Livina & Lenton: Consenseless Conclusion – Complete Cock-up

Posted: September 18, 2012 by Rog Tallbloke in Analysis, Incompetence, methodology, sea ice

Paul Vaughan has contacted me with information regarding an arctic ice paper which erroneously concludes that a ‘bifurcation’ or fundamental change to a two state dynamic took place in September 2007 when the Arctic ice minimum record was set. He says:

With less old ice there was a qualitative regime shift to a less damped and thus HIGHER AMPLITUDE annual cycle after 2007, BUT THERE WAS NO BIFURCATION.

Livina, V.N.; & Lenton, T.M. (2012). A recent bifurcation in Arctic sea-ice cover.
http://arxiv.org/pdf/1204.5445v1.pdf

Regrettably, the authors SEVERELY (cannot be adequately stressed) misinterpreted the stats they calculated. I sincerely hope this is instantly intuitively obvious to every reader here upon reading sentence 1 of section 1 (“bifurcation detection”).

After removing the mean seasonal cycle, the remaining fluctuations in sea-ice
area include some of order 106 km2 (Figure 2a).

Figure 2: Detection of a bifurcation in Arctic sea-ice area. (a) Sea-ice area anomaly, daily data with mean seasonal cycle removed.

Following this procedure, the authors have then gone on to use this residual anomaly data to draw the conclusion that ‘climate change’ has made a fundamental change to sea ice behaviour in the Arctic, flipping it into a two state dynamic, with unknown consequences. The schematic in fig 4 summarises their findings:

Figure 4: Schematic model of sea-ice dynamics with changes exaggerated: (a)
before bifurcation, showing range of seasonal forcing and seasonal response of normal ice cover state, and (b) after climate change forcing, the system passes a bifurcation in summer to an alternative lower ice cover state, but reverts later in the season to the normal ice cover state.

But according to the comment on Arxiv by Ditlevson, the authors have fundamentally misundertsood their data because they haven’t considered how the underlying baseline period affects the shape of the data, and this leads them to misinterpret the situation. He demonstrates this graphically by showing what happens with different baseline periods:

Ditlevson points out:

If the change in the signal is such that the amplitude of the seasonal cycle change[s] through the series a part of the seasonal signal will remain in the residual. In this case the bi-modality found in part of the record could be a result of the seasonal cycle rather than the result of a two state dynamics.

Paul Vaughan observes:

This is the most interesting thing that has come up in the climate discussion in many months. I hope there will be climate blog articles and university-course student-exercises focusing on this error because it is so painfully instructive about what can go hopelessly wrong with blind, TOTALLY COMMON-SENSELESS application of anomaly-think.

It’s almost unfathomable that authors of this methodological caliber could make such a mistake. Overlooking this is like failing to notice that a sine wave oscillates. It’s like never emerging from a lapse during which one is so thoroughly hypnotized (to the point of cognitive paralysis) by mindless pursuit of black-box computation that common sense awareness of the annual swing from summer to winter is TOTALLY suppressed. How this could happen to “climate experts” (and additionally make it all the way to publication) is a painful question to have to ask.

The lesson here for students is that rote application of statistical computing in the total absence of basic common sense can lead to EPIC misinterpretation. It’s sobering (to say the least) to see that the existence of the annual cycle (plain old summer vs. winter) was totally ignored by climate professionals during interpretation of algorithmic output. This error could be destined to become a classic climatology & time series analysis textbook example. [of how not to do it - TB].

Wake up call:

Swings between 2 different states – NAMELY SUMMER & WINTER – were interpreted as NEW phenomena indicating climate instability.

This falls into the “wow – just wow” category of excruciatingly sobering blunders. There WAS a regime shift in 2007, but let’s not common-senselessly mischaracterize it.

____________________________________________________________________

Livina, V.N.; & Lenton, T.M. (2012). A recent bifurcation in Arctic sea-ice cover.
http://arxiv.org/pdf/1204.5445v1.pdf

Ditlevsen (2012). Interactive comment on “A recent bifurcation in Arctic sea-ice cover” by V. N. Livina and T. M. Lenton. The Cryosphere Discussions 6, C1187-C1190.
http://www.the-cryosphere-discuss.net/6/C1187/2012/tcd-6-C1187-2012.pdf

Aslak Grinsted commented on Ditlevsen (2012) here:
http://judithcurry.com/2012/09/16/reflections-on-the-arctic-sea-ice-minimum-part-i/#comment-240690

UPDATE 19-9-2012

Livina, a senior research associate at UEA, has replied to Peter Ditlevsen:
http://www.the-cryosphere-discuss.net/6/C1584/2012/tcd-6-C1584-2012.pdf

Ditlevsen was very gentle & diplomatic. It appears his comment didn’t entirely sink in.

