Douglas Keenan has published an excellent comment on his attempt to replicate the prominent study by Chuine et al [Nature 2004] of harvest dates in Burgundy. The article being criticized is all too typical of a climate article in Nature – lurid headlines, lamentable statistics, ineffective peer review and obstructive authors.
It’s really nice to see someone else wade into this sort of study.
Keenan’s forthcoming article is here and two insightful additional comments are here and here.
You might want to take the opportunity to browse his website while you’re at it.
UPDATE: Keenan has added an Addendum to the page describing more technical issues with Chuine et al. This tells how the model performs no better than linear regression, and thanks Willis for pointing that out.
Climate Audit 
116 Comments
A lot of information in such an extremely short article. Wonder if the Teams will respond…Nature should be very ashamed, again.
Steve, I’ve much glad you’ve posted about this. If anyone has any recommendations for improving the non-technical description (linked to on my site), they would be much welcome. A goal is to get people without scientific training to be able to directly understand for themselves—without relying on anyone else—how severe the lack of quality is.
DJK: That work on the “untrustworthy” Anatolian tree rings is very interesting. Any chance of it being published soon?
#2. Douglas, thanks for the paper.
Inspecting Figure S1 of the Supplement to Chuine, et. al., it looks like modern times have a departure from September 1st that is just about the same as the 50 year average between 1650-1700, the Maunder Minimum, and the 50 year average from 1350-1450, the end of the MWP. Based on that alone, there is nothing to the exercise at all. In fact, ignoring the averages, I think that the time series simply looks like noise with no structure at all.
You might want to update your technical comments to add a problem that a high school student could spot.
Re: analysis of the data. My first step would be to histogram the departures and look at the probability distribution to get an idea of its shape. Is it Gaussian, Poisson, whatever, to get an idea of the nature of the fluctuations and to estimate the tails? There’s a decent amount of data here to do some reasonable analysis. Then you could make a statement about the extreme event at 2003 and see if it’s really so wildly different from say 1550 or 1890. There’s absolutely no need to fit a model to the data until that kind of analysis has been done.
The linked articles written by Douglas J. Keenan did an excellent job, in my estimation, of pointing to and articulating the problems with the basis for the grape harvest temperature proxy. They are so apparent from his explanation that one can only conjecture how Nature was able to allow the publication of that paper upon peer reviewing.
Subject confusion may have permitted grape chemists/biologists to peer review the paper? Input data could have been difficult to read by way of wine staining? If someone has better explanations, I would like to hear them. Seriously, one must wonder why Nature would not have allowed a rebuttal to be published if for no other reason than to perserve its own integrity.
And here is the quote of the day for me **I was told by telephone that the university had no procedures for investigating such allegations, because their professors always acted with integrity.**
From http://www.informath.org/apprise/a3200.htm
Is anyone surprised by this? Not any more.
The site seems to be down as of 11pm PST.
The site is indeed down. Apologies; I’m contacting the service provider.
Re #9
Could it be that the massed readership of Climate Audit brought down Douglas’s website by overloading it?
No.
My site is back up.
#3, bender: I want to; dendrochronologists have not been too happy with it, though, because it criticizes the field (see section 8, and also 5).
#6, Gerald: glad you appreciated that.
#10, John: apparently links to an IP backbone (Level 3) were down, but who can really say how powerful ClimateAudit is?
Love that formula …
How much money would the HT have to pay out if they followed Knuth’s policy?
I’ve just finished reading “The Smartest Guys In The Room” and my mind is full of metaphors that I’m currently struggling to resist making.
I don’t have access to the original paper but I would be interested to know how the following factors were handled.
1. The steady improvement of grape vine varieties both before and during the calibration period. This seems pretty complicated to me, because this is not simply a matter of increasing yields but also of adapting to changing tastes.
2. Changes in agricultural techniques, which started to accelerate with the agricultural revolution, well after the middle ages but before and during the calibration period.
3. The destruction of 90% of Burgundy grape vines by phylloxera between 1878 and 1900 (before the start of the calibration period). The vines were replaced by traditional vines grafted onto North American root stocks.
If you look at the raw harvest dates it’s clear that they are not exactly following the AMJJA temperatures and it’s also clear to me that it couldnt. Moderated by most probably precipitation and sun shine duration the harvest dates corresponds to in each year differently weighted averages of “summer temperatures” (the model which is used is a physiological model and not a statistical model). Why should the grapes react to an exact equal average of AMJJA (in this sense one could criticise the figur caption of the Chuine paper)? Its correlation with Paris temperatures is ~0.75. The Chuine article mainly discusses the mean record not the extremes. I guess it’s an easy exercise to find out how well a serie which correlates 0.75 with the temperature serie would describe extreme events. Do the authors of the Chuine article particularly stress the 2003 summer or make it the central point of their article? I dont think so. In my opinion the comments of Keenan are quite pointless.
Here is a comparison of the giss data of Dijon, Chuine’s model temperature and Hohenpeissenberg (a reliable long temperature series from Southern Germany)
The peak extremes of 2003 and 1893 are influenced by other factors than temperature: precipitation could be one of them (as the chuine temperature is derived from grapes). Furthermore IMHO Hohenpeissenberg can be considered as a fair predictor for Dijon.
see also:
http://home.casema.nl/errenwijlens/co2/europe.htm
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/historical/france/burgundy2004.txt
http://data.giss.nasa.gov/gistemp/station_data/
Do you have any statistics on the correlation between the three proxies, Hans?
How accurately did the average French grape grower know the actual date prior to say 1600?
Or anyone else for that matter?
