Originally tweeted by Stephen McIntyre (@ClimateAudit) on April 30, 2023.

Note from Anthony: For those of you that remember the yeoman’s work that Steve McIntyre did at Climate Audit debunking Michael Mann’s hockey stick and flawed methodology this should come as no surprise. Once again, hockey sticks get generated where the original data doesn’t show it. The only conclusion you can make is that the data and the method have been adjusted to fit a preconceived and desired result. This series of Tweets has been compiled here for easier access and readability. – Anthony

[M]ost readers are familiar with famous “hide the decline” from Climategate. Below are 1850-2000 parts of 5 series calculated from Asian tree ring data, explained below. I recently received some fantastic PAGES2k reverse engineering from @detgodehab and am re-visiting.

The data illustrated below comes from
(1) original Briffa 2001 Asian series with late 20th century decline (chopped off in Mann’s IPCC diagram);
(2) average of Asian series in gridded MXD series sent by Briffa/Osborn to Rutherford and Mann, ostensible input in Mann 2008

(3) average of (the 45) gridded MXD as used in Mann 2008. As discussed long ago at Climate Audit, Mann chopped off the offending declines and replaced them with temperature data. This was a different incident to the IPCC diagram or the 1999 WMO “hide the decline” diagram.

(4) PAGES2K (2013) introduced a novel Asia reconstruction (Cook et al) from tree rings in which late 20th C decline observed in Schweingruber data did not exist. Closing values were similar to high values at mid-20th century. The difference was not reconciled by PAGES2K.

(5) the current PAGES reconstruction – the “Woke Reconstruction” for short and used in IPCC AR6 – contained a subset of the PAGES2K Asia dset, the average of which yields a monster blade. The “decline” is in the Woke rear-view mirror.

A few years ago, I had noticed that some of the tree ring chronologies underlying the Woke Reconstruction had enormous closing blades that did not appear possible to replicate using standard methodologies.

I did a couple of Twitter threads as well:

I asked two lead authors of PAGES 2019 about the provenance of the Asian tree ring series, but got nowhere. They didn’t consider that they had any responsibility as lead authors of a Nature article to answer questions about their data.

The underlying reference (Cook et al 2013) contained only a single sentence as purported description of chronology methodology: that they took “considerable care” to avoid ‘segment length curse’, with “partial use” of a novel technique then recently introduced by UEA’s Tom Melvin

the keepers of these chronologies were at Columbia U which resolutely refused data when I was trying to figure out Hockey Stick mysteries. Jacoby: “Fifteen years is not a delay. It is a time for poorer quality data to be neglected and not archived.”

anyway, reader @detodehab got intrigued with the puzzling Asian tree ring chronologies and reverse engineered their calculation. He replicated the results to every detail. No one could have possibly imagined the actual calculation from details in PAGES2K or Cook et al 2013.

It’s hard for a statistical methodology to be so bad as to be “wrong”. Mann’s principal components methodology was one seemingly unique example. PAGES2K’s Asian tree ring chronologies are another. It’s worse than anyone can imagine.

unfortunately, exposition of the defective calculation are technical and will take some time. But for now, the monster blade of the “Woke” PAGES 2019 Asian tree ring data is bogus. PAGES2019 selectively chose the biggest blades, nearly all of which come from bogus chronologies.

I have a question for readers on order of exposition. Which should come first:
1) narrative of detective work by which calculation was reverse engineered
2) pathologies of PAGES2K Asia tree ring methodology
Not all Asia chronologies are pathological, but biggest blades are.

on the left is description of PAGES Asia2K chronologies and on right is my brief description of their actual algorithm as deduced by @detgodehab and verified by me. I’ll return later to how he figured this out. For now, the pathologies of the method.

I’ll describe the pathologies of the PAGES2K Asia algorithm more or less as we discussed them in DMs over past few weeks. @detgodehab had begun with analysis of paki033, the series that I had featured in a 2021 thread and blog article.

@detgodehab had ported the algorithm to Octave. The paki033 iteration stopped after 20 iterations. So let’s look at the evolution of the chronology. It opened as nondescript series on left and closed with big blade. On right is sequence of steps showing emergence of closing blade

what happened to individual cores? Tree ring “chronologies” are calculated as the difference between measurements and smooth (pseudo-model). Between start and close, the ‘model’ moved closer to zero at the close, so that contribution to chronology (shown on right) increased bigly

this looked very suspicious as a procedure. An obvious question was whether the monster blades in certain PAGES2K chronologies were some sort of artefact, as opposed to “climate”. So I suggested that @detgodehab see what happens when last 50 years of data not used? As a test.

bingo. Excluding the last 50 years of data, paki033 had an even bigger blade //50 years earlier//. For good measure, @detgodehab did test excluding 25 years and got same big blade //25 years earlier//.

so it was very clear that the big blade being produced by the PAGES2K Asia chronologies was bogus and some sort of artefact of their methodology and NOT due to climate. This doesn’t disprove global warming. It is only relevant to PAGES2K.

Also, PAGES2K uses much other data. Nor are all Asia 2K chronologies calculated with this bogus methodology. But PAGES2019 selected the worst and most bogus chronologies (claiming they were the best) and that’s why there’s the monster blade shown in opening tweet.

while it’s evident that the PAGES2K Asia methodology was pathological, when @detgodehab chopped off 100 years, it didn’t produce a blade. I haven’t parsed this to see why. There are many mysteries in the algorithm which I’ll continue to describe.

In an iteration, convergence is presumed. But PAGES2K algorithm did NOT converge for paki033. It was stopped at iteration 21. I ran D’s algorithm for 100 iterations and found that the Melvin “test statistic” (a weird one) increased up to iter ~68, dropped suddenly, then rose

this is NOT acceptable behaviour in a valid algorithm purporting to converge

what was happening to the chronology during these wild changes in “convergence” statistic? The blade (which had stopped at ~4) continued to grow reaching ~37 at iteration 50, then declined to ~15.3 by iteration 100. Obviously not climatic

the maximum of the blade by iteration for 500 iterations is shown below. It did actually converge under Melvin statistic – but to an implausibly large blade of 15.338.

@detgodehab observed “so clearly convergence doesn’t mean that a chronology is valid”. Clearly.

recall the diagram shown previously in which I had extracted the smooth ‘model’ for an individual core. At right are the ‘models’ for each core at convergence: they all approach zero at end.

a technical point: there are two main approaches to “detrending” each core to allow for juvenile growth: 1) a separate curve for each tree/core; 2) one curve (typically negative exponential plus constant) for site.

I did an experiment applying Melvin’s iterative method and supsmu smoothing as follows: 1) to individual cores as done in PAGES2K Asia; 2) one curve for entire site (“RCS”). The monster blade only appeared with PAGES combination of Melvin iteration and supsmu corewise smoothing.

thus far, I’ve discussed one site paki033.

@detgodehab has verified that same flawed algorithm was used for at least 8 other Pakistan sites. Note that these sites (together with Columbia U’s Mongolia chronologies) dominate PAGES19 list of heavy contributors to closing blade.

Also note that all these sites were preferential selected by the PAGES2019 screening procedures (which I’ve criticized elsewhere.)

Originally tweeted by Stephen McIntyre (@ClimateAudit) on April 30, 2023.