Mann 2008: the Bristlecone Addiction

I see that the BBC is taking the position that the “evidence” of an MWP-modern differential now is confirmed in so many “independent” studies that the matter is now “incontrovertible”. However these studies are not “independent”; the vast majority of the “independent” studies use the same stale proxy data. In my Erice presentation (which I’ll post up in a day or two), I showed the difference between updated versions of the Tornetrask, Polar Urals and Sheep Mountain data and the older versions used in Team studies, observing that use of the updated versions altered the medieval-modern differential in 9 of 10 canonical studies (the issues in Moberg involve a couple of different proxies, but are similar.)

In regards to the MWP-modern differential, Mann et al 2008 is once again not “independent” of the other studies – the entire suite of Graybill strip bark bristlecones rear up once again. Very few of the “new” proxies covering the MWP contain a relevant HS shape and there are fatal problems with the ones that do (e.g. the Korttajarvi sediments) making one question why these sediments are used in a supposedly serious study.

In a previous post, I made a flash summary of the non-dendro proxies in Mann et al 2008 and I urge interested readers to review that post.

Today I’m providing a flash summary of all the dendro series with SI start dates prior to 1010. While there has been considerable discussion of the use of “tree ring” series in Mann et al 2008, the real issue is the use of Graybill strip bark chronologies, which continue to be used without apology or discussion as to their validity. In the flash image, the HS shaped series labeled as ca534, nv…, co524 etc are invariably Graybill bristlecone strip bark chronologies. The version showed below are non-infilled in the latter portion. Other non-“independent” proxies used over and over are Briffa’s Tornetrask, the Jacoby-d’Arrigo Mongolia series. There is remarkably little that is both new and relevant to a modern-medieval differential, despite the puff.

The continued use of the questionable Graybill strip bark chronologies is highly objectionable, especially in view of Abaneh’s inability to replicate the most important Graybill chronology (ca 534.) It is also objectionable in view of the following puff by Mann et al:

We were guided in this work by the suggestions of a recent National Research Council report (35) concerning an expanded dataset, updated data, complementary strategies for analysis, and the use of thoroughly tested statistical methods.

As has been widely observed, the NRC panel recommended that “strip bark” chronologies be “avoided” in temperature reconstructions – a suggestion obviously not adhered to by the Mann group. This is doubly disappointing as the chairman of the NAS panel, Gerry North, is said to have been a reviewer of the Mann et al 2008 paper, but he seems to have taken no steps whatever to ensure that strip bark chronologies were avoided, though this issue should have been on his mind.


Figure 1. Mann 2008 Dendro proxies standardized to 1400-1980 with 21 year smooth illustrated. Non-infilled. There is one series in the list that doesn’t go back to 1010, but this appears to be an error in the underlying SI and has been left in.

The Pea under the Thimble
The Mann study claims that they can “get” a HS shape without dendro (i.e. without Graybilll bristlecones):

When tree-ring data are eliminated from the proxy data network, a skillful reconstruction is possible only back to A.D. 1500 by using the CPS approach but is possible considerably further back, to A.D. 1000, by using the EIV approach. We interpret this result as a limitation of the CPS method in requiring local proxy temperature information, which becomes quite sparse in earlier centuries. This situation poses less of a challenge to the EIV approach, which makes use of nonlocal statistical relationships, allowing temperature changes over distant regions to be effectively represented through their covariance with climatic changes recorded by the network.

A skillful EIV reconstruction without tree-ring data is possible even further back, over at least the past 1,300 years, for NH combined land plus ocean temperature (see SI Text). This achievement represents a significant development relative to earlier studies with sparser proxy networks (4) where it was not possible to obtain skillful long-term reconstructions without tree-ring data.

But watch the pea under the thimble: the completely defective Korttajarvi sediment chronologies are used in the comparandum series. Yes, the anthropogenically disturbed sediments have a HS pattern, but so what? Surely no one can seriously argue that this presents valid evidence on the modern-medieval differential (whichever way it goes.)

As noted before, in the SI to Mann et al 2008, they purported to account for the defective Korttajarvi sediments by doing a sensitivity analysis without them. But once again, watch the pea under the thimble: the Graybill strip bark chronologies are used in this comparison. Mann et al:

We therefore performed additional analyses as in Fig. S7, but instead compaired the reconstructions both with and without the above seven potentially problematic series [including the four Korttajarvi series], as shown in Fig. S8.

As far as I can tell, this is not a comparison of non-dendro proxies with and without the defective Kortaajarvi series, but before and after of a network which has multiple Graybill bristlecones in the network.

This is what passes as “incontrovertible” evidence in climate science.

