Comments on: McKitrick: What the Hockey Stick Debate is About?

By: Skiphil

Skiphil — Sat, 08 Dec 2012 20:47:41 +0000

AGU political leadership seems dominated by the Romm/Mann school of hype.

It would be interesting to know how long insiders maintained the view that Gleick’s misbehavior has “ravaged his reputation” (Romm’s phrase at linked book review), since it is still 2012 and Gleick/Mann seem to be at the peak of their reputational currency in many AGU circles.

By: Skiphil

Skiphil — Sat, 08 Dec 2012 20:37:51 +0000

This is a fascinating anecdote. This sort of setting must be even more “interesting” now for such seminars and meetings.

By: Science or political sophistry? | Squall Lines

Science or political sophistry? | Squall Lines — Wed, 13 Jun 2012 16:55:58 +0000

[...] there is no “settled science” on the theory of global warming. (link) In international surveys of hundreds of climate scientists “…half the scientific community [...]

By: Bernie

Bernie — Wed, 29 Feb 2012 22:35:42 +0000

William:
I assume that you are saying SVD methods are not very useful data exploration tools when you do not have well defined underlying processes and a very restricted data set. Apparent significant factors can emerge simply by chance with no or spurious relationships to the phenomena being explored.

By: theduke

theduke — Wed, 29 Feb 2012 20:21:42 +0000

Looks like the LA Times has hired Joe Romm to head their Opinion staff:

http://opinion.latimes.com/opinionla/2012/02/mann-climate.html

By: William Hayden Smithw

William Hayden Smithw — Sat, 11 Feb 2012 13:41:13 +0000

Ozjuggler,
SVD analysis is a mathematical tool for extracting information consistent with a data set. An answer is found, even if the data set is flawed. The main difficulty with the hockey stick is that it is not consistent with most other independent data sets. Other tree ring data, 10 Be data, orbital satellite data, borehole data, ice coring data, etc. show, for example, the lengthy medieval warming which is practically absent from the hockey stick. The tight correlation of the hockey stick with CO2 increase is largely absent as well. Every data set shows some warming. For example, the borehole data show warming since about 1500 AD which clearly was not anthropogenic, and in the latest decade, since the very warm 1998, the temperature trend is downward even in the Hadley Center compilations; the most ardent supporters of anthropogenic global warming. Ocean levels have fallen, arctic ice has increased, and so on. So, what was predicted is not consistent with the on-going observations of global climate.

By: OzJuggler

OzJuggler — Sat, 11 Feb 2012 10:48:26 +0000

Steve, sorry but that doesn’t really seem to answer the question. In the GRL paper your own diagram shows the example red noise hockey stick is 10 times smaller in scale than the final MBH98 hockey stick. The example was constructed from noise profiles with the same magnitude as the real data because they were based on the real data’s autocorrelation function, yes? And on top of that you divide the series (noise and real) by their SD because the PCA is made more robust by using normalised values, yes? Well the red-noise contrivance produced a hockey stick shape, but not The MBH Hockey Stick magnitude. Therefore “The Hockey Stick” (in both size and shape) was not an artefact only of the technique, so it must be inherent in the tree data. How is that not a fair judgement?

Steve: the MBH PC1 has EXACTLY the same scale as the synthetic PC1s. This is necessarily so because of how singular value decomposition works. The scale of both gets converted to degrees in the regression phase. There has been considerable disinformation about this from critics of ours, but you’ll notice that neither Mann himself nor wahl and Ammann raised this particular criticism.

I’ve read Ross’s description so I understand how the freak trees were amplified.
Is it fair to say the only reason The Mann Hockey Stick was not produced from noise is because no noise sequence had randomly produced a freak late 20th century growth spurt in any series that was fast enough to be exaggerated to that scale?
If you put a single freak Sheep Mountain tree into the rest of the hundreds of noise samples you used, do you get the no-trend PC suddenly convert into the MBH98 icon?

Steve: Mannian PCs are a very biased methodology. If you put a couple of HS-shaped series into a series with a different signal, you will get a HS. See our Reply to VZ.

Surely the physical reasons (eg the Divergence Problem) are a much greater reason to reject the MBH hockey stick than the statistical reasons (eg- the slight effect of non-standard centering and non-standard PC thresholding) ?

Sorry if this is old ground, but I’ve been a climate skeptic for many years and am only now hearing that some of the things I’ve been told over the years may not be the whole story. If this has been answered already in the intervening 7 years, just show me the thread.

