Juckes and the Moberg CVM

I vaguely remember an English joke about how certain Oxford colleges qualified prospective fellows. They’d invite them to a garden party and serve them a bowl of cherries and see how they disposed of the pits. I forget what the candidates were supposed to do.

In the Team Euro Moberg composite, Team Euro has constructed the composite using 10 of the 18 Moberg series, most of which you’ve heard about: Yamal (Briffa version), Tornetrask, the Yang China composite (with HS from two Thompson ice cores) and the Arabian Sea G. bulloides (showing coldwater upwelling), Taymir tree rings, Chesapeake Mg/Ca, a China stalagmite and two odd bristlecones, Methuselah Walk and Indian Garden – both lower border chronologies, related, if anything to precipitation, and without the 20th century growth spurt of Sheep Mountain, Campito Mountain, Boreal and Uppwer Wright, and finallly the GRIP borehole series.

I’ve accurately emulated the Team Euro Moberg composite as shown below using the coloring of yesterday’s exercise.


Team Euro Moberg composite. Black – archived; red- emulation; blue- discrepancy.

Now Moberg used 18 proxies – so what happened with the 8 "cherry pits"? These included the Sargasso Sea dO18 reconstruction – since this is an actual temperature reconstruction (rather than a measure of coldwater upwelling), to the extent that the inclusion of the proxy is being justified on the basis of representation tropical ocean temperatures, the Sargasso Sea proxy is a more defensible selection. Another one is the Indigirka, Siberia chronology. I’ll post more on this on another occasion, but, in my opinion, this chronology – which is the newest – should absolutely be included in any reconstruction as it is independent and offers an excellent opportunity to test the HS as a hypothesis of temperature history.

So below, I show on the left the Team Euro Moberg composite with 10 Moberg proxies and on the right an identically calculated composite replacing Yamal with Indigirka and Arabian Sea G. bulloides with Sargasso Sea temperature estimates. The medieval-modern relationship is reversed.

By the way, the correlation of the right panel reconstruction (the CA variation of the Moberg composite, if you will) is indistinguishable from the Euro Team Moberg composite in its correlation to the EHT instrumental record (r -0.57; EHT version r – 0.58).

9 Comments

  1. john lichtenstein
    Posted Nov 2, 2006 at 6:47 PM | Permalink

    SteveM please calculate RE as well. Remember it’s the RE maximizing proxies that are the *right* proxies. (Per Amman & Wall and also Gavin.)

  2. bender
    Posted Nov 2, 2006 at 7:29 PM | Permalink

    Nice. A bootstrapped reconstruction would have shown confidence intervals sufficiently wide that this outcome would have been entirely within the range of predictability. Instead we must rely on your keenly honed intuition. Thanks.

  3. Steve McIntyre
    Posted Nov 2, 2006 at 10:38 PM | Permalink

    #2. I haven’t parsed bootstrap definitons. But I can’t believe that the Team is doing bootstraps within any form of statistical definition – their “bootstraps” are as artfully cherry picked as their reconstructions.

  4. bender
    Posted Nov 2, 2006 at 10:58 PM | Permalink

    They give bootstrapping a bad name.

  5. Posted Nov 6, 2006 at 4:05 AM | Permalink

    We used proxies which extended from AD1000 to AD1980 (as stated in the manuscript). The Sargasso see series does not extend to 1980, so was excluded.

    Regarding “cherry picking”, I think we must have a language problem: our robustness test evaluated the effect of removing each of the 18 proxies. If you start selectively removing series and inserting series which don’t match the original selection criterea then you could obviously achieve almost anything.

  6. Willis Eschenbach
    Posted Nov 6, 2006 at 5:09 AM | Permalink

    Re #5, this would only make sense if you had selection criteria that applied to all proxies. But you don’t. You pick one southern hemisphere series without criteria. You choose one of two polar urals series without criteria. Finally, you omit the Indigirka series, which goes to 1993 and thus fits your criteria.

    w.

  7. TAC
    Posted Nov 6, 2006 at 5:24 AM | Permalink

    $5 Martin, thank you for responding to so many of the comments.
    I was surprised to read above

    If you start selectively removing series and inserting series which don’t match the original selection criterea then you could obviously achieve almost anything.

    Actually, if you do it right — repetitively subsample or even supersample (with replacement) the target population — what you actually achieve is the bootstrap variance. If, as you say, that could be “almost anything” — well, you’re not going to get any argument from me.

  8. Steve McIntyre
    Posted Nov 6, 2006 at 5:46 AM | Permalink

    We used proxies which extended from AD1000 to AD1980 (as stated in the manuscript). The Sargasso see series does not extend to 1980, so was excluded.

    Martin’s statement is, of course, false. Here ‘s a plot of the Sargasso Sea series taken from Moberg’s Supplementary Information.

    The original data is at ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/keigwin1996/fig4bdata and to show this more clearly, he is a re-plot:

    The Indigirka series also not used by Team Euro is also shown in Mober’g SI here:

    Here is a re-plot from digital data.

    As noted, swapping Yamal for Indigirka and the Sargasso Sea for the Arabian Sea yields quite different results – a point which Juckes has not refuted.

  9. TAC
    Posted Nov 6, 2006 at 5:56 AM | Permalink

    #7 was meant to begin with “#5″ not “$5″. I just noticed the error, and I apologize.

One Trackback

  1. [...] (Juckes and the Moberg CVM) [...]

Follow

Get every new post delivered to your Inbox.

Join 3,310 other followers

%d bloggers like this: