After the record-breaking global temperatures noted in 2010, Chip Knappenberger treated us to his Cherry-Picker’s Guide to Temperature Trends. It was a follow-up to his earlier Cherry-Picker’s Guide to Temperature Trends.
As a cherry-picker’s guide, it’s outstanding; those who want to characterize temperature trends based on an agenda rather than hard science will find what they’re looking for. But as an appraisal of actual temperature trends, it leaves a lot to be desired. The main conclusion is unchanged from his original post:
But I conclude that my original article’s summary remains applicable:
What I can say for certain, is that the recent behavior of global temperatures demonstrates that global warming is occurring at a much slower rate than that projected by the ensemble of climate models, and that global warming is most definitely not accelerating.
I too haven’t seen any evidence that over the last 30 years global warming has accelerated (except in the UAH lower-troposphere data, but that evidence is scant). But what about that claim of “a much slower rate than that projected by the ensemble of climate models” — a claim which is characterized as “for certain”?
The claim seems to be based on this graph:
which is an updated version of this graph from the original cherry-picker’s guide:
The dots mark linear-regression trend estimates, with filled circles indicating statistically significant trends and open circles non-significant trends. The updated graph plots trends from January of the start year to December 2010, the original version shows trends from September of the start year to August of 2009.
Note that the original graph shows a number of statistically significant negative trends, although all of them cover less than 10 years. Note also that in the update, those statistically singificant short-term trends are absent (except for the 9-year HadCRU trend). Perhaps Chip should have thought about why that is, and why it makes for a superior cherry-picker’s guide, but a poor characterization of actual temperature trend.
You may already have guessed that Knappenberger used monthly data — which shows very strong autocorrelation — but didn’t compensate for autocorrelation. So, his indications of “statistically significant” are, well, not. He does mention in the update that he omits this factor.
But let’s get to the heart of the matter. Are the observed trends really “a much slower rate than that projected by the ensemble of climate models — for certain”? Instead of just plotting trend estimates of observed data, let’s put some 95% confidence intervals on them. I’ll omit the compensation for autocorrelation, but use annual rather than monthly data so the estimates will be pretty good. Here’s the result for data from NASA GISS (plotted in red), compared to the actual mean trend from the given start years of the AR4 model runs (plotted in black) (click the graph for a larger, clearer view):
Gosh. Suddenly that “much slower rate” no longer seems “for certain.”
But wait! There’s more! There’s uncertainty not just in the observed trend, but in the model estimates as well! Let’s add the 90% coverage region (from 5% to 95%) for the spread of individual model runs:
Gosh. Suddenly that “much slower rate” is definitely NOT “for certain.” In fact, it’s not even plausible.
If your last figure, you are making a mistake by doubling up on the errors. The statistical uncertainty about the trend estimates is already included in the distribution of model trends, thus, showing the statistical error bars about the observed trend is inappropriate when showing them set against the model trend distribution.
How long can the observed trend lie below the multi-model mean projection before it is ok to say that the observed trend is less than model projections? 10 years, 20 years, 30 years, 100 years? Never? Given the intermodal differences, perhaps the latter answer is your preference? That seems unreasonable in a practical sense.
[Response: Either the spread of model results or the uncertainty in observed trends shows that your claim of “for certain” is wrong.
How long will you refuse to admit mistakes which are glaring, obvious to everybody, and indicate attachment to a political agenda rather than respect for scientific correctness?
Here’s some very sound advice for you — simply admit that you were wrong and you might get some respect, as long as you continue to cling to such obvious folly you only make yourself even more of a laughingstock.]
“How long can the observed trend lie below the multi-model mean projection before it is ok to say that the observed trend is less than model projections? 10 years, 20 years, 30 years, 100 years? Never?”
If you stop trying to focus on predicted short trem trends vs observed short term trends (both of which have notoriously large uncertainty ranges) you’ll see that the answer is obvious: compare the observed temperature record with the predicted temperature range. If the model predictions are way off, it shouldn’t take all that long (almost certainly a lot less than 30 years) to demonstrate this.
“If your last figure, you are making a mistake by doubling up on the errors. The statistical uncertainty about the trend estimates is already included in the distribution of model trends, thus, showing the statistical error bars about the observed trend is inappropriate when showing them set against the model trend distribution.”
This is a odd statment. For the real world we have one realization and GISSTEMP an estimate thereof. For the model world we have many realizations and the best characterization of the uncertainty is to use the model ensemble. This is of course not possible with the GISS data. Does your confusion stem from the fact that where an ensemble exists that is used for the uncertainty ? Or are you complaining that a different method is used for the models? In which case have you tried to use the same method on both models and GISS (and correcting for autocorrelation in both cases).
