Thompson et al [1993] on Dunde

Thompson’s Dunde ice core is an extremely important proxy in multiproxy studies. There has been an increase in dO18 levels in the 20th century. Whether this is a proxy for temperature is not at all obvious on physical grounds. The relationship between dO18 and temperature in monsoon ice caps is opposite to that of polar regions ” the lowest dO18 occurs in summer rather than in winter. Thus, a greater proportion of summer precipitation on the ice cap would cause lower dO18. The supposed connection between increased 20th century dO18 and temperature is entirely statistical. Thompson et al [1993] says:

The 5-year running means of dO18 from the Dunde Ice Cap are compared in Figure 5 with 5-year running means of Northern Hemisphere temperatures (Hansen and Lebedeff, 1987). For the period from 1895 to 1985, the correlation coefficient r is 0.5 (significant at the 99.9% level). This correlation suggests that the Dunde Ice Cap dO18 should serve as a good proxy for larger-scale temperature variations.

I’ve been meaning to check this for a while as it only takes a few minutes to do.

I checked Dunde against CRU NH temperature (I have this, not Hansen and Lebedeff, but the result should not depend on temperature data set). Of the various Dunde version in circulation, I used the Dunde version archived at MBH98, which visually looks like the version in Figure 5 of Thomspon et al [1993] and is available on an annual basis.

The correlation for the period 1895-1985 – the period cited by Thompson was 0.48, almost exactly the same as the reported coefficient of 0.5. Since Thompson is on the Hockey Team, you have to ask yourself why he only did the correlation from 1895 on. Any bets on what the correlation was for 1851-1895? Minus 0.36.

Reference:
Thompson, L. et al. "Recent warming": ice core evidence from tropical ice cores with emphasis on central Asia, 1993. Glob. Plan. Change, 7, 145-156.

13 Comments

  1. TCO
    Posted Sep 18, 2005 at 5:22 PM | Permalink

    1. What is the physical rationale for more isotopes when it’s hot? Does it have something to do with less stratification of the atmosphere of heavy molecules (boiling point of heavy water?) If so, this should be checked by atmospheric samples or by boiling point elevation experiments (or partial pressure observations).

    2. Maybe there is some good reason (related to physical reason for isotope seperation) for the proxy to work oppositely in polar winter and monsoon summer. If so, you could still get useful info from either, provided you had good foundational studies for each and that you used the appropriate calibration reference for each.

    3. I agree that leaving off the early stuff is cheesy and shows a tendancy to shading the truth that worries me. All that said, what is the total correlation (and significance)? Is there enough to make a somewhat reasonable proxy (and I think proxies need only be somewhat reasonable, since we use so many of them, use different types etc. in a big reconstruction.)

  2. Steve McIntyre
    Posted Sep 18, 2005 at 8:19 PM | Permalink

    In the tropics, the O18 effect is a rainout effect. O18 preferentially rains out. So by the time the monsoons get to the Himalayas (or to the Andes) the O18 is depleted – hence a larger negative anomaly. There have been lots of O18 measurements on the way from the East China Sea to the Himalayas and this is ironclad.

    Some ice cores show very detailed almost monthly O18 values. Visually the range of O18 in any given year looks surprisingly similar. A big part of the variation is simply the amount of negative O18 i.e. the amount of precipitation rather than temperature.

  3. TCO
    Posted Sep 18, 2005 at 8:24 PM | Permalink

    But if precipiation has any relation to temp (for a given ice field) then some signal of temp might be extractable. And if I actually care about science, I don’t care that the calibration curve I use for the tropics is different from the poles…what I care is that the calibration curves are valid (same as trees).

  4. TCO
    Posted Sep 18, 2005 at 8:26 PM | Permalink

    you still haven’t answered 3. How much does leaving that off change the overall answer. Yeah, it shows some moral issues either way…but what is the effect? We need to know that.

  5. Steve McIntyre
    Posted Sep 18, 2005 at 8:31 PM | Permalink

    Then somebody has to prove that it does. When they started the tropical ice cores, I think that they got blindsided when the dO18 relationship to temperature (on an annual basis) was the reverse of the poles. They have to argue that dO18 has a negative relationship to temperature on an annual basis, but a positive relationship to temperature on a century basis, which is a tough sell.

    The “amount effect” is now distinguished from the temperature effect by many specialists and at least a sizeable minority do not appear to believe that dO18 is a temperature proxy in tropical glaciers. I haven’t seen this caveat expressed by IPCC however.

  6. TCO
    Posted Sep 18, 2005 at 8:37 PM | Permalink

    Of course. But someone has to prove all the proxy foundational rationales. But if we find a tree that is impacted by too hot temp and we can do the right transforms and all that and end up using it as a proxy, that’s fine too. and would be no reason to cackle. It’s all about the foundational rationale. the quality of the calibration study. and the assessment of confounding factors.

  7. John G. Bell
    Posted Sep 18, 2005 at 10:05 PM | Permalink

    Re #4: TCO, the overall answer was 42. Does that help? :) Don’t know about moral issues. Try Google.

  8. John A
    Posted Sep 18, 2005 at 11:05 PM | Permalink

    What’s the r statistic for the entire dataset 1851-1985?

  9. Dave Dardinger
    Posted Sep 18, 2005 at 11:23 PM | Permalink

    BTW, John Bell, concerning the answer 42 did you realize that the ‘wrong’ question “What is 9 x 6?” actually is correct if you are working in base 13? I suspect lots of mathematically inclined people figured that out on their own but I was pleased with myself for being one of them many years ago.

  10. James Lane
    Posted Sep 19, 2005 at 5:33 AM | Permalink

    I’d also like to know the r statistic for the 1851-1985 data.

    Is it possible that Hansen & Lebedeff used 1895 as a starting point, and Thompson carried it over? That would be careless rather than cherry-picking.p

  11. Steve McIntyre
    Posted Sep 19, 2005 at 7:35 AM | Permalink

    James,
    Hansen and Lebedeff go from 1880 on; but CRU was available back to 1854. You can’t assume that they didn’t look at both.

    Actually, what I was originally interested in testing here was the impact of the smoothing in the calculation, rather than the impact of the truncation – which is just as interesting.

    For the unsmoothed data 1851-1985, the r is 0.19; it increases to 0.20 for 1895-1985. The OLS t-statistic is 2.3 and with Newey-West standard errors, it’s about the same.

    Smoothing obviously changes hte autocorrelation properties and the effective degrees of freedom. For the smoothed data, the r increases to 0.39 from 1851-1985 (0.48 from 1895-1985). The OLS t-statistic is 4.94. However, with Newey-West standard errors, which is much more realistic with smoothed data, the t-statistic is only 1.15.

    Smoothing to achieve “significant” correlations without allowing for the effect of the smoothing on standard errors is a very common Hockey Team statistical practice. Jones et al [1998] has a whole table full of “decadal correlations”.

  12. TCO
    Posted Sep 19, 2005 at 6:28 PM | Permalink

    which number is more relevant to understanding the situation, the usefullness of the proxy? The smoothed number or the unsmoothed one?

  13. Posted Nov 10, 2007 at 11:50 PM | Permalink

    Steve mentioned this thread on on 11/10/07, topic #2335, comment #50 —

    Steve — IMHO, Newey-West greatly undercompensates for serial correlation, particularly when the dependent variable has been moving averaged, eg with a 5-year MA as here. Fitting an AR(p) to the residuals and using it to compute the autocovariance matrix is a big improvement, though even that still understates the adjustment, even if the errors themselves are AR(p) — the residuals will show less s.c. than the errors, and even if we saw the errors, OLS estimates of AR are biased in finite samples away from persistence.

    Even if the original errors were serially uncorrelated, after taking a 5-year MA, only every 5th observation will be serially uncorrelated. This means that you would have to consider only every 5th observation in order for OLS se’s to be correct, but it’s roughly what has to be be done to undo the effect of the averaging.

    If you really want to estimate a time-averaged relationship, a more efficient way to do it than throwing out 4/5 or whatever of the observations would be to simply regress the unaveraged dependent variable on the averaged independent variable. This will not introduce s.c. into the regression residuals, and will let you use the entire sample.

Follow

Get every new post delivered to your Inbox.

Join 3,205 other followers

%d bloggers like this: