Michael Mann is out of bed, and all the action is on Twitter.
Oct 21, 2013 at 3:02 PM | Unregistered CommenterRichard Tol
Shot 1: Wilson on millennial temperature reconstructions at BH. Which include the phrase “crock of xxxx” which is explained in comments as
Lastly, the “crock of xxxx” statement was focussed entirely on recent work By Michael Mann w.r.t. hypothesised missing rings in tree-ring records (a whole bunch of papers listed below). Although a rather flippant statement, I stand by it and Mann is well aware of my criticisms (privately and through the peer reviewed literature) of his recent work.
So, Rob Wilson has not described everything Mann has published as a “crock of xxxx”.
Volleys on Twitter include: @flimsin , @MichaelEMann and others. Remember that if Mann has blocked you, you can probably still read the tweets by not logging in and loading his stream directly. I’m not going to try to search or organize especially since the volleys seem to be flying right now. Anyone who knows how to make well organized twitter streams, I encourage you to do so. I know it can be done, but I don’t know how to do it.
I lied. My favorite Mann tweet so far is this hour old tweet
Naturally, a declaration of this sort is followed by plenty of other tweets.
“The real fireworks came when Mann’s latest papers, which hypothesise that tree ring proxies have large numbers of missing rings after major volcanic eruptions, were described as “a crock of xxxx”.”
Mann has used conjecture previously in attempts to add to the credibility of his stand on climate issues. It was discussed a few years ago at CA when past and future tropical storm statistics were at the most controversial and was related to the observations that past tropical storm events were probably under reported in the past (thus making a spurious upward trend in the frequency of events) because the past events were mainly reported out at sea by passing ships. Mann come up with a conjecture in order to counter what seemed like a reasonable view by noting that modern ships will avoid tropical storms because of the detection equipment the ships carry and in the past the ships without that equipment would wander right into these storms and thus (if they survived to tell about it) would experience and report the storms.
A few us posting at CA snickered at this as a throw away line in a technical paper. I called it the dumb ship conjecture. Had it ended there I would be less concerned, but a few months later another paper referenced that line and in doing so seemed to add to its credibility. I suspect a tertiary reference might have added even more credibility.
I’ll have to find the recent Mann paper and determine whether the missing tree rings is hypothesized or conjectured.
Kenneth
If it helps, Rob wilson supplied a list of papers on Bishop Hill:
MISSING TREE-RINGS AND MAJOR VOLCANIC EVENTS
Mann et al. 2012. Underestimation of Volcanic Cooling in Tree-Ring Based Reconstructions of Hemispheric Temperatures, Nature Geoscience, 5, 202-205.
http://www.meteo.psu.edu/holocene/public_html/shared/articles/MFRNatureGeosci12.pdf
Anchukaitis, K. et al. (2012). Tree rings and volcanic cooling. Nature Geoscience. 5: 836–837. doi:10.1038/ngeo1645
http://www.st-andrews.ac.uk/~rjsw/all%20pdfs/Anchukaitisetal2012.pdf
Mann et al. (2013). Discrepancies between the modeled and proxy-reconstructed response to volcanic forcing over the past millennium: Implications and possible mechanisms. JGR. 118, 14, p. 7617-7627.
http://www.meteo.psu.edu/holocene/public_html/Mann/articles/articles/MRSTF-JGRInPress.pdf
Esper J et al (2013) Testing the hypothesis of post-volcanic missing rings in temperature sensitive dendrochronological data. Dendrochronologia. 31 (3): 216-222.
http://www.blogs.uni-mainz.de/fb09climatology/files/2012/03/Esper_2013_Den.pdf
Esper J et al (2013) European summer temperature response to annually dated volcanic eruptions over the past nine centuries. Bulletin of Volcanology 75, 736, doi: 10.1007/s00445-013-0736-z.
http://www.blogs.uni-mainz.de/fb09climatology/files/2012/03/Esper_2013_BullVol.pdf
St. George et al. (2013). The rarity of absent growth rings in Northern Hemisphere forests
outside the American Southwest. Geophysical Research Letters 40, doi:10.1002/grl.50743
http://www.st-andrews.ac.uk/~rjsw/all%20pdfs/Stgeorge2013.pdf
D’Arrigo, et al. (2013). Volcanic cooling signal in tree-ring temperature reconstructions for the past millennium, Journal of Geophysical Research, doi:10.1002/jgrd.50692
http://www.st-andrews.ac.uk/~rjsw/all%20pdfs/D'Arrigoetal2013.pdf
Thanks much, Diogenes, for linking those papers.
I should have commented in my previous post what a lucid writer Bishop Hill is. No James Joyce is he.
Kenneth,
“I’ll have to find the recent Mann paper and determine whether the missing tree rings is hypothesized or conjectured.”
.
Relevant papers reach the unanimous conclusion that Mann et al 2012 is indeed a crock…. The ‘missing tree rings’ paper was an attempt to explain why the tree-growth response to large volcanic eruptions indicates a much smaller temperature drop than the climate model ensemble predicts for major volcanic eruptions (or put more directly, that the tree ring data indicate the actual sensitivity of high latitude northern temperature to short term drops in forcing is much lower than the models indicate). Mann et al claim the cause is missing cold years (no ring!) and dating errors that “smear” a large temperature response in later years over multiple years. Pure poppycock. The paper falls in the large category of ‘you can’t prove the GCM’s are wrong that way either’ . Often Gavin is a coauthor of such papers, and I have to admit that I am surprised he wasn’t in on Mann et al 2012. Maybe even Gavin could see that the paper was rubbish.
.
So all subsequent papers show Mann et al 2012 is clearly wrong, leading to the obvious question: how did it ever pass peer review? Dozens of researchers ended up wasting time just to show that an obviously wrong paper in Nature is obviously wrong. Maybe the Nature Publishing group doesn’t care if what it publishes is rubbish so long as it thumps the drum of climate alarm. Hummm… sort of like Steig et al’s Antarctic warming nonsense.
.
The really surprising thing is that Mann’s seems always to get away with truly obnoxious behavior… he took down the offending comment when other climate scientists started complaining…. sort of like Steig ‘disappearing’ some of his most obnoxious comments on Steig et al Antarctic warming over at RealClimate. Never an apology given, just deceit to cover the offending behavior. Cowards all.
Thanks, SteveF. I have made it through the first 2 papers on Diogenes’ list and I see that at least Mann had more than a conjecture here – as he did with dumb ships. After my first read through the original paper I could see where he might get disagreement and the growth model was my first guess where. I admit on my first read, that is often the case with a Mann authored paper, I did not see how arbitrary some of the inputs into the model were and how they could bias the result. The second linked reply cleared all that up for me.
I have little reason to believe that TRW can faithfully proxy temperature and thus I would not conjecture one way or the other about climate sensitivity vis a vis TRW and volcanic eruptions. I have heard Rob Wilson in the past point towards TRW faithfully marking the eruptions of volcanoes. I would agree with him but at the same time also note that TRWs from locations in close proximity to one another most often have responses very different in amplitudes – and therein lies the rub.
Lucia
One suspects that Mikey took his ‘coming out of the closet’ ad hom tweet at Rob Wilson down because his lawyer suggested that providing ammunition for Steyn’s case wasn’t overly helpful…
Extraordinary how people with a high IQ can be so amusingly dim 🙂
Re “incontenstable”, my view:
David L. Hagen â€@DavidLHagen
@MichaelEMann #Nullius in verba backs @flimsin #McIntyre’s #climate #audits on data win #science #method vs ur #adhom http://v.gd/yBeHs3
Per Tamsin Edwards â€@flimsin 23 Sep
@stevebloom55 @MichaelEMann @wottsupwiththat @andrewadams99 Not sure what you mean. Models & data disagree -> uncerts of one/both too small.
My 2 bits:
@flimsin Y Way too small Not include #BIPM JCGM 100:2008 Type B #uncerts http://v.gd/qPP86i . @stevebloom55 @MichaelEMann @andrewadams99
5
Rob Wilson says:
30 Nov 2012 at 3:43 AM
w.r.t. “the cooling would be sufficient to saturate the growth response, and that SOME trees might `skip´ a ring for that year leading to a slight slippage in tree-ring datingâ€.
Some trees, all trees, or 50% of the trees? This is one of the main issues. At some sites, some trees indeed will express a locally absent ring at chest height where samples are taken. These can easily be identified through crossdating with the other trees in a stand that have this ring. Even in the unlikely event that all trees at a site may be missing a ring for a particular year, other sites are almost always sampled downslope or away from the high latitude tree-line at sites (as part of a wider network) that will not be quite as sensitive to temperature variability (i.e. growth will be less limited by temperature). These trees and those from other sites in a network can be used to identify the missing ring at that one site. In my 18 years of tree-ring work, I have NEVER seen a stand-wide missing ring.
Let’s finally focus on 1815/16. Accordingly to Mike’s variant of the VS tree-growth model and how he implemented it, ca. 50% of all trees will be missing the 1816 ring. There is simply NO evidence of such a phenomenon in tree-ring records in the regions affected by the “year without a summer†– NW North America and Europe. There are long instrumental and historical records in these regions which clearly show no “slippage†of dating prior to 1816.
Mike – please take the opportunity to speak to your dendrochronological colleagues and friends next week at AGU.
Rob
[Response: All very interesting points Rob. I hope you and your colleagues will watch my AGU New Fellows talk next week [Abstract Title: The Past as Prologue: Learning from the Climate Changes in Past Centuries (Invited) Final Paper Number: A32D-02, Presentation Type: Oral Presentation, Presentation Date and Time: December 5, 2012; 10:30 AM to 10:50 AM, Presentation Length: 20 minutes, Session Title: A32D. New Atmospheric Sciences Fellows Presentations II (Video On-Demand), Location: 3002 (Moscone West)] which will be livestreamed. I’ll be showcasing some very interesting new results based on real world data (the chronologies used to build D’Arrigo et al ’06) and, yes, Houston–we do very much appear to have a problem. I will also be addressing the recent claim by Esper et al that multiproxy reconstructions are underestimating a long-term cooling trend. This claim will be demonstrated to have been falsified 🙂 – mike]
7
Jan Esper says:
30 Nov 2012 at 5:36 AM
Sure, anything is possible. But it doesn’t necessarily have to be published in Nature.
[Response: As we should all be aware, what gets into Nature or Science (or Nature Geoscience) is a bit of mystery – and we could all list many papers that didn’t ‘have’ to be published there. I doubt however they would be the same! More to the point, everyone can submit stuff to these journals and kudos to them if it gets accepted. – gavin]
I recommend going out into the forest, develop a tree-ring chronology, and learn crossdating. There are hundreds of dendrochronologists that did this work in the past and could show you the basic techniques.
You could then check your ideas with real-world data and try to demonstrate that post-volcanic rings are missing, before concluding “the potential biases identified in our study necessarily impact all existing hemispheric-scale estimates†and “bolster the case for a significant influence of explosive volcanism on climate in past centuriesâ€.
[Response: I don’t have any dog in this particular issue and I wrote this to point out this this is part of a bigger discussion. It seems to me that the process of checking these hypotheses is well underway. I will be happy to see further discussion on the various issues. – gavin]
[Response: With regard to checking with real world data, you might want to watch my AGU New Fellows talk next week [Abstract Title: The Past as Prologue: Learning from the Climate Changes in Past Centuries (Invited) Final Paper Number: A32D-02, Presentation Type: Oral Presentation, Presentation Date and Time: December 5, 2012; 10:30 AM to 10:50 AM, Presentation Length: 20 minutes, Session Title: A32D. New Atmospheric Sciences Fellows Presentations II (Video On-Demand), Location: 3002 (Moscone West)] which will be livestreamed. I’ll be showcasing some very interesting new results based on “real world data†and, yes, Houston–we do very much appear to have a problem. I will also be addressing the recent claim by Esper et al that multiproxy reconstructions are underestimating a long-term cooling trend. This claim will be demonstrated to have been falsified 🙂 – mike]
Repeat: you are not in competition with each other.
Homer: I want to see you both fighting for your parents’ love! (Flicks light on and off.) Fight! Fight! Fight! Fight!
Graciousness is possible!
Michael E. Mann â€@MichaelEMann 5h
@flimsin @dougmcneall Fair enough Tamsin–lets leave it there. Apologies if I appeared to question your motives. For the record, I do not!
willard should go help mann
So the tree doesn’t grow at all for an entire year and then rises like Lazarus the year after to grow again?
Somehow I kinda doubt it.
Andrew_FL,
A tree can have ‘no ring’ in some places but have a ring in others… so it doesn’t die (if the core were taken in a different place on the trunk then there would be a ring). It is just that in a whole stand of trees you can check against others and weed out the obvious ‘no-ring’ samples. See #120373 above.
This idea crossed my mind.
Yes. Especially given that the volcano didn’t fully blacken the sky and kill everything including grasses, herbacious perennials and animals that feed on them.
SteveF-Ah, that makes marginally more sense.
Lucia-Makes one wonder the sort of conditions that would be necessary to halt all plant growth locally for an extended period.
What I see from the links that were provided above by Diogenes on the missing tree rings was that the Mann et al paper not only requires missing tree rings but a “stochastic forcing” i.e. “weather”, all in combination with a tree growth model to obtain what the Mann authors judge is the proper TRW growth – given an evidently preconceived temperature response from volcanic aerosols. By applying these inputs, and, which by the dendro replies to this paper were arbitrarily selected, into an untested growth model for trees in order to obtain the desired result, Mann and authors, perhaps unwittingly, put on display just how complex and poorly understood and tested the relationship of this proxy response to temperature is.
On the other hand, the dendros and colleagues show what a tightly knit group they are. It would appear they are willing to admit that tree rings and MXD do not necessarily provide a quantitative response to temperatures, but that all their work and effort cannot be refuted by a non dendro and that indeed the trees respond to extraordinary events like volcanoes and that not getting the year correct would be an anathema to dendrochronology whereas dating annually provides that proxy’s singular advantage.
Skeptics can learn from these discussions, but if they get too directly involved both these groups will remember who their common enemy is.
http://www.meteo.psu.edu/holocene/public_html/shared/articles/MFRNatureGeosci12.pdf
“There is consequently increased temporal smearing back in time in the hemispheric composite. This
smearing leads to a predicted delay of 1_2 years in the peak cooling for the 1815 eruption and an even larger delay of 4_5 years for the ad 1258/1259 eruption. It also leads to further reduction in the amplitude of the estimated cooling. These predicted features match quite closely with what is seen in the actual tree-ring series (Fig. 2d_see insets). The extent to which the predicted age-model errors are present in actual individual tree-ring series will need to be assessed by dendroclimatologists through the careful reanalysis of the original tree-ring data. It should also be noted that these age-model errors are predicted only for thermally limited treeline environments and for very large eruptions, and do not necessarily
imply more general problems in dendrochronology.”
http://www.blogs.uni-mainz.de/fb09climatology/files/2012/03/Esper_2013_BullVol.pdf
“Estimating the degree of cooling by eruptions prior to the era of instrumental observation necessitates the use of annually resolved temperature proxies that explain a fraction of temperature variance of which only tree-ring, and a few documentary records, have the temporal precision and accuracy to provide adequate information over the past millennium (Frank et al. 2010).”
I know nothing about tree-ring science.
In the 1930s CF Martin made guitars out of red spruce cut from the Appalachians. It’s believed they used trees harvested mostly from West Virginia and Pennsylvania. It’s possible some came from New York and Maine.
Not all trees in a stand would be large enough to make guitar tops, so while production of instruments was small (several thousand), the area from which the lumber came would be large.
A very high percentage of those guitars have a series of very narrow rings. Probably around 8 to 10. Something happened to the trees in the 19th century, and it happened to a large number of them.
They are fantastic instruments. They sound incredible. I have a 1937 D-18. They are one America’s rarest antiques, and are very valuable.
Get a bunch of scaffolding, plastic sheeting, long cable and ac units. Then erect scaffolding around trees you think give you a temperature signal, add plastic cover (pseudo-aerosol) and blast cold air into the base (cold winds).
After the summer, wait a year and see what has been laid down. Yearly, core the tree, and see what the effect on tree ring is.
The dendochronologists are very careful. They go to places that have had ‘events’.
Astrobiology. 2004 Fall;4(3):391-9.
The Tunguska event in 1908: evidence from tree-ring anatomy.
Vaganov EA, Hughes MK, Silkin PP, Nesvetailo VD.
Source
VN Sukachev Institute of Forest, Siberian Branch, Russian Academy of Sciences, Krasnoyarsk, Russia.
Abstract
We analyzed tree rings in wood samples collected from some of the few surviving trees found close to the epicenter (within 4-5 km) of the Tunguska event that occurred on the last day of June 1908. Tree-ring growth shows a depression starting in the year after the event and continuing during a 4-5-year period. The most remarkable traces of the event were found in the rings’ anatomical structure: (1) formation of “light” rings and a reduction of maximum density in 1908; (2) non-thickened tracheids (the cells that make up most of the wood volume) in the transition and latewood zones (the middle and last-formed parts of the ring, respectively); and (3) deformed tracheids, which are located on the 1908 annual ring outer boundary. In the majority of samples, normal earlywood and latewood tracheids were formed in all annual rings after 1908. The observed anomalies in wood anatomy suggest two main impacts of the Tunguska event on surviving trees–(1) defoliation and (2) direct mechanical stress on active xylem tissue. The mechanical stress needed to fell trees is less than the stress needed to cause the deformation of differentiating tracheids observed in trees close to the epicenter. In order to resolve this apparent contradiction, work is suggested on possible topographic modification of the overpressure experienced by these trees, as is an experimental test of the effects of such stresses on precisely analogous growing trees.
JCH, thanks for the note on musical instruments. You ever hear of the link between Stradivarius violins and the very narrow growh rings in the wood?
http://news.nationalgeographic.com/news/2004/01/0107_040107_violin.html
Does the tree ever get taller without adding rings?
j ferguson
“re get taller without adding rings?”
FYI A new inflorenscence starts each month and takes about a year to grow out into a bunch of coconuts. Correspondingly, the coconut palm grows upward with new fronds. Then sections get denser with age. The top is soft enough to stick a thumb into while the bowl at the base is denser than oak parquet flooring. In a volcanic eruption, a coconut palm will lose its fronds for about two years and then regrow. Palms are technically not trees.
Nature Publishing?
yes Nature. the magazine that published work about piltdown mann after it was outed as a fraud
“In order to resolve this apparent contradiction, work is suggested on possible topographic modification of the overpressure experienced by these trees, as is an experimental test of the effects of such stresses on precisely analogous growing trees.”
cool, blow things up by trees.
Mosher
On a more timely note, Nature Communication seems to own both Scientific American and Nature. SciAm is, of course, the magazine that up to very recent times employed Bora Z, who has resigned after the revelations of his harassment of aspiring young female science writers. ( Of course, could happen to any publisher. But.. that seems to be Nature Communications. )
Given the state of his head, Tol just woke up.
Feeling especially nasty today, Joshy?
“Some trees, all trees, or 50% of the trees? This is one of the main issues. At some sites, some trees indeed will express a locally absent ring at chest height where samples are taken. These can easily be identified through crossdating with the other trees in a stand that have this ring. Even in the unlikely event that all trees at a site may be missing a ring for a particular year, other sites are almost always sampled downslope or away from the high latitude tree-line at sites (as part of a wider network) that will not be quite as sensitive to temperature variability (i.e. growth will be less limited by temperature). These trees and those from other sites in a network can be used to identify the missing ring at that one site. In my 18 years of tree-ring work, I have NEVER seen a stand-wide missing ring.”
I think what Rob Wilson is pointing to here in general terms is that trees in the same stand or a stand nearby can make growth responses that are very different – perhaps caused by a micro event affecting just that tree or perhaps part of the noise in the response.
I think the dendros and those that use their TRW and MXD measurements for temperature reconstructions depend on the “averaging” effect of combining many samples to get around these problems. That approach sounds quite reasonable until you see those doing reconstructions selecting results that best fit the instrumental temperature. The averaging effect only works if you use all the data.
Until those doing temperature reconstructions realize and acknowledge the problem of selecting proxy results after the fact for the reason given above and others that should be equally obvious I cannot give these reconstructions any credibility.
Andrew FL,
I think the correct response to Eli Rabetts commenting on the state of Tol’s head is this:
ClimateGate e-mails also show scientists discussing Mann’s hockey stick studies in similar terms.
Earth to Mann… Making stuff up to confirm your prior findings is not science.
Lucia that is one angry looking rabbit. Some how I envisioned it as much more fuzzy.
Ken,
“That approach sounds quite reasonable until you see those doing reconstructions selecting results that best fit the instrumental temperature. The averaging effect only works if you use all the data.”
Don’t you mean,
“That approach sounds quite reasonable until you see those doing reconstructions selecting results that best fit the instrumental temperature, up to around 1960 when tree ring widths no longer track temperature. The averaging effect only works if you use all the data”
I just felt like it was a very unprofessional comment for an academic to make hiding behind a pseudonym about another academic.
But you are probably right.
Eli would stand no chance in an argument against Richard Tol, for obvious reasons.
Doc Martyn—actually the tree ring widths seem to do okay, individually.
Most of them have positive trends from 1960-1995.
The actual divergence appears to not be in th tree-ring data, but rather in the reconstruction. If so, there may be less to that problem than has been made out:
This is because we already know these reconstruction methods have a problem with loss of coherence of short-period signals (anything shorter than say 50-years). The so-called “divergence” could just be noise, and not signal, while the variations over periods of 100-years and longer could still be perfectly valid.
Carrick- Can you show a similar histogram for the tree ring widths in, say, 1915-1950 or a similar period?
Strictly speaking it may matter a great deal whether or not the same trees that have positive trends during the later period have positive trends in the other half of the twentieth century. And while we are at it whether they get the behavior mid century consistently as well.
Actually, maybe just show a histogram for 1900ish to the end of the data?
Lucia,
“I think the correct response to Eli Rabetts commenting on the state of Tol’s head is this”
.
Well sure, but the worse part is what goes on inside Eli’s head (or perhaps better, what fails to go on inside his head!). Really,the emperor of obtuse ought to get a grip on reality… but there seems no chance of that… he is just too, um… intelligent.. to make rational comments.
Speaking of rational discussion I suddenly have the urge to revisit that discussion of atmospheric circulation that happened at Jeff’s blog over three years ago. I never really got satisfactory answers to some of the questions I had…
I mean I know it has like nothing to do with what we were just talking about but it’s been bugging me for literally years.
Funny how Tol rubs academics like Eli (Josh) the wrong way. They seem too used to having people hew the politically correct (left-wing) line on all subjects. Tol doesn’t do that.
Andrew_FL,
Are you going to contact me by email?
Andrew—if I get time, sure. There is a complication because not all proxies are available for all periods, so you would need to make sure you only plotted proxies that had data throughout the range of interest.
(“Apples to apples” comparison.)
SteveF- Oh, sorry, I must not have seen where told me to contact you? I’ve been pretty caught up in a lot of stuff these past few days. Sure, I’ll send you an email.
Carrick- Thanks, I don’t mean to impose on you or anything. And I get what you are saying, we definitely want to make sure the data comparison is uh…comparable.
DocMartyn – I have read many articles about violins and various theories for why they are great.
In American steel-stringed guitars, which is my thing, I have never detected a consistent difference in narrow versus wide grain. I’ve had guitars made with very wide, new-growth (~7/32nds of an inch) red spuce, and they were incredible. but narrow has won a bias, deserved or not.
I leave the violinists alone with their psychiatrists.
Hmm…is the rabit a bouncer at a Playboy club?
