Obviously, there has been considerable controversy over the past few days over the Yamal data.
First, let’s observe the continued silence of field dendros on the dispute. None have stepped forward so far to support Briffa’s use of 10 cores in 1990 (and 5 in 1995). As others have observed, their silence is rapidly becoming loud. And it’s not as though Climate Audit is so well loved in dendro circles that dendros would not be willing to speak out against it.
The graphic below shows the disparity between core populations at Yamal and the other two sites in Briffa et al 2008. In 1990, both Avam-Taimyr and Tornetrask-Finland had 105 and 113 respectively (51 and 34 in 1995), whereas Yamal had only 10 cores in 1990 and 5 in 1995 (!)
Despite the lack of endorsement from field dendros, this hasn’t prevented Gavin Schmidt from opining that my observations are incorrect. However, in fairness, in one instance, Gavin included the caveat:
“I’m not a tree ring person, so my opinion on this is perhaps not worth much”.
On this point, Gavin and I can find some common ground. In the absence of any statement from dendros, including Briffa’s close associates at CRU, Gavin seized on an observation from CA and realclimate reader Tom P, who now seems to have become Gavin’s tree ring guru and authority on tree ring homogenization procedures, Gavin endorsing Tom P’s calculations as follows:
“Tom P. above showed that the Yamal curve was robust to homogenising the age structure”
I don’t know how much due diligence Gavin did on Tom P’s knowledge of dendro procedures to determine whether his opinion on these matters was also “perhaps worth not much”. We’ve known Tom P at Climate Audit for a few days and it is my understanding that until a few days ago, he was completely unfamiliar with dendro issues – in other words, well qualified to act as Gavin’s guru and mentor.
Tom’s own contributions to the discussion here have been Monty Python-esque. One could imagine John Cleese playing Tom P in a skit.
First, Tom P seized on the “excellent correlation” between two identical time series – the subfossil portion of the Yamal data set used in sensitivity studies with two different modern samples, one displaced microscopically below the other. Tom gushed that this increased “confidence” in both of them:
The excellent correlation between the two Yamal datasets strengthens confidence in both before 1850.
Next, Tom found once again that identical series have remarkable similarity. This time, Tom compared the post-1990 portion of a “combined” data set in which the only contribution came from 5-10 Briffa cores to the Briffa data set in which the post-1990 contribution came from the same 5-10 Briffa cores. The two were virtually identical!! Tom proclaimed the robustness of the Briffa data set once again.
Gavin had found a soulmate.
Then, Tom proclaimed that the Schweingruber series were no good:
Rejecting the Schweingruber series as a good proxy seems reasonable
While Gavin may agree with his mentor on this matter, I suspect that Briffa’s endorsement will not come quite so quickly. Y’see, Briffa has pretty much built his career on analysis of the Schweingruber network and the Khadyta River, Yamal (russ035) site is used as a proxy in virtually all of Briffa’s publications. The CRU website entitled “Keith Briffa & Tim Osborn: Tree-ring data lists a series of Briffa’s publications relying on the Schweingruber network disdained by Gavin’s new guru: Briffa et al (Nature 1998a); Briffa et al (Nature 1998b); Briffa (JGR 2001); Briffa (Holocene 2002) and Rutherford et al (2005), the last being a joint venture between the Jones-Briffa team and the Bradley-Mann team. The gridded version of the Schweingruber network developed in Rutherford et al (2005) was used in Mann et al 2008.
So I don’t expect Briffa to immediately join Tom P and Gavin in condemning the Schweingruber series as not being a “good proxy”. My guess is that his response, whatever it is, will take a different approach. Just a guess.
The next idea of Gavin’s guru was that joining Schweingruber’s Yamal data with the Yamal data with the Russians was an invalid comparison and was “comparing a signal to noise”:
The Schweingruber series is therefore of very limited utility for a valid comparison with the much longer-lived trees of the CRU archive. Your earlier sensitivity test is comparing a signal to noise…
All you have done is inject noise into the Biffra/H&S series by adding in much shorter lived trees. This also explains why the Schweingruber series did not well correlate with the instrumental temperature…
It’s the duration of the tree cores that’s important to extract a long term signal. The CRU archive during the overlapping period with the Schweingruber series has much older trees in it, as you have already pointed out. Your sensitivity analysis replaces longer- with much shorter-lived tree cores and hence obscures the longer-term signal.