Steve: if you read our articles rather than secondary sources, you’ll see that the red noise argument, though it attracted a lot of attention, is only one of several important criticisms. In retrospect, I would have placed additional weight on questions and issues arising on application of principal components (of any type, not just Mannian incorrect principal components) to tree ring networks. I look attentively at criticisms of our position and none to date have undermined our points. There’s a great deal of disinformation. I would have expected specialists in the field to be understand the criticisms better than they have and have become a bit bored with what seems to be wilful obtuseness.

By: William Hayden Smith

William Hayden Smith — Wed, 08 Feb 2012 18:00:16 +0000

As a user of SVD methods, I can answer generically the questions posed by OzJuggler.
In a scene, the number of components that can be resolved depends upon the spectral contract (variability) and the signal-to-noise (S/N)of the measurements. The more structure in the spectrum and the higher the S/N in the data, the more components can be distinguished in the data. So far, so good.
For example, a terrestrial scene compared with an ocean scene for the same S/N (assuming the S/N is high and not near unity) has more distnguishable components since the spectral contrast is always higher, or put another way, water completely dominates the ocean scene, as you might expect.
Tree ring data, along with other climate proxies, has a fundamental problem. The spectral contrast (ring variability) arises from multiple causes. SVD assumes an orthonormal basis set, and as a result, picks out the optimum basis for decomposition, which is NOT simply temperature, precipitation or other climate variables. Nonetheless, the observations ARE interpreted without KNOWING what the S/N is with regard to the climate parameters being evaluated. The proxy may show correlation but through what parametrization with regard to the climate is not known. One HOPES the basis vectors are known climate parameters such as temperature, precipitation, mean wind fields, insect infestatios, or other known components. Forcing the basis set to BE well defined climate components and then extracting the principle components will always achieve a best fit, but uniqueness does not exist.
Put succinctly, SVD assumes you KNOW the correct orthonormal basis upon which to decompose the measurements, which you never do. So, SVD must be taken with a LARGE grain of salt, however it is used. If a scene has only two or three major components, and you extract some vectors from SVD which resemble the expected components, then the agreement makes you feel good, but since it is not unique, I would not bet my house on it either way.

By: OzJuggler

OzJuggler — Wed, 08 Feb 2012 16:39:45 +0000

An SkS user called “caerbannog” says that McIntyre doesn’t know how to apply SVD.
http://www.skepticalscience.com/MichaelMann.html#70788

He says:

The bottom line is, unless you look at the singular values, you can’t say *anything* about your data. You can’t simply look at the principal components (aka singular vectors) without considering the associated singular value magnitudes and draw any reasonable conclusions about whether your data vectors contain a “common signal” or are just random noise. Without the information provided by the singular values, you simply can’t tell (no matter what your principal-components look like).

But that’s exactly what skeptics did when they attacked Mann by claiming that his procedure generates hockey sticks from random noise. They never bothered to compare their “noise hockey stick” singular values with Mann’s “tree-ring” singular values.

Could McIntyre respond, perhaps by saying whether this is a red herring argument about the validity of the temperature hockey stick, or whether caerbannog actually has a valid point.

Steve: In our 2005 GRL article, we comment on “eigenvalues” from simulations as compared to Mann’s. “eigenvalues” are the same thing as “singular values”. Not that the issue turns entirely on eigenvalue significance. The bristlecones in the AD1000 network are “significant” even without Mann’s erroneous PC method – a point made in our 2005 EE article. That doesn’t mean that they are magic thermometers. Other important issues in these articles are the failure of verification statistics said to have been used in MBH (verification r2) and the unreliability of RE as an arbiter of reconstruction validity in the face of spurious correlation.

By: The Liberty Papers »Blog Archive » Climate Gate 2.0 – What is it, why does it matter?

Wed, 30 Nov 2011 12:41:49 +0000

[...] The first controversy, “hiding the decline” is related to an attempt to create a global temperature record by Dr Michael Mann of Penn State, who used records of tree-cores collected at a handful of sites across the world to create a historical temperature record. By measuring the density and thickness of the rings, one can create a record going back about a thousand years of tree growth. Dr Mann used a statistical process that is a variant of Principal Component Analysis to generate identify which sets of tree-cores had growth patterns that most closely tracked temperature in the past hundred years. He presumed that these sets of cores would maintain a similar relationship with temperature throughout the entire record. By mathematically applying this transformation to the tree-core data, he produced the thousand year reconstruction known colloquially as the Mann Hockey Stick, which played a central role in both IPCC reports and in Al Gore’s movie, and Inconvenient Truth. At this point, I should digress to explain several critical flaws in Michael Mann’s work that doom this effort. [...]