More of the discussion can be found here:
http://www.the-cryosphere-discuss.net/6/2621/2012/tcd-6-2621-2012-discussion.html

The main hub for the article is here:
http://www.the-cryosphere-discuss.net/6/2621/2012/tcd-6-2621-2012.html

Comments
  1. JCrew says:

    Removing the seasonal cycle in the artic is pretty hard to do, but somehow they did it. So man can control the weather!

  2. tallbloke says:

    Heh, the problem is, they didn’t remove all of it, because as Paul says, the swings got bigger as the area got smaller. Less inertia, but still not flipping to any unpredictable ‘two state dynamic’ as they thought they could conclude.

  3. Roger Andrews says:

    To put things in perspective here’s a plot of Arctic sea ice extent showing monthly and twelve-month-smoothed means. We’re clearly a long way from having to push the panic button.

    Also note how smooth the twelve-year-mean plot is compared to Figure 2. This is because twelve-year averaging removes the seasonal variations while the “seasonal cycle” adjustments applied to Figure 2 don’t. Most of the wiggles on this Figure – and pretty much all of the wiggles after 2007 – are in fact generated by failure to remove all of the seasonal component.

    The next plot shows the amplitude of the seasonal cycle since 1988 (when the SMMI satellite became operational) with a best-fit regression line drawn through the data. Anyone who believes that the increase in 2007 represents a bifurcation is totally bifurcated up.

  4. Paul Vaughan says:

    I grew up near a large river. As kids we would swim in the river in summer and skate on it in winter. I could dig back through the weather records of my childhood home and pick an average summer and an average winter for comparison. By Livina & Lenton’s “logic”, such a comparison would senselessly indicate that the river was in the SAME state when we swam in it and skated on it.


    h/t to Judith Curry for drawing our attention to this stimulating case.

    After digging through the reviewers’ comments and the author replies, there’s plenty more that could be said (including about review failures), but there are too many other things to do with too little time. I suggest we at least keep an eye on the discussion at TCD to see what eventually gets published.

    Livina’s replies suggest she just doesn’t get (on multiple levels) what the reviewers are telling her. If the reviewers let her & her co-author Lenton get off with simply reframing the whole paper as a hypothesis, some quantitatively-inclined climate blogger will have effortless opportunity to rip Livina & Lenton’s “logic” to shreds (of shreds).

    I will volunteer a cordial warning to climate scientists: Expressing that you see promise in Livina & Lenton’s study claiming “a recent bifurcation in arctic sea-ice cover” INSTANTLY INCINERATES the ILLUSION of your quantitative integrity.

    One thing Livina & Lenton could be encouraged to do is stick to applying their beautiful methods to simpler time series. (Without substantial modification, the methods (in their current incarnation) can’t be meaningfully applied in 4D spatiotemporal contexts with natural periodic state alternation.)

    Alternatively, the authors could be forced to take a year or two to deeply (and I mean DEEPLY) reassess interpretation. The authors have shown something highly worthwhile, but it isn’t what they think it is. They’ve hopelessly (in a tragically extreme sense) misinterpreted the stats. Making already tragically hopeless matters only worse, the reviewers’ corrective efforts have been anything but commensurate with the severity of the error.

    Under favorable circumstances I would be willing to help. I can see FAR more important applications for the methods. If properly applied, I can see a way to use (suitable modifications of) the methods to isolate the very building blocks of climate. This would be serious, mature work demanding fulltime longterm devotion supported by substantial funding.

  5. Hello,

    The revised paper on sea ice has been accepted in the Cryosphere, you can see the new version in arxiv:

    http://arxiv.org/abs/1204.5445

    (pdf-file: http://arxiv.org/pdf/1204.5445v2.pdf)

    Best

    Valerie Livina

    [Reply] Many thanks Livina, we’ll take a look. Rog TB