the Air Vent

Because the world needs another opinion

Climatology School

Posted by Jeff Id on September 20, 2009

This is what I imagine climatology school is like.


Today as paleoclimatology is becoming a sophisticated and mature science we have thousands of young students scouring the earth looking for other proxies for which to measure temperature. It’s important that these students have the best education and understanding possible such that the mistakes of the past don’t get repeated. In this post we’ll review a few items which at first glance may contain temperature signals, however on careful examination are not ‘actual’ thermometers. The purpose of this study is to prevent otherwise potentially embarrassing mistakes.

#1 – A garden hose. Like Venus flytraps for the unwary climatologist, these things are laying around everywhere, tempting and teasing. However, no matter which way they are coiled their shape should not be construed as or mistaken for temperature. Don’t be embarrassed,you’re not the first to look longingly at its twists and turns. It’s an easy mistake for certain which can only be avoided through proper education.


2 – Bowl of noodles.  Noodles are visually intriguing. They make all kinds of interesting patterns and can even lie side by side apparently in agreement with each other. Despite a veritable bowlful of potential correlations, beware of this hornets nest young paleo’s, noodles are not temperature.


#3 Actual hockey stick. Despite the apparent advantages to a young paleoclimatology career, don’t make this same mistake often made by even experienced climatolgists. While actual hockey sticks mysteriously do contain a ‘ghost image’ of what appears to be temperature, sophisticated instrumentation has revealed that hockey sticks are actually NOT temperature. Source – Hakley et al.


#5 – Bag of clubs. This one is a little tricky now. It turns out that while this bag of clubs has a strong relationship and high correlation to temperature, the expected signal can only be extracted from the multiple heads all pointed in different directions through sophisticated principal component mathematical analysis. At first blush, it looks like an excellent prospect, however like the hockey stick above, further analysis has determined the actual signal in the club is related to solar output and therefore is obviously not in any way related to CO2 driven temperature.


Grass clippings are the hottest new proxy. While their curvilinear shape results in a high insensitivity ratio, it turns out that temperature can be extracted from grass clippings through a careful analysis of correlation to thermometers. This results in a high scrap data rate, however studies have shown that the high rejection rate can be overcome using sophisticated bulk sample collection technologies. Some have criticized these studies as they have disingenuously shown that a negative sensitivity to temperature is equally apparent. There is a substantial amount of peer reviewed literature to the contrary however and it looks like this proxy will be discussed hotly in the coming years.


That should be good for today’s lesson. Read chapter 10 carefully for tomorrow where we’ll study the mysteries of the thermometer and the foundation of the myth in the lesser sciences that it can only be read upside right.


We’ll close with this intriguing quote from Michael Mann AKA “the Architect”.

The claim that ‘‘upside down’’ data were used is bizarre. Multivariate regression methods are insensitive to the sign of predictors.


24 Responses to “Climatology School”

  1. John F. Pittman said

    “”The claim that ‘‘upside down’’ data were used is bizarre. Multivariate regression methods are insensitive to the sign of predictors.””

    You know JeffId, even hung over and sleepy, this one is always good for a laugh. It is such an advertisement that some do not look at their data.

    Perhaps you can help me. I do not know which sentence is funnier; sentence 1 or sentence 2. The only way it could be funnier is if the first sentence read “”The claim that ‘‘upside down’’ data were used is just not robust.””

  2. Jeff Id said

    I think they need to come together to get the full effect. 1+1 = 3 kind of thing.

  3. cogito said

    “Multivariate regression methods are insensitive to the sign of predictors.”
    Is this really true? Sorry to ask, but statistiques is not my field of knowledge.

  4. Jeff Id said

    #3 It’s true. They’ll flip the sign with no regard for what the curve means. Thats the reason behind my recent positive matrix factorization posts. No flipped signs.

    It’s hilarious that a PhD can put forth the claim that it makes no difference, equally hilarious as the trees, mollusk sphincters and varves are thermometers hypotheses. If people only understood the complete lack of verification or even interest in verification that the bowl of noodles are actually thermometers.

    I like that, bowl of noodles.

  5. Dave said

    “Multivariate regression methods are insensitive to the sign of predictors.”

    As are, apparently, some climate scientists.

  6. Charlie said

    #3, #4. Not all of us are as steeped in the details of proxy reconstructions as you are Jeff. We need the explanation in a bit more detail. Here’s my understanding of the situation. Please correct or confirm.

    My understanding is that the statistical technique of multivariate regression is insensitive to sign. You could randomly invert the proxy data and you would still come up with the same level of correlations.

    But for a dataset to reasonably be considered as a proxy for temperature, though, there should be some expected (or better yet, confirmed) relationship between the proxy data and temp.

    A multivariate regression without a constraint on the sign of the predictor can result in the proxy being used flipped upside down from what has previously been determined to be the proper relationship between the proxy and temp.

    So Mann’s statement that “Multivariate regression methods are insensitive to the sign of predictors.” is correct when the discussion is limited solely to the operation of the algorithm.

    On the other hand, that the multivariate regression flips the signs to the reverse of the expected relationship is a big red flag that something is wrong in the analysis.


    More crudely put, it would be as though someone found a strong negative correlation between atmospheric CO2 and temperature, and then used that strong correlation (but with inverted sign) as definitive proof that increasing CO2 causes warming.

  7. Layman Lurker said


    Wow this is serious deja-vu. Have you reviewed the thread(s) at tAV on Steig’s Antarctic negative thermometers?

  8. Jeff Id said

    #7 That’s why it’s so important to understand.


    Everything you said is correct as I understand it. The only detail I saw was

    But for a dataset to reasonably be considered as a proxy for temperature, though, there should be some expected (or better yet, confirmed) relationship between the proxy data and temp.

    I would say that an expectation of a temperature measurement is not sufficient for use of proxy data, I also don’t believe correlation is enough evidence for it to be used. There needs to be physical measurement of response in controlled conditions which is something the famous in climatology don’t care to do. Each time nature has carried the experiment out the proxies keep getting disproven. They use the term Divergence to describe this effect.

    I use the term – “ain’t temperature” its more clear.

  9. cogito said

    Thanks for these explanations, I’m learning …

  10. timetochooseagain said

    2-In the words of Orwell “Freedom is the freedom to say that 2+2=4”.

  11. wattsupwiththat said

    Top Ramen could be a proxy for temperatures. I think the noodles don’t plump well unless the water temperature is hot enough.

    Mann’s Cup-O Noodles. Now in pine flavor.

  12. Jeff Id said

    Top Raman, five favorite.

  13. j ferguson said

    Is all this proxy stuff really addressed to some profound need to use the word “unprecedented” or is it more to fabricate a long baseline so that the upward slope on the HS looks especially and unusually steep?

  14. Don S. said

    Top Ramen noodles also can be used to determine altitude. At 3250 feet in my kitchen Ramen noodles do not reach full plumpness due to the water not reaching 212 degrees. This is follow-on research re. WATTSUPWITHTHAT I’m applying for a grant and formulating a robust study.

  15. geoff chambers said

    Not only very funny, but I learned something new when you mention mollusc sphincters as a temperature proxy.
    Are there really people who can tell the future by looking up an oyster’s bottom?

  16. Sera said

    I prefer Smooth Rahman noodles (although the ends tend to pop if overcooked)

  17. Geoff Sherington said

    Re # 15 Geoff Chambers –

    “A poor virgin, sir, an ill-favoured thing, sir, but mine own: a poor humour of mine, sir, to take that no man else will. Rich honesty dwells like a miser, sir, in a poor house, as your pearl in your foul oyster”.

    Not mine. Shakespeare, “As You Like It”,(1599) Act 5, Scene 4.

    Strange how many messages lurk in that quote.

  18. jryan said

    So, to the layman (I’m one):

    Whether or not the data is flipped has no effect on correlation simply because the regression is testing for the changes in magnitude between two or more variable… so if you flip the data you get the same correlation because the “magnitudes” — for lack of a better term — don’t change, only the sign does.

    Where Mann is off his rocker is in assuming that correlation has anything to do with model accuracy when the data has been flipped.

    A simple example: I theorize that the sun rising causes the air to outside to cool. I take numerous measurements over time that show that there is indeed a strong correlation between the height of the sun in the sky and a change in temperature. The only problem is that I find the curve to be exactly opposite of what I expected. Instead of a drop in temperature I get a rise in temperature…

    Well, that won’t do!

    So, I flip the temperature graph so now the temperature graph shows a decline in temperature with the sun rise… I then run a multivariate regression against the flipped data and show, low and behold, that the correlation is exactly the same! So I then claim that my NEW graph has a high correlation so therefor must be accurate.

    Which is completely untrue because I flipped the data.

    Now imagine a whole host of data sets with tree rings and local temperature instead of sun height and local temperatures… what Mann is doing is finding any sets with positive OR negative correlation and flipping those with a negative correlation and inserting them into the model… this manufactures a final data set with high correlation, for sure, but does not if fact tell you anything meaningful about local temperatures… all it does is show that some trees, and not others, have had drastic changes in tree ring thickness over the last 150 years.

    In REAL science the REAL scientists (or statisticians) would be more interested in the multitude of trees that DON’T reflect the hypothesis rather than the handful that do. All Mann has done is assume that there is a handful of magic tree thermometers, and the rest are just stupid old trees.

    That is how I understand it, anyway. Feel free to correct me.

  19. Jeff Id said

    #18, I’d pass ya but I think you just flunked out of climatology school.

  20. jryan said

    Crap! Oh well… judging from the hearings on C-SPAN it’s no big loss, anyway. Oceanography will be the new hot job!!

    Acidification!! Acidification!!!! ACIDIFICATION!!!!!!

    (… money… month long sea cruises… attractive young Oceanography coeds in bikinis*… it’s a wonder they didn’t pick that first)

  21. s. geiger said

    question – if a tree ring (or other proxy) recon does not show good correlation during the calibration period (i.e., as the upside down series DOES for Mann), it is no longer retained or used for the final reconstruction (I think, at least this is how I understand it). What percent of these proxies do NOT show good correlation and are thus not used? Im sure this depends on the study/site/proxy, but is there a general percentage that do NOT act as good thermometers?

    thx for any info.

  22. Jeff Id said

    #21, You have asked exactly the right question. I would like to suggest the hockey stick posts button under the header at the top. There is a bit of math but all is answered.

    For Mann 08 60% was discarded.

  23. s. geiger said

    22, thanks. So of those 40% that did pass, its likely that AT LEAST some of them are also bad thermometers but that match ‘by chance’ fairly well in the calibration period. Do they run any sort of stats to at least compare how ‘close’ the remaining proxy series are in the non-calibration period? (or, are they all over the place and then just tend to ‘cancel’ out signal…leaving a relatively ‘flat’ stick handle?).


  24. Jeff Id said

    You are on the wrong thread, this is the fun one. You’ve already figured out the hockey stick. In fact all of them if you dig into the math.

    The point is that the noise in data like tree rings is created by non-temp effects. Half the time this creates upslopes which react quite well with correlation to temp. The negative noise is rejected in the recent century while it is unsorted in all previous. Thus high ‘variance’ in the recent times, low variance in the ‘historic’ times. Flat handle, strong blade – hockeysticization.

    The posts at have a detailed explanation of the effect.

    You sir have flunked climatology school and have been sent to skeptic school for rehab. hehe.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: