Climate modeler Matthew England still ignoring reality – claims IPCC models will eventually win

The FT isn’t going to be useful on multidecadal scales, but with 165 years of data it is probably not terrible for timescales up to a decade or even two. Again, the point of the analysis arises from the fluctuation-dissipation theorem. Chaotic or not, the FT is going to reveal important things about the dissipative mechanisms. For example, if one looked at the data on a monthly scale, the FT would clearly reveal the annual global temperature oscillation that is in counterphase with the eccentricity. The temperature data alone would tell us that a “year” exists and that something changes in the the way heat flows through the system in association with the year.
At some point the FT of a finite data segment will be dominated by artifacts, of course, so I wouldn’t take the 67 year cycle clearly evident in the data seriously even though it is a very strong peak. 165 years isn’t long enough to resolve a 65 year period, and chaos probably does do unusual things to the dissipative modes (like spontaneously switch attractors and hence multidecadal periods). But the point I’m making is that (most of) the models don’t get the short time dissipation right. They don’t get it close to right. They get it so wrong that it is perfectly clear that they don’t get the physics itself right. Whatever structures are emerging that transport heat and ultimately dissipate it in the fluid flow are the wrong structures with the wrong timescales and the wrong length scales, and I’m fairly certain that this extends all the way down to spatiotemporal scales we cannot see — I’d bet that they are off on an hourly, daily, weekly, monthly basis as it is the cumulation of this that we see as being spectrally off across the range of periods from 1 to 15 years (which ought to be resolvable with a 165 year base — that’s 10+ periods, enough to make artifacts from end effects a roughly 10% effect or less).
To give you a metaphorical model — suppose you have a marble that is rolling on a surface without slipping or dissipation. The surface it rolls on is not flat, however — it consists of many interconnected (smooth) nvalleys that wiggle from side to side and run roughly parallel and with ridges in between the valleys that can vary between no height at all to quite high. If you give the marble a random starting push along any initial valley, it will basically run up one side of the valley as it curves, then roll down and overshoot and roll up the other side, then roll back down and up the other side, repeat. Every now and then, however, instead of hitting a confining wall it makes it over the variable height ridge and into the next valley. We can imagine that the valley minima themselves are arranged on a larger scale, much flatter valley. We can even imagine a fractal distribution of the valleys and can imagine giving the marble a gentle push or a strong push, giving it the energy to easy jump between valleys at some scales but not others.
If you plot the motion of the marble transverse to its direction of motion as a function of time, local dynamics (within the current valley) gives you a “relaxation time” back to the local equilibrium at the bottom of the track. A period would emerge that tells you about the width and curvature of the locally confining valley. However, it also jumps over into nearby valleys, which might have a somewhat different width and curvature. Over time it can drift away from the initial valley altogether, although if the valley bottoms are themselves on a larger scale valley the marble will find itself confined on a larger scale and a new relaxation time will emerge as the marble “diffusively” bounces between the larger scale valley energy bounds. There can be many such times, and there is nothing that says that the width of single valleys has to remain constant or that the the larger scale valleys of valleys remain constant.
Still, as you accumulate data you learn a lot about the structure of those valleys. Things like an overall tip in the valley floors can emerge as the marble has a clear bias and is more likely to jump to the next valley on the right than on the left, although it can be difficult to differentiate this from simple cumulation of a random walk with no bias at all in a completely untipped set of valley floors.
That’s a pretty good metaphor for what the climate is doing and what we can infer from the temperature record. The climate is locally reasonably stable. It warms for a bit, then it cools for a bit, bouncing off the wall on the right that keeps it from running away in that direction, then off of the wall on the left that keeps it from running away in that direction. Sometimes it slips over into a nearby valley that has a slightly higher or lower bottom, but once there it finds itself still confined and the timescale of its confinement remains roughly the same. This results in a picture of “punctuated equilibria” — in a state of local equilibrium for a while, but with enough variability as it goes over bumps, finds holes in the confining walls, and is sometimes given a kick in a random direction by a pesky lizard that it jumps to a new nearby equilibrium now and then.
Discerning structure of the valley floors themselves is much more difficult, of course. For one thing, the valley floors aren’t usually very different in height. For another, Poissonian distributions have long tails, so it takes a lot of data to resolve bias in right vs left drift as the marble jumps valleys, and we cannot see the big picture of the surface on which the marble rolls.
