the Air Vent

Because the world needs another opinion


Posted by Jeff Id on November 27, 2009

Ben Santer had this unusual reply WRT a paper submitted by MM which used the same math that Ben Santer used to prove models were so good except they extended the climate model trend and temperature data after 2000. I haven’t read their paper cause it ain’t published yet but the Santer use of data pre-2000 only data was is pretty convenient to say the least. Consider that ‘skeptics’ or realists as we prefer suspect that models don’t relate the recent 10 years of no warming very well. As pointed out in other emails.

Ben makes the ridiculous claim that somehow McIntyre doesn’t realize that longer trends with the same noise have tighter confidence intervals but besides that truly goofy claim, he says it’s of no practical significance to look at model data for over what amounts to about 20 years. It seems a bit arbitrary don’t you think?

From: Ben Santer
To: P.Jones
Subject: Re: Good news! Plus less good news
Date: Thu, 29 Jan 2009 11:13:21 -0800
Reply-to: santer

Dear Phil,

Yeah, I had already seen the stuff from McIntyre. Tom Peterson sent it
to me. McIntyre has absolutely no understanding of climate science. He
doesn’t realize that, as the length of record increases and trend
confidence intervals decrease, even trivially small differences between
an individual observed trend and the multi-model average trend are
judged to be highly significant.
These model-versus-observed trend
differences are, however, of no practical significance whatsoever – they
are well within the structural uncertainties of the observed MSU trends.

It would be great if Francis and Myles got McIntyre’s paper for review.
Also, I see that McIntyre has put email correspondence with me in the
Supporting Information of his paper. What a jerk!

I will write to Keith again. The Symposium wouldn’t be the same without
him. I think Tom would be quite disappointed.

Have fun in Switzerland!

With best regards,



25 Responses to “???”

  1. Arn Riewe said


    OT somewhat, but as a layman, I’m trying to wrap my brain around the Tiljander issue. The argument of the RegEm’ers is that sign doesn’t matter once you turn on the sausage machine. But in my linear logic world, I would think a falling proxy would imply a negative impact on a model and vice-versa for a rising proxy. If signs have no impact, how is the trend of a composite model determined?

  2. Jeff Id said

    The issue is so simple I can’t beleive we still write about it. I think people are singularly shocked by the apparent idiocy. The Y axis of something assumed to be temp is multiplied by -1 turning it into anti-temp. It just happens to make a better hockey stick.

  3. Charlie said

    To say the same thing as Jeff, but in just words ….

    The program looks for correlation between the proxy measurement and temperature. And it also looks for correlation between temperature and the proxy measurement after it is flipped upside down. Whichever one is a better match during the calibration comparison period is used.

    It doesn’t matter whether or not it makes physical sense.

    For example, let us assume that an increase in temperature results in higher tree growth and either higher density or wider tree rings. If a particular core sample of a tree shows decreasing tree ring width as the temperature rises during the calibration period, then for _that_ tree, the program will treat thin rings as an indication of warmer temperature.

    Yes, I know it doesn’t make any sense. That’s the problem.

    It really is that simple and that dumb.

  4. PatrickG said

    RealClimate has put up a web page that claims to provide all of the data that they can publish at this time. Mostly links to other sites.

  5. Ron DeWitt said

    Arn and Charlie,

    The point that you may be missing is that the physics of the Tiljander series is not homogeneous. The early portion has been interpreted by the original authors to be a temperature proxy, with narrower banding indicating warmer temperature. A recent broadening of the bands was known to have been caused by construction that increased the amount of sediment, and the original authors rejected it as spurious noise. However, the recent broadening makes a nice hockey stick, so what was originally rejected as noise is now foolishly accepted as the signal.

  6. Ron DeWitt said

    I should have made clear that the foolish acceptance of the noise as signal was NOT done by the original authors.

  7. Jeff Id said

    #5 Nice way to make the point.

