Climate Audit

The 20th century warming counters a millennial-scale cooling trend which is consistent with long-term astronomical forcing.

MBH99

According to the UMass researchers, the 1,000-year reconstruction reveals that temperatures dropped an average of 0.02 degrees Celsius per century prior to the 20th century. This trend is consistent with the “astronomical theory” of climate change, which considers the effects of long-term changes in the nature of the Earth’s orbit relative to the sun, which influence the distribution of solar energy at the Earth’s surface over many millennia.

“If temperatures change slowly, society and the environment have time to adjust,” said Mann. “The slow, moderate, long-term cooling trend that we found makes the abrupt warming of the late 20th century even more dramatic. The cooling trend of over 900 years was dramatically reversed in less than a century. The abruptness of the recent warming is key, and it is a potential cause for concern.”

MBH99 Press Release

The long-term [northern] hemispheric trend is best described as a modest and irregular cooling from AD 1000 to around 1850 to 1900, followed by an abrupt 20th century warming.

IPCC TAR WG I: The Scientific Basis

The above figure has four familiar looking graphs. One of them is the original Hockey Stick, and three are “fake”. Can you tell which one is the real?

Although most of the original Hockey Stick methods have been uncovered, there are still a few remaining oddities. Apart from the confidence interval calculation there has been another mystery relating to MBH99. This is remarkable as the rather short MBH99 paper seems to be on the surface a simple extension of MBH98: a step (1000-1399) is added to the existing MBH98 NH temperature reconstruction using the same methodology. However, a wealth of material in the four page paper is devoted to “correcting” the (Mannian) North-American tree-ring series PC #1. How exactly or even why this was done has been somewhat a mystery. Two years ago Steve wrote notes about the issue (here, here, and here). It is worth reviewing those before continuing reading this post.

The problem with the methods described by Steve was that it was impossible that they had been actually used. The reason is that it is easy to see from the published data that the actual “correction”, or the “fix”, applied was piecewise linear. There is simply no way such a function could be obtained with any type of smoothing operations from the original data.

For the calculations Steve was using his private copy of Mann’s later destroyed UVA ftp archive infamously known for the CENSORED -directories. For the rest of us, data archived there has been unreachable — until now. The FOIA documents contain an MBH data directory structure obtained by Tim Osborn sometime back in 2003. It can be argued that the UVA ftp site was originally specially prepared by Scott Rutherford for Osborn, but that is another story. Anyhow, the files in Osborn’s archive seem to correspond to those originally located in UVA ftp site. The files in the directory TREE/COMPARE relate to the PC1 “fixing”.

While I was checking the files, I noticed a FORTRAN code “residualdetrend.f“, which I had not seen discussed anywhere. In the beginning of the file there is a comment:

c
c      regress out co2-correlated trend (r=0.9 w/ co2)
c      after 1800 from pc1 of ITRDB data
c

Wow! Exactly the same comment is found in “co2detrend.f” discussed by Steve here. Further down, we find

c
c      linear segments describing approximate residuals
c      relative to fit withrespect to secular trend
c

Indeed, there it was: a code removing a piecewise linear segment from the PC1, and further I found out that the segment matched nicely with the “secular trend in residuals” graph in MBH99 Figure 1(b). Mystery solved, well, kind of.

Now the question was, what the heck is then “co2detrend.f”?! I noticed that both these codes output to a file with the name “pc01-fixed.dat”. FOIA files include such a file, and its content matches to the output of “residualdetrend.f”. So IMO it can be safely assumed that Mann tried another CO2 “adjustment”, but for some reason ended up with the one described in “residualdetrend.f” (why to approximate the “secular trend” is another new Mannian mystery).

After establishing this, I had another surprise. I noticed that there is also a file “pc1-fixed-old.dat“, which I presumed to be the output of “co2detrend.f”. Well, it turned out that the “fix” contained in the file was neither of the methods described so far. Thus Mann had at least three methods for “adjusting” his PC1! Here is a plot of different “fixes” (to be subtracted from the original PC1) uncovered so far.

A natural question is now, why the fix used is “better” than the ones disregarded? Maybe the “skill” measures used by Mann contain the answer. MBH99:

The calibration and verification resolved variance (39% and 34% respectively) are consistent with each other, but lower than for reconstructions back to AD 1400 (42% and 51% respectively – see MBH98).

I (as Steve and UC) have been able to emulate the main MBH procedure for a while. Especially, my emulation of the AD1000 step is exact. So I ran the algorithm, but replaced the “fixed” PC1 with the other two “fixed” PCs. For the “co2detrend.f”-fix the calibration and verification REs are 0.37 and -0.09, respectively. So even according to Mann’s standards (negative RE) that “fixed” PC had to disregarded. For the “old fix” the RE scores were 0.37 and 0.20, so I guess they are not “consistent with each other”, and maybe this was the reason for trying yet-another-fix. However, the real surprise came when I tried the algorithm with the original Mannian PC1, i.e. without any “fixing”. The RE scores are 0.38 and 0.33, so based on these “skill metrics” there is no reason to “fix” the PC in the first place!

It gets more interesting: MBH99 has the linear trend (1000-1900 as in the IPCC figure) of -0.020°C/century, but without PC1 “adjustment” the cooling trend is reduced to less than -0.005°C/century! MBH99:

The substantial secular spectral peak is highly significant relative to red noise, associated with a long-term cooling trend in the NH series prior to industrialization (δT = -0.02°C/century). This cooling is possibly related to astronomical forcing, which is thought to have driven long-term temperatures downward since the mid-Holocene at a rate within the range of -0.01 to -0.04°C/century [see Berger, 1988].

Finally, the answer to the question posed in the beginning. The original Hockey Stick is Exhibit B. Exhibit C is obtained using the “old” AD1000 NOAMER PC1 “fix” keeping everything else the same in the Mannomatic. Exhibit D corresponds to the “co2detrend.f fix” , and Exhibit A is obtained using the original Mannian PC1 (no fixing). (click below to see an animated GIF of the different versions)