Rosanne D’Arrigo once explained to an astounded National Academy of Sciences panel that you had to pick cherries if you wanted to make cherry pie – a practice followed by D’Arrigo and Jacoby who, for their reconstructions, selected tree ring chronologies which went the “right” way and discarded those that went the wrong way – a technique which will result in hockey sticks even from random red noise. Her statement caused a flurry of excitement among Climategate correspondents, but unfortunately the NAS panel didn’t address or explain the defects in this technique to the lignumphilous paleoclimate community.
My long-standing recommendation to the paleoclimate community has been to define a class of proxy using ex ante criteria e.g. treeline black spruce chronologies, Antarctic ice cores etc., but once the ex ante criterion is selected, use a “simple” method on all members of the class. The benefits of such a procedure seem obvious, but the protocol is stubbornly resisted by the paleoclimate community. The PAGES paleoclimate community have recently published a major compilation of climate series from the past millennium, but, unfortunately, their handling of data which goes the “wrong” way is risible.
The PAGES 2017 collation is a successor dataset to the PAGES 2013 collation, aspects of which I discussed a few years ago. Not included in my previous discussion was their North American tree ring collection, which stubbornly included the same stripbark bristlecone chronologies of Mann et al 1998-9, while claiming to be “independent”. In total, there were 146 North American tree ring series in PAGES2K (2013).
PAGES2K (2017) contains almost exactly the same number (150) of North American tree ring series, but, if you look at the second tab (Table S2) of its Supplementary Information – an excerpt of which is shown below, one series after another was rejected because it had a “negative relation to temperature”
The next table provides an inventory of changes between 2013 and 2017. Of the 146 North American tree ring series used in PAGES2K, 84% (!?!) were discarded because they had a “negative” relation to temperature. Only 23 series were carried forward (nearly half of which were stripbark bristlecones or foxtails). Replacing the 123 discarded series were 125 new series all of which were said to have a “positive” relation to temperature (though many were admitted in the SI to have “no low-frequency signal”.
Although only 15% of the PAGES (2013) North American tree ring data had a positive relation to temperature when reconsidered in 2017, 100% of the series added in 2017 supposedly had a positive relation to temperature. Given the prevalence of negative relations in the 2013 version, one can only assume that the 2017 additions were screened from a considerably larger dataset and that numerous series were examined but not selected.
In the union of the 2013 and (screened) 2017 additions, there were 273 North American tree ring series, of which 45% went the “wrong” way (and were discarded in 2017), while only 55% went the “right” way. To achieve a 50-50 split, it merely requires that 27 series were screened out in the selection of 2017 additions. My guess is that many more than 27 series were screened out in the 2017 addition.
One of the PAGES 2017 coauthors, Julian Emile-Geay, claimed that the new study shows:
But what this latest PAGES 2k compilation shows, is that you get a hockey stick no matter what you do to the data.
If (1) you start with an extended dataset half of which goes up in the 20th century and half of which goes down and (2) from that extended dataset, select only those series which go up, one trivially will get a hockey stick with simple composite methods (which do not assign negative coefficients i.e. flip the underlying series), [added] as illusttrated by the following cartoon (h/t CTM):
As Rosanne D’Arrigo explained years ago, you have to pick cherries if you want to have cherry pie. Nothing has changed.
There are other points of interest in the PAGES 2017 proxies which I’ll try discuss if I have time, inclination and energy.
Climate Audit 
272 Comments
Steve,
Glad to see you back. Give them hell! I tried to do my part while you were gone, but they are squirmy. 🙂 I had to get them down to yes and no questions and to use mathematical norms to quantify their curves.
Jerry
Steve,
Welcome back !
I’m hoping you might take a look at:
Fire Ecology Vol. 5, No. 3, 2009
doi: 10.4996/fireecology.0503120
Swetnam et al.: Fire History of the Giant Forest
There seems to be a GREAT deal of giant sequoia tree ring samples and data
sitting in the lab in Arizona, and except for Swetnam, Brown, et al. has
languished there for over a decade.
You’ll note in the study, Swetnam isn’t fond of bristlecone pines as
indicators of temperature shifts in the NA west.
Some of the cores and sample slabs go back over 3,000 years.
Are there no brave doctoral candidates who want to brave the cold storage
in Arizona to measure rings for carbon or beryllium isotopes, calibrate
volcanic ash inclusions, or just count the rings?
Another pearl from Julian Emile-Geay @ strange weather: “But that would take away from the main point: I promised you hockey sticks, and hockey sticks you shall get.”
It doesn’t help to take quotes out of context. What the quote is referring to is a potential digression in the previous paragraph:
“Much could be said about the cat-herding, the certifications, the revisions, the quality-control, and the references [..]”
The “promised” hockey sticks refers to the content of the article, which is hockey sticks, not any kind of predetermined synthesis of the data.
I just wrote a polite comment @ “Strange Weather”/ The Hockey Stick is Alive; long live the Hockey Stick notifying readers about Steve’s news Climate Audit’s post above. As “reward” my comment was removed 2 hrs later: par for the course in data massage land. I call it “Strange Science”.
I wonder how many other “Climate Studies” suffer from this cherry picking syndrome?
Probably most of them. p-hacking is an important ‘scientific’ methodology in pseudo sciences.
I greatly appreciate being able to read terse and pointed analyses that shed light on the validity of efforts to measure past climate. They do not seem to learn.
seems like Steve’s reading was very terse as well, since he completely misread what we did.
JEG: I believe it has been a long time since you commented here. It seems to me that your only defense of your selection method is that it adheres to Steve’s requirement that the inclusion or exclusion of the particular proxies be independent of any analysis of dendro variables with temperature.
Beyond the above, I think I will wait for Steve’s response…it should be very interesting.
JEG, you people use strictly statistical methods and claim a physical result.
Your entire field is no more than pseudoscience.
(Best put at the outset for this one) JEG: I used to be a deni@r until I confirmed CAGW on my own. I found a few dozen rocks in my yard, cracked ’em open and saw lines. Some went up, some went down, some were all squigely. Tossed back the non-temperature related rocks, and all 5 of the rocks left were temp related. Confirmed this was global by graphing my results from the 5, flat lines for the last 1000 yrs of rock temp lines, with a swoosh up in the 20th century rock lines. In all 5! Exactly what they found in temp related trees! So I know it’s worse than we thought and I had to join the cause. They were very strict on the admission at first, but were quite welcoming after they learned my data was lost (put the rocks in my new koi pond I could afford after the NAS grant!).
Sorry, best put “sarc” at the outset…
By rejecting so many series that go the “wrong” way they tacitly admit that much more than temperature affects tree ring metrics. Until and unless they can clearly determine the relative contributions of various environmental components to each series, they are practicing more art than science in the interpretation.
Reblogged this on ClimateTheTruth.com.
At first glance it seems that the PAGES team either still don’t understand the screening fallacy, or do understand it but are trying to hide it, despite the Gergis et al fiasco five years ago.
It’s not just the North American series that are screened. Several records in Africa, the Arctic and Australasia are rejected because of their “unclear relation to temperature”.
I’m looking at how PAGES2017 handled the Gergis proxies. Not pretty.
I like cherry pie, but it is useless as a model of climate. Great to have Steve as a steadying influence and holding the feet to the fire of those who think science is a political hobby.
“But what this latest PAGES 2k compilation shows, is that you get a hockey stick no matter what you do to the data.”
Let me fix that for you JE-G:
“But what this latest PAGES 2k compilation shows is that when you discard/ignore the half of the available data that shows no hockey stick result, you can tell people “you get a hockey stick no matter what you do to the data.”
This is a useful study in that it shows that no relationship between tree growth and temperature can be established through the study of tree ring widths. Plausible though it seems that a positive correlation should exist, the data show otherwise.
This, of course, is not the conclusion intended by the authors. If this study used public funding, the public has finally gotten some useful science from the tree-ringers.
One fundamental issue is overlooked: trees grow significantly better under elevated CO2. So how does this new PAGES selection know that it did not pick CO2 proxies instead of temperature proxies?
Same problem with CO2. One half of the data show no positive correlation of growth.
Very good point, Hans. Any valid calibration of tree ring growth to temperature must incorporate CO2 as an explanatory variable. Because of the strong correlation between CO2 and instrumental temperatures, this will likely reduce the partial correlation of temperature and tree ring growth and could even render it insignificant.
Steve has invented a word. Lignumphilous. No result from Google besides Climateaudit.
Is this related to the game Lignum?
Look up “Stick” in Latin
Since the stick in question is a “ho[c]key stick” possibly “clavaphilous” would be more appropriate. “Clava” translates to “cudgel” among other similar meanings.
BTW, the square braces around the “c” are humour not a correction.
Lol.. but “philic” etc is Greek.. Not an expert, but google offers “xulon”. Xulophilic?
But could Steve be accused of Lignumphobia?
Lignumphilous would be closely related to Grantumphilous
Tweeted:
Julien Emile-Geay @el_nino_waves responded 12:38 PM – 12 Jul 2017
Responded: Enlighten us why “84% (!?!)” 2013 #tree #data with “negative” relation to temperature were replaced with “positive” tree data @ClimateAudit
Steve, I appreciate the publicity you’re giving us, but it would have been preferable to spend just a few more hours digesting what amounts to years of work. Dismissing it as “nothing has changed” shows all too clearly that you want climate scientists to respect you without doing them the basic courtesy of carefully reading their work. A few points to pierce through the cloud of confusion that you once again so artfully created:
1) one of the main points of this exercise was to include many archives other than trees. What do you make of Fig 8 (if you read that far), which shows that hockey sticks appear on composites based only on ice, lake, coral, and documentary records?
2) Every paleoclimatologist knows that there are multiple factors influencing tree growth; hence why it is so important to carefully select sites where temperature is the limiting factor. The basis for selection, obviously, was not whether trees exhibited a hockey stick shape or not; in cases where there are enough instrumental data, the criterion is whether they correlate well to that – too bad the temperature data show warming over the 20th century, I suppose.
Moisture is known to be the primary control at many sites, and in version 2.0.0 those sites were excluded so as to retain what is currently thought to be the sites where temperature exerts the primary control. I find it remarkable that you manage to complain about both the 2013 the 2017 versions, even though they used these different approaches.
Lastly, there is this pearl:
“My long-standing recommendation to the paleoclimate community has been to define a class of proxy using ex ante criteria e.g. treeline black spruce chronologies, Antarctic ice cores etc., but once the ex ante criterion is selected, use a “simple” method on all members of the class. The benefits of such a procedure seem obvious, but the protocol is stubbornly resisted by the paleoclimate community. ”
Has it ever occurred to you that paleoclimatologists might know a shred more about paleoclimate than a mining executive? Perhaps, having studied paleoclimate proxies their entire life, they understand that “Antarctic ice cores” isn’t a specific enough criterion to ensure that temperature is the main control on a proxy at the timescale of interest.
However, the data are all public, and so is the code, so unless you want to make a staggering display of bad faith, you can’t complain about obfuscation. I’m sure that you’ll find plenty to complain about, however. If it well-reasoned and backed by evidence, we’ll do our best to correct mistakes that might have slipped through. But please spare us the lecturing from the mountaintop – you need to crawl out from that puddle of paleoclimate ignorance first. However, while you’re there, please play with the data and tell us if you can find something that’s not a hockey stick!
(spoiler alert: using a completely independent approach using ALL tree-ring chronologies we can get our hands on, and a bivariate proxy model that takes moisture and temperature sensitivity into account, we also get a hockey stick. You’ll break your knee on it before the hockey stick breaks).
JEG Thanks for prompt response and Fig 8: https://www.nature.com/articles/sdata201788/figures/8 You state: “bivariate proxy model that takes moisture and temperature sensitivity into account, we also get a hockey stick.” “Moisture is known to be the primary control at many sites, and in version 2.0.0 those sites were excluded so as to retain what is currently thought to be the sites where temperature exerts the primary control.” Your paper states:
What “temperature” evidence were they selected to fit rather than moisture? How is this not just reinforcing the hockey stick presupposition?
How do we know that this is not CO2 fertilization rather than temperature? The only ref to CO2 is Stott et al. 2007.
I find no mention of “low pass” or “high pass” filter effects of paleo evidence vs modern instrumental data. (PS why the ad hominem attack: “than a mining executive” “crawl out from that puddle of paleoclimate ignorance”? Have you considered Steve McIntyre’s statistical expertise or his extensive examination of the statistics of paleo data?)
JEG,
So was your 2013 paper wrong? If not, why did you exclude 123 proxies from it in your 2017 paper? Steve , with his meticulous analysis
of many paleo data tweaking schemes, has shown far more mathematical and scientific credibility than the climate “scientists”
that foster questionable practices.
Jerry
Supplemental data explanation:
strange:
Sorry, the requested URL ‘https://images.nature.com/full/nature-assets/sdata/2017/sdata201788/extref/sdata201788-s2.pdf’ caused an error: Internal Server Error
John Bills Try accessing the Supplemental via the main Nature page for Pages2K http://www.nature.com/ngeo/journal/v6/n5/full/ngeo1797.html or the link to the supplemental information page http://www.nature.com/ngeo/journal/v6/n5/full/ngeo1797.html#supplementary-information
The paleoclimatological community cannot even now acknowledge just what selecting individual time series that conform with their hypothesis from a larger set of time series is: ex post selection of data.
From the article:
“Relationship to Temperature
Here we examine the extent to which the database as a whole captures the observed temperature variability at local and regional scales. We do so via correlation analysis, which makes the common assumption that the relation between the proxy value and temperature over the twentieth century is representative of the entire record (stationarity).”
I’m hoping someone can explain why the assumption of “stationarity” should be accepted as plausible, especially given the number of series that fail to correlate with observed temps in the last century. I understand that certain series were filtered out altogether, for particular, site-specific reasons, while others were multiplied by a factor of -1.
I can understand perfectly well why one might make such assumptions in making a best-case argument for the hockey stick — and I don’t mean that in a snarky way.
Still, how can it be assumed that in the entire pre-observational period that the proxies capture temperature rather than something else? Given the known lack of correlation for certain series in the observational period, how can you possibly hope to avoid false positives (for temperature correlation) in the pre-observational period? What am I getting wrong?
“What am I getting wrong?”
#####
Your viewpoint. You must not adopt such a critical mindset. You will never make an acceptable tree-ringer unless you leave off such hypercriticisms.
The global temperature anomaly certainly shows an increase since systematic thermometer records began in the tail end of the period of extreme cold known as the ‘little ice age’, but just as in the medieval warming period, local temperature series need not correspond to the global trend. They may indeed have a rising gradient, but they may also be indeterminate, or even show a cooling trend over a given period.
The question is whether you have calibrated local proxies against local temperature readings or global. Given that a local series need not follow the global one, then only choosing series that follow the global curve is tantamount to using the hockey stick as the selection criterion.
“The basis for selection, obviously, was not whether trees exhibited a hockey stick shape or not; in cases where there are enough instrumental data, the criterion is whether they correlate well to that – too bad the temperature data show warming over the 20th century, I suppose.”
This is the crux of the problem. You don’t actually know if these are actual “treemometers” … or if you just picked those series out of random noise that happen to match what you are looking for. The fact (or do you dispute it?) that a huge percentage of those trees you thought WERE “treemometers” just 4 years ago were not seems to strongly indicate that these series ARE just random noise.
This is the very simple point that, from my reading on this subject over the years, is NEVER addressed. If you could do so here, you might actually convince someone.
You have succinctly raised the most salient point; viz., how can any confidence be ascribed to a proxy selection criteria which utterly rejects over 100 samples deemed suitable just four years earlier.
Naturally, optimists could ascribe the delta to the self-correcting nature of Science. Yet, even this modest wit is left wondering how many rejected samples were excluded solely for deviation from brief instrumental records. Absent, a detailed explanation of exclusionary methods, we are left wondering if Dendrochronology is Alchemy disguised as Science.
I suppose we should give credit where due….at least there’s no decline to hide.
JEG,
With respect, re your words “sites were excluded so as to retain what is currently thought to be the sites where temperature exerts the primary control” –
This would be better with the change of one word. Then, “sites were excluded so as to retain what is currently KNOWN to be the sites where temperature exerts the primary control”.
Papers using ‘think’ are ok, so long as it is realised that they are but a stepping stone in the evolution of a hypothesis and that no ‘policy’ type conclusions should be put on them.
Papers using ‘know’ indicate that the coverage of extraneous variables is sufficiently under control to allow ‘policy’ implications to be drawn.
We cannot make policy on the basis of what people ‘think’ about data, without knowing.
BTW, while you are here, can you expound on why Australia is so under-represented in the PAGES work? It is a given that the Southern Hemisphere could do with more data, but in PAGES2K there was virtually nothing from the Australian mainland. Still to read the latest PAGES, hoping that there is more, but still very puzzled why Australia has been almost a blank map.
Geoff.
JEG,
one of the main points of this exercise was to include many archives other than trees
From the paper: The majority (59%) of the records are based on tree rings because they are annually resolved, precisely dated…”
What do you make of Fig 8 (if you read that far), which shows that hockey sticks appear on composites based only on ice, lake, coral, and documentary records?
It is a reasonable inference that the how the data was selected for the majority of proxies, trees, was applied to others.
Also,
a. Marine sediments and “documents” show a decline in the last fifty years.
b. Glacier ice show no warming for 100 years.
c. Lake sediment rise begins about 1750. No anthro acceleration.
d. Why use “historic” documents at all? 11 of the 15 are Chinese beginning around “1470.” River floods or something.
e. I think they are still using some of those very suspect Australian non-treeline studies. There is no basis for such science outside of Australian Climate Science, Ltd.
SteveM’s criteria method before-the -fact selection of proxies applies to all types of temperature proxies not just tree rings
If these factors are well known then following SteveM’s a prior criteria selection and using all that selected proxy data becomes an easier task. Different approaches are not the point here but rather using an a prior selection criteria.
The advice given by SteveM can be made without having expert knowledge but rather has to do with statistical advice concerning situations where a test cannot be repeated as in hard science. In these matters the expert knowledge required to find a prior criteria may well not include climate scientists who merely use the data to make temperature reconstructions . It has to do with the basic approach of a prior proxy selection that gets away from the HS biases.
Playing with the data is not the answer but rather the hard work of determining a prior selection criteria for temperature proxies.
I would think a spoiler alert should have a paper reference.
JEG might make a case for why some climate scientists appear to ignore the correct approach in selecting proxy data and avoiding the HS biases. They simply miss the point.
JEG, thank you for responding to Steve’s article.
“Moisture is known to be the primary control at many sites, and in version 2.0.0 those sites were excluded so as to retain what is currently thought to be the sites where temperature exerts the primary control.”
Having been raised in a family which once owned America’s largest independent dairy, I’d like to think I know something about plants and trees, so I am always puzzled when someone says temperature exerts the primary control, because different factors are all major contributors to plant/tree growth, water availability arguably being the most important factor facilitating nutrient uptake for a given non-extreme area.
One very simple example – forty years ago, two trees were planted 100 ft. apart, one in my lawn and the other in an unwatered area. The tree in the lawn is 90 times more massive than its twin.
I am only concerned about the math when the subject matter makes horse-sense, so temperature proxies based on tree ring width don’t don’t pass muster.
All you are saying is that to prove your hypothesis you use only the proxies that prove your hypothesis. You can dress it up in all sorts of ways, but since you are trying to reconstruct a variable, you cannot assume you know what that variable was. And of course you don’t what the other variables were either, so your rejections make little or no sense either.
This is no different from the majority of papers in climate science which produce a model to show what will happen if the model assumptions prove to be correct and then claim that the model shows their assumptions are correct.
Climate science seems to be stuck in a loop of circular reasoning.
Julian, thanks for commenting.
Because I wrote first about North American tree ring proxies, please do not assume that this is the only issue that I noticed. There are many other issues, but each one takes time to write up.
As to the issue of choosing series ex post, this has been a longstanding concern, originating with my first encounter with Jacoby and D’Arrigo, discussed in one of the earliest Climate Audit posts in Feb 2005 here https://climateaudit.org/2005/02/06/jacoby-1-a-few-good-series/. Jacoby and D’Arrigo had collected data from 36 northern sites, from which they selected the 10 “most temperature sensitive”. They purported to test for statistical significance but did not test the effect of selecting 10 of 36 series.
Jacoby and D’Arrigo archived data for the 10 series that they used, but refused to provide me with data for the other 26 series when I requested it.
In 2004, Climatic Change had asked me to review a submission by Mann. In my capacity as a reviewer, I asked for the data which Mann had refused to provide me as a critic. Schneider said that no reviewer had ever asked for data in 28 years of running the journal. I was unimpressed with this precedent and re-iterated my request. Schneider said that he’d have to consult with his editorial board to establish a policy; I said fine. Ulitmately they agreed that authors would have to provide data. Mann continued to refuse and abandoned the article.
Under the new policy, I requested data for the other 26 series for JAcoby and D’Arrigo, which had been published in Climatic Change. Schneider made a halfhearted effort to get data from Jacoby, who sent the remarkable refusal letter replicated in the CA post linked above.
From a statistical perspective, you’re doing exactly the same thing as Jacoby and D’Arrigo in your tree ring data – perhaps even worse. If you’re screening series, you need to keep track of how many series you tested and rejected. Unfortunately, with JAcoby and D’Arrigo, who are important contributors to your project, we do not know how many series were thrown out because they didn’t have the “Jacoby signal”.
If you use biased statistical methods, your results become untrustworthy. It seems quite possible to me that the modern warm period is somewhat warmer than the medieval warm period, but you cannot demonstrate this with ex post screening.
“Schneider said that no reviewer had ever asked for data in 28 years of running the journal.”
Here we have all we need to judge the field of climate science. What an admission!
Mann, Jacoby, d’Arrigo, and seemingly, JEG fail to understand how damming it appears in the eyes of others when they refuse to be forthcoming with their data selection methods, withholding as they do the datasets which they rejected. I am reminded of Phil Jones of CRU, East Anglia, when he responded to Steve Mc’s request “Why should I provide the data to you when you only intend to find something wrong with my work?”. I dream that some day these types will be forced to publicly defend such practices.
mpainter, that was Phil Jones’ notorious reply to Warwick Hughes, not to Steve McI, when Warwick asked Jones to supply details of ground station sites he used in his global temperature record.
Warwick has discussed these events here on his blog.
You can see from the email to Teunisson that Jones’ personal response was immediately hostile, no matter that his first replies to Warwick seemed cheerful enough.
Pat Frank, thanks for the correction and link. I had always believed that it had been Steve that Jones was responding to.
No collegiality within the other camp for those who intend to look too close. Science in the dark, and how they squawk when you apply some light to their doings.
JEG,
If there is true correlation of growth with local instrument temperature records, then it seems to me you can only rationally justify selecting tree series on that basis by testing if that selection method works as well when you do the selection based on holding out half the instrumental data, and then seeing how well the trees correlate with the held out data. If you select based on correlation with the whole of local temperature records, then people are going to always conclude the process is simply picking cherrys. Of course, the instrumental period held out would have to be randomly selected to minimize the risk of correlation with other non-temperature factors. That is, if an instrumental record is 100 years long, you need to hold out a 50 year block randomly (50 early years, 50 late years, 50 years in the middle, etc), and make the include/reject selection based only on the included temperature data… if the trees are truely responding to temperature, then it should not matter what block is randomly selected. What you can’t do is select series based on the whole of a local temperature record, and then apply a hold-out test, which would just be a slighly fancier cherry picking routine.
It isn’t obvious to me that you need to work on blocks of years if you deal with autocorrelation in the temp series.
There is no simple way to reliably account for autocorrelation. I think blocks are a reasonable approach to minimize the influence of aurocorrelation…. and a robust test of whether or not the selected trees are credible indicators of temperature.
I had thought there were techniques for fitting ARIMA models with missing values. My rusty recollection had it as an application of Kalman filters, but no doubt statistical theory has since moved on from there. Compared with blocks you’d get more robust results this way wouldn’t you?
stevefitzpatrick Posted Jul 14, 2017 at 6:03 PM | Permalink |
Steve, you might be interested in the method of Koutsoyiannis for accounting for autocorrelation, which I independently discovered and reported on here.
Regards,
w.
Willis, semi-circular reference.
HAS
Posted Jul 15, 2017 at 12:46 AM | Permalink
HAS, lack of any reference.
Seriously, what on earth are you referring to? Please QUOTE MY EXACT WORDS THAT YOU DISAGREE WITH. I have no interest in guessing what it is you are on about.
w.
try clicking on your reference
HAS
Posted Jul 15, 2017, at 2:32 AM | Permalink
Ah. My bad. The correct link is here.
Regards, and thanks for letting me know.
w.
Still broken, but embedded in it is https://wattsupwiththat.com/2015/07/01/a-way-to-calculate-effective-n/ which works. Will have a look.
Thanks, HAS. Finally fixed, I think. It wants the “http://” in front of the rest. In any case, let me know what you think. Dimitris Koutsoyiannis commented down below the main post to graciously point out that he was the first one to discover this method.
w.
Willis — Sounds interesting. Will take a look.
Willis wrote: “Steve, you might be interested in the method of Koutsoyiannis for accounting for autocorrelation, which I independently discovered and reported on [at WUWT, 7/1/15, https://wattsupwiththat.com/2015/07/01/a-way-to-calculate-effective-n/ ]”
Willis — I think that the calculations you cite are only relevant to the problem of estimating the mean of a series with long-memory errors. When estimating a multiple regression, the effective sample size for each coefficient can be different. For example, when fitting a trendline with random walk errors (H = 1), the intercept is unidentified and hence has an effective sample size of 0, while the slope is identified with a finite (though perhaps large) standard error. Comments are closed on your WUWT piece, so I can’t comment there.
Thanks for the Koutsoyiannis reference. I’ll take a look at it. A few years back, UC tried to teach me about fractional integration, with only partial success!
Cross validation is way of looking at data that does not have out-of-sample data for model testing and that would include such comprehensive methods as k-fold cross-validation.
https://en.wikipedia.org/wiki/Cross-validation_(statistics)
With proxy data that has a high auto correlation or long term persistence one can obtain extended lengths of time where a stochastic trend can appear. In such cases the cross validation will give a false validation of a deterministic trend.
Another problem that occurs is where in the case of temperature reconstructions we are in the end looking for trends and unfortunately high frequency correlations will not always translate to low frequency correlations (trends).
Even with these weaknesses in obtaining reasonable proxies for temperature reconstructions, I seldom or never see papers by climate scientists that even consider cross validations of any sophistication.
Wow. JEG, what is the matter with you? You are associated with PAGES, surely you are aware that PAGES2013 made major (uncredited) corrections to their results, corrections that may have restored the MWP, based on McIntyre’s critiques? That other studies (Gergis,…) were completely withdrawn, based on McIntyre’s critiques? Surely you are aware that Mann’s original statistical techniques were discredited and eventually abandoned, based on McIntyre’s critiques?
I’m leaving out all the politics – in terms of just the science, it seems obvious that McIntyre is a major contributor to your field. You need him – as an essential corrective.
Sneering and throwing ad hominems is a good way to convince the casual reader that you aspire to join the ranks of those climate researchers who really see the jobs as being politicians. If that’s your goal, go for it; play to the fanbase. If you want to be respected as a real scientist, behave like one.
I think it would be very safe to credit Steve with having had a second career after publishing McIntyre & McKitrick (2003) and also starting Climate Audit in 2005. Almost two years ago the long-time director of paleoclimate research funding for the NSF, Dave Verardo, commented here defensively as his name came up regarding past disputes and then afterwards made a remarkable admission.
Type I error.
Thanks for posting back, JEG.
I am a plant physiologist, by training and I would love to know how you can confidently ascribe growth changes to one factor, namely temperature, in some trees and not others.
Please explain the physiological theory that underpins your selection choice.
If you cannot, then you are not practicing science.
Since tree growth is dependent on many variables other than temperature, one can only take it on faith that rejecting certain sample sets is valid. For example, you are assuming that a sample set is invalid because the moisture showed some kind of correlation whereas temperature did not, but it might not have been primarily the moisture that impacted the growth – you just assume it is. It could be nutrients, wind patterns, temperature, moisture, sunlight/clouds, or other unidentified factors.
So unless you can demonstrate that all of the tree ring sample sets included for analysis are due mainly to temperature, and all of the sample sets you reject have no significant temperature component, your analysis is just a faith-based conclusion.
More convincing is demonstrating that there are multiple proxies that all show the same response, but if you reject sample sets that do not support your conclusions in the analysis of each proxy, you are just practicing more faith-based analysis. (Bad methods beget bad results)
By the way, how does the recent 18-year pause in rising temperatures factor in to a hockey stick? Increases in CO2 certainly did not pause. Your proxies do not explain reality. That is another problem.
Please make a temperature versus time graph using only the rejected data sets.
JEG — I’m confused. You say: “2) Every paleoclimatologist knows that there are multiple factors influencing tree growth; hence why it is so important to carefully select sites where temperature is the limiting factor. The basis for selection, obviously, was not whether trees exhibited a hockey stick shape or not; in cases where there are enough instrumental data, the criterion is whether they correlate well to that – too bad the temperature data show warming over the 20th century, I suppose.
Moisture is known to be the primary control at many sites, and in version 2.0.0 those sites were excluded so as to retain what is currently thought to be the sites where temperature exerts the primary control. I find it remarkable that you manage to complain about both the 2013 the 2017 versions, even though they used these different approaches.”
If I do a research study on a new drug that saves lives but discards the data from the patients that die, I’ve not accomplished anything. You appear to have done exactly that. Could you help me, as a technically trained member of the general public, understand how your selection criteria allow you to come to any statistical conclusion? Steve has claimed you picked cherries. And that appears to be exactly what you have done.
Harriet,
Excellent scientific point!!!!
The new paper is quite open about the fact that they screened for temperature:
“Proxy records were gathered from archive types for which previous understanding of the proxy system indicated that the records are temperature-sensitive. Records were only included when the original study described the relation between the proxy value and one or more climate variables, including temperature, or when the correlation with nearby instrumental temperature data was high enough to reject the null hypothesis of zero correlation at the 5% level”
So is their position that the trees that are “correctly” temp sensitive now are the only ones that were temp sensitive in the past? And do they provide any biological hypotheses and/or evidence why this should be so?
A good way to do that, btw, would be to measure their consistency with one another in how the earlier period compares to instrumental temp recordings. I think the field of statistics has tests to provide a confidence and uncertainty level for that. 🙂
Haven’t read the paper, but it would make sense to run that test I think?
“….to measure their consistency with one another in how the earlier period compares to instrumental temp recordings. ”
– ostensibly, a nice quality of these series is that when averaged, the ‘earlier periods’ tend to form a relatively flat line…a fine ‘lumber’ to be used for placing pucks between pipes.
The screening is stated even more clearly later in the methods section (Steve quoted this bit in a tweet 2 days ago):
“Of the 641 records that together comprise the previously published PAGES2k datasets, 177 are now excluded, of which 124 are tree-ring-width series that are inversely related to temperature. To be included in the current database, tree-ring data were required to correlate positively (P<0.05) with local or regional temperature"
It's astonishing that the circa 100 authors of the paper and the Nature reviewers apparently didn't see a problem with this.
The response from JEG is notable only for the fact that it ignores the main issue raised in Steve's post, spelt out clearly at the end of the post and illustrated in the cartoon.
Yes, it is easier to find positive correlation after you throw out everything that is negatively correlated …
Sheesh. You say one hundred authors gave this a pass … go figure.
w.
It could be argued that samples that agree with instrumental records are more reliable than those that do not. Yet, the 64K question is; how can any one or more of the screened samples be a reliable temperature proxy throughout the reconstructed period. Confidence in the screening process infers certain knowledge of dominate growth influence for the period covered. Hackneyed logical inference at best.
RobRic said
You mean if you throw out all the samples that disagree with the instrumental record, what remains agrees with the instrumental record? Who knew? …
Rob, I fear you’re making the same mistake the authors made. That’s “cherry picking”.
w.
WE is correct. These guys draw conclusions from data half of which contradicts the other half. They simply throw out the half that they dislike. With any training in science, they would know the futility of such an approach. But, no training in science for these wannabes. Or, if they were trained, they threw the training out with the unwanted data.
So true. I could get population records of 300 cities, throw out 250 cities that do not correspond to instrumental records and keep 50 cities that correspond well with instrumental records. What could possible go wrong if I used those 50 cities to reconstruct temps in 1500? Obviously a lot.
And did they use any series that suffer from the dreaded Divergence Problem? Y’know, those that correlate with modern temps, except when they don’t.
RobRic, for more than half the time span considered in the PAGES document there IS no instrumental record. That is WHY they are using “proxies.” The methodology shows a guaranteed, high potential for a Type I error, and I think it may be the very first time I ever saw Expectation Bias as a documented, deliberately employed, element in a data-screening methodology.
But they’re also using Bristlecones, which are known to have bad juju, but they didn’t seem to care.
Shades of Keith Briffa: “Has it ever occurred to you that paleoclimatologists might know a shred more about paleoclimate than a mining executive? ”
JEG, I urge you to study the Climate Audit archives searching “yamal” and “Briffa” before you make any more such statements.
Many archives? Apples oranges and orangutans?
Or were measures applied to “equalize” archives, so they could be included.
Exactly what measures were used to verify accuracy and applicability for each archive paleo source to temperature profiles local to the source?
How is CO2, a definitive plant enhancement component isolated from other environmental factors?
You use the word “knows”, but the description casts doubt on that statement. That researchers use their personal knowledge or discretion smacks of subjective decisions, not objective data.
A) Did the original researcher have the exact same understanding of those factors when they recorded the data?
B) How were those factors applied to selecting source archives?
C) How does a field researcher “know” what ground conditions existed over any historical period?
a. Aquifer levels rise/fall, Beavers build dams, Rocks fall, streams change course, ground cover burns, pack rats build nests, marmots store food, bears water the tree; there are numerous factors that change ground water availability to any plant.
b. Non-field researchers then “judge” another person’s meta data to select proper data sources?
Slip in an ad hominem? A little derisive condescension and pomposity?
Well, it greatly reduces any import you are trying to convey. Another climate whomever relying upon Argumentum ad verecundiam (argument or appeal to authority); a ruse that has worked so well for climatologists.
Our impressions? Think of puffed up self important nasal discharge.
More ad hominems, all fallacious.
Steve McIntyre’s data studies are meticulous and dispassionate. Unlike the climastrologists that prepare much of the morasses they call data; as you demonstrate here, so well.
Steve adheres to facts; preaching is what you are doing.
What an interesting challenge gauntlet you toss around, “tell us if you can find something that’s not a hockey stick”
That sure sounds like someone who knows the game is rigged.
Which brings up the ugly question, exactly how you verified and validated your model?
Climastrologist climate models have worked so well to date.
Given a choice between the mining executive and a ‘climate scientist’, I would go with the mining executive everytime.
Someone is about to get cherry pie all over their face.
JEG,
As a former mining executive, and a science graduate, I was in the company of many management people who had degrees in engineering, metallurgy, science (geology/physics/chemistry), some with several degrees, commonly an MBA (Harvard). They were in top management because they were good scientists as well as managers.
You seem to be missing a measure of success. Money can be used. The small group of scientists I helped manage for 20 years found 13 new mines, whose products to date have been sold up to 2015 for $Aust 62 billion (in 2015 dollars and metal prices), with an estimated $79 billion still to be mined and sold.
A substantial part of this went into general Government revenues and a substantial part of that was allocated to funding things like the salaries of government scientists. Geoff.
“Has it ever occurred to you that paleoclimatologists might know a shred more about paleoclimate than a mining executive?”
Has it ever occurred to you, JEG, that a universally recognized and respected expert in statistics might know a shred more about statistical manipulation than a woodcutter?
Phi Howerton
Reminds me of my time in grad school – I spend a few years working in my profession, went to grad school at night. Took a class from the professor that had written numerous textbooks books used through out most universities. I became the defacto instructor clarifying significant parts of the subject matter that the professor had not actually encountered in a real situation. Point being that work product is much improved when real money is on the line.
Phil, that’s an unfair slur on woodcutters, some of whom deserve a great deal of respect for their skill and accuracy:
I just had to thank you for posting that awesome video! Crazy…
i would be interested in seeing any north american tree ring series that produce a hockey stick. the temperature record for north america in recent decades would suggest this might well be an impossibility.
Dang the R developer on the project is doing commits as we speak
Are you referring to Ms. Wang? https://dornsife.usc.edu/jianghaowang/cv/
Pure math education. No science. Another wannabe.
‘Pure math education. No science. Another wannabe.”
The majority of the project is assembling the DATA
This is a DATA paper.
you are welcomed to go check the Git.
you wont.
I did.
If you read the paper the paper you will see that is data paper.
Do you know what that is?
Do you know what skills it takes to create a data repository in a feild where there is no “standard”?
Get back to me when you understand LiPD.
Here is the thing. The data looks to be published in the new format. I am suspecting that you have to
know matlab, python or R ( work in progress) to read the data.
no more text files on the internet to just grab and go.
pluses and minuses.
Data aren’t data absent a falsifiable theory or hypothesis, Steve Mosher. Learning how to run numbers around in R never made anyone a scientist.
“Data aren’t data absent a falsifiable theory or hypothesis, Steve Mosher. Learning how to run numbers around in R never made anyone a scientist.”
Do you ever actually read anything?
1. The claim was made that the WRONG PERSON wasnt a scientist
2. I never claimed, nor did they ever claim to be a scientist. so strawman
3. Go ahead and put forward a testable hypothesis about what makes a scientist,
your ad hoc definitions are un convincing.
4. The paper is a Data paper. It’s a collection. Did you ever record something?
What does that act of recording and reporting constitute?
5. Philosophically, Yes, there is no such thing as data without theory.
Pat I’m sorry your crazy ideas about uncertainty have been laughed out of every credible journal
and every credible blog discussion.
That must suck at your age.
Retire in peace, I can suggest some caves where you will be the smartest person in the room.
by default of course
Gavin Schmidt is an example of someone with a math education and none in science who somehow drifted into climate science. The Wang person would have shunned this study if he had any insights into scientific rigor. That is the whole point, Mosh. They are all wannabes, but in fact are clueless.
Steve Mosher, your response to “Pure math education. No science. Another wannabe,” was, “The majority of the project is assembling the DATA. This is a DATA paper. … If you read the paper the paper you will see that is data paper.”
To which I replied, “Data aren’t data absent a falsifiable theory or hypothesis.”
Near the end of your (ignorant) rant you even agreed, “Philosophically, Yes, there is no such thing as data without theory.” although your agreement is fatally incomplete.
Nevertheless you’re now on record agreeing with me that Julian Emile-Geay’s PAGES 2017 paper is not science.
Just to add, science is not philosophy and does not depend upon philosophy for anything. Theory in science does not emerge from philosophy, is analytically falsifiable, and is not axiomatic.
As to the rest of your sneering, you have never shown the slightest understanding of my work.
Your dismissal is uninformed, fatuous, and wrong.
Much of my work is posted on Watts Up With That, where I have carried every debate about it.
I also carried the debate about measurement error at Jeffid’s.
Just as I carried the debate about you folks at BEST negligently ignoring instrumental resolution.
You probably didn’t understand that, either.
mpainter ITS A DATA PAPER.
quite simply. You take the data. You develop a screen. You publish the output of the screen.
Thats screened data.
You dont want Screened data?
Simple. Go the original sources. Compile it. Publish your own damn data paper.
For example. Over 5 years ago Anthony and Steve Mc did an analysis paper where they published some results on the web.
Key to that was a dataset Anthony created.
I asked for a copy. They said No.
I complained that Steve and Anthony could hide this data forever. They said dont worry it will be released.
5 Years. during taht time Gergis republished their mistaken paper and data.
I asked Anthony to at least publish a data paper. Data papers make no scientific claims. They are just data.
Still no data.
Even if they screened their data I would happy to get a subset of the data. So, I find little standing to critique
a mere data paper.
Its numbers.
The paper is not just a data paper. Julien has asserted results based on the data.
Further, the particular screening used by Julien is integral to the selection of data, so it’s entirely reasonable to criticize the selection of data. Indeed, it’s a long standing issue.
Nor is it unreasonable for a reader to criticize the selection of data without a saucy retort that they should collect their own data.
Nor do I understand why you’re ragging me about temperature data. I haven’t collected temperature data and do not control any. Nor is it fair to say that “Anthony and I” wrote his paper. He and Evan wrote a paper. I was asked to review and sent him a short comment, which he included and, out of excessive courtesy, added me as an author without asking me. His intentions were generous but I hadn’t been involved in the design or writing of the paper and would not have accepted authorship. I almost immediately noticed an error in the analysis and asked that the paper be withdrawn for re-working. Anthony invited me to help with the analysis but, as you know, I long ago resolved to spend time on paleoclimate data not temperature data and did not pursue the topic. I understand that you would like to look at Anthony’s data. I also understand his problem – he’s trying to do a lot of things: operate a business in difficult times, operate an enormous blog and it leaves little time for analysis. While the delay is regrettable, it’s far more justified than (say) Lonnie Thompson’s withholding of data for decades. I have zero interest in arguing about Anthony’s data. Take it up with him.
Hi, Mosh. I think the best response for your comment is from Geoff Sherrington, below, which I copy:
The basis of this cherry picking post by Steve would plausibly disappear if the PAGES papers had proper error analysis, because none of the studies showed any physical significance. I dislike noise dressed up as data.
Geoff
What you fail to recognize is that certain assumptions are made in the selection of the data, which assumptions fall to the ground when critically examined. That is why any mathematician who wants to dabble in science should acquire some scientific or statistical training, or both.
Mosh,
Has Mann given you a date on releasing his Hockey Stick data yet? You must be livid about that by now! Almost 20 years!
“Further, the particular screening used by Julien is integral to the selection of data,..”
Steve McIntyre makes a worthwhile point, one that I neglected to make because it did not seem that any reasonable distinction could be made between the “screening” employed by PAGES 2017 and the regular methods of selection used in these types of studies.
“Over 5 years ago Anthony and Steve Mc did an analysis paper where they published some results on the web…etc.” Mosher, I don’t think you’re making sense here. The obligation to provide data (and code) isn’t some kind of a “data wants to be free” thing. It’s, “If you’re claiming a scientific result you’d better be prepared to back it up.” Their paper came out, a mistake was found, they withdrew it. Done, good job all around. The way science is supposed to work. Why are you asking for data afterwards? Do you want to prove it wrong a second time?
Asserting ‘results’ outside a paper that basically documents how a dataset was built
does not transform a data paper into another more than a data paper.
I dont find any claims in the data paper.
If making claims outside a data paper could somehow change what it is, then I’d
like to know how.
Except you didnt criticize the selection of data or the method, you seem to criticizing
the use of that data subsequent to the screening.
1. you did the statistics on the paper.
2. An error in not using TOBS data was pointed out to you.
3. you promised to re do the analysis with the right data ( Recall that Gergis had a
comparable data boo boo the exact same time frame)
4. Re running that analysis should have taken a short period
5. Those results were never published
6. Subsequently ‘non TOBS stations” have been selected for additional
analysis ( sound familar)
7. At the time I asked for the station data– Not the temperature data BUT the
metadata ( kinda like Willis asking Jones for station ids )
8. Your comments from over 5 years ago below.
But Hu.
1. Anthony has put it out for blog review and cited muller as a precedent for this practice. that practice included providing blog reviewers with data.
2. Anthony brought Steve on board at the last minute even though hes been working on this paper for a year. Steve has a practice as a reviewer of asking for data. Since we bloggers are asked to review this, we would like the data.
3. if, they want to release the data with limitations, that is fine to. I will sign a NDA to not retransmit the data, and to not publish any results in a journal.
4. You have to consider the possibiity than Anthony and Steve could now stall for as long as they like, never release the data and many people would consider this published paper to be an accepted fact.
Steve: Mosh, calm down. this is being dealt with.
Mosh, you say: “1. you did the statistics on the paper.” I did not. Please stop saying this.
Evan did the statistics. I was not involved in its writing.
I was asked to comment a couple of days earlier, quickly sent some comments on Friday as I had family obligations. The TOBS issue was a topic that I had discussed at Climate Audit in 2007 and was very familiar with. So I was annoyed that this hadn’t been dealt with – annoyed with myself that I hadn’t noticed it. I asked that Anthony withdraw his draft paper and he did. Anthony needs someone with statistical skill to work up his results but I don’t have the time or energy to wade into temperature data. As you know, it has never been a primary topic at Climate Audit. I have many unfinished topics of greater interest to me.
You know that it’s not “my” data. I don’t know why you keep ragging me about it.
Mosher, seek help.
Hi Mosh,
I’m not sure what your point is when you keep insisting that “it’s a DATA paper” but if you mean to say that the study is outside of a science context, well, that is simply wrong. Data are data only within a context of science and the methods of collection, screening or selection and presentation of data is science, and any conclusions drawn from data is science with such conclusions dependent on the methodology employed. So the data paper is a science paper, but done with little understanding of proper scientific procedures or scientific rigor.
The point that I attempted to make above is that such a study is authored by those who lack understanding of scientific rigor.
