On Recent Criticisms of My Research
By Dr. Roy Spencer
One of the downsides of going against the supposed “consensus of scientists” on global warming — other than great difficulty in getting your research funded and published — is that you get attacked in the media. In the modern blogging era, this is now easier to do than ever.
I have received many requests recently to respond to an extended blog critique by Barry Bickmore of my book, The Great Global Warming Blunder. The primary theme of my book was to present evidence that scientists have mixed up cause and effect when diagnosing feedbacks in the climate system, and as a result could have greatly overestimated how sensitive the climate system is to our addition of carbon dioxide to the atmosphere from fossil fuel burning.
For those interested, here is our most extensive peer reviewed and published evidence for my claim.
But for now, instead of responding to blog posts, I am devoting all the time I can spare to responding to peer-reviewed and published criticism of my work. The main one is Andy Dessler’s paper in Science from last fall, which claimed to find positive cloud feedback in the same 10 years of NASA satellite radiative energy balance (CERES) data we have been analyzing.
In his paper, Dessler dismissed all of the evidence we presented with a single claim: that since (1) the global temperature variations which occurred during the satellite record (2000-2010) were mostly caused by El Nino and La Nina, and (2) no one has ever demonstrated that “clouds cause El Nino”, then there could not be a clouds-causing-temperature-change contamination of his cloud feedback estimate.
But we now have clear evidence that El Nino and La Nina temperature variations are indeed caused in large measure by changes in clouds, with the cloud changes coming months in advance of the temperature changes.
And without going into detail, I will say it now appears that this is not the only major problem with Dessler’s diagnosis of positive cloud feedback from the data he presented. Since we will also be submitting this evidence to Science, and they are very picky about the newsworthiness of their articles, I cannot provide any details.
Of course, if Science refuses to publish it, that is another matter. Dick Lindzen has recently told me Science has been sitting on his critique of Dessler’s paper for months. Science has demonstrated an editorial bias against ’skeptical’ climate papers in recent years, something I hope they will correct.
In the meantime, I will not be wasting much time addressing blog criticisms of my work. The peer-reviewed literature is where I must focus my attention.
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
T(t0) does not have unlimited range, and does not lead to silly graphs like the one showing all the different paths.
Arthur’s “Roy Spencer’s six trillion degree warming” is ridiculous. He’s still confused about starting times, using different ones for the convolution integral and the exponential term, and separating unstable parts which cancel each other out when propagating into the past, claiming that the unstable excursion of the one demonstrates the model is wrong. That’s just stupid. We implement electronic circuits which evolve according to similar differential equations all the time, and they don’t fail to work because propagating a part of the difeq solution backwards in time blows up.
You guys are amateurs, and you are just trying to cause trouble. I’m done here.
One more thing: “I’m not saying he did that–he says in his book that he got his answers on the first try–but a “statistical” technique that can give an infinite number of solutions is absurd.”
Even if this were true, it is a criticism of form over substance, and I do not see the point, except to get up on a box and shout “Me smart! He dumb!” But, more to the point, it appeared to me that it was YOU who separated Cp into two components, leading to the ambiguity you decry. (If that is not the case, I don’t care. As I have shown above, your discourse is so crude, and seemingly calculated to score cheap points with your choir, that I have no desire to delve further and teach you things you should already have learned when you set yourselves up as authorities).
One more, and this really is the last…
“The real point of my analysis was to show that Spencer’s model was, as you note, trivial, solvable by any first-year control student. I wanted to make that painstakingly obvious.”
Simple models are used to demonstrate complex topics all the time. And, many, many models of complex phenomena can be replicated by simple first and second order models which simulate the dominant modes in a particular operating regime. So, all you’ve made painstakingly obvious is that Spencer is using a common framework for teasing out the properties of the system he is studying, and that you were unaware of this. Congratulations on your victory, if you call it that.
“Do you think Spencer understands this? Did he understood it when he touted solving it in his book?”
More chest thumping. What about YOU? Do you understand that the model shown can be a reasonable linearization and simplification of the underlying complex nonlinear dynamics? Do you know how the time constant and equilibrium temperature can be derived via partial differentiation of those dynamics (if they are known), and might change over time, depending on your local operating condition? Do you understand … oh, never mind. If I asked you, I’d have to explain it to you.
Bloody… this popped up in between my posts. Really, this is the last.
“As I understand it, you were saying that A0 ought to be zero… which is what Arthur and I both said in our critiques. So in essence, it seems you agree with at least some of our criticisms of Spencer’s work?”
Based on what? Was T(t0) at equilibrium, or is it now?
In the absence of that knowledge, it is fully reasonable to use it as a parameter to fit the data. It doesn’t say that is TRUTH, but it does say it COULD be. And, the burden is on the AGW fanatics to prove it isn’t, or at the very least, to explain RIGOROUSLY (not, “oh, we just don’t think so, and we’re real smart”, or “we don’t know any reason why it would be” – those are ad verecundiam and ad ignorantiam arguments) why it is unreasonable.
Bart,
I didn’t make up the model, and I didn’t make up the parameters I used. That was all Roy Spencer. The graphs with all the different model lines were meant to show how important it was that Spencer started his model so far out of equilibrium. The point was to show that he could match the first half of the 20th century pretty well simply by manipulating THAT. He could never fit the second half very well. So if you came away with the impression that it’s absurd to start this model really far out of equilibrium to fit the temperature data, you were right. You just failed to recognize that this is exactly what both Arthur and I were saying.
I used Roy’s model in essentially the same way he did to obtain THE SAME RESULTS. I just took the extra step of showing why his methods were nonsense, and how one could get a wide range of answers using such a technique.
Form over substance, indeed. His entire case for the PDO-climate connection was based on his modeling efforts, so the “form” WAS the “substance” of his claim.
Bart,
By 1900 the temperature had been pretty stable for at least 50 years, according to the HadCRUT data set Spencer was using, so it seems reasonable to assume that if any period in that temperature record was near equilibrium, it was then. In contrast, the period Spencer chose to normalize to was 1961-1990, which had a strongly positive trend.
So why did Spencer choose that as his base period? The only reason he gave in his book was that the climate experts who put together the temperature series chose that base period (even though their choice had nothing to do with any assumed “equilibrium), and he made a big deal of it being right in line with his curve fit! Fitting a model with 5 adjustable parameters just isn’t impressive, however.
You keep accusing Arthur and myself of blustering to make ourselves look smart, but the fact is that in my critique I explained in some detail why Spencer’s choice of equilibrium temperature had nothing to do with reality. You just failed to read (comprehend?) what I said.
Bart, you are not free to set T(t0) to anything you like. That was one of the points I made – as did Barry B. Some values of the starting temperature – (1900 in Spencer’s choice) – are physically plausible. Some are not.
Spencer’s choice of starting at a very low temperature in 1900 requires, if his model was intended to be valid before then (with the exponential A0 term zero), that low starting temperature could only be appropriate if the PDO was exceedingly negative for most of the 19th century. If you look at Figure 3 in my “6 trillion degree” post, you’ll see that the 30-degree PDO convolution was close to zero or positive in 1900, not negative. So Spencer’s choice of starting temperature in 1900 is very different from the behavior of the PDO up to then.
And because he has a 30-year time constant, that low 1900 starting temperature essentially completely governs the behavior of his model until almost the middle of the 20th century. That’s the only reason Spencer’s model even appears to resemble 20th century temperatures – because he started from an unphysically low 1900 temperature.
Can you think of any valid justification for Spencer’s choice of temperature in 1900?
You guys have got me all worked up now. I really should follow Spencer’s example, and just stop. And, I will. But, there is one more point I wanted to make. Arthur stated ‘I’m not sure why you call that a “mistake”’ in equation (15). Arthur, look at our equations. There is another key difference beyond your subsuming the initial temperature anomaly into an arbitrary parameter. And, that difference is the start time in the exponential term and the convolution integral. Your equation is wrong.
Your latest comment popped up when I came back to explain this: “Can you think of any valid justification for Spencer’s choice of temperature in 1900?” Yes, certainly. It fits the data. It does not mean it “governs the behavior of his model until almost the middle of the 20th century” at all. It means that, with the given model, that is how the system would have evolved from that point onward. It is, in essence, a curve fit to the data based on a particular 1st order linear lag model involving the parameters T(t0), Te, tau, et al. Crying foul on it makes as much sense as saying the slope on a linear fit to temperatures in the latter half of the 20th century has no meaning because it accounts for the bulk of the upward movement.