32 Comments

  1. Gary
    Posted Sep 15, 2008 at 12:57 PM | Permalink

    This situation poses less of a challenge to the EIV approach, which makes use of nonlocal statistical relationships, allowing temperature changes over distant regions to be effectively represented through their covariance with climatic changes recorded by the network.

    And just how do we know this covariance isn’t spurious?

  2. Steve McIntyre
    Posted Sep 15, 2008 at 1:00 PM | Permalink

    The correlation between Korttajarvi sediments and temperature change is totally spurious. Red meat for the Mannomatic II which doesn’t care.

  3. mugwump
    Posted Sep 15, 2008 at 1:05 PM | Permalink

    As noted before, in the SI to Mann et al 2008, they purported to account for the defective Korttajarvi sediments by doing a sensitivity analysis without them. But once again, watch the pea under the thimble: the Graybill strip bark chronologies are used in this comparison.

    This was exactly my first thought on reading the paper: leaving out one defective proxy at a time proves nothing if there is more than one defective proxy in the network and their algorithm mines for defective proxies.

    I suggested this somewhat tongue-in-cheek on an earlier thread, but in all seriousness a more robust test of their approach would be to use artificial modern temperature data with, say, the trend reversed, and then see if their algorithm can discover a “signal” from the modified data. If it can, you know they are overfitting, and their technique is not robust.

  4. Steve McIntyre
    Posted Sep 15, 2008 at 1:10 PM | Permalink

    I agree that this would be a useful exercise, but it takes a bit of time to parse through what they are doing. It’s helpful that they archived quite a bit of code, but the methods are not standard ones and it may tkae a bit of time to understand their properties. Ian Jolliffe recently said that he did not fully understand the properties of Mannian principal components, about which a great deal was been written. So what are the properties of Mannian RegEM EIV – well, we know that it weights HS shaped series more heavily than CPS.

    • mugwump
      Posted Sep 15, 2008 at 1:16 PM | Permalink

      Re: Steve McIntyre (#5),

      I agree that this would be a useful exercise, but it takes a bit of time to parse through what they are doing.

      I’ll say. I thought about doing it myself but after spending a couple of hours with the code, the paper, and the SI, I shelved it. It would take me a couple of weeks to reproduce what they are doing which I just don’t have right now.

  5. Jordan
    Posted Sep 15, 2008 at 1:11 PM | Permalink

    Slightly OT. The BBC is screening a Horizon edition on Tuesday (16 Sept, 21:00 BST). Leading scientists share advice on global problems (advice to presidential candidates on AGW by the look of the trailers).

    It was Horizon that gave us the “Global Dimming” edition. Once a great TV programme, but has gone down hill in my estimation. Like a quality magazine turned into a comic.

    How many complaints will be provoked by tomorrows screening? Probably none – but at least OFCOM did not rise to the risible “peer-reviewed complaint”.

  6. Steve McIntyre
    Posted Sep 15, 2008 at 1:32 PM | Permalink

    #7. ACtually it would take longer, because then you have to think about what it is actually being done, and it takes a bit of time.

  7. Craig Loehle
    Posted Sep 15, 2008 at 1:50 PM | Permalink

    Mann claims that CPS uses local temperature records:
    “We interpret this result as a limitation of the CPS method in requiring local proxy temperature information, which becomes quite sparse in earlier centuries.” but WHAT prior centuries? Isn’t the CPS method just calibrated to local data from the 20th century? By using EIV which calibrates “nonlocally”–ie to HadCRU(a hypothesis that has never been tested and has no physical reality) the mining for hockey sticks is even easier.

  8. Jeff Alberts
    Posted Sep 15, 2008 at 1:57 PM | Permalink

    I’ve got a Bristlecone in my front yard. Do I now qualify as a Dendroclimatologist?

    • Gary
      Posted Sep 15, 2008 at 2:16 PM | Permalink

      Re: Jeff Alberts (#11),
      Only if you measure it, hide your data, run a mystical statistical routine, and claim it predicts the rain in the Seine.

      Steve: As readers of MM2003 know, the rain in the Seine falls mainly in Maine.

      • Jeff Alberts
        Posted Sep 15, 2008 at 2:48 PM | Permalink

        Re: Gary (#13),

        I can do that. But I’ll need a LOT of funding. Highest bidder gets the results they want. Oh, and my tree is as old as it needs to be.

  9. jae
    Posted Sep 15, 2008 at 2:13 PM | Permalink

    Pea under the thimble is right. snip

  10. John A
    Posted Sep 15, 2008 at 4:12 PM | Permalink

    This is doubly disappointing as the chairman of the NAS panel, Gerry North, is said to have been a reviewer of the Mann et al 2008 paper, but he seems to have taken no steps whatever to ensure that strip bark chronologies were avoided, though this issue should have been on his mind.

    He probably just “winged it”. That’s all the attention it requires.

