My Photo
Blog powered by Typepad

December 2018

Sun Mon Tue Wed Thu Fri Sat
            1
2 3 4 5 6 7 8
9 10 11 12 13 14 15
16 17 18 19 20 21 22
23 24 25 26 27 28 29
30 31          

« Phil Jones on Phil Jones | Main | Tiger Woods drives car into tree, is taken to hospital »

November 26, 2009

Comments

jj mollo

You can visually inspect adjacent tree stumps to identify some correspondences and make conclusions which seem reasonable, but tree ring evaluation is actually a very tricky area. Using them as proxies for temperature is even more tricky.

For one thing, the relationship between temperature and ring width is not linear. It's more like the Laffer curve. If the temperature is below a certain minimum, the tree will not grow at all. If the temperature is above a certain maximum, the tree will not grow at all. These curves are different for different species. This obviously represents the extreme case, but, as with the Laffer curve, it illustrates the non-linearity problem.

To compound this complexity, there are multiple dimensions. Precipitation has its own Laffer curve, which might be interdependent with temperature. Solar irradiance could conceivably be another factor. More distressingly, CO2 itself could be a factor, making the data somewhat self-generating.

Patterns of ring growth, first of all, have to be established for a particular area. An annoying feature of the data is that ring growth changes normally depending on the age of the tree. This has to be corrected by making the assumption that older trees had the same life cycle as trees today, despite all the other changes possibly effecting the health of trees. The sites where you can get a good longitudinal sample of overlapping tree lives are limited and not necessarily representative. The sample is biased because areas where people live are less accessible to trees and it's hard to find places where the same species has grown for a long time under generally the same conditions.

All of the vagaries of the data are subject to statistical correction, but that opens you up to issues of statistical competence and quality of judgment. It's very easy to argue that any particular approach is suspect, and there is good reason to do so.

One thing you can do is find places where trees are at the margin, under extreme stress because of the cold, but probably not impacted by changes in precipitation, whatever. For instance, at the tree line on mountains or approaching the tundra. In places where we have records of the tree line, we could use that as a proxy for temperature, and we would expect the tree rings to provide more exaggerated results in those places. It still doesn't do anything to help you with the sampling problems, though. If the Sierra Nevadas and Novosibirsk are the only places where you can get decent data, then it's pretty hard to justify generalizing your results to a global level. I wouldn't want to do it, but you've got to start somewhere.

I commend you for going to the ultimate sources, by the way -- the best choice from each side, I think. You keep showing us new facets of your talent, Frank.

Neo

Meanwhile back at the White House we find out that

Barack Obama’s radical socialist climate czar Carol Browner on Wednesday rejected claims that e-mails stolen from a British university show climate scientists trumped up global warming numbers, saying she considers the science settled.
.. but in one of those “shades of Haliburton” moments, we find out that she …
was a board member of one of the leading carbon offset trading companies, APX.
Neo

Here is the latest wonder of "peer review" .. the lastest Mann paper has some of the data inverted. Seems they used the X-ray density data instead of temperature. The proxy relation is that it should have been inverted, but Mann and his co-authors and the "peer reviewer" let it go by without inverting it before using it as a temperature proxy.

jj mollo

Mann is quoted in his defense as below in a number of places, but mostly as fuel for ridicule.

“Multivariate regression methods are insensitive to the sign of predictors. Screening, when used, employed one-sided tests only when a definite sign could be a priori reasoned on physical grounds.”

Having used regression analysis pretty extensively myself, I am puzzled as to where this quote goes astray. Terms are added to a multivariate regression with other proxies already in place. The new term may be used to moderate the previous over-response by another proxy. Moreover, the regression analysis, itself, will simply reverse the sign if that's what is called for. Also, in the face of no prior information, a two-sided test is called for.

Is it possible that the critics are faulting Mann for not starting with the same bias that they start with?

Frank Warner

Here's a basic question. How much of Mann's hockeystick chart depends on tree-ring data? Everything from 1000 to 1960? Or just some of those years? Or does tree-ring data overlap with other data? My impression is the tree-ring stuff is from Keith Briffa. Anyone know if all those years are tree-ring temperatures?

Mark

I did read where Dr. Keith Briffa wrote that by using tree data "preserved in old buildings or naturally in river gravels, peat bogs or lakes, enables chronologies to be extended backwards, sometimes for thousands of years, even though the individual trees that make up the chronology may have lived for only a few hundred years."

Research "has produced continuous temperature-representative chronologies stretching across the whole of the last 2000 years. Sub-fossil wood preserved in lakes in these regions (northern Eurasia) will eventually allow us to extend these series to more than 7000 years."

George

Mann et al. used a "multiproxy" technique that combined a variety of proxies. The most numerous, and influential, proxies in their data set are tree ring chronologies. And in the Mann's 1998 paper that originally gave us the hockey stick, it was found that a tree ring chronology from a stand of bristlecone pines at Sheep Mountain, California was the most influential in creating the hockey stick shape. A test found that, if these particular tree rings are removed, Mann's hockey stick is gone.

Mann had even performed this test himself. The results were discovered in folder called "CENSORED" on Mann's FTP site.

It also was evident that Mann's paper passed "peer review" without any significant level of checking and no attempt to reproduce his work. A list of discrepancies provided by McIntyre and McKitrick to Nature, the original publisher of the paper, led the publisher to order a Corrigendum (corrections) from Mann.

jj mollo

I find myself at a loss when trying to evaluate all the back-and-forth in the tree ring controversy. Let me just talk a little about something I do know about.

Least squares regression analysis (as well as many associated methods) is just a mathematical process. It is a mechanism for searching through data, often large quantities of data, looking for patterns of correspondence among multi-dimensional observations. What matters most is whether the data represents legitimate sampling protocols. While regression analysis tends to be relatively forgiving, and it can actually be used to correct a lot of deficiencies in the data, it makes certain assumptions.

One thing you have to be careful about, for instance, is data snooping. If you build your model based on a particular set of data, most likely you should be evaluating it on another set of data. This is very important when building such complex models as climatologists work with.

Another assumption is that the experimenter is not tweaking the numbers. I read an old article in Science, for instance, by David Pilbeam and Stephen J. Gould, of all people, where a simple regression line was drawn over three points involving sizes of bones. The R-square was reported as .99, or somesuch thing, which attracted my attention as being a little too close to perfection. Well, it turned out that they had actually estimated the value for the middle observation, a fact that they reported in the article. One couldn't help but suspect that their estimate had been informed by the data itself.

In other words, the article was substantially based on a breath-takingly stupid mistake. And it passed peer-review on the basis of the authors' reputations, I imagine. I'm sure that this mistake was entirely innocent, but it provides a demonstration of just how inept people are, even world famous scientists, with statistical concepts.

Now, one feature of regression analysis that most people might not realize is that most practical results are heavily influenced by a relatively small number of points. The bulk of the data points are middling sorts of numbers within the structure of the equation, but a few key points have an outsize leverage on the result. If I were to take those points and deliberately tweak them, I could have just about any result that I desired. But please notice, this is a post facto process. The iterative exploratory multiple regression approach can be seen, in a way, mostly as a search for these critical points. The assumption is that all of the points are equally representative, but that some of the points tell us more than other points.

From a AGW point of view, I don't know whether McIntyre or Briffa are entirely justified in their analyses of the tree ring data. I do know, however, that it is not at all surprising that 12 data points or 17 data points, whatever, when removed could change the result dramatically. If McIntyre used regression analysis to find such critical points, then I suppose he is right to be surprised if all of the high leverage points were in one segment of the data, but he shouldn't be surprised at all that some such points exist.

George

I find Briffa's response misleading and possibly disengenuous.

1) He claims he was merely using the data provided by Russians, Rashit Hantemirov and Stepan Shiyatov. In the Russian version of the data, Yamal had little by way of a twentieth century trend. Strangely though, Briffa's version, which had made it into print before even the Russians', was different. By the way, it took Briffa nine years to release his data such that we could finally understand why it was different.

2) Briffa claims that he does not understand the "basis for McIntyre's selection" of tree rings and wonders why McIntyre excluded his 12 tree rings as if McIntyre is up to something underhanded here. The basis was simple and very clear. McIntyre chose the 18 tree rings that Briffa omitted for unexplained reasons. There was no reason to include Briffa's 12 although they were reexamined for comparison to the other 18.

3) Briffa actually seems to suggest that he selects tree rings based upon their data. He initially says, "We do not select tree-core samples based on comparison with climate data." However, he goes on to say, "Chronologies are constructed independently and are subsequently compared with climate data to measure the association and quantify the reliability of using the tree-ring data as a proxy for temperature variations." You be the judge here. He is saying he checks the rings for "reliability." Were any rings discarded because the were deemed "unreliable"? I can't imagine that he would keep such a ring. He sounds as if he may actually be saying that he cherry picks -- the original problem he is implicitly accused of.

jj mollo

There's nothing wrong with discarding data that are deemed unreliable, as long as it's done according to strict rules that prevent bias. It's also a good idea to examine the discarded data statistically in order to see whether a bias has been inadvertently applied. Cherry picking happens all the time, but is seldom a conscious effort.

jj mollo

Here's a cherry-picking accusation (pdf) from the other side of the AGW debate.

George

I agree that unreliable data should be discarded. And this part of the discussion becomes hypothetical when we don't even know if any data was discarded.

However, to my knowledge, the Russians never indicated any problem with any of the 40 sample sets. Briffa used only 12 without explanation. It is now known that the other 18 (sometimes referred to as "17+1" as one was gathered at a later date) sample sets show the exact opposite trend that Briffa's 12 sample sets showed.

If 18 out of 30 were deemed unreliable, how could you even trust the remaining 12?

Neo

Ben Santer predicts the future ...

I believe that our community should no longer tolerate the behavior of Mr. McIntyre and his cronies. McIntyre has no interest in improving our scientific understanding of the nature and causes of climate change. He has no interest in rational scientific discourse. He deals in the currency of threats and intimidation. We should be able to conduct our scientific research without constant fear of an "audit" by Steven McIntyre; without having to weigh every word we write in every email we send to our scientific colleagues.
1226451442.txt
Kevin

Scientists have counted on tree rings to give them a record of global temperatures, particularly for the 20th century, but also for earlier centuries for which tree samples can be found.

Didn't someone disprove reading tree-rings and chicken-entrails centuries ago?

jj mollo

To my knowledge, it's never been tested with chicken entrails. Etruscan haruspicy using sheep livers, however, has been rolled out and developed into a software product (payor).

George

I have a high confidence that if you take the chicken entrails data and process it with the hard-coded "fudge factor" array found in the leaked Climategate code you will find a hockey stick-shaped curve with the blade in the 1990s.

jj mollo

Here's a tree-ring web-site from University of Arizona.

I also found an explanation for why more recent tree-ring data might be unreliable. For each designated year, it is required to have a systematic sample or a reasonably random sample of whole trees where the year in question is represented at every stage of tree life cycle. The trees normally selected would not have recent years represented from early in the life cycle because the trees normally selected would be the larger trees, those which tend to be preserved best and comparable over the centuries. In other words, to get current data, the sample would have to differ qualitatively from the sample associated with earlier years.

Mark

JJ, you think for the years they did use that they have this systematic sample or a reasonably random sample of whole trees where the year in question is represented at every stage of tree life cycle? From what I read they had very limited "samples". Old wood from buildings. A tree found in a bog? Seems like if they could not get it for recent trees, then older would be much harder.

Dwayne Germaine

The whole premise that tree rings provide a reliable proxy for temperature is just absurd. Several years ago, I was personally involved in studies that used tree-ring thickness as a proxy for COOL weather. With just a bit of searching, one can find hundreds of such studies, based on the premise that tree-ring thickness is most affected by availability of water, meaning the abundance of precipitation, resulting from the cooling of abundant cloud cover.

But both premises are flawed, because the relationship between temperature and ring thickness is not a direct correlation in either direction. The thickest tree rings don't grow in the hottest weather or in the coldest. They grow best in the optimum conditions (doh!), and when you find a ring that's narrower than the thickest, you have absolutely no way of knowing whether that means the next growing season was cooler or warmer than the season that produced the thickest ring. And the same, obviously, holds true for every ring examined. If it's thicker than the next, one cannot say whether than means this year was warmer or cooler than the year before.

jj mollo

Good points all. I don't really know what's been done with tree ring studies. I do know that it's got some serious challenges if you're going to do it correctly, which is not to say that it can't be done correctly.

Addressing Dwayne Germaine's point about optimum conditions, I think some research has been done by focusing on trees that are, or have been in the past, at the extreme limits of conditions where trees can be sustained. For instance, if we look at trees just below the tree line on mountainsides, we can assume that any increase in temperature would be beneficial and would increase the ring size. We can also refer to historical records in some places to find out where the tree line itself occurred in the past and whether it has moved.

Of course, none of that evidence would be decisive, but it would add to the body of evidence. Everything is murky to one degree or another. Nothing is indisputable about this field except that the C02 concentration has inexorably increased over the last 50 years. But I'm convinced that there is enough evidence that we should at least start taking the easy steps, especially when those actions can be supported for other reasons.

For instance, we should be demanding that artificial obstacles that impede the development of nuclear power should be eliminated. And imo, we should move some of our national revenue collection away from taxes that discourage employment, in favor of taxes that discourage consumption of limited or polluting commodities. Petroleum consumption, in particular, has so many negative consequences, political, economic and environmental, that it should be targeted for special attention.

The comments to this entry are closed.