In this metaphor, CO_2 plays the role of a tip in the valley floors. As it increases, it makes it a bit more likely to jump to the valley on the right with slightly higher mean (local equilibrium) temperatures than the one on the left with slightly lower temperatures. But HadCRUT4 makes it pretty clear that there is other large scale structure that can produce at least as much of a slope in the opposite direction, hence the “pause” in warming (or even cooling) from 1940 to the mid-1970s in spite of steadily increasing CO_2 throughout that period, hence the current “pause”. A correct dynamic model should capture both the gradual accrual of the tip, and the decadal dynamics of sloshing from side to side around the local equilibrium. A really good one would capture the multidecadal variation as well, which is almost certainly not (just) chaos per se, but is rather reflective of the named self-organized dissipative structures that we know have a significant impact on the climate because that’s how they were discovered and named — they are usually climate structures! These are things like the multidecadal oscillations (especially), the thermohaline circulation, the solar cycle, hurricanes and typhoons, thunderstorms, “winter”, monsoon(s), blocking highs, with timescales ranging from minutes through several decades out to as many as a thousand years (that we know of). There are even much longer timescale motions (Milankovitch stuff) and quite possibly unknown stuff thrown in on top. We have a pitifully short base of instrumental observation of any sort, and an even shorter base of global observations made with good/modern instrumentation from e.g. satellites or ARGO buoys or halfway decently sited weather stations. IMO, no more than 60 or 70 years, and arguably only 30 or 40 or even 10 or 15 (ARGO hasn’t been online very long yet).
The climate models fail this simple test long before they fail in other ways! They don’t oscillate with the right sub-decadal spectrum, and that spectrum is well resolved in the data! Worse, they overshoot badly — the marble careens up too-steep sides too fast and too far, and overreact to lizard kicks in the form of e.g. volcanic eruptions. Finally, the worst of the models just tip the damn plane to where the marbles just jump to the right every year. They never even regress to their local equilibrium because the model creates so much tipping from CO_2 that it exceeds the slope of the local valley wall. You can easily see these trajectories above — they just steadily get warmer, year after year, without ever cooling at all, without ever having a “pause” as long as a single year!
This is what is absurd in the figure above. How can any sane scientist with the slightest understanding of dynamical trajectories on rough landscapes take a look at reality over 165 years and the predictions of the CMIP5 models and not just pitch all but four or five of them directly into the dustbin? They are obviously wrong. Not because of chaos, not because of a lack of knowledge of initial state, but because they have the wrong sub-decadal dynamics. They also don’t match the multidecadal dynamics, but there we have inadequate real world data to positively resolve the problem and can only say that it is is highly probable that they have the wrong multidecadal dynamics. This isn’t a small problem, it is a large one. It is perfectly obvious that (most of) the models generate an absurdly large warming bias compared to the regressing local dissipation that makes local temperatures generally stable rather than otherwise and hence generate diffusive transitions to higher temperatures at far too great a rate, a random walk that is no longer random because instead of taking steps to the right (say) 55% of the time as opposed to 45% to the left so that a random walk on average drifts right, they’ve just built a model that takes steps to the right 100% of the time.
Seriously? And you want us to take this seriously?
Let’s reduce it to even simpler terms, Bayesian terms. You are given a coin. The coin has some probability of heads and some probability of tails, but you don’t know what they are! They could be literally anything, because a clever magician built the coin and you have no idea how he weighted it. Worse, the coin has a memory — successive trials aren’t independent! And you don’t know what the autocorrelation behavior is, how much getting heads on one flip influences the probabilities of the next flip.
You also have a computer. You wish to build a model of the coin that you can use to predict its long term behavior.
You start by flipping the coin 165 times. In that sequence, roughly 55% of the time the coin comes up heads, and roughly 45% of the time it comes up tails. But the heads and tails are clustered — you are more likely to see (say) three heads in a row or three tails in a row than you would expect if you just flipped a pure bernoulli trial coin, although in the long run there are slightly more runs of three heads than three tails and so on. As you always expect, there are a few runs of much more than three heads or three tails, and if you form the cumulated distribution function as you flip with 1’s given to heads and -1’s to tails, you see a cumulant that gradually grows but has stretches where it goes up several times in a row as well as down several times in a row.
Now you build your supposedly microscopic model! Curiously, however, your coin always lands on heads, and even more amazingly, lands at an angle so that instead of getting occasional tails, it just gets a fraction of a head. It never produces a single tail, let alone three or four in a row.
Forget about whether or not you have tweaked this model so that the angle distribution manages to smoothly interpolate the observational cumulant with its ups and downs. The model is not a model of the coin! It is the model of an imaginary coin, a coin that doesn’t exist. It doesn’t have the right probability of heads or tails, it doesn’t have the right autocorrelation behavior. It is a Bad Model. Its cumulant trace will never look like any data collection built with the real coin, and one cannot even reasonably argue that it is a predictor of the mean behavior of the cumulated distribution of coin flips because it has absurdly different and nonphysical dynamical behavior. You would do just as well fitting an arbitrary function with roughly the right shape to the cumulated distribution and not bothering to call it a microscopic model and pretending that this function can be extrapolated to predict the future of the CDF.
That’s how bad the worst CMIP5 models are, but even the not-quite worst models don’t come close to having the right decadal spectrum or fluctuation properties or dynamical responses to e.g. aerosol boluses. And yeah, then they fail to predict LTT, rainfall, drought, frequency and violence of storms, the rate of SLR, and the list goes on. Basically, they don’t correctly predict much of anything run to run, and the statistical distribution of outcomes (per model) doesn’t come out anywhere near the one result we actually can measure, reality.
This is not repaired by superaveraging the models together. Why would it be? You don’t get closer to a correct model of the coin flip by averaging any number of absurdly incorrect ones that allow coins to land at an angle and never produce a tail.
If England really wants the models to be taken seriously, he can start by rejecting models that fail a fairly detailed hypothesis test, one model at a time, in comparison with reality. I won’t even begin to address the nonsense of picking model traces that have “some” hiatus and extrapolating them to show that it isn’t quite impossible for the models to be right and yet for there to be such a profound disagreement between model results and reality all the way back to 1850 and still have crazy-talk 5 C warming by 2100 because you cleverly retain the models that have a slightly higher chance of landing the coin on the edge only slightly tipped towards heads so that they can generate a “pause” of ten or fifteen edge-on flips in a row while still cumulating those head-biased large angle flips like crazy most of the time.
rgb

I’m not sure what you are asking — why CO_2 is increasing in the first place or why temperature is increasing. Personally, I think that CO_2 is increasing at least in part because we are burning large amounts of coal. I think nearly all other forms of burning are nearly irrelevant in comparison. I have no opinion on the half-life of carbon in the atmosphere, but think it is plausible that the burned coal will stick around in the carbon cycle for quite a long time with at the very least more carbon in play before nature “sequesters” it in reservoirs at a new semi-stable equilibrium. But what “a long time is” I could not say, and am not convinced by any of the carbon models so far as I think that we still lack too much information to assert that one is “correct” and alternatives that also “work” are “incorrect”.
As for why the temperature is on average increasing — it is very plausible that a significant fraction of that increase, as much as “all of it”, is due to the increase in CO_2. If you look at the simple model I fit up above, this assumption fits the data remarkably well, and is backed not by complex microscopic model physics but by very simple, mean field physical arguments. On average, more CO_2 in the atmosphere should warm the surface in fairly predictable ways within some range of possibilities, and the observed best fit is within the general ballpark of that range. However, the data also clearly demonstrates that there is more going on than just CO_2, and since we cannot reasonably separate out or predict this unknown component in a simple radiative model fit, the fit must necessarily lead to a large uncertainty. The total all-feedback climate sensitivity to CO_2 could be as low as (probably just above) 0 C, say a few tenths of a degree C, although this low is pretty unlikely. It is probably in the ballpark of 1 to 1.5 C, and it also could be as high as 2 to 2.5 C, although at this point this high is pretty unlikely as well.