JCH- Maybe it matters more in a smaller instrument?
JCH,
“I leave the violinists alone with their psychiatrists.”
.
Indeed: http://blogs.discovermagazine.com/notrocketscience/2012/01/02/violinists-cant-tell-the-difference-between-stradivarius-violins-and-new-ones/#.UmfbPRC6R8G
SteveF,
I enjoyed reading the blog about the Strad debunking. It makes one wonder how all of the studies aimed at discovering the cause of their superiority could have resolved anything if they weren’t superior to all other violins.
There was a lot of wine mixed in with the violins – how even experts couldn’t distinguish plonk from the good stuff. I don’t buy that but then I don’t buy the good stuff either. When I was turned loose from uni in 68, one could buy almost anything for about $5/bottle at the liquor store corner of Armitage and Lincoln avenue in Chicago. I’m astonished at the present prices of the things I used to drink – mostly estate bottled Medoc’s.
My current practice is to require that a wine not taste bad. This saves a lot of money since the cost delta between not-tasting-bad and actually tasting really good can be significant.
I also agree with the commenters who suggested that weeks could be spent by a competent violinist coming up to speed on how best to equip and play a particular violin. It certainly was true with French Horns which is what I used to play. I was loaned a Sansone by one of the symphony guys for a weekend and was unable in that time to improve on what i was able to do with a considerably less expensive Olds. But then i wasn’t any great shakes with the Olds either. I also agree that the room accoustics make a lot of difference and imagine that what the musician hears is not the same as the audience which is certainly true of horns. There’s a lot of acoustical stuff going on the instrument/player interface which doesn’t travel. I’d be interested in hearing if this applies to guitars.
j ferguson,
I agree that there are big differences between good and bad tasting wines. There is some correlation (not very strong) between good tasting and higher prices, but even very expensive wines can sometimes resemble a poisonous mix of venom. That being the case, I find the best thing is to try a bunch of inexpensive wines and see which really are good, then buy many bottles when you find a good one. Most are not far in quality from what their price would suggest, of course, but once in a while you can find a gem. Not surprisingly, when I discover an inexpensive wine which is very good (say a flavorful, buttery Chard under $8/bottle) it tends to disappear pretty quickly from the store shelves, so I suspect I’m not the only person who selects this way. If a wine is identified via publication as an excellent value, I find it is rare to then find it in a store… other years from that maker, yes, but almost never the excellent value wine.
.
Many years ago I flew business class to Brazil on American Airlines. They treated their business class travelers pretty well in those days, and they even had a wine expert who selected wines for serving in business and first classes. They served a bottle of Murphy-Goode Chardonnay which was nothing short of astounding… the best Chard I ever had! And the flight attendant said it was under $10 per bottle. When I got back to the States I contacted Murphy-Goode to try to buy a case…. Nope, not available. American Airlines had apparently purchased their entire stock. I (foolishly) let them talk me into buying the next years vintage… $12 per bottle, stemmy, bitter and horrid. Live and learn.
SteveF,
Well not-tasting-bad is pretty effective. It makes purchasing quicker, costs lower, and disappointment infrequent. After all, what did you expect?
There used to be another issue which was the 3 1/2 liter boxes. These would fit in our little refrigerator and we bought Franzia Drinkable White (I think they had another name for it) in that size. Unfortunately, this size was not readily available outside Florida, so we would stock up in Stuart and Jacksonville on our way north lining the bilge in the stateroom with as many as 20 boxes – enough for the summer.
I suspect that this does not speak well of us, but then we’re pretty cheap. Also, and this applies to my efforts at reading here, at CA, and Jeff’s, a 5% comprehension makes a lot of things a real struggle.
DocMartyn (Comment #120422)
October 22nd, 2013 at 4:05 pm
“Don’t you mean”
I mean if you want to take advantage of averaging out noise from a temperature signal in any reconstruction using proxies, you cannot select out data that does not agree with the instrumental period whether that data is under or over trending. Otherwise you are biasing the noise and you no longer have an average condition.
It may turn out that the averaged result has little of no temperature signal but at least you have used the averaging process correctly.
Beyond abusing the averaging process, selecting proxies based on how well those proxies match the instrumental period is wrong once it is realized that any series with a high level of auto correlation that does not contain a deterministic trend can have a series ending upward stochastic trend that spuriously shows a correlation with the instrumental record. Selecting for that relationship amongst sufficient proxy data could well produce a reconstruction that would appear to show a modern warming and a rather meandering response around some average temperature for the remainder of the series.
It should also be noted that the situation with TRW and MXD proxies is not quite as black and white as I represent it above. These proxies do respond to temperature as marking a volcanic event can show. The problem is that these proxies also respond to other variables and that can result in a very high noise to signal ratio.
Regardless, selection of proxy data after that fact is just plain wrong headed and is something that the climate science community as a whole evidently does not see as a show stopper.
“The actual divergence appears to not be in th tree-ring data, but rather in the reconstruction.”
Jury is entirely still out on that question- “divergence” is not uniform but rather varies from place to place and is not explained by any single, known mechanism. The basis of Loehle (2009) takes as its working assumption that that observed divergence might very well be “in the tree ring data” as you phrase it, and then gives the biological basis for it, and the mathematical repercussions of it.
“This is because we already know these reconstruction methods have a problem with loss of coherence of short-period signals (anything shorter than say 50-years). The so-called “divergence†could just be noise, and not signal, while the variations over periods of 100-years and longer could still be perfectly valid.”
No, this is exactly backwards. The short term variations have generally high coherence; it is estimating long term trends that is most problematic. This issue has long been recognized–it was raised and discussed in detail by Cook etal (1995; i.e., the “segment length curse”), in cases where individual cores/trees are detrended, and then expanded upon by me, to also apply to cases where Regional Curve Standardization (RCS) is used, as coded in existing, software algorithms. As explained in great detail starting here:
http://ecologicallyoriented.wordpress.com/2012/11/10/severe-analytical-problems-in-dendroclimatology-part-1/
I edited the above comment but the edits did not appear. It’s in response to comment 120426.
SteveF (Comment #120462)
October 23rd, 2013 at 10:32 am
“There is some correlation (not very strong) between good tasting and higher prices, but even very expensive wines can sometimes resemble a poisonous mix of venom.”
My wife is more of wine drinker than I, but I drink enough wine to know what tastes good and bad – or at least to my and even my wife’s palates. When we were much younger and with a somewhat limited budget for wine we began as you suggested with lower priced wine that we thought tasted good. We then would go onto to more expensive wines to see what we were missing. Our experience was that the best tasting wines were the most expensive and that generally while one could get a bad tasting expensive wine it was a much easier task to get a bad tasting inexpensive wine.
I think what we have settled on as a compromise is finding “reasonably” priced wine that we like and getting pleasure from relaxing with those wines and letting the setting have a positive effect on the drinking experience while knowing full well there are some very expensive wines out there that would make incredible drinking if just taste were involved.
The ritualistic part of drinking has been firmly established with my evening martini where I use a good vodka or gin made from a bottle kept in the freezer along with my martini glass. It has spoiled for me ordering a martini from most eating establishments as they just do not measure up to my expectations.
Ah Kenneth Fritsch,
We ran aground on Tequila Joint margaritas. I would go through the descriptions on the list and not find my favorite mix. I would then ask the attendant if I could specify the mix at of course an adjusted price. I would be asked what I wanted. One to one to one, Jose Cuervo “Gold”, Triple-Sec, Rose’s Lime Juice.
For some reason no place was ever willing to do this despite having dozens of “margaritas” on offer. They would admit to having all the ingredients but for some reason it was always against management policy. I would try to buy a glass of each of the ingredients and roll my own, but they wouldn’t allow that either. I sense attorneys.
I think I was surprised when I realized that the taste of a margarita was the tequila not all the other stuff and yes I know there are tastier tequilas out there, but i guess it’s like our wine theories.
We, too, like Vodka from the freezer – sort of syrupy.
Kenneth, 120467:
Since the long-beaten horse has been resurrected therein…
Yes you **CAN** do what you are saying one cannot. Indeed, how else could one EVER, in any way, determine whether there’s a relationship between the rings and the climate variable, except by so doing? Answer: you can’t; It’s the only way to do it!!!
Potentially spurious relationships can indeed arise at individual sites, as you state, but that is addressed–as it always is in such cases–by taking into account that multiple comparisons/tests are being made and adjusting the acceptable p value criterion accordingly, or alternatively, comparing how many statistically signifcant results actually do arise, compared to how many *should* arise, for a chosen p value (typically, p=.05).
Why people continue to not understand this very basic issue I really don’t understand. Selecting or omitting the chronologies from a set of candidate stands is **NOT**, in and of itself, cherry picking or an otherwise nefarious practice, as long as you do what I state above.
Kenneth,
On a more positive note, thanks for bringing the discussion back to substantive issues, and away from the raging food fight.
Jim Bouldin:
I agree. If you plot the trends on a map, this is what you see.
There is clearly a geographical bias sitting on top of any putative temperature signal, indicating that other factors (e.g., sunlight, precipitation changes, etc) are playing a role. That’s hardly surprising of course given what we know about what influences tree growth.
What surprised me though was the predominance of positive trends when I had been told that the proxies were categorically showing a divergence. Most, it turns out, don’t.
Thanks for the input, I hadn’t seen Cook’s paper. I will go back and look at it.
However, you were to align the individual proxies with respect to the long period portions of the signal, getting a loss of coherence in the short-period signal in the reconstruction only requires that the relative phase response for long and short periods for each proxy varies between proxy.
In other words, were I to use a method that preserves the long-period portion of the signal, when the statement the proxy responses to long period and short period stimuli have the same relative phase is false, the reconstruction will lose coherence for short period signals (the signal components for these periods will tend to average out).
Were I to use a method that preserves the short-period portion of the signal, e.g. detrend before calibrating, under these same circumstances the reconstruction will lose coherence for long period signals.
Is this clear?
If not, I write out maths and expand the discussion a bit, if that will help.
Note I’m not making any claims about whether tree rings are “better” or “worse” for short-period versus long period temperature variability, either way. I simply don’t know enough to have a useful opinion on that.
Jim Bouldin (Comment #120468)
October 23rd, 2013 at 11:29 am
“No, this is exactly backwards. The short term variations have generally high coherence; it is estimating long term trends that is most problematic.”
Interesting that the detrending and not detrending issue was a major problem in a paper submitted and withdrawn for a temperature reconstruction of the Austrialasia region. The authors did the selection after the fact by what they claimed was detrended data correlation of proxies with the instrumental record in the hopes that the detrending could somehow more legitimize the after the fact selection process. Their error was revealed at CA as one of not detrending the data before making correlations. It turns out in their case that the correlations without detrending met their correlation criteria better than the detrended data did. I would think, that since the end purpose of using the proxies for reconstructions is to establish trends in the series, that the lower frequency part of the series is the more important part for those purposes.
Better higher frequency coherence is something that one might expect from a proxy that is responding to temperature variation but not correlating with amplitude because other variables are affecting response. It is a rather simple exercise to construct two series which have a very good high frequency correlation and produce very different trends.
Kenneth Fritsch:
I think this is true as well. The focus should be what happens to the high frequency portion of the signal when you align the proxies for the low-frequency portion.
Make sure to add noise as well. If you align the trends to match your temperature record (e.g., calibrate them individually with temperature), you’ll find that the noise components from the proxies with the smallest trends will get a disproportionate weight in the reconstruction.
Technically this is a different issue than what I was describing above, since the signal portion isn’t getting attenuated. Rather the problem is the noise component isn’t getting attenuated as much.
Thanks Carrick at 474; hang on, will have to respond later when I have more time to digest. Not 100% sure what you mean and we will have to clarify terms so we are exactly on the same page. In the mean time, if you have time, perhaps you can clarify exactly what you have in mind by:
“individual proxy” (individual core, tree, or site?)
“reconstruction” (the period preceding the calibration period?)
long vs short term
“aligning” proxies
Jim Bouldin, here you go:
individual proxy
I was looking at Mann proxies. So the individual time series as here:
data set.
I’m only using the “series 9000” which is the identifier for tree ring (I exclude MXD because these are not the original MXD series, but are the modified version used by Mann et al),
reconstruction
I’m referring to the time series that is the output of the software that combines the proxies to create a temperature-like series. E.g., Mann’s EIV series from his 2008 paper.
aligning
There is a more technical description that I don’t want to go into, unless it’s necessary and useful—these sorts of things take a while to write up and proof for errors—but the simplest version of this would be to compute the regression coefficients of the individual proxy series against temperature (or vice versus) over the calibration period, e.g. 1900-1950. I’m describing this process as “aligning”. The “reconstruction” would be using these individual regression coefficients together with the proxy series to “reconstruct” the temperature over the “reconstruction period” (e.g., 0000-1900).
A few technical points:
If you use the proxy series without any filtering (e.g., detrending), because the signal has a 1/f character, this will weight the reconstruction towards the lower frequency/longer period portions of the signal (unless you pre-whiten it before performing the regression analysis). I am describing this as “aligning the series to the long-period portion of the signal”.
Were you detrend or otherwise high-pass filter the signal before performing the regression analysis, I would this as “aligning the series to the short-period portion of the signal”.
Jim Bouldin (Comment #120472)
October 23rd, 2013 at 12:02 pm
Kenneth, 120467:
Since the long-beaten horse has been resurrected therein…
Yes you **CAN** do what you are saying one cannot.
Jim, I would suppose you can yell about doing a number of things but that does not make it correct. I suspect that those who have been trained in the hard sciences do not see the error in this approach because in hard sciences you have an opportunity to run a confirming experiment under controlled conditions (out-of-sample testing) after doing some preliminary in-sample testing. Unfortunately in climate science and in validating proxies this is a task that is very difficult to impossible to do. You can split the instrumental period into a calibration and validation periods but this is not the same as doing a purely out-of-sample confirmation test and it is open to peeking and to further selection based on a reasonable agreement between calibration and validation.
The only reasonable approach to this problem is to establish a well understood and physically based criteria for selecting proxies prior to any measurements and then retaining all the data for reconstruction unless one has a very obvious reason for rejecting it – and after the fact poor correlation between proxy and temperature is not an obvious reason.
Your rationale of using a p.value is not clear to what the p.value test is referenced. Are you referring to a correlation between a proxy response and temperature and the probability of that occurring by chance and then further calculating, based on models of the temperature and proxy series, the frequency of that many proxies having a statistical significant correlation with temperature occurring by chance? That is what, I recall Mann et al attempted to do did in Mann (08) – and used invalid data to do it. If what I described is what you propose I would ask how you would account for the fact that the proxy population from which you are testing significance is already biased towards a statistically significant correlation with temperature.
Kenneth Fritsch (Comment #120486)
I should have added to my previous comment that a p.value of 0.05 for a monthly correlation of temperature to proxy response, i.e. a probability that the correlation differing from zero occurring by chance is 0.05 or less, with all those degrees of freedom might well show a correlation of 0.2 or less as statistically significant. A correlation of 0.2 would imply that the temperature variation explains just 4% of the proxy response variation. How confident should we be that a response that is influenced so little by temperature in the instrumental period (on average) would hold into the pre-instrumental period?
Oh and by the way, I do not see after the fact proxy selection as nefarious but rather simply wrong headed for what could be from one or more of a multitude of reasons.
Thanks Carrick. We are on very different wavelengths here and there are a whole slew of things that need to be clarified before we can get anywhere on this. Hopefully there will be time for it. I’m not sure why you are comparing procedures in which detrending is, and is not, performed. Almost all data is detrended first unless one has a good reason for not doing so. The questions revolve around the legitimacy of the detrending procedure, and then subsequent calibration/validation issues.
It is also imperative that we decide whether we are discussing methods generally, or with specific reference to Mann 2008. I don’t know why the latter is being brought into the discussion–we need to get general principles straight first.
Kenneth, I don’t agree with this, and I’m pretty sure no statistician would either. You seem to be taking ideas that you have gathered from some other science or engineering discipline, and thinking they are strict laws of statistics that must therefore be applied in dendroclimatology as well.
What you call “out of sample” testing, in a tree ring sampling context, would equate to none other than sampling another set of trees, or if you like, splitting the sample you do have into two groups. So there is indeed a direct analog, and indeed, this is the underlying reason why a number of trees are sampled in the first place.
This is a very simple concept. For a given number of chronologies, you know how many you’d expect to give a spurious correlation under a model of strict randomness (no relationship between ring response and climate), and you simply use that as your reference point against the number you actually observe. You can also do additional things like split validation and calibration periods that add more information.
As to your latter comment, if you don’t like using a p value, you can always use an R-squared criterion if you like, or just drop the p value down to .01 or whatever if you like.
“If what I described is what you propose…”
Yes, you understood it correctly.
“…I would ask how you would account for the fact that the proxy population from which you are testing significance is already biased towards a statistically significant correlation with temperature.”
That last assertion is absolutely *not* correct and indicates why you are confused on this topic. Where did you get that idea? The proxy population is not biased toward anything other than what drives the ring response, which is not what bias is.
Jim Bouldin, what do you think of TomP’s attempt at a Yamal robustness analysis:
http://climateaudit.org/2009/09/30/yamal-the-forest-and-the-trees/#comment-196134
I’m more interested in whether you can verify a specific criticism I had of his methodology, that he needed to rewrite the RCS function, but I can’t find that criticism in those threads yet.
Jim Bouldin, what do you think of TomP’s attempt at a Yamal RCS robustness analysis:
http://climateaudit.org/2009/0…..ent-196134
I’m more interested in whether you can verify a specific criticism I had of his methodology:
Tom P, your code doesn’t properly implement the sensitivity test you described. You have to rewrite the RCS chronology function, not just edit out some trees.If you want to only have above 100 years, then you have to cut out the first hundred years of EVERY tree. As you have it, you still have some trees that are in their tenth year of life.
Is this a valid criticism?
Re: Jim Bouldin (Oct 23 18:24),
And what statistical model are you using for the randomness to determine the percentage of spurious correlations? Mann, if I remember correctly, used AR(1), which is not a very good model at all.
Jim Bouldin:
Gergis is recent example of leaving in the trends. I think there are other recent examples, but I don’t have time to go back and dig.
If you don’t detrend, you leave in lower frequency information. If you do detrend you remove it. There of course are other operations that are equivalent.
The point is unless you whiten the signals that you are regressing against, the largest frequency components will dominate your fit (“most variance explained by …”). Since the spectra have 1/f shapes, the lowest frequencies that survive whatever operations are going into this portion of the reconstruction algorithm will dominate.
The issue is whether your different series don’t have the same phase behavior with frequency. If they don’t, then analyses that amount to some variation on linear regression will always have issues with loss of coherence outside of the frequency band where the signal is dominant.
Mann 2008/2009 is the data source I used for the proxy series to generate the above graphics.
This gets to be pretty hairy stuff to try and explain unless both people can understand each others terminology. It may be hard for us to achieve that here.
“That last assertion is absolutely *not* correct and indicates why you are confused on this topic. Where did you get that idea? The proxy population is not biased toward anything other than what drives the ring response, which is not what bias is.”
Jim Bouldin, perhaps you know in detail what the population from which the proxies were drawn (by selection after the fact) was for various published temperature reconstructions. Your replies come across as hand waving. Perhaps you could take us through the process as you see it of a temperature reconstruction that you have seen that follows your prescribed approach.
I have never seen a temperature reconstruction paper published where data was taken from trees that were sampled and all the cores were measured and described and a precise accounting was provided for which measurements were retained and which were not. All the temperature reconstructions from TRW and MXD that I have read are one or two steps removed from the original core sampling and measurements.
You appear to miss my point on the correlation coefficient where the p.value is used for statistical significance and what that can translate to when an R^2 is calculated from it. The point is that one can readily find a statistically significant correlation between two variables when the number of degrees of freedom is large. The problem (on which you do not reply) is that the statistically significant correlation can be quite low, as it often is in these reconstructions and that low correlation in turn indicates that a very small portion of the proxy response is driven by temperature. In turn this means that the temperature signal in the proxy response could change in other periods by being overwhelmed by the responses to other variables.
“Gergis is recent example of leaving in the trends. I think there are other recent examples, but I don’t have time to go back and dig.”
Carrick, Gergis actually thought that their selection process was based on detrending when in fact they found after revelation at CA that they had not detrended. For Gergis, detrended correlations were lower than for the case with no detrending. Gergis, as I recall, was touting the detrended correlations (which were not) as something not normally done in reconstructions and something that would add credibility to the selection after the fact process.
As I stated previously in this thread it is a simple exercise to produce (with the corgen function in R) two highly correlated series with noise, both white and red, with significantly different trends. I do not know why one would want to use the high frequency part of these signals to correlate with temperature when the trends are the important end product of these exercises. Mann has stated the same preference in different words – but I agree with him. I think those who want to use the detrended series for correlations are concerned about the spurious correlations that are easier to come by with a highly autocorrelated series when no detrending is done.
Kenneth Fritsch (Comment #120517)
October 24th, 2013 at 8:48 am
” Perhaps you could take us through the process as you see it of a temperature reconstruction that you have seen that follows your prescribed approach.”
Jim Bouldin, that was a serious suggestion. I think many of the misconceptions about selecting proxy data after determining how well it fits the instrumental temperatures come in large part from looking at the process piecemeal. Simulations can show the biases that that process produces, but a simple thoughtful reflection on what is assumed in the process and how that assumption leads to bias should be sufficient.
In order to even contemplate using proxies for temperature reconstructions, we have to assume that tree RW and MXD (use the terms temperature proxies here for the more general argument ) do respond to temperature but that those tree variables (TV) also contain influences from other variables. If these other variables have a large influence on the TV compared to temperature then one must be concerned about the effect those variables would have on a historical temperature reconstruction where those other variables influences were not occurring randomly over time. In fact one has to assume that those other variable influences are occurring randomly over time or the rationale of using that proxy for temperature reconstruction is no longer valid – as predictability is lost.
The approach that most climate scientists apparently take to these assumptions, or is one that is at least implicit in their methods, is that some trees carry a temperature signal and others do not based on their reactions to the temperatures recorded in the proxy locale. Selection after the fact of only those trees and cores with a given apparent temperature signal during the instrumental period and exclusion of those deemed not having one precludes the rather reasonable assumption that all TVs contain a temperature signal that can be randomly obscured at various points in time by the non temperature variable influences. That selection also would preclude the reasonable assumption that the variability from other non temperature variables can at times enhance a temperature signal and make it appear larger than it should be.
The assumptions of those that select proxy data for temperature reconstructions after the fact also assume that that response is rather stable going back in time. For those who might assume that there can be a proxy response to the temperature signal that is also responding to non temperature variables randomly but differently over time the assumption of apparent stability going back in time is not required providing one takes sufficient samples to “average out” these random effects. The obvious and large differences in approaches here can be seen in that the antithesis of the selection after the fact approach is to use all the data in order to average out the random effects and avoid biasing the final results.
It is critical here to add that this averaging out and using all the samples would only work to the extent that there is a reasonable temperature signal in the proxy response and that is where one must establish a criteria for selection of proxies before any measurements are made and that criteria should be reasonable and physically based. A search for such a criteria might show that these proxies do not provide a reasonable temperature signal for doing temperature reconstructions, but that is way ahead of the game where one merely selects after the fact.
A numbers/computer model guy asserting expertise on tree growth is like a preacher, with a degree in theology, waxing eloquently regarding engineering based on his interpretation of scripture. The interesting insight Mann gives so many times is that of someone who, far from following evidence, instead slices and dices and interprets and ad hoc conjectures data to fit his pre-ordained shape.