Home/BlogThe e-mail debate I have been having with Andy Dessler over his recent paper purporting to show positive cloud feedback in 10 years of satellite data appears to have reached an impasse.
Dick Lindzen has chimed in on my side in recent days, but Andy continues to claim that – at least during the 2000-2010 period in question — I have provided no evidence that clouds cause climate variations.
This is remarkably similar to how Kevin Trenberth rebutted my last congressional testimony…”clouds don’t cause climate change”, is approximately what I recall Kevin saying.
So, let’s return to Andy Dessler’s main piece of evidence, which is Fig. 2 from his paper, showing how monthly, global-average changes in (1) clouds and (2) surface temperature relate to each other, in the satellite observations (top panel), and in the ECHAM climate model (bottom panel, click for large version):
Andy has fitted regression lines to the data, and both have a slope approaching zero (for some reason, I can’t even find correlation coefficients in his paper). He claims these regression slopes support positive cloud feedback, in both the satellite observations and the climate model.
Now, why do I (and Dick Lindzen) disagree with this interpretation of the data? Because, while feedback is — by definition — temperature change (the horizontal axis) causing a cloud-induced radiative change (the vertical axis), NO ACCOUNTING HAS BEEN MADE FOR CAUSATION IN THE OPPOSITE DIRECTION.
And as shown most recently by Spencer & Braswell (2010, SB2010), any non-feedback source of cloud variations will (necessarily) cause a temperature response that is highly DE-correlated…just as we see in the satellite data! In fact, we showed that a near-zero regression slope can occur with even strongly NEGATIVE cloud feedback.
The bottom line is that, you can not use simple regression to infer cloud feedbacks from data like those seen in Dessler’s data plots.
This is not a new claim…there have been earlier papers cautioning against inferring cloud feedback (a specific kind of causation) from such data. The first two papers that come to mind are Aries & Rossow (2003 QJRMS), and Stephens (2005 J Climate). Nevertheless, researchers continue to use such statistics to try to justify the claimed reality of continuing climate model projections of strong global warming.
I’m sorry, but finding some statistical relationship with a near-zero correlation in BOTH the satellite data AND in the climate model behavior is (in my opinion) nowhere near proving that climate models are useful for long-term predictions of the climate system.
If that makes me a “denier”, so be it.
I don’t see why it’s so difficult to understand that the cause and effect relationship is not well understood or clearly defined. The modus operandi seems to be to assume a relationship and press forward with it, never looking back to validate that assumption.
I don’t understand what is so difficult about your point Doc. It seems clear that the plots above are not proof or even evidence of positive cloud feedback. Although feedback affects the graph, feedback cannot be determined from this graph.
I guess Andy’s logic is that the regression slope would have to be remarkably different from the model for it’s feedback to be inconsistent with the observations. But whether the observations at such careless, first-order analysis, can unequivocally rule out the model feedbacks is not a very good question-doubtless we don’t have enough data to do something like that even if we had reason to expect that an immediate inconsistency should be the case if the model’s feedback was wrong. A better question is whether one can find a method which finds the correct feedbacks for both-I say the latter part because I have a feeling that the ECHAM diagnosed feedback is NOT consistent with it’s own sensitivity, which I would bet is overestimated by this method. Then it becomes interesting to ask whether the data are different; that is, find if they match after finding the signal in the noise.
Please excuse me if I get this wrong, but when 2 variables are correlated it does not guarantee there is a cause and effect relationship. I’m not convinced either. Comparing complex computer generated model output with satellite data does not, it seems to me, prove a firm relationship. I create complex computer models, and internal errors can be just as important source of model behavior as the rest of the program.
Here’s a fascinating lay discussion from the New Yorker of the bias we (humans) have in trying to uncover the truth through experimentation and testing. The area is behavioral and medical sciences, but it applies to any field where experiments have to be validated.
http://www.newyorker.com/reporting/2010/12/13/101213fa_fact_lehrer?currentPage=all
Dear Roy,
Don’t we get to see the rest of the emails? 🙂
Best regards,
Alex
It looks like you could fit an ‘efficient frontier’ (a la a Markowitz curve) to a scattergram of points to determine, e.g., the theoretical ‘optimum’ positive or negative change in surface temperature on the X-axis in response to cloud variability on the Y-axis.