On Sep 29, Tom P observed that he was “patient enough to let Steve to plot his own data when he’s ready”. Awfully generous of him. However, Tom did not live up to this undertaking. A couple of days later on Oct 2, not resting on the laurels of his two previous proofs that identical series had “excellent correlation”, Tom suggested that I carry out a sensitivity analysis only using “trees with ages above a certain value”.
Of course this begs the question of a sensitivity analysis based on recalculation of the Briffa Yamal plot only using trees with ages above a certain value. It would be very useful to see how sensitive the shape is tree age – we’d see how the snake bends as its bones grow older…
A little later, he asked Roman or I do the analysis, noting that it could probably be done easily with the tools that I had provided online (and indeed it could.)
RomanM and Steve McIntyre, Are either of you willing to do the sensitivity test I propose above? I believe this would be unbiased and potentially publishable work. I’d love to get stuck into R and do this, but I face a steep learning curve and little time. I would guess it wouldn’t take many tweaks to the code to filter on the tree-record field to achieve his.
In the early morning of Oct 3 (3:51 am blog plus two hours), less than 24 hours later, Tom started getting impatient that room service had not delivered his requested sensitivity analysis, while still acknowledging the possibility that the kitchen might be otherwise engaged:
The more I think about this core-age sensitivity test, the less patient I am to see the results, but I sympathise with your time constraints!
Less than 7 hours later, Tom P reported to realclimate that he had “lost his patience” with the totally unacceptable delays from room service and “kludged” my code to the sensitivity analysis that interested him:
3 October 2009 at 10:47 AM
I’m afraid I lost my patience and have kludged Steve McIntyre’s code to do my sensitivity analysis (code is posted on Climate Audit).
No wonder Tom was outraged. After all, it was 24 hours since he put in his order. Next time, I guess Tom will order up his data and sensitivity studies over at CRU. I’m sure that he’ll be pleased at the service.
In making this derogatory remark at realclimate, Tom failed to mention that, despite the fact that he had never previously used R, that the tools and turnkey scripts that I had provided enabled him to carry out the desired sensitivity study with an almost immediate turnaround time. Didn’t even leave a tip for room service.
Indeed, Tom was so dissatisfied by the performance of room service that he published his complaint at realclimate before he published a virtually identical complaint at CA a few minutes later (CA blog time plus 2 hours). “Tom’s” code wasn’t actually posted at CA when he notified realclimate. Nor was it available online for a number of hours as it was caught in the filter until I manually retrieved it later in the day. Nor did “Tom’s” code as posted actually generate his sensitivity study. His code as posted was a copy of my code except for one line that didn’t do anything in the code as placed online.
Tom’s supposedly “demanding” test is set out in this comment , linking to a set of figures, of which a recon using Yamal trees over 75 years is compared to the Briffa chronology. In the latter portion of the graphic, the 10 Briffa trees are all well over 75 years in age so that this portion is identical between the two comparisons, a similarity that Tom, once again, finds remarkable.
Figure 1. Tom’s comparison. Original here.
As noted above, Gavin Schmidt, applying normal Team standards of due diligence to results purporting to contradict me, immediately endorsed the findings of his new tree ring guru at realclimate.
Tom P. above showed that the Yamal curve was robust to homogenising the age structure, and frankly I have a lot more confidence in Keith Briffa to do this right than I have in McIntyre. – gavin]
This finding quickly spread around the internet.
However, while Gavin might well have “more confidence in Keith Briffa” to do this calculation right, the calculation was neither done by Keith Briffa nor endorsed by Keith Briffa. It was, in effect, a calculation that, in the terms of Gavin’s head post, appeared “randomly on the internet” with, as far as I can tell, neither Gavin nor any of his colleagues running the calculation by any practising dendros before endorsing them. Briffa might have given them a different answer, as I’ll discuss below.
The underlying complaint of Gavin’s guru about my inclusion of the Schweingruber data was that it had a younger population distribution than the CRU Ten. Drawing on several hours of experience in internet debate on this matter, Gavin’s guru (GG) opined that the inclusion of this younger data added “noise” to the “signal”. Tom proceeded to further complaints about room service, observing that my comparison was “rather uncooked (to be kind)” and that I had “left behind quite a mess”.
Steve McIntyre’s critique was rather uncooked (to be kind) but he appeared to be in quite a rush to find fault and has left behind a bit of a mess.
