Over at RealClimate, a commenter by the monicker of manacker insists that global temperature follows a cyclic pattern. His “analysis” to establish this consists of saying, “Looks pretty cyclical to me.”
Whether he knows it or not, he’s just repeating a long-standing meme from the denialosphere. It reached its “peak” (hohoho!) in a paper by L.B. Klyashtorin & A.A. Lyubushin, On the Coherence between Dynamics of the World Fuel Consumption and Global Temperature Anomaly, published in — surprise! — the “journal” Energy & Environment (2003, 14, No. 6).
Klyashtorin et al. model global temperature as a superposition of a linear trend and a sinusoid. They treat temperature from 1860 to 2000, but let’s go one better and use HadCRUT3 data from 1850 to 2000 to define the model. They also extend their model beyond 2000 to predict that global temperature will soon decline (surprise!), so let’s extend that model from 2000 to the present — to find out whether or not it has any predictive power.
The best linear+sinusoid fit to HadCRUT3 data 1850-2000 has a linear slope of 0.003901 deg.C/yr and a period of 63.5 years. And here’s the data, together with a lowess smooth in red and the Klyashtorin model in blue:
There’s sizeable misfit for this model even during the time span used to define the model (1850-2000), and the fit beyond 2000 isn’t very good. This is clearer if we look at residuals:
The most notable discrepancy is that after 2000, the model gives temperature estimates which are way too low. I’m shocked. Not.
We can try the same thing with GISS data, using 1880-2000 to define the model, then extending it beyond 2000. This time the best linear+sinusoid fit has linear slope 0.004947 deg.C/yr and period 55.5 years:
The model fits much better to the GISS data 1880-2000 than it did to the HadCRUT3 data. But after 2000, it fails utterly:
Again, I’m shocked. Not.
Has global temperature evolution over the last century+ been cyclical? Not.
36 responses so far ↓
guthrie // December 22, 2009 at 2:16 pm |
I ran into someone a couple of days ago that explained away all the warming with reference to sunspots and the PDO, and could make nice curves to prove it. Unfortunately he forgot about things like stratospheric cooling which rather blow a hole in that, not to mention assuming no CO2 effect, but anyway:
http://climaterealists.com/attachments/database/TwentiethCenturyTemperatureCorrelationupdate.pdf
[Response: We have a new leader in the bag-of-hammers race.]
Jim Galasyn // December 22, 2009 at 4:44 pm |
“monicker of manacker” — nice!
Sekerob // December 22, 2009 at 5:00 pm |
The sum of the denialoose easy explain away each one readily accepted by the denailoose as the reason, would turn earth straight back to deep freeze. What’s new?
carrot eater // December 22, 2009 at 5:13 pm |
Did Klyashtorin et al provide any plausible physical mechanism for their linear trend and superimposed sinusoid? or were they just scribbling lines on graphs? I don’t feel like taking the time get an article from E&E.
At least those authors allow an underlying trend. It’s unclear if Manacker does.
Didactylos // December 22, 2009 at 6:10 pm |
Does it even matter if climate has a cyclical pattern? After all, the fit still requires a linear component (so it doesn’t even pretend to question global warming), and we know that among the many natural forcings affecting climate, more than a couple are cyclical.
Pattern matching and correlation-mining is just numerology.
I won’t trot out the old one about fitting an elephant. Oooops. Too late.
Hank Roberts // December 22, 2009 at 6:38 pm |
> Over at RealClimate
And many, many other places. Google.
One site ranks that name (of course it may be legion) among the top 50 or so posters globally of denial themed material.
Nick Barnes // December 22, 2009 at 7:26 pm |
There are obviously oscillations, but you show very clearly that they are not regular. I have an intuitive theory about this (essentially that the excess heat is stored in sub-surface Pacific waters and then comes back to the troposphere in El Nino). So I would expect deviations from the linear trend to be something like a random reverse sawtooth. This is what is going to kick us in the butt in the next big El Nino (quite possibly the one we are in now).
Anyway, I think this sawtooth notion only applies on decadal and sub-decadal timescales, not the “60-year cycles” which manacker sees.
Deep Climate // December 22, 2009 at 8:15 pm |
What if sinusoidal (or Nick’s sawtooth) were superposed on a (rising) exponential fit? At the centennial level surely AGW is accelerating.
On the other hand, it might not fit much better, just eyeballing it.
P.S. How about a new open thread?
Riccardo // December 22, 2009 at 8:45 pm |
Didactylos ,
what these people try to demonstrate is that the linear trend is much lower, it would be 0.4-0.5 °C/century; hence the effect of CO2, if any, would be much lower.
DrC // December 22, 2009 at 9:28 pm |
Ah! Vintage Tamino. Keep them coming.
