 series selected into the H and S chronology:

Figure 2. Series Counts in H and S Chronology 2002

The CRU Archive
While the Yamal measurement archive is a quantum addition to (indeed, the first useful) information on Yamal, it lacks any sort of metadata as to where the individual samples were obtained.

I think that some information can be gleaned from the nomenclature of the ID numbers. There are 252 distinct series in the CRU archive. There are 12 IDs consisting of a 3-letter prefix, a 2-digit tree # and 1-digit core#. All 12 end in 1988 or later and presumably come from the living tree samples. The nomenclature of these core IDs url (POR01…POR11; YAD04…YAD12; JAH14…JAH16 – excluding the last digit of the ID here as it is a core #) suggests to me that there were at least 11 POR cores, 12 YAD cores and 16 JAH cores. The JAH and POR cores (7 in total) were over 200 years, while all of the YAD cores were under 200 years – so there may be some difference between the Briffa selection and the Hantemirov selection.) YAD presumably stands for Yadayakhodyyakha River (see map and Figure 3); POR for Porzayakha River (see Figure 3); JAH for one of the unlabelled tributaries in Figure 3.

There are 235 IDs consisting of a 1-digit alphanumeric followed by a 4-digit tree number and a 1-digit core #. The tree numbers appear to be unique within this group – the lowest is 0008 and the highest is 2258. There are three different prefixes: L, P and _, with L having by far the most. Perhaps these prefixes correspond to three different subfossil provenances: alluvial, peat and surface. (I don’t rely on this; they might relate to something geographic.)

There are 5 IDs that are 4-5 digits (X13, X02S,M021,M022,M331): these in 1963, 1978 and 1982 and look like they are results from previous coring of living trees. Perhaps these 5 cores plus the 12 cores from living trees (with 6-digit IDs) are the same as the 17 cores from living trees selected in the H and S chronology. But maybe this is a coincidence. One never knows – it’s climate science.

In any event, the following graphic shows the number of cores by year in the CRU archive. The counts decline from 24 in 1956 to only 10 in 1990 and 5 in 1995-96. While the appearance has much in common with the corresponding H and S graphic, there are noticeable differences: the number of series in the MWP is dramatically reduced in the CRU sample – counts were consistently in the 30-40 range in the H and S graphic between 800 and 1400, while the CRU count is more like 15-20. On the other hand, the CRU count for the first part of the 20th century is visually higher than the corresponding H and S count (which holds at about 17.) What accounts for this difference? I have no idea.

Figure 2. Count of archived cores for Yamal at CRU by year.

Further confusing matters is the information on the number of cores in each year attached to the Hantemirov Yamal chronology at NCDC, which has an entirely different appearance in some respects, but other features seem to match. The count is exactly 17 from 1875 to 1963, which seems to match the 17 living trees. The count in 1990 is 10 (the same as the CRU count, dropping to 5 in 1995-1996, as CRU.) Unlike the CRU count, the count is much more homogeneous – so the data sets are definitely not the same.

url, illustrated below:

Figure 4. Counts from Hantemirov 2002 (NCDC)

I provide this data not to provoke reactions like – How can anyone conclude anything about Yamal climate from only 10 cores in 1990? Or 5 cores in 1995? (While I understand the sentiment, please don’t post such comments.) Especially the last few years, where the number of available cores falls below minimums advocated elsewhere in the Briffa corpus.

To show the tininess of the subset representing Yamal in the 1990s, the next graphic compares Yamal counts with counts from the Polar Urals version, set aside by Briffa in favor of the Yamal series (and even I’m a little shocked by the discrepancy). In 1990, there were 57 cores in the Polar Urals and only 10(!) in the Yamal subset.

So while CRU may have archived the data that they “used”, we run once again into a problem that we run into over and over again in paleoclimate studies. What about the data that was in their sample, but which they didn’t use? 10 trees going to 1990 and 5 trees to 1995-1996 is impossibly small for a dendro expedition. Rob Wilson could do that in an hour. What about the rest of the Yamal data? Where is it?

To what extent is the Yamal HS a product of the selection process and to what extent is it climatic? Without the complete data set, it is impossible to set aside the troubling thoughts that one is faced with in these circumstances.

I think that I’ve figured out a way to crosscheck the modern portion (yielding some disquieting results) and will discuss that in a forthcoming post.