# 19 Very exactly since it was a major and very official affair. Just read the article.
# 15
1) Apparently the main goal was allways: alcohol. And there is only one way to get it. Prost
2)Technological progress is rather taking place after the harvest, not before.
3) Could you give us some reference for that (ie the 90%). In any case there is no discontinuity in the serie.
No paleo record is perfect. This one makes first time something quite quantitative from the so often cited wine arguments.
i thought the keenan piece was a real beauty. It showed what was needed to demonstrate very elegantly that the paper was busted.
very nicely done
per
Chuine et al. highlights the potential of grape harvest dates as a proxy for summer temperatures. The correlation of their reconstruction with measured temperatures at Paris (a couple of hundred kilometres away) is 0.75. This is fairly good for a temperature proxy, and certainly interesting enough to merit publication in Nature. Allegations of scientific fraud made by Keenan are absurd. The raw data are available from http://www.ncdc.noaa.gov/paleo/pubs/chuine2004/chuine2004.html
Re # 23 – I think you missed the point of Keenan. I do not hink he was disputing the correlation.
He was disputing the conclusion that 2003 was the warmest.
Re #24
Then why is the title of his paper:
“Grape harvest dates are poor indicators of summer warmth”
It is worth noting that Chuine et al. use a “process-based phenology model developed for the Pinot Noir grape” to relate harvest date to temperature fluctuations. In contrast, the tree-ring people do not use process-based phenology models of tree growth to relate annual ring width/density to temperature. Big difference. The Chuine et al. approach is far more sophisticated in terms of its biological scientific foundation.
2003 was an especially warm and dry year in the region this study is based on.
As usual, temperature, precipitation, desease and insect infestations need to be taken into account in any of these tree ring or harvest date studies (but they don’t seem to adjust for them for some reason – the reason being bias of course.)
http://en.wikipedia.org/wiki/2003_European_heat_wave
Did y’all read the supplementary information? I thought not.
RE: #23
did you read the paper ? either paper ?
chuine et al are claiming that they can detect hot years. Their data show that they cannot detect three of the four hottest years, and the one they do (2003) is 2.4C too hot ! All four of the four hottest years are considerably mis-assessed, in a model which is meant to detect hot years- is this sinking in yet ?
confused
per
per,
Maybe the reason you are confused is that you misunderstand (or just misrepresent?) the authors’ goals. You describe the model as
but this is your interpretation (or spin?), not theirs. They are attempting to get general agreement between warmer/cooler than average observations and warmer/cooler than average predictions. To focus on the fit of the model to four data points – and four extreme data points at that – is to miss the forest for the trees. The model works pretty well overall. It doesn’t do so well at predicting extremes. But what model does?
The authors could have done a better job pointing out the limitations in their model, no question. But then their work probably would not have been published in Nature, would it?
Put it in perspective. The dendroclimatologists would be drooling at r = 0.75. They are more used to r = 0.2 or less. Face it – it’s a pretty good paper.
Re # 25 and #30 – That is the point I was making – there is some correlation, but you cannot use that to prove that a year is the hottest because of the errors. I do not know why Keenan titled his paper as he did. However a significant point is that you cannot say that 2003 was the hottest(or some of the other years as he noted). I had only mentioned the one year because of the “hockey stick” effect.
#15, Chris: the original paper is linked to from the comments on my site.
Several people have pointed out that the modelled temperatures of Chuine et al. correlate well with observed temperatures. My paper does not dispute that, e.g. it states that the model “might give reasonable results on average”. So for broad-brush events (for instance, studying what happened during the supposed Little Ice Age, or the last few decades of the 20th century), the method might be worth consideration (but see some of the problems described on the linked subpage).
The point of Chuine et al. was not their method. That method had been largely published some years prior: see the references in the authors’ Supplementary Information [Chuine, J. Theoretical Biology, 2000; Chuine & Beaubien, Ecology Letters, 2001].
Chuine et al. applied their method to some interesting data (long harvest series). The point of their paper, though, and the reason that it got published in Nature, was that it made an assertion about the scientific topic—global warming. The assertion was that 2003 (an individual year) had by far the warmest summer in over six centuries.
The point of my paper is that the assertion was unfounded and, more importantly, the authors must have known it. (See too the authors’ response to my criticisms; they would be very unlikely to write such a response if they had high integrity.)
#24 & 30, Gerald: kindly appreciated.
#32 Douglas,
Thanks for the link but the brief doesn’t go into sufficient detail and I don’t have access to the full paper. I did find a useful overview of the research project on Isabelle Chuine’s website, which addresses all of my questions at a simple level. Without access to the actual model used, I won’t be following up on this further.
From what I have access to, this seems like a good research project. The team appear to have explored the significant issues sensibly and they seem to have built a pretty good model. It’s unfortunate that they felt obliged to undermine their own paper by including the 2003 claim in the very first paragraph of the brief.
1. The claim is relatively understated (“2003 probably higher”).
2. The paper is not alarmist in tone.
3. The paper does not address temperatures during 900AD-1300 AD, which is the period of greatest concern in the AGW debate over magnitude/existence of MWP.
4. IMO the paper got published in Nature because it was topical, novel, and well-constructed – not because of any alarmist element.
Of Grapes and Hockeysticks
Volume 7, Number 48: 1 December 2004
http://tinyurl.com/ykj44w
Idso Co2 Science website had a yet another way of looking at this paper.
Re #35
Exactly.
Isn’t part of the problem here that its not so much the paper itself but the interpretation of moviations that the major sticking point.
The concept is interesting and should be encouraged. The fact that their model has a problem reconciling with actual data should be explored further to tease out the robustness of their assumptions and the model.
#33 Chris H, the previous sentence contexualises your quote.
I don’t think that you can call this small paper misleading as Douglas does. It would have been perfectly acceptable to comment on the problematics of methodology. However, to subscribe motivation is not justified in my opinion based on the paper itself.
I would like to see more discussion of the points made in third link of the lead in to this thread which I have excerpted below:
Perhaps D. Keenan would give us more details on the following comment:
Re #38
Yes, me too.
1. There are many good reasons why the authors’ minimum SSE might differ from Keenan’s. What are they?
2. Keenan’s parameters [t0,Tb,F*] = [101,10.3,302] are not all that different from the authors’ [t0,Tb,F*] = [92,8.63,412.97].
3. Keenan does not say what the impact of his parameterization would be on the paper’s conclusions. A difference in SEE of 40 may or may not be significant.
4. One must be careful not to throw the baby out with the bathwater. Steve M complains of Martin Juckes picking at warts. Is this not what is happening here?
http://www.ncdc.noaa.gov/paleo/pubs/chuine2004/chuine2004.html
Still, that wording does remind me of:
http://www.ncdc.noaa.gov/paleo/pubs/mann_99.html
By the way check 1998 in this graph:
🙂
#33
I think the Summer of 2003 (which saw the death of 35000 due to heat) was the warmest Europe saw since at least the end of the LIA. What is interesting, is that the following winter was also severe in terms of temperatures and precipitation (Lower Rhine froze over, heavy blizzards esp in Eastern Europe).
Of course, the Hohenpeissenberg observations agree with you. However, the overhyped 1998 doesn’t even show up in Europe, which to my memory was then still in the Northern Hemisphere.
Re #40
The parallels are there, but they’re weak. Chuine’s language is far more guarded and probabilistic, yet their methods are far more robust and their reconstructions far more accurate. No comparison here as to who is being responsible in their writing and who is being alarmist. No contest.
These disputes over 1998 vs. 2003 as the warmest year in NA vs Eur seem to miss the obvious points that (1) in terms of a global average both years were much warmer than normal and (2) they occurred within a short time of each other. So stop being silly, people. You have much bigger fish to fry over in the other threads.
(Keenan’s work on Anatolian tree-rings and tree-ring cross-dating, on the other hand, is intriguing.)
The dispute clearly shows that global warming is not the synchronous event we are led to believe. But we digress…
IMHO Chuine is less cautious than MBH (count the bold words).
Re #45
I disagree strongly with your opinion. First, it matters a great deal how the words to bold are chosen. I would have bolded quite differently. Second, tone is not just about a count of words, it’s about the weight of each word. Third, and most important, it’s the weight of those words relative to weight of evidence in support of them. Chuine et al’s methods (process-based phenology model) are robust and their results, although local, are far stronger (r = 0.75) than anything you get from tree-rings. Their conclusions are appropriately scaled to the weight of evidence they present.
As for your digression, no serious climatologist believes global warming is noiseless and synchronous. That you lump the climatologists in with everyone else with an opinion is more than just a little tendentious. An AGW trend, if there is one, will be noisy enough (overlain with numerous non-AGW effects, such as PDO, etc) that detection and attribution over short time-scales will be tricky. It’s not a matter of just eyeballing the data. They are far too noisy for that.
Keenan’s point was:
Keenan’s request for data was evidently met with resistance from the authors of the paper:
Nature did not request supporting data from Dr. Chuine:
I would personally like to see more of the details of the grape ripening temperature proxy. There are some things that would appear to be calling for another look (most here would probably agree that Nature did not look into these details) and perhaps D. Keenan can guide us through them.
Chuine et al. note that they used a corrected and updated harvest-date series from Burgundy with a footnote to a paper published in French.
There is a second graph at the top main graph in Figure 1 that evidently shows the number of stations (reporting stations for grape harvest dates?). The numbers seem to be 2 for the most of the period but rise 10 for one extended period and 4 or 5 for a very early period. I can find no explanation in the paper or caption under the figure.
I would also like to see more on Keenan’s reference to daily temperatures used for the regression model and then outputting growing season temperatures (annual) for the reverse regression.
It is proven beyond doubt that Chuines method falls apart on the extreme events.
You cannot use a non-linear data spike as a proxy to tell something valid about an extreme event. The only thing that the harvest proxy can tell you with certainty is that since 1370 there never was a pinot noir harvest as early as 2003.
enter crop management, fertilisers, improvement of the pinot noir grape etc….
I can’t argue with the lack of due diligence. But that is their only serious fault. Again, it is going to take the community some time to wake up to the new reality that they are being watched more closely than in the past.
Re #48
You got a personal axe to grind? You and others are so focused on predicting extreme events you’re either missing the point of the paper, or misrepresenting it. You are very wrong about what can and can not be concluded from this data set. Why you so distort its objectives and conclusions is beyond me. Especially when there are other papers so much more deserving of critique.
Re # 44, bender, thanks for your post. You say:
Perhaps you could expand on this a bit and tell us
1) What temperature is “normal” in a temperature time series containing a trend?, and
2) What is unusual about the most recent data being higher than “normal”, however you might define that, in a series containing an upward trend?
w.
Re #51
1. “Normal” has a formal definition, with the exact definition varying among jurisdictions. Of course, these definitions are aritrary. A long-term average, whether based on 10, 20, 30, or even 50 years is not suitable as an expectation if the system is not in some statistical state of equilibrium. Chaotic systems are prone to doing nasty things like avoiding the long-term average for prolonged periods of time.
2. Nothing. That is the point. These observations are what, in part, constitute the trend.
What’s with the obsession here over extreme values? Models are designed to fit the bulk of the data that occur frequently, not the extreme observations that occur rarely.