Try starting the analysis at 1979.
Chip criticizes tamino for showing both. I’ll criticize Chip for showing neither.
Fair enough. The full context of my quote from my updated cherry-picker guide is :
As you can see, I was not professing complete certaintly at that time.
In the editing process, the phrase “more or less” apparently got dropped from qualifying “applicable” before leading into the quote from my original article. It should have read “But I conclude that my original article’s summary remains more or less applicable.”
Thanks for pointing that out. I have gane back and made note of this and re-inserted it where it should have been all along.
While, as recently demonstrated by Santer et al., the average observed rate of warming (at least in the lower troposphere) has consistently has fallen beneath the multi-model mean projected warming for at least the last 32 years, Santer et al. show that is has not done so with standard (95% level) statistical certainty. I have argued on your previous thread, that if natural variability were accounted for, that the perhaps that may not be the case—but again I do not know so “for certain” at this point in time. I also presented evidence a few years ago that the 95% level of statistical certainty was achieved over some trend length periods for some datasets (see here for discussion of that work). We are in the process of updating that effort, so I can’t say “for certain” whether our results through 2011 will indicate the same thing as our results through 2009 did.
However, while I can’t, at this moment, say for certain with 95% confidence statistically that the observed rate of warming lies beneath the multi-model mean warming, there are indications that it has been (at least in some data sets during some time periods) in recent months.
So while virtually all evidence shows that observed trends (on average and currently) are numerically less in magnitude than the numerical mean value from the collection of climate models over many periods during the past 1-3 decades, not all the differences are statistically significant at the 95% level. That’s why I characterize climate models as “being on the verge of failing” rather than that they “have failed.”
Actually what Santer et al. showed is that the ratio of TLT (based on UAH and RSS data) to surface warming is less than projected by the models.
Rather than simply assuming this means models are over-predicting TLT warming, an arguably more likely explanation (which fake skeptics refuse to consider for obvious reasons) is that UAH and RSS estimates of the TLT trend are biased low.
The smaller the trend difference, the longer the time. So, patience :-)
Well, the influence from natural variability will presumably eventually shrink down towards nothing, but the influence from intermodel differences could reach a minimum that is not close to zero. After 32-yrs, by my reasoning presented in the previous thread, intermodel differences are responsible for about +/- 0.1°C (2 * std dev) of variance about the multi-model mean trend—which seems to be close to what it was after about 20 years—which means that it may not shrink much more over the time over which the trend is computed. If that is the case, then an observed trend of ~0.15°C/decade would remain “consistent” with the multi-model mean of ~0.25°C/decade forever, even though it was only 60% of the projected rise. That is a long time to be patient before admitting to what would seem to be a fundamental error in the models (in this hypothetical situation)
Chip said… “If that is the case, then an observed trend of ~0.15°C/decade would remain “consistent” with the multi-model mean of ~0.25°C/decade forever, even though it was only 60% of the projected rise. That is a long time to be patient before admitting to what would seem to be a fundamental error in the models (in this hypothetical situation)”
There’s a lot that’s hypothetical there, Chip. How about telling us up front which “observed trend” you’re talking about. Or, perhaps you’ll have to fess up that you’re adding more cherry picking to your previous cherry picking.
Chip, you’re beginning to get there ;-)
Of course the models contain systematic trend errors. Already the fact that model trends differ from each other (even after removing natural variability by ensemble averaging) should tell you so. At most one of the models could be exactly correct in expected trend behaviour, but I’m pretty sure none of them are. So, testing the model distribution against “data” (itself the result of a difficult modelling exercise) is trying, in a more difficult way, to find out something you already know. What it does give you is a handle on the common-mode error, what most or all of the models have in common. And note that one source for this is likely to be the forcing data, not the model logic.
The proper thing to do then is to start considering where these differences might come from. Santer et al. do so in their article, presenting the outline of a sensible hypothesis.
To conclude that the models are (or are starting to be) “falsified” (whatever that means… do you really mean “mainstream atmospheric physics is all wrong”? Surely not. What is your hypothesis?), is just silly, a dog-whistle for your denialist audience. That the models are “wrong”, George Box style, is obvious, and signifies little besides the need for further study.
By the way, is there a PDF of Santer et al available somewhere? Paywalls still thwart some of us. . .
Click to access Santer2011.pdf
So what Chip is trying to avoid saying is that the models have not failed. So he agrees then that they have not failed– he is coming around to reality.
I would expect that using monthly data and not accounting for the autocorrelation would way overstate the significance of Chip’s “trends”
Within the ensemble of models is most likely one that will exactly match Chip’s “trends”. So using his logic his headline finding is “SOME MODELS ARE PERFECT!!”
Just to confirm, given that you are using the GISS data, am I correct in assuming that you are using the near “surface” (i.e., 2 m AGL) trends for the models and not the synthetic MSU TLT trends estimated from the model profiles?
What is the predicted mean rate of increase in “surface” temperatures from the models? I know it changes slightly, but it should be near 0.2 C right?
[Response: It depends on the time frame, but yes, very close to 0.2 C/decade.]
If so, then why the is the paid disinformer now talking about 0.15 C/decade (satellite estimates for TLT) versus 0.25 C/decade (synthetic MSU TLT estimates from the model profiles)?
Chips says in his reply to you that:
“As you can see, I was not professing complete certaintly at that time”
But he was! He clearly says in his post:
“What I can say for certain, is that the recent behavior of global temperatures demonstrates that global warming is occurring at a much slower rate than that projected by the ensemble of climate models…”
There is a disconnect here, a breakdown in Chip’s logic. If what he writes here is true, then he should have prefaced “certain” with a word that made it clear he was not “completely certain”,in fact he should not have used the word “certain” period. The word games these fellas play are ludicrous. These word games are nothing more than an attempt by the paid disinformer to cast doubt and feed fodder to the “skeptics”.
As for your advice:
“Here’s some very sound advice for you — simply admit that you were wrong and you might get some respect, as long as you continue to cling to such obvious folly you only make yourself even more of a laughingstock.”
Agreed. Note that he cannot bring himself to admit that saying “what I can say for certain” was wrong, so he then promptly shift the goal posts to the lower tropospheric data, even though the topic of discussion is the surface data.
Chip claiming that the models are on the verge of failing is like him telling people that his wife is almost pregnant. No, she is either pregnant or she is not pregnant. Or as for “certainty”, another analogy is him bragging telling people that he is certain that his wife is pregnant, and when it turns out she isn’t, instead of admitting error, he tries and make excuses by saying that he was not professing “complete certainty” at the time. People would laugh at such nonsense, as they should.
I invite you to see how the climate models are faring compared with surface observations and MSU LT observations (through 2009) over at James Annan’s site at via this link:
I have referred to the MSU in recent comments made at Tamino’s because that is the comparison that we have the most up-to-date data from, thanks to Santer et al., 2011.
Thanks for proving my and Ray’s point– you just can’t admit that you were wrong to say “What I can say for certain…”.
To divert attention from such inconvenient critique, you suddenly seem to have forgotten the topic of this post and start floating red herrings. Well I’m not chasing your blimp (I did Dr. Annan’s post– and you seem to have already forgotten his “hopping list of possible reasons for the results”).
Try staying on topic and speaking to the subject of this post.
You can keep posting and tying yourself in Gordian knots, but with each post you only further erode what precious little credibility you think you might have :)
And then check how it looked four months later.
Gish gallop anyone?
Maple Leaf, I would note that Chip’s tactics have been adumbrated by the creationists–who have been saying the theory of evolution is “on the verge of failing” for about 160 years now.
Chip is much more into the fawning adulation of the denialati than he is into anything as unremunerative as simple respect. He’ll never back down. He’ll always double down.
Every time a new fossil is discovered one ‘missing link’ is removed to be replaced by two ‘missing links’; we’re getting further and further behind!
Yes, and every year that has a max in global temp or min in arctic sea ice provides more opportunity for the pseudoskeptics to claim “recovery” until the records are again broken …and the “recovery cycle” begins anew.
Ray said, “He’ll never back down. He’ll always double down.”
His paycheck depends on it.
Well, Santer et al. 2011 showed that the multi-model mean warming trend in the LT was ~0.25°C/decade, and Foster&Rahmstorf showed that the RSS and UAH LT observed trends from 1979-2010 were ~0.15°C/decade.
My hypothetical situation was that if these trends were representative of even a 100 yrs or 1,000 yrs (rather than just 32 yrs), that they still could be consistent with one another (because of differences between models) even though one was only 60% of the other–a discrepancy that some folks may consider to be large enough to have policy implications (then again, some may not).
But you’re still avoiding the fact that GISS and NCDC both show trends closer to the models. Why would you automatically assume it’s the model trends that are wrong instead of the satellite data? It’s equally as likely that it’s UAH/RSS that are on the verge of failure. In fact, there’s more of a precedence for errors in the satellite data (remember, orbital decay?).
