The two proxies that carry the water in the Trouet NAO reconstruction are the Baker speleothem in Scotland and Esper’s tree ring chronology in Morocco.
The briefest examination of the Scotland speleothem shows that the version used in Trouet et al had been previously adjusted through detrending from the MWP to the present. In the original article (Proctor et al 2000), this is attributed to particularities of the individual stalagmite, but, since only one stalagmite is presented, I don’t see how one can place any confidence on this conclusion. And, if you need to remove the trend from the MWP to the present from your proxy, then I don’t see how you can use this proxy to draw to conclusions on relative MWP-modern levels.
For reference, the following graphic shows the Trouet version:
As Andy Baker explains in his article and in a CA comment, speleo widths are believed to be narrower in warm and wet periods and wider in cold and dry periods (with temperature and precipitation not being independent due to NAO). Based on this, I’ve plotted the original data in an inverted sense (narrower at the top) in the top left. I’ve plotted the full speleothem (Trouet cuts off at AD1075 or so). They transform the data to z-scores even though the data is highly non-normal. The z=score distribution is truncated at an sd of about 1, corresponding to the minimum width of about 0. Also shown are 25-year and 50-year binned averages. As noted before, although Trouet et al say that they used 25-year averages, they actually used 50-year averages.
 |
 |
 |
 |
Figure 1. Versions of SU967 Speleothem.
The bottom left shows the same plot for the “detrended” widths, while the bottom right shows the same plot for the “precipitation reconstruction” which is a re-scaling (linear transformation) of the detrended width series. This is the version that is carried forward into Trouet. The “detrending” (also called “normalisation”) is described in Proctor et al 2000 as follows:
Personally, I find this justification underwhelming. The adjustment leads to important reversals of medieval-modern relationships – all in the direction of enhancing 20th century levels relative to 11th century levels. Even if the adjustment subsequently proves justified (and there are circumstances in which it could come onside), you can’t assign any confidence to any product of this data set without assessing potential errors in the detrending. Unfortunately, Trouet et al neglect this error source:
These individual error terms include (i) dating uncertainties of the speleothem record, (ii) chronology error of the tree-ring record, and (iii) calibration error of the residual record NAO_ms.
Actually the non-detrended Scottish speleothem record reminded me of the Polissar glacier discharge proxies from Venezuela of all places, as shown below (Discussed a couple of years ago at CA), both showing transitions from ‘MCA’ to LIA around 1250-1300. I submit that this is no less “remarkably similar” than any of the Team comparanda. This is not actually inconsistent with the actual language of Trouet et al where a global reorganization of wind fields at the ‘MCA’-LIA transition is posited.
 |
 |
Reference:
Proctor et al 2000: http://www.barlang.hu/pages/science/angol/CD2000_815.pdf
Climate Audit 
51 Comments
Let me see if I’ve got this straight (please assume I didn’t).
Specific aspects of a certain data set were deliberately removed, forming a new data set.
The new data set was used to construct a new model.
The model was said to demonstrate the insignificance of the removed aspect because it did not contain the removed aspect.
What did I misunderstand?
Re: Dishman (#1), there’s a bit more to it than that. Regardless of the detrending, the speleothem as interpreted is evidence of a warm-wet MWP in Scotland. It’s more so without the detrending.
I’m merely noting this problem for now. My greater puzzlement is exactly how Trouet’s interpretation purports to set aside Lamb’s view of the world, since they seem to unconsciously adopt so much of Lamb’s view.
What might cause the ring width to increase with time? A steady change in precipitation/temperature or even the gradual growth of the blanket bog on the surface, which causes waterlogging of the soil and it holds water better. Blanket bogs were not always present so you need to know when they originated. Some started after humans cleared the forest. The effect of bog growth would be a valid thing to detrend, but ppt/temperature would not.
What is your measure of the weight of “water carried” by these two proxies?
Re: bender (#4), the TRouet NAO is the difference between these two series, both scaled in some way that I haven’t decoded yet.
Re: Steve McIntyre (#23),
Ok. I guess the weight couldn’t be any heavier, could it!
I’m still trying to figure out how such a thin reed can support the enormous logical structure built atop it.
RE Steve #3,
Isn’t their point (rightly or wrongly) that the MWP in the N Atlantic was offset by equal and opposite cooling elsewhere, so that the net effect on the global avaerage must have been 0?
Of course Craig Loehle’s reconstruction strongly indicates that the MWP does show up in the global avaerage.
Re: Hu McCulloch (#6),
If that’s their point, then where is their proof? They need to *show* “cooling elsewhere”, not assume it. NAO is the pea under the thimble this time. Prove half of it (the warm half) persisted through MWP, assume the rest of it (the cool half) pesrsisted also.
i.e. No offsetting cooling “elsewhere”.
So, if I’m reading this right. It has to be either warm and wet or cold and dry? No other combination occurs? I know this is the NAO and Scotland, but coastal areas can vary greatly, and going inland just a few miles can also change things dramatically. Here in the PNW it’s generally Warm and Dry or Cold and Wet.
Re: Jeff Alberts (#7),
If warm+dry and cold+wet register ambiguously in that kind of proxy then they are simply not reconstructable. In general, the independent components of heat and moisture can not be deconvoluted from any kind of proxy that is sensitive to the interaction between heat and moisture. That is why it is such a significant observation that most tree species are sensitive to drought, the interaction of heat and dry. The only species-site combinations eligible for temperature reconstruction are the ones where temperature is far more limiting to growth than moisture. Hence the bias toward alpine/treeline tree-ring based proxies. Problem: If those alpine tree species ever become more limited by moisture than tenmperature then the proxy loses accuracy during that period. Problem: was MWP one of those periods? Hence the extreme interest in imprecisely dated physical proxies as opposed to precisely dated but flaky biotic ones.
I assume the authors have never been to Scotland? Its wet even when cold and dry! Its never warm even when wet and warm!
Re: Andy (#9), Relatively wet/dry or hot/cold to what it “normally” is.
The spikes of stalagmite growth at 1450 and 1620 look to be of suspiciously short duration for their amplitude compared to the rest of the record. Although it doesn’t speak to the detrending issue directly, one has to wonder if the record is reliable enough to provide climatic information. Is the degree to which CaCO3 was limiting known? If not, it adds another uncertain variable to the mix.
One observation means infinite standard error means the data inspires zero confidence. What sorry reviewer let this sad fact slip through and what shall be his punishment?
I think Ed Cook and Rob Wilson would be disgusted to hear of such an abuse of detrending and detrended data.
#12 bender
I realize that it was a rhetorical question , but there are two answers to what kind of reviewer?
1. A peer reviewer of course as in a member of a group of equal stature and ability (which could obviously be a problem)
2. Only the finest as in Home Alone 2
Re: Thom Scrutchin (#14),
Not at all. I’m inviting the reviewer to self-identify and to explain.
Proctor, et al., detrended by fitting with a 2nd order polynomial, and subtracting the polynomial from the data. This is a curved fit, the effect of which was to greatly reduce the magnitude of the MWP relative intensity.
You can see this effect by comparing the region of year 1000 in the upper and lower leftmost plots, in the first set of four. Note in the lower plot how the down-spike at year 1000 is become at least as negative as the down-spike in year 1800. This stretch is due to the downward concavity of the 2nd order polynomial fit.
So, they didn’t just detrend. With a 2nd order polynomial normalization, they also did a non-linear intensity re-scaling across the data.
Re: Pat Frank (#15),
Thank you. I was just about to ask whether or not this was linear detrending or not. Note that when tree stands are detrended using such fits, there is a physical reason for the detrending (growth patterns of trees). Also even if there are errors in such detrending, there are several to many trees in a stand and the errors will be washed out by overlap (a number of trees present at any one time period and in various stages of growth.)
But when we have one speleothem, and no obvious physical reason (AFAIK) for the detrending, it wouldn’t seem likely that the result is of any value for comparing temperatures over long periods of time.
Re: Dave Dardinger (#18), While I understand your point about the tree-ring detrending using multiple trees as being superior to this detrending, I have long been bothered by it as well.
From what I have read on the subject, tree-ring detrending also involves a curve fit (often exponential), followed by subtracting out the fit curve, leaving the residuals for further analysis. I have never figured out how this can distinguish “standard” trends in tree growth from long-term variation in whatever environmental conditions (precipitation, temperature, etc.) affect the tree-ring parameters. I’ve done some searching, without any success. Can anyone point me to a reference that addresses these concerns?
Re: Curt (#25),
One difference: AFAIK, tree-ring detrending is per-tree, based on the idea that a young tree gains radius more quickly than an old tree. So with trees of varying age, the “starting point” may be anywhere in time.
ONE stalagmite strikes me as more of an anecdote than data.
Silly question: if wide==warm and narrow==cool, why plot upside down? Or is that a typo?
Steve: My bad. The hypothesis is narrower in warm and wet and wider in cold and dry: hence the inverted orientation. I’ve corrected the language in my sentence above. Thx.
Human development goes back at least as far as the 1500s in the Inchnadamph area (~3m from the cave), Ardvreck Castle dating from 1597. This might explain the wide deviations. Plenty of human activity, since it is close to a large freshwater loch.
Some things of possible interest.
This is background which illustrates some of what has gone on, by whom and so on, is not in isolation.
Click to access ASCRIBEpresentation.pdf
The following papers might be of interest since it is about validation. The above hints at why an Irish connection. They were looking at the whole Atlantic seaboard, not more inland such as continental climate. Far north west Scotland is not representative of for example England which is more affected by continental weather. Even to the west, compare say river flow records between the Wye and Thames, not so far apart in catchment, very different. (Wallingford mentioned later, is of the government hydrology station there, is in the Thames valley)
Click to access hessd-5-547-2008-print.pdf
or
Click to access hess-12-1065-2008.pdf
Now I see a spot of bother. Why would it be they can’t get a match with Valentia also maritime, so switch to non-maritime, very different location to make a to me dodgy looking validation?
Why this particular stalagmite, an outlier, the only one like that and with admitted water storage, such as where, doing what?
Scotland has changed a lot over the years and today is misleading, seeing it after depopulation.
Above is a personal opinion. I am not an expert on the above subject.
Re: Tim Channon (#21),
In browsing the first URL, I found no graph linking temperature to any other measurement derived from a speleothem. One would have though that a temperature/other quantity graph would be fig 1. One supposes that by now a myriad of conveniences have been found to correlate with temperature and it is no longer required to show a temperature relationship when presenting a temperature paper; unmentioned literature suffices in the manner of perceived wisom. Or perhaps I confuse a convenience with a bog.
Re: Geoff Sherrington (#27),
My apologies, I did not wish to be read as ‘piling on’.
A quick question.
Why is one speleothem different from one Stevenson screen? You wouldn’t put all your faith in the results from the latter so why would you do so for the former?
Re: detrending of Scottish stalagmite?
Apologies if there is any confusion around detrending (or otherwise) and thank you for the interest in this work. I’ll try to explain via the history of the research. Back in the late 1990s, Chris Proctor and I produced the stalagmite ring width series under discussion here. And in the 2000 Climate Dynamics article we presented both the raw and detrended data, and used the detrended data to reconstruct precipitation. As many have commented, why? Well, as caves and fissures in limestone evolve through time then theoretically they will widen due to dilution of the limestone over time. So we did wonder if the trend that we were seeing was due to an increase in water flow rate through time that was non-climatic.
We went on to develop longer stalagmite series from the cave in a follow up paper in 2002 (also in Climate Dynamics, and I sent Steve a copy on request a few days ago and I can happily send pdfs on e-mail request). This longer record showed that stalagmite ring width had been just as fast several times previously, and showed that our previous detrending wasn’t sensible. So in that paper we didn’t detrend, mentioned this briefly within the paper (admittedly confusingly using the word ‘normalised’ rather than ‘detrended’) and lodged the extended data series and the metadata at the WDC for Paleoclimate too (at ftp://ftp.ncdc.noaa.gov/pub/data/paleo/speleothem/scotland/assynt96.txt). It is this series that is used by Valerie Trouet and co-authors. I’ll ask the WDC to update the page to make sure there is a reference to the fact that this is the data used in the Trouet et al paper.
So, back to the questions raised here – Trouet et al use the raw data (not detrended) in the paper (the use of raw data is stated, although admittedly briefly, at the start of the SI). So the plot shown in this post (both Figure 1 (top left) and last Figure, left) is that of the data used by Trouet et al.
Whilst mailing, a couple of other comments which I will hope answer some other questions raised.
The bog history was published by Charman et al 2001 in Quaternary Research, so we have good evidence that this is about as an undisturbed site that you could possibly have in the United Kingdom with respect to human influence, and we know that the bog significantly predates the time period considered here.
Yes, one sample is not ideal, of course, but there are practical and conservation issues. You just can’t sample stalagmites in the quantities one can obtain tree ring samples. For example, the site in question is a Site of Special Scientific Interest and in a National Nature Reserve and we had very limited sample permission. So yes, I’d love to be able to have replicates, but in this case it has not been possible. But, you can get paleoclimate data from one sample, but to be confident it means that you really do have to understand how the climate signal is transferred to the individual sample.
Charman, D.J., Caseldine, C.J., Baker, A., Gearey, B. and Hatton, J., 2001. Palaeohydrological records from peat profiles and speleothems in Sutherland, NW Scotland. Quaternary Research, 55, 223-234.
Proctor, C.J., Baker, A. and Barnes, W.L., 2002. A three thousand year record of north Atlantic climate. Climate Dynamics, 19, 449-454.
Re: Andy Baker (#29),
Andy, thanks for the on point comments. I definitely cannot verify the following comment. You say:
You also say that you
Here is an excerpt from Trouet showing the version used in the article. Notice that the 20th century upnotch is above the zero reference.
If you squint and compare to the corresponding graphic for the raw data (top right) – which matches the assynt96 version – for this case, the 20th century upnotch is below the zero reference, eliminating this version as a candidate for the provenance of the Trouet version. On the other hand, the upnotch in the detrended version matches the upnotch in the Trouet version.
Perhaps you could verify your understanding with Trouet et al. before asking WDCP to make any changes. Obviously, this sort of guesswork would be eliminated with proper digital citations. In your case, you had archived your data and one at least has a chance of trying to assess matters. Other Trouet graphics refer to articles with no digital information, leaving a total mess.
You might also try to verify whether the data has been binned in 25-year intervals, as stated by Trouet, as it seems evident to me that it’s been binned in 50-year intervals.
In addition, I received Proctor et al 2002 from Andy (thanks); I determined that the SU967 data in P2002 matched the “raw” version in P2000; however, Trouet et al said that they used Proctor et al 2000 and the analyses described above showed to my satisfaction that they used the detrended data from Proctor et al 2000. In my reading of Proctor et al 2002, I hadn’t noticed their observation (mentioned above by Andy) that they had determined that their previous detrending wasn’t justified, or else I would have included this observation in my criticism of the Trouet data. I appreciate Andy’s bringing his to our attention.
Also to be clear for readers, Andy Baker and colleagues have provided a commendably complete archive of measurements – the criticism of the non-archiving of the Esper tree ring data does not extend to them.
Re: Steve McIntyre (#30),
the results in the above graphic can be verified through the following script:
Re: Andy Baker (#29),
A post from a scientist with a very readable and comprehensive explanation of the details in question and showing enough confidence to retreat from a previous conjecture. Good stuff.
Re: Andy Baker (#29),
This is a lame excuse. The hard reality is this: one sample is not enough for calibration & reconstruction. Period. Get two and then we’ll talk about confidence levels. My apologies, but this is really a cheap dodge.
Re: bender (#36),
Why ?
Re: UC (#37), I could see bender’s point if it was about ensuring the robustness of the relationship implied in one sample, but he seems to be suggesting that instead of needing more samples to test your hypotheses about the first one, you need more than one to formulate your hypotheses. I’m perplexed by this to. But I will wait for bender to clarify before I pass judgement, however.
Re: UC (#37),
I’ve already explained why in #12. How do you proceed when you have infinite standard error and hence zero confidence? You proceed by publishing the anecdatum (great word) and not moving one iota further until you have another anecdatum. With two, you can start to hypothesize, with the shakiest of calibrations and the most speculative of reconstructions.
.
Is a hypothesis sufficient to support trillion-dollar policy decisions? Wow, great question. But rather than ask captain lecture pants, let’s ask Ross McKitrick what he thinks is a wise precautionary approach. Whoops, he’s already published on the subject. Why not read what he has already written?
Re: bender (#39), I tend to agree with bender here. Sure you can do the reconstruction, and you can even write a paper about it, but what does it matter without real confidence intervals? Does it mean anything? All you can do is shrug your shoulders. That’s not to imply that Andrew did anything wrong by not getting more samples, I believe him when he said that it was not allowed. Under the circumstances though, I’m not sure it was worth the effort for just one sample.
Re: Paul Penrose (#40), You mean Andy. I’m Andrew. We are totally unrelated.
Re: Paul Penrose (#40), From Andy at 29:
I’d have thought that there should be a pretty strong “Special Scientific Interest” in getting this right. Maybe if the request is made for a second sample it would be granted? That way the value of the first should also increase – the info. gained from it would not be negated by getting a second sample.
Re: bender at 36 – how much further would/could a second sample take things?
Re: bender (#39),
Univariate calibration = no CI ?
Re: UC (#43),
UC, there’s sampling error contained in the proxy itself. That’s precisely why Andy admitted:
I would like to thank Andy Baker for his detailed explanation. This is how I think the scientific inquiry should go without name calling and finger pointing that seems to happen often, more so from AGW side then on mine but we (including me since steve has snipped me a couple times) are guilty as well.
While I appreciate Andy’s comments, I disagree that the explanation as it stands is a “comprehensive” explanation of the details. As I observed in my notes above, my own calculations indicate that Trouet used the detrended data and NOT the raw data. Until this is resolved – and no offence is intended here – the explanation doesn’t meet my standards.
Re: Steve McIntyre (#34),
I knew that and do not want to confuse intentions with deeds. Let us see what the next round brings – if there is a next round.
The important question for this thread is what Steve M noted above and I was hoping that by being nice we could get a second reply from Andy Baker and let the auditor do his work.
Re: Kenneth Fritsch (#45),
Quite so.
Re: Steve McIntyre (#34),
No offense should ever be taken. If you accept the principle of “trust, but verify” you should have no qualms releasing all code and data. Turnkey would be nice.
Discussion points re: the value of additional samples:
Click to access baker.pdf
Unfortunately the last page is pixelated.
Re: curious (#47),
Good grief. And y’all thought tree rings were a nightmare.
Andrew,
Yes, I did mean to say Andy. Sorry for the confusion.
On April 9, 2009, notes were added to all the Scottish speleo data sets. The earliest data set ftp://ftp.ncdc.noaa.gov/pub/data/paleo/speleothem/scotland/scotland_data.txt had the following note added:
Given that the ring widths are bounded below by zero and occasionally spike high (when the axis is inverted so that wider is up), a log transform would probably give a more meaningful series.
But removing a time trend, whether quadratic or just linear, would seem to defeat the whole idea of a paleo proxy, which is supposed to have a uniformitarian relation to what it’s trying to measure.