#51, 52 — maybe Willis’ point is that in an underlying upward temperature trend, such as we’re having on emergence from the LIA, there are bound to be a series of ‘highest recent temperatures,’ because the yearly oscillations will be superimposed on the positive trend. Such records will have no particular climatological importance.
When the trend eventually turns negative, we can expect a series of ‘lowest recent temperatures,’ also with no particular climatological importance, but nevertheless eliciting concommittant worries of the coming ice age, and sea-level falls. The national wealth of the citizens of Tuvalu will then be threatened by the cold-induced deaths of their cocoanut palms and their receding water-tables, with the blame falling on the enviro-extremist capitalist societies that cruelly cut back on their climate-benignifying CO2 emissions. They should be made to pay, by jimminy.
Re #53 Yes, that was the point that I understood Willis was making. Reconstructions don’t tell you anything about cause. They only tell you the time scale over which a given temperature level or trend might be “unprecedented” relative to some “normal” baseline. If the natural post-LIA rebounding trend is noisy, then each new high that is randomly reached is natural too.
Course, this paper is NOT ABOUT ATTRIBUTION of cause. It’s only about observed effect.
Bender, thanks for your explanation. You go on to say:
The obsession is not ours, but Chuine’s. In their abstract, they say:
and in the paper they follow up with:
They are the ones making the claim that their method can properly describe extreme events, not any of us. Since they have made the claim, it is proper to examine it.
There are a few other curiosities about the paper:
1) There is no data for 1978, and no explanation for the lack.
2) They say that their 95% confidence intervals based on “the regression between observed and reconstructed temperature in Dijon, are in purple. These were obtained by regressing the reconstructed temperature with the observed temperature over 1880–2000.” However, the GISS data for Dijon only go back to 1950 … we’re back in the land of spliced datasets. Nor is GISS alone in this, the ECA dataset for Dijon only goes back to 1958. Man, I hate hidden data … Keenan says
Oh, man, the SOB ate my post again … I gotta start working offline. Reconstruction follows, starting at where the post stops above …
w.
————————————–
… [M Mestre (Meteo France), private communication, November 2005]
Doug, where did you get your temperature data? And if data to 1883 is provisionally available … how did they do the regression to 1880?
3) I find a number of reports in the 2003 news stating that the farmers were “rushing” to harvest the grapes before they were damaged by the lack of rain … dang confounding variables raising their ugly heads again.
4. This problem with the rain is supported by a graph of the Dijon temperature, and the reconstructed temperature from the grape harvest data:
Note that while most of the peak temperatures are overestimated by the grape thermometer, the 2003 is underestimated. This supports the idea that the grapes were harvested extra early because it was so dry.
Given all of that, I find their claims about 2003 being “probably” warmer than any year since 1370 to be unsupported by the data.
This is a shame, because basically, it seems to be a decent plant-based proxy.
Interestingly, it also reveals the “U-shaped” response of plants to temperature. While both extra high and extra low temperatures slow plant growth, the assumption in plant-based proxies is that slow growth = cold temperatures. The problem can be seen by splitting the record into two parts, years warmer than average, and years colder than average. The error between the instrumental data and the grape regression data (intstrument – regression) is larger in the warmer half of the data. For warmer years, the error is 0.27°, and for colder years, it is 0.20°. While the difference is not statistically significant because of the low number of data points, it is interesting nonetheless. When I get time, I’ll repeat the analysis with a tree-ring dataset.
w.
#33, Chris. For the equations, see the author’s Supplementary Information. (I have not criticized the equations.)
#38, Ken. I wouldn’t really expect reviewers of the Nature paper to be much familiar with global optimization; so I didn’t want to criticize here. Instead, I think that those problems should have been detected when the model was first published [Chuine, J. Theoretical Biology, 2000; Chuine & Beaubien, Ecology Letters, 2001]. There are a few problems to consider. First, the algorithm used by the authors is not guaranteed to find the optimum, or even a near-optimum (and in this case it clearly failed). Second, the input data is inexact and some small perturbation of the inputs would very likely change the optimum. Third, the inputs are really only a sample from a much larger population; even if they were exact, it might be that the optimum for the population was different than the optimum for the sample. I found the optimum for the flowering-date model via comprehensive search. The véraison-date model has a five-dimensional parameter space, which has to be evaluated for each near-optimum of the flowering-date model. That’s a lot of cpu time. Given that the model is fundamentally ill-founded (see especially the fourth problem in my list), I didn’t see much point in computing this.
bender. Regarding the title of my paper, I think that you are correct: a better title might have been something like “Grape harvest dates are poor indicators of extreme summer warmth” (i.e. inserting the word extreme). Your question 39.1 makes no sense to me. Question 39.3 is answered above. Your repeated comment suggesting that the model failures are near-inconsequential is something that seems seriously incorrect, as others have also indicated. I encourage you to reread the main comment on my site.
#55–56, Willis. That the authors’ data for 1978 is missing seems bizarre to me too. As for the temperature observations, those are from the weather stations in France (I purchased the data from Météo France, as mentioned on my site). The pre-1922 data is being manually entered from written notebooks; currently only monthly highs and lows have been entered (which is why monthly means for that time are inaccurate). Note that, as per the fourth problem in my list, what should be used is seasonal data. (That data is available on my site back to 1922: I have to be cautious about publishing paid-for data.) As for the discrepancy between the authors’ claim of data back to 1880 versus the actual data only going back to 1883, I wondered about that too: the authors told me that “there is a typo in the starting data year”. There are other small discrepancies in their work—some of which they never explained, despite my repeated asking—but I wanted to focus on the bigger issues.
Re#56
Exactly.
But here’s the conundrum. You have a fine piece of work like this, and your probability of it getting accepted in Nature is directly related to how topical and extreme you can spin it. So they insert a few lines – what maybe four of them? – that could be called into question by a skeptic, but won’t be called into question by Nature reviewers and editors. The writers have played the game as everyone else does. In fact I would say given the strength of their data they were generally more modest than most.
The urge to academic one-upmanship is so strong it brings writers very close to a line that puts their credibility at risk. All papers should be interpreted in that context. These are not factual reports. They are spun-for-market pieces designed to achieve maximum exposure and readership, all to keep the wheels of natural science lubricated with grant money.
It’s a very good, modest paper.
Re #57
If you can’t understand #39.1, what can I say? The question is: what did the authors say to you when you offerred to show them a better optimization method that provided a more convincing parameterization? What did they say when you showed them you could lower SEE by a certain amount? Did you strike up a collaboration to see what the impact would be on their reconstruction?
I understand why some people might want to attack fradulent work and expose it for what it is – AFTER having pursued the more productive, collegial route – but this destructive attitude makes no sense when the work is basically integral.
For example, now I am very curious to see what the impact of YOUR parameterization would be on their reconstruction. But how will I ever get to see that as a science consumer if you don’t work WITH them?
BTW, knowing as much as you clearly do about optimization, you’ll understand that the lowest SEE is not necessarily the “best”. Your parameterization may be stuck in a divot in the error surface, whereas theirs might lie in a more broadly shaped bowl. In which case it is arguable whether your parameterization is as robust as theirs. [Depending on their level of experience with optimization, the authors might never even have considered such a thing.]
Zweifel et al have an article testing CO2 impact on alpine conifers and 2003 happened to be in their study period. Surprise, surprise – 2003 had narrow ring widths. I need to post on this.
There was a good paper in Nature showing 2003 to be a year of tremendous reduction in tree “leaf area index” and rates of photosynthesis (and therefore C-fixation) across Europe. From this one would expect narrow rings. The negative responders. Divergence. The diminished MWP. etc. I’ll dig out the ref. I’m sure CO2 science will have it.
This is where I was coming from with that jabber about bristlecone and pinyon pine. Pinyon pine were hammered during the recent drought in the SW US. I sure wish someone would update the SW US bcp chronologies.
Ciais P, Reichstein M, Viovy N, Granier A, Ogee J, Allard V, Aubinet M, Buchmann N, Bernhofer C, Carrara A, Chevallier F, De Noblet N, Friend AD, Friedlingstein P, Grunwald T, Heinesch B, Keronen P, Knohl A, Krinner G, Loustau D, Manca G, Matteucci G, Miglietta F, Ourcival JM, Papale D, Pilegaard K, Rambal S, Seufert G, Soussana JF, Sanz MJ, Schulze ED, Vesala T, Valentini R. 2005. Europe-wide reduction in primary productivity caused by the heat and drought in 2003. Nature 437(7058):529-33.
Future climate warming is expected to enhance plant growth in temperate ecosystems and to increase carbon sequestration. But although severe regional heatwaves may become more frequent in a changing climate, their impact on terrestrial carbon cycling is unclear. Here we report measurements of ecosystem carbon dioxide fluxes, remotely sensed radiation absorbed by plants, and country-level crop yields taken during the European heatwave in 2003. We use a terrestrial biosphere simulation model to assess continental-scale changes in primary productivity during 2003, and their consequences for the net carbon balance. We estimate a 30 per cent reduction in gross primary productivity over Europe, which resulted in a strong anomalous net source of carbon dioxide (0.5 Pg C yr(-1)) to the atmosphere and reversed the effect of four years of net ecosystem carbon sequestration. Our results suggest that productivity reduction in eastern and western Europe can be explained by rainfall deficit and extreme summer heat, respectively. We also find that ecosystem respiration decreased together with gross primary productivity, rather than accelerating with the temperature rise. Model results, corroborated by historical records of crop yields, suggest that such a reduction in Europe’s primary productivity is unprecedented during the last century. An increase in future drought events could turn temperate ecosystems into carbon sources, contributing to positive carbon-climate feedbacks already anticipated in the tropics and at high latitudes.
I take it “et al” has fallen out of favor in the science rags?
1. This is a reference, not a citation. “et al” is not used in refereces.
2. It came from a database and it was faster for me to copy and paste than re-edit.
3. This was a large-scale co-operative study. Like the FACE experiments in the US, all such papers typically have many authors.
Willis E, in comment #56, do not you mean that the grape thermometer underestimates the Dijon instrumental temperature but overestimates the 2003 instrumental (unless the temperatures where mislabeled, which I doubt). Thanks for presenting the Dijon instrumental temperatures as they are the grape harvest dates that are used as the basic reference series as described in the SI in Nature. I was surprised not to find the Dijon temperatures in the paper or the SI for the paper.
The phenological model development (for flowering and ve`raison) outlined in the SI is impressive but brings into view the very limitations that it possesses. It is tied to only the April, May, June , July and August temperatures and the models used daily mean temperatures in constructions that depend on discontinuous functions (flowering) and third order polynomials (ve`raison) for inputs while the output is given in an average seasonal temperature. The models make use of parameters to fit 16 and 17 year periods of harvest date to temperatures, using evidently daily mean temperatures, and then cross-validated using a held out 11 and 10 years where the output is necessarily the seasonal mean. While the R^2 values are extremely high for this cross-validation one would have to study these models further to determine how they can be so apparently capable of estimating mean seasonal temperatures.
The model constructions and cross-validations were carried out over the 1960 to 1989 time period for the Dijon region which would leave a considerable time period for looking model performance out-of-sample providing the authors had not already peaked.
I also want to re-ask the question of D. Keenan about what the graph within the main graph in Nature paper is referencing. Is that reporting stations for grape harvests dates? On squinting at the original I can see what appears to be maybe a single reporting station for most of the period (Dijon, as explained in the SI?) and then unexplained additional stations for short periods over the entire time span. The SI mentions a total of 18 reporting locations with Dijon being the single location available over the entire time period.
#58 — It’s actually possible that the few extremist lines were added by the Nature editors. Publishing in Nature is not like publishing in specialist journals. Nature’s editors involve themselves with composition and often re-write parts of the paper for effect. A careful author can have a hard time holding the line against them. Someone less careful, or less familiar with the process, or perhaps more ambitious about getting a paper in Nature, may let the Nature editor get away with inserting something that is not consistent with the data, or that over-plays the data. I was a co-author on a fairly prominent paper published a few years ago in Nature (on archaeological sulfur). Our lead author was pulling out his hair, trying to keep the Nature editor from changing things around to be more spectacular.
“Unprecedented” — there’s a word that is used and abused! It always makes things sound scary, yet almost everything we see in our world today is “unprecedented”. The world population is unprecedented, and, in fact, has always been! So everything we touch will be as well.
And in an era where we are coming out of a little ice age, it’s also likely that “unprecedented-during-the-last-century” climate-related phenomena will occur on a regular basis.
Still, for journals such as Nature, the trick always works: add an “unprecedented” somewhere, and an average paper suddenly becomes an important and significant new report!…
Re #64, Ken, thanks for pointing out my typo when you say:
You are correct. After half my post was eaten by the blog, I re-wrote it in a hurry at around midnight. Your revision is the right version.
w.
Re #65
I did not know that. I have published solo in Science and had the same experience: much interference from the Editorial staff, all for spin. I thought my case was unique. I would like to hear Chuine’s account of their experience with Nature.
Re # 57, Doug, thanks for your prompt response. In it you say:
My concern is that both the GISS data and the ECA data only go back to the fifties, while they are claiming data back to the 1880s. GISS and ECA have fairly strict policies about data quality, and will cut off the early part of a series if it does not meet their standards. GISS also will not commingle two datasets under one name.
Thus, the reliability of the pre-fifties data is questionable. It is the best we have, to be sure, and should be used … but only with an accompanying disclaimer.
Re #58, bender, your comments are always welcome. However, your theory that the offending claims of “extraordinary” and “probably higher than in any other year since 1370” were inserted by the editors at Nature, rather than by the authors, is a theory without evidence to back it up.
In particular, one should always be very careful when drawing conclusions from outliers with huge sigma values. My father used to tell me “Son, if something seems too good to be true … it probably is”. In this case, a bit of research should have showed the authors that a large reason for the early harvest was not the heat, but the lack of water.
In addition, further reading of the paper’s supporting online material reveals the following:
The authors use a mathematical model which includes an equation for the “photosynthetic activity curve”. This describes the amount of photosynthesis performed by the plant (growth). The equation is
with
the mean temperature of day t and m, n, o, and p being coefficients specific to the plant. The authors say the function varies “between 0 when temperature is superior or equal to 40°C or inferior or equal to -5°C, and 100 when temperature reaches the optimal temperature 
which is comprised between 10°C and 25°C.”
This, of course, is the infamous “U-shaped” response of plants to temperature which we have discussed here previously. Note that the temperature in question is the mean daily temperature, and not the mean monthly temperature.
As noted elsewhere in this blog, this equation does not return a unique answer when it is inverted. There are two possible answers, a high one and a low one. This to me is one of the underlying, unsolved problems with paleodendroclimatology “¢’¬? which one do you pick?
The authors say that:
Now, we need to watch the pea under the thimble very closely here. They are taking an equation which depends on the mean daily temperature and inverting it to solve for temperature. This gives two answers, a high one and a low one. They are then comparing that inverted daily temperature to a five-month mean temperature, and adjusting the parameters of the model so that the 5-month average temperature agrees with the higher value from the inversion.
This problem is particularly great when the possible temperatures fall on opposite sides of the “U-shaped” photosynthetic response curve. However, since agricultural crops are generally grown where the average local temperature is optimum for the particular plant, with ag crops we are almost guaranteed to have this situation.
Here are some approximate photosynthetic curves for different temperature optimums, using the author’s equation:
The black curve, with an optimum at about the Dijon average temperature, is nearly horizontal in the range where 95% of the Dijon April-August temperatures occur. This makes the inversion of the curve even more problematic.
Since Doug has provided more temperature data, I am able to calculate the average error for a longer period. Looking, as before, at the average error for high temperatures vs low temperatures, the difference is more pronounced. With the additional data, it is also now statistically significant, showing that in fact the “U-shape” of the temperature response curve does in fact result in a general underestimation of temperature of the the warm periods compared to the cold periods.
Finally, the authors say:
This reveals a couple more problems. First, the issue is the standard deviation of the delay (3.8 days), not the standard error. With a standard deviation of 3.8 days, the 95% confidence interval of the delay between veraison and harvest is ±7.6 days.
Now, in their model, a 1 day change in the harvest date equates to a temperature change of 0.1°C. This yields a 95% confidence interval on the veraison to harvest dates of about 3/4 of a degree. As their claimed overall 95% confidence interval for the model is 1.15°C, it is probable that they have not included this factor in their uncertainty estimate.
The second problem is that the difference between veraison and harvest dates depends on rain. If the grapes are a bit moldy and rain is forecast, they will be picked early. (This is what the authors mean by “the interaction between the phytosanitary state of the vineyard after véraison and the occurrence and amount of rain.) But on the other hand, they will also be picked early, as we saw in 2003, if there is not enough rain … another “U-shaped” response curve.
w.
Re: #69
Willis E, thanks for looking into the models in more detail. I do have a couple of questions/points to ask/make.
The model that you analyzed above is for the veraison date and involves a 3rd degree polynomial. The curves that you show appear to more those expected for a 2nd degree polynomial. My questions is did you actually calculate the coefficients for the x^3, x^2 and x and find that the coefficient for x^3 was relatively small?
This model also involves a second term related to 1/(1+e^a(x-b)), but my mind is not sufficiently abstract to see if it is this term in combination with the first that helps produce the parabola like curves in your graph.
The authors have used the flowering date in a second model and that uses what I call a discontinuous function, and that would not involve solving for roots or at least as I see it. The cross-validation R^2 for the veraison model was 0.913, and that for flowering model was 0.818.
When you say, “However, since agricultural crops are generally grown where the average local temperature is optimum for the particular plant, with ag crops we are almost guaranteed to have this situation” this may explain some simplification that makes the mean daily temperature a more practical predictor of maturation. I, however, have a problem seeing how a mean temperature could give a validation this good when remembering that the maturation has to be affected by the sum of momentary temperatures during the day as applied to some equation similar to one used in the models. If these variations around the daily mean were more or less constant over time I could see the mean being a good predictor, but I am not sure this is the case. Finally when the output is provided as a growing season average I have an even larger problem with changes in the variation around the mean for the season, as well as daily.
As you say these models results seem almost too good to believe. True out-of-sample results would certainly help in the belief department. The following excerpt from here is also bothering:
Are you sure that is exactly what they did? Or is that an assumption/first guess? These folks are not amateurs, you know. If I were to tackle this problem I would not do it the way you describe, and so I doubt they did it that way. And if they did I am willing to bet they thought about the importance of the location of the optimum. These people are not hockey-sticking cherry-pickers. They know their system.
Not to dissuade you from the critique though. This will illustrate for all exactly what the difference is between a good proxy and a bad proxy. Careful science and reckless science.
P.S. Editorial intervention is not “my theory”. I have no idea what happened. That’s why I said it would be interesting to talk to the authors.
Re 70, Ken, you say:
As you can see if you inspect the left and right curves closely, there are two inflection points. I did actually calculate the coefficients. Here’s the coefficients for two of them:
______Optimum______ m_____ n_____ o_____ p_____
_____________16____0.0004___-0.23____7.47___38.64
_____________35____0.005____-0.46____8.45___53.41
If you use a 2nd degree polynomial, you can’t vary the optimum point and still keep -5° and 40° at zero. (I did a quick emulation, didn’t bother to get them exactly to zero.)
Bender, you ask about the inversion. Here is their description. I have already established that they used daily temperatures for the initial calculation (see the equation above). About the inversion, they say:
As we know, there are two temperatures that would provide the same harvest date … I don’t know how they dealt with this. However, the fact that the error is larger in warmer times than in colder times suggests that they used the “earlier harvest = hotter” relationship. Unfortunately, we don’t have the full 1883-2003 temperature dataset to verify this, but it sure looks like what they did.
Sorry for calling it “your theory”, you are right, the best would be to talk to the authors.
w.
http://www.rense.com/general59/reveal.htm
http://www.gsenet.org/library/11gsn/2005/gs050218.12.html
So Chuine’s an alarmist, is that it, Ken?
Willis, I have a question for you. Having to do with the number of days during the calibration period that the observed temperature exceeded a certain threshold. I’ll ask it in full tomorrow. Meanwhile, see if you can guess what I’m thinking.
Re 75, good question, bender, but I haven’t a clue. I don’t have any daily temperature data at all …
On another matter, Chuine is an alarmist regarding extinctions, and if she wrote the “extraordinary” and “probably warmer” parts of her report on wine, regarding temperatures as well. However, looks like the wine could be a good proxy if treated right, coundfounding variables removed somehow, you know, the usual …
w.
And by “alarmist” I should specify – an alarmist who lets their alarmism color their science? The author of this paper may be alarmist as evidenced by materials outside the paper. But I don’t see the alarmism creeping into the methods, results, conclusion.
Re #76
Unfortunately, these data will be necessary to answer my question, just as they are necessary to support the view that the argument in #69 is relevant. (I grant that the argument contains no logical flaws. The issue is quantitative: what percentage of days during the growing season will this problematic nonlinearity kick in?)
If daily temperatures spend most of their time on the left side of that optimum point, rarely dipping into the abyss on the right, then the linear reconstruction model is going to work pretty darn well. If you want to dismiss the Chuine reconstruction you’re going to have to prove that the nonlinearity argument comes into play during a significant number of years in a signficant number of years. I don’t expect it does …
My very first post at CA was on nonlinearity of short-term responses of plants to temperature, and conditions under which these scale up to seasonal and secular time-scales. So I warn you that I’m experienced in this area. The Chuine paper is pretty darn good …
Looking at Figure 1 in their paper does anyone see to what Yiou is referring? Besides this reconstruction was primarily from a single locale, Dijon, and was purported to cover only the months of April, May, June, July and August of each year. He may have been misinterpreted, of course.
Bender, whatever the views of the authors on this paper (I think they can be helpful to know sometimes), I think their work deserves more analyses than they presented in the paper themselves. My interest was piqued by what D. Keenan has presented about the paper — more in general terms than about the extreme annual events — and then further on going back to the Nature SI for the paper. I found that in delving into the basis of their models that more new questions became apparent than old ones answered.
