by Judith Curry
The CMIP5 decadal simulations are now available for seven climate models. The first intercomparison results have just been published.
Evaluation of short-term climate change prediction in multi-model CMIP5 decadal hindcasts
Hyemi Kim, Peter Webster, Judith Curry
Abstract. This study assesses the CMIP5 decadal hindcast/ forecast simulations of seven state-of-the-art ocean- atmosphere coupled models. Each decadal prediction consists of simulations over a 10 year period each of which are initial- ized every five years from climate states of 1960/1961 to 2005/2006. Most of the models overestimate trends, whereby the models predict less warming or even cooling in the earlier decades compared to observations and too much warming in recent decades. All models show high prediction skill for surface temperature over the Indian, North Atlantic and west- ern Pacific Oceans where the externally forced component and low-frequency climate variability is dominant. However, low prediction skill is found over the equatorial and North Pacific Ocean. The Atlantic Multidecadal Oscillation (AMO) index is predicted in most of the models with significant skill, while the Pacific Decadal Oscillation (PDO) index shows relatively low predictive skill. The multi-model ensem- ble has in general better-forecast quality than the single-model systems for global mean surface temperature, AMO and PDO.
Citation: Kim, H.-M., P. J. Webster, and J. A. Curry (2012), Evaluation of short-term climate change prediction in multi-model CMIP5 decadal hindcasts, Geophys. Res. Lett., 39, L10701, doi:10.1029/2012GL051644.
Links: full paper [Kim et al. 2012_GRL] and auxiliary material [Kim et al. 2012_GRL_Auxiliary]
The CMIP5 simulations were described in a previous post.
Background information from the Introduction:
The prediction of decadal climate variability against a background of global warming is one of the most important and challenging tasks in climate science. Not only does natural variability have a large-amplitude influence over broad regions of the globe, it is an integral component of climate variability that modulates low-frequency climate phenomena as well as extreme climate events such as trop- ical cyclone activity. On decadal timescales, some aspects of internal climate variability may be predictable . However, the actual prediction skill of natural climate variability on decadal timescales using various current climate models has received little attention.
The Coupled Model Intercomparison Project Phase 5 (CMIP5) has devised an innovative experimental design to assess the predictability and prediction skill on decadal time scales of state-of-the-art climate models, in support of the Intergovernmental Panel on Climate Change (IPCC) 5th Assessment Report [Taylor et al., 2012]. The decadal pre- dictability and prediction skill of individual models have been analyzed separately for multi-year prediction horizons over different time periods and regions [Pohlmann et al., 2009; Fyfe et al., 2011; Chikamoto et al., 2012; Mochizuki et al., 2012]. However, the CMIP5 decadal predictions from different models have not been evaluated and com- pared using the same evaluation matrix. The choice of one model over the other, or the use of sets of models in a multi- model ensemble (MME), requires information that compares the predictions of individual models. Here, we compare the ability of currently available CMIP5 decadal hindcasts to simulate the mean climate and decadal climate variability from individual coupled models and a multi-model ensem- ble. We focus on the surface temperature and two dominant internal climate modes: the Atlantic Multidecadal Oscilla- tion (AMO) and Pacific Decadal Oscillation (PDO). This study addresses how well the CMIP5 multi-model decadal hindcasts simulate the spatio-temporal climate variability.
JC comment: this post is cut short by my hopping on a plane to head back to the US. This is a technical thread, comments will be moderated for relevance. I will be back online tomorrow and will provide further comments and participate in the discussion.
Does the exaggerated variability occur only over the short term or are long term trends also exaggerated?
The problem, jim2, is that models of reality replaced observations of reality in government-funded science after 1945, when Sir Fred Hoyle led science astray with a model of Earth’s heat source that was adopted without debate.
All experimental observations that conflicted with that sacred model were brushed aside for sixty-four years, until November 2009, when Climategate emails and documents revealed deceit that most world leaders and leaders of the Western scientific community ignored.
It’s called science-fiction lolwot. Move on…
Science based on observations, like this:
Nature <b.277, 615 – 620 (22 February 1979)
Not puesdo-science propaganda, like IPCC reports:
Please describe what this means (from your personal bio):
“I accepted powerless over all things controlled by cause-and-effect in 1996 and grasped the great benevolent reality outside my own ego cage.”
Humans are powerless over all things controlled by “cause and effect”?
Doesn’t seem too scientific to me. And what about this “ego cage”?
It is interesting to see the models have predictive skill in some regions but not others. I wonder if the regions of predicitve skill will change once natural variability does. For instance, will they still have good predicitve skills regarding the AMO once the AMO changes direction.
dice and a dart board also have predictive skill. Much more successful than any climate science is this expert:
Interesting. Bit puzzled by the CFSv2 results in Figure 1 – the uptick at the end of each forecast can’t be real, can it?!
Another question might be whether we can continue to assume that a rational analysis is even possible in an area where ignorance affects expectations and where the range and interrelatedness of events are misunderstood if not entirely unknown.
How to tell if prediction in this is better than random chance? If global climate is as teleconnected as many claim, how can failures in huge areas of the globe translate into viable results for other parts?
Do you mean, “is this skill or luck?”
By the way, welcome back to the US. Traveling at the end of a semester must be tough.
I’m puzzled by fig 1 too. Models (b) and (c) seem to always start off cooling then level out. (g) is weirder with uptick at start and end, as Ed says.
And why don’t they all start on the black observations lines?
Presumably the models are told about the major volcanoes, hence cooling in early 60s, (Agung) 80s (Chichon) 90s (Pinatubo)?
Maybe I need to rtfp.
Fig 2 is also puzzling – it suggests the observations show a warming of over 0.4/decade?
If climate is chaotic and the models’ emergent properties mirror this property, then one wouldn’t necessarily expect the emergent global temp average to mimic what we have seen in the past, other than to the extent external events like volcanoes are input into the model.
Paul – the different behaviours in (b) and (c), say compared to (a), are due to the way the models are initialized. In HadCM3, they use observed anomalies added to the model’s own climatology, whereas the models in (b) and (c) use the raw observations. For (b) and (c) the models drift towards their preferred state after the forecast is started, and this mean drift is removed afterwards. There is not yet any agreement about which methodology is best!
These simulations do know about the ‘future’ volcanoes – some previous simulations have not done this.
Agreed – 0.4K/decade seems far too large from a glance at Fig. 1 – looks more like 0.4K/50 years.
Thanks Ed, I can see arguments for both methods. That also answers my next Q, which was why does (c) CNRM seem to show a cooling in all runs yet shown as warming in Fig 2.
Yes – that is the explanation for CNRM’s trend in Fig. 2 as well. The multi-model mean drift corrected predictions are shown in the Supp Info, but it is a shame that the individual models are not shown there.
I find in the conclusions the following “All models show high prediction skill for
surface temperature up to 6–9 years over the Indian Ocean,
the North Atlantic and the western Pacific Oceans, while
showing lower predictive skill over the equatorial Pacific
and North Pacific Ocean.”
I am not at all clear what this means. To me “prediction” means forecasting what is going to happen in the future. That is, we must predict what is going to happen, when we dont actually know the answer. In order to do this, one must predicit what is going to happen in at least 2012, but more importantly, what is going to happen from 2013 onwards. How can anyone have possibily assessed the prediction skill of these models? Or does “prediction” mean something other than saying what is going to happen in the future, when we dont actullly know what the answer is?
Jim Cripwell | May 17, 2012 at 9:25 am |
Not speaking to this particular case, but just in general:
The data don’t know what year it is now. Testers can pull a fast one on data by pretending it’s 10, 20 or 30 years ago and predicting from then matching results to what actually was recorded. Keeps the data honest.
Those sneaky testers.
Unless there’s a bias in their methods. Or, say, the dataset itself is inadequate in any number of ways. Or they’ve made invalid assumptions.
That sneaky data.
Bart, you write “The data don’t know what year it is now.”
Sorry, Bart, but this is not true in my book. What the paper must be describing is what I term “calibration”. All models have lots of fudge factors. One calibrates such models by changing these fudge factors, until you get the best fit for all the known data. Once the model is calibrated, then you use it to predict what you dont know. Then you wait until the answers are known, and compare prediction with actuality. When you have done this a sufficient number of times that the match between prediction and actuality could not have occured by chance, then you know you have a model that can make predicitons.
If what you say is the proper interpretation of what this paper is all about, then it is, IMHO, completely and utterly useless. All it is describing is the calibration process; there are no predictions.
Let me add, Bart, this is the difference between this paper and the one I quoted recently, namely
N.A. Kilifarska, Journal of Atmospheric and Solar-Terrestrial Physics, http://dx.doi.org/10.1016/j.jastp.2012.03.002.
The paper I have just quoted, about which you were rather disparaging, does the correct thing. It matches the known data, and then does what any good model does, it forecasts the future. Yes, we must wait to see what happens. But this is the correct use of the word “prediction”.
Jim Cripwell | May 17, 2012 at 10:59 am |
What, I’m _not_ disparaging enough when I say “pull a fast one”, or “sneaky” or bias, inadequate or invalid?
I find all the sun-worshippers’ articles interesting to read, and look forward to many more of them. And I must say, JASTP (or as I knew it when I still read it right after the Journal of Irreproducible Results, JATP or UFOlogy Today) has been rife with Scafetta and like entertainments and diversions practically from its inception at the dawn of the science fiction era.
Why without it, Ray Bradbury, Isaac Asimov, Arthur C. Clarke and Gene Roddenberry’s writings would have been much poorer, and we never would have seen the Twilight Zone and X-Factor on television.
Open Arctic during warm times cause massive snows that do increase the Albedo of the Earth. Apparently this has been measured. They have not realized that open Arctic and increased Snow is responsible, but they do now have more evidence that Albedo is increasing.
Dr Judith Curry does know this. You likely recall the paper she published recently that linked open Arctic to More Snow and Cold Weather.
This does mean that Snow Extent is increasing, every year since 1997.
The ice retreat is over and Earth Warming is over for this warm cycle. We may bounce in and out of this, some, but the warm oceans will keep the Arctic open enough to keep the snow coming. We will cool until the oceans get cold enough to freeze the Arctic with multi-year ice and turn off the Snow Monster.
Jim – agreed – in this sense they are using ‘prediction’ to mean retrospective predictions, i.e. mimicing forecasting in the past by only using data up until the forecast start time.
There is of course no pretense about modelers capturing the ‘game changers’ as can easily be seen in other historical contexts like, the frozen Bering Strait, a longer spear, cavalry and the double-back bowy, Jesus, the Black Plague and Spanish flu, the Irish…
Great paper Dr Curry. Thanks so much for taking this on. This is how the unsettled science improves. Lots to absorb in your paper but throwing your efforts into this technical issue in a professional manner is much appreciated by us interested observers.
Congratulations on an excellent paper! I have one question. What fraction of the predictability is due to persistance from the initialization (i.e. nowcasting) as contrasted with changes in the diagnosed metric?
Roger – the paper itself compares the skill in predicting AMO and PDO with ‘persistence’. The models tend to do better than persistence for the AMO, but not PDO – did you have something else in mind?
@Abstract: The multi-model ensem- ble has in general better-forecast quality than the single-model systems for global mean surface temperature, AMO and PDO.
OK, single-model systems are terrible, ensembles are just poor. Therefore, ensembles are “better”.
That “in general” qualifier is a nice slight of hand too. In other words, sometimes the ensemble is worse.
Things were going swimmingly until I got to this paragraph:
The results from the ERA40 and the ERA Interim are NOT SURFACE TEMPERATURE DATA. They are not data of any kind. The ERA40 and ERA interim results are computer model results. They are not data in any sense of the word, and I am astonished that you would refer to them as “data”.
You are simply comparing computer model results to other computer model results, and declaring victory where they chance to match up …
Why in the world would you want to do that, when we have at least passable observations covering the period in question? You claim that you are comparing the model results to “natural variability”, when in fact you are merely and trivially comparing them to the variability in the outputs of other computer models.
The underlying problem is that nature is discrete. Nature loves edges, and sudden changes. Generally, temperatures don’t go smoothly from temperature one to temperature two in Nature. There are jumps, discontinuities, and sudden changes. There is no gradual change from being in a cloud to being in clear sky. There is cloud, and more cloud, and then suddenly there is clear sky.
The ERA40 and other computer results, on the other hand, give us nice, smooth, gradual curves from temperature one to temperature two. They don’t do edges, they don’t do sudden jumps.
This inability to distinguish between actual observations on the one hand, and ERA40 and other computer model results on the other hand, is a very worrisome trend in computer science. There is no “surface temperature data from the ERA40” as you claim. There is no data of any kind from ERA40.
The only thing the ERA40 computer model puts out are computer model results. You seem impressed that these ERA40 computer model results can be approximated by other computer model results … me, not so much …
PS—I would have thrown the paper out in any case because you have not even mentioned autocorrelation … you can’t just compare two datasets without considering autocorrelation, Judith. Next time, hire a statistician. What is it with climate scientists, are you guys allergic to statisticians or something?
Psst.. Anthony used re analysis data in his paper. I missed you pitching a fit on that one
Thanks for the input, Mosh, but a more neutral way to bring that up would have been to say “Did you know that Anthony used reanalysis data in his paper”?
Then I could have said, “No, Steven, I didn’t know that, haven’t a clue if he did or didn’t, but if he did it would seem like a mistake to me”. Then I could ask you “Which paper and where?”, and the beat could go on.
Instead, you go all snarky, asking why I’m not “pitching a fit”, as you quaintly describe my scientific objection to calling computer model results “data”. You want to know why I haven’t commented regarding something I’ve never even noticed … which doesn’t improve your reputation for mildly unpleasant drive-by posting.
Nor does it address the question of Judith’s use of ERA40 computer model results, much less speak to the question of her calling computer results “data”. Instead, all you have done is call them “data” yourself. They are nothing of the sort.
Yes, Computer Model Output is not Data! That abuse does occur, time and time again. Thank you Willis for Pointing that out!
You must have a much more limited definition of “data” than I have. Seeing as tens of thousands of programmers would disagree with yours, I think you ought to reconsider it.
Data doesn’t have to be observational. It doesn’t even have to have any connection to reality.
Thanks, Brandon. If data “doesn’t even have to have any connection to reality” on your planet, you definitely are using a very different definition of “data” than most people use.
On the other hand, if your data has absolutely no connection to reality, you would be perfectly equipped to take up a new profession as an AGW alarmist …
Would programmers disagree as you state? Sure, because they think that their computer models are reality. It’s a well known problem in the field, read the writings, they talk about “the world” or “the global temperature” when they are talking about the computer-generated world and the computer-generated temperature.
However, it is vital to compare our theories, not to computer model results, but to actual observational data. Comparing the output of one computer to the output of another computer, as the paper has done, tells us nothing about the real world.
And that is my point, not the terminology regarding “data”, but the comparison of models to models rather than comparing them to the world.
**model data** is still data
It would be data only if one were investigating the behavior of code, then it would be data about the code. But the numbers output by climate models is in no way data concerning the climate. I’m not sure anyone knows what exactly it is.
I had initially agreed with you that ‘data’ equates with facts. But that apparently is just one definition
That it has a number of other somewhat post modern, computer related meanings and can include statistics, goes a long way to explaining why Mosh and I sometimes disagree when I complain that some ‘data’ (like SSt’s and tree rings) don’t seem to be rooted in what passes for reality in the real world that the rest of us inhabit (obviously excluding politcians and economists as well).
climatereason, there’s an article on Wikipedia discussing the use of the word data in the way I describe, and the main article on data acknowledges data doesn’t have to be tied to observations. Despite that, Willis Eschenbach says:
His position is tens of thousands of people use the word one way because they’re delusional, not because it’s a common usage. Entries in dictionaries and articles on Wikipedia support my claim of common usage, but he does have snideness and blind dismissal on his side.
I’m reminded of a time on this site where a bunch of people insisted one cannot do experiments with computer models.
I don’t doubt what you say. I just think that many of us naively automatically equated ‘data’ with ‘factual’ and that is obviously not the case. I will try to use the words ‘factual data’ in future as other uses of the word data obviously has quite a different meaning.
I don’t consider a Wiki to be in any position to define what data in a scientific context should mean. Data used in science has always meant numbers resulting from a measurement of some sort, no matter how tortured. There is no reason to change that for the convenience of modellers.
I understand, and I don’t have any issue with that. I’d never be bothered by someone not knowing a particular definition. It only starts to become a problem when they’re told that definition, and they dismiss it without consideration. And then it becomes a major problem when they call people delusional for using that definition.
Disagreements over what words should mean is one thing, but disagreeing over how words are actually used is a totally different thing. It’s silly to be upset with people writing programs (such as computer models) for using a word in the way programmers use the word.
Mind you, I’d approve calling for them to be clearer about their data. Instead of saying something like “temperature data,” they could easily say something like “modeled temperature data.” That would allow them to use their standard terminology while being clear on just what is used.
Brandon Shollenberger | May 18, 2012 at 3:22 pm |
Programmers call numbers generated from programs “output.” Not data. There is input, processing, and output.
This discussion is ridiculous. Data is a broad term. There is “model output data”, “observational data”, etc., but to claim that output from a computer is not data is just plain stupid. But “data” does need to be put into the proper context. “Reanalysis output data” is not completely disconnected from the “observational data”. It is a lot like the old hand analyses that meteorologists used to do, except now it is done by a computer. Oh, horrors!
Instead of engaging in semantic quibbling, I would be interested to see someone address the actual argument Willis is making:
“However, it is vital to compare our theories, not to computer model results, but to actual observational data. Comparing the output of one computer to the output of another computer, as the paper has done, tells us nothing about the real world.
And that is my point, not the terminology regarding “data”, but the comparison of models to models rather than comparing them to the world.
Well, what is missing is the knowledge of how the ECMWF matches the purely observational datasets. I assume EMCWF (ERA) was used for a 4-D interpolation (basically filling in missing data in elevation and time fields). It’s missing from this discussion but perhaps Judith can address this when she gets back, as I’m sure it exists (whether satisfactory or not is another issue).
RealClimate posted a nice concise description of reanalyses last year. The post was marking the launch of a renalysis wiki site, which will probably have more information. Essentially they seem to involve assimilating observations to constrain an atmospheric (or atmosphere-ocean) model.
Don Monfort, it’s cheeky of you to accuse me of semantic quibbling. Willis Eschenbach made a clear assertion in regard to the point I’ve been discussing:
This is a derogatory comment which impugns the credibility of thousands of people, and it is entirely based upon him misunderstanding and/or making things up about definitions. It is not semantic quibbling to point out somebody else is using bogus definitions. It’s basic communication.
And really, if people can’t even agree to what the words being used mean, why should they try discussing something more complicated? If we can’t agree on simple points, is there any reason to think we can agree on complicated points?
Brandon, my point is quite simple, and you continue to ignore it and focus on the semantics.
Judith is not comparing computer models to observations. This fact is being hidden, whether deliberately or not, by the misleading use of “data” as in “reanalysis data”.
Look, I don’t care what you call it, the point is she is comparing one computer model to another computer model as if that meant something. In no case is she comparing model results to observations.
Perhaps when you get done with being concerned about the words used, you could comment on the actual issue that I am pointing to.
willis, the reanalyses provide a sensible gridded analysis of surface temperature. the grid spacing, nominally 100 km, is not directly comparable with with point observations over land. The reanalyses (other than the coupled reanalysis) are directly based upon the satellite observations of sea surface temperature. You cannot usefully compare point measurements with the coarse resolution model results.
I see that Judith has deleted my comment in reply to you calling me “cheeky”, Brandon. I guess it is OK to call someone “cheeky”, but “baby”. is verbotten.
You are engaged in useless semantic quibbling and nitpicking. Your silly claim that you are defending thousands from an assault on their credibility by Willis is ludicrous and amusing. And we all do not have to agree with a particular definition of the word “data” to carry on a discussion. Judith has replied to Willis without quibbling over or even mentioning the word “data”. That’s how the grownups do it.
I’m sure you’ll consider this “semantic quibbling,” but there’s an enormous difference between calling someone’s behavior cheeky and saying they’re a baby. Namely, talking about behavior is not talking about a person. There’s no comparison between our two comments.
You’re welcome to ask what “has happened to” me, but the answer is, nothing. I haven’t changed. My behavior is the same as it was six or twelve months ago. You’ve just begun to take issue with it for whatever reason.
Seeing as you’ve resorted to insults (getting yourself moderated in the process) I think questioning your behavior would be more productive. That idea is reinforced by your misrepresentations, as you did not just call me a “baby.” You called me a “big pretentious baby.”
This site has an explicit rule against personal insults. The fact you are bothered/confused/whatever by such a simple rule being enforced says a lot about you.
You fabricated an issue and made derogatory comments based on that fabrication. You, for all intents and purposes, called me delusional because I acknowledge a definition one can find in dictionaries. And you complain that I don’t just ignore this.
Do you have some other point? Yes. Am I obligated to respond to every single point you make in any comment I respond to? No. It is perfectly acceptable for someone to respond to a single point made in a comment. That’s what I have done. It’s unsurprising. It’s perfectly normal. It only bothers you because you won’t acknowledge you were wrong.
And that’s what this comes down to. You were wrong. You made offensive remarks which were wrong. I called you out. You now try to divert the discussion rather than just admit you were wrong. Ironically, if you just admitted you were wrong and apologized for your offensive remarks, the conversation would go back to the point you want me to discuss!
It would appear it always those who abuse semantics who complain about people discussing semantics.
Thank you, Brandon. You understand the rules better than I do. Perhaps I should have said “your behavior is like that of a big pretentious baby”.
Look Brandon, Willis explained what he was getting at and Judith answered him, without engaging in useless semantic quibbling. Didn’t you see that? All your yammering in the interim, about dictionary and wiki definitions, and tens of thousands of wronged programmers was just comic relief. Carry on, without me. i had enough of this foolishness with josh.
Model OUTPUT (data) is ONLY as good as model INPUT (data).
JC, please address the synthetic data issue. In what ways is it likely to be superior and/or inferior to observed data.
Yes, the problem of the true and only meaning of “data” is fascinating, but it may be of more interest the reason behind not using “factual measurements”.
I don’t normally like reposting verbatim a comment I made on another site, but as it seems directly relevant to your comment here goes;
‘I wonder if the constant fiddling with figures has its genesis in studies of Historic temperatures carried out by such as Phil Jones. In a book of his I recently read, in which he re-analyses a number of the very earliest instrumental temperature records (18th Century) he made a comment (either in the book or a related article) something along the lines of;
‘The instrumental records we examined seemed to be showing warmer temperatures than our computer models indicate should have ocurred. We have therefore adjusted the instrumental record.’
I was dumbfounded when I read this. If we have ‘factual data’ we should generally use it or explain why the substitution is better, so I would be interested in any reply to your comment above on synthetic data.
Do you have a source? Context?
Here is the book in which the comment was made (or in one of the articles links/comments that accompanied its publication)
The context is that, as you know, this was an EU funded project that looked at 7 long temperature records and tried to ascertain their accuracy-difficult for a vast variety of factors including physical movement, different observers, accuracy of instrument etc.
I read it and then tracked through the analysis in order to write an article about the Mannheim Pallatine.
As I said in the other thread I don’t believe in a conspiracy theory, but using models instead of observed data needs careful explaining and all in all the case for the observed data being biased warmer was not that convincing. As you know historical temperatures are fraught with diffculties but are generally better than other proxies such as tree rings which are a very blunt instrument. (pun intended)
I see PJ listed as editor but not as author.
But context is needed, because as it stands it doesn’t make sense. The only point of adjusting an instrumental record is if you want to do some processing – eg look for cycles. You’d need to know why whoever said it needed the adjusted temperatures.
Camuffo was the main author who does some very nice, detailed but sometimes rather tedious work which makes for difficult reading. I don’t think the point was to look for cycles but to try to determine the likely temperature at a given time in a given place.
However I don’t intend to spend another three weeks ploughing through it again as it was peripheral to my research subject, so perhaps you could comment further on the use of synthetic data?
Here is the actual book I read;
I think the book has been repackaged several times as in one version both Camuffo and Jones are listed as editors but also wrote articles. I have read several books from Jones and Camuffo on very similar subjects (early temperature records) but think the comment I posted was in reference to this book rather than the other similar works on early records.
I think Jones in particular writes in a very interesting manner and there are echoes of Hubert Lamb in his material. He also comes over as much more equivocal in his comments about warming in general than is generally ascribed to him. Bearing in mind the apparent diffculties he has with spread sheets I assume he looked at graphs of adjusted temperatures presented to him for this book rather than actually created them.
Again, on the account you’ve given, it makes no sense. You’d only want a modified estimate of the temperature at some time/place for a purpose – to fit some theoretical postulate. And then the adjustment would be an arithmetic device. That’s why context is essential.
Where synthetic data is created somewhat as you say is in the NCEP reconstructions. There the idea is to take a dataset with lots of gaps and, using a model, produce a consistent series. In the process, some actual readings will be modified. But I don’t think anyobe regards the NCEP as a replacement for the original readings at some time/place. Instead it is a usable source for other types of inference and calculations.
Nick, I think you’re begging the question. As I see it, climatereason has shown direct evidence that he modified observational data to conform to “computer model data.” You’re saying “Wow, that would have been aweful–he must have done something else.” It seems to me that the burden is on you to show that there is some “context” that would justify his behavior.
‘… on the account you’ve given, it makes no sense. You’d only want a modified estimate of the temperature at some time/place for a purpose – to fit some theoretical postulate.’
The postualte is that there is a warming bias in early instrumental records and they weant to identify and correct it.
I’m not saying it would be awful, I’m saying it would make no sense. We’re talking here about a long established, well published European data series. It isn’t PJ’s to change. He may have calculated some modified numbers for some analytic purpose. Until you know what he really did, it’s pointless.
Your paraphrased quote somewhat explains Willis’ bellicose cynicism – it has useful predictive power!
OTOH JC is not Jones. Re-analysis done carefully ought to allow more meaningful regional comparisons between models and instruments. And I hope JC will fill us in.
You say it was not in Jones gift to change the historic records but that is the purpose of a number of EU funded projects-no doubt for what they believe are good scientific reasons. I’ll quote you a little of what I wrote
“Discovering the Societas Meteorological Palatine of Mannheim has enabled a few pieces of the climate jigsaw to fall into place for me. I noted in one of my articles;
‘Frederik became King of Prussia in 1701 and immediately set up a measuring station that became Berlin Tempelhof, one of our oldest records and this started a rash of similar stations that caused Samuel Horsley to comment in 1774: ‘The practice of keeping meteorological journals is of late years becoming very general’.
This proliferation of stations using good quality instruments and often under royal patronage was no doubt the reason for the establishment of the Mannheim palatine from 1780, as a great deal of information was becoming available, but no one was collecting it together. The 1780 start date for the worlds first ‘global’ temperature record precedes James Hansen’s establishment of Giss by no less than two centuries.
I am trying to find the original Mannheim temperature data for the 40 stations, as all those stations that I recognise as probably belonging to the Societas have had their data adjusted by modern researchers, including Phil Jones. He (and others) got several very large lumps of EU funding to look at a variety of historic sites. For example this study by him of a site in Austria from 1760 in which he reduced the temperatures by 0.4c.
He then got a separate 5 million Euro grant to investigate a further 7 Historic European early records and believes their temperatures were overstated by an average of up to 2 degrees.(mostly somewhat lower) This resulted in a book.
“Improved Understanding of Past Climatic Variability from Early Daily European Instrumental Sources
The link immediately above is the brief that the various researchers worked to under the auspices of the EU grant. The intention was to identify and correct any shortcomings in the old records. As I say there may be genuine reasons to change the old temperatures (although there is an awful lot of conjecture and supposition in the analysis) but the net result was that it was agreed there was a ‘warm bias’ and as Bohm noted ;
‘However, the constantly warmer instrumental summer temperatures before the late 19th century in the CET data suggest the presence of a warm bias in the early English observational data in
this season, similar to what has been concluded for Central European stations (Frank et al. 2007a, b) and also proposed for Stockholm and Uppsala summer temperatures (Moberg et al. 2003).’
The observational instrumental data did not agree with the models and the observational data was amended. I am not saying there is a conspiracy theory or scientific fraud, just that the older records (such as those from the Mannheim Palatine) are being amended with official sanction and public money.
Your second link gives the context I was seeking. It sets out the specs for the IMPROVE project. It says that the product will be a CD with both the original temperatures and the “corrected” temperatures.
It seems to have several parts. One is simply the collection of the original readings. I presume this involves some QC. Then there is the use of metadata. This is one kind of correction, but it isn’t model-based. Then there is homogenisation. That’s not modelling exactly, though they might use the help of models, since the data is so sparse.
And they say what they are really after:
“characterise climate variability and determine teleconnection behaviour across Europe: high-frequency temperature variability; seasonal temperature extremes; recurrence intervals; growing seasons.”
That’s the point. No-one really cares (scientifically) what the temperature was at Tempelhof on 25 Dec 1701, so there’s no intrinsic purpose in correcting it. The point is that that measurement is a clue to things that we really want to know. But it needs to be interpreted for those purposes. One way to do that is to explicitly calculate corrected series corresponding to those sites, and work from there. Another is just to do some reweighting as part of the averaging process, if you want, say, a central European index.
The corrected series has the merit that anyone can then use it for whatever similar purpose they have in mind. But they don’t have to – it’s an option. They are choosing to use the IMPROVE interpretation for their purpose.
Shouldn’t we distinguish the use of various statistical techniques used to backfill times/regions of missing data from modelling? They aren’t really the same thing. Or did I misunderstand Nick?
So these models were able to agree with other models, which has nothing or perhaps little to do with reality. Yippeee
DEEBEE | May 18, 2012 at 6:33 am |
Willis has grossly simplified a nontrivial question.
Computers produce nothing that in any qualitative way is differentiable from what he claims is ‘real’ data.
A thermometer is a device. A computer is a device. Many thermometers are analog models that correlate measured temperature by volume of a liquid in a sealed container to heat in contact with the structure the thermometer comprises part of. A computer generates a digital model that correlates calculated temperature by weighted inputs from.. thermometers, and other instruments.
Tempest meets teacup. You can’t actually get ‘real’ data. You can collect samples of measures of proxies in models and interpret them into datasets.
The better dataset is the one that can be validated and verified to other objective standards meaningfully and readily. What Willis is arguing for is pretty much the Luddite opposite, from a point of view that appears to fail to recognize the difference between a thermometer and temperature.
In figure S1 of the supplementary material I note that the standard deviation of the ensemble mean is higher than in other periods.
Any hypotheses as to why?
Since each of the decadal “predictions” were supposed to rely only upon data up to the beginning of that prediction, it would seem that the real predictions for our actual future should have the same approximate model to model variation as all of the previous predictions for which we actually have results. Correct?
In my above comment the period with higher than average std dev of ensemble mean is the latest period, which is the only decadal prediction for which we don’t know the correct answer.
For all of the earlier “predictions”, where we DO know the actual result the Standard deviation of the ensemble mean is smalle.
Hi Ed – Yes – For example, see the paper
How Much Skill Was There in Forecasting the Very Strong 1997–98 El Niño?
Christopher W. Landsea, John A. Knaff Bulletin of the American Meteorological Society Volume 81, Issue 9 (September 2000) pp. 2107-2119. http://journals.ametsoc.org/doi/pdf/10.1175/1520-0477%282000%29081%3C2107%3AHMSWTI%3E2.3.CO%3B2
As they wrote
“A …….simple statistical tool—the El Niño–Southern Oscillation Climatology and Persistence (ENSO–CLIPER) model—is utilized as a baseline for determination of skill in forecasting this event”
“….more complex models may not be doing much more
than carrying out a pattern recognition and extrapolation
of their own.”
By persistence, I mean a continuation of a cycle whose future is predicted based on past statistical behavior, given a set of initial conditions.
Dynamic models need to improve on that skill (i.e. accurately predict changes in this behavior) if those models are going to add any predictive (projection) value.
(i.e. accurately predict changes in this behavior)
Perhaps you should remove this thought from the merely parenthetical to something like–e.g., all caps or perhaps scratched into the underside of climatology’s coffin lid.
Let me add a general comment. It is exceedingly difficult conducting double-blind experiments, and ensuring complete neutrality, when you dont know the answer before you do the experiment. How on earth anyone could conduct the equivalent of a double-blind experimen t when you already actually know what the answer is, and still ensure complete neutrality with no bias whatsoever, I cannot imagine. But if Kim et al, including our hostess, has somehow performed this miracle, then HOW they did this ought to occupy at least 90% of the paper. Until I see how this was done, I dont believe a word of what is in the paper.
In any event, why doesn’t the paper include a prediction of what is going to happen in the next 5 years? That is what we want to know. Yet, it simply is not there. Incredible!!!
Jim Cripwell, what you consider incredible is exactly what I would expect. Why would testing models against each other and past data require making new predictions? And when the paper indicates biases across the models, indicating future predictions made at this point would be wrong, why would it make future predictions?
“why doesn’t’t the paper include a prediction of what is going to happen in the next 5 years?”
The papers DOES have the 2010 to 2015 prediction, or at least the ensemble mean forecast. As I noted in a comment above, the standard deviation on the ensemble mean is significantly higher than for the hindcasts. If the hindcasts were truly independent hindcasts without knowledge of the “correct” answer, I would expect the ensemble std dev ( I.e. the differences between model predictions) to be the same for the hindcasts and the true forecast to 2015. It is not.
The ensemble mean projection is for temps to rise 2010-2012 and then to flatten and slightly decline to 2015.
See figure S1 of supplementary material
charliexyz you write “The papers DOES have the 2010 to 2015 prediction”
Yes and no. Sure the graphs extend past 2011. But what I am looking for is a number I can compare with something that happens in 2015. In Smith et al Science August 2007 I find “with the year 2014 predicited to be 0.30 C +/- 0.21 C warmer than the observed value for 2004”. So, when it comes to 2014, I find an observed value for 2004, add 0.30 C, and that is the forecast for 2014. What is the equavalent for this paper? What number do I look for in 2015 to compare with what is forecast in this paper? That is what is not at all clear to me.
Judith – I am all for developing better models. However, a model should be ultimately compared to reality, not just to other models. Why should I trust some 100-year (10-year, 5-year) predictions from climate models? Where is a reliable 100-hour weather prediction? Maybe we should start there.
First, there are pretty good weather predictions for a hundred hours out.
Second, comparing models to each other doesn’t prevent them from being tested against “reality” as well. This is one test, and others can be done in addition to it.
Thanks. Great! Where can I find them?
Brandon Shollenberger | May 17, 2012 at 5:06 pm | Reply
From the climate models? I’ve never seen any, but they might exist. Citation??
Seeing as I didn’t say anything to indicate those weather predictions were coming from climate models, and climate models aren’t used for predicting weather, I don’t know why you’d ask about climate models.
Anyway, the entire subject is a stupid diversion as predicting short term changes is radically different than predicting long term changes. The only reason to compare the two is ignorance or a desire for cheap point-scoring through rhetorical tricks.
Thanks, Brandon. Since the subject under discussion is climate models I assumed that’s what you were talking about. If you are switching gears to talk about weather models, you should say so, as otherwise in a discussion about climate models your words will be misinterpreted.
You go on to say:
Before your hyperbolic language gets too extreme and your abuse goes entirely over the top, consider that a climate model time-step is typically on the order of half an hour. Then the climate model, just like a weather model, calculates what will happen in the following half hour, and then the half-hour after that.
In other words, a 100-year prediction is nothing but 1,753,190 half-hour predictions strung end to end. So a prediction of a long term change, far from being “radically different” from a prediction of a short term change, is comprised of nothing but short term predictions.
In other words, a climate model is simply a weather model run for more years and with more variables (e.g. slow changes in ice cover, land cover, and many other climate phenomena).
So I see no reason to ignore the question of how the climate models do at the short term as you would like us to do. The climate models are supposed to be good at the half hour scale. The claim is often made that despite being no good at annual or decadal scales, they miraculously become accurate again at multi-decadal scales.
However, I’ve never heard any explanation of how many years it takes for them to finally regain accuracy at the long end of the time scale … or how fast they lose accuracy at the short end … or why they should be accurate at both ends of the time scale but not in the middle. Do you have such an explanation?
I also know of no other field of science where there is a prediction method that claims accuracy at the half-hour and the century-long scales, but which is useless at annual and decadal scales. Perhaps you could enlighten us on that question as well.
“I also know of no other field of science where there is a prediction method that claims accuracy at the half-hour and the century-long scales, but which is useless at annual and decadal scales. Perhaps you could enlighten us on that question as well.”
It’s common. Systems that satisfy differential equations can be tracked for a while, and then often have useful asymptotic behaviour.
Take a pendulum clock. If you start it in a certain way, you can predict with regular nechanics for a while exactly where the pendulum will be at any instant. But after a minute or two that fails, if only because you just can’t know the initial condition accurately enough. And the clock is useless for timing a footrace.
But longterm the pendulum clock can still keep excellent time, even though you can’t now predict the location of the pendulum at an instant. You know the asymptotic oscillatory behaviour.
Nick, nice bait and switch here. You pendulum example starts out as predicting where the pendulum will be accurately, then fails after a minute or two. Yet over the long term, you claim, the clock tells excellent time.
So what happened to knowing where the pendulum will be over the long term? Because the early claims have nothing to do with telling time, only with where the pendulum will be located.
At what point – minutes, hours, days, weeks – does your model allow you to again accurately predict where the pendulum will be?
I think you made Willis’ point quite well. It is EXACTLY what the climate modelers say: we can predict something in 30 minute intervals, then it falls apart in the mid term, but over the long run a different variable is completely accurate!
Dear Brandon: First, please tell me kindly where to find reliable 100-hour weather predictions you have referred to.
Second, I am ignorant enough to think that most climate models – at least those modeling the Atlantic Multidecadal Oscillation – are in fact attempting to forecast weather on a very large time scale. A highly ambitious task; in addition to a meteorological situation, you also have to model seasons, glaciers thawing, sea currents changing, and so on. If we can not reliably model even the meteorological component, why should I take the big model seriously?
Here are BoM rain forecasts, detailed, quantitative, for 8 days. They use numerical weather forecasting models – closely related to climate models. I have found them to be very good.
Nick, thanks. Very nice indeed. Are you aware of anything similar for the USA? I know that Dr. Curry lamented here some time ago about a state of the numerical models in the USA.
FWIW I don’t understand your clock example.
My understanding is that the pendulum paces an escape mechanism and this in turn increments gear movements which effectively sum the no. of pendulum swings and output the integral via the mechanical indicators of hands. If you know the mechanism design, initial position of the pendulum and the hands then knowing the position of the hands at some future point in time will tell you the position of the pendulum. It is determinate mechanical system. I don’t think it’s behaviour is asymptotic. Calibration is the issue – IMO it will not tend towards your chosen reference measure of time but continue to diverge in line with its loss/gain.
I guess wrt climate models I would say the calibration issue would be akin to having the correct sensitivity coefficients for the correct driving variables.
Probably the clock example would be better if I supposed a model for the pendulum mption, which would be more or less a simple differential equation solved by timestep. It would predict the exact position of the pendulum for a short time, but would soon fail to predict the real position at a prescribed time. However, it would, as asymptotic behaviour, have the correct frequency oscillation.
But maybe a better example in routine use is computational fluid dynamics. This solves basically the same equations as GCMs. You often solve for flow patterns, which may be steady or transient like like vortex shedding. It’s simulating the flow from moment to moment. Even though your knowledge of the initial conditions (short term) may be poor, it successfully describes the long term pattern.
Curious George and Willis Eschenbach, even if you want to (ridiculously) pretend forecasting ~100 hours at is comparable to making predictions 10+ years out, despite the monstrous difference (averaging alone is enough to reject such), there is also the enormous difference of scale (i.e. climate models don’t attempt to describe individual weather patterns). With such a massive difference in resolution, the two aren’t remotely comparable.
As for weather predictions, try checking your local news stations. It’s free, readily available, and has well over 80% accuracy. I’d say that’s pretty good. There are much better predictions, but when all one needs to prove my point is turn on a television, I don’t see a reason to look for them.
The foremost problem is that forecasting is in the realm of arbitrary axioms. The problems with weather forecasts,and the increase in model error are well described in the literature eg Nicolis and Nicolis 2006
This is evident when seen in the evolution of the weather forecast model ability of the ECMWF.A widely used model producing forecasts in the range for a few days to a number of weeks. The preparation base is a n-day forecast with n= 10 days of the global atmospheric state.
In any forecast there is an error dependent on initial conditions (due to arbitrary assumptions/estimates of unknown qualities) with the ECMWF model over the last 20 or so years in a paradox the model error has increased.
In 1982 in a seminal paper in which ECMWF data was first used ,to measure predictive ability. Edward Lorenz found the mean error evolution (doubling time of initial error) was two days, presently has dropped to 1.2 days.
This suggest that there is a limiting of predictive capabilities for long range weather forecasting with models of increasing sophistication ,owing to interconnected complexity in the atmospheric dynamics.
Sensitivity to the initial conditions-the principle signature of deterministic chaos-is thus not an artifact arising from when lower order models are used but is, rather, deeply rooted in the physics of the atmosphere.
Similar findings in Orrell 2002
The atmosphere is often cited as an archetypal example of a chaotic system, where prediction is limited by the models sensitivity to initial conditions (Lorenz, 1963;Pool, 1989). Experiments have indeed shown that forecast errors, as measured in 500 hPa heights, can double in 1.5 days or less. Recent work,however, has shown that, when errors are measured in total energy, model error is the primary contributor
to forecast inaccuracy (Orrell et al., 2001).
Brandon, like many weather geeks, I don’t take any notice of computerised forecasts. I just look at the maps, and the calendar. My predictive accuracy using these crude and simple techniques for the next day or two is at least as good as what the weather bureau churns out of its computer. And, my long term forecasting using these techniques is just as poor as the stuff their computers produce. However you approach the problem, the inaccuracy of forecasting increases exponentially with the timeline.
All dynamic systems have this characteristic. That’s why economic forecasting, for example, is voodoo – as has been demonstrated time and time again. According to your rationale, it should be possible to forecast what the economy will be doing in 20 years, even though it is demonstrably impossible to forecast what it will be like in 2 years’ time with any confidence.
johanna, I can’t speak for your forecasting abilities, but I do know weather forecasting in general is quite good at the range of five days. If that can be done without modeling, it just speaks even more against the point Curious George made.
As Johanna writes:
Nassim Taleb covers this well in his The Black Swan.
Is source code of HadCM3, CanCM4, CNRM-CM5, MIROC4h, MIROC5, MRI-CGCM3 and CFSv2 published? Archived? Documented? Available for download, modification and redistribution? If not, there is nothing to talk about.
In a computational model the theory is the code base itself. As long as the theory is kept secret, it is not science, but an arcane ritual.
There are manuals describing the algorithms
I looked there for a calculation of the influence of lunar tides, could not find any. Is it negligible or neglected?
I don’t know how to say your name but another useful web site for model tutorials is liinked; http://www.cesm.ucar.edu/events/tutorials/071210/
Best to not try to learn them all simultaneously but lots of information on the web.
His first name is Peter. The Hungarians put the family name first then the Christian name.
That’s correct, thanks for the clarification.
Codes don’t differ that much. NCAR’s CAM 3 has very complete documentation.
CAM 3.1 code
You can get GISS Model E here.
Great, thanks. What about the set of computational models used in the study, that is HadCM3, CanCM4, CNRM-CM5, MIROC4h, MIROC5, MRI-CGCM3 or CFSv2?
Berényi Péter | May 17, 2012 at 5:19 pm |
It seems you’ve asked a question that’s produced a lot of positive and useful information all in one place.
Thank you for that.
One hopes soon people won’t need to ask, but such information will always be provided from cradle to grave as a matter of course, throughout science.
This is interesting. Not being a climate scientist, this may be a silly question, but here goes:
On the “prediction skill” global maps (Fig. 3) it appears almost across the board that the correlation coefficients for the annual mean surface temperature anomaly over land masses are cooler while over oceans, particularly the tropical Pacific and Indian Oceans, they are warmer.
Am I looking at this right? If so, what does this mean, if anything?
Thanks in advance for satisfying my curiosity.
The AMO is easy to predict, it is generated in the N. Atlantic subpolar gyre, and it has a near perfect natural synchronizing driver.
Every 60 or so years it synchronization breaks down, the AMO gets delayed by about 6 years then re-locks onto its synchronizing driver.
See post : http://judithcurry.com/2012/05/17/cmip5-decadal-hindcasts/#comment-201201
Just to put matters into perspective, irrespective of the all the possible sources of error endemic in these studies – from posing the problem to be solved to the gathering and processing of information – there really is no crisis except involved except the damage to society that has and will result from the government’s use of such studies to manage a non-existent crisis.
On prediction of global mean temperature:
1) El Nino and La Nina appear to be random
2) El Nino causes global warming by up to 0.2 deg C from the neutral position
3) La Nina causes global cooling by up to 0.2 deg C from the neutral position
4) As a result, a +/- 0.2 deg C variation in the global mean temperature is random and cannot be predicted
Do you all agree?
After El Nino and La Nina, the global mean temperature returns to its neutral position
Girma | May 17, 2012 at 7:30 pm |
When examining the past of GMT:
1) El Nino and La Nina (component states of ENSO), although far from random (for instance, either appear more likely within two years of the last time ENSO was in that state) are non-linear and both spatially and temporally correlate weakly with well-understood factors, and/or in some unknown way with poorly-understood factors, such as circumpolar currents, atmospheric influences, Pacific circulations above and below the equator, the complex PDO, biota and teleconnections with continental systems.
2) El Nino is associated with GMT rise of unknowable amount from expected levels ceteris paribus.
3) La Nina is associated with GMT drop of unknowable amount from expected levels ceteris paribus.
4) As a result, variation in the GMT of some dynamically changing amount and direction is poorly understood, and moreover due sensitivity to initial conditions we do not know how much data at a minimum would need to be collected before attempts at many types of prediction could be considered valid, though we can be fairly certain we collect at least two orders of magnitude too little data in terms of granularity on a spatial grid, plus inadequately frequent observations and inadequate different types of measures.
5) As such, there is no ‘neutral’ position for the GMT. Volcanoes, forest fires, dust storms, cyclonic activity of sufficient level, aerosols from unpredictable botanical or man-made changes, and ocean biota changes can each in some way measurably influence GMT on sub-decadal levels enough to render the very notion of such a ‘neutral’ position meaningless.
It would take a special sort of incompetence to agree with you, Mr. Orssengo.
That is a a good try. Why do you have to call me names? I am here here to learn.
No, you are not here to learn.
If you were you would be asking more questions rather than making statements claiming to understand and predict the climate.
And you would be posting fewer funny graphs.
Judy, just to ask a technical question, to stay on topic.
Wouldn’t longer runs of the models you have used be more appropriate to determine the skill of models in making decadal predictions (forecasts)?
Wouldn’t longer runs of the models you have used be more appropriate to determine the skill of models in making decadal predictions (forecasts)?
The forecasts are a decade long. Perhaps I’m misunderstanding the question but that seems to me the ideal length of time for determining skill in decadal forecasting.
I doubt this is anything Judith can control anyway. These were the experiments that were run in conjunction with CMIP5, so that’s the information available.
Bob, there are also some 30 year simulations, but we haven’t looked at those yet.
It appears that YOU are NOT here to learn (since you appear to feel you already know it all), but I have learned quite a nit from the analyses, which Girma has posted here.
curryja: Bob, there are also some 30 year simulations, but we haven’t looked at those yet.
That was going to be my question, but he beat me to it, so I’ll just say:
Thank you for a good paper, and I look forward to reading the sequel.
See note below on the number of runs needed. Fred Singer’s testing suggests 20 runs are needed for run lengths of 20 years; 10 runs are needed for 40 year lengths. Thus expect 40 runs needed for 10 year lengths.
Girma | May 18, 2012 at 1:20 am |
Mr. Orssengo, I do not call you names. I call the situation as I see it.
I’ve been out of this sort of analysis for decades and still can spot dozens of basic errors in technique, specious usages, mistakes, fundamental flaws and simple falsehoods in almost everything you present.
A halfway competent professional should be able to find much more wrong with your technique, and provide much better advice on its correction. You’re a professional engineer and a PhD with access to exactly that sort of help and an obligation to uphold a higher standard, and yet at every turn you fail to avail yourself of such simple remedies.
I have no problem with people making mistakes, even those who ought be more careful. I have a real issue with people clinging to their mistakes and spreading them around, confounded well-meaning but easily influenced readers who lack the background to tell the legitimate item from such inferior and wrong depictions.
In other words, if someone is actively harming others through their conduct, and I see it, don’t I have a moral obligation to say so?
Though perhaps asking you moral questions, considering your track record, is wasting my time. Again.
It didn’t return after the 1998 El Nino, did it?
It is at the neutral position (along the long-term trend of the secular GMT) as shown in the following graph => http://bit.ly/HRvReF
GMT for 1998 => 0.53
GMT for 2011 => 0.34
A drop of about 0.2 deg c
A further drop of 0.2 deg C below the current neutral position in the next two decades exactly like after 1955.
How do we know the position is neutral? It is on the graph.
How do we know the graph is correct? It shows the neutral position.
There is a remarkable degree of skewness and kurtosis in your graph for your ‘neutral’ line to be considered really neutral. One notes that , the frequency of excursions below the ‘neutral’ line are rather lower than the frequency above the ‘neutral’ line. Which makes it a pretty poor candidate for ‘neutral’.
Here is the frequency histogram for the residual GMT (= Observed GMT – Smoothed GMT)
Uni-modal and symmetric!
Girma | May 18, 2012 at 1:12 am |
Tch. 102 samples. Is that for 102 years, then?
What about for 1224 months? Or better, for the last 204 months? For the 204 months before that?
From 1910 to 2011
There is an annual climate signal. We are talking about the climate. Forget about the huge monthly variation. You don’t consider months in climate analysis. You consider the annual value.
Mr. Orssengo you suffer the Goldilocks problem.
Indeed, for the climate signal you don’t consider months in isolation. You consider what span of months gives a confident strength of signal.
The confidence level of 95%, 19 times in 20, is roughly 17 years, not one single year.
30 years is better, but we then have the following problems:
We have too little data of sufficient quality to produce a valid demonstration of signal in the sense you use it. Before 1960 the standard deviation overwhelms the natural variability. (Let’s not even mention error bars.) After 1960, half a century is a really short span of time.
Which is all distraction in any event. One does not throw away information about goodness-of-fit with so slim pretext as ‘because we seek the climate signal’, even if one wishes to highlight that signal. One always uses so much information as one has available. Compress:12 throws away 92.5% of the information about fit. It’s simply technically wrong to do it as you have done. Either you do not understand this simple mechanical limitation and are practicing beyond your competency, or you do and are scamming. Both are shameful.
IPCC uses only annual values => http://bit.ly/b9eKXz
The name calling is baseless.
Girma | May 18, 2012 at 2:57 am |
What namecalling? Shameful isn’t a name. It’s a remark on how disappointing it is to face shameless falsehood.
The graph you link to, for example, while it uses annual data also shows clearly the general region of monthly values, and attempts no fit shorter than a quarter century.
You, on the other hand, show from annual data fits that exceed the precision of the monthly source data — a mathematical impossibility!
And on top of this, you speculate on the CI of the compressed annual data only and call this a foundation for your R^2 — a strict no-no.
That you repeat these techniques shows either complete mathematical incompetence, or gross lack of ethics.
Which one is it?
Bart R: “Either you [are incompetent], or you are [a] scamm[er]. Both are shameful.”
Girma: “The name calling is baseless.”
Bart R: “What namecalling? Shameful isn’t a name … [I repeat. Your] techniques shows either complete mathematical incompetence, or gross lack of ethics.”
Bart, please comment on IPCC’s AR4 GMT graph. Do you expect it will be updated and in AR5? Why or why not.
blueice2hotsea | May 20, 2012 at 10:47 pm |
The actions, not the person. I don’t know the person. I know what he’s posted. http://yourlogicalfallacyis.com/strawman – If you disagree with the description of the actions, back up your counterargument with evidence, not changing the topic into ad homs that just aren’t there.
IPCC’s AR4 GMT graph. Do you expect it will be updated and in AR5? Why or why not.
I expect GMT graphs will be important especially in chapters 9 & 11-14 of IPCC AR5; further, we have seen five new years of data so far, BEST’s very significant contributions, HadCRUT4, new satellite developments, new thinking on GMT trends, serious blowback from communications and reasoning problems with IPCC AR4 GMT, and an expansion of the expertise applied to the topic; as such, I’d expect considerable updates of the graph next year.
I even expect someone will take a whack — probably in conjunction with the paleo reconstructions — at cyclic trend topics in the report, too.
Of course, I can’t see the future, don’t know the people, and was surprised by some of what was in each of the previous reports, so I’m not the best source of insight into the question.
Thanks for your comments wrt the AR5 GMT graph. I
also think it ought to be important.
Sorry, I have no idea what your comment means. MY comment consisted of two parts: a reduction of your exchange with Girma & the curiosity which it provoked in a possible GMT AR5 graph. (Basically, a disingenuous front is cover for a disingenuous back.)
Finally, your criticisms of Girma’s efforts would be more effective if also accompanied by criticisms of the GMT AR4 graph. WHT for example is not exactly impressed by that graph and he says so. For that reason and others, WHT is building some serious credibility. You are obviously very intelligent so when you resort to bs-ing it’s very weird and seemingly unnecessary.
blueice2hotsea | May 24, 2012 at 8:59 pm |
You want me to confuse Mr. Orssengo even further by describing mistakes he _didn’t_ make? ;)
Finally, your criticisms of Girma’s efforts would be more effective if also accompanied by criticisms of the GMT AR4 graph.
Have you noticed me disagreeing with WHT’s criticisms?
(I mean, other than where he’s clearly wrong?)
I cover problems with IPCC claims and statements routinely. When I disparage Mr. Orssengo’s projections in cases he includes the 0.2C prediction, I seldom mention the prediction without prefacing it with the words “very silly”.
When I mention the data available, I generally preface it with phrases like “inadequate by one or two orders of magnitude for the uses it is put to.”
My criticisms of the AR4 graphs aren’t lacking. However, as those problems are of too ambitious use of the evidence on hand, generally, and Mr. Orssengo’s are of elementary competency, there’s nothing I can do about the problems with the IPCC graph other than promote greater data collection by two orders of magnitude. For Mr. Orssengo, I can actually help by pointing out substantive errors of technique and reasoning.
Besides, the authors of the IPCC graph are working on fixing their worst mistakes, one presumes, and will release their graphs in AR5, scant months from now. They’ve had the chance to read my criticisms plentifully over the past four years and take them into account.
I combined the names of race horses and jockeys and then examined the letter combinations that produced profanities; the model was that the pair with the largest number of swear words would perform best.
Upon testing I find that I have considerable skill in predicting the winners of steeplechases of less than 3 miles, p=80%, but less skill when predicting loner races or on the flat.
I am seeking Federal funding based on this initial success to expand the model by increasing the size of the profanity index; essentially by three mechanisms;
1) visiting places where unusual profanities are uttered (justification for line item 2 in the Budget; “The Three Year Global Pub Crawl),
2) persuading the compilers of dictionaries that words such as ‘Brity’ and Thomst’ are in fact profanities (justification for line item 3 in the Budget; “The Russian Mafia and Linguist re-education”),
3) rerunning races which have provided the wrong outcome to reduce the statistical variability (justification for line item 3 in the Budget; “The Italian Mafia and the Fix-Is-In”).
I have high hopes.
I hope Judith responds to the above comment from DocMartyn when she gets time – very eloquent (even if it does not necessarily meet the ‘technical’ standard she indicated at the opening.
I’ve seen an application very like this once.. Well, more than once.
Rejected due to failure to attach ANOVA to the grant request form each time.
One hopes yours is in order, and wishes you all the best..
Doc, you remind me of one of the classic old “white ethnics” jokes of my youth:
How do you tell the Pole at the cockfight? –He’s the one with the duck.
How do you tell the Italian at the cockfight? –He bets on the duck.
How do you know the Mafia is there? –The duck wins.
Apologies in advance to everyone whose ears sting at such ethnism.
Further apologies to animal rights folks. We all know that cockfights are by definition NOT FUNNY. Sorry.
Since we’ve drifted off-track to “white ethnic” stories about “fights”, I’ll add this one about Spanish bull fights (also with apologies to the animal rights folks), taking the risk that our host will squelch it..
Bullfighting has a long history in the Spanish hilltop town of Ronda in Andalucia, southern Spain. This is said to be the place where modern bullfighting began. One of the oldest bullrings in Spain is located in Ronda. It is the venue for many of the most spectacular corridas of the year.
The bullfighting season usually starts on Easter.
It has been a long tradition in Ronda that after the first major bullfight of the season, the mayor of Ronda was presented with the “cojones” of the slain bull, which were cooked in a traditional sauce and served with a bottle of the local red wine. This comes from the mountainside vineyards located at around 750 meters above sea level, a necessary requisite for making high quality tables wines so far south in Spain.
On one such occasion the mayor was disappointed to see that he was being served what appeared to be two shriveled meat balls rather than the traditional “cojones de toro” that he was so fond of.
When asking the waiter “what happened here” (¿que se pasa aqui?), he was told:
“Sorry, señor, sometimes the bull he does not lose…”
Very good. :)
This seems to me progress in the right direction — i.e., actually using the scientific method.
I have a technical/methodological question, though, for you or anyone else with knowledge on the subject:
If a model does a good job passing this sort of test, that would, of course, be good news. Even better would be passing a prospective test of this sort (i.e., on future data over a decade or two).
But it seems to me that is quite possible that a model could legitimately do a great job on this sort of test, but, as CO2 levels continue to climb, the model might end up failing. I.e., it might be that the model truly does do a good job for CO2 levels at the present times and for recent decades, but that the model does not extrapolate well into a higher CO2 regime.
Of course, this is most likely if the model uses simple curve-fitting, data-fitting, semi-empirical techniques to establish the model equations and parameters. But, even if the model equations are based on first principles, it could well turn out that effects that are now negligible and not included in the model become important as CO2 levels rise.
I’m not sure how one could rule out such a possibility.
Naturally, if a model gives absolutely stellar performance in every respect, one might hope that it would continue to perform well. But, it is easy to come up with examples from modeling in other fields (basically, non-linearities that kick in strongly) that show that this hope is often misplaced.
I’m not criticizing your paper but just wondering how to go beyond it to really nail down if a model can be trusted.
Dave Miller in Sacramento
or anyone interested in the 300 years (1710 – 2010 ) AMO reconstruction from its natural driver
Is the AMO a player at it’s own in climate? My doubts increased after this operation: Take the TLT-Data from RSS from 1/1979…4/2012 and calculate the least-square 204-month trends according to Santer et.al (2011) (http://muenchow.cms.udel.edu/classes/MAST811/Santer2011.pdf ).
Do the same with the monthly AMO-Data and plot both series ( centered at 102 month): http://www.dh7fb.de/reko/amorss.gif .
You see that the AMO-trends follow the TLT-trends with a timelag of 12 month. The correlation is 0.95. It seems to me that the AMO is a result of the TLT and not a driver?
AMO is just a temperature index like any other. I don’t understand why it’s often de-trended. Makes no sense to present it like that. There is a pretty strong correlation between solar cycle frequency (or inverse SCL) and AMO.
These are what I call “idle thoughts of and idle fellow”. I am trying to get my mind around the discussion of what is, and what is not, “data”. The word I find missing in this discussion is “replication”.
What I was taught in Physics 101, was that the key to having good data was replication. I go out an make some measurements, I put +/ – values on the number, and I publish the results. Then someone else repeats the measurement somewhere else, and does the same sort of thing and measures what I measured. If the numbers agree within the limits, then we agree that we have some reliable data. If the numbers dont agree, something is wrong somewhere, and more measurements are required to resolve the issue.
So, I hope we can agree that there is no such thing as good observed data, if there is no replication process. Having good replication is a sine qua non of good, reliable data.
But what is the meaning of the term `replication`, if we are talking about the output of computer models? If I develop computer code, add the input data, I will get an answer. If someone else gets my computer code and input data and repeats what I have done, then he will get exactly the same answer.
So here is my question that I hope someone can comment on. I know what “replication“ means when I do actual measurements. What does “replication“ mean when we are discussing the output of computers? How do we, properly, replicate the output of computer programs?
And that, Jim, in a nutshell, is why the output of running computer models is not ‘data’ in the scientific sense, and running the model is not an ‘experiment’ in the scientific sense. Even if the output of the model hindcasts correctly, or matches what actually happens after the event, it does not prove that the assumptions are correct. It only proves that the answers match real, measurable information, which used to be called ‘data’ in the days before post-normal science.
If someone goes to the same spot as you and measures the same thing as you, they should get the exact same answer as you. The only reason they wouldn’t (assuming nothing in the environment has changed) is human error.
Did they replicate your answer? If so, how are computer models any different? Are we to say human error is key to replication? Of course not. Replicating results isn’t done by repeating the exact same process on the exact same thing.
You replicate answers with computer models in the same way you replicate any other answers. You repeat the process with alterations. You may alter the sample (data) used, or you may alter the methodology (code) followed, but you must alter something.
Saying you’d just run the same code with the same data is like saying someone would try to replicate my efforts to find the size of soda cans by coming over, using my ruler, and measuring the same cans as me.
Physics use to have the two branches of theoretical and experimental. In theoretical physics, replication means that a theory is described well enough that the calculation could be independently performed by an independently researcher. You don’t get independence by using the same computer code.
The theory is tested against measurement by calculation of a prediction whose measurement can be made in the future (i.e., Einstein predicted light would bend by gravity; experimentalists developed measurements to confirm the theory). None of these steps seem to be use for climate theory.
Judith, in Kim et al (2012), the text and correlation maps typically describe the eastern tropical and North Pacific as having the poorest multiyear correlation. I would have to assume the models are showing too high a temperature and too great an increase in temperature in these areas. Is this correct? If so, that would be consistent with the failings of the CMIP3 model hindcasts/projections for the past 30 years.
more generally it would be very interesting to know how the specifics of these CMIP5 runs, which do not have a complement in CMIP3, correspond with the generalities of the CMIP3 model runs?
I hope Judith will respond to Willis’s questions and points raised. I was writing a comment very similar but think he captured the concerns well.
Here’s a test for the CMIP5 models, assuming that there are 1×1 monthly gridded datasets available for ocean temperature.
1. For each gridbox in the ocean between 0-2000M, detrend the temperature timeseries and measure the variance.
2. Calculate the zonal mean of the variances (i.e. average out the longitudinal dimension).
3. For each depth level, normalize the data by standard deviations.
4. Produce an image plot of the variances by latitude and depth.
5. Do the same using interpolated Argo data and compare the plots.
Why do I think this could be important? Because one can use the variance of the timeseries to calculate the average power of the seasonal signal. Perhaps this can be used to assess heat uptake? I did this last year against the CMIP3 models. In my opinion, none of the models adequately matched the Argo pattern. See for yourself:
Then again, maybe I messed up or it doesn’t matter. All the usual caveats apply.
Further to my previous comments about reanalysis “data” being nothing but computer model output, I just compared the ERA40 and the NCEP/NCAR reanalysis model output with reality in the form of the HadCRUT3 analysis.
The R^2 of the surface temperatures in both cases is about 0.85, with the errors having a standard deviation of a tenth of a degree, and a maximum of about a quarter of a degree. The R^2 of the first differences, on the other hand, is only about 0.5.
I also looked at the two reanalyses 700mB lower troposphere data vs UAH MSU … the R^2 is worse, only about 0.7.
Did you sample the reanalysis to match the sampling of hadcrut both in the temporal and spatial domain?
A much better test would be to use a more complete dataset such as Berkeley earth which is more complete in the spatial and temporal domain.
If you dont mask the model results to match the collection structure of the observations you are comparing apples and oranges
Thanks, Steven, but I’m not sure what you mean. I have used the HadCRUT global data, and compared it with the NCAR and ERA reanalyses, again global, for the period of overlap (1958-2002 from memory).
How would you suggest I subsample it?
What if you wanted to make direct comparisons of regional temps at similar gridcell size and observation intervals – but they didn’t match up. Might be easier (and cheaper) to “normalize” observation data rather than model output.
What would you do?
More Lukewarmer stuff on sensitivity
That’s some hotstuff, there, moshe; the interplay of albedo. Sure, for a half century or so, but what about the last decade’s albedo and radiation records?
steven mosher and kim
OK. It’s only a model simulation – BUT (as I understand it) it does show that a long-term climate response (1,000-year response to 2xCO2) = 2degC gives a reasonable fit for reconstructed conditions of the past millennium taking into consideration all the other factors we THINK we know.
So, if we believe that CO2 will increase to 780 ppmv in the next 100 years or so (sort of an upper limit of reasonableness), this means we might have 2 degC warming above today’s level sometime beyond this time.
Not very alarming.
(Living in Switzerland I say “bring it on!”)
Stop being stupid.
When Girma and others post their toy regressions of observations to get sensitivity YOU SAY NOTHING. But when that same calculation is done in a much more systematic fashion you call it a model.
I guess the problem is that “more systematic” doesn’t rule out more fudge factors to reach the pre-conceived result. That’s a “system,” but it’s not one I care to trust.
In any event, these brief and nasty posts that have become your modus operandi recently aren’t really helping me understand things, the way your posts used to do. So help me out. Is the entire point of this post to complain about his use of the word “model” in his first sentence? Or are you trying to make a more substantive point?
Steven Mosher has been cranky ever since the moment he discovered There’s No Raw Data in Climate Science.
moshe’s cranky ‘cuz he’s figured out that warming is a blessing and cooling a curse. He may also wonder that the sun is acting up, in a possibly meaningful climatic sort of way.
dont forget my nickname is moshpit. My primary concern here is the lack of skepticism on the part of skeptics. If you want to estimate ECR and TCR from observations there is a way to do it. You have to start by specifying an underlying physical model. Girma’s approach is NON PHYSICAL. in his approach you have Temperature on the left hand side of the equation and Time and Angle on the right hand side. We know this cannot be physically correct. We know that the curve has merely been fit to the available data. It explains nothing.
When you have an underlying model ( equation) that is physically constrained you can derive many hypothesis from it. So, very simply
you should be able to ask Girma “Given your model, what will the temperature be if the sun increases its output by 3 watts? and , if he has a physically realistic model he should be able to answer. Or, Girma, If we have a huge volcano or more S02 in the atmosphere what does your model predict? But He cant answer that because he doesnt have a model that is quantified over known parameters. His model is quantified over
Year and Angle. That is the temperature is a function of the year and the sin() and cos() of an angle. no physical parameters. you should ask him, what does you model predict for the holocene? for the LGM?
To do that you have to model the PHYSICAL SYSTEM. We know on inspection that Girmas model cannot be true because it does not take the correct form. It’s not an explanation. An explanation would show how the earths final temperature is a function of known quantities. It might be a incomplete explnation, it might be inexact here and there, but it should at least have the correct form and should operate over parameters known to influence the climate.
For example: here is a nice little simple equation to get the Top speed of a car
Those equations are physically realistic. Yes they are simplified, but
you will note how they are dimensionally correct. You will also note
that they allow you to make various predictions. predictions that are tied to the quantification structure of the formulas. What happens if I increase the drag? what happens if I increase the frontal area? the horse power..
As a model its pretty simple. a model is nothing more than a set of equations that capture physical laws. By capturing physical laws one phenomena ( speed) is EXPLAINED in terms of other phenomena
Horsepower: frontal area, friction, drag, etc. That is what science does.
It explains X in terms of y,z,p,q,r,z.. We test this by varying the parameters: what happens if we increase y? decrease p.
Girma’s approach is NON PHYSICAL. in his approach you have Temperature on the left hand side of the equation and Time and Angle on the right hand side.
You are correct.
However, to solve any problem, you can use a theoretical or empirical method. Man used to build a working bridge before Euler’s equations were discovered.
In engineering, when we face a problem whose governing differential equation is unknown or is hard to solve, what we do is to experimentally establish a repeatable relationship between relevant variables. In my empirical model, that is what I have done. It establishes a pattern between the global mean temperature and time. Statistically as nearly 100% of the observed global mean temperature lie within the GMT band for the last 100 years, it is reasonable to assume it will continue to do so at least for the next 20 to 30 years.
The results from both the theoretical solution by climate scientists and my empirical model must converge on the same solution.
Steven Mosher is a good guy but he is bad company. His great analytical ability is channeled in wrong direction, if he could devote part of his intellectual ability to synthesis of ideas, rather to dissection of a ‘dead sheep theory’ he could made important contribution to science.
Steve me boy,
Stop being sassy.
Girma’s regressions were not made with a computer model, they are simply attempts to analyze physically observed data (in this case the historic HadCRUT3 temperature series). I think Girma does a good job of this.
I have nothing against computer model simulations – as long as one recognizes exactly what they are (and aren’t). The ones you just cited are interesting, as I commented.
Girma’s “projections” for the future make as much sense to me as the “projections” that come out of multi-million dollar computer models.
If I were a betting man, I’d give the edge to Girma, as a matter of fact, for a very simple reason..
If you’d like to know WHY this is so, I’d be glad to tell you.
There is no difference between fitting a series of points with a formula and using a formula to estimate coefficients. Dont be innumerate.
The proof of the pudding is in the eating: 100% of the last 100 years global mean temperature lie within the GMT band as shown => http://bit.ly/HRvReF
Need I say more?
Let the above graph defend itself. It does not need any explanation. The AGW advocates’ blog have not allowed me to post it. They just delete it.
Empirical physical data (even a temperature record, which we all know has several flaws) provide information.
If someone like Girma takes these data and (pardon the expression) “connects the dots” to arrive at a mathematical correlation, then this may provide us even more information than the raw data themselves can.
Do we presently know everything there is to know about what makes our climate behave the way it does or could there be some as yet not clearly defined factor or combination of factors, which is causing the observed multidecadal cyclical nature of the observed temperature, which Girma plots?
My personal conclusion is that we are still in the very infancy on our learning curve regarding our planet’s climate. As our host here has reminded us frequently, “UNCERTAINTY” is writ large.
Climate models are wonderful tools and they help us immensely in analyzing the world around us, but in reality, Steven, they are nothing more than extremely expensive, very rapid versions of the old slide rule of yore and their biggest shortcoming is not that they are not yet powerful enough, but the old GIGO problem.
Those are my thoughts on this , Steven, and just because you and I may not agree does not make either one of us “stupid”.
In this case Imo you are the one writing something stupid.
Not criticizing Girma’s forecasts does not equate to believing it is some outstanding form of modeling.
Someone stating that the model’s results do not demonstrate a probable harm to humanity also does not mean that a person either agrees or disagrees with a process of model development.
You are wrong. We have a name for this over at CA it is called the silence of the lambs. we criticize people all the time for failing to stand up and be counted WRT the mistakes that Mann made. I hold all people to the same standard. when you see BS if you fail to speak up, I discount your opinion on things. If Girma said 2+2 = 5 and you sat by silently, I discount your opinion. His approach to things is nothing less than the equivalent of 2=2 =5. I repeat my invitation to him to put his stuff up at Lucia’s and take a beating. He wont. There is a reason for that. figure out what it is and get back to me. he doesnt even understand dimensional analysis
You should be careful Steven Mosher. There are plenty of people who don’t talk about what Girma does because they think it’d be a waste of time. For example, what would be the point of me saying anything? I’m just a random person on the internet. It wouldn’t change anything.
The silence of the lambs referred to professionals remaining silent while their field of work was abused. That’s a far cry from random people ignoring other random people because they have better things to do with their time.
If and when Girma is relevant to a point I’m making, I’ll discuss what he says, but until then, I’d rather just pretend he doesn’t exist.
There is so much dross in the climate debate.
If the relationship between global mean temperature and CO2 concentration can not be determined from the observed values after removing from the global mean temperature components that are oscillating or cyclic, how are you going to determine it?
By the way, IPCC’s 0.2 deg C per decade warming includes a 0.12 deg C per decade warming due to ocean cycles and it is wrong => http://bit.ly/HRvReF
The current observed global warming rate is only 0.08 deg C per decade, which is less than IPCC’s estimate by a factor of 2.5. IPCC’s projection of for a warming from 2.4 to 6.4 deg C warming by the end of this century is baseless.
This one goes a little farther than Gillett and Held because it has low estimate of TCR and a low estimate of ECS. It would seem to add fuel to the “show me the cloud feedbacks” fire. Is it inconsistent with Spencer & Braswell?
I’d hardly call spencer and braswell the gold standard here.
There is a certain mathematical attractiveness to this fractal approach, and I especially like the fact that it does include solar forcing (which drives much of the ocean cycles), aerosols (natural and anthropogenic) as well as of course greenhouse gases. It does explain pretty well the cooling observed between the 1940’s to 1970’s, as well as the relatively flat atmospheric temperatures of the past decade. What I like best about it is the specific range of forecasts given by the fractal model going forward and the role that anthropogenic aerosols will play. How quickly China and India get some controls on the aerosol emissions will have an impact, and this model predicts a nice set of scenarios for that.
As this is a new mathematical approach to modeling, I would appreciate someone with knowledge of fractal mathematics to have a look at it and maybe decipher it for some of us without this specific background.
Nonsense. Fractal relates to the supposed multiple scales of the power law and not to the method which involves fitting a power laws.
Compare this with Lean’s multiple regression model fo instance – http://s1114.photobucket.com/albums/k538/Chief_Hydrologist/?action=view¤t=lean_2008.gif
It is simply a different way of fitting curves to temperature data and doesn’t mean a thing unless all of the processes are identified – and they are not if clouds, snow and ice are not included.
You are entitled to your opinion, but, based on other recent studies on the role of both aerosols and solar modulation as they relate to the LIA and the NAO, I think there is something worth considering here, and I especially like the attributions to these from the conclusion:
“On assumption of the correctness of a strongly modulated solar irradiance (Shapiro et al. 2011) and by using recent data on SO2 emissions (Smith et al. 2011) the model provides tentative explanations for conspicuous trends in global average temperature from Middle Ages up to now (Figs. 5, 7a, b, 8). The Medieval Warm Period and the subsequent Little Ice Age are primarily attributed to a decreased solar radiation in the latter period. The rise of the temperature from the early 19th century up to about 1950, including the fast 1910–1940 rise, is for about 70 % attributed to an increase in solar radiation. The increasing warming by CO2 up to 1950 is partly offset by increasing cooling by SO2. The slightly cooling climate of 1950–1970 is attributed to SO2 cooling overtaking CO2 warming because of fast economic growth without much pollution control. The warming of 1970–2000 is attributed to increasing warming by CO2 and decreasing cooling by SO2 because of stringent air pollution policies. Finally, the post-2000 period with an apparent lull in temperature rise seems to replay the 1950–1970 events, with now China displaying fast economic growth with, initially, little pollution control.”
Other studies have also found the cleaner air that began in the 1970’s by way of reduced aerosols had a warming effect, that may be coincident with or even causally related to warm ocean cycles that began about the same time.
The 1940-1970 cooling is natural, not man made:
“Interdecadal 20th century temperature deviations, such as the accelerated observed 1910–1940 warming that has been attributed to an unverifiable increase in solar irradiance (4, 7, 19, 20), appear to instead be due to natural variability. The same is true for the observed mid-40s to mid-70s cooling, previously attributed to enhanced sulfate aerosol activity (4, 6, 7, 12). Finally, a fraction of the post-1970s warming also appears to be attributable to natural variability.
A vigorous spectrum of interdecadal internal variability presents numerous challenges to our current understanding of the climate. First, it suggests that climate models in general still have difficulty reproducing the magnitude and spatiotemporal patterns of internal variability necessary to capture the observed character of the 20th century climate trajectory. Presumably, this is due primarily to deficiencies in ocean dynamics.”
You may shift the goalposts if you like buy I like to be precise in mathematical terminology. The fractral refers to the similarlity of the power law behaviour at all scales.
‘In this article, a multi-scale climate response model was fitted to temperature records encompassing time scales ranging from a year to a millennium. On assumption of the correctness of a strongly modulated solar irradiance (Shapiro et al. 2011) and by using recent data on SO2
emissions (Smith et al. 2011) the model provides tentative explanations for conspicuous trends in global average temperature from Middle Ages up to now (Figs. 5, 7a, b,8).’
21st century SO2 forcing is 0.1 W/m^2 – either from volcanic or anthropogenic aerosols. It is a little more complex for anthopogenic aerosols because of the interactions of sulphate and black carbon. http://www.nature.com/ngeo/journal/v3/n8/full/ngeo918.html
So it doesn’t explain the period for which there is the best data available. It is curve fitting – and as I say is reliant on knowing all of the forcing functions. This is obviously not the case. It doesn’t for instance reference the CERES data at all at the near scale and we don’t know about snow, ice and cloud going back to 1200AD. Don’t you see? The mathematics doesn’t change anything fundamentl about what is known about forcings. So tentative is entirely the correct interpretation. Claiming more than this is I am afraid just climate warrior claptrap.