Fitting a periodic model would be interesting if:
1. The period was chosen to match a real oscillation (and not a quasi oscillation like the PDO), and
2. there was a mechanism to support such a model
Recall that the reasons that A.E. Douglass essentially invented what we think of as dendrochronology was to find the 11-yr sunspot cycle in tree growth. Despite what his Wikipedia page says, he never found it. But is wasn’t a bad idea!
carrot eater // December 22, 2009 at 10:39 pm |
Deep Climate:
“What if sinusoidal (or Nick’s sawtooth) were superposed on a (rising) exponential fit? At the centennial level surely AGW is accelerating.”
that’s exactly what Mojib Latif did, wasn’t it, in that presentation that got totally misinterpreted by the blogosphere?
David B. Benson // December 22, 2009 at 10:40 pm |
Even some otherwise highly competent researchers are sometimes snookered by “cycles”.
Variability of El Niño/Southern Oscillation activity at millennial timescales during the Holocene epoch
http://www.nature.com/nature/journal/v420/n6912/full/nature01194.html
is quite good and based on some rather impressive limnological research. Unfortunately, they detect a “statistically significant” 2000 year cycle. At least the abstract states “However, the millennial oscillation will need to be confirmed in other ENSO proxy records.”
Rattus Norvegicus // December 23, 2009 at 5:56 am |
Looks interesting, but I don’t subscribe and $32 is a bit much for one article.
Caviedes uses archeological and historical data in his analysis, but it only goes back a few hundred years. The book is well worth reading.
Riccardo // December 22, 2009 at 10:53 pm |
Deep Climate,
I can hardly imagine a skeptic doing an exponential fit on the temperature data ;)
I tried, though, and got a better fit in terms of chi^2 than with the straight line; but then the modulation of the sinusoid (amplitude about 0.1 °C) disappers quickly you are left with just the exponential rise. In the end, the (eventual) sinusoidal modulation would be irrelevant.
carrot eater // December 23, 2009 at 12:44 am |
Riccardo,
You might get a skeptic to do an exponential fit, if he’s trying to tell you that the warming is part of some magical ‘recovery’ from the Little Ice Age. When in that mode, then they’re perfectly happy to admit there’s been warming. Of course, five minutes later they’ll be back to 1998 or the PDO or cosmic rays or whatever.
David B. Benson // December 23, 2009 at 1:51 am |
Detecting low frequency oscillations of the pacific ocean by the ocean upper layer temperature data
Mingqiang Fang et al. 2003
find power spectral density peaks which may be related to ENSO at periods of 13.8, 8.1, 5.0, 3.8, 2.5 and 2.1 years, in order of declining height.
Deep Climate // December 23, 2009 at 3:48 am |
Riccardo,
Good one! Well, of course, a second order exponential over 100 years is different from a 6th order over 30 years. The skeptic algorithm appears to be try stuff until you get a downturn at the end. BTW, some time I might revisit the exponential on UAH – at some point it will curve up again quite scarily if it hasn’t already.
c.e. ,
Yes Latif started with exponential – but I think he superposed some kind of random noise rather than a sinusoid.
chriscolose // December 23, 2009 at 4:36 am |
The much more important point is that global temperature anomaly follows radiative forcing, and the RF evolution over the 20th century has been such that a warming to slight cooling to strong warming are expected. Volcanoes and CO2 rise is not “cyclical” stuff, it’s just how it worked out.
Lamont // December 23, 2009 at 8:35 am |
Ricccardo, try a log-periodic function like Sornette uses in:
http://www.amazon.com/Why-Stock-Markets-Crash-Financial/dp/0691118507/
He uses basically a sinusoidal wave on top of an exponential, but as you approach where the exponential blows up the frequency increases.
Do I get some kind of obscure-bag-of-hammers runner up prize?
Douglas Watts // December 23, 2009 at 9:30 am |
This raises the question of trend overprint. It is perfectly possible to have a natural trend and an anthropogenic trend acting simultaneously and creating an overprint in your trend. And unless you can completely eliminate any chance of an anthropogenic effect from a trend, you cannot call the trend wholly natural. Some part of it, large or small, is likely to be due to anthropogenic effects.
So even if you were extremely generous in ascribing some natural cause to the existing late 20th century global temperature trend, its existence is not a de facto negation of an anthropogenic overprint.
You can have both.
carrot eater // December 23, 2009 at 12:08 pm |
DC: I didn’t mean to imply Latif used a sinusoid. That’d be strange.
I think he used stochastic noise that had the same properties as the residual between the exponential and the observed temps.
Riccardo // December 23, 2009 at 12:28 pm |
Lamont,
well, i don’t think that a physical system will ever show the weird behaviour of the stock markets :)
J SMITH // December 23, 2009 at 4:40 pm |
Manacker infests many a site, spouting his nonsense copied from one to the other.