Some folks not far from here would jump a hole in the air if not the discrepancy, but the 60% of IPCC projections itself would lead to real, honest-to-goodness ‘policy implications’ :-)
This may be of interest to you,
Annan et al. (2011) published an interesting paper in which they propose a new approach to validate the CMIP3 ensemble of GCMs. They find that,
“However, once observational uncertainty is correctly accounted for, the values observed appear entirely unremarkable, and the hypothesis is not close to being rejected.”
“Figure 2 (right) also shows the outputs of the CMIP3 ensemble of GCMs, calculated from the 20th century simulations. The models represent a broader spectrum of behaviours than can be represented by the energy balance model, possibly due to forcing differences, as well as more sophisticated dynamical behaviour. However, the observations of climate change fall well within the ensemble spread.”
Here is a version that uses monthly data with autocorrelation correction. You can get the equivalent of Chip’s graphs by reading off the right vertical axis. You can select any of various indices.
There’s also a Foster/Rahmstorf version.
— by Horatio Algeranon (with a little help from Chip)
As certain can be,
About the model
The ground truth!
See Figure 2 at http://www.wunderground.com/blog/JeffMasters/article.html
Places that should be covered with snow resulting in a high albedo, are bare, resulting in a lower albedo, and more warming.
Did somebody cherry pick the date? It does not matter. Any date in winter should have more snow on the ground than shown. I think we can say that it is likely that the (2007) climate models have understated the expected climate change by the start of 2012.
So given that the Suomi NPP is “only” 512 miles above the Earth (diameter of 8,000 miles, like in take a cone from the satellite’s position, find the tangency line, etceteras), that we can’t even see Antarctica, or even much of the Arctic for that matter, since you know, like it’s totally dark up there.
Given that the CG of that photograph is somewhere’s near Mexico’s east coast, mind you.
That NA has seen very low winter precipitation THIS YEAR (to date) versus LAST YEAR, but we already knew that, now didn’t we?
Sum all that up and what do you get?
One winter’s worth of regional weather data != long term global climate change.
So if we had an image from LAST YEAR, showing almost the OPPOSITE of the first image taken from a brand new satellite THIS YEAR, your logic would suggest, to me at least. that you would have reached the exact OPPOSITE conclusion.
I mean, I’m all on the bandwagon and what not, but seriously your instrument needs some major tuning, as in, get you act together before you make such silly slapshot statements.
Your larger point is correct: however unusual, it’s just one day.
However, you may have missed the fact that the image is actually a mosaic, made up of multiple shots stitched together? It’s described in the commentary.
Completely understood implicitly.
But it is a mosaic of the entire Earth as seen from the perspective of a satellite 512 miles above the Earth.
Field of view, at that height, to image the entire Earth is 125 degrees. Area of Mexico? 761,606 sq mi. Area of USA (CONUS)? 3,119,884 sq mi. Ratio of CONUS/Mexico? Approximately 4.1 to 1. Rough eyeball estimate of CONUS/Mexico as seen in the aforementioned composite perspective image? Approximately 2 to 1.
Which, at that height, can only see about 11% of the hemisphere (or <6% of the globe).
Spreadsheet geometry? Check.
Google Earth "Eye at xxxx.xx mi" fly through? Check.
Work experience? Four years of high school drafting, two year degree in civil engineering technology (two surveying courses, private practice land surveying, TA in surveying at university level (while being at university as an undergraduate myself), coastal surveying, DGPS ship navigation surveys, laboratory surveys), Landsat, MODIS, SPOT, Geoeye, and aerial photography (most recently during the 2011 Mississippi River Flood, specifically related to the USACE's Birds Point Levee breach).
Touche. . .
Santer et al: “Because of the pronounced effect
of interannual noise on decadal trends, a multi-model ensemble of
anthropogenically-forced simulations displays many 10-year periods with
And what of the data? The same and then some, as shown by Dana’s “Up The Down Staircase” riff on BEST data. That’s an interesting agreement.
This may be a bad time to criticise climate models but …
Am I right in thinking that all the AR4 models in this ensemble have missing feedbacks? These include CO2 and CH4 emissions from melting permafrost and the emissions of methane from Arctic seas. For projecting future climate this means they are underpowered.
The underlying force of this discussion seems to me to be about how much effort should be put into climate mitigation. I guess that on the current performance of AR4 models, Chip Knappenberger feels climate mitigation is not worth that much effort.
Does Chip Knappenberger recognise these models are underpowered for predicting the future because of the missing feedbacks?
I have been told that these feedbacks are missing from the models being used in the preparation of AR5. Is this true? If so, is it serious?
Can I just know the final score?