I think there is still room here to make it clearer that the IPCC concept of “feedback” used by Dessler is about the shift of a dynamically fixed point, caused by a single step change of an otherwise time-invariant parameter, while the ordinary notions of feedback, such as those cited to Bode 1945 and elsewhere, and used by Dr Spencer here, are about dynamical structure and refer essentially to time variation of dynamically determined state variables while the parameters remain time-invariant. The innuendo cleverly fostered by the IPCC, that these two concepts are somehow equivalent, has to be recognized as one of the finest masterworks of the art of propaganda spin of the twentieth century; but surely now we are in the twenty-first century, it can be dismissed for what it is.
What about the fact that Dessler reports the regression coefficient in the first graph of the measured data was reported to be only 2%. In my professional experience with statistical modeling, I would stop there and look for another variable pair. This seems very poor before even talking about cause and effect, etc.
The problem is that there are two regimes present; clouds (ascending moist air, low pressure) and no clouds (descending dry air, high pressure). Each has its own physics, and trying to overlay the two regimes makes an awful mess. The problem hasn’t even been defined.
Any Roy, you are as guilty anyone with your attempted analysis in terms of “stuff that just happens.”
I’m no statistition, but if I remove the lines from the first plot and perform the squint test, I see a slope of zero. The fact that the limited number of points seems to have caused a negligible positive slope means nothing to me.
What you must realize, Roy, is that you are talking science and Dessler is talking politics. He will never agree he is wrong. The warmaholics play the “refute game”. When a paper like yours appears, they desperately need a paper to say you are wrong. Whether you are, or not, is irrelevant.
When Svensmark published, Lockwood and Frohlich produced a rebuttal. So when anyopne tries to quote you or Svensmark, the warmaholics can say “Yes, but we have Dessler and Lockwood”. It has nothing to do with science, and everything to do with politics.
So, as long as you talk science, you will not win. And if you degrade yourself to play the warmaholic game, you will lose as well.
I would not worry. Dessler has already lost.
The fact that the 100 year plot is based on the suspect GISS or HADCRU temperatures should be sufficient to be suspect of any correlation in my opinion.
Dr Spencer writes:
“… feedback is — by definition — temperature change (the horizontal axis) causing a cloud-induced radiative change (the vertical axis), …
And as shown most recently by Spencer & Braswell (2010, SB2010), any non-feedback source of cloud variations will (necessarily) cause a temperature response that is highly DE-correlated …”
By this reference to “any non-feedback source of cloud variations”, Dr Spencer seems to mean, amongst other things, that el Niño and la Niña events include effects on clouds that are not completely explicable as consequences of changes over relatively short times of global spatial-average land-sea surface temperature (if I am reading aright what is meant here by ‘temperature change’).
For example, he seems to mean that perhaps el Niño and la Niña events include effects on clouds that require for their explanation a spatial component re-distribution of local land-sea temperatures with a component that does not alter the short-time global spatial average land-sea temperature. For example, perhaps one part of the ocean gets hotter and another part gets cooler, without altering the global spatial average, and this component re-distribution can fully explain some partial effect on clouds.
So far as I understand, el Niño and la Niña events include local changes in ocean temperature that precede the global spatial average temperature changes that are associated with those events. Such local changes would qualify as “non-feedback” cloud variations, as I read the usage of words above.
On this reading, Dr Spencer seems to be proposing the possibility that el Niño and la Niña events include cloud phenomena that are not explicable in terms of the IPCC “forcings and feedbacks” formalism (as set out for example in Bony et al. 2006), which insists that all its “feedbacks” are fully explicable by changes in global spatial average land-sea temperature. In contrast to the IPCC usage, ordinary scientific language would regard such phenomena as properties of a dynamical system that, loosely speaking, has feedback, though not in the very narrow usage demanded by the IPCC “forcings and feedbacks” formalism.
To me, it seems likely that cloud changes associated with el Niño and la Niña events are not their primary external triggering causes. For example, so far as I know, no one is proposing that changes in cosmic rays cause cloud changes that trigger el Niño and la Niña events. On the other hand, it seems likely that el Niño and la Niña events will include cloud effects, which ordinary scientific language would regard as feedbacks in response to whatever is the (currently unknown) primary external triggering cause of el Niño and la Niña events. I am saying that ordinary scientific language would under some circumstances regard cloud changes that are part of el Niño and la Niña events as feedbacks, even though they are “non-feedback” in the usage prescribed by the IPCC formalism, that Dr Spencer’s above definition seems to accept.
On this reading, Dr Spencer’s citation of Aires and Rossow 2003 and Stephens 2005 suggests that perhaps he may be open to usage of the word ‘feedback’ outside that prescribed by the IPCC “forcings and feedbacks” formalism. Personally I find the IPCC prescription to be ludicrous, and I would be happy to see Dr Spencer go outside it. I think it would be helpful if he explicitly said so, and if he gave his own definition of ‘feedback’, showing how his extended usage departed from the IPCC prescription, with a thorough explanation of his reasons. This would mean his stating a definition alternative to the one he gives when he writes above “… feedback is — by definition — temperature change (the horizontal axis) causing a cloud-induced radiative change (the vertical axis), …”. Such a new definition would not necessarily exclude the effects referred to by IPCC prescription, but would not be limited to them, and would be more in accord with the usual scientific understanding of the word.
I happened to run across a reference to “Simpson’s paradox” elsewhere. I am not a statistician, but the gist suggests a possibility: there are one or more other variables that confound Dessler’s correlation.
Simpson’s paradox
http://en.wikipedia.org/wiki/Simpson's_paradox
“In probability and statistics, Simpson’s paradox (or the Yule-Simpson effect) is an apparent paradox in which a correlation (trend) present in different groups is reversed when the groups are combined. This result is often encountered in social-science and medical-science statistics,[1] and it occurs when frequency data are hastily given causal interpretations.[2] Simpson’s Paradox disappears when causal relations are brought into consideration (see Implications to Decision Making).
“Though it is mostly unknown to laymen, Simpson’s Paradox is well-known to statisticians, and it is described in a few introductory statistics books.[3][4] Many statisticians believe that the mainstream public should be informed of the counter-intuitive results in statistics such as Simpson’s paradox,[5] in particular to caution people against the inference of causal relationships based on the mere association between two or more variables.[6]”
? For what it is worth, if anything.
Jim Cripwell is right, it’s politics. Dessler’s F-grade science is being used as a cover story, and you cannot win with merely competent science, even if the evidence is all on your side. You need the ruling political party behind you for that. But fighting the fight is necessary to learn, and to teach others, what is really driving the “consensus” climate science. Cloud feedbacks are clearly a secondary effect, and a side issue now; currently they have put their scientific marbles all on the “unassailable” radiative transfer theory. I suggest everyone assail it, since it is obviously disconnected from the thermodynamics that really determines the atmospheric temperatures. Identifying changes in the atmospheric CO2 is not the same as establishing a CO2 radiative forcing effect on temperatures. If you decide you really want to be a “denier”, you will have to become honestly (scientifically) comfortable with the idea (I say fact) that there is no greenhouse effect whatsoever. There is IR absorption and emission, but no warming (or cooling) effect. That really should be obvious by now, but few want to consider how a whole field could be so corrupted, so wrong.
The problem is that the data are very noisy, making any conclusion about causality problematical. Dick Lindzen deals with this problem by considering only the largest temperature/radiation fluctuations and showing that the result does not depend sensitively on the choice of threshold. Roy uses all the data and uses the multi-month segment method to extract an estimate of the underlying trend. Neither is altogether convincing, but the fact that they give similar results is very suggestive of a negative feedback at these time scales. Dessler’s correlation claims are simply mindless.
Roy,
Has Dessler, at any time, said anything about your phase space plot results and the conclusions you draw from them? I’m curious if there are any other interpretation that can be drawn from that specific result?
I am a materials scientist, so this is very far from my experience, but I wonder if it is normal in climate science to try to infer anything from such a graph.
In a review I would reject as foolishness any attempt to fit a cloud of points like the satellite-based one. Is this the quality of data over which people are building climate models, or is just Dessler without any better argument?
Kjai January 25, 2011 at 5:25 PM
Welcome to climate science.
But even if the correlation were better, the assignment of cause and effect would be ambiguous.
As has been pointed out in an earlier reply by someone else,Correlation is a necessary but not sufficient condition to establish cause and effect. It appears to me that any correlation in the case of the two scatter plots is weak at best. Both plots look more like shotgun scatter patterns than representations of any significant degree of correlation between the variables. Clearly there is much variation that is not accounted for by the regression.
