Matt Ridley in the WSJ: Whatever Happened to Global Warming?

You have committed two more logical fallacies, the “Straw Man” and the “Red Herring.” I did not claim that millions of people around the earth are in great danger from the effects of GW/CC, yet that is the argument you attacked. I said, “if this article appeared in the New York Times, a Comments thread having a differerent tone and slant would be seen. There would be experts substantiating GW/CC models and agreeing that millions of people around the earth are in great danger from the effects of GW/CC. Is there no objectivity anymore?”
That does not mean that I imply that the danger to people around the world is inevitable or imminent. I may not believe that at all.
Indeed, your minimization of the effects of GW/CC is dangerously ignorant of the facts on the ground; emperical reality; that is, climate data collected and analyzed, and projections made based on clearly defined assumptions. This is not model data, but model validation data from the real world. This doesn’t mean these projections will happen with certainty; each has an associated likelihood based on its set of assumptions about future climate events and remediation implementations. .
In fact, it is you and those of like OPINION who are exaggerating, thus lulling the public and officials into a false sense of security. It is basically propaganda. You have no idea when global warming will pick back up and you have no idea how long it will last or how severe it will be, or what its acceleration will be.. There are climate scientists and atmopheric physicists who study this everyday and collect and analyze data everyday. Their job is to report their findings in scholarly journals with their research reports and findings subject to peer review, thus gaining a good measure of objectivity. .
Your self-appointed job seems to be to take pot shots at these experts from the sidelines, calling into question their character, competence, honesty and motives for doing their work. Why don’t you learn something about the subject and people at which your throw darts in a way reminiscent of “Blind Man’s Bluff”? That way your criticisms would carry more weight.
If you were competent to speak with athority on this subject, you would be a practicing climate scientist, and you would present your findings to substantiate your claims and be in line to win the Nobel Prize. Get with it; I wish you good luck. I you need an expert mathematical modeler, simulation expert, and big data analyst, I stand at the ready to lend you a hand.
G.

Dearest Mr. Grubbs,
Since you questioned my credentials, I am happy to provide them. Ph.D. in theoretical physics (dissertation on multiple scattering theory solution to single electron problem in crystal band theory). As an undergrad and grad student, after getting AP credit for calculus I completed two “introductory” math classes (multivariate calculus and linear algebra, the latter in the advanced section reserved for math majors). I then jumped to graduate courses in complex variables, PDEs, and functional analysis, all taught by Mike Reed (who literally wrote the book on functional analysis along with Barry Simon). In grad school I added two more graduate math classes — one on topology and one on the mathematics of classical mechanics (action principles, group theory, etc) taught in the math department, not physics. Six graduate level math classes (four taken as an undergraduate) was not a math major (I skipped e.g. number theory, ODEs, “advanced algebra” all required for a major anyway) and Duke did not offer minors at that time but OTOH they really covered everything but number theory to excess. I would honestly self-assess my competence in math (my research and dissertation were basically solving advanced PDEs) as easily equivalent to a BS in math, maybe even a masters, since at this point I’ve taught a two semester survey of math methods in physics that covered probability theory, ODEs, group theory, and so on, in addition to teaching quantum and electrodynamics, which is basically advanced vector calculus and algebra and group theory and right up against the edge of differential geometry (I sat in on around 1/3 of a course in differential geometry before research pulled me away as well).
I’m still weak on number theory although I’ve worked through about half of a decent textbook on the subject, but I’m strong on things nobody even offers courses in, such as geometric algebra (e.g. complex number, quaterions, generalized clifford algebras — graded division algebras)). There are probably a few holes compared to an undergrad major who completed all of the sequence but the holes are where I’m not that interested as a physicist more interested in solving problems than formal proofs concerning abstract objects. OTOH I have lots of places where my knowledge is likely beyond most math Ph.Ds. What I am not, credentialling aside, is incompetent in math. Quite the contrary. I can walk to a board and start teaching calculus through (common) ODEs and PDEs without notes, and with a tiny bit of prep could teach a lot more.
I’ve had only three courses in computer science — one intro, one on computer architecture, one on numerical methods — but again I, like many scientists whose work involves a lot of computation, have acquired professional competence in the field. If you google “rgb beowulf duke” you’ll still get around 15,000 hits, although back when the beowulf archives were online and the topic was happening it would have been closer to 250,000. I’ve been a professional systems/network administrator for 27 years, although I’m no longer particularly active in administration and my knowledge of current tools is probably limited (although easily refreshed). My personal subversion-based code tree is deep and wide and gigabytes in size. I periodically teach selected computer science majors independent study courses in CPS that count towards their majors with permission from the department. I’ve trained a number of systems admistrators. I wrote the first two versions of Duke’s security and acceptable use policies and was a primary faculty advisor to deans and vice-chancellors responsible for the development of Duke’s network as it evolved from twisted pair 56 kb serial lines to gigabit and internet 2.
I am co-founder (CTO) of an internet security company that is just getting off the ground. I have written a book on the engineering of beowulf-style compute clusters and parallel scaling, wrote a daemon-based open source cluster monitoring software package that was popular for a while. I have written a small mountain of predictive modeling software — primarily advanced, genetically optimized neural networks but a smattering of nonlinear regression code back before R came out and made it silly to write your own code to do that sort of thing — and have founded two companies (CTO again, contributing both code and knowledge of statistics, probability theory, modeling, AI, pattern recognition, etc) doing predictive modelling for money, and the current incarnation of the second one is still extant: http://www.arcametrics.com/ and making money.
The last 15 years of my research career were spent doing large scale numerical simulations on homebrew linux-based parallel supercomputers (hence my connection to beowulfery) — primarily importance sampling Monte Carlo in condensed matter studies of critical phenomena (second order phase transitions) and autocorrelation, numerically solving Langevin/Generalized Master Equation sets to model quantum optics, and so on. I would say that I’m fairly expert on many, but of course not all, aspects of large scale computer modelling. It’s a big field. I’m hardly a novice, though, either in statistics, probability theory, computer science, predictive modeling, in particular applying Bayesian methods in the construction of super-advanced, nonlinear, multivariate models — models of extremely high dimensionality (as in up to hundreds of dimensions) — not to publish papers in the field but for the real payoff — to get paid for it, in competition with lots of other highly competent folks trying to do the same thing.
Just for grins, I’m also the author of dieharder (you can google that, too), an open source suite of random number generator testing tools. One of my favorite books is E. T. Jaynes excellent “Probability Theory, the Logic of Science” — in a sense you could call me a disciple of Jaynes as he was also a primary reference for some of my work in optics although I never met him professionally. You can take it for granted that I am highly knowledgable about hypothesis testing, the Cox axioms as the basis for axiomatic probability theory and epistemology, quantified doubt as the basis of sound knowledge, the Bayesian “loop” of priors, models, and posterior probability correction based on empirical outcome, stuff like that, and not in any sort of ivory tower sense. In recent years, I have become something of a devotee of William Briggs, a professional statistician who spends much of his time, curiously enough, tearing the statistical basis of much of climate science to shreds because its statements are all too often indefensible nonsense.
My credentials in climate science per se are still fairly limited. I’ve been studying the field for five or six years at this point. I’ve worked through Grant Petty’s excellent book, “A First Course in Atmospheric Radiation”, although I may have mislaid my copy while teaching physics at the Duke Marine Lab this summer and will probably have to purchase a new copy. Most of the physics in it I’m already familiar with from working in and teaching E&M, quantum mechanics, thermodynamics/stat mech, and so on, so the book is comparatively easy for me to understand. I’ve also worked through Caballero’s textbook on climate science, which walks one through things like basic thermo (already known) and up to in-context discussions of things like the adiabatic lapse rates dry and wet, reduced temperature, atmospheric potential, the greenhouse effect. The one thing I haven’t really done much with personally is fluid dynamics — I know a lot about it (from teaching the intro levels if nothing else) but Navier-Stokes and the solution of actual physical fluid models near the various nonlinear critical points (onset of turbulence, for example) I have not done actual computations in. That does not stop me from reading about it, understanding the general structure of the PDEs involved, and appreciating the incredible difficulty of solving them. Mathematicians don’t even have existence proofs for NS — it is a grand challenge problem.
Which brings me to my own opinion on the subject, you can judge on any basis you like if it is ill-founded or ignorant. Climate models attempt to solve two nonlinearly coupled NS-systems — the atmosphere and the ocean — externally driven by both nearly periodic drivers and by what amount to “random” modulators, some of which are long-period internal feedback loops within the highly non-Markovian system itself with its broad spectrum of autocorrelation times. They attempt to do so on a grid no smaller than 100x100x1 km in size at the equator, with timesteps of ~100km/340 m/sec = 5 minutes (the time required for sound to propagate across the non-uniform lat/long cells, but absurdly long compared to the vertical propagation time or the oceanic propagation times). Since the Kolmogorov scale for nonlinear dynamics in the atmosphere is order of 1 mm, these spatiotemporal cells are order of 10^7 x 10^7 x 10^5 x 10^7 = 10^26 — 26 orders of magnitude — too large to have a good chance of actually solving the NS equations.
One can argue that one can coarse grain at this scale, create smoothed approximations of the cell dynamics, and solve systems like this and have the solutions obtained end up being meaningful, but there is no evidence that this is the case. In fact, there is substantial evidence that this is not the case, both from many other nonlinear chaotic systems where the minimum scale for microscopic dynamical models is the minimal scale for a reason (that reason being that you get garbage if you try doing the computation at coarser resolutions), from plain old numerical computation theory of quadrature where we know perfectly well that we cannot even numerically solve completely deterministic and boring systems like planetary orbits with a coarse timestep and little error control and end up with anything like the actual answer, especially after many timesteps, and above all, empirically — because the models suck at predicting anything outside of their reference period.
I strongly suggest that you download Chapter 9 of AR5 and read it in its entirety, but especially sections 9.2.2 and ff — where they acknowledge that the multimodel ensemble mean is completely meaningless as far as being supported by the axioms of probability and statistics are concerned — and that no effort is made to subject the individual models of CMIP5 to a simple hypothesis test by comparing their individual runs to the data.
This is not done because if it were done, the game would be instantly over. Turn to figure 9.8a in AR5. This figure presents a spaghetti graph of the CMIP5 models (sufficiently jumbled together that it makes it difficult to see instantly how terrible they really are individually at predicting the climate) against both HADCRUT4 (primarily) and the meaningless MME mean. Note well that each curve in the spaghetti is already a perturbed parameter ensemble mean of anywhere from a very few to hundreds of runs (yes, this asymmetry of data contribution is a problem, see 9.2.2 and ff). Since the variance per model is nevertheless much greater than that of the actual climate (and since none of these averages come anywhere near tracking the actual climate) we can fairly safely assume that if we were to compare individual model runs to the climate their variance would be around an order of magnitude too large. This alone would be instant cause for rejection in the specific sense that the model in question is clearly inadequate to use at all as a basis for formulating public policy that is killing millions of people a year now by misdirecting funds that could be used to (for example) end world poverty and the preventable deaths of children into a hypothesis of “catastrophe” that has no sound theoretical or quantitative support.
The models do not just fail in the present. They also fail, badly, to hindcast the past. Note well that the collective MME mean spends roughly 90% of its time warmer than the actual global average temperature even in the past all the way back to the beginning of HADCRUT4. It utterly fails (as to the individual models) to reproduce the signficant climate variations of the early 20th century as well as the entirety of the 21st century. Indeed, the only time it does a good job is in the reference period!
You seem to claim to be a professional predictive modeller. How impressed are you by a model that fits its training data but fails to predict either trial data from the past or future data with any skill? What would you estimate the p-value is of the null hypothesis “this model is a perfect, valid model of the climate” comparing the outcome of each of the models in the spaghetti to reality, one at a time, in 9.8a? Eyeballing an crude estimate based on the probability of the MME being too high 90% of the time instead of 50% of the time given at least 10-12 independent “trials” (with a presumed autocorrelation time of a decade) we get what, exactly? And the MME is going to be an upper bound on p for almost all of the PPE means, and the PPE means themselves aren’t even the proper basis for the hypothesis test — one really has to determine the probability of getting the actual climate given the envelope and distribution of PPE results, not just global surface temperature anomaly but results in several other dimensions — global distribution of rainfall, prediction of LTT, etc. Seriously, anyone who knows anything at all about modelling would reject (almost all) of the CMIP5 models out of hand, one at a time or collectively, if their skill were rigorously compared to reality. If you were shown this data, told that it was a prediction of some stock in the stock market, would you invest in it? Only if you were a serious gambler. If you’d bet on AR3 predictions to within almost 100%, you’d be broke today./
In real science, rather than fund-me-to-save-the-world science, nobody would be surprised by the failure of the models to have any skill. We wouldn’t expect them to, as they are just modified weather models and we already know weather models have no skill as little as weeks out. We wouldn’t expect them to, as we already know that dynamical models of this type do not fare well when integrated at a scale twenty-plus orders of magnitude too large. We wouldn’t expect them to, given that the scale of integration is too large to resolve lots of meso- not even micro-, meso-scale energy transport phenomena that are critical to the climate system, such as thunderstorms, clouds, the water cycle in general, soil, vegetation, rivers and reservoirs, the UHI. All of these phenomena are rolled into ad hoc phenomenological terms for the cells that we are then told are “physics based” as if this is either true or relevant when solving the NS equations at an absurd resolution. The models as solved aren’t even conservative and have to be renormalized regularly to prevent drift, and all of this is with parameters for critical process that aren’t even set to be the same between models. Finally, the models in CMIP5 aren’t independent, aren’t unbiased samples drawn from some sort of pool of models, so we have no a priori reason to trust either the mean or any of the moments of collective statements about the climate. It would be a miracle if they worked.
This is why climate scientists are back-pedalling as fast as they can as the climate has finally started to deviate from the models by enough that no amount of statistical band-aiding or obfuscatory graphing of the models compared to reality can conceal this simple fact from the politicians who have been more or less deliberately misled for nearly twenty years now by a comparatively small, but powerful, group of other politicians and climate scientists who have effectively controlled the funding of the entire multidisciplinary science. The oceans are not rising any faster than they have from time to time over the last 140 years. The temperature is varying with the same general pattern it has followed for roughly the last 165 years. LTT’s have been flat for most of the time we have measured them with satellites. The global surface temperature anomaly has been flat for long enough that AR5 has an entire Box devoted to ex post facto explanation of this inconvenient truth, and the continuation of flatness for another whole year and continued semi-fizzle of the ENSO event that was supposed to “rescue” the models has — so far — encouraged the number of “explanations” for this pause to double in the literature in the meantime.
“Climate sensitivity” is in free-fall, down from Hansen’s absurdities of 5 C or more by 2100 to under 2.0 C in the current warmist literature and the data is looking like it is supporting an even lower estimate of 1 to 1.5 C. Indeed, all of the warming in the era where CO_2 has substantially increased — call it 65 years — has occurred in a single 15 to 20 year burst from 1978 (or 1983) to 1998. All of the rest of that stretch temperatures have been flat to slightly descending.
There is one simple way to understand this — a way outside of the complexities of climate models. If we assume even a single layer model like that expressed in Petty, we can easily — for a given set of absorptivities — find the fixed point, including the warming due to the GHE. This model is basically a linear model — the absorptivities and albedo are held to be independent of the temperature, and if one makes it into a dynamical model by assigning some heat capacity to the various reservoirs if one increases the initial temperature of the reservoirs over the fixed point it will be driven back to the fixed point. If, OTOH, the absorptivity is itself a function of temperature, it isn’t so simple.
Suppose absorptivity increases (approximately) linearly with temperature, also known as “positive feedback” (from e.g. water vapor) in the vicinity of an initially assumed, consistent, fixed point. Then increasing temperature increases absorptivity. There is no longer any clear fixed point. If average temperature increases monotonically with absorptivity, and absorptivity increases monotonically with temperature, even if the system has fixed point for some initial conditions it isn’t clear that that fixed point will be stable. Natural fluctuations will drive the fixed point itself into a random walk. Random walks are themselves not stable — they diverge like the square root of the number of steps. The system must have negative feedback to be globally stable to fluctuations in a Langevin-type model of driven dynamics plus random noise — its response to any positive fluctuation in absorptivity driven “gain” has to be to decrease, not increase the effect of the fluctuation on the average (fixed point) itself. Otherwise in a warm year, net water vapor content would (on average) increase, which would increase absorptivity from water vapor as a GHG, which would shift the fixed point up, so that the system did not return (on average) to the same fixed point it started from but to one a bit higher. Fluctuations up and down would not average out to zero, but to a steady drift either up or down, like a random walk in 1D. Only if increasing water vapor has a compensating nonlinear effect, such as increasing average albedo, to make net feedback negative will the system return to its former fixed point. Indeed, the nonlinear dynamical fixed point would have positive feedback if it was too cool, negative if too warm. It would resist both perturbations in the water vapor channel itself and additional forcing from CO_2, not augment it.
In other words, there is a very good reason — long before one builds models — for thinking that the system is moderately stable, and that the average temperature is in fact the sort-of fixed point where things like water vapor feedback change sign, so that warmer cools, cooler warms, by modulating the entire water cycle — not just absorptivity, but albedo and latent heat transport. If this collective nonlinear cycle produced positive warming feedback in the neighborhood of its fixed point(s), the system itself would probably be unstable and the Earth would be Venus. It does not. On the cooling side things are more worrisome as the Earth is at least bistable, maybe tristable, with both the current deep glacial oscillation as a known stable cold phase and a spectrum of shorter period not-so-deep glacial oscillations from the earlier part of the Pliestocene to worry about as a third locally stable state of the system. There is strong evidence in the ice core data that the Earth can plunge into cold phase in as little as a decade when conditions are right, given remarkably little alteration in the driving/forcing. Since we do not properly understand this bistable behavior (well enough to predict it with much skill) and we absolutely do not understand the underlying multivariate dynamics, quasi-particle large scale structures or dynamical variations in either driving (such as ENSO, the PDO, the NAO, the solar cycle, Milankovich stuff) it is very hard indeed to say if the Earth’s global state on geological time scales is unstable to cold phase transitions. The LIA suggests that it might be. There is little evidence that even the interglacials have a warm phase that is accessible that is more than a degree or so warmer than current temperatures, either in the record of the Holocene, in ice core data, or in longer term radiometric proxies, although the 600 million year record is one of steady, systematic variation right down to the current all time low of global average temperature.
rgb

rgbatduke
September 9, 2014 at 9:42 am
IMO the warmest interglacials, such as the Eemian (MIS 5), MIS 11 & MIS 19, were globally at least a few to several degrees hotter than now, & even in the Holocene, its warmest intervals have been more than one degree warmer than now.