It’s not even a “paper”. The authors call it a “community sourced database of temperature sensitive proxies”. They also admit that (only) “nearly half” of the proxies show temperature sensitivity.
The initial argument here doesn’t rely on what Mosh or anyone else decides to call it. It’s about the methods used to build the database AND the DATA excluded from being included in it. I’d call that a “Selective database” or “biased database”.
Nope.
Do you know how to check Git, moron.
Different guy is working on the R code.
It is quite good
But you didnt check.
No Dang contributions in R on Github to this project: https://github.com/search?l=&p=1&q=Dang+language%3AR&ref=advsearch&type=Repositories&utf8=%E2%9C%93
Plenty of code from Julien Emile-Geay though: https://github.com/CommonClimate/PAGES2k_phase2/commits/master
Antony, Mosher The Recalcitrant was using “dang” as an expletive, like “damn”. Not sure if you’re aware of that.
Typical morons. read the paper. use the links in the paper. click on the link. find the git. find the R. If you want to read the data that is. you dont.
Mosh, you’re sounding like a grumpy old man yelling at kids to get out of your yard. Lighten up, man.
Mr Mosher
It is evident that punctuation, comprehension and civility are not three of your strong points. If you go back and read your top comment and Antony’s two responses you might figure out why you owe him an apology for your childish name-calling.
Mosher is now communicating via Staccato Encyclical.
Lignophilous. Gotta love it.
Steve, as always, thanks for your clear and readable analysis.
w.
Truely puzzled why 123 tree ring series were good temperature proxies in 2013 and have become anathema in 2017?
This once again raises the question of a parabolic temperature response and the inverse problem. If half the series go down with warming,and half up, this seems like a good indicator that the species can be water limited (as they say). Thus for any given site, if you warm it enough the trees may become water limited at some point and become negative responders. Also, in the past, precipitation need not fluctuate in synchrony with temperature, and it can fluctuate a lot. Thus the assumption of stationarity seems only to be valid for short time periods (maybe a century), not for 1000 year studies.
“Thus the assumption of stationarity seems only to be valid for short time periods (maybe a century), not for 1000 year studies.”
And that really understates the problem, doesn’t it? When c. 45% of tree ring series are not temperature responsive during the instrumental period, how can there be any confidence about temperature sensitivity for any series in any period before the instrumental period?
And out of that 45%, how many act like the Briffa series which tracks well with temperature…until it doesn’t (the “decline”)? (I don’t know, but I’m asking.) Relatedly, how many of the series from the 55% go in the “right” direction during the instrumental period, but exhibit their own “decline” in an earlier period? We simply don’t know. And I don’t see how that question is even answerable, at least not with any method currently known.
Not even with dowsing rods broken off ancient dry tree trunks (so not lignophilous 😉
How can you test that something is specifically a ‘temperature responder’ other than to see if its data correlates to “known” temperature?
Split your sample in to parts with sufficient power for a reliable measure that hypothesized relationship exists, then use the non-tested portion of the sample in your analysis. Or at least that is the way psychometricians do it when building metrics.
Bernie: that is not going to help. Precipitation varies over time but is not known 500 yrs ago and so can’t be factored out. The temperature response depends on precipitation. If it rains more, the same tree won’t go negative with warming. The pretense that precip effects can be ignored if the tree is CURRENTLY no water stressed is simply not tenable.
The assumption of confounding variables cancelling out over time is not a good one per what Craig says.
If one has to work with confounding variables in doing temperature reconstructions, one must then assume that at least spatially there is some cancellation of those confounding effedts. But that acknowledgment only leads to a further weakness in the current post fact selection of proxy data which amounts to throwing out data that could well have a necessary cancelling effect on the data retained.
Craig and Ken: I thought I was answering a different and simpler question. I absolutely agree that sorting out signals that likely covary is very difficult. That is like having a large number of interaction terms which vary with a large number of to be defined dummy variables.
With biological stuff, you have to realize that there are multiple variables that affect growth and development. At any one point in time, the limiting variable may not be the limiting variable previously or later or both. Also well known in biochemistry circles, i thought, was that biological organisms have a quadratic response overall. When conditions are below optimal, moving toward optimum enhances growth. Once near or at optimum, the improvements stop. Past optimum, you don’t get enhancement at all and get suboptimal responses. Far enough away from optimum, you get losses of individuals. The survivors will bias your sample.
You would have to first determine reasonable a prior criteria for selecting proxies and then test the validity of those criteria over a period of known temperatures. No one has said the proper selection would be easy, but one first must admit that selecting post fact is going to bias the data – and towards a hockey stick.
The well established problem of tree rings and other temperature proxies that appear to respond reasonably well to temperature for a period of time and then do not later in the series throws up another caution flag in selecting criteria a prior.
JEG — Bear in mind that Craig is a certified lignologist and knows what he’s talking about when he says that any temperature response is likely to be parabolic and non-monotonic!
It always wondered me how they can know that Medieval temperatures were not higher than today, because of the “optimum” temperature for tree ring growth…
If the same species now still is growing with higher temperatures near the optimum growth, there is no way to know that a ring width drop of 1,000 years ago was caused by higher or lower temperatures…
Another one I have read some years ago: some tree species did show a good correlation with moisture for a large part of the altitude of the forest, where the upper few hundred meters up to the tree line did correlate well with local temperatures.
As the tree line in the Medieval period was several hundred meters higher than today (thus showing higher temperatures…), the same trees used as “temperature proxy” today at that time were moisture proxies…
Link to PAGES 2K 2017 in the first post would be helpful.
With respect to disregarding tree rings because of a negative relationship to temperature, can’t they just flip it?
Give them time. They will flip it, eventually.
Isn’t that exactly what was done?
“Sign adjustment
Records were multiplied by −1 if their values decrease with increasing temperature (i.e., if their interpDirection parameter is negative); by +1 otherwise. This step ensures that all proxy values point upward (downward) in response to warming (cooling).”
Looks like PAGES 2K gave an explanation for excluding tree rings with a negative relation to temperature:
“Trees whose growth increases with temperature (e.g., direct effect of temperature on physiological processes and photosynthetic rates) are more likely to produce a reliable expression of past temperature variability compared to trees that respond inversely to temperature, for which the proximal control on growth is moisture stress (e.g., evapotranspiration demand)”
So they give a decent reason, although Steve’s McIntyre’s red noise argument has some merit. What is the geographic distribution of excluded tree rings due to negative temperature relation to non-excluded tree rings? If they have different geographic distributions (i.e. excluded tree rings are primarily in warm dry places) then the PAGES 2K approach is probably good, otherwise there might be an issue.
Anyway, congrats to the PAGES 2K team for publishing their new data.
They say in their methodology that they account for autocorrelation when determining the significance of the relationship. So maybe the red noise argument isn’t an issue.
How was the CO2 fertilization effect or ocean acidification taken into account? As I understand it, most of their proxies (except ice core, borehole and document) should be affected by CO2 fertilization.
I do not see their reason as stated as a “decent reason” at all. The criteria for excluding trees has to be specific and not based on an ex post facto measurement of the relationship with temperature. It is one thing to exclude a specific species of tree or trees of a specific species from a particular altitude or region or location because they are not good temperature proxies, but it is “cherry picking” to exclude trees based upon their ex post facto relationship with temperature. They have essentially determined the average height of Dutchmen by excluding all Dutchmen who are below the average height of Dutchmen 5 years ago. It is a bogus measure.
Bernie 1815 wrote:
“They have essentially determined the average height of Dutchmen by excluding all Dutchmen who are below the average height of Dutchmen 5 years ago.”
No they didn’t. You are wrong and McIntyre is off in the weeds on this. The dendros think they are excluding Danes who ended up in the data. Why do you think they should include the Dane data when they are studying Dutchmen?
I’ll admit that I fell for the red noise schtick when it was first described here. But it only applies if you are studying red noise, not if you are studying a signal. A classic red herring.
Suppose an ornithologist wanted to analyze the song of a rare bird. She puts up an array of 50 audio recorders in the woods and lets them record. Later, she reviews the audio tapes and hears the bird song in two of them. She discards the other 48 records and analyzes the two with song. Now McIntyre comes along and criticizes her for not analyzing all 50 recordings. After all, how can she be absolutely sure that there is not a note of birdsong here or there that might influence her results?
Here is the thing: do you accept that temperature can be recorded in tree ring width? I do, because I have seen dendro records that convinced me at Dr. Mauri Timonen’s site (unfortunately no longer available). This was a cherry-picked example of the best treemometer he could find, and it was very compelling. Regional reconstructions bring their own set of thorny issues, but that does not change the basic fact that treeline trees can be a good proxy for temperature.
Dendros have seen these compelling data sets as well, the readership of CA not so much. So if you accept that it can happen, that becomes an assumption for paleoclimate reconstructions using tree rings.
When McIntyre derides the dendros for leaving out trees that exhibit no temperature signal, he is basically asking them to prove once again that trees can record temperature. This is akin to asking every researcher who wants to contribute to the literature on evolution to begin their papers by proving evolution exists.
I would venture that McIntyre’s childlike insistence on reinventing the wheel in these reconstructions is the real reason he got a collective shrug from the dendros, and disinvited to their symposia.
That being said, two entirely different points are conflated in this post. The second point, that folks like JEG should not tout their hockeystick when they screened out other shapes, is entirely valid. Everyone needs to remember paleo reconstructions of climate are about the handle, not the blade. They are trivially uninformative about the blade.
Matt Skaggs wrote:
“When McIntyre derides the dendros for leaving out trees that exhibit no temperature signal, he is basically asking them to prove once again that trees can record temperature. This is akin to asking every researcher who wants to contribute to the literature on evolution to begin their papers by proving evolution exists.”
This is a fascinating comment. It’s as if researchers are entitled to pursue research programs on particular questions, even when those questions can’t be answered with the methods currently being employed. The frustration, on a personal level, and even on an intellectual level, is understandable. But that’s irrelevant as to whether the questions are answerable or not. However “compelling” a particular dendro study might be (whatever “compelling” might mean), what bearing does that have on the assumption of stationarity across both the instrumental and pre-instrumental periods? Or on the assumption of a linear response to temperature? (And we know that linearity doesn’t hold — everyone admits that, regardless of the implications they might derive.)
The arguments in Loehle, C. 2009. A Mathematical Analysis of the Divergence Problem in Dendroclimatology. Climatic Change 94:233-245, are devastating on these points. If there’s been a satisfactory critique, I haven’t been able to find it.
To take the rare bird analogy, the problem isn’t knowing which sounds are birds and which aren’t, but knowing which birds were singing the right tune in the distant past. At the same time, some birds may have sung the right tune in the distant past, but ceased to do so in the recent past (i.e. the instrumental period). Without recordings of past eras, these questions can’t be answered. Wishing it so doesn’t make it so.
Matt Skaggs: “compelling data sets”.
Did it ever occur to you that all growth factors might fortuitously combine to yield your “compelling data sets”? Showing a spurious correlation to the instrument record? But the dendros reflexively assign such “compelling data sets” only to temperature factors. What is it that “compels” them to do this?
Astrology can also be compelling, if the signals of the heavens align themselves in fortuitous ways. For those who believe, that is.
And no, Matt Skaggs, we are not “asking” for proof that trees can record temperature. We are asking for proof that the temperature component of growth can be isolated from other components of growth, so that we can feel assured that the “compelling” correlation to site temperature history is not simply fortuitous and spurious.
As long as the dendros ignore these cogent issues, arguments like yours carry no weight with skeptics.
Finally, why fool with tree rings when ice core d18O is a far, far superior proxy for paleoclimate reconstructions? The tree ringers never learn anything about proper science, but mulishly plow ahead with their shabby methods.
Matt: Yes, trees can record temperature and we can show it for the historical period. But can that same tree show temperature 800 years ago when precipitation and other factors were different? Prove this. This is the stationarity assumption.
My understanding is that the problem is more basic than the problem of extrapolating derived relationships into the past. It relates to the way in which the instrumental period models have been derived. In terms of Matt Skaggs’ two analogies: verification of the models we are using to separate Danes from Dutch, and separating the rare bird’s song from the common clods.
What needs to be done with each instrument is that we need to be satisfied that it is a valid instrument using a predefined criteria. The typical test is that the model derived from a subset of known relationships adequately predicts the balance of known relationships. If it can’t do that what use is it for domains where the instrumental temp is unknown?
My understanding is that this isn’t done.
Suppose an ornithologist wanted to analyze the song of a rare bird. She puts up an array of 50 audio recorders in the woods and lets them record. Later, she reviews the audio tapes and hears the bird song in two of them. She discards the other 48 records and analyzes the two with song.
Suppose a dendro wanted to analyze tree rings. She goes into the woods and cores 50 samples. Later, she reviews the samples and sees rings in only two of them…wha???
–
Are you claiming that the ornithologist did not study all 50 recordings?
Yes.
You are also claiming that the ornithologist only studied two “known” recordings with bird songs.
Cherry pick cherry pick cherry pick.
–
We can accept that you are gullible and easily convinced. Proof is required to convince most of us.
–
–
Plants, including to trees respond to sunlight, water, CO2, and nutrients. Plant photosynthesis occurs in sunlight.
The higher the temperature, the faster the photosynthesis.
Regardless of temperature, deficits of sunlight, water, CO2 or nutrients are reflected in growth. When deficiency’s are reduced, growth increases, especially when temperature is higher rather than lower.
–
Temperature, by itself, is not what grows a tree or builds wood.
–
Judgmental subjective decisions, e.g. deselecting series that are judged “wet”, or the wrong direction are quite absurd; when other substantive growth factors are ignored in some devotional belief that only temperature drives tree ring width.
Craig, I’ve never disagreed with you before, but tree rings correlating with recent temperature is not tree rings recording recent temperature.
Matt Skaggs wrote: “Here is the thing: do you accept that temperature can be recorded in tree ring width? I do, because I have seen dendro records that convinced me at Dr. Mauri Timonen’s site (unfortunately no longer available). This was a cherry-picked example of the best treemometer he could find, and it was very compelling. Regional reconstructions bring their own set of thorny issues, but that does not change the basic fact that treeline trees can be a good proxy for temperature.”
Here is the thing: do you accept that intellectual capability can be recorded by bumps on the skull? I do, because I have seen phrenology records that convinced me at Dr. Franz Joseph Gall’s site (unfortunately no longer available). This was a cherry-picked example of one community where phrenology was able to identify the smartest men, and it was very compelling. Regional reconstructions of intelligence bring their own set of thorny issues, but that does not change the basic fact that bumps on the head can be a good proxy for intelligence.
Of course, we do know that the rate of photosynthesis (the rate-limiting step in the growth of most plants) depends on temperature. Or do we. Has anyone grown trees in a greenhouse under properly controlled conditions similar to those encountered at high latitudes or high elevations? How big is this temperature effect on tree rings when all other factors are kept the same?
This is what differentiates science from pseudoscience: One develops hypotheses AND tests them.
JEG claims to screen temperature-sensitive sites from those that are not. So how would scientists in other fields test JEG hypothesis that he has done so (and not cherry-picked)? Well, first he would note that CO2 and temperature are co-linear variables in the second half of the 20th century, so he wouldn’t use that period. And looking for correlation in high-frequency (annual) temperature change would be far more convincing than correlation arising from slow change (where auto-correlation is a bigger problem).
Then he might test the hypothesis that he has successfully separated sites that are ALWAYS (or most of the time) temperature sensitive from those that are NEVER (or less of the time) temperature sensitive. How about looking into the years before the instrumental period when there were major volcanic eruptions? Or local documented disruptions in a particular year. Are these always, often, rarely or never recorded in by temperature “sensitive” AND non-sensitive sites. One could use the part of the instrumental period to screen and the remainder to validate this assumption.
They would also carefully compare sensitive and non-sensitive sites to see if they could identity a common characteristic that distinguishes between the two. High-elevation sites lacking morning sunlight may be less temperature sensitive due to the local microclimate.
Finally, an independent colleague would be asked to prepare artificial data with both temperature sensitive and insensitive sites ahead of time – based on a composite of real temperature records totally unknown to the investigator. The investigator would blindly check to see how well his screening strategy and reconstruction performed.
Regarding Matt Skaggs’ comments: Surely, some trees show a clear temperature signal, but that does not mean that all trees with a positive correlation to temperature are showing a signal. That should be one of the main conclusions of the red noise study. Even for trees that have a signal that is causally linked to temperature, simply having a positive (even a large positive) correlation does not mean that the low frequency component of that signal is represented without bias or non-stationary in sensitivity. The low frequency signal that is being sought is very small compared to daily, seasonal and other short to medium time-scale fluctuations.
Matt Skaggs – A more relevant illustration: In 2013, after reviewing thousands of recordings, an ornithologist claims to have discovered that 146 of them contain the song of a rare bird. In 2017, it is proven that this rare bird can only produce sounds between certain frequencies … and that the songs in 123 of the 146 recordings exceed that range. In other words it can be proven that they do NOT in fact record the bird’s song.
Why on Earth would you believe that there is any validity to this “expert’s” opinion on anything else? Why would you think that the 23 remaining recordings actually do record the bird, just because they have not been conclusively disproven?
Matt Skaggs,
There’s a massive logical problem with your rare bird analogy:
In order to equate to the tree ring situation here, your ornithologist would have to be looking for times when this rare bird would sing only because it was sad. After collecting all 50 recordings, and hearing the bird’s song on ALL of them, she then tosses out nearly HALF of them because she determined that in those, the bird was singing for a reason other than sadness.
NOW do you get the point??
Matt Skaggs
Posted Jul 14, 2017 at 1:13 PM
There is no diagnostic measure linking tree growth to temperature absent thermometers. Plant growth, including trees is governed by many things including the availability of CO2, water and other nutrients. The SOLE reason that heat would come into play is through affecting one of those other factors. In alpine and circumpolar environments where trees deal with soils frozen for part of the year, temperature would be a reasonable consideration. Anywhere else on the planet and that assumption is prone to failure. In fact, increases in available water or CO2 would lead to increases in annual growth. Water not only increases wilt resistance, it improves the availability of soil-born nutrients. Increasing CO2 is well known to facilitate plant growth. Increasing CO2 over the 19th and 20th centuries would, in fact offer a justification for increased tree growth without recourse to any other factor.
Well, if the study were about whether the rare bird’s song was different in the past, it would not be valid to assume that the 2 records which caught the song in the present were the ones that were valid for the past.
“When McIntyre derides the dendros for leaving out trees that exhibit no temperature signal, he is basically asking them to prove once again that trees can record temperature.” No. This doesn’t work. Even if trees can record temperature, once you admit that there are trees that do not record temperature you are left with the problem of deciding which are which. Checking correlation with recent instrumental temperature will yield (a) trees that record temperature, and (b) trees that don’t but happen to correlate for other reasons. You can’t tell them apart, you can’t tell if 99% of your trees are in the second group.
And as McIntyre has already implied, the larger the total set of your excluded trees compared to included, the larger the second group is likely to be – more candidates. And it sounds like we no longer can even keep track of all the trees excluded somewhere along the line, just keeping the ones that happen to have had a good century so far.
Perhaps another point: I didn’t follow the details, but if the set of included trees changed drastically from PAGES2K to PAGES17 – well, that is what you would expect if most of the trees are type (b). They fit one measure the first time, but any change in the measure and it’s a whole new set of trees that happen to fit.
“We use the Cowtan & Way version38 of the dataset, which corrects for missing values and incomplete post-1979 Arctic coverage via the use of satellite observations. Even with the correction, the HadCRUT4.2 dataset is incomplete, with about 60% of the monthly values missing, so the remaining missing values were infilled via the GraphEM39 algorithm.”
What do they mean by Cowtan & Way is incomplete? I thought it infilled everything.
Oh wait, did they only use Cowtan & Way satellite version after 1979 and HadCRUT4 before 1979? That would explain by what they mean by ‘missing values’. Why not use the Cowtan & Way long version prior to 1979?
The divergence problem and upside down parabola:
Loehle, C. 2009. A Mathematical Analysis of the Divergence Problem in Dendroclimatology. Climatic Change 94:233-245.
Many thanks. Look forward to reading it.
For those interested, some key points from the abstract:
“If trees show a nonlinear growth response, the result is to potentially truncate any historical temperatures higher than those in the calibration period, as well as to reduce the mean and range of reconstructed values compared to actual. This produces the divergence effect. This creates a cold bias in the reconstructed record and makes it impossible to make any statements about how warm recent decades are compared to historical periods.”
But, as a practical fact, the problem of sorting temperature factors of growth from the other factors are insuperable. Treemometers are the dreams of simpletons.
It’s possible in principle by way of tree-ring C-12/C-13 ratios, mpainter, but the realization of that possibility has so far frustrated everyone who’s tried it.
Craig,
The parabola is most relevant. So are fertilizer levels, with at least half a dozen major and a dozen minor. The application rate is an obvious variable but the availability is even more so over multi-decadal terms. Soils and the fertilization factors cannot be assumed to stay constant over centuries with variable rainfall etc. Also important are anthropogenic contributions particularly CO2 and sulphur incl airborne SO2 from the smelting and coal burning era. Plus the exotic mix from volcanic ash falls. Others above have noted these complications but there seems difficulty in accepting them.
It is a scientifically naive idea that past temperature can be assumed to be the dominant cause of variable growth. There are few rigorous studies addressing this in the instrumented era and no papers seem possible to cover earlier times.
(I once did full-time science of plant nutrition and soil science for 9 years with a CSIRO lab and 2 other subsequent labs).
Not to mention aeolian transport, Geoff. 🙂
Why discard tree rings that don’t correlate well with temperature? You can use them upside-down like Mann did with Tiljander.
The absence of a correlation means that it makes no difference if you use it upside down or downside up. If a particular proxy correlates negatively with temperature, then yes you use it upside down – BUT you must have an explanation for the relationship before you analyze the data. What Mann did was flip the the established relationship because it was convenient for his argument. There was no defensible rationale – just dust bowl empiricism.
Yeah I should have specified negative correlations…
And if the same proxy sometimes “correlates” negatively, and other times positively, then it’s an awfully suspicious “proxy”.
I’m currently debating with Katharine Hayhoe about paleoclimate reconstructions, among other things. She posted this graph, without providing her source. Does anyone recognize it?
The temperature curve is invention. Tell hayhoe it is a lie. Like all warmunist paleoclimate reconstructions.
from that graphic. temps are only ~0.9F hotter than 6000 and 3000 years ago!
that is only 0.5C.. what a dumb graphic
(ignoring resolution of proxies, vs thermometer readings)
James, in your graphic, the recent warming is about 1.5°F. This is about 0.8 °C.
On the other hand, the recent change in CO2 corresponds to an effect of just over half of a doubling [ log( 400 / 275 ) = 0.54 ]
This would mean that according to her data, climate sensitivity is a mere 1.6°C per doubling.
HOWEVER … over the previous 5,000 years, temperatures were going DOWN and CO2 was going UP … say what?
Overall, a most curious graphic.
w.
I doubt that she will find any that agree that we are warmer than the Holocene Optimum, as portrayed by her fantastical plot. I’ll bet that Michael Mann would reject it.
James McCown, I published a paper demonstrating that paleo-temperature reconstuctions are pseudoscience, here. If you’d like a reprint, email me at pfrank_eight_three_zero_AT_earthlink_dot_net.
I also posted an essay, “Proxy Science and Proxy Pseudo-Science,” at Watts Up With That that does the same and further takes apart d-O18 reconstructions by assessing the neglected experimental error.
Try Knudsen:-https://www.nature.com/articles/ncomms1186
top curve looks like a variation of Marcott et al, discussed here a few years ago.
The title should be changed to “global warming for dummies”.
I posted Loehle & McCulloch’s 2008 reconstruction and asked Katharine why it differs so much from the chart she posted. She responded with this link to the Skeptical Science site:
https://skepticalscience.com/Kung-fu-Climate.html
All that essay does is take issue with the fact that Loehle & McCulloch’s chart ends in 1935. So, I also posted Werner et al’s reconstruction of Arctic temperatures and pointed out that it looks similar to Loehle & McCulloch’s reconstruction.
It will be interesting to see how long this debate lasts before she blocks me on facebook. I tried debating her on twitter a couple of years ago, but she blocked me after two exchanges. I’ve also been blocked by Michael Mann.
Maybe we could get little hockey stick tatoos to signify how many times a “climate expert” has blocked us online or in person.Like teardrop tatoos in prison, it would alert others to how many times the self proclaimed “experts” have killed a reasonable, logical climate discussion…
Jim — Thanks for the reference. I’ll take a look at it.
One thing that struck me about this Nature article is how the graphs for composite proxies and particularly lake sediment begin forming a “blade” well before 1900 (about 1750 for lake sediment). Wouldn’t this contradict a major point of the “consensus,” that industrial-era C02 caused accelerated warming sometime AFTER 1900?
Maybe this has been discussed somewhere, but I can’t seem to find it here or elsewhere.
CL
Most of such curves take on a completely different complexion when read with the knowledge of errors detailed e.g. by Pat Frank at WUWT of April 3 2012.
Steve Mc has said that he does not choose to delve into error analysis and we enjoy his emphasis on more lively topics.
It would be great if a group of error formalists gathered to add to Pat’s work. I can’t, health issues. Analytical Chemists have to know the error business.
The basis of this cherry picking post by Steve would plausibly disappear if the PAGES papers had proper error analysis, because none of the studies showed any physical significance. I dislike noise dressed up as data.
Geoff
Best wishes, Geoff.
CL: “Wouldn’t this contradict a major point of the “consensus,” that industrial-era C02 caused accelerated warming sometime AFTER 1900?”
Two plausible answers:
1) The regional climate trend where many lakes where proxies are available started warming 100yrs before the industrial revolution. Or…
2) Finding cherries that ripen at just the right time is too constraining on one’s selection for making a pie.
Steve: ” My guess is that many more than 27 series were screened out in the 2017 addition.”
If 150 N. Am. trees were positively correlated with temperature at the 5% level (1-tailed test), that is consistent with no relationship at all if the original universe was 3000 trees. If there was an inadequate correction for serial correlation, which is easy to do, the universe could have been much smaller. (416 significant trees in all would be consistent with a universe of 8320 trees.)
If one did have say 8320 treeline trees and wanted to establish a significant correlation with temperature, a multiple regression would fail with only 150 years or so of temperature data. An appropriate fix, consistent with the work of early dendroclimatologist Preisendorfer, would be to find the first few Principal Components of the data set, and then regress temperature on them, including CO2 as a variable and adjusting appropriately for serial correlation. If the first r PC’s pass an F test, then at least one of them is a valid indicator of temperature. (Testing them individually with t-tests would raise the cherry-picking problem again. Searching over r for a good F statistic is not a pure method, but is easy and adequate, I think.) Only the few PC’s that look good on a scree plot (or fancier method) would need to be considered.
Since in fact tree ring growth is a function of temperature rather than the other way around, the regression described above would give an inconsistent estimate of the strength of the relationship, and would give too weak a temperature reconstruction. CA contributor UC called our attention to multivariate “CCE” (Classical Calibration Estimation) several years ago, but I think that it would be equivalent to running the above regression, using the estimated linear combination of PC’s, adjusted for C2, as the proxy index, and then regressing it on temperature, but preserving the numerator degrees from the original regression. The reconstruction is then obtained by inverting this regression line. My preference would be to compute a confidence interval (or rather credible interval since Bayes is invoked) using the Ratio of 2 Normals distribution, as discussed in my paper on Lonnie Thompson’s 2003 ice core index, rather than the traditional CCE method.
As Nick Stokes would say: “it’s nice to have”.
This is proof of willful ignorance or stupidity.
Hu, no matter the sophistication of the method used to analyze the data and test a model, the fact remains that the data used are in-sample and thus all the caveats in not using out-of-sample data must be applied.
This whole process brings to mind a memorable quote of mathematician David Berlinski, when commenting on the results of attempts to demonstrate the existence of self-replicating RNA:
“They began with what they needed and purified what they got until they got what they wanted.”
What an excellent quote! Might you have a source for it? Would love to use it in lectures.
Never mind – found it. Thanks so much for mentioning this.
http://bit.ly/2v5rE3u
JEG, thanks for your comments on this matter. I always respect a man willing to publicly defend his own work. This open discussion of the issues is at the heart of science.
Inter alia, you say:
Perhaps I’m misunderstanding, but it appears that you are using outcome data (correlation of recent tree rings to temperature) in order to select your input data (which trees you are going to analyze).
Surely you must see that this is what is known as “data snooping”? From UTexas:
This is a very clear description of exactly what you are doing. First you are looking at the data, seeing if it fits your hypothesis. After snooping the data, you are then eliminating those proxies that don’t fit.
If you still don’t see it, let me give you an example of data snooping.
Suppose I give some medicine at intervals to a group of people. It’s supposed to affect their body temperature. Some people appear to respond to the medicine, while other people don’t respond. However, we may also just be seeing the natural variations in daily bodily temperature.
So I toss out the records of the people that don’t respond to the medicine, and I proudly announce that my medicine works in 100% of the cases …
Here is the deal with data snooping. You can’t look at the variable of interest, in this case tree ring widths, in order to select which trees are INPUT to the procedure.
So what can you do? Well, you can use separate identifiable variables unrelated to the variable of interest (tree ring width) to select the input to the procedure. For example, you could say “I’m going to look at ponderosa pine trees in the Southern Cascade Mountains that are growing above 6,000 feet elevation and within 2,000 vertical feet of the nearest ridgeline”. That is a legitimate ex ante proxy selection criteria.
But you can’t say “I’m going to sift through all the trees in the Southern Cascade Mountains, throw away all of those whose tree rings don’t fit my criteria regarding correlation, and analyze the rest”. If you do that you are using the outcome to affect the input, you are making up your proxy selection rules post ante, and that’s data snooping.
Best regards,
w.
WE, your example of the medicine study is not only data snooping, it will get your application to the FDA handed back to you forthwith, and perhaps a g-man visit with a list of questions he wants to ask you.
So where are the g-men on climate science? Billions of dollars to Mannian fiascos parading as meaningful and informative insight when in reality they are just a big bowl of mashed cherries!
Still waiting for a rebuttal from Julien Emile-Geay of Willis’ data-snooping critique. Or perhaps from one or other of Julien’s 97 co-authors (I’ve only done a rough count). I’m sure Julien and co. won’t mind the publicity, and I hope Steve’s house rules aren’t infringed, if I copy the complete list of authors of the PAGES2K17 paper:
Julien Emile-Geay, Nicholas P. McKay, Darrell S. Kaufman, Lucien von Gunten, Jianghao Wang, Kevin J. Anchukaitis, Nerilie J. Abram, Jason A Addison, Mark A. J. Curran, Michael N. Evans, Benjamin J. Henley, Zhixin Hao, Belen Martrat, Helen V. McGregor, Raphael Neukom, Gregory T. Pederson, Barbara Stenni, Kaustubh Thirumalai, Johannes P. Werner, Chenxi Xu, Dmitry V. Divine, Bronwyn C. Dixon, Joelle Gergis, Ignacio A. Mundo, Takeshi Nakatsuka, Steven J. Phipps, Cody C. Routson, Eric J. Steig, Jessica E. Tierney, Jonathan J. Tyler, Kathryn J. Allen, Nancy A.N. Bertler, Jesper Björklund, Brian M. Chase, Min-Te Chen, Ed Cook, Rixt de Jong, Kristine L. DeLong, Daniel A. Dixon, Alexey A. Ekaykin, Vasile Ersek, Helena L. Filipsson, Pierre Francus, Mandy B. Freund, Massimo Frezzotti, Narayan P. Gaire, Konrad Gajewski, Quansheng Ge, Hugues Goosse, Anastasia Gornostaeva, Martin Grosjean, Kazuho Horiuchi, Anne Hormes, Katrine Husum, Elisabeth Isaksson, Selvaraj Kandasamy, Kenji Kawamura, K. Halimeda Kilbourne, Nalan Koc, Guillaume Leduc, Hans W. Linderholm, Andrew M. Lorrey, Vladimir Mikhalenko, P. Graham Mortyn, Hideaki Motoyama, Andrew D. Moy, Robert Mulvaney, Philipp M. Munz, David J. Nash, Hans Oerter, Thomas Opel, Anais J. Orsi, Dmitriy V. Ovchinnikov, Trevor J. Porter, Heidi A. Roop, Casey Saenger, Masaki Sano, David Sauchyn, Krystyna M. Saunders, Marit-Solveig Seidenkrantz, Mirko Severi, Xuemei Shao, Marie-Alexandrine Sicre, Michael Sigl, Kate Sinclair, Scott St. George, Jeannine-Marie St. Jacques, Meloth Thamban, Udya Kuwar Thapa, Elizabeth R. Thomas, Chris Turney, Ryu Uemura, Andre E. Viau, Diana O. Vladimirova, Eugene R. Wahl, James W.C. White, Zicheng Yu, Jens Zinke.
Do they all go along with Julien’s defence?
A few familar names catch the eye: Rapheal Neukom, Joelle Gergis, Eric J. Steig, Chris Turney, Eugene R. Wahl and others.
Much quoted, but never–in my opinion at least–out-of-date: “The first principle is: You must not fool yourself, and you are the easiest person to fool.” –Richard Feynman I totally get the temptation to make a splash in science, as that will probably lead to more and better future funding, etc. But isn’t there, somewhere, in our collective scientific soul, a commitment to be true scientists, to make every possible effort not to fool ourselves? So when we screen ex post, does not the ghost of Feynman arise and disturb our sleep at least a little?
I think this paper can be readily ignored due to this revealing statement by the Prime Minister of Australia.
http://www.independent.co.uk/news/malcolm-turnbull-prime-minister-laws-of-mathematics-do-not-apply-australia-encryption-l-a7842946.html
All mathematical contributions by any Australian must be considered unreliable since the laws of mathematics don’t apply in Australia.
Matt Skaggs is right. Scientists such as Lindzen, Curry, Tsonis got it wrong when identifying the source of #ENSO and #QBO. Have to identify the analogous “bird signatures” FIRST before producing the models, otherwise it will stall research for generations. http://ContextEarth.com/2017/07/06/confirmation-bias/
There is active modelling to hindcast and forecast yields of major crops with data from decadal time spans under climate change. Lately, wheat, soybean and maize in USA are studied by Mistry et al in ERL 12,7 of 2017.
Study this paper to see how hard it is to weave climate factors into this major crop production with much supporting data.
Then think about how confident one should be when projecting tree growth back for centuries, often with no information about variables interacting with temperature and yield.
It requires a leap of faith that has no place in hard science. It is prudent to dismiss all of the tree series outlined by Steve above, instead of picking and choosing.
Geoff
Tree line timber at high altitudes can be shown to be more subject to moisture, as a growth factor, than to temperature, as in the situation where there is a slope. Soil is thin and holds limited moisture, water drains downslope by gravity, scant snowfall means less moisture and higher temperature means higher rates of evapotranspiration. Hence, changes in the tree line may reflect changing climate with respect to moisture availability instead of temperature changes. This is climatology, but climate scientists are not climatologists. Climatology is a multi-discipline science, you see.
Indeed. Maybe that sparse treeline you cored today in a proposed “temperature limited” region belonged in a healthy forest centuries ago. The surviving trees are maybe the least sensitive to cold, but how sensitive are they now to warmth ie the parabola effect? Additionally, where once they had to compete for sunlight and nutrients, their competition is dead and decomposing: more sunlight, more nutrients.
It’s all fine and dandy claiming you know a treemometer because right now you can test to make sure it correlates with temperature and the current environment suggests it could be one, but without accounting for the entire physical and climatic history of the tree and area, how can you even begin to claim you know it was a treemometer centuries ago? Never mind assume you know how much temperature change those wiggles correspond to to make comparisons? Far too much wishful thinking and not enough science.
Consideration of slope brings other potentially important factors into play. Spring snow melt timing will have a large effect on moisture availability and the further consideration of whether the slope is faced to North, South, East or West will be important as well as the degree of slope. For anyone who looks at these factors closely and openmindedly, the exercise would seem to be completely pointless. I can only conclude that these proxies are only used because they can be manipulated to produce what is desired.
JEG says: “Perhaps, having studied paleoclimate proxies their entire life”
Here’s another example of exaggeration used by alarmist advocates. He must think that chewing on the wood of his baby bed railing is studying “paleoclimate proxies”. He appears to be another milennial who thinks he knows more than the generation before him.
The Pages2017 data base contains only 3 of the 6 Thompson ice cores that were used to construct the famous “Dr. Thompson’s Thermometer” that Al Gore cited in his book and movie as providing “independent” evidence confirming the validity of the MBH Hockey Stick. The three included are Quelccaya, Dasuopu and Guliya. Huascaran (core 2), Sajama, and Dundee have been rejected, evidently for lack of correlation with instrumental temperature. See my 2010 post here, “Calibrating Dr.Thompson’s Thermometer”, at https://climateaudit.org/2009/12/10/calibrating-dr-thompsons-z-mometer/ , and Thompson et al, Climatic Change 2003.
As it happens, the graph Gore claimed in both his book and movie as being based on Thompson’s ice core data was really the MBH Hockey Stick itself, spliced together with instrumental temperatures, but that’s another story.
And as it happens, one of the key series used by MBH was Quelccaya, so that any ice core index containing it is not completely independent of the Hockey Stick. Thompson et al 2003 uses the Quelccaya Summit core. There is also a Quelccaya Core 1. I’m not sure which is in PAGES, but I’ll assume it was the Summit core for now.
In his 2006 PNAS article, Thompson also used a seventh series from Puruogangri, which is included in PAGES2017, but I was unable to figure out from his description how he had incorporated it, and so did not attempt to calibrate the resulting 7-core composite index.
I am curious about the Figure 8, Tree Records at Pages2017. It is a global proxy agglomeration. At Pages2k 2013 the regions were plotted separately, see its Figure 2. At this Figure 2 it looks like North American tree data is handled up unto the year 2000, but this would be an untrue reading. At the Pages2k 2013 Table 1 it is shown that the North American tree data “Time Period” is cut off at the odd date of “1974.”
I pointed this out in 2014 here at Climate Audit.
Given that the North American trees was their biggest record base, it was hard not to suspect that they found the post-1974 data problematic to their hoped for findings. If Pages2017 now uses N Am data up to 2000 (I cannot tell what the cut off date is) it may be the trees thrown out were the ones showing no warming after 1974, and the new ones added are some that do show post-1974 warming.
Follow the Money —
2017 Figure 1c shows that the number of TR series drops off precipitously after 1980 or so, so perhaps it was felt that after 1974 the sample was getting too small to be representative.
A similar problem occurs in the data base of Loehle and McCulloch 2008: Most of the last 2000 years is represented by 18 proxies, but this number falls off on both ends. In 1935 the number of tridecadally averaged series drops suddenly from 11 to 8 (less than half), so we decided to draw the line there. The effect of declining sample size on the standard deviation of the average is discussed and illustrated in the SI I wrote and posted at http://www.econ.ohio-state.edu/jhm/AGW/Loehle/ .
Mann (2008), nor his paper referees nor his defenders have ever had a problem with lack of proxy data nor even truncating and replacing proxy data that goes against the modern day warming trend (divergence).
Mann(2008) Figure 2 shows compilations of reconstruction proxies with divergence during the modern warming period for dendro and non dendro proxies. If the paper were written in the true scientific spirit the main topic of the paper would have been: Given the general divergence problem, and even in the face of a post facto selection process, how can we rationalize the proxy responses as reliable thermometers for past climate?
http://www.pnas.org/content/early/2008/09/02/0805721105.full.pdf+html
The paper briefly and begrudgingly admits to a general problem with divergence, but without addressing the problem head on.
“Interestingly, although the elimination of all tree-ring data from the proxy dataset yields a substantially smaller divergence bias, it does not eliminate the problem altogether (Fig. 2B). This latter finding suggests that the divergence problem is not limited purely to tree-ring data, but instead may extend to other proxy records.”
If one is willing to wade through the individual proxy series divergence can be seen in many proxy responses to temperature. The Briffa MXD series is certainly not a one-time thing in the world of proxy divergences. Divergence is something to be expected from reconstructions where proxies that are truly not reliable thermometers are selected after comparing the response to the instrumental record and running out of proxies – as chance would predict – that turn upward in the modern warming period.
Attaching the instrumental record to a reconstruction merely takes attention away from the divergence and allows the implication that the proxy responses are on par with thermometer responses without having to explicitly state such nonsense.
“Because of the evidence for loss of temperature sensitivity after 1960 (1), MXD data were eliminated for the post-1960 interval. The RegEM algorithm of Schneider(9) was used to estimate missing values for proxy series.
The RegEM algorithm of Schneider(9) was used to estimate missing values for proxy series terminating before the 1995 calibration interval endpoint, based on their mutual covariance with the other available proxy data over the full 1850–1995 calibration interval. No instrumental or historical (i.e., Luterbacher et al.) data were used in this procedure.”
“Because of the evidence for loss of temperature sensitivity after 1960 (1), MXD data were eliminated for the post-1960 interval.”
That is one of the most hilarious and telling lines of all time. It never gets old! So…the MXD trees suddenly “lost their previous sensitivity to temperature” in one exact year…1960. Inexplicably, and completely. And yet, it doesn’t even cross your mind that such an event could have occurred prior to 1960? Maybe even often??
Courses in logic and critical thinking should be REQUIRED to graduate from both high school AND college.
Thanks for the reply Hu. Here is more:
From Table 1 of Pages2k (2013) here are the breakdowns for end periods (“Youngest”) for tree rings by region: Arctic, 2000; Europe, 2003; Asia, 1989; N Amer, 1974; S Amer, 1995; Aust, 2001.
As you can see, the cut off for North American trees is quite aberrant. Such is not explained by the 145 N. Amer. tree proxies used. Here their “Youngest” data is categorized by decade:
1970s: 1 (“1979”)
1980s: 51
1990s: 63
2000s: 30
There is no good reason in the N. American data for selecting an end date of “1974” except by inference that the warming cycle since 1974 was not reflected well in the rings of these proxy trees.
Also, the majority of these North American proxies are listed as negative (see Column J: “Sign Relation to Temp”). 60 of these proxies are positive with temperature. 85 are negative. This could very well explain why the N. Amer tree proxies were cropped to the year 1974. What other could be?
It does sound more like a case of “hide the decline” than small sample concerns.
I have a suggestion for one of the experts here:
It seems like Steve Mc argues that if you take a bunch of random data, and screen out everything but the random data that goes up at the end, you’ll automatically get a hockey stick (true).
JEG comes back with the argument that he’s not screening for going up at the end, he’s just trying to screen for what actually correlates to temperature, which happens to have increased in the 20th century.
Even if JEG is correct there are huge issues there, but nm that for now.
What if the proxy selection criteria was exclusively based on correlation to NEGATIVE changes in the 20th century. That is, despite the fact that the temp went up in the 20th, there were many year-on-year negative changes in particular locations. Pick only those trees which accurately correlate to THOSE changes and utterly ignore positive changes.
Then what do you end up with?
“Pick only those trees which accurately correlate to [negative] changes and utterly ignore positive changes.Then what do you end up with?”
A study with the same methodology that comes to the opposite conclusion based on its assumptions, i.e. the study merely restates its assumptions (at least to the extent they filter out data based on the same variable they’re measuring).
It would be merely bad if they chose an ex ante criterion that also happened to correlate to temperature. To measure temperature by throwing out records because they don’t correlate to assumed temperature trends is so gobsmacking it lends credence to every criticism of the whole field.
Nobody said anything about selecting records based on correlation with assumed temperature trends. Rather, they are selected based on correlation with instrumental temperature, which in fact has a trend.
My point is, suppose a zone where temp went up 1910-1940, down 1940-1970, up 1970-2000. Then pick ONLY trees that went down 1940-1970. Ignore the rest.
What happens then? A hockey stick or… ?
Some years ago when experts were writing papers to explain the cause of the C21 global temperature hiatus, I noted the opportunities this created for proxy students. First, if temperature does not change, you cannot construct a time response curve or calibration graph for T=constant, no matter what you regress against T. Second, for constant T, variables usually linked Go T are simply giving a neat expression of their prevailing noise patterns.
Then Tom Karl popped in (not poppered in) and this happy scene was spoiled – if he was right that there was no pause.
Given that the temperature change in C21 could still be quite small, this poses more questions about whether the proxy you are measuring is actually proxy noise or is recalcitrant trees, hence a need to reject or not. If it is just noise, the both plus and minus excursions are valid and no trees should be rejected. Only the result of the study changes – instead of measuring temperature by proxy one measures inherent variability of the proxy, which is also useful. Geoff
“Nobody said anything about selecting records based on correlation with assumed temperature trends. Rather, they are selected based on correlation with instrumental temperature, which in fact has a trend.”
Sorry if that was unclear. The implicit assumption in doing such a selection is that the instrumental temperature trend is the actual temperature trend, hence “correlated with assumed temperature trend.”
Again, as Steve points out and you correctly surmise, selection on the dependent variable amounts to assuming your conclusions, and in your example you could use the same methodology to arrive at opposite conclusions.
eloris: Year-to-year variation in local temperature at many sites is larger than the long-term change in warming. (Where I live, mean summer temperature varies by by more than 1 K from year to year. So if one detrends both the tree ring and temperature data before looking for a correlation, that correlation is likely to be telling you about the response to annual changes in temperature, not longer trends. Detrending has been a popular subject at ClimateAudit, but it doesn’t appear to have been discussed for this paper. (Given the large number of sites that have been rejected as temperature proxies, I’d guess that the correlation test was done on detrended data.)
I suspect that it should be possible to qualitatively assess the validity of a screening procedure. Years with major volcanos could be used to compare “temperature sensitive” and “temperature insensitive” sites. If one’s screening procedure affords “temperature-sensitive” sites that aren’t any cooler after the 20 largest volcanos in the last 1000 years, that would be a good reason to be skeptical. If the average of 20 post-volcanic periods is colder than the pre-volcanic decade at temperature-sensitive sites and not at temperature-insensitive sites, that would provide independent evident that the screening procedure was working. You may remember Willis challenging BEST with “Spot the Volcano”. One volcano may be hard to spot in the historic record, but an overlay of the periods around 5? volcanos produced an unambiguous signal.
If it was done on detrended data, wouldn’t that eliminate Steve Mc’s argument that the process would automatically create a hockey stick?
This raises an interesting point. Year to year changes in temperature are most probably very small on average, perhaps 1 or 2 degrees C. Some Spring days may be warmer this tear and some Summer days warmer the next year. And yet we are expected to believe that ring growth correlates with temperatures. What temperatures? Spring? Summer? Early Spring? Indian Summer? I realize this is the basis for some people’s entire careers but the whole idea is absolutely preposterous to me. Casting chicken bones is just about as meaningful. It’s a joke!
As Iowahawk might say, “Your Honor, that incident where I drove my truck into the liquor store should be thrown out as it is negatively correlated with my all my other driving that did not result in damage to liquor stores.”
I always miss an understanding of the methodolgical ideas behind the use of tree rings. From the discussions it seems that dendrologists are stupid peaople. I looked a little bit into it.
Tree Ring Data for Climate Reconstruction and Indication of Shifting Climate Regimes
by Valerie Barber, Institute of Marine Science:
“The three different parameters of the annual tree rings (width, density, and ð13C isotope concentration) measured in this study produce distinctly different climatic information. Ring width is the parameter most commonly studied. Large, thin-walled cells are laid down early in the growing season when conditions are favorable for growth. These are followed by smaller, thick-walled and denser cells late in the growing season which form due to the onset of cooler temperatures, lack of soil moisture, and shorter days. The production of latewood cells terminates abruptly, followed by larger cells the following growing season. This abrupt termination at the end of one year and the beginning of the next year marks a ring boundary. Ring width is measured between two successive ring boundaries. In previous Alaskan studies, ring width has correlated with annual temperature.
Cell density is measured across the length and width of the tree ring, and there is intra-annual as well as inter-annual variation. Maximum density is the parameter that is typically measured and is usually correlated with summer temperatures here in Alaska. Wood density may be affected by factors other than those which affect tree ring width and may be more sensitive to environmental changes.
Stable carbon isotope ratio (ð13C) is correlated with moisture availability. This occurs because stomates remain open for longer periods when moisture is available, allowing for greater exchange of CO2 and selection of lighter carbon. Thus under moisture-limiting conditions or arid climates, reduced CO2 availability results in less discrimination against the heavier isotope and higher ð13C values.
The combination of width, density, and isotope analysis offers a powerful set of independent but mutually reinforcing tools for reconstructing climate because the effects of temperature and precipitation can be distinguished, and seasonal changes can be observed. A better understanding of how the boreal forests are affected by different climate parameters is necessary to predict response to future climatic changes.”
Yes, following on your point, searching tree ring isotope ratios brings up hundreds of truly scientifically dedicated dendro specialists. I wonder about the spectrum of emotions of those like Fritz Schweingruber regarding the hockey team’s use of their extremely complex biological studies. It certainly would be great to hear an opinion of a life-dedicated dendro who didn’t appreciate Mann and the hockey team’s use of their work, sometimes fitting it upside-down, and grabbing world headlines, and crediting dendro science. Searching Schweingruber on CA I can see Steve Mc is pretty well read-up on the science.
Looking at his CV, Julien Emile-Geay, seems to be a general practitioner climate scientist in the mold of Mann and Trenberth.
If the USC press release for JEG (2017) he sums up his views.
From his publication list I would class Emile-Geay, like Mann, but unlike Trenberth, as having specialised in multi-proxy paleoclimate reconstructions. This makes Emile-Geay (and to some extent his co-authors) career-threatenlingly vulnerable to serious critiques of his techniques such as those of Steve and Willis. I wouldn’t expect hime to make concessions any more than has Mann. Like Mann, he can be expected to go to great lengths to protect his career.
As far as i know of “calibration” it does not matter how it correlates.
You just “flip” the responce to the temperature. I still wonder if there are any investigations of why and under whitch conditions that tree rings responds to temperature. I belive there are some conditions that could give a clear responce, but can you be sure those conditions were present so many years ago.
Steve McIntyre, how do your reconcile your objection to screening with your various posts where you complain about use of a proxy that is not responsive to temperature or perhaps negatively correlated?
That’s easy! There should be a physical theory that specifies the direction of the proxy’s response to temperature. If the proxy’s response correlates oppositely to the physical theory in the calibration period, then it should not be inverted. Doing so is the most egregious case of using a lucky statistical noise to create a meaningless series. BTW, I do not suggest that the proxy should be omitted because of the negative correlation; if it satisfies the ex-ante selection criteria then it should be included. If the result is an unsatisfactory representation of temperature in the calibration period, then either the ex-ate criteria are incorrect OR you cannot reconstruct temperatures from tree rings.
Thanks for that succinct reply, rwnj. While I am well aware that most scientists working in the climate science area of temperature reconstructions with proxy data do not “get it”, I am always in hope that those who read/post at these blogs such as Climate Audit where SteveM has spelled out the case against post facto proxy selection would “get It”. Your reply shows that some here “get it”.
Yes, I agree with Ken Fritsch, a succinct reply, right on the mark. There is no contradiction in the criticisms of the faulty methods used by the tree ringers.
A tree’s ring width, at best, is an indicator of how close to optimal a growing season was, at the time and place, for that particular tree. Optimal growth occurs when all of the necessary and sufficient factors are within the optimum’s tolerance. Since exact correspondence will almost never happen, it will boil down to light, nutrients, water and luck with respect to timing of adverse weather and predation. Because of this, any effect that temperature has will be an inverted quadratic. Dr Bouldin had a series discussing dendrochronology a few years back. It is one thing to use rings for dating purposes. One best use here is cross-validation of other methods. Temperature reconstructions, beyond the grossest of approximations with large error bands, are not possible without having much more information at hand. Climate is a human abstraction that means nothing to biological life, including humans. Weather is what we live in and have to survive. The climate, on the other hand, means very little in that respect.
cdquaries: Yes, the effect of temperature on tree ring growth is likely to be an inverted quadratic, but that doesn’t mean that local temperature ever gets near the optimum or crosses over to the other side. Most sites used in reconstructions are at the northern or high elevation limit for survival of any trees precisely because the scientists who choose them are attempting to collect data from trees whose ring widths are most likely to vary linearly with temperature. To some extent, their success depends on whether trees fail to grow further north or at higher elevations because it is too cold in winter (when tree are not growing) or too cool in summer (when the trees normally would grow). There was a paper that showed that trees growing at the tree line on one mountain were temperature-sensitive, but those growing lower on the same mountain were not. So, during the MWP even trees growing at the current tree line might have been too low to respond fully to warmer temperatures.
At the infamous Yamal site, “fossil wood” from trees north of the current tree line that grew during the MWP suggests to me that the Yamal was “warmer” during the MWP than it is “today”. I put quotation marks around “warmer” and “today”, because we have no idea how long it takes a forest to start growing at a site that used to be too cold. Did the areas north of Yamal become warm enough to support trees in 2000 or 1980 or 1950 or 1900. Or perhaps It isn’t warm enough even today.
One fact all of us amateurs should be aware of is that one enzyme, RuBisCo, comprises almost half of the protein in the leaves of most plants. That tell us that the reaction catalyzed by this enzyme – incorporation of CO2 into organic intermediates – is the rate-limiting step in growth, including the growth of tree rings. When CO2 is entering leaves through stomata, water is leaving. So there are trade-offs between rapid growth and water consumption. This problem should be worst in late summer (after seasonal snow has melted) and and early fall (latewood density). TRW may be a rainfall proxy during these months. On the other hand, warm moist air from the south may produce rainfall at these sites, making TRW at proxy for both temperature and rainfall. Unfortunately, variation in temperature is correlated over relatively long distances (1000? km), but summer rainfall is not.
It doesn’t mean that it doesn’t, either. If we are talking about temperatures with plants, one must consider the leaf temperature, not just the air temperature, which may be quite different. Plants modify the local environment. When leaf temperatures get too high, remember that leaf temperature will vary greatly from sun angle and local shading, that leaf will emit VOCs as a consequence of local stress. These VOCs will undergo photolytic reactions if conditions for that are there. This will ‘shade’ the leaf, cooling it. Doing this, though, shunts energy away from building plant mass.
About RuBisCo, that enzyme is subject to oxygen poisoning, resulting in photorespiration.
About correlation, recall that correlation is not causation. To the extent that air temperatures correlate over distance, that would be due to local limiting factors and not-so-local bulk transport, in addition to other factors.
At least they are measuring tree rings. In the event of Her Majesty’s command to study her swans, there would need to be experimental adjustment for male homosexual swans andvtheir designs when swimming about with their heads submerged looking for cygnet rings. Or if gay geese, having a gander.
But is it science?
Geoff
Julien Emile-Geay (JEG) wrote: “However, the data are all public, and so is the code, so unless you want to make a staggering display of bad faith, you can’t complain about obfuscation. ”
The problem is that PAGES only gives the cherry-picked data, and not the universe of data from which it was selected, so that it does not, in itself, allow one to assess the effects of temperature on the proxies and therefore the ability of the proxies to reconstruct temperature, as claimed by Julien.
Here is a suggestion for climatologists that would allow them to legitimately down-weight, if not altogether eliminate, series that are only weakly or even negatively correlated with a common climate signal present in the majority of the class under consideration, without cherry picking the data as has been done by PAGES:
Assuming for simplicity that all series in the class under consideration (say treeline treering widths) have the same coverage over time, standardize each series to zero mean and variance over the entire period (not just the instrumental calibration period). Average these Z-scores as Julien is apparently doing in his post “The Hockey Stick is Alive” at https://strangeweather.wordpress.com/2017/07/11/the-hockey-stick-is-alive-long-live-the-hockey-stick/ and as Pages2017 itself apparently does in its summary illustrations.
But then, if there is a common temperature signal in the data, it will be partly obscured by the presence of several series that aren’t particularly correlated with the signal, and perhaps even some that are accidentally negatively correlated. If we look at the residuals of each Z-score series about the date-specific means, the “good” ones will have variance less than 1, the “indifferent” ones will have variance still approximately 1, and the “bad” ones will have variance greater than 1. These variances may then be used to re-average the Z-scores, not with equal weights but with “Weighted Least Square” (WLS), which weights each observation inversely to the variance of its first-stage residuals.
The resulting series will place the highest weight on the “good” series that correlate well with the consensus signal and very low weight on the “bad” ones that are contrary to the consensus signal, without using any instrumental data to perform the weighting. The WLS series may then be regressed on temperature and CO2 (in the case of TRW) and the relation solved for temperature to reconstruct pre-instrumental temperatures. Since the signal will be clearer in WLS average than in the original OLS average, any correlation with temperature should come through more clearly and significantly using the WLS series.
The approach can be refined for spatial autocorrelation by replacing WLS with “Generalized Least Squares” (GLS), aka Aitken’s rule, which takes correlations into account. Unequal temporal coverage is messy, but not insurmountable.
Hu, I’m reminded of a method that I invented a while back to determine the amount of signal in a dataset. What I did was sequentially remove from the dataset the individual timeseries that most poorly correlates with the average value. Hang on … OK, I found it, Can’t See The Signal For The Trees, from ten years ago.
Thanks for your interesting comment,
w.
Willis,
Prophecy fulfilled!
“6. At some point, after Steve figures out Mann’s method, the proponents of Manns work are sure to claim that the hockeystick signal is really there, regardless of the method used … yes, it really is there, but only in the Tiljander and bristlecones. Garbage in, garbage out …”
one of my long-standing disputes with present paleoclimate is the fixation on trying to find a magic multivariate method to extract the “signal”. If there’s an actual “signal” in a reasonably large data set, it will be extracted with pretty much any method. A principal components calculation (singular value decomposition) also calculates weights to explain the most variance. Your method might be a form of algorithm that yields something close to a PC1. Principal components work fine when the data set contains a “signal”.
The problems arise because the data isn’t consistent. In my opinion, progress in the field can come only from careful parsing and reconciliation of regional data.
While JEG has sneered at mineral exploration, procedure of working out from good data has much to recommend to paleoclimate.
Thanks, Steve. I was thinking of my method regarding Hu’s thoughts about trimming out the “deadwood”, to coin a phrase. My method allows you to pull out those trees that do not agree, not with the temperature, but with the signal in the other trees. As such, it’s not data snooping … at least I think not.
The advantage to my method is that it shows graphically the datasets that do or don’t contain the signal, so you can break apart the dataset properly.
w.
PS—I’d be very interested to hear a bit more about what you are referring to when you say a “procedure of working out from good data has much to recommend to paleoclimate.”
PPS—One of your earlier insights has stuck with me, which was when you said that all that the various methods are doing is putting different weights on each of the various proxies … that cleared up a lot of things to me regarding a “magic multivariate method” you mention above. I’d looked at it from the front end, and not from the final result. No matter how complicated the methods, they are all just adjusting weights.
Willis: If the criteria you use to exclude is based on an observation/metric that is independent of temperature then I agree. For example, strip bark trees have problematic growth history because they have been or are damaged. Similarly, you might exclude/include trees that are too far below or above the current tree line. I am less sure about the notion that they are excluded because the “signal” is different without reference to some biological or physical characteristic. It seems to me that comes close to snooping.
Willis —
The method you describe in your 2008 CA post is very interesting, and is somewhat like the proxy “culling” I experimented with in my SI to Loehle and McCulloch (2008), at http://www.econ.ohio-state.edu/jhm/AGW/Loehle/ .
The major difference is that whereas you are looking at the correlations of each proxy with their average, I am looking at their variance about the average. In fact, both are relevant and should be taken into account unless we are dealing with only a relatively small number of well-distributed, equal variance proxies.
Your graphs show that the results of Mann 2008 depend disproportionately on 3 Tiljander series (thickness, lightsum and darksum), and after that on several SW US Graybill treering series. These series may contain some climate information, but clearly they are going to be positively correlated, and therefore not give as much information as if they were uncorrelated. Generalized Least Squares (GLS aka Aitken’s formula) would correct for both unequal variances in proxies (as I assumed in my comment above) and correlations across proxies.
The error bands on the composite series shown in PAGES2017 are supposed to be based on the bootstrap procedure of Effron and Tibshirani. Even apart from the cherry picking issue, it is likely that these error bands do not take the spatial correlations into account, which, with the much larger number of proxies, are more severe than in Loehle and McC. The PAGES (and Julien’s) error bands therefore may be much too small.
Another problem, as you may recall from the Lake Tanganyika study, is that climatologists sometimes confuse the bootstrap and jackknife procedures of Effron and Tibshirani, and get the formulas wrong. PAGES does not explicitly give the formula they use in either the text or the SI, so it should not be taken for granted that they got the formula right.
Frank Posted Jul 22, 2017 at 12:52 AM
Frank, thanks for the comment. Despite how logical that sounds, as someone who grew up in a forest up near the tree line, I can assure you that this is not true. I suppose this is an advantage of doing science from the bottom up rather than the top down.
The missing link that you’ve left out is moisture. Plants, in general, are not just temperature limited. They are limited by a combination of temperature and water. Think of the plants in your garden. If they have enough water a hot afternoon doesn’t do anything to them.
But if there is not enough water, they will droop and sag, suddenly they are on the downhill side of the inverted quadratic. And now, that same temperature at which they grew happily with when they had water may now kill them.
Next, consider that up near the treeline we usually have a number of things going on:
• The soil is thin and rocky.
• Thin rocky soil holds water very poorly.
• The air is often dry, with the water wrung out of it on the way uphill.
• Wind is often high.
• Evaporation is a function inter alia of wind.
• Thin air lets in strong sunshine.
• Rainfall is often low compared to the foothills.
And as I used to see every summer, the combination of all of these factors means that the trees up near the treeline are often visibly stressed, branch tips turning brown, far into the downhill side of the inverse parabola, while trees a few hundred metres lower were doing fine at essentially the same temperature.
Don’t feel bad, though. I’ve seen lots of stuff from respected dendroastrologists demonstrating they hadn’t thought about this either.
Best regards,
w.
WE is correct. I append my comment from five days earlier:
Posted Jul 17, 2017 at 1:45 PM | Permalink | Reply
Tree line timber at high altitudes can be shown to be more subject to moisture, as a growth factor, than to temperature, as in the situation where there is a slope. Soil is thin and holds limited moisture, water drains downslope by gravity, scant snowfall means less moisture and higher temperature means higher rates of evapotranspiration. Hence, changes in the tree line may reflect changing climate with respect to moisture availability instead of temperature changes. This is climatology, but climate scientists are not climatologists. Climatology is a multi-discipline science, you see.
Thanks, MPainter, I’d missed that one.
w.
Willis, the construction of ring width “chronologies” according to standard paleoclimate practice loses one of the most important pieces of information: the average ring width. This is an important and unnecessary loss of information. And JEG expects us to tug our forelocks in deference to such abysmal practice.
Several years ago, I extracted average ring widths for bristlecone sites according to altitude. Rather than the highest sites having the narrowest ring widths, the exact opposite. The ring widths declined with lower altitude – indicating (unsurprisingly) that bristlecones (in a high desert) are moisture limited with the moisture limitation least at altitude. That bristlecones are moisture limited is easily seen in photographs in which they compete with sagebrush.
Dendroastrology … the gift that keeps on giving.
I think there’s a typo in your comment though. You say:
Thanks,
w.
Steve: ugh. fixed.
Willis: I completely agree with your comments. (Been backpacking in the High Sierra.) In fact, immediately below I wrote:
“So there are trade-offs between rapid growth and water consumption. This problem should be worst in late summer (after seasonal snow has melted) and and early fall (latewood density). TRW may be a rainfall proxy during these months. On the other hand, warm moist air from the south may produce rainfall at these sites, making TRW at proxy for both temperature and rainfall.”
My mistake came in this sentence: “Yes, the effect of temperature on tree ring growth is likely to be an inverted quadratic, but that doesn’t mean that local temperature never gets near the optimum or crosses over to the other side.”
I should have said that the local temperature never gets near the optimum or crosses over to the other side given adequate rainfall. And “adequate rainfall” rises with temperature. The problem isn’t high temperature per se. For tree rings to grow wider in warmer years, trees may need more rainfall than in an average growth year. Ex post screening for “temperature sensitivity” at high altitude sites may mean screening for sites where warmer temperature is usually associated with higher rainfall.
Instead of picturing an inverted quadratic plot of growth vs temperature, a three-D plot of growth vs both temperature and rainfall is needed. High latitude and high elevation sites could be very different.
One might remember that trees are simply little factories that turn CO2 into the carbohydrate found in tree rings. The factory is always starved for CO2 and getting CO2 in lets water out. One needs a little nitrogen and phosphorus to make amino and nucleic acids (which may be adequately supplied by decomposing dead trees. And potassium for ion channels. I’m not sure whether the factory works faster when it is warmer, or whether the factory stays open longer when it is warmer in the spring and fall.
Dozens of variables, of which temperature might well be among the least important. Slope and side of the mountain the tree grows on relative to sunlight are others, as well as the side of the mountain relative to prevailing winds which tend to drop more rainfall on the downwind side. The notion that this is a science is ridiculous! It is a symptom of government fiscal diarrhea.
IMHO, it is past the time to state unequivocally that it is scientifically invalid to use tree rings from naturally-grown trees older than the instrumented era as a proxy for the air temperature in which they grew.
In other words, classical dendrothermometry as we know it is dead. Geoff.
I agree, Geoff. Without bad science they would have no science. This has always been my view.
Re bad science, there is a good test for that:
1) Randomly core four trees from each location of a past study.
2) Re-label them with a reference code and send them for dendrochronological analysis.
3) Create a plot of local instrumental record for each location of study.
4) Re-label the instrumental plots with reference codes.
5) Allow JEG and others to independently match the corresponding set pairs.
6) Statistically determine the degree of scientific validation.
Tree ring “calibration” is better known as “selecting on the dependent variable”. It is FORBIDDEN in statistics and it INVALIDATES your results.
It is unfortunate that Climate Science hasn’t taken the time to apply
fundamental statistical controls to their methods.
“Technically, sampling on the dependent variable is when you select cases on the basis of meeting a criteria and then use those cases as evidence for the criteria.”
Sampling on the Dependent Variable
gabrielr.bol.ucla.edu/soc210a_f09/w9.pdf
Don’t select on the dependent variable in studying the science …
http://www.culturalcognition.net/…/dont-select-on-the-dependent-variable-in-studying-the-s...
Nov 19, 2013 – “Selecting on the dependent variable” refers to the practice of restricting one’s set of observations to cases in which some phenomenon of interest has been observed and excluding from the set cases in which the phenomenon was not observed.
Selection bias — Crooked Timber
crookedtimber.org/2003/07/13/selection-bias/
Jul 13, 2003 – But it still sounds as though these books commit a serious social-science sin – “selecting on the dependent variable.” What does this mean?
The problem with selecting on the dependent variable is that your underlying hypothesis is that tree rings are a reliable proxy for temperature.
By selecting those trees that appear to correlate with temperature you are artificially making appear that the hypothesis is true. However, the trees you reject are telling you your hypothesis is false.
Thus, the trees that appear to correlate with temperature may not be responding to temperature at all, they may be correlating accidentally, of in response to some other environmental factor entirely.
Thus the conclusions you make statistically from the data will appear much more reliable than they actually are.
ferdberple: “Tree ring “calibration” is better known as “selecting on the dependent variable”. It is FORBIDDEN in statistics and it INVALIDATES your results.”
I think we all agree that the field’s validation practice of selecting a certain component of the time series for statistical correlation is unsatisfactory as it can be easily snooped (given enough time) on every core sample.
To achieve scientific validation a hypothesis must demonstrate statistical skill that is proven independent of the investigator’s skill. The several methods to achieve this include: setting up analysis blinds, making skillful predictions based on data yet to exist, having skillful use reproduced by adversaries, and preferably a all three.
Tree rings may have been abused but this doesn’t mean they don’t hold out the brightest hope for an eventually proven reliable temperature proxy. If all ring growth influences can be correlated to markers systematically and predictably to the point that an investigator can decipher the full climate history, drought, flood, predation, infestation, sunlight, wind, average temperature and range of temperature extremes, then they can demonstrate that skill with blinds and predictions. Then they just need to demonstrate that growing season temperatures are indicative of year-round temp.
Thanks for that, Ron Graf, you gave me a chuckle.
“Then they just need to demonstrate that growing season temperatures are indicative of year-round temp.”
Ron – that is a point that has baffled me. With a growing season of 3-5 months, how do you extrapolate the remaining months to arrive at annual temps.
I think they tend to report that this reconstruction is of temperatures in those months.
ferdberple: This is not exactly true. The problem is that at some sites (dry sites) the tree mainly responds to moisture. At other sites, temperature is the limiting factor. You want to select those sites where temperature is the limiting factor. The problem is not that they are “selecting on the dependent variable” but that they are assuming that all else (mainly precipitation) stays constant going back 2000 years, whereas we know this is unlikely. Over a 2000 year period in Sweden, even elevation can change due to isostatic rebound.
RE: Craig Loehle
“. . .The problem is that at some sites (dry sites) the tree mainly responds to moisture. At other sites, temperature is the limiting factor. You want to select those sites where temperature is the limiting factor. . . ”
It was my understanding that the ‘thing’ that is most significant to growth (or not) is the ‘thing’ that is most lacking. Whether that’s sunlight, moisture, temperature or whatever else needed for ideal conditions. Perhaps some expertise from those who know would help here.
-barn
CO2 might also be the limiting factor.
Speaking of Sheep Mountain, it appears that Linah Ababneh is now out of Witness Protection and a Senior Research Associate at Cornell!
See https://climateaudit.org/2007/12/13/malcolm-hughes-and-the-witness-protection-program/ ,
vivo.cornell.edu/display/lna22
Rosanne D’Arrigo once explained to an astounded National Academy of Sciences panel that you had to pick cherries if you wanted to make cherry pie …
Climate change politics is really just motherhood and cherry 🍒 pie – how can anyone be opposed to viscous class war and profiteering based on trumped up and vacuous scare stories? What’s not to like? After all – we’re all human, aren’t we?
Thanks, ptolemy2. I gotta say, “viscous class war” is one of the better descriptions of the climate farrago that I’ve seen, even if inadvertent.
w.
But many scientists worry that the 0.05 threshold has caused too many false positives to appear in the literature, a problem exacerbated by a practice called P hacking, in which researchers gather data without first creating a hypothesis to test, and then look for patterns in the results that can be reported as statistically significant.
https://www.nature.com/news/big-names-in-statistics-want-to-shake-up-much-maligned-p-value-1.22375
Thanks for the link to the Pvalue and .05 threshold. One of my first experiences with false positives is the concept of “premature mortality” due to ground level ozone. Numerous studies showed a link/ high correlation with increases in ground level ozone an premature death. However, there was a much higher correlation with increase in heat with premature death, yet the heat was ruled out. In several cities with relatively low ground level ozone, the increase in premature death was greater than the premature mortality of cities with high levels of ground level ozone.
While my response is off topic, it does point to the sloppyness of the use of statistics. (if a layman can find basic & massive errors in peer reviewed studies, then how reliable can the study really be)
I’m also reminded of the analysis by actual statisticians of standard practice:
“In this paper, we assess the reliability of such reconstructions and their statistical significance against various null models. We find that the proxies do not predict temperature significantly better than random series generated independently of temperature. Furthermore, various model specifications that perform similarly at predicting temperature produce extremely different his- torical backcasts. Finally, the proxies seem unable to forecast the high levels of and sharp run-up in temperature in the 1990s either in-sample or from contiguous holdout blocks, thus casting doubt on their ability to predict such phenomena if in fact they occurred several hundred years ago.”
Click to access aoas_mcshanewyner.pdf
“The degree of controversy associated with this endeavor can perhaps be better
understood by recalling Wegman’s assertion that there are very few mainstream
statisticians working on climate reconstructions [Wegman, Scott and Said (2006)].
This is particularly surprising not only because the task is highly statistical but
also because it is extremely difficult. The data is spatially and temporally autocorrelated.
It is massively incomplete. It is not easily or accurately modeled by
simple autoregressive processes. The signal is very weak and the number of covariates
greatly outnumbers the number of independent observations of instrumental
temperature.”
yeah…that
The cherry picking issue already invalidates any conclusions from the PAGES 2017 data set. However, there are a couple of statistical issues that would invalidate the summary graphs in Figure 7 as well as in Julien’s strangeweather blogpost “The Hockey Stick is Alive; long live the Hockey Stick”, even if the data had not been pre-selected:
First, the 95% bootstrap confidence intervals are undoubtedly based on the usual bootstrap assumption that the errors are independent. In fact, many of the 692 records are concentrated in a few regions — tree rings in western N.Am. and central Asia, ice cores in Greenland and Antarctica, marine sediments in the N. Atlantic, etc., see PAGES figure 1. This means that many of the errors are in fact highly correlated and the confidence intervals much too small. I don’t see any way to modify bootstrapping to take this into account.
A reasonable parametric approach would be to assume that correlations die off exponentially with great-circle distance as in “exponential kriging”. Most of the cross-sectional non-normality of the errors can be accounting for without bootstrapping by estimating the variance of each series across time as in Loehle and McCulloch 2008. See the SI at http://www.econ.ohio-state.edu/jhm/AGW/Loehle/ for details.
Second, the graphs PAGES and Julien show are composites constructed, as is common in climate studies, by averaging series that have been normalized to have zero mean and unit variance. This causes no distortion if each series is observed for the entire time period under consideration. However, if some are observed only for a shorter period, normalizing them to have zero mean over their own period tends to smooth out any long-run shape that might be present. In particular, it will tend to smooth out the LIA and MWP (if present) relative to recent fluctuations.
If the series are all progressively shorter, this can be easily corrected by zeroing out the longest ones over the entire series, and then centering each shorter series to have the same mean over its period as the longer ones do over its period. If they overlap randomly like weather station data, it is possible to find consistent offsets by solving a system of N+T equations in N+T unknowns, where N is the number of series and T the length of the total period. (There are actually N+T+1 equations in N+T unknowns, but one equation is redundant so it works out.)
In Loehle and McC 2008, most of the 18 proxies averaged ran for most of the 2000 period, so the second problem was not a big issue. However, Calvo quit in the 15th century, so that ideally it should have been centered to have had the same average over its period as the average over the other 17 series over its period.
Reblogged this on Quaerere Propter Vērum.
7 Trackbacks
[…] Steve McIntyre’s critique [link] […]
[…] Steve McIntyre’s critique [link] […]
[…] https://climateaudit.org/2017/07/11/pages2017-new-cherry-pie/ […]
[…] Source: PAGES2017: New Cherry Pie « Climate Audit […]
[…] about to vanish in one or two years, or ten years; climate scientists continue to be accused of selecting data sets to create hockeysticks and manipulating data; and teams of climate scientists keep producing reports saying almost exactly […]
[…] about to vanish in one or two years, or ten years; climate scientists continue to be accused of selecting data sets to create hockeysticks and manipulating data; and teams of climate scientists keep producing reports saying almost exactly […]
[…] Pages 2017 New Cherry Pie […]