Really interesting article post. Your post is very interesting. I liked this post. This post is good! I have to say that after continually reading awful as in, same old material, little quality content, etc. blog posts on other sites, it’s nice to actually read something that has some thought put into it. It’s a pleasure to read good content, especially after continually seeing rehashed crap that writers and bloggers are throwing out nowadays. It’s always nice when I come across content that actually has value, I’ve been seeing a ton of subpar writing attempts lately. Anyways, thanks again and I’ll check back regularly to see what else you have to offer. I’ll check back in the future to see what else you have up your sleeve. Keep up the excellent work! BTW, I like your site design, but your header image was only loading half way for me. P.S. Your header is messed up a bit in IE. The problem could be on my end but I thought you might want to look into that.
Dear Dr. Spencer, A. Dessler, C. Game and …
Please kindly consider what follows.
Firstly, let me quote something from my own post sent to “Does CO2” drive …?” debate last October:
“Clouds´ impact on Earth´s climate is among the uncertainties so far. Their effect varies a lot with their type, location, etc.?Although they reflect back into space a lot of sunlight that hit them, we have to keep in mind that this is important only when sunlight beams hit the Earth on a not too big circular area around the point where they hit perpendicularly, let us say of a diameter equivalent to 6 or 8 hours, and for the clouds there in each moment… The effect of the rest of the clouds at the day side, yet alone the ones at the night side, is to warm the climate system”.
Let us suppose, for the sake of the ease of the exposition, that for the real kind and time distribution of clouds, those opposite effects were compensated over the whole daily cicle.
At noon, due to what exposed above, clouds´warming effect would be clearly negative: the more clouds, the lesser warming. If we introduce (or increase) GHGs, the subsequent decrease of cooling rate – directly due to the GHGs – will mean that part of the energy will have to make its way to outer space some hours later… when cloud´s effect would be at least less negative.
And this delay of cooling also hapens during the afternoon, evening and night, with positive clouds´ effect most of that time.
So, as far as I can see, GHGs globally decrease cooling effect of clouds and/or increase their warming effect.
It could be argued that in the morning the contrary happens, but then there is much lesser heat to be removed, and that happens during only a few hours.
Am I right?
Yours,
RMN
I was only partly right…
After sending previous post I realised that at late afternoon, evening and night, although clouds have a warming effect and GHGs delay cooling too, for what exposed it´s the same some hours sooner or later!
What exposed would be right only during app. the mentioned six to eight hour time zone circle.
I see now less effect than what I initially thought: the amount of energy to be removed since noon until three or four hours later is much higher than during the same time before noon… GHGs added to clouds would have the exposed efect, but more slightly.
Better now?… Thank you.
RMN
What would the IPCC projections be like if you were to substituted the CERES trends and your final interpretation into their models? Could you show us graphically what the comparison would be like to the year 2100?
I recognize that there are many factors in the IPCC model – though other work I have read has suggested that most are irrelevant, and the whold “black box” boils down to a very simple, couple-of-variable equations. Still, what would be the effect of this view of sensitivity? Significant, as per your implication, or not significant, as per what the warmists might say?
Dear Dr. Spencer, A. Dessler, C. Game and …
Not seeing any reply to my last Sunday posts, perhaps they haven´t been clearly understood. I´m going to put it in another way.
For any given scenario of clouds in a zone at noon, any increase in clouds has a clearly cooling effect – especially if not too far from tropical belt – due to the fact of beeing much more important the subsequent increase in energy reflected back to upper atmosphere, than the increase oin IR radiation energy captured at their lower side and sent back down.
But some time later, due to the higher inclination of the Sun´s beams, less energy reaches the considered Earth´s zone, even without clouds. Any increase in clouds can´t significantly diminish the amount of Sun radiation reaching the zone.
But that increase in clouds increases the energy captured at their lower side and sent back down, without this beeing affected by the higher inclination of Sun beams.
If due to GHG´s, part of the energy that could have been sent back to upper atmosphere at near noon moments has to make his way later, the above mentioned increase in clouds will have a warming effect.
As expressed in previous posts, the opposite would happen in the morning, but the amount of energy “suffering” the delay in its way back to space would be much smaller.
That´s why I consider that the sheer presence of GHGs slides clouds´effect towards a warming or less cooling one (could I say a positive feedback?).
RMN
I found this interesting and I did not regret reading your article.
I found this interesting and I did not regret reading your article. https://northwesttermitepestcontrol.com.au/termite-control-ermington-north-west