This is indeed what Webby does – fits a power law to a curve. This paper is world’s more sophisticated. Webby’s relationships are hopelessly simplistic based always on some insanely misconceived conceptual model. Fat tailed ot thin tailed is entirely irrelevant is the context of a curve that is fitted by incomplete or incorrect forcing functions.
Chief is misguided and a bit lost in the ozone. He doesn’t realize that one of the standard solutions to a class of diffusion problems is also scale-invariant. Here are three views of a well-known diffusional growth law at three different time scales:
They all look exactly the same, and this has nothing to do with fractals. Chief always lashes out at things he does not understand. Right now he is upset that a guy is modeling a system by using slab diffusion approximations. But because he doesn’t understand what exactly van Hateren is trying to do, he goes after me. Excellent.
We get back to the definition of a fractal – which is typically that of similarity at any scale and this is the sense that it is used in this paper. The mathematics involve ‘multi-scale’ power laws as in the quote I previously provided.
I don’t have any particular problem with the approach – nor indeed with the Lean paper earlier cited. Simply that it is incomplete if all of the forcing functions are not included – and they are not. If this is treated as ‘tentative; – well this speaks for itself. Exploring techniques is not the problem – it is drawing hard and fast conclusions from the tenuous nayture of the data.
Webby I have a problem with. His functions are not merely incomplete but physically incorrect. He then treats this as a law of nature and is insulting and abusive at the same time. I suppose I should just ignore it – but it is especially eggregious intimidation and not at all discourse in good faith. Stupid and abusive is not something I respond to well.
Chief is so ignorant that he believes everything must be complex. Andy Grove wrote his PhD thesis around Fick’s law, started Intel with that knowledge, and the rest is history. Chief blubbers away saying nothing can be explained because of chaos and complexity.
Many years ago I struggled through a paper on a 14 compartment carbon model for Chesapeake Bay. I was pleased wit myself that I worked through the detail systematically – it took me weeks – until I got to the last paragraph which said that they needed a model with a couple of dozen new trophic parthways.
With this paper I have simply agued that the approach is inadequate unless all the forcing functions are defined,
Webby first of all claims it is too difficult for me to understand and that then I ask for impossible perfection. I ask for realistic representation of the world. Webby for instance has a global so called carbon model that consists of one pathway he calls diffusion. His insistence that hard and fast conclusions can be drawn from this is beyond absurdity.
“Webby for instance has a global so called carbon model that consists of one pathway he calls diffusion. His insistence that hard and fast conclusions can be drawn from this is beyond absurdity.”
Chief, I really don’t mind the attention, because what van Hateren has done with this paper is something I am very much in favor of. Tsonis is your guy, van Hateren is much more my style.
There is a difference between van Hateren and yourself. His language is of tentative suggestions – quite proper scientific restraint. Yours is climate warrior nonsensical certainty coupled with self-aggrandisement and bullying abuse. The van Heteren model I suggest lacks a critical albedo function – although both theory and data suggest a changing albedo. Your single parameter curve fitting has no worth at all.
It is not a competition between Tsonis and van Hateren. Tsonis demonstates numerically a new dynamical mechanism for major climate shifts – van Hateren uses curve fitting in a way that seems to me to be incomplete. I have nothing against either.
You on the other hand have a breathtaking ignorance that refuses – or is incapable – of evolving beyond the nonsensical.
For those interested in Van Hateren’s paper, you might be interested in what I posted at The Blackboard. It would appear the model was fitted to two temperature reconstructions, using 781 data points a piece. That was less than half the data available. When I reran his code with it set to use almost all the data, the resulting climate sensitivity dropped by ~0.3 C.
The paper mentions he only uses a portion of the data for those reconstructions, but no reason is given, nor is the effect of the decision disclosed. I personally would argue the reconstructions shouldn’t be used at all, as they are without merit,* but if they are going to be used, they should be used in their entirety. If not, an explanation needs to be provided, as should a disclosure of the impact. The author should not be allowed to arbitrarily truncate his input series, especially when doing so changes by an amount roughly equivalent to his error margins.
*I won’t say more as I assume nobody wants a discussion of the hockey stick controversy.
Fractal modeling is essentially the same as random walk that can contain even more randomness in the distribution of the hop rate. The adjustment time response of CO2 could be considered fractal because the tails are fat. All that is well known and already accounted for. These kinds of models show a collection of exponential responses with coefficients chosen to approximate a fat-tail or fractal response (as an example in Archer’s models and the BERN model).
He is mainly doing the thermal response via a very simple fat-tail approximation, which is to place a slab layer between the upper surface and to the deeper heat-sink. This approximates a diffusional process in as simple a manner as one can imagine. It essentially fattens out the mid-range of the tail, and if you need more fat in the far-reaches of the tail (i.e. long durations) then one keeps adding slab layers to create a multi-compartment model. This is the fast-slow model approximation of diffusion that I have been working on for the last year or two. He picked a response that is right between a first-order damped response and the classic Fick’s law diffusional response. I think it is a fairly modest fat-tailed response.
I appreciate this kind of model because it follows my own intuition. Climate response is more a factor of disorder, dispersion, and diffusion than that of chaos, criticality, and complexity. The only negative is that I don’t use the term fractals myself very often because it brings in a certain amount of baggage.
“The adjustment time response of CO2 could be considered fractal because the tails are fat. All that is well known and already accounted for. ”
But so is the null hypothesis, a constant e folding time. I haven’t seen anything that refutes it, yet. For example, a two month lag in the annual cycle, a four month lag in the ENSO cycle, and a ~zero time lag in the Milankovich cycles are all consistent with a constant exponential decay.
What is the observational basis for varying e-folding time? Perhaps the asymmetry between the long march into a glacial period and the quick rebound out. But that just points to one factor in one direction and another factor in the other.
I think the tail is skinny.
So CO2 levels in the atmosphere and the ocean, at all depths, are supposed to be in ‘equilibrium’, but no one wonders why the oxygen levels in the atmosphere and the ocean, at all depths, are not at all in ‘equilibrium’.
Are you serious? All pure diffusional processes have a kernel solution that follows a power-law temporal dependence. It’s actually rare that you will find something that is exponentially damped with a diffusional process mechanism. Pekka hasn’t been around here for awhile, but he would tell you the same.
I will give you an example: How fast did that oxide grow on the silicon-based CPU that you are currently using to spew out garbage? You probably don’t have a clue.
Hint, it wasn’t based on an “e-folding” time, ha ha.
I will grant that van Hateren is perhaps a bit naive in his understanding of the way that physicists model physics, but he is doing a fine job of bookkeeping and creating something that is tractable.
Theory says it is fat. And the experiment evidence is bearing that out — and of course it has to, otherwise all of our previous statistical mechanical models would be proven false. And your CPU would never have worked because all of the theoretical oxide growth tables would be wrong.
How many PPM is the O2 concentration, compared to CO2?
But more salient than that is the arrow of entropy. What on earth do we expect to happen when we unearth huge deposits of previously sequestered carbon and turn that into CO2? Clearly, that CO2 has to make it slowly back to a similar sequestering site, with the only catalyst or accelerator available the evolution of the biotic system to take advantage of the excess. As it stands the central limit theorem says that billions of entities will not instantaneously change the rate of CO2 update. It will remain the same until evolution occurs, and since that can take an indeterminate amount of time, we have to wait for the sequestering adjustment time to pass
AJ is right, some of us understand how the carbon cycle works.
BTW, no one really cares about the semantic haggling over “steady-state” versus “equilibrium”. That is a deadly bore.
Insert “back” and define “huge”.
WHT, I’ve pointed this out to you once before, but the WEC tells us that the total optimistically estimated extractible “previously sequestered carbon” on our planet only amounts to around 2,800 Gt. This compares with a total carbon sink in the atmosphere, ocean, biosphere and lithosphere of 44,000 Gt and would be enough to get us up to a bit more than 1,000 ppmv in the atmosphere when it has all been used up.
The practical upper limit for the atmospheric concentration (to be reached in the next 150-300 years) is probably closer to 800 ppmv, or around 2x the current concentration.
So let’s not paint the Devil on the wall, WHT.
Relax. Panic is not in order.
It’s not about panic. It’s about buying a clue and doing the models correctly.
“What on earth do we expect to happen when we unearth huge deposits of previously sequestered carbon and turn that into CO2?”
Love of theory. We do the same thing to the “previously sequestered” hydrogen, so what?
Here’s an interesting model retrofit based on a “fractal response”. It fits data all the way back to about the year 1200. Uses solar forcing, aerosols (natural and human), and of course greenhouse gas emissions. I’ve gotten private responses from both Dr. Curry, Dr. Trenberth, et. al on this. Take a look:
It was posted above, but thanks. Is this some kind of 1D simulation? Sorry for not using the correct term.
OK. From the introduction:
“In addition to uncertainty of model structure and parameters there is also uncertainty with respect to the energy flows driving the climate, the forcings. In particular, solar irradiance and the effects of anthropogenic aerosols
Ah the forcings. Their solar forcing is not serious. It’s a nice play, but someone will take it seriously. That’s how we got here.
Steven Mosher | May 18, 2012 at 3:19 pm |
I have read some of the article. The RC analogy and therefore formulae are inappropriate in this context. I would suggest for the oceanic heat-temperature consideration application of hysteresis equations would be more appropriate.
His solar cycle fitting is more than poor.
Here is my example of a good comparison
you can see that the AMO rather than sunspot cycle is a synchronizing factor. To pursue the AMO a bit further in time, I did this reconstruction
which IMO is superior to what Mann or Grey achieved. I draw your attention to the highlighted anomaly in the current data for the early decades of 20th century.
Note that the AMO (if correction 1900-1930 is true) gets a bit ‘adventures’ only after 1960 (should cheer up any warm-monger).
The RC model is very appropriate.
When you have something legible and understandable to publish complete with the source data and the source code, maybe you can join the debate.
As it stands you are not joining the debate. This is your choice. You can join the debate and show your work so that others can reproduce it, or you can
sling pixels onto screens. Posting a chart and waving your arms does not count. It doesnt count as science and won’t be consider by people who actually want to debate.
There is a kind of contagious nervousness spreading among ‘warming bunch’, it is reminiscent of state of mind among ruling classes of East Europe before fall of communism. Dinosaurs of the system ended extinct, while the rats who jumped the ship early ended as new big beasts of the jungle, apparently the evolution has seen it before, K2 or something.
As far as RC circuit analogy is concerned it is fine for a one off DC event, very appropriate for temperature behavior at bottom of a mine shaft where day/night or summer/winter don’t exist.
No need to shout.
(woe, but for the knowledge of the bold typeface on a blog)
@DocMartyn | May 18, 2012 at 11:11 pm |
“So CO2 levels in the atmosphere and the ocean, at all depths, are supposed to be in ‘equilibrium’, but no one wonders why the oxygen levels in the atmosphere and the ocean, at all depths, are not at all in ‘equilibrium’.:
On the earth that I live on, everything is trying to equalbrate, but never does. Ergo, we see cycles. What planet do you live on?
I live on a biotic planet; on my planet the whole of the atmosphere, the whole of the ocean and the first 30 meters of soil are sculptured by living processes.
The reason we have 23% oxygen and 0.036% CO2 in the atmosphere is due to water splitting photosynthesis. At the top of the ocean the water is supersaturated with O2, a product of photosynthesis. The organisms fix carbon and so the same layer which is supersaturated with O2 is denuded of CO2. The fluxes of CO2 to the surface layer come from the atmosphere and from the lower layers of water.
Dying organisms and excreta, carbon rich, rain down from the surface. Some of these are intercepted, and oxidized using molecular oxygen, hence the hypoxic zone below about 30 meters. Much of the organic matter makes it to the bottom; this is either oxidized aerobically or anaerobically, or ends up in the mud and becomes mineralized.
The oxygen concentration at the bottom is greater than the middle due to oxygen rich waters coming in horizontally. The cold melt water from the poles sinks and pushes its way to the bottom.
The two major biotic gasses at the surface of the ocean are not in equilibrium with the atmosphere; the oxygen concentration of the water is greater than its equilibrium constant because O2 is being generated by photosynthetic organisms. CO2 is far lower than its equilibrium constant because it is being consumed by photosynthetic organisms.
Reason they switched from every year to ”decadel” is my fault. I kept pointing to the leading Warmist, politicians and editors, academics, bureaucrats, that: Swindlers pretending to know to one thousandth of a degree; if the GLOBAL temp is gone / down from one year to another – when nobody is monitoring on 99,999% of the planet, is admission of credibility, nothing to do with the global temp.
So, they started to talk about ”decadel” as part of their ”science on the run” Which means: ”in 10y, needs less sensitivity than for every year”. Even though: multiplying zero by 10, or by 100, is still zero. But that is irrelevant for them. B] (note that on their charts is still vibrating ”for every year” as if the planet has hi-fever, it’s not DECADEL…?
Jim D pointed to me; how they get it correctly to one thousandth of a degree: as his example: if I give him the temp for the beginning of the hour, and the end – they can get in a thousandth of a degree for every second…? I.e. if is 22C in the beginning of the hour, 24C on the end, approximate temp, easy!
Or is it easy? If the first 5 minutes was 22C, -then gone to 21 for 40 minute – then to 22C for next 6minute – and finishes the hour on 24C. It wouldn’t match the above data, would it? What about on the 99,99% on the planet where the temp is not monitored? Don’t ask!!! DECADEL IS ONLY THEIR KICKING AND SCREAMING ON THE WAY TO THE CONFESSION BOX. Silencing stefan, doesn’t silence the truth – the cat is already out of the bag / documented!!!
So, very simply you should be able to ask Girma “Given your model, what will the temperature be if the sun increases its output by 3 watts? and , if he has a physically realistic model he should be able to answer. Or, Girma, If we have a huge volcano or more S02 in the atmosphere what does your model predict? But He cant answer that because he doesnt have a model that is quantified over known parameters.
The solar energy that reaches the earth does not suddenly change. That is a theoretical question. The practical reality is that the huge reservoir of heat contained in the oceans maintains the earth’s temperature according to the climate pattern of the last century (http://bit.ly/HRvReF), and the effect of volcanoes on the global mean temperature is short (less than 13 years) and therefore negligible.
Again, some idle thoughts of an idle fellow; this time on forecasting. To me the essence of any attempt to explain what has happened in the past (hindcasting), is to then go on and try and estimate what is going to happen in the future, to test if the exercise in hindcasting proves to be correct. As I noted earlier in this discussion, this is one of my criticisms of the paper which our hostess used to in intitiate this thread, such forecasts need to be easily testable on a short time frame.
I remember when I first read RealClimate, Gavin claimed that it was not possible to forecsast for a short time into the future, that was weather, but one could accurately forecast 30 or 50 years into the future, that was climate. I remember thinking what a wonderful piece of propoganda that was, but hopelessly unscientific.
So, could I please put in a plea, that when anyone, like Girma or our hostess, has a way of explaining the past, by whatever means, they include a forecast of what is going to happen in the future. But this forecast should be as short term as feasible, and easily testable. That is, one should select something that is being routinely measured now, estimate what it’s value will be in the short term, put the numbers up for all to see, and then everyone can see how well the forecast is doing. A good example of this is the annual estimates of minimum ice cover in the Arctic.
If this sort of approach is adopted, then it would enable us to sort out which ideas seem to have the most merit.
Jim, whether we will have an El Nino or a La Nina is not known in advance. However, they are oscillations relative to the longterm ocean pattern and their magnitude is +/- 0.2 deg C. As a result, we cannot be more accurate than +/- 0.2 deg C in prediction of GMT without using probability.
For 2012, my graph (http://bit.ly/HRvReF) for the smoothed GMT, which is the neutral ENSO position, gives a value of 0.36 deg C. As a result, with 68% confidence, the GMT for 2012 will lie between 0.26 and 0.46 deg C (+/- one standard deviation).
Also, with 95% confidence, the GMT for 2011 will lie between 0.16 and 0.56 deg C (+/- two standard deviation).
Note that these estimates are for HADCRUT3.
Girma, You write “Jim, whether we will have an El Nino or a La Nina is not known in advance.” etc.
Thanks, Girma, what you have written I completely understand. It is simple, straightforward and easily testable. I am sure the second part has a typo; you mean 2012, not 2011.
Now, if only others like out hostess would do the same sort of thing, we might be able to judge how good the various models, etc. are at predicting what is going to happen to our climate in general, and surface temperatures in particular.
I record the past which perhaps is different to ‘explaining the past.’ Bearing in mind we have seen a long slow thaw lasting 350 years (with several advances and retreats) the odds are on a continuation of the warming over the next decade.
WebHubTelescope | May 19, 2012 at 12:07 am |
“Are you serious?”
Not really. There are a lot of smart people, including yourself, that say otherwise.
The example you give, however, is one of a solid and not of a stirred fluid. Doesn’t seem applicable to me.
“Each decadal prediction consists of simulations over a 10 year period each of which are initial- ized every five years from climate states of 1960/1961 to 2005/2006”
I would be interesting to see the drift comparison between the same models run both with and without the 5 year re-initialisation, as it might give some idea of the skill of longer term projections.
curryja | May 19, 2012 at 10:34 am |
Thank you for your answers, Judith. I fear, however, that I don’t understand your explanations … which could easily be my lack, not yours.
First, the HadCRUT and the GISS results are gridded. So it seems to me that I’m comparing gridded to gridded when I compare HadCRUT to the NCAR/NCEP or the ERA40 analyses … what am I missing?
Next, I’m not clear what you mean when you say that the reanalyses are “directly based on upon the satellite observations of sea surface temperatures”. Both the NCAR and the ERA40 reanalyses start before 1960, and your own analysis starts in 1960 … which satellite observations of sea surface temperatures would the ERA40 folks have been using during the 1960’s?
Finally, you say you “cannot usefully compare point measurements with the coarse resolution model results” … yet the GCM model results are routinely compared to e.g. the HadCRUT or GISS surface temperature results, or the gridded precipitation results, or the like.
How can they do that routinely, yet you say it can’t be done? Again, what am I missing?
Or perhaps I misunderstand what you mean by “point observations” … are you referring to the HadCRUT or GISS analyses as being “point observations”? Or perhaps the HadSST observations as being “point observations”? Because all of those are routinely compared to “coarse resolution model results” from the GCMs …
In any case, I’m afraid that I am more confused than ever. Likely my fault rather than yours, but if you could clear it up I’d be most grateful.
If you are advocating that HADCRUT or GISTEMP should be used instead of ERA40 for evaluating models, I think that would not work because they are designed to look only at temperature change, and I don’t expect that their methods derive an absolute temperature in each grid box, just the change over time, so ERA40 with its actual temperatures is a better comparison.
Jim D, HadCRUT absolute temperatures are here …
So do you believe they represent a true area average? Or are they just a flat average of all the non-uniformly distributed stations in the cell? Would you use this for anything?
Jim D, per this, they are the average of the stations within each gridcell.
I also note that, rather than admitting that your claim that HADCRUT does not “derive an average temperature in each grid box” was 100% wrong, you now want to change the goalposts to the quality of their average …
Look, the surface temperatures have problems, a host of them, I’ve discussed a host of them in a number of posts.
But that doesn’t mean that they should be thrown out, and climate model results be substituted in their stead. If the surface records are bad, then the reanalysis based on those records can only be worse, not better.
I said ‘I don’t expect’ they provide an average because if they did it was not likely to be very useful. Do GISTEMP do this too? If it is a 5 degree square it could contain quite a mix of stations (urban, coastal, valleys, etc.) or a few non-representative ones, while ERA40 would give something like 1 degree squares with an attempt at representativeness while fitting existing observations to the extent that they are representative. This is what Judith was getting at when saying ERA40 is better than stations for comparing with model data that also represent grid averages. Makes sense, doesn’t it?
Well, it’s not true that a reanalysis will be worse.
In CRU take a cell that is half ocean and half land.
The final value will be a simple average of the SST and the Land. If there is one station in the land then the whole 500km by 500km will get a single
value. Call it 15 C for example. If the half that is SST is 20C the whole cell will get a value of 17.5. If the next inland cell has a value of 20C
The delta between the two cells will be 2.5. It will be a step change.
We know this is not real. In fact we know it is wrong. Physically Wrong.
If we want to estimate the value on the coastline with a 5*5 grid well that answer is 17.5. we know this is wrong. We know it is wrong.
If we run a reanalysis code that is constrained by the surface temperature reports and the laws of physics we know we will get better answers.
they wont be observations, but they will be closer to the truth than a
5*5 grid that proposes unphysical changes every 500km
I also suspect the reanalysis is less prone to growing urban heat island effects that may affect some stations, which makes it better for comparison with climate models that likely also don’t include this effect.
“Thank you for your answers, Judith. I fear, however, that I don’t understand your explanations … which could easily be my lack, not yours.
First, the HadCRUT and the GISS results are gridded. So it seems to me that I’m comparing gridded to gridded when I compare HadCRUT to the NCAR/NCEP or the ERA40 analyses … what am I missing?
Next, I’m not clear what you mean when you say that the reanalyses are “directly based on upon the satellite observations of sea surface temperatures”. Both the NCAR and the ERA40 reanalyses start before 1960, and your own analysis starts in 1960 … which satellite observations of sea surface temperatures would the ERA40 folks have been using during the 1960′s?
1. The gridded product of NCAR and ERA40 is at a higher resolution
and it is spatially complete.
2. The variance of CRU grids and GISS grids varies greatly depending upon how many stations are in the 5 degree grid. You also have to correct for portions of the earth not sampled by CRU/GISS gridding.
3. Read the ERA40 documentation. pay special attention to the appendices.
4. I will expect you to object to every use of UHA and DMI, since that is not “data” but rather output from a model.
Bart R | May 19, 2012 at 6:33 pm
Thanks, Bart, and I understand that. However, the difference is that the computer can only output what the programmer tells it to output. It is ruled entirely by the programmer’s beliefs and understandings.
A thermometer, on the other hand, cares nothing about the beliefs or understandings of the person who is operating it.
Yes, I understand that a mercury-in-glass thermometer doesn’t measure the temperature. It measures the expansion of a liquid, which we know from experimentation to vary linearly with temperature. Once it is built and tested, it will work until the laws of physics are repealed.
A computer, on the other hand, is under no such physical constraints. It can make an error as easily as it can make a correct calculation.
To illustrate the difference, consider a mercury-in-glass thermometer. When’s the last time you heard of a thermometer like that having a “bug”, getting “hacked”, being subject to a programmers foolish error, being spoofed by a “Trojan horse”, having a “Y2K problem”, accidentally using Fahrenheit instead of Celsius, or being affected by GIGO? Yet all of those happen with depressing frequency to computer programs … and when any of those things happens, the computer output is wrong.
Your belief, that there is some kind of equivalence or equality between the reading of a mercury thermometer, and the output of a complex computer program containing a million lines of code, is simply not true.
I can trust a thermometer. When I take my temperature to see if I have a fever, I don’t have to examine the code to see if the programmer made a foolish mistake. I don’t have to verify that the thermometer hasn’t been unintentionally corrupted by the accidental transposition of a couple of numbers in a line of code. I just put it in my mouth and take my temperature.
If you trust a computer in that same way, however, you don’t understand computers.
PS—Suppose you build a computer program that you think can accurately report the temperature. Do you test it against another computer … or against a thermometer?
That is my point here. Judith is comparing one computer to another computer, there’s no thermometer anywhere in the comparison.
Nicely put, Willis. Another point worth considering is that thermometers don’t make either predictions or hindcasts. Given that even measuring multiple temperatures using thermometers in the here and now is fraught with problems, despite the simplicity, universality and accuracy of the instrument, calling anything that comes out of a computer model ‘data’ is a pretty big leap.
Willis Eschenbach | May 19, 2012 at 11:12 pm |
A thermometer is made by human hands. I’ve seen countless of them fail utterly in no way repealing the laws of Physics. Ask any nurse with enough experience about the reliability of medical thermometers. As any salesman about his competitors’ products in the field of medical instrumentation.
No one who’s done sufficient lab work could credit what you say as anything but bumpf, either. Analog devices are not sacred, untamperable (read any book about the history of tricks servicemen tried in field hospitals to malinger), or in any way special. You should not trust them any more than you trust a computer. And I say this with the greatest disparagement toward computers.
As it happens, I’ve worked in the field of computers for a while. I’ve worked on computer systems so much larger and more complex than what the models come out of that by comparison, we might as well be discussing the browser on your desktop PC as the climate models. It’s very likely you’re using lines of code I’ve had a hand in every day, without knowing it.
Do you use a telephone anywhere in America? Drive a car? Invest in American industry? How’s your trust in those systems that you don’t even see in action?
Sure, there’s such things as errors, hacking, GIGO; there’s also bounds checking, UAT, performance testing, specification testing, validation and verification cycles, user feedback, system security, monitoring of logs, and a hundred other functions analog thermometers lack.
Your faith is misplaced, and your point specious.
PS – If I build a computer program, I get teams of qualified SME’s to confirm or test it in each phase as one stage, but I certainly don’t start or stop there.
You are not getting this stuff. Maybe I can help you by relating to something you know about:
A thermometer is a device, a computer is a device, and a hammer is a device. When you hit a nail (or your thumb) with a hammer, you are producing data. Are we straight now?
Wasn’t Captain Kirk monitoring SSTs in the ’60s?
Don, I don’t care what you call the output. My point is that the output of a computer program is not the same thing as the reading of a thermometer. Sure, you can use a computer to calculate a temperature … but when you go to check your computer program, you check it, not with another computer program, but with a thermometer.
Willis E is correct. Computers ALWAYS give output predetermined by human beings. Thermometers do not.
I agree with you, Willis. I was just being facetious to kill time, while we awaited Judith’s reply to your request for clarification of her brief cryptic remarks on the post she obviously wanted us to read and discuss. But it appears that she has moved on to Copenhagen. Whatever. Her wishy-washiness is becoming tiresome. Post after post with hundreds of predictable comments, most of them by the half-dozen usual suspects, and Judith rarely wades into the fray. I am beginning to wonder what her point is here.
My bad, I missed the sarcasm.
As to Judith, she posts when she does, and the rest of the time it’s a great forum for discussions. I do hope, however, that at some point she clarifies my questions that I posted above.
“Thanks, Bart, and I understand that. However, the difference is that the computer can only output what the programmer tells it to output. It is ruled entirely by the programmer’s beliefs and understandings.”
Wrong. think harder about this Willis and you’ll see your mistake
Steven Mosher | May 21, 2012 at 1:20 am
Oh, stop with the coy bullshit and clearly state what’s on your mind. Your Socratic nonsense is wasted on me, I’m not interested in your ‘catch me if you can’ games and questions. You may well have a valid point, but I’m not going to try to guess what it is, so I invite you to either make it or go away. Your cutesy drive-by postings with the implicit claim that you are so much smarter than everyone else are losing you traction and respect.
PS—Upon reflection, I’ve never yet seen a computer that did something I didn’t program it to do. I often have not realized that I was programming it to do so, but I’ve never found one that acted on its own with no instructions from me. So now I’ve thought harder about it and I don’t see my mistake … you see why your method sucks?
So what is your point?
I will have to help you again, Willis. Your mistake is that you trust the computer programs and systems that have been developed to run elevators and the air traffic control system, but you don’t trust the amateur programmer-statistician climate scientists to produce models that actually are acceptable and useful representations of climate.
Bart R | May 19, 2012 at 11:42 pm
Fine, Bart … when’t the last time you didn’t trust the thermometer that you used to take your own temperature?
I fear that given your point of view, that you can trust a computer’s output in the same way that you can trust a thermometer reading, that the gap between us is simply too large to bridge.
I note that you didn’t answer my question about whether you test a computer with a thermometer or with another computer. Instead, you retreated into bafflegab. That’s fine, but I’m not interested in folks who won’t give simple answers to simple questions.
So I’ll let you go your way, and let the reader decide whether a computer program can be trusted in the same way that we all trust a thermometer … I know I don’t trust a computer program that way, particularly a program with hundreds of thousands of lines of code, but then I’m a practical man and a programmer with fifty years experience in the game … but YMMV.
Computers outputs have an involvement in every single aspect of modern life to the point were every second of your waking hour has some interaction with a computer output. I hope you don’t spend your time distrusting all those computers, that sounds like a stressful life. I’m guessing you mainly distrust climate computers? Why they so different?
I am with HR on this one. We rely on computers to run our daily lives. Where would we be without them? It would be chaos, like the 1940s. Therefore, we must trust that the genius climate scientists and their super-computers have got it right. Decarbonize the world economy now! Then the climate scientist geniuses can turn their efforts to financial and econometric modeling. The people handling that now apparently don’t have a clue. We should probably put the genius climate scientists and their computers in charge of everything. It would be a model world.
HR | May 20, 2012 at 12:26 am | Reply
HR, I wrote my first computer program in 1962, which means that I have been programming computers for half a century at this point. Please don’t try to school me on computers. Yes, I know that computers are used to run everything from submarines to elevators.
I also know that computers used in mission-critical situations are subject to extensive verification and validation procedures that compare them, not to the output of another computer, but to reality.
Unfortunately, climate models are not subjected to the same V&V procedures, nor are they tested against reality. Take a look at the output of the GCMs with regard to precipitation … they do no better than chance. Perhaps you are willing to trust such a computer program in the same way that you trust a thermometer.
I’m not. Why are climate models so different? Partially because they are iterative models rather than physics based models which directly calculate results. Partially because they have a number of tunable parameters. Partially because they are tuned to match the physical world. Partially because they have not been subjected to the normal V&V that mission-critical computer models must undergo. Partly because they perform so poorly with respect to actual observations of the real world.
In other words, not all computer models are created equal. Blanket trust in computer models is as foolish as blanket trust in human beings …
Gosh, you mean that Bart has jumped sideways and changed the topic when successfully challenged?
I may have to lie down in a dark room with a cool cloth on my forehead.
Of course, I won’t be sure if the room is dark or the cloth is cool without using a computer model to verify these things. Human agency and all that.
johanna | May 20, 2012 at 12:30 am |
Before you go off for your lie-down, could you explain what challenge, what success, what topic and what jump?
Because all my jumping sideways, I seem to have missed it.
Willis Eschenbach | May 19, 2012 at 11:59 pm |
What you call bafflegab is the parlance of the people who make computers trustworthy. Thermometers? The average analog thermometer has less than a tenth the reliability of a well-engineered computer system.
But see, I’m asking you _not_ to trust me. I’m replying that you should discuss thermometers with SME’s (that’s subject matter experts), of your own choosing. Check with them, and if they’re qualified and experienced, they’ll tell you the error rates and precision, failings and limits of the devices you hold in such superstitious regard, like a coke bottle fallen from the sky.
And I don’t care if you’re interested in folksy chat about computers. You take on the topic of computer models and their reliability, you better not turtle up into “but only in folksy words”, or you’re going to look ridiculous. More.
So, if you have 50 years experience in the game, and don’t know the term SME, then it’s likely your game has been stale for about .. ever.
In my world, SME means ‘small to medium enterprise’. And, anyone who describes themselves as a ‘subject matter expert’ is well into pompous ass territory.
The computer systems we use to make phone calls, do banking, post on this blog etc get tested to destruction by millions of users every day. There is no comparison with climate models at all. It is like saying – this 500 year old cathedral is built of bricks, and this retaining wall with sand instead of mortar is built with bricks – so why don’t you trust my wall?
johanna | May 20, 2012 at 5:32 am |
Funny. We call people pompous if they’re too full of themselves to have their systems tested by people who work intimately with the subject matter.
I don’t particular speak to the engineering of the climate models; I don’t know the detailed engineering of the climate models — though not for want of trying; information often is inadequate or closely held or not public so far as I can tell for the most part, which could mean it is either just not there or the work hasn’t been done to properly validate and verify. Even the few studies published about the subject seem to rely on error/fix logs, which seems.. of low reliability.
I’m just pointing to the false dichotomy of ‘computer’ data vs data.
The climate models could go far to improve their value by release of more information about their testing and validation.
SME means subject matter expert, at least since 1985 when I learned the term and gained the title
You ‘gained the title’? Gosh, who awarded it and where?
I think you have just proved my point.
simple johanna. you gained the title by proving through your actions that your judgement in an area was superior to others. For example, if you were designing an airplane and you wanted to know what displays a pilot should have, you had to consult the SME: pilots.
Its pretty simple. If you want to know how to speak Korean you consult a subject matter expert. They will know how to speak korean.
So’ll I ask you: how does someone as dumb as you survive?
Assume you had 2 identical earths. The 2nd earth an exact model of the present day earth, in all respects, down to the sub-atomic level.
Now play these two earths forward in time. Would they both have the same climate in 10 years time, 100 years time?
Physics tells us the answer is no. The two identical earth’s would rapidly diverge and have different futures, even though they started out identical.
My question is this. If an identical earth cannot model our current earth, how can a much coarser and less perfect computer model provide a better prediction?
simple. your two earth model should show you the difference.
The point is that no matter how much the two earths diverge they
remain constrained within physical boundary conditions.
Both earths have to obey certain laws: energy out = energy in.
Climate modelling is a boundary condition problem. While the temporal evolution of the models may be different here and there both will be constrained by the fundamental laws. Given the same input from the sun
neither will exceed the temperature of the sun. Neither will fall below absolute zero. Both will differ, the question is how much?
I agree with Mosh.
If they were the same distance from the sun and remained at the same orbit, which is accurately known to be true, then after 100 years timespan, the average temperature of the two bodies would be nearly the same. That is energy balance for you, and wing-flapping butterfly chaos has nothing to do with the outcome on that time and energy scale and those initial and boundary conditions.
Re: “energy out = energy in”
Technically, need to add changes in earth’s energy. Chaotic variations in albedo, especially clouds, and in chaotic ocean circulation will likely result in significant differences between the two models. As you and WebHubTelescope note, they will tend towards the same average temperature.
Note however that Fred Singer shows substantial differences between model runs due to starting conditions and choatic variations. The few model runs from reported IPCC models runs were insufficient to provide average results. See: Overcoming Chaotic Behavior of Climate Models S. Fred Singer
NIPCC vs. IPCC Addressing the Disparity between Climate Models and Observations: Testing the Hypothesis of Anthropogenic Global Warming (AGW) By S. Fred Singer, to be presented at Majorana Conference, Erice, Sicily, August 2011
Willis Eschenbach | May 19, 2012 at 11:12 pm | Reply
That is my point here. Judith is comparing one computer to another computer, there’s no thermometer anywhere in the comparison
The first space shuttle launch stopped at something like T-15 seconds because of something that was relatively unknown to computer scientists at the time. Two identical computers, even though they are digital, do not give identical results.
Elliott’s Law: Every computer program with more than 1 line of code has at least 1 undiscovered bug.
Bart R | May 19, 2012 at 11:42 pm | Reply
Bart R., of course thermometers fail. However, I doubt greatly that you have seen “countless” mercury-in-glass thermometers fail in normal use. In addition, as the EPA notes, “failure usually is visually apparent” … which is the opposite of the condition with computer programs, where failure is never visually apparent.
In addition, they are more reliable. Here’s a typical analysis (emphasis mine):
Please note that they do NOT compare electronic thermometers to other electronic thermometers … they compare them to glass/mercury thermometers. There’s a reason for that, Bart, no matter how much you might try to talk around it.
Willis Eschenbach | May 21, 2012 at 1:04 am |
Superstitious and arbitrary. You speciously compare analog nonelectronic device to digital electronic device that in no means involve computers.
Is it electrons that frighten you, or binary? Because that’s not reasoning, that’s a pathology you’ve got.
I have to wonder, what is it about the PDO that makes it so much less susceptible to model skill than virtually every other component of the ensemble?
Is it the influence of the Arctic? The size? The poor data? Something more fundamental?
Steven Mosher | May 21, 2012 at 1:41 am |
All of the models use gridcells. Both the ERA40 reconstructions and the model results are interpolated to the same grid size:
So whatever is accurate in in your analysis applies to the models just as it applies to the HadCRUT or other surface analysis. As a result, it’s not at all clear to me what your point might be.
In addition, the globally averaged annual-mean surface temperature anomaly from the reanalysis is shown as hitting its highest temperature in 2005 … how does that relate to your claims?
read it again and see if you can comprehend. the difference is clear
Steven Mosher | May 25, 2012 at 12:59 pm
OK, steven, I read it again. The difference is not clear. Stop with the drive-by postings, they’re nothing but a pain and do not enhance your reputation, and explain yourself.
For example, you say:
No it won’t, that’s not how the averaging works at all. There is a good description of the actual procedure used in the HadCRUT dataset, as opposed to your simplistic fantasy, located here on page 13:
As you can see, it’s not “simple averaging”.
You are still missing the point.
Even the enhanced blending is wrong. RomanM and I discussed this at length on CA some time ago. A physics driven model like ERA40 will have a much reduced error over a simple or even a complex blending algorithm. Both the complex blending of CRU and the physics based blending of ERA40 ARE MODELS… That is the point.
There is no RAW DATA.
There are no observations at the cell level.
There are modelled results from the GCMS
There are TWO MODELS of gridded temps
A) the cru MODEL which models a 5 degree grid by averaging
averages, to quote william briggs, are NOT OBSERVATIONS
B) the ERA40 MODEL which models gridded temps using
1. point observations
So, your notion that comparing GCM grids with Cru grids is somehow comparing model results with observations is confused at the depths of its soul. Its confused because the CRU priduct is not observations it is a MODEL of observations. all averages, all means, all anomalies are data MODELS. they are not observations.
Thanks for the explanation, Steven. I’ll need to think about this more deeply, but I finally understand what you are driving at, and at first look it seems reasonable.
I thought that indices such as ENSO had no physical meaning, being merely a collection of measurements reflecting a high degree of subjective choice. One could choose an infinity of different components to create an index. But perhaps I’ve misunderstood how they are defined. Perhaps someone can help.
I would be confused, and not a little concerned, about the provenance of a model that claims predictive skill of indices based on arbitrary human choices.
How many times do I have to post this comment here and elsewhere before you guys finally wake up and pay attention to this seminal monograph.
RE: Climate Change, What the Russians Say.
The English translation of “Cyclic Climate Changes and Fish Productivity by L.B. Klyashtorin and A.A. Lyubushin can be downloaded for free thru this link:
NB: This mongraph is 224 pages and is not about climate science. The Russian edition was published in 2005. The English translation was published in 2007 and was edited by Gary Sharp, Center for Climate/Ocean Resources Study.
By analyzing numerous time series of empircal data (e.g., temperature records, sediment cores, fish catches, etc), they found that the earth has several global climate cycles with periodicities of 50-70 years and that the average of these cycles is about 60 years which has a cool and warm phase of 30 years each.
The last warm phase began in ca 1970-75 and ended in ca 2000. The global warming from ca 1975 was due in part to this warm phase. A cool phase started in 2000 and their stocastic model predicts that it will last until 2030.
Several others studies have found this 60 year cycle. See ,for example, Alan Cheetam’s “Global Warming Science” at:
NB: You should bookmark this page. This is _the_ one-stop-shop-until drop store for global warming and climate change info.
The monograph was preceded by: “Climate change and long-term fluctions of commercial catches: the possibilities of forecasting.” by L.B Klyashtorin, FOA Fisheries Technical Paper. No. 410, Rome, FOA, 2001, 86p.
Note the date of publication. Was this report and the mongraph forwarded to the IPCC and cited in AR4. I don’t know but I’m going to find out.
Also check out:
“On the Coherence between Dymanics of World Fuel Consumption and Global Temperature Anomaly” by L.B. Klyashtorin & A.A. Lyubushin, Energy & Enviroment, Volume 14 No.6 2003.
Briefly, they found no correlation between rising world fuel consumption and the global temperature anomaly over the interval 1961-2000.
The Russian have demolished the IPCC years ago. When their works become more well-known, the People will storm the UN and univerities, capture the bureaucrats (i.e, con men) at the IPPC and climate scientists (i.e, the white coated wiseguys) at the universities, douse’em with sweet Diesel and burn them all at the stake!.
When the People learn the true objecctives of the UN as set forth in the B.C. Climate Action Plan, the New Communist Manifesto, they shall go after the politicians and slaughter them all.
It looks like Judith does not want to participate in a discussion of her paper, Willis.
Thanks, Don. Judith, like all of us, does what she does subject to the constant pressures of time and obligations.
In addition, I’m sure that she is aware that silence is as much of an answer as a discussion of the issues.
And a lot of non-silence is neither discussion nor answers.
“JC comment: this post is cut short by my hopping on a plane to head back to the US. This is a technical thread, comments will be moderated for relevance. I will be back online tomorrow and will provide further comments and participate in the discussion.”
I believe CMP5 has a fundamental problem. That modern climate started in 1960-61. This is a common error. The starting date should be about 1905. Between 1905 and 1940 global temperature rose about 0.5C due to CO2. There was then a transport delay of about 30 years during which global atmospheric temperature actually fell despite higher CO2, but 100% absorption. After 1970 the earlier temperature rise had propagated through the oceans, producing the present atmospheric temperature. See my ‘An alternative theory of climate change’ at http://members.iinet.net.au/~alexandergbiggs.Obviously ignoring a large permanent chunk of heat pre-1940 would make prediction difficult.
Steven Mosher | May 25, 2012 at 1:05 pm
I keep reading this, but what I have never been able to get from any proponent of the idea is a complete list of what you call the “boundary conditions”. I mean it’s obvious that energy in must equal energy out … but as the existence of the “greenhouse effect” clearly demonstrates, the surface temperature is obviously not constrained by TOA energy balance.
So your claim desperately needs backup in the form of a list of all of the “boundary conditions” you are referring to, and an estimate of the values of those conditions over the 21st century. Without that, the “boundary conditions” claim is just a modeler’s security blanket.
I’ll wait for your answer … unfortunately, if your claim is like the other times I’ve asked about this, the boundary condition on how long I’ll wait for an answer may involve the temperature in the place of eternal perdition …
the surface temperature is constrained by the TOA energy balance.
Its really pretty simple. If you add opaque gases to the atmosphere
the ERL increases. The height at which the earth radiates increases.
That is, the ERL is that height at which the concentration of GHGs is low enough to allow radiation to escape to space. You add GHGs the ERL increases. It increases because the concentration of GHGs above the ERL is a constant. Its that concentration which is low enough to allow the escape to space. When the ERL moves up the earth is radiating from a higher colder place. hence it radiates at a slower rate. This means that the surface will cool less rapidly.
There are two unknowns. What is the time period it takes the system to achieve balance from the pertubation and what are the exact rearrangements the system undergoes to achieve balance. This much we know. If you add 3.7 addition watts at the top, be it from solar forcing or GHG forcing, the system will respond over time by a reduction of cooling at the surface. Will the system rearrange it self quickly? slowly? in fits and starts? will positive feedbacks enhance the 3.7watts? how? over what time scales. All of those questions cannot be answered exactly. The best we have is estimates. Estimates from past reconstructions of how the system responded and estimates from the best physics we have today. A physics that is incomplete and uncertain. But we do know that if the sun were to increase by 3.7watts that the surface temperature is morel likely than not to respond by increasing in temperature. Over time, in the long run as the system responds, the trend will be up it will not be down.
because the concentration of GHGs above the ERL is a constant.
Is that true? Perhaps the total mass of GHG’s above the ERL is a constant?
yes its true. its true by definition. the ERL is that altitude at which the concentration of Co2 is such that radiation can escape.
Here is the simplest version I could find for you
That is just basic physics. You raise the ERL. The earth radiates from a cooler higher place. It must radiate at a slower rate. fundamental physics.
A slower rate of cooling at the top means a slower rate of cooling ( “warming”) at the surface.
There is no getting around this. The issue is feedbacks. period.
Get the book referenced at the end of the slide deck
I think you should double check the references you supplied.
By your own use of the definition, CO2 concentration varies by elevation, otherwise how could a change in ERL make any difference?
As concentration of absorbing gases such as CO2 are increased, the ERL rises, decreasing the total mass of the air and keeping the opacity of that air constant.
ERL = effective radiating level, such that total CO2 above is fixed
That spells mass to me.
What units of concentration were you using? ppmv, ppm, or tons/km2?
CO2 will thoroughly mix as it is similar in molecular weight to N2 and O2. Thereafter it will follow the first-order barometric density profile of an exponential decrease of pressure (and therefore density) with increasing altitude. Both the complete mixing and exponential decrease are maximum entropy effects. Second-order differences will be due to the convective effects.
Which doesn’t address the issue.
How Is ERL defined?
(1)a key concentration level of GHGs at the ERL
(2)or a total mass (or mass per unit area) of CO2 above the ERL
(3) or is (1) and (2) equivalent and if so, how. Seems like too many degrees of freedom.
Steven Mosher | May 25, 2012 at 3:53 pm | Reply
Steven, I think what you are trying to say is that if you add opaque glass AND EVERYTHING ELSE REMAINS THE SAME then the ERL (the “effective radiation level” for those in the bleachers) increases … but since nothing ever remains the same in the climate system, your statement is a huge oversimplification.
For example, I have shown that as temperatures in the tropics increase, the clouds form earlier and thus negate and can even reverse that temperature rise. Sure, if there were no cloud thermostat, your exposition might be physically based …
But since there are clouds and a host of other natural thermostatic mechanisms, I fear your explanation is woefully incomplete. It’s like saying that if the temperature drops outside my house, simple energy calculations can prove that the temperature inside my house has to drop as well. Basic physics, no? But the existence of thermostatic mechanisms, in my house just as in the climate, guarantees that overly simplistic explanations such as yours simply don’t work.
I encourage people to read my post The Details are in the Devil for a more in-depth discussion of these issues involving emergent homeostatic phenomena.
The erl is determined by the concentration of gases.
Why you always fighting with Willis?
Willis may be wrong on the ERL. Is it correct that even if all else is not equal, that increased GHG will result in the ERL moving higher? That the ERL is solely dependent on the concentration of GHGs? If the answer is yes, can you cite the evidence that the ERL has moved measurably higher, since let’s say 1950?