This is the second of our two-part series on the flawed science behind the famous "Hockey Stick" chart of historic global temperatures that forms the basis for claims that the world climate is in the midst of unprecedented warming. In yesterday's first installment, Dutch science journalist Marcel Crok outlined the story of two Canadians researchers, Stephen McIntyre and Ross McKitrick, who found serious flaws in the statistical methods used to construct the chart. The McIntyre/McKitrick critique of the chart, created by U.S. scientist Michael Mann, will be published in February by Geophysical Research Letters, the eminent scientific publication. In today's report, Crok explores the rest of the story of how the Canadian researchers uncovered the serious failures that cast the whole chart into doubt.

- - -

There was yet another important discovery. McIntyre:"When we compared data as used by Mann with original archived data, we found one and only one example where the early values of a series had been extrapolated -- a cedar tree ring series from the Gaspe peninsula in Canada. The extrapolation, from 1404 back to 1400, had the effect of allowing this series to be included in the critical early 15th-century calculations. When we did calculations both including and excluding the series, we found that the difference was considerable. In some cases, the temperature was as much as 0.2 degrees Celsius lower using the modified Gaspe series as compared with the archived version.

"More strangely," said McIntyre, "the series appears twice in Mann's data set, as an individual proxy, and in the North American network. But it is only extrapolated in the first case, where its influence is very strong." McIntyre and McKitrick went back to the source of the Gaspe series and then to the archived data at the World Data Center for Paleoclimatology."We found that although the Gaspe series begins in 1404, up until 1421, it is based on only one tree. Dendrochronologists (tree ring researchers) generally do not use data based on one or two trees. The original authors only used this series from 1600 onwards in their own temperature reconstructions. This series should never have been used in the 15th century, let alone counted twice and extrapolated."

McIntyre and McKitrick submitted a paper to Nature in January, 2004. Mann and his colleagues were invited to respond. McIntyre: "They raised an interesting point. They stated that the critical North American PC1 [a technical term: the first Principal Component (PC1) of a Principal Component Analysis (PCA)] was not just based on the top-weighted Sheep Mountain series, but that 14 other series were also highly weighted in it. In late March, we sent in a second version of the article in which we demonstrated that these 14 tree rings were all from highly controversial bristlecone pine series, studied by Graybill and Idso in 1993, which showed an unusual growth spurt in the 20th century. Graybill and Idso themselves attributed the growth spurt to higher concentrations of CO2 in the air, because they were able to show that it was not caused by increased temperatures. Oddly enough, in their 1999 article, Mann and his colleagues had actually admitted the same thing: 'A number of tree ring series at high altitudes in the western part of the United States seem to show a prolonged growth spurt that is more pronounced than can be explained with the measured increase in temperature in these regions.' "

Now, a number of years later, Mann's defence includes the remark that these same series form the "dominant" part of the Northern American PC1, and accordingly, justifies their inordinate influence on the temperature reconstruction of the entire Northern Hemisphere.

As the story unraveled, more intrigue came to the surface. McIntyre: "On Mann's FTP site, the directory for the North American network contains a subdirectory with the striking name BACKTO_1400-CENSORED. The folder contains PCs that looked like the ones we produced, but it was not clear how they had been calculated. We wondered if the folder had anything to do with the bristlecone pine series: This was a bulls eye. We were able to show that the 14 bristlecone pine series that effectively made up Mann's PC1 (and six others) had been excluded from the PC calculations in the censored folder. Without the bristlecones sites, there were no hockey sticks for Mann's method to mine for, and the results came out like ours. The calculations used in Mann's paper included the controversial bristlecone pine series, which dominate the PC1 and impart the characteristic hockey stick shape to the PC1 and thereafter to the final temperature reconstruction. Mann and his colleagues never reported the results obtained from excluding the bristlecone pines, which were adverse to their claims."

McIntyre finds some irony in Mann's response. " After we published our findings in Energy and Environment, Mann accused us of selectively deleting North American proxy series. Now it appeared that he had results that were exactly the same as ours, stuffed away in a folder labeled CENSORED."

When McIntyre and McKitrick submitted the second version of their article to Nature, they discussed the dubious role of the bristlecone pine series and reported the CENSORED subdirectory." Nature then asked us to shorten our article to a mere 800 words and we did. Months went by and then we were told that they were now only willing to permit us 500 words and the content was too 'technical' to be dealt with in 500 words."

In January, 2005, an adapted version of McIntyre and McKitrick's paper was accepted for publication by Geophysical Research Letters (GRL). Judging by the reactions of the referees of GRL, which McIntyre made available to us, the tide may be turning in the climatology field. One referee stated: "S. McIntyre and R. McKitrick have written a remarkable paper on a subject of great importance. What makes the paper significant is that they show that one of the most important and widely known results of climate analysis, the 'hockey stick' diagram of Mann et al.,was based on a mistake in the application of a mathematical technique known as principal component analysis (PCA)."

The same referee also writes: "McIntyre and McKitrick found a non-standard normalization procedure in the Mann et al. analysis. Their paper describes this procedure; it was an apparently innocent one of normalization, but it had a major effect on their results. The Mann et al. normalization tends to significantly increase the variance of data sets that have the hockey-stick shape. In the Mann et al. data set, this turned out to be bristlecone pines in the western United States. Thus the hockey stick plot, rather than representing a true global average of climate for the past thousand years, at best represented the behavior of climate in the western U.S. during that period.This is an astonishing result. I have looked carefully at the McIntyre and McKitrick analysis, and I am convinced that their work is correct."

The referee ends with: "I urge you not to shy away from this paper because of its potential controversy. The whole field of global warming is currently suffering from the fact that it has become politicized. Science really depends for its success on an open dialogue, with critics on both sides being heard. McIntyre and McKitrick present a cogent analysis of the global warming data. They do not conclude that global warming is not a problem; they don't even conclude that the medieval warm period really was there. All they do is correct the analysis of prior workers, in a way that must ultimately help us in our understanding of past climate, and predictions of future climate. That makes this a very important paper. I strongly urge you to publish it."

Climate researchers can now no longer dismiss McIntyre and McKitrick's efforts with the remark that they didn't publish in an authoritative journal. Mann, Bradley and Hughes, meanwhile, continue to defend themselves quite aggressively. One of the Nature referees noticed this as well: "I am particularly unimpressed by the MBH style of 'shouting louder and longer so they must be right.' "

Mann has obviously decided to defend his graph to the bitter end. Not too long ago, he and his team launched a weblog, www.realclimate.org, in which they strike back very aggressively. Mann's main criticism of McIntyre and McKitrick's previous calculations is that they should have expanded the list of North American PCs from two to five, so that the bristlecone pines in the fourth PC (PC4) could be included.

Not surprisingly, McIntyre is unfazed by the criticism: "Mann claims that his PC1 (essentially the bristlecone pine series) represents a dominant trend in the North American network. Using his incorrect standardization, the PC1 does account for 38% of the NOAMAER [North American] network variance. However, in a correct calculation, the bristlecones are demoted to the PC4 and only account for 8% of the variation. Hardly a dominant trend, like Mann claims. His argument to increase the number of PCs is simply a desperate move to salvage the hockey stick. Look at this from a robustness point of view: Mann has claimed in print that his result is so robust that even removing all his tree ring data will not overturn it. Now all of a sudden, he insists that a single PC4 based on the controversial bristlecone pine data plays the deciding role in the temperature history of the entire Northern Hemisphere."

When we put forward some of the criticism to Mann, Bradley and Hughes in an e-mail, we received an elaborate response within the hour. Apart from the stock arguments that McIntyre and McKitrick are not real scientists, Mann rationalized the presence of the directory BACKTO_1400-CENSORED on his FTP site: "After publication of the first hockeystick in 1998, we ran a number of sensitivity tests to determine if we could come to a reliable reconstruction without having to correct certain tree ring series at high altitudes for non-climatological effects, like the influence of CO2. We reported on this in the publication of 1999."

McIntyre is not satisfied: "In his second publication, Mann mentioned problems with the bristlecone pines, but only with regards to the period of 1000-1399 and not the 15th century that is in this file. More importantly, if you know there are problems with the bristlecone pines, the obvious test would be to eliminate them from the calculation and see what the effect is. This is exactly what Mann did in the directory BACKTO_1400-CENSORED. When he did not like the results, he did not report them and proceeded to include the bristlecone pines in his final analysis."

We asked Mann about the apparent inconsistency between the claimed robustness and the evidence that the shape of his hockey stick relies heavily on the bristlecone pines. Mann responds that he can reach the same results even without doing a PCA, arguing that you could simply use all 95 proxies individually in the calculations: "There is no clearer proof that McIntyre and McKitrick claims are false."

"Mann is a clever debater," McIntyre points out. "That he can produce a hockey stick with another method that also allows the bristlecone pines to dominate is completely irrelevant. The bristlecone pine series are still essential for this new result. When you do the calculation without the bristlecone pines, the result does not resemble a hockey stick in any way."

Mann further argued that he is not the only scientist to have found the hockey stick graph: "Over a dozen other estimates based on proxy data yield basically the same result." That argument is not new to McIntyre.

At this point, McIntyre has growing doubts about the other studies as well. His initial impression is that they are also dubious. It is almost certain, or so he states, that the other studies have not been checked either. McIntyre: "Mann's archiving may be unsatisfactory, but other researchers, including Crowley, Lowery, Briffa, Esper, etc, are even worse. After 25 e-mails requesting data, Crowley advised me that he had misplaced his original data and only had a filtered version of his data. Briffa reported the use of 387 tree ring sites, but has not disclosed the sites. Other researchers haven't archived their data or methods or replied to requests."

"Mann speaks of independent studies, but they are not independent in any usual sense. Most of the studies involve Mann, Jones, Briffa and/or Bradley. Some data sets are used in nearly every study. Bristlecone pine series look like they affect a number of other studies as well and I plan to determine their exact impact. I'm also concerned about how the proxies are selected. There is a distinct possibility that researchers have either purposefully or subconsciously selected series with the hockey stick shape. I'm planning to use simulations to test if the common practice of selecting the so-called "most temperature sensitive" series also yield hockey sticks from red noise."

McIntyre and McKitrick draw far reaching conclusions from their research: "When the IPCC decides to base their policy on such studies, triggering the spending of billions of dollars, there should be more thorough checks. At some point, some one should have done an elementary check on the principal component calculations. This never happened and there is no excuse for this."

Rob van Dorland of the Royal Netherlands Meteorlogical Institute has read the article that will appear in Geophysical Research Letters and is convinced it will seriously damage the image of the IPCC. "For now, I will consider it an isolated incident, but it is strange that the climate reconstruction of Mann has passed both peer review rounds of the IPCC without anyone ever really having checked it. I think this issue will be on the agenda of the next IPCC meeting in Peking this May."

This brings climate research back to square one. McIntyre: "Our research does not say that the earth's atmosphere is not getting warmer. But the evidence from this famous study does not allow us to draw any conclusions about its extent, relative to the past 1000 years, which remains as much a mystery now as it was before Mann's article in 1998."