  8. Jeff Id said

    #4 Is there anything new in this? It looks like the stuff we already have and not related to the raw data we need to verify?

  9. Raven said


    Santer and his buddies take the position that you do not calculate confidence intervals by calculating the variance of the data set because if you use too many data points the confidence intervals become too small and the models look bad.

    Instead the use the outlying model runs as the confidence interval – i.e. as long as the actual trend is higher than the trend of at least one run from one model in the ensemble then they claim that the trend is consistent with the models.

    This approach is probably justified if you have many runs from the same model. The ensemble of runs would create a probability distribution for the process being modelled. The trouble is I do not believe such an approach makes any sense for ensembles of runs from many different models with different parameterizations/physics. But what do I know – I am just an engineer.

  10. crosspatch said

    “The trouble is I do not believe such an approach makes any sense for ensembles of runs from many different models with different parameterizations/physics”

    Easy to test. Does it work for hurricanes? Sounds like just the kind of ensemble modeling that is done with them. While the last hurricane of the season was still in the Gulf of Mexico, every single one of the models in the ensemble had the storm going ashore in Florida, tracking across Georgia, and moving out to sea almost in a due Easterly direction. Not a single one of those models was correct. The storm tracked right up the East coast of the US. So much for models.

  11. Phillip Bratby said

    Myles is presumably Myles Allen (review editor for IPCC FAR). Who is Francis?

  12. Manuel said

    These model-versus-observed trend
    differences are, however, of no practical significance whatsoever

    In other words, models can’t be used to make long term predictions as we suspected.

    McIntyre has absolutely no understanding of climate science

    Or, the people behind IPCC that have used models to project climate until 2100 have absolutely no understanding of climate science. Thank you very much for confirming it Mr. Santer.

    Common, you can’t have it all. Either models are accurate or they are not. If you can’t test them comparing their output for the future with reality, you can’t use them to predict what will be the future.

  13. Ursus maritimus said

    Just noticed that you spelled Chiefio (Cheifio) wrong in his blog link. Funny, I regularly link through to his blog from there, and just noticed now.
    Just a head’s up. No need to post this.
    Keep up the great work. The cubs and I are counting on you!

  14. Kenneth Fritsch said

    Jeff ID, the background on the Santer controversy to which you refer is primarily this:

    When Douglass et al (2008) did their calculations they were comparing the observed and climate modeled tropical troposphere temperatures using standard error whereas some of the previous researchers had made comparisons using standard deviation and range (Tom Karl). Obviously what Douglass was attempting to avoid using were model results where some of the very unrealistic models could provide the range or standard deviation that make the comparison with the observed very uncertain. In Karl’s case using range he only had to include some ridiculous model results to show that the model range covered the observed results. Douglass complained that in some cases the model inputs for the troposphere had associated surface temperatures that were obviously unrealistic but were included to give a range that encompassed the observed results.

    Obviously anyone can see that if you have infinite sample size that the standard error (SE) goes to zero and then any slight difference between observed and modeled results will be significant. This is an argument that Gavin at RC was putting forth against the Douglass use of SE. If anyone can find a statistic book that says the number of finite samples used by Douglass would preclude using SE I would like to see it. As far as I know Gavin’s argument came out of whole clothe. SE is used everyday in these kinds of statistical comparisons.

    Anyway when we had a chance to read the Santer article, surprise, surprise they also used the SE and not standard deviation for a comparison, but unlike Douglass used the SE for the observed results. I doubt that anyone pointed to the egg that should have been on Gavin’s face.

    They were many other issues with the Santer article amongst which were the failure to extend the time period and to make a direct comparison of troposphere temperatures to surface temperatures as a ratio or difference. They included some new adjustments of radio sonde data that widen the observation range. All in all, what Santer provided in his paper was that in order to show that the observed to model differences were not significant they had to show that model results were so wide ranging that great doubt should be placed on the model results as a whole.

    It sounds to me that Santer is back into the Gavin argument about using SE – perhaps he was never convinced of its use by his associates in the first place. Now he wants to say with the same false authority that I hear so often from these people that the few added data points that Steve M used makes use of SE no longer valid. Unfortunately, I think the game these people play is for a criticized author like Santer to make generalized replies, that because they do not deal with the specifics sound authoritative, and then the consensus fellows use that reply as their authoritative reference.

  15. Matt Y. said

    So… he’s whining that the confidence interval is tighter because SM used a longer time period. Aren’t they always claiming that longer time spans are where the models excel? That while they might mess up short term weather, they are as good as gold at predicting long term climate? Yeesh.

  16. PatrickG said

    #8 Re #4

    You are correct. There is nothing new in the “data” that RealClimate has posted. My guess is that they posted this to make it appear that they are complying with the request to “free the data” and “free the code”. The comments are very self-congratulatory.

  17. MikeN said

    Is using personal e-mail valid science? Shouldn’t you be able to get them on the record, or have the journal verify? It’s annoying having to explain Michael Mann’s error when Steve uses the source pers comm in his comment. At least the graph is pretty clear in the Tiljander paper.

  18. timetochooseagain said

    So basically…What they are saying is that smaller confidence intervals are bad, because they give a false impression of the sureness of the trend if it’s “too long”. Bullshit. And this “structural uncertainty” handwaving is starting to piss me off. Does anybody have a actual reason why the satellites “must” be wrong that hasn’t already been addressed????

    This deal with the satellites disagreeing with the models has been going on a long time. The modelers have screamed victory no less than three times as all the issues were uncovered-each time the satellite data had to be revised, they triumphantly declared that now the models were vindicated. It never quite happened. But without a reason for the satellites to still be wrong, I am really baffled at the legs the modelers still think they have to stand on.

    In this case, Santer prides himself on making this hypothesis tests really easy to pass.

  19. Charlie said

    Mike N — citing a “personal communication” is a normal way of giving credit to someone for either data or an idea that has not been published.

    Nothing unusual about it at all. If Steve McIntyre hadn’t cited the personal communication with Santer, then Santer would probably be complaining about McIntyre failing to give him credit.

  20. Layman Lurker said

    #18 TTCA

    Also consider the email references to sattelite metrics. IMHO this is the most blatant case of confirmation bias that I have seen. They obviously single out UAH with contempt, yet don’t seem to fully understand what each metric represents, or the possible sources of bias – diurnal drift, etc. UAH = bad; RSS = not so bad.

    Some of the discussions of RSS I think can be tracked back to recent posts done by Chad and Jeff. Land / ocean / ENSO etc.

  21. curious said

    4,8 – they have said:

    “…but if anyone has other links that we’ve missed, note them in the comments and we’ll update accordingly.”

    which is not quite the same as saying they’ll give you it if you ask for it but I think specific requests should be worth a try.

  22. […] By timetochooseagain The fallout from Climategate continues. Today, Jeff Id has found this gem: From: Ben Santer To: P.Jones Subject: Re: Good news!  Plus less good news Date: Thu, 29 Jan 2009 […]

  23. timetochooseagain said

    Exactly. They just say, “that data must be wrong, don’t know why, but it must be”. One is left with the distinct impression that it is “wrong” because it disagrees with the models…

  24. I think “Francis” is Francis Zwiers.

  25. Jason said

    Santer is sort of correct here.

    if the models predicted warming of 0.02 degrees per year, and the actual warming was 0.019 degrees, after a great many years of data were collected, it would be possible to claim that the difference between the models and reality is statistically significant.

    The again, we aren’t talking about thousands of years of data. Elsewhere, Santer complains that the calculations are being made after just 20 years. Either the calculations are too soon (in which case Santer 08 is fatally flawed) or too late, but not both.

    This paper by Steve is a very big deal. It proves the models wrong, and it uses Santer’s own math to do it. Santer needs to do better than this.

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: