Here’s something amusing in the New Scientist article, which includes a defence of the Stick. Their jpg is compiled by one Robert Wilson of the University of Edinburgh, who includes the reconstruction from Juckes 2006. Has Juckes been accepted or doesn’t this matter any more?
Like IPCC, Rob has failed to show 1960-1994 values of the Briffa 2001 reconstruction. Rob, if you’d like to issue a correction to this graphic, it would illustrate the Divergence Problem very nicely.
Also if Rob has used a common version of Crowley, as I discussed at CA some time ago here , Crowley spliced the instrumental record in the latter portion of his “reconstruction” so the yellow line shown by Wilson is not a “reconstruction” in its latter part but just a rendering of the temperature record giving a misleading impression of the success of this particular proxy set.
New Scientist Spaghetti Graph
Climate Audit 
244 Comments
Nice graph, lots of pretty coloured lines. My question is this:
Why, if all of the reconstructions fail to match the instrument readings, (they seem to be about 1 degree C too low) do we not see ajustments to the proxies? Wouldn’t it be fair to asume that the proxies are inadequate?
It seems that the constraint employed by the Direct measurements filter preserves the late 20th century trend ( sorry, insider joke 🙂
Anyone else notice that the graphs are arranged so that the MBH construction is invariably on top of any other proxy line it crosses — given pride of place — and that the MBH line is about twice as heavy as any of the rest? I’d call that a bit of subliminal propaganda.
Was it Rob who did that, or someone at New Scientist? Whoever did it is clearly wedded to the MBH construction, no matter that it is thoroughly proven to be entirely BS.
I read the New Scientist article. Can someone explain this comment?
Hasn’t the temperature rise flattened out in the last 6 or 8 years?
Question: Why is it that all the reconstructions (at least the data that is included) are in fairly good agreement from 1850 moving forward, but prior to that they’re not?
On further thought, I suppose they all calibrated their data to the instrumental record for that period, yes?
Re#3, it also calls-out Mann’s as “the hockey stick graph,” so I presume that’s why it was given this emphasis. Even a few lay-people are familiar with “the hockey stick.”
If you cherrypick from red noise series using 20th century trends as a criterion, you can produce stuff like this.
In my opinion, the low covariance in early periods relative to calibration period covariance is evidence of cherrypicking and is probably even a statisitcal test for it. The Team views the breakdown of covariance as evidence of “regionalization” of past climate not present in the modern period.
6: PK
If they calibrated their data to the instruments for that period, should’nt they do the same calibration backwards through the data?
Forgive me if I make uneducated statements or ask stupid questions as I am just a layperson trying to get their hands around all this. I’ve been following this website for the past few months trying to get a balance from what I’ve read elsewhere. It seems that if a person like me wants to learn what’s really going on they simply can’t depend on the media and have to dig a little deeper into some things that are difficult for the layperson to understand. I’m an engineer, but I’ve been away from the technical side of things for quite awhile so digging up what’s in the memory banks is sometimes a bit slow.
#8. If there’s that much difference in the regional climates (as would be seen in the low covariance of the earlier periods), then wouldn’t it make more sense to calibrate to the regional records and determine a past global temperature on a geographically weighted geometric or arithmetic mean of several reconstructions?
#9. I’m not sure I understand your question, but how could one do that without instrumental records going back that far?
I guess I must be dumb: I can see charting a single proxy against the local temperature record to demonstrate its reliability but what purposes does it actually serve to construct a spaghetti chart. It leads to the mistaken impression that if all proxies show the same pattern then the pattern must be correct independent of how well the proxies map to the instrumental record for the regions from which they are drawn. It seems that there is a huge amount of variance among the proxies though collectively they do suggest a MAWP and LIA. Once we have agreed that past temperatures are or are not on a par with today’s temperature doesn’t the focus shift to the adequacy of the construction of the global anomaly trend itself? I keep looking for the predicted strong GG driven rise in temperatures in the Arctic where thee trend is supposed to be the most pronounced but don’t see them in any of the individual weather stations (with the exceptions of those where a UHI cannot be ruled out). If the “Direct Measurement” is wrong, what value have the proxies?
To date these charts serve only to reinforce the basic notion that the proxies have limited value as indicators of temperatures given the level of precision demanded by the current debate over AGW.
10: PK
I too am a simple layman in this venue. My question made a lot of sense to me, until you questioned it that is. I guess what I’m trying to wrap my head around is this, if corrections were needed to fit the proxy data into the instrument values, then it begs asking how valid the proxy data is at postdicting past temperatures. Yet the authors of those charts are confident that their reconstructions are very valid.
I’m not sure that made sense either.
In econometrics, the Divergence Problem would invalidate the hypothesis that the “proxies” were proxies for temperature. If an article in the American Economic Review truncated the last 34 years of a record a la Mann, Briffa and IPCC, it would be a scandal of epic proportions.
The NAS Panel discussed the divergence problem – Cuffey clearly asked : if the proxies don’t record present warming, how can we be sure that they did to in a possible MWP? However they drew back from this particular abyss.
Thanks.
Steve:
How about posting a close up of the end of the Wilson graph to highlight the truncation of Briffa?
If we assume that TRs are local temperature proxies, how can a dendrochronologist who is carefully monitoring local temperatures, mostly away from UHI, feel comfortable looking at the global temperature reconstructions when their regional temperature signals diverge from NH and Global signals? Can’t most of this confusion and obfuscation be traced back to the instrumental temperature reconstructions by Jones and Hansen?
Again, apart from the PR value of the HS and the debatable research practices of some paleoclimate types, why should we pay special attention to these TR proxies as part of the AGW debate? I think the efforts to re-examine the station data is where we get to “size” the issue or scope the problem.
My boss pointed out a rather sarcastic statistical saying recently: if you haven’t found a correlation, you’re not looking hard enough.
Mark
#13 — Even if the proxies duplicated the modern temperature record, there’d be no valid scientific reason to suppose that they are a record of past temperatures, because there is no valid falsifiable scientific theory that can extract a metrical temperature from a tree ring width or density.
There is such a theory for ice cores, for example, involving thermal oxygen isotope fractionation. But even this method is fraught with climatological imponderables such as regional rain out. Even so, isotopic fractionation is theoretically sound in that it’s based in physics, no matter that the regional loop-de-loops of climate make it less than reliable as an accurate temperature probe (as opposed to climate probe) of the distant past.
Tree rings temperatures have no such physical theory. They rest on hand-waving qualitative arguments about temperature-limited growth (and the abuse of principle components). Normalizing them against the modern instrumental record (itself a highly suspect metric) is no more than a specious empiricism. By “specious,” I mean that it looks like a proper empirical justification of conclusions, but in fact is not. It’s a pseudo-justification.
What is so odd about this article is that the authors have to know the fallacies in their methods by now. Do they really believe that they can save this hokey stick by continually republishing the same garbage? Sort of like continually saying “the science is settled” so often that everyone will be forced to agree?
It shows a huge lack of integrity on New Scientist’s part. We would normally expect publications like Nature, Science and New Scientist to push hard for proven science and true replication, but that just does not seem to be happening with the global warming field.
Perhaps there will be a few Mea Culpa’s in the years ahead as we are now seeing with some of the biological sciences.
Mark #17
After working all day playing the role of a code monkey (which I’m not), I read that comment and laughed my ass off for 20 minutes. It really ought to be a tag line for CA.
#16 bernie: the paleo work is important because the use of the word “unprecedented” (trend or level) hinges on it. That’s why they need to leave the error bars off to make their case. Correct error bars would include the possibility of a very much warmer MWP.
No matter what I hear about the HS showing the unprecedented Modern Warm Period and in that one aspect, all other proxies agreeing, I continue to see the Mann’s original attempt as the idealized and unique reconstruction for evidencing the AGW. The HS is most unique in that it shows little variation in temperatures until one can safely connect temperatures to the A in AGW and does that with the best of conditions by showing a slight downward trend leading up to the steep instrumental increase.
If I were actively promoting/marketing AGW, the HS would no doubt be given top billing. I think these not so subtle differences are sometimes lost on the skeptics and the advocates. Some of those other reconstructions show rather large variations prior to the Modern WP that would indicate that nature can provide some rather large shifts in temperatures and particularly if one sees beyond the handy “unprecedented” terminology and looks merely at the variations.
bender (#22)
I get that issue but I guess I am more surprised at the apparent absence of empirical support for the supposedly clear trend for increasing temperature in station data that conforms to AGW theory.
As the 2001 IPCC TAR pointed-out explicitly (not sure if the AR4 did so, haven’t looked hard enough), these reconstructions should show significant warming prior to 1850 in order to correspond with widespread glacial retreat. However, most reconstructions, including the surface record, are several decades to a century (or more) behind. The TAR refers to this issue as “remains unresolved.” Gavin responded to me once with something along the lines of, “I don’t know what they mentioned it, it’s overblown.” Yet it seems important that there’s a physical record (glaciers) in direct conflict.
Of the above, Moberg’s seems to the one which may show warming in a time appropriate with the onset of glacial retreat…interestingly enough, his does show some extreme warmth in the area of the MWP, the coolest valleys of the LIA, and the lowest 20th century temp at the time of the end of the reconstruction.
I’d like to be able to reproduce New Scientist’s “Direct Measurements” plot, but I’m struggling.
Both NCDC and HadCRUT3 show about a 0.5C or 0.55C at most rise from 1975 to 2000. New Scientist shows 0.8C.
I wonder what “Direct Measurement” New Scientist uses.
Uh, not really. NS is not in the same league as Nature or Science. NS is more like Discover meet Enquirer. Integrity is the last thing I’d expect out of them.
Mark
why doesn’t the rutherford et al. (2005) recon make it into these ensemble plots? the team made such a big deal about it when it was first published. remember how much it was used to claim that m&m was discredited (http://www.realclimate.org/index.php?p=10)? i realize that it showed basically the same history as the mann et al. (1998) series, but it was supposed to be the new and improved version…
sorry, i forgot to make the last link active: Rutherford et al 2005 Highlights
Re #24
bernie, that’s a bit simplistic a characterization of the theory. It’s basically YOUR supposition, not the theory’s. If AGW serves to strengthen heat flows thorugh major heat disspiation pathways (PDO, NAO, AO, etc.) then you exxpect some cycling around the trend, and therefore some difficulty in distinguishing what is the GHG trend vs. what is the natural cycle. This is perhaps why the upward trend leading up to 1998 has since flattened out. Wait until PDO & NAO flow warm again and you’ll likely start to see some more land surface records broken.
Figure SPM-4 on page 18 of last IPCC summary for policymaking:
Click to access 02_02_07_climatereport.pdf
presents regional and global temperature instrumental trends of 20 century, with overlay of blue shaded bands representing 5-95% range for 19 simulations from 5 climate models using only natural forcing due to solar activity and volcano. Overlayed red shaded bands show 5-95% range for 58 simulations from 14 climate models using both natural and antropogenic forcing.
Both bands equally follow temperature record up to 1950, and then red (antropogenic) bands begin to disengage and head up. That’s to say, according to IPCC climate models, antropogenic climate forcing begins to influence global temperature only in second part of 20 century.
It corresponds well with actual emissions of carbon dioxide from fossil fuel combustion and total flux in atmosphere, presented for example here:
Emissions were small prior to 1950 ‘€” less than 1 Gt carbon, and since 1950 increased steadily to about 6 Gt carbon per year at year 2000.
I fail to understand how presented “hockey sticks”, showing steady rise from 1900 to 1950 and then hesitate and diverge, could be considered as proof of AGW.
Forgetting for a moment that none of those lines is what it purports to be … just take the whole mess at face value. What I see is this:
1) The caption is proven false by the graphic it allegedly describes. “… all suggest that it is warmer now than at any time in the last 1000 years”. No they dont all suggest that. Their ‘Esper 2002’ line suggests that it is no warmer now than at any time in the last 1000 years, and the ‘Juckes 2006’ graph says that it is cooler now than at least three past temp peaks. And many of the rest show current temps within ~0.1C of some pre-SUV peak … certainly within error.
2) Speaking of error, apart from the ‘Crowley 2000’ splice job, all ‘temperature reconstructions’ miss the ‘Direct measurements’ of the latest temperature by ~0.8C.
3) The error in 2) is pretty well matched by the typical discrepancy between the high and low estimates for any particular time in the last 1000 years, which seems to hover around 0.6C.
4) Uh, arent the errors in 2) and 3) approximately the same size as the alleged measured warming that is going to kill us all?
5) (-1.0C) – (-0.8C) = 0.4C
This is ‘The New Scientist’? Seems like ‘The Emperors New Tailor’ to me.
Dear All,
The two figures (2006 and 2007 versions) I drafted for New Scientist utilised data archived at the World Data Centre. I used what was available – therefore the criticisms against the Crowley and Briffa data are irrelevant as far as I am concerned as I was only plotting what was available.
I was asked a few weeks ago to update the figure for yesterday’s issue. Gabrielle Hegerl and Martin Juckes kindly sent me there reconstructions for this. The Juckes version shown here is an ‘update’ from the original 2006 submitted version and should probably actually be referred to as Juckes 2007.
In drafting the figures, I essentially sent NS a simple spaghetti plot in a format that the NS graphics department could ‘edit’ for their purposes. The emphasis of Mann 1999 was their choice, not mine, but I guess was done intentionally to highlight this ‘famous’ series. Note that NS have actually mixed the colours of Hegerl 2007 and Briffa 2001.
Rob
How was MBH99 smoothed? And Direct instrumental ?
Yes – sorry – I had forgot to say.
The time-series were smoothed with a 50 year cubic smoothing spline
For the 2006 figure, I used the smoothed versions of the original archived reconstructions
In the 2007 version, I scaled (same mean and variance) the smoothed proxy records to the smoothed instrumental data (annual extra-tropical land only temperatures) over the 1856-1960 period.
I thought this might be a slightly better way to present the data in the 2007 version as the original reconstructions had different instrumental targets (i.e. summer vs. annual / land only vs. land/sea etc). It does not make a huge difference to the final figure though.
Rob
Such a shame that they all fail tests for statistical significance, isn’t it?
Was your goal to emphasize the divergence problem, or to preserve the late 20th century instrumental trend?
bender,
If AGW is of such a strength, that it can only show up when all the trends (PDO, NAO, AO, etc.) turn positive, then how strong a trend could AGW be to begin with?
What happens if you do this to the “Direct Measurements”?
Why? Has it been published? Maybe it should be referred to as (Juckes, in review)? More check-kiting?
Jim Johnson (#33),
This is a surprising point and actually pretty funny 🙂
Having followed SteveM’s careful dissection of the HS here on CA and the many problems it revealed in existing temperature reconstructions, and having read the NAS and Wegman reports which confirmed SteveM’s findings, I had fully discounted the HS and stopped paying attention to it. I had not noticed that if you look at the HS graphic one line at a time, what stands out is that there was cooling from the MWP to the LIA and warming since then. It’s not clear at all that we’re back to MWP temperatures.
To Rob Wilson
Its good of you to comment. What do you think of this story? If Steve McIntyre is correct in this and recent posts, this is sensational news. The proxy curves are absolutely fundamental to the AGW case – in that they represent THE argument to say that current warmth is unprecedented in the past 1000 years.
Do you think Steve has a case, or do you think he’s got it wrong? If you think he has any kind of a case, you would presumably be discussing it with your New Scientist collaborators. If there’s a case, then surely science journalists should be jumping on it. Its news!
Your views would be appreciated.
This is an important point because those of us not immersed in the AGW faith feel there is some kind of self-sensorship in most of the news media regarding any evidence that questions or challenges the ‘consensus’.
Re #39
I’m not going to engage in a parsing war with you over your straw man. Ask a straight question that is on topic and I will answer it. I don’t have time to debunk your misconceptions.
#40
I think it has been applied to Direct measurements. Without using 2000-2007 data? Let’s see,
help splineI wet myself laughing reading the NS article- particularly the simplistic links made between CO2 in the atmosphere and warming;
1) “These lags show that rising CO2 did not trigger the initial warming at the end of these ice ages – but then, no one claims it did. They do not undermine the idea that more CO2 in the atmosphere warms the planet.”
No mention, of course, of the work by Rothman, D.H., Atmospheric carbon dioxide levels for the last 500 million years
Proceedings of the National Academy of Sciences 99 (7): 4167-4171, (2002).
In which he states “The resulting CO2 signal exhibits no systematic correspondence with the geologic record of climatic variations at tectonic time scales”
2) “We know CO2 is a greenhouse gas because it absorbs and emits infrared. Fairly basic physics proves that such gases will trap heat radiating from Earth, and that the planet would be a lot colder if this did not happen”.
Again no mention of the Beer-Lambert law- more basic physics- which states that the intensity of transmitted radiation decreases exponentially with increasing depth of absorbing material. Or that the principal IR absorbance bands of amospheric CO2 are already at, or near to, saturation (Thomas, G.E. and Stamnes, K Radiative Transfer in the Atmosphere and Ocean. Cambridge University Press, 1999.)
What happens if you do this to the “Direct Measurements”?
I ask because the highest temperature differential in the GISS Surface Air annual average temperatures is 1.33°C.
If you do a simple rolling average this differential is reduced to:
– 1.08°C for a 5 year rolling average – similar to the “Direct Measurements” in Rob’s compilation
– 0.89°C for an 11 year rolling average
– 0.43°C for a 50 year rolling average
When Rob Wilson says:
What is the centre of the 50 year period?
It looks as though the reconstruction extend to ~1980. If the 50 year period is centred on 1980 then annual data most extend to ~2004. If the 50 year period is “centred” on the last year of the period then the earliest data must be from ~850 AD.
And whatever happened to the NAS report’s statement about reconstructions exending back more than 400 years? Is everything prior to 1605 simply plausible? (or is that 1580?)
Re #46 Why would you wet yourself laughing when a journal of that nature fails to discuss the science in the depth you seek? When you read derivative material that’s inavariably what you are going to get. That their presentation is sophomoric does not imply the argument they’re trying to present is incorrect. They’re journalists, not scientists.
News flash: it is easy to find fault with the glossy derivative AGW material. Once you’ve conquered that skill, it’s time to move on to the primary literature. And once you’ve conquered that, it’s time to write up your own GCM with your own brand of physics and make your case that way.
And when you are there, I will make sure to have plenty of towels handy for when I wet myself laughing.
Re #44,48, Bender
Hmmm.
RE #36 Thanks, Rob, for the information on the instrumental data.
A question for anyone: what is the basis/significance of using a 50-year cubic smoothing spline versus other possible methods of smoothing?
Not only is the Mann green line bolded and on top of all execpt the Direct Temperature Readings line, it is also boardered in white, so it is further set apart. None of the other 11 lines received this special treatment. — John M Reynolds
#50. David, I don’t think that much turns on the different methods of smoothing per se. Cubic splines seem fair enough to me and don’t use end-period padding. The key issue with these spaghetti graphs are the things that we’ve talked about: deletion of data that doesn’t fit the model (the Briffa truncation); use of condemned series (strip bark bristlecones) and condemned methodology (Mann’s PC1); cherrypicking of (say) Yamal rather than Polar Urals (and then calling it Polar Urals). I am prepared to stipulate that if you mix foxtails/bristlecones, Yamal with noisy data with negligible common signal, you will get something that will not offend the standard spaghetti graph. It’s just new mascara.
Hi,
last comment before I go climbing over the long weekend.
I used a 50 year spline to highlight the decadal to longer term variability in the series.
I could have used a 40 year spline or a moving average or Gaussian filter of varying length.
As the NS figure is schematic in nature, I do not think it makes a real difference. The point of the figure was to compare the instrumental data (the recent period) with the proxy data.
Anyway, as a dangled carrot for all your tree-ring non-believers, the figure below compares a new TR based reconstruction (1750-2000) of extra-tropical NH temperatures with the other records shown in the NS figure. My new series utilises 15 local/regional TR reconstruction from various regions around the NH which DO NOT express divergence at the local regional/scale. Steve saw my presentation at AGU on this work and it will be published later this year. The smoothed series (50 year splines) have been normalised to the common 1850-1960 period and are therefore expressed as standard deviations. My new NH reconstruction uses a completely independent (i.e. not used in any previous NH reconstruction) data-set of tree-ring data and is purposely short as I was focusing on the divergence issue and why it is so difficult to model large scale temperatures in recent decades. Although the instrumental data are not directly included in this figure, they do form the recent end to the Crowley and Hegerl series. As you can see, my new NH series still shows some divergenve. The paper discusses many reasons why this is so. It is a complex problem.
#53. Rob, it sounds to me like you’re doing the same thing as Gordon Jacoby did in 1989. Jacoby studied 36 sites and picked the 10 most “temperature sensitive” sites from which he derived his HS version. (He’s refused to archive or disclose results from the other 26 sites but that’s another story.)
Readers of this site know that if you select the 10 most upward trending series from a network of 36 red noise with persistence properties of tree ring sites, you will get series that look like Jacoby’s or for that matter yours.
The impasse that we get to is the after-the-fact selection of sites. If larch or Engelmann spruce at treeline (or whatever) are valid temperature proxies, then one should be able to assemble a network of all such sites and find the supposed signal across the board. Once you start picking ones because they don’t diverge, you’ve fallen into the same trap that all these after-the-fact studies have. From my vantage point, it seems that for every site that doesn’t diverge, there’s at least one that does.
Re: 52
Steve, Is there a summary or chart somewhere listing the various potential faults of each reconstruction (condemned series, condemned methodology, cherry picking, truncation, etc…)?
Re#53, unfortunately this new TR series doesn’t extend as far back as many would like to see.
The timing/magnitude of warming also seems to fail what I referred to in #25.
Then why did it cover 850-1750?
P.S. Thanks for NOT starting the graph at AD1000.
#55. Unfortunately not. It’s one of many things that would be a good idea to do. IF you look at the Categories in the left frame, there are categories for several of the reconstructions and that will give you an idea.
The main problem is that the “proxies” don’t have much of a common “signal” if indeed the metaphor of a “signal” makes any sense. As a result, the medieval-modern relationship depends on highly stylized decision-making by very biased designers. There are a couple of key decisions that I look for in any reconstruction: bristlecones and foxtails. HS-manufacturers are addicted to bristlecones and foxtails. These are highly problematic with a 20th century growth pulse believed by many specialists to be nonclimatic; NAS recommended that these be avoided. However the HS dealers are addicted and continue using them despite the NAS panel. Sometimes not just once. For example, Moberg in a small sample of 18 series, used 4 bristlecone/foxtail series. Amusingly he used one site twice in both an older version and a newer version. Hegerl used both a bristlecone and foxtail series out of 12 sites, with one of them being Mann’s discredited PC1, which seems to be used more often by other climate scientists now that it’s been discredited – seemingly a show of solidarity.
The other key decision is Yamal versus Polar Urals, One has a big MWP and one has big 20th century trend. You know which series the Team will choose. But the MWP-modern relationship is not stable to using one rather than another.
I’ve enough experience with accounting to be very suspicious when the Modern Warm Period index consistently comes out a hairs-width warmer than the Medieval Warm Period in Team studies. I’m aware of too many accounting decisions when companies juggle the books so that they manage to show a profit. You can think of famous examples.I’m not saying that the Team has done this, but that these hair-breadth differentials are very much a red flag.
Steve,
having not read the paper, you have missed the point of it. I will send you a copy soon.
I will readily admit that I have cherry picked. I did that on purpose to increase the chance that I could develop a divergence free reconstruction. However, despite this approach, the new series still diverges from the instrumental data. I expected this and the paper really focuses on possible reasons why. Amongst the many reasons I discuss, the number of proxies is by far one the the most import. I do not believe that 6 or even 15 proxy series are enough to reconstruct NH temperatures robustly. How many would actually be needed is the $1,000,000 question and is the focus of my ongoing work.
anyway – enough – please – CA is a huge distraction and the mountains call
Rob
snip
Rob, what is the purpose of picking the TR series that show a divergence free reconstruction?
Are you trying to select trees or areas that correspond closer to the measured temperature record? If so, I think this is actually a good strategy. But the end-point of that strategy means that we should eliminate the non-responding TR series in all the other reconstructions as well.
That would also mean we do not have a 1,000 year reconstruction at all, we just have yours from 1750 onward. The rest is just spaghetti on a plate and should not be used as evidence of historical climate. I hope you see my point.
Bender ree #48 The reason I laughed is because it is so typical of the “on message” pro-AGW mush written these days in both the popular media and the soft (and not so soft) science journals. As you say knocking this glossy stuff is simple. FYI I do try and keep pace with the primary literature and in doing so have arrived at the conclusion that AGW is anything but proven. One area I am not familiar with is climate models – I don’t have the mathematical or computing background to understand the formulae and their implementation. What I do know, however, is that they are trying to make sense of very complex non-linear and chaotic systems. They do not model various critical forcing agents (e.g. clouds) effectively and the number of iterations that they need to perform to arrive at a conclusion is so large that any small initial error will be greatly magnified. I’m sure that you will argue that in the context of climate modelling, that I am pontificating from a point of limited knowledge. With that I agree. However others with rather more knowledge in this area would argue (and I would agree) that you cannot actually prove anything with a climate model and that it is therefore folly to change our whole economic, industrial and social structures on that basis.
Perhaps all of these proofs should be submitted to the Journal of Non-Reproducible Results. The proxy measurements of temperature would seem to be right up there with cold fusion and Korean stem cells.
Re #62 That is a fair response and I respect that.
Re #63 Non-reproducible? What do you make of the fact/statement that the Wilson 2007 reconstruction (thick black line in #53) is completely independent of all previous Team work? That looks like reproduction to me. No? And please note that his self-admitted cherry-picking (to bias against divergence) only affects the last 20 years, not the first 250.
Precisely. If these tree ring reconstructions are meant to be reproducing a single phenomenon then why are they all so different?
This is the IPCC AR4 presentation on past climate:
snip
These assertions are false. What a time sink this is becoming.
#John A, the idea of estimating a global mean temperature anomaly is not a foolish idea. I don’t want to debate this topic and I would urge you not to keep interjecting this particular notion into discussions of proxy reconstructions where the salient issues are quite different.
RE #52, #53 Thanks for the replies.
My curiosity is about spline smoothing methods and how they handle the regions near the end point. I’ll move that question to Unthreaded.
Step 1) Throw out any data that doesn’t match your theory.
Step 2) Publish an article showing that the data matches your theory.
Step 3) Only archive the data that matches your theory and refuse to tell anyone anything about the data that was discarded.
Re#59
You’ve covered quantity of proxies, but not explicitly coverage. I would assume that’s one reason why you’d want to see more proxies, but I could be wrong.
In any case, do you think the limited spatial coverage of proxy locations shown here, regardless of the quality of the proxy data and methodology uses for the reconstructions, would be “enough to reconstruct temperatures robustly” – robustly enough to conclude that the resulting reconstructions “indicate that late 20th century warmth is unprecedented for at least roughly the past two millennia for the Northern Hemisphere?”
These assertions are false. What a time sink this is becoming.
Aside from the assertions by an anonymous blogger going by the name of a cartoon character could someone please explain why john A’s assertions are false?
If there is evidence or methodology that is generally agreed upon by skeptics and the AGW consensus would it not be a good idea to spell it out?
Re: #67
OK. Let’s forget Global Mean Temperature for the moment. If they’re supposed to be measuring the same thing, some variable that is meant to be a measure of climate, why do they not converge on what that single thing is?
Because just because the “proxies” don’t consist of signal plus white noise or low order red noise. If the proxies were even decently good, then it really wouldn’t matter very much which proxies were selected or what multivariate method you used. You’d be able to randomly sample from the universe of proxies and get roughly similar results.
If one’s hypothesis is that the “proxies” are merely persistent red noise series, selected on the basis of 20th century trend, then you can make spaghetti graphs effortlessly that look much the same as these ones. I’ve posted some up a while ago, which would probably make sense to bump back into play.
Looking at the spaghetti chart in #65, it is apparent that there are no strands rolling off the table (below the zero line) at about year 1400, similar to the late 1900’s. So that suggests that the MWP may not be regional in nature. Now if only the keeper of the keys would “correct” those temperatures a bit we could have a real MWP.
Rob,
Thanks for posting. Plus thanks for being a gentleman about things. At some point it would be
nice if you could provide the actual data, so folks dont have to digitize spagetti graphs.
Prove it. Show me the histogram of pairwise correlations among all series. Does this distribution have mean of zero? (Hint: No.) That they are in fact consistent with each other is precisely what Steve M is pointing to in #74. And I happen to agree with his concern. You sometimes go too far, John A. I was happy to see your retraction in #73.
#76. The data that Rob used in the spaghetti graph is mostly available at WDCP as he says. I’ll post up a script to replicate this spaghetti graph adding in the
deleted Briffa portion.
Actually bender, I don’t agree with that. Their only consistency is that they have a non-zero correlation with the instrumental average because that’s how they were selected. Outside of that, I’d defy you to show me that those individual studies have an R2 significantly different from that of similarly scaled red noise.
Regarding how similar the graphs are and the evidence for the MWP, if you scale them all to have unit variance over the period (see Box 6.4, fig 1 legend above) then any larger peaks (like the MWP) are simply squashed down.
Re#48 Bender
My personal opinion is that the whole debate over AGW will never ultimately be decided by GCMs. If you’ve been following the exponential growth thread, several things have come out of the debate that are important. As Dr. Browning has pointed out many times, NWP (used for weather forecasting, not climate forecasting) models have been produced largely by trial and error. This is possible because of the short time frame of the predictions. The predictions can be tuned by measuring the error and “tuning” the model. The model is repeatedly tuned in this way until it gives fairly accurate prediction. Actually, trial and error is a big part of computer intelligence (artificial intelligence if you like) since computers can systematically test out a vast number of possible solutions to a problem in a short period of time. You simply keep trying until you find a solution to a problem. Some AI languages such as prolog are designed for just such a purpose.
Obviously, long term computer climate models can’t be “tuned” in the same way weather models can. This means trial and error is out, and you are stuck with a purely analytical approach where your model has to be nearly perfect from the start. With all the unknowns and complexities involved with climate, I don’t see this being possible.
This is the reason why things like temperature reconstructions based on proxies will continue to be very important to the debate.
Rob Wilson:
I recall that, during the Senate Energy and Commerce hearings w/ Wegman, North, Steve M., et al, the shameless journalists at the Wall Street Journal were blamed for leaving the error bars off the hockey stick. Mann and company were, of course, innocent of any resulting public confusion about the claimed certainty of their scientific claims.
What’s your excuse for misrepresenting Mann’s and others’ work when producing a spaghetti graph for public consumption ? Why no error bars for the lay public ?
#82. I wish that at least here people wouldn’t dignify the MAnnian error bars with the term “error bars”. They are simply 2-sigma standard errors in the calibration period. These are not error bars in any valid statistical analysis.
Re #82
The real error bars are illustrated in white with a thin horizontal blue stripe.
Craig: I am anxiously awaiting your new paper. Has it been accepted for publication?
Nick-named “floor” and “ceiling” respectively? 🙂
Mark
Re #79
I defied you first. 🙂
(The answer will come from of Steve M’s script when it appears.)
Re #81
I know what an overfit is. Thanks though.
#83, Steve M.
Fair enough, it appears when I look at them that they’re arbitrary constructs and not based on any statistical analysis of underlying annual data. That being said, Wilson, et al. should not claim higher certainty than Mann et al do – allowing the authors to defend at the next hearing that it’s not their fault Wilson misled the public by inaccurately representing their work.
Even if the “error bars” are complete BS, they still represent the extent of Mann’s claims. Mann supporters should take pains to illustrate Mann’s diminishing confidence 400 yrs BP.
Would you prefer the term “confidence limits” ?
#84, John A
I interpret your comment as referring to the entire area of the graph. Very good.
I don’t fully understand what Rob Wilson is aiming to show in his ‘proxy graph’ (53, 59). but it looks to me like he is showing a trend that started around 1830. If this is the case, this is well before significant modern CO2 emissions, and is fully in line with the evidence that glaciers have been receding since around 1880. So: to Rob Wilson – what do your results show? Is it that they support a temperature rise that started around the mid 1880s that continues, or do they support modern manmade global warming – starting in the 2nd half of 20th century?
Rob has transformed the archived data in a curious way that I’m trying to figure out. All of these recons have a 1961-1990 center. But look at the graphic: NONE of the reconstructions are centered at 0 in the 1961-1990 period. They’re all below it. This is not the case with the archived versions most of which are already in a 1961-1990 basis. So Rob’s rescaling and recentering has altered this.
As I catch up on a newsworthy couple of days at CA, I want to thank Rob Wilson for participating here! Dialog with dendro people is long overdue, even if it’s only from an army of one.
That his contribution to the New Scientist piece inevitably draws attention, goes without saying. In these overheated and testy times, I hold fast to Popper’s commendation that our scientific ideas improve through the fires of criticism: here’s my thanks, also, to the participants for generally keeping it constructive here at Steve’s web site. If only this was widely emulated.
And Rob, good climbing to you!
#53 Rob
Thanks to you, especially posting, others here, and on RC. Hope you had a good and safe time in the mountains.
I had a real problem with a low signal that is categorically real, with intermittant signals that could be as much as an order of magnitude higher, lower did not count, as much. This is a control problem. Some of the reset times had to be able to respond in 20 seconds and would controllably effect the desired result. One of the resets had to respond at 360 seconds to controllably effect the result. We needed a feedforward. For estimators, we set the constant/gain at .64 (that golden mean fellow, for old codgers like me). And other response times in between 20 and 360 seconds. They had in common that they were different, and they were known to be real.
So I decided to use as best possible (as described here) a Mannesque solution. The important point is that when I told the control scheme we knew what the answer was, it became unstable. When I told the control scheme, we knew about half as much as we though we knew, and left control loose, control became the best we have had. (Simplified for post.)
The point is, I have a suggestion. As noted in several threads, it appears that part of the problem is over-tuning. Why not run your proxy like this control problem. If you don’t know (and assume you don’t), have it explain only half or .64 of the problem. Perhaps if you do this, your R^2 will go down in one sense, but the divergence problem will become much less, such that the loss in R^2 is more than made up by getting rid of divergerce. I am assuming divergence, in this case, is like my control becoming unstable due to overtuning (thinking I could solve the equation, rather than letting the system have more freedom). If you do this, could you use the 11 year that some argue about from Swindle?
snip
#93,
I believe Roger Pielke Sr has said something very similar.
#77 “Show me the histogram of pairwise correlations among all series. Does this distribution have mean of zero? (Hint: No.)”
The non-zero correlation could be due to a sample bias error. The number of proxy series is very finite and not well distributed, globally.
Some years ago I had reason to look at the emergence spontaneous chirality in natural quartz crystals. In a totally stochastic process, the number of righties and lefties should average out, making the total chirality zero. This turned out to be a question that many people had been interested in, and thousands of bits of crystalline quartz had been sampled world-wide. Indeed, the chirality tended toward zero as the historically tested sample size increased, but even after 27,053 crystals checked, there was still a 50.17% to 49.83%% bias to right-handed excess. But the trend was toward zero (or perhaps toward a tiny effect).
So, the question as regards proxies is not whether the pair-wise correlation is zero or not, but rather whether the correlation with respect to measured temperature tends toward zero as the number of proxies increases. This, I think, could be checked, even locally.
Re: #85 yes it is in press in E&E and as soon as I have Galleys I will forward to Steve
Eh… excuse my lame interjection, but… where are they finding all these 1000 year old trees?
Oh come on now.
1. Well mixed greenhouse gases will have global-scale effects. The issue is what strength of effect.
[snip]
KM, read the blog. There are dozens and dozens of threads devoted to tree ring research. These trees are located in every part of the world, usually cold, dry places (alpine deserts) where trees grow slowly and decay slowly.
I agree that a complete census is better than a sample. Who wouldn’t? The point is John A’s ridiculous assertion that the proxies are “wholly inconsistent with each other”. I would have refuted this earlier today, but did not have time until now.
John A is a pretty good rhetoritician, and he’s used a clever trick here, turning the dendros quantitative argument into a qualitative one: “THEY’RE ALL CORRECT” are the words he’s putting in their mouths. I won’t call everyone on their rhetoric, but John A knows better than this. The dendros do not argue in the black and white about their reconstructions. They know they are imperfect and error prone. So why is John A trying to present such an absolutist position? It is dishonest. News flash: a correlation coefficent varies continuously between -1 and +1. It is not a binary variable, so quit arguing in absolutist terms.
If Rob Wilson really has produced an independent* reconstruction, this is an important development. It would be a first. Or at least a major departure from Team work. He is to be credited for attempting to maintain some independence.
*On the issue of statistical independence, I fully understand what Steve M is getting at, I just think he’s wrong on this one. Series that are red noise but are correlated with one another must have SOME basis to the correlation. What could it be? We know what it’s NOT, because there are plenty of variables that DON’T correlate with tree growth. If temperature exhibits a weak correlation among a wide variety of species across a wide variety of locations, well, hmmm, do you think, maybe …???
Finally, this assertion that the proxies are not well distributed globally is misleading. They’re well-distributed in terms of continents, but not well enough distributed among countries to have even global coverage. Well, duh. You don’t find cherry trees just anywhere. (And yes, I know there’s an irony there. It’s a joke.)
Never mind. You are the same guys who wouldn’t be satisfied that a global mean field exists until we put a thermometer in every household. Even then, I can hear the naysayers.
Guys, you’ve got to pick your battles better.
I do agree with the spirit of #84, however, and have been saying so for some time. The true “error bars” or “confidence limits” (neither of which may be applied to what I’ve seen so far, as these are well-defined statistical terms) are probably very, very wide, and I am particularly skeptical about the accuracy of these reconstructions during the megadroughts 1000 years ago. So is Rob Wilson, for that matter.
Divergence, by the way, does not prove that tree rings are invalid as proxies. It may suggest the models used to build the reconstructions are somewhat deficient. I have heard someone here complain of “overfit” reconstructions. Nothing could be further from the truth. The calibration models are pathetically overly-simplistic. There is no opportunity for “overfitting” whatsoever. A correlation coefficient, for example, is one parameter. If a tree-ring time-series gives you 1000 observations and a hundred degrees of freedom (heavily reduced, due to red noise), that’s still alot of degrees of freedom with which to estimate a single parameter.
The uncertainty is not only ENORMOUS (in both paleo AND computational cliamtology) – it is the one thing the warmers are truly trying to hide. It’s the one thing they are very afraid of. They have no handle on it whatsoever.
[snip]
Regarding, “absolutist terms” , and not to make much of a fuss one way or the other but these reconstructions with their caveats and nuances are presented by the MSM in “absolutist terms” with hardly a word of “clarification” from the dendros. Hints and allegations of a perceived phenomena are great for research that has a chance to search for the truth unhindered by political influences. Unless you proffer a theory and test it rigorously, you really don’t have squat. Plant little stands of trees all over the world, take careful measurements of the temps, precip, etc then core the little buggers after X years, maybe do that in a lab environment first, prove your theory then take it to the field. Until that is done, what you have is just a little more rigorous than sociology.
Go on and rant a while and then I suppose I’ll have to spend more time refuting you tomorrow. Which won’t be difficult, I assure you. One refutation just leads to another challenge, and so on … it never ends.
Such as the incredibly controversial theory that some growing seasons are shorter than others, and less favorable for plant growth? Don’t garden much, do you? My advice: get over your indignation, get a grip, and let’s talk facts.
[snip]You know what an Iowa farmer would do if you told him he had to define a corn heat unit before he could be allowed to grow corn?
If you’re in denial over such simple things, I’m afraid I can’t help you. I’m not sure what’s worse: blind faith or blind skepticism.
So you are saying that testing the theory of tree rings and temps in a controlled environment is a bad thing? Or maybe unnecessary? I can understand a group of researchers having a hunch and telling whoever will listen that “hey it ain’t perfect, we really don’t understand it completely but we are working on it”. Except that is not what guys like Wilson are saying to the public at large, in fact quite the opposite.
The MSM is stating various metrics in an absolute context with no attempt at clarification, no talk of nuances by the various paleos, dendros, etc. So one can hardly blame someone like John A for taking a similar tack in defense of his own views.
[snip] An existential argument won’t work here however, showing that such a thing exists is meaningless unless you can write it down so that you can use it for something.
#105. once again, please don’t convolve discussion of proxies with the “issue” of global mean temperature, something that doesn’t trouble me and that I would rather not expend bandwidth here.
If a researcher was going to use a chu data to predict how much corn would grow in a given season then he had better know what a chu is.
You sound confused and you are definitely projecting. Accusing me of indignation when I ask for a definition of something that should be defined if it is going to be of any use and suggest that some controlled experiments might be in order is over the top, imo. Again you sound confused and angry.
105 didn’t mention global mean temperature, 102 did. 106 and 108 hinted at it. But I cease talking about gmt on this thread.
snip
re 100
*
Unless the biological parameters were of equal or greater sensitivity. Say for example with the diurnal cycles of photosynthetic assimilation which are more strongly affected by spatial variations in vegetation parameters than by meteorological variables. This indicates that topography induced variations in vegetation parameters are of at least equal importance for the fluxes as topography induced variations in radiation, humidity and temperature.
Temperature has a correlation with growth,in so far as there is suitable photosynthesis available radiation ie photon flux at readily assimilative wavelengths.
#110, willis. I don’t expect bender to engage in this discussion on mean temperature. The conventional method of anomalies seems quite reasonable to me as a means of discussing issues relating to temperature change, while avoiding the issues that you’re worrying about. I don’t want to spend time engaging on this topic nor do I expect bender to.
#101
Multivariate calibration models are very prone to overfitting. Univariate models are not. But if you have divergence problem, it cannot be overfit at the same time 😉
#110 and #112 Steve M
This is a very interesting thread.
The issue over what is defined as global mean temeprature (GMT) and how it has changed over the last 1000 years and whether this is indisputable evidence (as it clearly is to the pro-AGW crowd) of man’s effect on our planet is FUNDAMENTAL to the whole AGW debate.
Why therefore can we not have a debate on this blog (under a different thread please) about what is the definition of GMT and whether or not it is a physical entity or not (which IMO it clearly is not i.e. I’m with John A and Willis on this one)? This also applies to the so called temperature anomally regurgitated ad nauseam in all these (IMO pretty useless bloody spaghetti graph) proxy reconstructions.
KevinUK
A General comment:
Temperature anomaly plots are essentially plots of the variance about a mean, by definition.
Say we have a large dataset of measured temperatures spanning a couple of hundred years and we arbitrarily decide that a period, say 1960 to 1990, is a baseline. This baseline is a mean temperature which, when subtracted from observations produces variances from that predetermined mean. These variances seem to be called anomalies.
What they are not are temperatures, and given the magnitude of the temperature anomalies, those are far less than the instrument accuracies.
They are, in essence, mathematical artefacts, not real measurements.
Yet we read so much debate about such ephemeral deductions here.
KevinUK, bender, Willis et al
Can we not deal with the definition of global mean temperature here, since its not on Steve list of things to discuss?
In a few days, I will put down my own thoughts on GMT on my own blog and then I’ll invite allcomers to debate the question to their heart’s content.
bender (#100) argues that
I think bender knows this that this statement is not correct (this topic has been covered extensively on CA). Specifically, independent red noise series will, on average (if a mean exists), exhibit zero correlation. However, they tend to exhibit large absolute correlations with surprising frequency in comparison to white noise series. In short, high absolute correlation can result from either a common deterministic factor or from the time-series structure of the (independent) errors. In the latter case, there need be no causal relationship between the series.
#116 John A
Steve has said that he doesn’t what to discuss the definition of GMT here on this thread. Your offer to discuss it elsewhere is welcome but your blog is unlikely to attract as many visitors as here on CA.
I suspect that a large number of the most recent visitors to CA (say overthe last 6 months) have neither the time nor the inclination to read back through previous threads to find out where certain subjects have been previously discussed. I think its time for a frank discussion on/revisit of the ‘pillars’ that hold up the AGW temple, the whole concept/idea of a GMT and its anomalous variation (the spaghetti graphs) being just one of them. What do other visitors think?
Rob W
It good to see that my taxes are being spent well (not!) north of the border (as well as in Oxford) :-).
Re # 118:
It seems to me that it is up to Steve M to decide. And he was pretty clear about it.
#117 TAC, I’m talking about LONG series. Not short series. Therefore I’m not incorrect. (I’m not one to knowingly make incorrect statements. You should know that by now.) A signal mixed with red noise still carries signal.
#108 Apologies to Jaye on the one instance when I used “you” in the generic sense (y’all). I was not “confused”. Maybe I was “projecting”. I was trying to get numerous people here who seem bent against proxies to realize that they do have LIMITED value. They’re not worthless. They’re not baseless, as #111 indicates. It’s because the proxies have LIMITED value that they need to be audited. Deny that value, and you deny the value of CA.
#108 *I* don’t need to define a ‘chu’, that’s homework for you (=y’all) who want to reject the theory of growing season length.
[snip]
Let’s just simmer down, get back on topic, and wait for Steve M’s script. The issue here is whether the proxies are meaningless spaghetti or whether they have limited meaning. Y’all are avoiding the import issue here: Wilson claims to have an independent reconstruction. Why would you reject this work when we have the option of auditing it? (I wonder if he used CA or Wegman to figure out what combination of proxies would yield an independent reconstruction?)
#114 Louis H
Good post! It always good for someone (in this case Louis H) every now and then to remind us exactly what it is that all the fuss is being made about i.e. a manufactured mathematical artefact. It beggars believe that billions are being spent on this (academic debate on this artefact) rather than on alleviating real global problems like poverty.
KevinUK
Also why can’t we discuss Beck’s work here on CA?
KevinUK
Mathematical artefact?
More junk.
Take it to unthreaded. You’re polluting an important thread with garbage.
#123
bender, I have a lot of respect for you and all the work you’ve done and the arguments you’ve presented on this blog in the past. I’ve learned a great deal from your past posts. With all due respect to you, calling proxy temperature reconstructions mathematical artfacts is not junk.