  11. Robinson
    Posted Sep 15, 2008 at 4:24 PM | Permalink

    Indeed #24, I do remember their “Earth Day” being shelved due to its perceived political bias. The story was reported in The Times here. It seems to me, that a TV documentary about the HS debacle would make excellent viewing, but whom in our media would be willing to fund and produce such a program?

  12. John F. Pittman
    Posted Sep 15, 2008 at 5:35 PM | Permalink

    C Brook: I do not know if you have seen Lucia’s posts on rank-exploits. It is hot-linked in the CA blogroll. Perhaps the easiest way is to show how similar they are, such that if the satellite data is bad, one would assume the surface data is as well.

  13. Ben
    Posted Sep 15, 2008 at 6:12 PM | Permalink

    From the OP:

    But watch the pea under the thimble: the completely defective Korttajarvi sediment chronologies are used in the comparandum series. Yes, the anthropogenically disturbed sediments have a HS pattern, but so what? Surely no one can seriously argue that this presents valid evidence on the modern-medieval differential (whichever way it goes.)

    As noted before, in the SI to Mann et al 2008, they purported to account for the defective Korttajarvi sediments by doing a sensitivity analysis without them. But once again, watch the pea under the thimble: the Graybill strip bark chronologies are used in this comparison.

    I therefore stand by my remark that the recon is not sensitive to the removal of any one series. But are you suggesting that the robstness test was inadequate? That the rcon is not robust to the removal of any two series?

    AH, CRAP. I can’t play this role any more!!! I’m movin on …

    • mugwump
      Posted Sep 15, 2008 at 7:34 PM | Permalink

      Re: Ben (#32),

      you might want to see my comment at #4.

      Leave-one-out cross-validation requires pretty strong algorithmic stability to establish any kind of useful error bounds. See eg here and references therein. I very much doubt Mann’s procedure qualifies.

      • bender
        Posted Sep 15, 2008 at 9:18 PM | Permalink

        Re: mugwump (#35), I know you’re right. I knew it from the start. But I had to play the role of bender_team_member. It was simply a question of how many days it would take before someone put their finger on IT – the flaw that no one can reasonably defend and which kills the stick.

        Also, the failure to use Ababneh’s updated data is incomprehensible. I would like to hear that one defended.

        [P.S. I see at RC that you are now getting the silent treatment. That's too bad. Your questions make sense to me.]

        • mugwump
          Posted Sep 16, 2008 at 6:27 AM | Permalink

          Re: bender (#38),

          Ah, team-member-bender.

          Mind-bending.

          I keep forgetting.

    • DeWitt Payne
      Posted Sep 15, 2008 at 9:25 PM | Permalink

      Re: Ben (#32),

      It was fun while it lasted. Thanks.

      They tend to get rid of Devil’s Advocates in industry too. Something about not being a team player.

  14. Sylvain
    Posted Sep 15, 2008 at 7:16 PM | Permalink

    I find it interesting that for something that is not supposed to be important, many people gives it a lot of importance.

    The journal Nature proudly claim that the hockey sticks is holding true. It seems to me that they should show more reserves in supporting any new paper by Mann.

  15. John Lang
    Posted Sep 15, 2008 at 8:03 PM | Permalink

    So, once again, Mann is using the bristlecone stripbark pines for the dendro reconstruction and he is using a contaminated Finnish lake sediment core for his non-dendro reconstruction.

    Just looking at all the gif animations supplied by Steve, it is apparent that these two series must be heavily weighted in the two different reconstructions.

    Any idea yet Steve, how much weight was included for these two.

  16. deadwood
    Posted Sep 15, 2008 at 8:26 PM | Permalink

    As far as I can tell, this is not a comparison of non-dendro proxies with and without the defective Kortaajarvi series, but before and after of a network which has multiple Graybill bristlecones in the network.

    I detect some uncertainty in this statement. Or are you as flabbergasted as I am that Mann can be so brazen in his deceit?

  17. Posted Sep 15, 2008 at 10:17 PM | Permalink

    The enriched uranium tips of your wry and gentle humour, Steve, are a pleasure to read… and savour. You make “devastating” so easy on the eye…

  18. Posted Sep 15, 2008 at 10:29 PM | Permalink

    mugwump #6,

    I’ve spent some time trying to do a hack approach to make as inverted a hockey stick as I can. The best the latest mann data has anyway (that’s all I have to work with now) is a flattish curve. I just found that the HS shape was magnified in this data prior to insertion in the statistical meat grinder though. The infilling produced a huge kick at the end of the data series. I did a big post on it tonight.

    I guess my point is that for this modified data, I doubt any negative signal can be extracted.

    At least from this post, I understand now how series which stop 500 years ago can be calibrated. The NV037 series had half its data lopped off, I bet it didn’t calibrate well before pruning.

