The WUWT Hot Sheet for Saturday August 31st, 2013

.Luther Bl’t:
Thank you for raising the penetrating question. To paraphrase this question, is it not true that a climate model makes a conditional prediction (e.g., if GHGs rise at rate A, then surface temperatures will rise by rate Y, plus or minus amount Z)? The answer is NO.
In other terms, a “conditional prediction” is a “predictive inference.” Logic contains what we know about the rules by which correct inferences may be distinguished from incorrect ones. Thus, your question has logical significance.
In response to this question, I’ll point out that a “predictive inference” is an extrapolation from an observed state of a system to an unobserved but observable state of the same system. By convention, the observed state is called the “condition” while the unobserved but observable state is called the “outcome.” An example of a condition is “cloudy.” An example of an outcome is “rain in the next 24 hours.”
A pairing of a condition with an outcome is a description of an event. The complete set of these events is an example of a statistical population. No statistical population underlies the IPCC climate models. Hence, the predictive inference that is made by such a model is undefined.
In a model that possesses an underlying statistical population, events of a particular description possess relative frequencies. The numerical values of these relative frequencies relate to the numerical values of the associated probabilities. The various probability values map to the information that is conveyed to a policy maker about the outcomes from his or her policy decisions in advance of these outcomes.
For the IPCC climate models, there is no underlying statistical population. Hence there are no relative frequency or probability values. Thus, these models convey no information to a policy maker about the outcomes from his/her policy decision in advance of these outcomes (http://wmbriggs.com/blog/?p=7923 ) . .
It follows that these models are worthless for the purpose of making policy on CO2 emissions. The appearance that they are worthwhile for this purpose is a product of repeated instances of the equivocation fallacy on the part of global warming climatologists. Among these climatologists are skeptics as well as their opponents.

“@terry Oldberg”
If the ipcc is only making projections and not predictions, then 1] it is not doing real science, and 2] there is therefore no basis upon which to argue that we should modify anything we humans are doing.

Bruce Cobb above misread what I wrote. I didn’t write “statists.” I wrote “stasists,” my term for folks eager to maintain existing energy norms in America. More examples: http://dotearth.blogs.nytimes.com/?s=stasists

richardscourtney:
In your post of Aug. 31, 2013 at 10:15 am you contend that I have yet again posted untrue nonsense and deplore the injustice of this infraction. However, your argument fails to support your contention.
My contention is stated in unambiguous and logical terms in the peer-reviewed article at http://wmbriggs.com/blog/?p=7923 . The referee for this paper, William Briggs, holds a master’s degree in atmospheric science and a PhD in mathematical statistics; he is a professor of statistics at Cornell University; formerly, he sat on the standing committee of the American Meteorological Society on probability theory and statistics. In a previously published peer reviewed paper http://judithcurry.com/2011/02/15/the-principles-of-reasoning-part-iii-logic-and-climatology/ I go over similar ground. The referee for the latter paper, Dr. Judith Curry, is a professional climatologist and is the chair of Earth Sciences at Georgia Tech University. In articles that are cited in the latter article, Gray, Green and Armstrong and Trenberth reach conclusions that are similar to mine. Gray holds a PhD in physical chemistry from Cambridge University and is a longtime IPCC expert reviewer. Armstrong is a professor at the Wharton School of Business of the University of Pennsylvania; he edits the Journal of Forecasting. Green holds a PhD degree and is a university professor. Trenberth holds a PhD degree and is a high-level professional climatologist.
As I reach my conclusion via a syllogism, this conclusion is true unless it can be proved that one or more of the premises to my argument is false or unproved. Your post of Aug. 31, 2013 at 10:15 am is devoid of such a proof.

If you haven’t seen Dr. Murry Salby’s utter destruction of the AGW hypothesis through a stunning mathematical analysis of the data, it is worth your time. You will need some physics and Math background. Even if you don’t have that, you will be able to follow the logic. I don’t think the IPCC will be inviting him for a presentation any time soon. Here’s the link:

part of the introduction in German but the rest in English.

Terry Oldenberg writes: For details on the equivocation fallacy in global warming arguments, see the peer-reviewed article at http://wmbriggs.com/blog/?p=7923 .
Rarely have I ever had the dubious pleasure of reading a piece of sophist crap like this, “peer-reviewed” (whatever that means in the context of a blog post) or not.
In addition to openly inventing terminology to create an absolutely artificial distinction — model vs “modele” — the only thing this article does is carefully delineate the following circumstances:
a) If the climate programs, built on top of what are loudly trumpeted to be physical laws and a variety of stated and unstated assumptions, each of which can be treated as a bayesian prior and weighted with a prior probability of being true, are models, they can be falsified by ordinary hypothesis testing and validated to the extent that they prove successful in predicting the future. They are science. They aren’t even “good” or “bad” science — science doesn’t use value labels — they are merely computational models of a general theory that should be given weight according to their predictive success, in their own way no different from abstruse model computations in condensed matter physics or predictive models built with empirical methodology that predict the stock market, customer preferences, or tomorrow’s weather. Even models built without an underlying theory — neural networks, for example — are without exception called “models” by every expert in predictive modeling in the world, where yes, I am such an expert. Show me a single textbook that supports your utterly artificial, bullshit “disambiguation” and we’ll talk. I’ll cheerfully show you a small stack of very current textbooks on pattern recognition, predictive modeling, scientific modeling, and Bayesian statistics where no such distinction is ever made because it does not matter.
b) If the climate programs (specifically, of course, the General Circulation Models or GCMs), built on top of physical laws incorporated in code and a variety of stated and unstated assumptions (bayesian priors) are modeles, you assert that by definition, they cannot be falsified by ordinary hypothesis testing or validated. This does not make the slightest bit of difference in how sensible people should treat their “projections”. They are computational “modeles” of a general theory that should be given weight according to their predictive success because mere common sense suggests that it is stupid to give much weight to non-falsifiable models that don’t work, whether or not they are technically “falsified” (technically because of course they are falsified — all one requires to be able to apply a hypothesis test is a null hypothesis, and model or modele, there is always a null hypothesis. One can, and people do, apply hypothesis testing to far more abstruse “modeles” than GCMs — for example a “modele” that claims that prayer affects outcomes in disease, claims of pure prophecy, claims of astrology. Note well the “modele” company you are putting GCMs into — religious claims are all modeles, are they not? Otherwise, what is it, precisely, that differentiates the two beyond a mere statement of purpose, a desire to protect a modele from falsification (as do the purveyors of any religious belief)? Not a damn thing.
Not only is your distinction, then, between modele and model absolute nonsense — we already have plenty of adequate terminology for predictions made on a religious basis, even those that “claim” to be a science such as astrology or other numerological nonsense and we do not hesitate to reject the use of those predictions in common political discourse on the grounds of mere common sense when the predictions thus made fail to come to pass — it is incorrect in the specific case of climate science where the GCMs are without any possible doubt science based predictive models that are without question subject to falsification, one at a time. If this were not the case, they would not be science at all, you twit.
Let me be specific. In both AR4 and (leaked) AR5, there are graphs of the output of certain specific GCMs when a Monte Carlo method is used to add noise to the inputs and various possible future trajectories are computed. The spread of these trajectories is then used to determine both a mean future climate and the envelope of statistical uncertainty around that mean, with the implicit null hypothesis of “this model is correct”. Let us state this in clear statistical terms. If this model is correct (as a prior assumption) and if we perturb the initial conditions enough to obtain a reasonable statistical spread of future outcomes, then the future trajectory should lie withing that spread with a probability directly related to the spread. In other words, we expect the actual future climate to be in the densest part of the distribution of possible futures, and to spend relatively little time out on the wings of the distribution of possible futures.
For most of the models that contributed to the predictions of AR4 and AR5 — where AR4’s summary for policy makers did not hesitate to make statistical claims containing percentages, even though those claims are utterly indefensible using any form of statistical theory I’ve ever heard of and hence the IPCC is treating them like models, not “modeles”, they are called General Circulation Models, not “Modeles”, and where their predictions are being used to make statistical assertions about the future as we expect of scientific models, not religious “modeles” — the actual future climate trajectory, relative to the time of the model’s construction and application, is at the outside envelope of a spread consisting of hundreds of runs. It is even worse than this, as none of the runs have either the right noise or variation, and even the runs that generate the envelope to not spend all of their time down where the actual climate has been, but rather individually drop down for short periods to help collectively define the range. The p-value that can absolutely be estimated per model, for most of these models is well under 0.01. The models can, for the most part, be rejected at the 99% or better confidence level even over the last 15 years alone, one at a time.
Collectively, they aren’t even a model or modele, they are just a bad mistake, an utterly indefensible application of the terminology of statistics in circumstances that do not satisfy the axioms of statistics as prior conditions for its relevance.
Note well that there are in the literature genuine efforts, underway at last, to test the GCMs. A recent paper, for example, compared four GCMs applied to the same toy problem — a uniformly heated, untilted “water world” — and got four future climates that bore not the slightest resemblance to one another. This too, is an elementary hypothesis test and we can thus say that each model is at most 25% likely to be correct because they cannot all be correct and a toy problem like this probably does, in fact, have a unique solution (we just don’t know what it is). Or perhaps it does not — which is even worse news for climate modelers, because if we cannot even get a single consistent answer for a toy problem, then what is the point of modeling, or “modele-ing”, any sort of future with GCMs and pretending that it has some predictive, projecting, or prophetic force?
To conclude, the issue is not sophistry like an artificial and arbitrary distinction between “models” and “modeles”, it is a much simpler distinction between science and religion. Science is by definition strongly falsifiable and weakly verifiable — any claim, whether “predictive”, “projective”, or “factual” made by a numerical computation implementing a scientific theory or hypothesis, can be disproven easily and rapidly by contrary observational evidence, and while it can never be fully proven by observational evidence, to the extent that it is in continual agreement we accord the theory a greater degree of belief. Good science is not dogmatic, it is flexible and it changes to accommodate contrary evidence so that it incrementally approaches a consistent theory that works to cover all or nearly all observations (where it is “nearly” all we consider the science to be “unsettled” — that’s where “research” comes in).
Personally, I think that GCMs are science. Science that has, in many specific cases (there are many different, distinct GCMs) been shown to be in very serious conflict with observation, conflict so profound that we should correctly doubt both the implementation of known physics and the stated and unstated hypotheses and the code itself (there are many places a complex computationall theory can fail). That is, they are clearly models as their name asserts (unsurprisingly, since nobody ever heard of “modeles” before you just made the term up as a religious apologia). If they are not science, they are religion (which might offend the climate scientists who think otherwise) but as far as politician, scientist, or lay public alike are concerned it does not matter. Call them anything you like, they are still wrong, still fail elementary hypothesis tests at the level of 99% on up, they are still things that mere common sense suggests should not be taken particularly seriously as they almost certainly contain serious errors in fact and implementation and are not even sufficiently consistent to give a single answer in implementation to a problem that is far simpler than the Earth’s actually noisy, tilted, highly elliptical orbit and its crazy array of oceans, solid surface, and biosphere.
As a professional predictive modeler, one that uses modeles such as neural networks to obtain results that cannot be connected in any way outside black (bayesian neuromathematical) magic to the predictions, this really is the bottom line. A model, or modele, either works or it doesn’t. If it doesn’t it is pretty useless no matter what term you invent to try to excuse it from the necessity of actually working. If it doesn’t work, you don’t get paid.
rgb

RACookPE1978:
Your argument is illogical. If this topic is of interest, I’d be happy to expend upon it.