Cherry picking proxy data, using made-up statistical techniques to analyse that cherry picked data, hiding (well being stupid enough not to hide) the non cherry picked data in a folder called CENSORED on a public FTP server, including TR proxies which are known to be proxies for precipitation and NOT temperature and publishing that reconstruction in the IPCC TAR and continuing to defend it is JUNK!
KevinUK
#122. I don’t want people coming to this blog, seeing that’s preoccupied with things like Beck and GMT. Neither of these meet my hurdle of an issue worth analysing. If the topic gets online, I’ll either have to waste time refuting the discussions or at some point, some critic of CA will surely criticize me for letting these observations pass unchallenged on my blog and demean the blog that way.
I ask people from time to time to raise their game. And getting involved with Beck and GMT is the complete opposite.
Re #124
Now who’s “projecting”? That was MBH98. This is Wilson07. He has a new “independent” reconstruction and I, for one, can’t wait to see what is in it.
#113. UC. The statistical model for MBH is a unique combination of univariate methods and multivariate methods that hasn’t really been explored. It’s a combination of univariate spurious regression (bristlecones) and overfitting (the network of 21 other series that more or less function as white noise). Mann’s particular “genius” was developing a method to combine spurious regression with overfitting. But I think that you absolutely can have overfitting in the calibration period 1902-1980 or whatever and out-of-sample divergence afterwards.
re:#31
bender,
I agree the PDO and NAO play a very significant role in climate. It is nice to see someone else who thinks so. When I read the TAR or 4AR, it does not sound like internal climate variability plays any role at all.
I just want to point out that the PDO just went into the cooler phase last year (right on time according to the Bratcher and Giese prediction) and since the PDO has a 30 year cycle, we may not see the 1998 record broken for quite a while. I am not certain about the NAO, but I think it is cycles within a year or two of the PDO. Is that not correct?
#126. bender, there’s a list of sites in the Divergence paper. The trouble with his new paper is that it only covers the period after 1700 and doesn’t go back to the MWP, which is the period in controversy. Right now only a couple of the sites in the Wilson short recon are archived. Hopefully Rob ensures that the data is available on Day 1 so that it’s possible to see what he’s done. While he may get cross with us, at least he knows we’ll read past the abstract. What I anticipate will be the problem is that for whatever sites he chooses, I suspect that there will be equally plausible sites left on the sideline that have a Divergence Problem (or that don’t have a divergence problem but don’t have a HS either). Unless he can define an objective criterion for sites that diverge and sites that don’t diverge – a criterion that can be applied to new sites without data snooping – I’m afraid that it’s going to be the same old Jacoby-type picking. But I’d like to be disappointed and if Rob can break the problem, it would be great for him and the discipline.
bender, you should read or re-read Ferson’s articles on the interaction of data mining and persistence in yielding spurious regressions in attempts to predict the stock market. It’s a story of people always “moving on” to new indicators as the old ones fall apart. The type of statistics carries over and, if the Team actually wanted to raise their game, they’d familiarize themselves with this field, unrelated as it may seem/
What stock market prediction and tree ring analysis have in common is that they are stochastic time-series forecasting problems. I agree that they are not unrelated. I agree that dendros need to pick up their game in this area. I think Hal Fritts would too! It would be fun to be in an investors club, the dendros vs. the skeptics, and see who goes broke and who gets rich.
While I was re-collating data for the New SCientist spaghetti graph, I looked at the data for the Juckes spaghetti graph. Guess what. Juckes et al (Briffa, Moberg, Zorita, Hegerl) deletes all values for the Briffa reconstruction after 1940 , neatly terminating it at its highest level. I’ll post this up later.
These guys don’t leave a scrap on the table. They’ve gotten bolder since there’s been no blowback on the deletion of values after 1960 and now they’ve gone even further. I wonder if Rob Wilson used this version as well.
bender (#120), first, as you note, I am fully aware that you do not knowingly make incorrect statements. I apologize for inadvertently suggesting otherwise.
I also thank you for clarifying that you were
I understand your point. For simple stationary ARMA models, one need only consider short-term persistence, where correlations die out quickly (asymptotically exponentially). If this were the relevant case, I think we would agree.
However, long proxy records of geophysical time series typically reveal long-term persistence (e.g. fractional Gaussian noise). Such series cannot be modeled as ARMA processes (other models, e.g. FARIMA or arfima, are sometimes used). For such long-memory series (which, by the way, are stationary), correlations do not die off exponentially. In such cases, to escape the “redness” problems may require record lengths of thousands or milliions of years, or even longer.
Where does this lead? First, IMHO, we should always assume we are looking at “short” records when considering geophysical phenomena (see Koutsoyiannis). Second, going back to the original question, I remain skeptical that observed correlations are as meaningful as (I think) you suggest.
The other issue to keep in mind and separate is that the recons are not proxies, but averages/weighted averages of proxies. Because the recons have so many overlapping selections, there is going to be correlations between them (which despite the spaghetti and despite John A, there is). If you take slightly different combinations of bristlecones/foxtails, Yamal, Tornetrask, Fisher’s Greenland, Yang’s China composite, you will get series that have a certain amount of correlation. The issue for the spaghetti graphs is whether their “remarkable” similarity rises above what one would expect from composites made from slightly different selections from a small population of already stereotyped proxies. Because the Team incorrectly asserts that the studies are “independent” and deals in armwaving concepts of similarity, the Team doesn’t consider this problem.
We’re back on track. I agree with all these points in #132 and #133. Independence (i) among observations within a series and (ii) among recon series are critical issues that I think have not been dealt with adequately (in the primary literaure or in AR4). THIS is precisely why I am skeptical about the recons – NOT because I am skeptical about the concepts of growing season length or mean fields.
Divergence is consistent with a number of alternative hypotheses:
(i) a new forcing agent (Briffa, AR4) such as “global dimming”
(ii) nonlinear response to multiple climate inputs (D’Arrigo & Wilson)
(iii) proxies not responding to temperature at all (John A et al.)
As Jaye says, controlled experiments are the most logical means to figuring this out. The problem is cost. Even greenhouse space costs. Yet you have some other commenters suggesting that any experimentation at all is a waste of taxpayer money.
Well, we can’t do NOTHING. The AGW hypothesis is too critical to not do ANYTHING about it. The least we can do is study it.
Re #132, TAC, nonstationary AR models can also generate long tailed histograms.
#134 bender
I beg to disagree. Doing nothing is an option. I don’t believe (yes I’ve used the word believe) for one minute that the current warming is about to suddenly accelerate to the point where we’ll go beyond (or even approach) the much politically hyped ‘tipping point’. Just as we had at least another 20 years to go at least back in 1988 when Hansen hyped up AGW which kicked off the whole IPCC mess, I’m pretty sure that based on the evidence that we have available today that we have at least another 20 years before we’ll get to a point when we can be sufficiently confident that man is the cause of the current warming trend, that we should then take action. In the mean time we can spend the billions of dollars that are currently spent on climate change research on global problems that we know FOR SURE already exist e.g. poverty, disease etc.
KevinUK
In that spaghetti graph on top, how many of the proxies are from the southern hemisphere?
I hope Rob is OK on his climbing trip. The weather forecast is for high winds and freezing temperatures on the Scottish hills this weekend.
Steve (or perhaps John A)
Why have you removed my post in which I provided links to Rob Wilsons bio and list of publications? They are relevant to this thread. The publications list shows that Rob W has clear links with Rosanne D’Arrigo and Gordon Jacoby, whose work you have criticised several times on this blog. D’Arrigo and Moberg and esper et Al who Rob W has also published with are all authors of reconstructions in the New Scientist spaghetti graph above.
KevinUK
Mr McIntyre-
Please delete physical meaning…G-M-T posts like #98 — the thermodynamic ignorance is obvious and violates your sanction against the topic. … Feel free to delete any others you feel initiated the G-M-T distraction — one would argue, consistent application of your policy demands it!
Steve: I’ve spent some time snipping discussion of a topic that I’ve repeatedly asked be discussed elsewhere. There are probably still a few references up because I got bored doing this.
Re#88
Unfortunately, the authors’ text doesn’t shed as much uncertainty as the “error bars,” and the textual conclusions are what gets references, quoted, used in press releases, etc. It wasn’t until 2006 that I heard Mann or his supporters back-off from the general “confidence” of his HS.
#101 — “Divergence, by the way, does not prove that tree rings are invalid as proxies.”
I don’t know why it’s so hard to understand that nothing is a valid proxy for temperature unless there is a rigorous derivation of a temperature metric from the observable. This is the case for oxygen isotope fractionation. It’s not the case for tree ring widths or densities. Divergence raises the issue that the correlations could well be empirical happenstance. What divergence “proves” is that maybe tree rings correlate with temperature, and maybe they don’t. It proves that tree rings, as such, are not worthy of blind trust or qualitative justificationisms.
Only a derivation from theory will establish the issue, one way or the other. Until then, it’s all just shouting.
Some time ago on CA I discussed with Paul Dennis a 13-C kinetics approach to derivation of a true temperature from tree wood. If that worked out, it would be a method of deriving a valid temperature metric from ring wood that is independent of ring width and ring density. It would suffer from its own suite of confounding variables, primarily to do with night-time respiration, but it would be a physically valid metric on the same order as 18-O fractionation in ice cores. But I’ll bet no one is working on any such thing. If Rob Wilson or anyone else really loved their field of dendroclimatology and wanted to bring quantitative rigor to it, they’d be working on a project like that. Derive a valid temperature from wood from quantitative physical theory. Not doing one more hand-wavingly justified, speciously normalized, pseudo-temperature publicizing, tree ring study. Those things are nothing more than mathematically embellished propaganda for dendroclimatology groups — look guys, at what we did this time! Isn’t it fun!
In other areas of science, people who publish conflicting results argue about them in terms of theory until a clear winner emerges. And the winning idea is ultimately the one grounded most firmly in objective theory. Those spaghetti graphs all claim — each and every one — to tell a single story. However, they clearly have different story-lines, and the set we see doesn’t exhaust all the possible, equivalently pseudo-justifiable, story-lines. They are conflicting results that should cause the groups of origin to argue vigorously about who is right or wrong in terms of applicable theory. But that doesn’t seem to happen, perhaps because there is no applicable theory. Instead we get uncritical composite plots like Rob Wilson’s, or like the lovely IPCC hash that John A reproduced in #65, and various new proxy studies that merely present some new compilation of trees and cores representing yet one more soon-to-be-bypassed statement about past pseudo-temperatures. It’s a scientific scandal.
#120 — “I was trying to get numerous people here who seem bent against proxies to realize that they do have LIMITED value.”
But none of that value presently includes deriving a physically valid temperature metric.
So you don’t believe in empiricism; everything must derive from first principles. That’s fine. You do realize that you’ve just dismissed ecology as a scientific discipline, don’t you? Hokay. Carry on without me.
Gesundheit.
Mark
Bender is your
a comment to 142 and 143?
Re:#142
Pat says
I’d differ slightly and say that of course a valid proxy for temperature can exist without such a derivation (just as something can be true without a proof). That is, there’s a difference between saying something isn’t a valid proxy and saying it hasn’t been properly shown to be a valid proxy.
How seriously you take results using that proxy certainly depends on the degree of proper scientific justification supplied.
#144 — Ecological conclusions and research depend exactly upon evolutionary theory. For a very interesting evolutionary theoretic approach to a central part of ecology take a look at “Modeling Extinction” by Newman and Palmer
Empiricism lends us potentially valuable inferences — some we may not get any other way. It doesn’t produce conclusions. Nor do qualitative judgments produce quantitative results, intervention of math or no.
#147 – Agreed, Armand. Your distinction is between that which is and that which we know. All we can positively talk about is that which we know. We can speculate about that which is. But that which is, is always defined for us by that which we know. If a valid temperature proxy has not been determined for us by physics, we can’t now positively say that one exists even if later it turns out one is discovered to have now existed. There’s no such thing as precocious knowledge.
It was an analogy. I’m not rejecting anything about growing season length.
#143
Contrary to your skepticism, several people are working on isotopes in tree rings, as a cursory search of google will reveal. This is an actively developing field, which will allow multi-proxy dendroclimatology to develop (width, density and isotopic composition). Unfortunately it is not without its problems.
Your rejection of empirical science is unwarranted, a rhetorical device to avoid having to accept unpalatable truths.
#151 Science is theory plus empirical result. Empiricism by itself is only half of it. It’s not science, and purely empirical conclusions are not warrantable by objectively predictive test. If you don’t understand that, you understand neither science nor the difference between objective knowledge and everything else.
I’m happy to wait and see what comes out of isotopic analysis of tree rings. After all, I’m long time on record here at CA in favor of that approach. Until that or its equivalent appears, supposing that a physically valid temperature metric can be extracted from qualitative judgments about tree-ring causality is so much bushwah.
#152
The problem with theory is that there is seldom only one. Predictions from sound empirical relationships may well be better than those from one of several competing theories.
Pat Frank, could you help me out here?
“I don’t know why it’s so hard to understand that nothing is a valid proxy for temperature unless there is a rigorous derivation of a temperature metric from the observable. This is the case for oxygen isotope fractionation.”
I wondered about the oxygen isotope fractionation as a proxie. Doesn’t it depend on the assumption that wind patterns and ocean currents remain the same?
If the direction that the rain/snow clouds come to its place of entrapment from changes, shouldn’t the oxygen isotope fractionation?
Is there any correlation between the Earths plate tetonic activity and the O16/)18 ratio; particularlly during the isthmus of Panama?
Re: #129
It would be great to be shown just once that the scientists doing reconstructions, and particularly the dendros, understand just how readily numerous correlations can be obtained data snooping past data for stock picking schemes and how so many of these schemes fail so badly out-of-sample.
I could easily misread what Rob Wilson is saying, but from what I see it would appear that he is almost advocating data snooping as part of a selection system — and does so without mention or consideration of what this can do statistically to the correlations derived. In the mean time pardon me while I remain skeptical.
Steve M., thanks for your post. You say:
While I am happy to discuss this elsewhere, the reason that it is important to the current topic is that we have several “global mean temperature” dataesets, which show both different trends and different anomalies. Because “global mean temperature” has no agreed upon meaning, none of these datasets is theoretically superior to any other. This has a couple of effects.
1) People are free to choose which “global mean temperature” dataset they wish to use to compare and fit their proxy data … which in turn changes the result of the proxy exercise in whatever direction they may prefer.
2) It increases the uncertainty of both the data and the proxy reconstruction. For example, even using a single dataset, an average of all of the stations in the world shows a different trend than averaging the hemispheres individually and then averaging the two hemispheres. Which one is correct? We can’t say, there is no theoretical reason to prefer one over the other, but it certainly must increase the uncertainty of whichever one we may choose.
For example, were all of the various proxies in the graphic above done using the same “global mean temperature” dataset? I would doubt it, although I don’t know … but if they are not, it must perforce increase the uncertainty.
w.
Willis, I presume you are comfortable with statistical inferences about populations based on representative samples. Why are samples drawn from a field (temperature, whatever) any different? Two independent random samples will be close to one another in mean if the samples are sufficiently large. So why must there be a single standardized method of assaying a field or a population? As a counter-example, pollsters don’t formally define “public opinion”, yet they don’t doubt its existence. Are they foolish for daring to compare between polls?
Steve M is right that I’m not interested in metaphysical GMT mumbo jumbo. But if the core issue here is sampling theory, then that is not only interesting, but relevant to every aspect of climate science, including proxy-based reconstructions.
#153 — Predictions “from sound empirical relationships” are no more explanatory than a predition that the sun will rise tomorrow because it rose today and yesterday. You’re arguing for the general validity of inductive conclusions. 250 years ago, David Hume showed that method is useless.
The only time predictions from inductive knowledge pan out, other than accidentally, is when an unknown but deterministic process is operating. However, only the later appearance of a falsifiable theory tells us whether that is the case, or not. Inductivist predictions tell us nothing about cause, and those that pan out are indistinguishable from accidental correlation.
Predictions from competing theories are far more useful than inductivism, because if the theoretical predictions differ then one theory or another will be disproved. When competing theories all predict the same observables, they are still more useful than induction because they indicate the direction of objective knowledge. Induction — “sound empirical relationships” — tell us nothing about causality.
So far as inductive generalization is concerned, “God did it!” is as good an explanation as any other. From an inductive standpoint one may as well say that tree rings conforming to temperature are guided by angels, whereas the divergers show the interference of malicious imps. So, we should believe the corresponders because angels obviously tell the truth.
Absent a falsifiable theory of tree rings and temperature, by objective argument would you reject that explanation?
#154 — DocMartyn, you’re right on all counts, though I can’t offer an opinion about 18-O and tectonics. My point about 18-O was that the dependence on temperature of the oxygen isotope fractionation process itself is thoroughly grounded in physics. There obviously are seriously confounding problems because of stochastic climate processes, such as changes in monsoon rain patterns, etc. If someone were to normalize out these things, such as by finding an independent measure of paleo-rainfall, then the physics of isotope fractionation would allow a qualified 18-O paleo-temperature to be calculated.
This is not the case with tree wood, however. There is no underlying physics to tell us of a potentially quantitative relationship between ring-wood patterns and temperature. It’s all just handwaving about temperature limited tree lines.
#156. willis, I’m with bender on this. If one sticks to specific issues with the data and methods and stays away from the abstract arguments, then useful things can be said. Think about the construction of a price index. There are lots of warts on every index but you can’t say that just because there’s no absolute way to make a weighted index, there’s no such thing as inflation. And this is a much more complicated situation than anything we’re dealing with here. If people want to have a thread on the construction of consumer price indices, then we can host that. Or if people want to discuss issues in the GISS or CRU indices, that’s fine. But nothing abstract until concrete situations are fully itemized, please.
Pat Frank, you are a chemist seeking low-level chemical explanations for biology & ecology. That’s fine. Just don’t expect the field of environmental science & policy to wait for chemical (& biochemical & molecular) explanations of the world’s most pressing ecological phenomena. Personally, I agree with the bottom-up approach to science. Causative, all from first principles – it makes for wonderful models. The reality is you’d never get anywhere if you didn’t also work form the top down to try to make the ends meet.
I bet Francois Ouellette would agree.
You need to read more of the primary literature and less of that armchair science philosophy stuff. Let me compile a list of REAL papers for you, showing you what you’re missing.
Steve M. and bender, thank you for your comments. I see I have not expressed my thoughts well. Steve, you say:
I agree, I have no problem with that.
My point, which I clearly didn’t get across, is that if we take three different price indices, along with a single proxy (say GNP, or whatever), if we use the proxy to project the price indices backwards in time, we’ll get three different answers. As I said, this must increase the uncertainty of any single historical reconstruction of prices.
This is not “metaphysical GMT mumbo jumbo” as you seem to think, bender. It is simple math. The three price indices will give three historical reconstructions. This has to increase the uncertainty of any one of the three reconstructions.
Now, if you don’t want to deal with that mathematical uncertainty in this thread, that’s fine.
But it’s not metaphysics … it’s mathematics.
w.
#161 — Bender, when you make your usual strong arguments here, they’re always in terms of fact, theory, and methodology. I always learn something from you, then. Arguing by ad hominem, as you do in #161, is unworthy of you. It shows you haven’t a strong argument. If you did, it would be to the point.
Science is theory and result, and nothing else. The strength of that observation resides in the contents of science journals, and not in the opinions of people you find respectable. There’s nothing wrong with empirical studies. But those who do them should be modest in their conclusions. Dendrothermometry has no causal physics to bring analytical rigor to the conclusions promoted in its name. The conclusions do not reflect analytical modesty. They reflect promiscuity. You, among the best people here, should understand that. My argument, unlike your #161, stands on its own merits.
I’d bet John A would agree. 🙂
#162 — You’d only need to show me one paper, Bender. The one outlining the rigorous physical theory that predicts a temperature metric from a tree ring observable.
Re #164
I would agree with you Pat but absent an argument from theory, my agreement would be merely an opinion. Bender has enunciated, in my opinion, some commonly held misconceptions about sampling of thermodynamic systems which would be interesting to consider.
I am going to write up my own understanding on temperature on my own blog. Because I’m away on business in Europe during the week, this will have to wait until next weekend for me.
#83
Let’s see… We need to find a case when these would be valid error bars. First we need to note that they were computed in temperature domain, not in the proxy domain ( doh.. ). Secondly, in the case of ICE, the prior knowledge of past temperatures must be correct. But the most important hidden assumption is that calibration errors are completely neglected.. They kind of assume that calibration period is infinitely long. And all this in the univariate case, equations get bit more complicated with multivariate case, but I’ll try 😉
It seems to me that we need to go through same kind of discussion as Hoadley, Krutchkoff, Williams et al did in 60-70s. Confidence sets in linear calibration problems.
#157
The problem is that there is strong evidence that the climate samples are neither random, nor representative.
Re 154:
Could a biologist enlighten me? When a plant metabolises CO2 it preferentially selects 12C as it dislikes heavy C isotopes. Does it similarly select against heavy O isotopes?
C4 metabolism is less C isotope fractionating. Is it similarly less O isotope fractionating?
What are the implications of the above questions to detecting temperature data from historical organic material?
JF
I agree that THAT is not mumbo jumbo. (Don’t put words in my mouth.) It’s some of the other stuff I was referring to. I thought you were trying to go there. Maybe you weren’t. Sorry.
By calling you a chemist? And outlining your domain bias? I thought by clarifying your POV I was elevating your argument – from one of unhealthy skepticism to one of exceptional idealism, in terms of your expectations for a reductionist model of the biological world.
Re #167
UC, can I ask how close you are to publishing something definitive on uncertainty & error bars?
[snip- no more on GMT please]
Let me argure just a bit on Willis’ side of the discussion. The biggest, though not the only, aspect of varying temperature measurements is humidity. Thus heat input can take the form of increased temperature which is characteristic of dry land areas or it can take the form of increased absolute humidity, which is characteristic of ocean areas, with the caveat that the formation of storm cells result in varying areas of humidity. Given just the humidity vs temperature differential there is certainly the potential for different station groups to give different temperature reactions to an identical forcing.
Even more problematic is the fact that even the same putative suite of stations, existing station sets have undergone changes in the actual stations used (not to mention location movements of individual stations, which we assume have been properly accounted for), during the history of the suite. This last means that it’s entirely possible that we’re not seeing just time changes in an unchanging network, but a time changing network which may move from a more land based to a more sea based status or vice versa.
Though the above may seem to be applicable only to temperature measurement stations, it would also be true of sets of proxy data. We’ve had quite a bit of discussion here about how much certain sites are rainfall proxies vs temperature proxies, so if a proxy suite changes over time, say from only certain sites going back before 1500 say, then if a longer-term site tends to be more a rainfall proxy than a temperature proxy, then it might well reflect lower temperatures, everything else being equal, than a proxy which is a purer temperature proxy but which only goes back 500 years instead of 1000. This could cause a warming bias. Of course the reverse could also be true, but my point is to support the position that different temperature systems, and also different proxy networks, could both have biases, and as Willis says, could mean that the expected errors of any individual temperature network could be greater than otherswise expected after alternative networks are taken into consideration.
Re #127
Good point. Given the trend, that is a very short calibration period, even for a very simple two-parameter model. It brings TAC’s #117 argument about short time-series back into play. I had momentarily forgotten how short the calibration period is. I wonder: what is the longest-term calibration on record and what are its statistics? Presumably there are some 250-year records that would be long enough to get over the problem of the red noise trend caused by the LIA.
Re #176 LOL.
I am going to reply in unthreaded. DD makes a point that cuts quicker to the bone than some of these others. There’s something there to build on that gets us away from arm-waving – which I can’t stand – toward a more concrete test of the hypothesis.
Re: #166 Now who’s projecting?
#127
In #113 I mean that if calibration method overfits, there shouldn’t be divergence problem in the calibration period.
Correctly computed confidence regions may look very interesting. Disjoint semi-infinite lines perhaps 😉 In those early papers, the largest problem in CI computation seems to be the possibility of
, proxy = zero * temperature + noise in this context.
RE:172
Steve M,
That was poorly done, I have to to conclude that bender is above criticism in your eyes.
My comment had nothing to to do with GMT. I was criticizing bender for skating on comments made by Willis and Pat. Willis’s comment explained the the relevence of his earlier GMT comments and Pat’s comments had nothing whatever to do with GMT. I don’t understand your conflict with John A. over thermodynamics, but you need to resolve it. This issue should not bleed over to the point that anything to do with temperature is off topic.
You have spent great energy trying to get Phil Jones to come clean on the temperature record, why is GMT forbidden and the temperature record on topic? I don’t see the difference.
Cliff
Re #178 Don’t sweat it, Cliff Huston. I saw your comment (see my #175), thought it was funny, and have replied substantively in unthreaded #11.
As for the undiscussable topic, John A is setting up a forum for that. I’m not above criticism; you (and others) just keep posting OT in this thread. Don’t you get it? We have to take this elsewhere! The topic is New Scientist, Juckes, and Rob Wilson.
171,
I’ll try first to find an easy-to-read paper on the topic. At this stage this is just thinking aloud. Meanwhile, maybe someone knows how to compute this:
I’d like to see how MBH9x system predicts proxy readings, given the reconstructed temperature. With multivariate models this seems to be important diagnostic (*). Not sure if it is applicable to MBH though.
(*) Note that in univariate case this prediction will be perfect
#170 — Your argument turned on your characterization of me, Bender: “Pat Frank, your are a chemist [who is]…“. That’s an ad hominem argument (not necessarily an insult). You didn’t argue the point at all. You merely assigned some strawman POV to me, and then argued against that.
Make a convincing argument that science is not theory and result, and I’ll agree that you can make an inductivist conclusion, assigning your preferred explanation to whatever correlated data set you’d like, and call that science.
#181 is OT. Comment elsewhere and I will reply.
RE:179
bender,
Ok, I’ll give you a pass on everything GMT (even though your unthreaded #11 reply is to Dave Dardinger, not Willis), but that does not include a response to Pat Frank – if that is also off topic, do it on unthreaded and please leave a note here pointing there.
Thanks,
Cliff
I’ve been following this thread in disbelief and amazement. Steve has been deleting and snipping, IMHO very unfairly, several comments by several respected regulars. 😦
re #132 (TAC): Additionally, it has been argued even in statistics literature (see Shumway&Stoffer: Time Series Analysis and Its Applications: With R Examples, Springer 2005) that even the annual thing-you-are-not-allowed-to-say is integrated of order one.
#183 http://www.climateaudit.org/?p=1588#comment-110357
#184 It’s our own fault. We’re being very bad, breaking many of Steve’s guidelines.
It’s amazing how all the spaghetti lines up well during the calibration period. Is that due to cherry picking?
Jean S (#184), thanks for the reference to Shumway & Stuffer — I’ll take a look. My analysis of several records of the annual thing-you-are-not-allowed-to-say shows a fractional differencing parameter of approximately d=0.48. This corresponds to a stationary, though adventuresome (i.e. prone to prolonged excursions like ice ages and multi-century warm periods), stochastic process. A value of d=1 (a random walk) would be seriously non-stationary — is that reasonable? I know the TYANATS wanders around a lot, but it also “returns home to visit” pretty regularly, too.
This paper is for Pat Frank, who in #165 seeks one low-level explanation of the relationship between tree growth and temperature (but not because he’s a chemist). This is interesting because Picea glauca in northwestern Canada is one of the chronologies that Rob Wilson & Steve M have corresponded on.
Danby, R. & D. Hik. 2007. Responses of white spruce (Picea glauca) to experimental warming at a subarctic alpine treeline. Global Change Biology 13: 437.
Abstract:
From 2001 to 2004 we experimentally warmed 40 large, naturally established, white spruce [Picea glauca (Moench) Voss] seedlings at alpine treeline in southwest Yukon, Canada, using passive open-top chambers (OTCs) distributed equally between opposing north and south-facing slopes. Our goal was to test the hypothesis that an increase in temperature consistent with global climate warming would elicit a positive growth response. OTCs increased growing season air temperatures by 1.8°C and annual growing degree-days by one-third. In response, warmed seedlings grew significantly taller and had higher photosynthetic rates compared with control seedlings. On the south aspect, soil temperatures averaged 1.0°C warmer and the snow-free period was nearly 1 month longer. These seedlings grew longer branches and wider annual rings than seedlings on the north aspect, but had reduced Photosystem-II efficiency and experienced higher winter needle mortality. The presence of OTCs tended to reduce winter dieback over the course of the experiment. These results indicate that climate warming will enhance vertical growth rates of young conifers, with implications for future changes to the structure and elevation of treeline contingent upon exposure-related differences. Our results suggest that the growth of seedlings on north-facing slopes is limited by low soil temperature in the presence of permafrost, while growth on south-facing slopes appears limited by winter desiccation and cold-induced photoinhibition.
188: Another unenlightened chemist might observe as follows: What does this prove? As you well know, the response of plants to temperature is well established. What we don’t know, as you keep pointing out, is the interaction of all the other variables with temperature and with each other. Under ideal conditions, the theory works…
Re #189 So you’re conceding the point, and shifting the topic? Excellent. That’s progress.
#141, Michael Jankowski:
Yes, the text gets quoted, but it’s the graphs that become “iconic” and capture the public’s imagination. The original graph looks almost like it was prepared without any confidence limits, but Mann added an arbitrary standard of + / – 0.4 degrees C to make a reviewer happy and get published. It looks like an afterthought; but became Mann’s defense when he was brought before Congress. Even if the “confidence limits” only became part of Mann’s claims post facto it’s important to hold Mann and those who would use his work within Mann’s publicly proclaimed “confidence limits”, lest policy makers and members of the public be even more misled by extraordinary claims.
[snip -please stop trying to get the last word in on global mean temperature issues]
Re: 118 — I’m for having a place off CA to discuss it, have done so many times on other blogs. But it is likely to become a thread of those who understand therm’s, and those who are ignorant of it.
M&M found bad statistical practice of a bad [snip] statistic. M&M deserves accolades, because even after creating a statistic, some climatologists and much of the team blew it statistically, and continue to do so! In addition to learning a bit of R and gaining statistical depth, that’s what keeps guys like me coming back!
Actually, I can even see a rationale to select only those trees that show a correlation between LOCAL temperatures and temperature during the calibration period. The problem is in demonstrating that that relationship holds during previous centuries. The out-of-sample checks so far indicate a lot of “divergence” problems.
re #187: Nothing spectacular in Shumway&Stoffer, TYANATS (GISS) is just an example data set there. Yes, a random walk is non-stationary.
Gavin Schmidt at RC says a common fallacy is that errors do not accumulate in climatological forecasts. Hmmm. If the errors do not accumulate, but get folded back into the circulation (and weeded out, when “non-physical” outcomes are realized), then maybe the circulation is not as fixed as GCMers think it is. Maybe the circulation shifts when the exponential growth in deviations accumulates to exceed some threshold? This would imply that structural features of the circulation cannot be taken as fixed, and that the circulation itself is a sort of random walk – even if the temperature field, over a short time-scale, is not.
We talk a lot about red noise. How about 1/f noise? Is that a more appropriate null model?
Re: #188
bender, how do they separate the wind effects from the temp effects of the OTCs?
re: 192 — sorry, but that is not my intent. Any sense of humor out there?
Hmmm … how about the even more neutral, “Mr Eschenbach — hold post 6 (decimal) = ??? (binary) for the off-CA discussion? I have.” 😉
Koutsoyiannis and Montanari (here) have just published another interesting paper that is relevant to the discussion here: “Statistical analysis of hydroclimatic time series: Uncertainty and insights.”
The abstract states:
#188 — Bender, from the abstract that paper doesn’t relate at all to the point of my objection to current dendrothermometry. The issue is not whether tree growth responds to warmth, or even whether tree rings are specifically sensitive to temperature when all else remains constant. The issue is whether anyone can take tree ringwood and determine a specific temperature number from it. In principle, one can do that with 18-O from ice cores, because of a coherent physical theory of isotope fractionation. There is no such physical theory for tree wood. That’s the bottom line.
#195 — Bender, do you have a link to Gavin Schmidt saying that? Thanks.
http://www.realclimate.org/index.php/archives/2007/05/fun-with-correlations/#comment-33384
Article on divergence , this time tropical. My apology if this has already been linked.
It mentions something I’ve wondered about – the impact of cloud cover changes on tree growth. Cloud cover, while somewhat related to temperature and precipitation, is independent enough to constitute yet another variable in tree growth. It is possible to have increased cloud cover without changes in average temperature or precipitation.
I’ve been trying to keep an open, undecided mind on the use of tree rings to derive temperature patterns, but that is becoming increasingly difficult. Trying to deconvolute a temperature record from the various and poorly-understood variables looks to me, to an increasing extent, like wishful thinking at best.
Thanks, Bender. 🙂
Hi
a lot of responses about a figure that took me about 10 minutes to draft. Just remember I was asked to plot the ‘swath’ of available NH reconstructions – that is all. I will not make any more comments about it except say that #36 states how the 2007 figure was derived if Steve is having problems drafting his own version.
Pat Frank – isotope dendroclimatology is really coming of age at the moment as the cost and speed of deriving isotopic measurements from tree-rings are coming down. Despite your own arm waving of the superiority of this approach, there are still many problems. This is outside of my field of discipline so I will not comment further. I guess however, when a C13 or O18 reconstruction agrees with a ‘standard’ RW and or MXD derived version, you might believe them?
The mountains were great – the weather was great and I even sampled a new pine chronology – you know – all part of bringing them proxies up to date.
Rob
Which ones: those that diverge, or those that don’t, positive responders or negative responders? Seems to me an impossible task.
Mark
Re #202
Way back in the mists of time on this blog, I speculated that the key missing parameter that would affect tree growth would be relative humidity. It seems to me sensible that trees grow most strongly when the relative humidity is high, and would be relatively insensitive to temperature.
So a general drying in the atmosphere (which happens during cold periods) would produce less tree growth and smaller treerings than during warm periods. It is increasing relative humidity that would cause trees in dry areas to put on growth spurts.
Increasing carbon dioxide would work in the opposite direction, allowing trees in marginal regions (like bristlecones) to grow strongly by reducing water loss due to respiration.
Of course, relative humidity is weakly related to temperature at the best of times.
Here’s one NH reconstruction:
The data is from Juckes et al, JBB INVR reconstruction.
Note1: CIs are probably underestimated, as I used 2X calibration residual std. I didn’t smooth the data, because I generally don’t want to use any prior information of past temperatures.
Note2: Juckes INVR would be maximum likelihood estimator, if proxy scales and noise covariance (between proxies, zero between years) would be known and used in the estimator.
(Wrote this post in less than ten minutes, thus it is probably full of mistakes 🙂 )
Maybe there would have been fewer comments if you had taken longer and had provided an archive showing what you did. I don’t understand why all the proxies supposedly centered on 1961-1990 have values below 0.
Re: 208: Same reason why everyone is smarter than average 😉
re 209
No everybody thinks they are smarter than average. The fact is that 50% is less smart than average.
This reminds me of comments by Mann (was it him?) regarding his initial shock that the hockey stick got so much attention yet it was only a minor piece of the overall claim. If it is so minor, why do these guys keep putting such graphics together allowing them to be headlined?
Mark
Re: #210
OT, but I can’t resist. It depends on the task. If the task is relatively easy, and/or everybody does it like driving a car, most people think they perform better than average. For a very difficult task, the response is opposite, most people think they perform below average.
See here for example: http://www.chicagocdr.org/cdrpubs/pdf_index/cdr_560.pdf
Re #211 Back on topic:
It was a shock because they weren’t used to real scrutiny. It’s still that way to some degree. Apparently.
Re: Hans, maybe the sarcasm was not transparent. Here is something very important, however: I think you are thinking of the median rather than the average. — Sinan
No kidding, bender. These guys don’t seem to realize that the graphics, not the numbers, are the visual “sound bites” that drive the public’s perception of science. The public doesn’t notice that all of the diverging tree-ring reconstructions are cut off at some arbitrary date. Along come those that do notice those things, those who attempt to point out the numbers are flawed, and suddenly it’s “we didn’t realize it would get such attention!” Of course such things should get such attention, it’s just unfortunate the public doesn’t see that.
Mark
re 214:
Mean, median and average are very confusing concepts for the average (mean?) non-native english statistical amateur like me.
#204 — I obtained a research MS degree in physical organic chemistry prior to a Ph.D., Rob, and so my comments about the quantitative superiority of a 13-C isotope effect temperature metric over qualitative judgments concerning temperature limited growth are not “arm-waving.” Don’t take my word for it, though. Temperature Effect on Reaction Rates.
Here’s a shocker: Given the quadratic response of tree growth with temperature, a standard MXD or ring width approach will agree with a quantitative 13-C isotope effect temperature (assuming it’s possible to extract one) only on the positive slope of the width or density temperature growth response, if at all, and not anywhere else. Perhaps explaining the divergence problem.
Re 202 & 206 and what about wind-run?! – this not only impacts on evaporation/moisture stress but also [as the the abstract in #188 highlights in an extreme way] impacts on the temperature in a micro environment -Were the windruns the same in the little ice age or MWP ? Or were the prevailing wind directions the same?
As a posting test, here is a link to a paper that illustrates how well physiological process models perform (ARCWheat is the Rolls-Royce of the mechanistic wheat model world)
http://www.esnips.com/doc/bd6b37c7-bba3-4201-b5bd-b557b661a1f5/Landau-arc-wheat
#206,
Perhaps a simplistic observation but one that springs to mind. Growing up in OH, we were a part of the “corn belt” of the Midwest. In summers of higher RH, you could almost “hear” the corn grow, especially at night. This despite more clouds and less sun. In years of lower RH, you couldn’t. The old saying knee-high by the 4th of July was generally true, but in years of higher than normal RH, the corn would be more like waist high by early July, even if there was not more precipitation.
Read this
http://www.pnas.org/cgi/content/full/104/11/4249#otherarticles
re #216 (Hans): Don’t worry, they are be very confusing concepts for many people, see here for a nice description. Even highly regarded statistics references contain mistakes about these concepts, for instance, Kendall’s advanced theory of statistics (Volume I) contains the following totally false statement:
Try to construct a counter example! (Hint: starting from the Gaussian pdf, “cut&paste” in order to create a distribution with the quantities in the following order: median, mode, mean)
[Reference for this observation/finding: Dudewicz&Mishra: Modern Mathematical Statistics (1988), p. 216. They refer to K&S 1969 edition, but the statement is still there in my 1994 edition]
#221: As easy as 1-2-3 🙂
0 0 0 1 1 1 1 2 2 2 2 2 25
Median: 1
Mode: 2
Mean: 3
Hans, don’t feel bad, the notion that the mean (average) of a set of numbers corresponds to the center of the series can be hard to let go (well, it does correspond to a center, the center of gravity).
Jean S., thank you for pointing out the error in Kendall’s book. Goes to show you how carefully(!) I read.
Aaaaanyway. Back to the regularly scheduled programming 😉
Sinan
That’s why Gaussian distributions are so nice to play with. 😉
Mark
I am puzzled as to what purpose spaghetti graphs serve. Which of the graphs are relevant and which aren’t? If I attach equal relevance to all of the graphs then I see a departure of about 0.7 degrees on the positive side and about -0.9 degrees on the negative side (hard to determine as the scale changed from one tick per degree to two below the -0.8 departure). So what does this tell me? There is less departure on the positive side than on the negative when compared to 1961-1990?
“Spaghetti” is an epithet suggesting that all multiproxy reconstructions are fundamentally incoherent, and that any superficial coherence is due to arithemtic non-independence among series (e.g. sharing of bristlecone pines). But to answer your question, their purpose is to illustrate a data-based “consensus” – convergence on truth through independent approximation.
A scientific graph, however, would show error bars, and if this were done on spaghetti graphs what you would see is a consensus … not on trend, but on massive uncertainty. Instead, when there are no error bars – which is often what you see in the derivative literature (e.g. newspapers, policy talking points) – what you see, somewhat ironically, is a total lack of consensus. Skeptics love “spaghetti” graphs in part for all the irony they possess. Also, they stand in utter distinction to the IPCC TAR hockey stick.
If Briffa’s global dimming effect is responsible for 21st century diverence (of tree rings from the instrumental temperature record), then, being a staunch uniformitarian, what does he have to say about the 10th century?
#226. The etymology of the term “spaghetti graph” came up once before. As I recall, someone found a prior usage in another discipline, but own usage was independent and appears to have been the proximate etymological origin in much paleoclimate usage.
RE: #202 – I would surmise that places that have a “dull” (typically Marine West Coast) climate, that have reasonably even annual rainfall, where the soil never freezes, have soils with some level of drainage, have lots of cloudiness during all seasons, and never incur truly high sun angles, probably support a number of tree species which would tend to express annual average temperature as tree wing width. Humid and perhumid continental areas a couple of hundred to a hundred miles south of the the northern limits of tree growth may also be candidates, but I’d be concerned about soil freeze thaw variation causing distortion. For example, a year that had an intense soil freeze followed by a very warm summer may not provide a wider ring. Elsewhere, I see numerous issues. In arid and semi tropical climates, tree rings are dominated by precipitation. In tropical climates, there is not much of a meaningful temperature variation, and trees there are responding to things like precip, cloudiness and other factors. In most continental climates, there is too much annual and lesser period variation in all factors to know what the ring width is a signal of.
Re #228 The award for creative etymology has to go to the inventor of the phrase “Mann-o-matic”. Every time I say it or read it I chuckle. Maybe one day I will invent the bender-o-matic. heh heh.
That would be Ross McKitrick.
Mann’s origination of the term “Hockey Team” was also felicitous.
#230 — “bender-o-matic” lacks the poetic rhythm of Mann-o-matic (which also carries echos of those 1950’s TV commercials for eccentric kitchen gadgets always purveyed by Popeil). So, if you ever develop a method for levitating conclusions, bender, may I suggest calling it ‘the bender suspender’? That has a more attractive alliteration. 🙂
benderithm ?? (if Al Gore invented the algorithm, why not?)
bend-o-matic.
Mark
It didn’t exist.
Mark
bend-o-matic! that’s it! now i’m gonna HAVE to go measure me some treeline conifers so that I can show y’all how the bend-o-matic brings the mwp back. (the trick is in bending the response curve the way it’s supposed to be: nonlinear upside down quadratic, and then blending in the true moisture response.)
mmmm, cherry.
#226
Thanks Bender. I appreciate the response: very Benderian.
I assume you mean skeptics love to refute the validity of spaghetti graphs. I suspect New Scientist did not publish it as skeptics.
I accept by your description and others on this blog, the Wilson spaghetti graph is not “scientific” as it does not show the error bars. Then I might assume it was meant for a layperson such as myself. When I look at the graphs, I would not come up with the heading, “… all suggest that it is warmer now than at any time in the last 1000 years”. I would be more inclined toward, “temperature departures greater on the negative side than positive when compared with 1961-1990”.
It seems the New Scientist graph was not well thought out. Even Rob Wilson does not seem too excited about it. I do think they should at least correct the lowest ordinate from -1.0 to -1.2 as it would tend to exaggerate the curves below -0.8 (which further contradicts their title).
Re #237 Skeptics love the graphs themselves because they are self-refuting. i.e. Warmers think they mean one thing, when what they actually mean is quite the opposite. (What is the martial art where you defend yourself by turning the offending force back on itself? Very satisfying.)
It’s just like in Back to the Future. It’s been erased….from existence.
#238
Aikido.
Benders are capable of bending anything, even unbendable bars! John Brognell (is that spelled right?) of numberwatch calls himself “your bending author.” Maybe there’s a connection. 🙂
Mark
It would appear that the interest in dendrochronology, and specifically Rob Wilson’s approach to this science, got sidetracked by other discussions including some observations on the philosophical basis of scientific inquiry and what that inquiry emphasizes. I thought what was being discussed was more on emphases (empirical and experimental versus theory and hypothesis or emperical versus semi-emperical as described in the excerpt below) than basically different approaches. I’ll give my take on the subject from the excerpts from Wikipedia below and let it go at that because I would rather get posters’ takes on, specificaly again, Rob Wilson’s approach to using TR and MXD for temperature reconstruction.
The tendency in these observations posted to a blog are that they can be over simplistic and with that precaution in mind I will attempt to summarize what I find lacking in Wilson and company’s approach. The item with which I believe most here would agree is that Wilson and company do not strictly adhere to when applying the scienticfic method of inquiry is:
What I see as a perhaps a lacking to which posters are in less agreement is:
Here, I think the lacking involves using hypotheses that would provide criteria for selecting trees, months, lags, TRs, MXD, QC indexes, etc. for temperature reconstructions prior to testing and then using all the data to test whether a model can predict temperatures in the instrumental time period using both calibration and validation. What bothers me about the current approach as I understand it (I will stand corrected if someone provides better information) are that the criteria selected are based on the results of the testing after the fact and the dangers of data snooping are completely ignored. In my view, if these dangers were well understood the dendros would be making a major effort to compare the calibration and validation periods for a reduction in correlation on going from calibration to validation, and finally, since data snooping can always look ahead to validation performance, they would want to be making every effort to look at out-of-sample periods using the same criteria that was applied to the original calibration and validation periods.
http://en.wikipedia.org/wiki/Empiricism
http://en.wikipedia.org/wiki/Scientific_proof
Did anyone mention that the caption says “all suggest that it is warmer now than any time in the past 1000 years” (emphasis mine) yet both the Esper 2002 and Moberg 2005 peak about 1000 years ago and do not return, even today? Perhaps the caption should have been “nearly all…”???
Mark
Re #241 Brignell.
Does anyone know where to find a good set of spaghetti graphs depicting the individual proxy records (i.e. one line per tree or other series), each shifted and scaled to “best-fit” the instrumental record over the period of overlap? I recall that somebody did this (SteveM?), but I don’t recall where. Thanks! 🙂
I realized it might be Brignell after I posted (should’ve just gone to his website)… thanks bender.
Mark
One Trackback
[…] very practical too! Visit Climate Audit and SDA for a recap of the comments. It seems that little is agrred to, except the […]