  19. Hoi Polloi
    Posted Sep 16, 2008 at 2:25 AM | Permalink

    I don’t understand Mann being a recidivist. Why didn’t he move on as earlier stated? He must have known that all his new papers will be painfully scrutinized.

    At the other hand:

    ad·dic·tion (-dkshn)
    n.
    1.
    a. Compulsive physiological and psychological need for a habit-forming substance: a drug used in the treatment of heroin addiction.
    b. An instance of this: a person with multiple chemical addictions.
    2.
    a. The condition of being habitually or compulsively occupied with or or involved in something.
    b. An instance of this: had an addiction for making hockey schtick graphs.

  20. Posted Sep 16, 2008 at 4:53 AM | Permalink

    The proxy formulation that results in the reconstructed hockeystick is definitely one of the undead. However, the most important proxy, the attribution of each molecule of CO2 as natural or anthropogenic is not even born. The calculations behind the isotopic signal of ACO2 are impenetrable to me — perhaps someone somewhere has a reasoned argument which is accessible. Without that scientific reasoning then we are left with post hoc ergo propter hoc, CO2 is going up, humans are emitting more, therefore humans are causing atmospheric CO2 to rise. I find this less than convincing. One argument, the measurement of various flows, looked promising until I noticed one uncertainty of plus or minus 50 gigatons per year. I laughed and moved on to ‘we’ve accounted for all the flows so we know it must be anthropogenic’. I moved on again but this time did not laugh.

    When I saw the first headlines of ‘hockeystick confirmed’ I was alarmed: when I saw the name of the author I felt relieved — nothing to see here, just more strange proxies. It would be a great relief to be able to feel the same about ACO2 isotopes, the most telling proxies of all.

    There is hope. Much of the AGW business is political and when the politicians feel it is expedient to look deeply into this farrago of nonsense then it will vanish. Perhaps a depression with its concomitant need for good news will trigger the change.

    JF

  21. Posted Sep 16, 2008 at 6:01 AM | Permalink

    Steve,
    I have really enjoyed reading your blog. Thank you for doing this. Are there basic descriptions of each of the proxies available?

  22. Rusty Scott
    Posted Sep 16, 2008 at 6:14 PM | Permalink

    Way off topic:
    What’s up with the numbering of comments? The numbers in the links don’t match the numbers to the right of the comment. It makes it hard to follow the responses.

    Great information here though, so I’ll keep lurking in spite of the difficulties.

    Steve: I moved a lot of BBC comments to a new thread.

    • DeWitt Payne
      Posted Sep 16, 2008 at 10:47 PM | Permalink

      Re: Rusty Scott (#30),

      It makes it hard to follow the responses.

      That’s why it’s so important to use the reply and paste link when replying to a post. As long as the post hasn’t been moved to a different thread or deleted entirely, clicking on the generated link will take you to the original post. Even if it as been moved, the comment number, which shows up in Firefox at least at the bottom of the page when you mouse over it, won’t have changed. That makes searching other threads for a moved comment a lot easier.

  23. Plimple
    Posted Sep 17, 2008 at 11:41 AM | Permalink

    Steve,

    In figure 1 the start and end points appear to drop to zero. Is this the nature of the actual data, or a product of your plotting code? I’m interested in looking at the data and this is preventing me from making any sort of judgment. Can you explain this please?

    Also, having looked at your earlier plots of the proxies and having commented (thank you for the reply) I notice that the plots now seem to end at different points, but before they all ended in 2000. It’s difficult for me to understand why Mann 2008 dendro proxies end at different points but that Mann 2008 non-dendro proxies all end at 2000. Any clue as to why this happens?

    Steve: Yes. Good question. I plotted up the non-dendro proxies using the archived proxy data. We discovered to our surprise that Mann had “infilled” data up to 1998. Since then he archived his “original” data, in two versions. I did the dendro plots using non-infilled data. So there’s a little inconsistency – arising from Mann’s bizarre archiving. At some point, I’ll re-issue the non-dendro flash chart using “original” data.

  24. MartinGAtkins
    Posted Sep 16, 2008 at 2:12 AM | Permalink

    Re: MC (#30), “On a general note can anyone help me get real station data. Its not that easy. I’m looking to start comparing temp, precipitation, CO2 levels and the like against proxies.”

    No CO2 though. Use the little map on the left to navigate you way to a surface station.

    Weather Underground,

2 Trackbacks

  1. [...] Mann 2008: the Bristlecone Addiction [...]

  2. By It's on... on Jan 17, 2009 at 11:58 PM

    [...] on 15 Sep 2008: While there has been considerable discussion of the use of “tree ring” series in Mann [...]

Follow

Get every new post delivered to your Inbox.

Join 3,298 other followers

%d bloggers like this: