Data Comparisons

Climate Prediction Quote of the Day

9 Comments

This is from Taylor and Gekler’s (2007). (Power Point Presentation.)

The graph prediction index vs. climate prediction skill is also interesting.

It’s generally dangerous to try to conclude too much from power point presentations. I can’t help but wonder, how does “J. Murphy” even know the absolute predictive skill in the earth’s climate sensitivity? Does Murphy know the earth’s climate sensitivity? Or is everything compared to some other climate sensitivity?

I guess I’ll have to hunt the J. Murphy reference down.

9 thoughts on “Climate Prediction Quote of the Day”

  3. Good God, the dreaded quadrant graph.

    There must be someone from a Human Resources Department involved here someplace.

  4. I would say the choke on that shotgun is worn. Better replace it.

    They should try the center of the target also.

  6. Note Taylor’s classic summary (emphasis added):

    The scientific community, funding agencies, and policy makers are interested in “which model is best?”
    * This question is not specific enough.
    * Although single “performance indices” can be proposed, there is currently little rigorous scientific justification for paying much attention to them.

    K. E. Taylor & P. J. Gleckler (2007), Summary, pg 27

  7. Ultimately, how can there be a difference between model “performance” and “quality of prediction” ? If one has really good “predictive skills” with respect to what climate will actually do, how can that person be said to have low “climatology skill”?
    —Gentlemen, our findings show that Model Variant #R17b.2.1 rev.4 performs really well, it’s just that it’s always wrong…
    I get a spooky sensation that there may be climatologists who simply resent reality for lacking the elegance and clarity of models…all climate behavior is just an ongoing, rather crude exception to a rule that is obvious to the modeler.

  8. Yes, in that graph, everything is compared to some other climate sensitivity, not the Earth’s climate sensitivity.

    The graph says “perfect model test”. In climatology, a “perfect model experiment” (called a “simulation study” in statistics) is where you start with a model with known properties (like climate sensitivity), generate hypothetical observations from it, and see how well your statistical technique can reconstruct the known, “true” value.

    The purpose of a perfect model experiment is not to say anything specific about the real climate sensitivity, but to see how accurately your analysis methods can determine it in a “best case” scenario (one where you don’t have to worry about model structural error, weather noise which deviates from your statistical assumptions, etc.).

  9. George,

    I don’t know what they’re doing, but it looks to me like they’re trying to evaluate some climate skill measure (perhaps one they’ve invented). They then compare how well that measure agrees with the actual error made in inferring climate sensitivity. The last can’t be known in reality, but it can be known in a perfect model experiment (see above). This way they can use it as a “reference truth” against which they can see how well their skill index works.

    P.S. “J. Murphy” is probably James Murphy of the Hadley Centre (UK Met Office), who works on the ClimatePrediction.Net project.

Comments are closed.