He and his pals hang out at :
http://ccgi.newbery1.plus.com/blog/
Check out any of the articles to see a wealth of self-confirming denial, many involving manacker. I believe he is Swiss or Austrian.
jerryg // December 28, 2009 at 2:08 am |
He signed one of his messages on CA as being from Maienfeld, Switzerland
george // December 23, 2009 at 5:09 pm |
Riccardo said
Yet Black–Scholes, based on a physical model ( assumes the price of a stock follows Brownian motion), has been used extensively by financial analysts [sic] to price derivatives.
and we all know how well that worked out.
Hammers seem to be a mainstay of the wall Street financial analyst’s bag of tricks.
In fact, one could argue pretty convincingly that the hammer is the only thing they have in their bag and that Wall Street itself is a gigantic hammer looking for a plate Glass-Steagall window to shatter.
David B. Benson // December 24, 2009 at 12:21 am |
Another view of temperaure data is that of (1/f^a) noise for a in the interval [0,2]. The paper
Links between the Annual, Milankovitch, and Continuum of Temperature Variability
by Peter Huybers & William Curry in Nature suggests that annual to multidecadal data is power law noise with a smaller than 1 while on centennial and longer scales the power law increases to as much as a=1.6.
For values of the power law larger than one, there is a tendency to “see” cycles in the data. However, these go away once a longer time series is obtained. Wavelet analysis enables one to visualize the waxing and waning of such “cycles”.
Hank Roberts // December 25, 2009 at 10:51 pm |
> Sornette
Regrettably he’s predicted far more crashes than happened, and he missed the timing on the big one by predicting it several years early. He declared defeat and gave up predictions well before it happened. He was a lot of fun to watch though.
TCO // December 26, 2009 at 1:57 am |
If you could segue this into a more interesting general discussion of what statistical tests to use (and I recognize there may be choices) to test for cyclicality, it would be more interesting.
Just picking another of the weaker denialists and spending your time as some sort of PR advocate blog-warrior to oppose him is not so interesting.
Realize that there are some of us here who have our union cards. While we may not be statistical experts in terms of math, we are well able to follow logical, sophisticated discussions.
Would be better if you went after the better arguments of denialists…not the weaker. And/or just used the discussions as ways to segue to interesting general stats discussions. Who knows…perhaps even avoiding some of the general bile that both sides have for each other and just saying…this guy is wrong…but it brings us to an interesting topic to discuss.
I will admit though…that I am a genuinely intellectually curious person. Like the sidedness tussles too. But prefer to learn than just find things to bash.
Btw…I have very similar critiques of McI…that he is more interested in gotchas than in the sometimes interesting issues he brings up and in wrestling them to the ground whichever side they end up supporting.
elspi // December 26, 2009 at 2:25 am |
“Would be better if you went after the better arguments of denialists…”
WHICH BETTER ARGUMENTS??????
dhogaza // December 26, 2009 at 4:53 am |
Oh, gosh, and what would those be? That there’s no evidence that CO2 absorbs long-wave IR? That it’s all GCRs? That natural variations proves that GHGs can’t force an underlying signal?
What, exactly, do you have in mind?
What is your evidence that Tamino hasn’t just gone after the *best* of denialist arguments?
Hank Roberts // December 26, 2009 at 5:16 am |
> another of the weaker denialists
By volume output, prolific authors of repetitive misinformation are worth tagging, because their bogus claims are effective by repetition.
Inhofe, Morano, and their ilk.
george // December 26, 2009 at 5:56 pm |
dhogaza:
I think we all learned long ago (through repetition) that TCO is an omniscient, totally unbiased, genuinely (Feynmanly) curious climate oracle whose Raison d’être is to inspire us all to ask the “right” questions, discover the right answers and thereby find and follow the path of enlightenment.
Hence his assessment of which are “the better arguments of denialists” is worth infinitely more than that of Tamino.
Do not question it, my son.
Lab Lemming // December 29, 2009 at 11:24 am |
Is there a weak ~20 year sinusoidal in the residuals?
David B. Benson // December 29, 2009 at 8:51 pm |
Lab Lemming // December 29, 2009 at 11:24 am — There is power at all frequencies. Approximately, there is little but pink (1/f) noise.
Joseph // December 30, 2009 at 4:05 pm |
The thing about models is that they should not only fit the data that was used to derive (or train) the model. You can come up with all sorts of models to explain series, but the question is: do they generalize well outside of the training data set? (Or do they forecast well?)
This is a common problem in machine learning. You can train a program using heuristic techniques, and your program might learn how to make predictions within the training data set quite well, but when you try different data, it fails. This is related to the concept of overfitting.
Gavin // January 1, 2010 at 1:04 am |
Possibly an interesting question would be what temperature data did Klyashtorin and Lyubushin use in their 2003 paper and 2007 book?
It does not look like the global mean – it has too much decline in the 1970s and too similar a peak in the 1940 to the 1980s. It purports to go to 2000, but the 1998 peak is clearly missing. They cite Jones et al (JGR, 2001) – but for some reason I can’t a copy of that today from AGU.
Any clues?
[Response: I couldn't identify their data either, although I noticed the Jones et al. reference. Your guess is as good as mine.]