We saw above that Briffa might not be quite as quick as Tom and Gavin to trash the Schweingruber network. Nor do I expect Briffa’s eventual line of response to follow this approach. Y’see, Briffa’s own protocols for carrying out RCS standardization explicitly preclude the sort of age truncation proposed here by Gavin’s guru, requiring both a “large population” and “a wide range of different tree ages, each distributed widely through time and all drawn from a single species in a relatively small region”. Here is an excerpt from Briffa et al 2001 url :
I also urge readers to read three other Briffa publications on RCS standardization:
Briffa et al 1996 stated:
Unfortunately the RCS approach is far from being a general panacea for the loss of long-timescale information in dendroclimatology. …
The crucial factor, as so often in dendroclimatology…is replication…. In general, much larger sample replication is necessary to achieve an accurate estimate of long-timescale variability in chronologies …”
Briffa and Cook 2008, among other caveats, stated:
We need a different mind set as regards sampling: sample numbers an order of magnitude greater than the “commonly perceived” need for 15-20 trees should be targeted, even at a single site level.
Briffa and Cook recommend the development of regional networks – including Khadyta River would seem to be precisely the sort of thing they have in mind (and did over at Taimyr):
Regional networks of such well-replicated data should be developed – and if possible from different ecological situations (range of elevations, aspect, substrate type, etc.).
Aside from issues of sample size and age classes, Briffa explicitly warns against exclusive reliance on codominant trees in the modern portion, urging a representative age class distribution condemned by Gavin’s guru. Briffa and Cook state 2008 state clearly:
We should not sample only dominant or co-dominant trees and not sample only the oldest trees.
They go on to observe that such sampling will result in “modern-sample bias” (one manifestation of which is described at length in Briffa and Melvin 2009):
The results may, therefore, be affected by a modern-sample bias, bearing in mind the common practice of sampling dominant or codominant trees (Schweingruber and Briffa 1996).
And yet the CRU Ten (originally selected by the Russians) clearly consist of 5 dominant or codominant trees from two different sites – exactly the sort of truncated network criticized here by Briffa on this second count.
There are many other interesting observations in both Briffa and Cook 2008 and Briffa and Melvin 2009. Given Briffa’s failure to archive his Yamal data for nearly ten years after his original publication, his refusal to provide the data to the authors of D’Arrigo et al 2006 and his serial refusal to provide the data to me, the following statement to the PR Challenge conference is remarkable:
The ITRDB is a great resource. It needs to be continually improved to allow easy storage of other than “usual” tree-ring width data. Improved meta data should be sought for all submissions, including tree dimensions and architecture and information on context of measurements (routinely including estimates of missing rings to pith). When standardised indices are archived, precise details of standardisation options should always accompany them. This should include detailed output from the programs used for standardization, auch as the ARSTAN program. Only in this way can others replicate how standardized tree-ring chronologies were developed.
As I’ve observed to readers in the past, when dealing with arguments from the Team (and, I guess, Team groupies and gurus), you have to watch the pea under the thimble. The Gavin guru graphic presented above shows a chronology for trees older than 75 years, but did not show the corresponding graphic for trees younger than 75 years that would support his assertion that the younger trees contributed nothing more than “noise”.
The two graphics below compares chronologies for old (more than 75 years) and young (less than 75 years) for the combined Yamal and Khadayta River data sets. The first shows them for the period from 1800 to the present. As you see, up to the 1970s, the young tree chronology is, shall we say, remarkably similar to the old tree chronology. Thus, if we stipulate along with Gavin and his guru that the young tree chronology is simply “noise”, a similar characterization also applies to the remarkably similar old tree chronology. The most distinctive feature of this graphic is something quite different than the guru reported to us: given the similarity of the two series up to 1970, their divergence thereafter really is quite remarkable. Readers should not presume that it is the young tree chronology that is the exception. Tomorrow I’ll show this with a graphic comparing these two series to the NSIB temperature reconstruction of Briffa et al 2001.
Here’s the same plot for smoothed versions (21-year smooth) of the two series, again showing coherence up to the 1970s, once again showing remarkable coherence between the two series until the 1970s.
As noted above, while Tom P has an undoubted knack for showing that identical series have excellent correlation, his “sensitivity study” proves nothing relevant to the issues surrounding the Yamal data set, except perhaps that Gavin’s opinion on the matter is “not worth much” and their total lack of due diligence in failing to spend even a few minutes examining whether Tom knew what he was talking about, whether his claims held up or even whether his claims were consistent with policies set out in Briffa’s own publications.