You are still treating having trouble with the concept of when the convolution starts. It starts at the point t0 (tee-zero), which you have set to zero in the exponential term (examine my equation again if this confuses you). When the upper limit becomes less than the lower limit, the integration turns negative. In order to propagate back from that time, you must have a really, really, really precise record of the integrand, because the backwards propagation is unstable. That does not generally work because any small initial error, and any error in the data, will cause rapidly diverging results. In light of these considerations, when you say “low starting temperature could only be appropriate if the PDO was exceedingly negative for most of the 19th century”, I seriously doubt you have demonstrated it (and I’m not going to do the legwork to explain whether you have or have not – to say that you have failed to inspire me with confidence in your methods would be kind, and I do have productive uses for my time).
Don’t even get me started on the infamous tree ring proxies.
“By 1900 the temperature had been pretty stable for at least 50 years, according to the HadCRUT data set Spencer was using, so it seems reasonable to assume that if any period in that temperature record was near equilibrium, it was then. “
You are begging the question. That is not reasonable. The reasonable parameter is the one which gives the best fit with the data for the selected model! If the temperature profile then matches the observations, you have some indication that your model is reasonable.
“You just failed to read (comprehend?) what I said.”
You know, I’ve been nice. Given the shoddy work and ignorance, not to mention bravado, you two have displayed, it’s been difficult to show as little snark as I have. But, if that’s the way you want to play it, well, I was leaving anyway, and you two can go stew in your own effluvia and snuffle your armpits to your heart’s desire for all I care.
Bart,
You say, “The reasonable parameter is the one which gives the best fit with the data for the selected model!” But this statement is absurd in this case. Spencer was using the model to see if, by adjusting several parameters, he could get the PDO index to produce something like the 20th century temperature trend. He had to use 5 adjustable parameters, but he did it! Does that mean all 5 of those parameters were physically reasonable? Not even close. He had a 700 m mixed layer in the ocean, for instance, when it is well known to be more like 100 m. Of course, he could have fit the data just as well (exactly as well, in fact,) by using a 100 m mixed layer, but then his alpha and beta terms wouldn’t have supported his hypothesis about low climate sensitivity. If there are an infinite number of possible “best fit” solutions, then how does it make any sense to say that the reasonable parameter is the one that gives the best fit to the model?
This is all just basic regression analysis, which anyone who has taken a single college statistics course should have picked up.
As for your contention that Arthur’s equation is wrong, I don’t see how that can be. His integrated form of Spencer’s model gives exactly the same results as I got by doing a numerical integration of the differential equation, which were the same results Spencer got by doing a numerical integration. Therefore, Arthur got it right.
P.S. Yes, Arthur did demonstrate that the PDO had to be quite negative over most of the 19th century for Spencer’s model to produce a suitable starting anomaly in 1900. Keep reading his piece down into the comments, if you want to see where.
“Therefore, Arthur got it right.”
I can scarcely imagine a more damning quote. It is laughably wrong, which only a cursory examination by someone who knew what they were doing would show.
Bart,
Care to explain how Arthur’s integrated form could give the same answer as Spencer’s numerical integration of the raw differential equation, if Arthur got his math wrong?
I’m also on the edge of my seat waiting for you to explain how “The reasonable parameter is the one which gives the best fit with the data for the selected model” in a system that has an infinite number of equivalent solutions.
Bart – Let’s make something very clear here: what we are discussing is temperature *anomaly* – difference from some reference temperature. You can pick anything you like as the reference temperature, and the curves *look* the same, just shifted up and down a bit depending on what the difference is between your choice of reference temperature and mine.
However, you cannot pick anything you like as Te in Spencer’s model, because that has a very special meaning – it is the “equilibrium” temperature the model exponentially returns to after it has been perturbed by some forcing (PDO in Spencer’s case). This is due to his definition of “feedback” (eq. 2 in Barry Bickmore’s original post) as proportional to delta-T. And when you pick a starting temperature T(t0) for a particular point in time (say, the year 1900) then you are asserting something about the difference between that starting temperature and the equilibrium temperature.
That is, if you choose a temperature T(t0) = Te, you are making a (reasonable) assumption that before the start of your model simulation, the temperature was in its equilibrium state. If you choose a temperature T(t0) = Te + 0.1 C, say, then you’re saying it started slightly out of equilibrium on the positive side. If you choose T(t0) = Te – 0.1 C, you’re saying it started slightly out of equilibrium on the negative side.
In Spencer’s analysis, he chose to use a starting temperature in 1900 T(t0) = Te – 0.6 C – so his starting climate state was very significantly out of equilibrium. There is no evidence that climate in 1900 was way out of equilibrium – as I presented, what evidence we have shows it should have been close to equilibrium or above, not significantly below.
So – the point is, whether he did this by a “curve fit” or just picking convenient parameters for some purpose, the resulting fit is highly unphysical given his stated claim to be showing that temperature can be modeled as resulting from changes in the PDO index. For the first half of the 20th century, his modeled temperatures are mainly a result of his choice for T(t0) – Te. And as Barry has shown, it’s not just his T(t0)-Te that was bad – he had an infinite number of choices for his other parameters, and happened to end up with ones that required a hugely unphysical portion of the ocean to be participating in Earth’s short-term heat capacity. But conveniently his choice showed low sensitivity. Other parameters which would have given an equally good fit (as Barry and I showed) allow for a much more reasonable heat capacity and much higher climate sensitivity.
Barry – I told you what was wrong with the equation. I wrote out the correct equation. Have you never taken a course in calculus? Can you read? Let me explain one more time – I will type very slowly… What does Art’s equation give for T at time t = zero? Does it give T(0)? No? Then it is WRONG. This has to do with the existence and uniqueness of the solutions of differential equations, and the necessity of matching the initial conditions, but I’m sure that is way over the angels dancing on your head.
Art – A) you do not know the equilibrium temperature B) your rigid model parameter definitions are not compulsory. How the real world system works is what is being investigated. You and Barry up there do an awful lot of question begging. Read this. Maybe it will help. Maybe you can read it to Barry.
The important point here is that the temperature very closely follows the output of a system with commonly encountered magnitude and phase characteristics driven by the PDO. You guys are so busy jerking off, you missed the forest for the trees. Or, maybe you are too busy lounging in those little cesspools you call your blogs flinging feces and sucking up the praises of mindless sycophants that you just let it go to your heads. So, let me make sure you get the truth from an objective source widescreen: You are not smart, you are dumb. You are not pro-Science, you are anti-Science. You are throwbacks to the medieval dark ages when the Church employed brutes like you to browbeat and bully anyone who stepped out of line.
And, you should be ashamed of yourselves.
Wow Bart. Anthony, Roy, is this the type of dialogue you encourage?
Bart, if you have anything real to contribute, state it clearly without the insults.
T(0) is not a free parameter in my equation, so by definition my equation (18) gives T(0) = T(0). More specifically:
T(0) = Te + A0 + β Q(0)/c h
For physical reasons as you described earlier, the exponential term A0 really should be zero. Or you can think of it as a correction to whatever the estimate is of the starting value for the convolution Q at time 0. If we agree A0 = 0 (to avoid 6 trillion degrees etc), then the starting value for Q must be:
Q(0) = c h (T(0) – Te)/β
Since Q here comes from Spencer’s model as a 30-year convolution of PDO, the choice of T(0) – Te implies something about the PDO for time t before 0 (before 1900 in the case of Spencer’s fit). Spencer chose T(0) – Te to be highly negative – which would require the PDO to have been at its extreme negative limit for many decades before 1900. There is no evidence this was the case.
Or if you want to think about it another way, what Spencer’s model does is model temperatures through a dependence on PDO values several decades into the past. But he starts modeling at a particular year (1900) without any source of data on the PDO many decades into the past. Therefore his model cannot be considered a valid representation of the effect of PDO on temperatures until you have run it many decades after the start point, to get rid of transients. And it doesn’t actually match the rise in temperatures in the second half of the 20th century very well at all.
“Wow Bart. Anthony, Roy, is this the type of dialogue you encourage?”
It is the kind of “dialogue” found on your blogs. I am only returning the favor in kind. Maybe YOU should consider toning down YOUR rhetoric, if you want to be taken seriously.
The equations don’t work that way, Art. Solutions to Lipshitz continuous differential equations are UNIQUE. By your definition, A0 = T(0)-Te-β Q(0)/c h. So, then you’ve got β Q(0)/c h evolving exponentially thereafter. It doesn’t work. The solution I gave is the UNIQUE solution.
If you do not understand this, how can I take anything you have to say seriously? This is precisely why I keep trying to leave this discussion. It is very painful to acknowledge the existence of such smugly clueless people.
Actually, it can work that way, with a proper rearrangement of terms and splitting up of the integral. But, then you have destroyed your claim that A0 should be zero. And, you have no means of evaluating it, because you have to integrate Q from minus-infinity to zero, and you do not have that information.
“If we agree A0 = 0 (to avoid 6 trillion degrees etc), then the starting value for Q must be…
Dumb, dumb, dumb. I explained this to you at April 5, 2011 at 1:56 pm.
“Spencer chose T(0) – Te to be highly negative – which would require the PDO to have been at its extreme negative limit for many decades before 1900. There is no evidence this was the case.”
No! By your accounting, he chose T(0)-Te-β Q(0)/c h to be highly negative. You are mixing and matching your terms. As you shift your definitions to keep up with my criticism, you have to reevaluate everything you have previously done.
On this level, you are wrong. And, on the forest level, it is immaterial.
You DO NOT KNOW the history of Q from minus infinity to zero. And, YOU ARE NOT TAKING INTO ACCOUNT that this system does not exist in isolation, and other inputs can drive it to the starting condition.
You guys are like Rainman – totally focused on a model which only approximates reality. You are suffering from the delusion which has gripped all of climate science, that models ARE reality. They ARE NOT.
The model MAY however be a reasonable approximation to reality over all time. But, even so, the values of Te and tau and all the other parameters are partial derivatives, which hold in a particular local operating regime, but may have to be shifted over time to continue matching reality. You DO NOT have enough information to say. All we DO know, is that it gives a reasonable approximation in modern times using Spencer’s calibration.
This has got to end. You guys have no idea of the complexities involved, and I am getting wrapped around the axle trying to explain reality to an unsympathetic duo who will never see beyond their own noses.
“I am getting wrapped around the axle trying to explain reality to an unsympathetic duo who will never see beyond their own noses.”
And, that is precisely why Dr. Spencer was right not to wade into your fever swamp with a reply.
Bart – your comment April 5 at 1:56 pm claims I used different starting points for the convolution integral and the exponential term – however, you give no definition of what a “starting point” would be for the exponential term. Exponentials don’t “start” anywhere! The equation works going forward no matter what starting point you choose to use for the convolution, because you can fold any neglected parts of the convolution (from before the starting point) into the exponential term.
I.e. if you insist that the starting point for the convolution integral be t0 (1900, say), then Q(t0) has to be zero, because it’s a convolution integral from t0 to time t, and for a zero time interval the integral would be zero. All that means is that A0, rather than being zero, likely will have some non-zero value.
But if this model is a true representation of the physical system, rather than a mathematical toy, it must not have any special dependence on particular times. So if we knew the true value of PDO back into the past before time t0, then we could calculate the value of Q(t0) for a convolution with starting point centuries in the past, back to t = -infinity as I did in my Eq. 14. But in that case, the value of A0 would have to be zero, and the choice of starting temperature in time t0 would be determined by this Q(t0) integral.
So you have a choice to start the integral at t0, implying Q(t0) = 0 and A0 is free to fit things, or start the integral at -infinity, implying A0 = 0 and this full Q(t0) is some nonzero value determined by the actual behavior of the PDO. Since both choices give the same result going forward, that gives a relationship between the “full Q(t0)” and that choice of starting temperature, as I indicated.
“But, even so, the values of Te and tau and all the other parameters are partial derivatives, which hold in a particular local operating regime, but may have to be shifted over time to continue matching reality.”
Please note, in addition, that as partial derivatives, these are slopes. They are sensitivities. They do not necessarily correspond directly with physically meaningful quantities. If you do not know how differential equations are commonly linearized about a set operating condition, do some reading.
Bart – yes, I certainly agree “this system does not exist in isolation, and other inputs can drive it to the starting condition. ” But if other inputs are important in setting the starting condition, they will likewise be expected to be important in the behavior of the system going forward. At the very least, these other inputs, if they are to explain the very negative T(t0) – Te that Spencer chose for his model, then it is these *other inputs*, and not PDO, that explains roughly the first half-century of behavior in Spencer’s model. So his explanation that he has modeled 20th century temperatures with PDO is wrong – he has modeled it with PDO plus “other inputs”, unspecified, that made T(t0) – Te so negative to start with.
Either way, Spencer’s model claimed to be explaining 20th centuries with PDO, but it does not.