Getting aerosols right will substantially improve this sort of estimate. Figuring out thunderstorms and the water cycle and feedbacks will improve it still more. But ultimately, in a chaotic system, our best efforts are limited by a kind of “ceteris paribus” assumption that may not be justified — the assumption that the system itself cannot confound our best efforts to describe it in simple terms by spontanenously shifting one of its major modes of dissipation. For example, the emergence of the so-called “Pacific Hot Spot”. Suppose that this is a new, stable mode for the entire Pacific Ocean circulation. Suppose that it has the effect of basically permanently blocking ENSO phenomena, or else limiting them to being consistently weak and biasing the basin towards La Nina events. It would simultaneously shift the entire average global distribution of heat and change everything in the long run, quite possibly causing the planet to heat, or to cool, completely differently than models that assume a simple progression of a predictable PDO and semi-predictable ENSO would predict.
There is strong evidence that the planetary climate does this sort of thing without help all of the time on millennial timescales. California was mostly desert for a very long time, then it switched to become comparatively wet for most of the last couple or three centuries. It could switch back to being desert dry in a heartbeat. It may have already switched. Not because of CO_2, but because the multidecadal oscillations in the Pacific switched, because of pure chaos. We might never be able to predict such a switch, except to be almost certain that within some sufficiently long timeframe the switch will occur. That’s chaos for you.
CO_2 might (paradoxically) actually help suppress such switches and eventually give us a much more predictable and stable climate. Its increase is already directly responsible for roughly 15% of the food put on the world’s table every day, and it may be a factor in the slight decrease in the frequency and violence of storms, although that isn’t a well-supported conclusion statistically.
It may well be that increasing CO_2 will eventually lead to a “climate catastrophe”. Up to now, the additional CO_2 has almost certainly been beneficial and we might well have chosen to burn lots of coal just to get the benefits, as over a billion people are dining today courtesy of the additional CO_2. The world has never been lusher or greener in recent history, except where we persist in cutting down trees to burn them for fuel, and burning coal instead helps out with that as well. It is quite plausible that our climate “optimum” would be 450 or even 500 ppm. It is not at all clear that running it up to 1000 ppm will be totally beneficial, or even beneficial in excess of catastrophic. Personally, I think that RCP 8.5 is fairly absurdly unlikely even if we do nothing, or nothing more than is already being done. If thermonuclear fusion has indeed been cracked by e.g. Lockheed-Martin or others working on the problem (or is cracked in the next 10-20 years) the issue is totally moot for all time — we will not in this case come anywhere close to 8.5. Even without fusion, thorium and uranium fission, if we ever get over the collective fear associated with nuclear energy, would keep us from getting close to 8.5. Photovoltaic solar will probably keep us from getting close to 8.5 as its differential cost drops and the differential/marginal cost of coal based energy increases, especially if there is any sort of breakthrough in storage. We need to wait on and invest in these technologies, not engage in panicked and expensive measures that even if implemented at the cost of an ongoing human catastrophe right now would have little impact on our eventual “carbon footprint” by the time we quit using coal one way or another.
In the meantime, there are roughly 2 to 3 billion really, really poor people around the planet that are starved for energy, the fundamental resource that puts limits on nearly all other economic scarcities. Given enough, cheap enough, energy, we can literally make the deserts bloom by desalinating the oceans, we can mine scarce material out of seawater or build them out of the air. We can make fertilizers and develop access to clean water and sewage treatment in countries where the bathroom for hundreds of millions of humans is the field behind their back door, or it would be if they had a door. I am fortunate to live in an air conditioned house with clean water and super health care and more food than I can or should eat every day, all built with and enabled by cheap, copious energy. Who am I to tell those people “Sorry, you have to live another couple or three generations in abject poverty because we won’t permit the coal burning generation of electricity that would convert your poverty to plenty“?
Who is anyone? Especially with the CO_2 score so far at Benefits — 1 billion fed as a statistical certainty, every day; Penalties — absolutely statistically invisible against the background of confounding variables.
rgb
[Well summarized. Thank you. .mod]

“Penalties — absolutely statistically invisible against the background of confounding variables.”
Or Penalties: “drought, 9 feet of snow in 4 weeks in an urban city, mass migration to cities in the middle east, increased terrorism, border tensions between Bangladesh and Pakistan, riots in Sao Paulo (forecasted), no more skiing in Tahoe (sorry 2-3 billion poor people, we gave at the office), and potentially what is most catastrophic is the risk of dramatically lower grape harvest in Napa. Florida can flood, not a catastrophe.”