Their models are something like I envision for the computer climate models in that they start from some basic first principles of biology and then do some heavy parameterizing to fit the model to some historical results. I think their models may be more readily tested out-of-sample and analyzed to determine whether there are some fundamental limitations to predicting past temperatures.
bender, let’s consider some of your assertions.
“your probability of it getting accepted in Nature is directly related to how topical and extreme you can spin it … These are not factual reports. They are spun-for-market pieces designed to achieve maximum exposure and readership.” We agree on this point. Where we disagree is that you believe making non-factual claims in order to get published is ethical. It is incredible to me that you would support such unethical behaviour, which I believe is destroying science.
“It’s a … modest paper”. In fact, it falsely claims to show that the summer of 2003 was extremely warm relative to any other year in the past 600 years. So we disagree on this point.
“The question is: what did the authors say to you when you offered to show them a better optimization method”. I exchanged tens of e-mails with them, as stated on my web site; the first several were cordial. An extended quote from the authors is included on my site, which should make clear their attitude when I began pointing out some of the problems. Here is another quote: “We have read your comments carefully. We do not wish to comment them more than necessary …… our time … is rather limited”.
You suggest that perhaps the claim that 2003 was extraordinarily warm might have been due to the editors of Nature. As has been pointed out, you have no evidence for this. The quote from Chuine et al. on my site is evidence against, as is the quote from the Globe and Mail cited above by Ken Fritsch.
“These folks are not amateurs, you know.” So now you are defending them on the basis of their “authority”. Perhaps you should do the same in defending Michael Mann.
Your claim about GO parameterization is a red herring, given the issues I listed. Moreover, this misses the point that I was making, explained on my web page.
“I don’t see the alarmism creeping into the methods, results, conclusion”. They have failed to provide any grounds for the conclusion in their paper that drew the attention. They have smoothed the graph without any justification; they have erroneously used standard errors; they have incorrectly inverted their model. All leading to falsely-small errors in validation. I, at least, do see.
.
.
.
Wikipedia has had issues similar to the above, with biased editors resorting to sophistry in order to twist articles to their extreme points of view. Wikipedia has mechanisms to deal with vandalism, but nothing to deal with sophistry. (ArbCom does not intervene.) The problem seems to be growing; see too this Slashdot comment:
http://slashdot.org/comments.pl?sid=195779&cid=16044499
A friend wrote a letter to Jim Wales (founder of Wikipedia) a couple weeks ago about all this, but did not receive a reply. I see that Wales has just resigned as chairman.
bender, I don’t want to play sophistry, okay? I ask you to respect that.
A few people indicated that they would like to have more data. If anyone does, feel free to e-mail me.
#64, Ken. Regarding your question about the authors’ graph, I did not ask the authors about this.
Re 78, bender, you ask an interesting question above regarding the non-linearity of the plant growth, and what percentage of the time a plant has to spend on the “other side” of the hump to make a quantitative difference. Your implication is that the difference is not significant. I disagree. Here are some points that will help to clarify this:
We can begin by noting that we’re not talking about a percentage of 24 hours here. No photosynthesis goes on at night. We’re only talking about daylight hours.
Next, the answer will be different at different times of the year. During April, extreme heat is much less common than in August. And it will be different on each farm, due to the differing microclimates.
It is also different at different stages in a plant’s growth – the optimum temperature for flowering is different from the optimum for veraison, which in turn is different from that for harvest.
The key point, however, is that agricultural plants such as wine are generally grown where the average temperature is about right. This implies that on average, about half the time it’s a bit too cold for the the plant, and half the time it’s a bit too warm for the plant, yielding maximum growth.
It also implies that in an unusually warm year (or on an unusually warm day), it will be too warm (the “other side” of the hump) for more than half the time. This is supported by the data, which shows a statistical difference in model error between warm and cold years which is consistent with an error arising from non-linearity.
A final point about this. The photosynthetic response curve I plotted above is a function, not just of temperature, but of water as well. Lack of water pushes the curves to the left, meaning that they begin suffering heat stress at a lower temperature, and spend more of their day heat-stressed (on the “other side” of the curve. If it is dry enough, the plants do not even recover overnight, and start the day out already wilted and heat-stressed.
It is also worth noting that a plant is more stressed by excess heat than excess cold. When it’s cold (barring frost) the plant just grows more slowly. But when it’s too hot, plants can get physically damaged. This stunts the growth more than a corresponding cold time.
So yes, I do think this is a significant effect, and the thesis that it is significant is supported by Chuine’s data.
w.
D. Keenan, I have resolved my search for the explanation of the upper graph in Figure of the Chuine paper in Nature. My old eyes failed to pick the correct color and small print explanation in the Figure 1 caption. The graph indeed, as I expected, indicates the number of yearly harvest dates used in the reconstructions. Now if I could resolve the graph into actual numbers I would have more pertinent information. This experience is why I keep thanking Willis E for his very legible graphs even though I must take full responsibility for not being sufficiently observant in my first readings of the Nature graph.
Bender, you seem to be much impressed with Dr Isabelle Chuine’s models and I was curious if this results primarily from her evidently better use of phenology in constructing reconstruction models than the linear simplifications made by dendrochronologists (some say that her attractive models run rings around the dendrochronologists’). At the same time her models give evidence (as demonstrated by Willis E) of their own limitations. I personally think it would be fun to carry the evaluations of her models to out-of-sample years in Dijon and to other areas where pinot noir grapes are grown such as Oregon and Australia. I would be particularly interested to what degree the inverse regression reduces the prediction of past variations in temperature. It would be interesting to pump simulated hourly temperatures into the model with differing patterns but with the same mean.
Below are some excerpts that I find in perusal of the internet contents on pinot noir grape maturation.
http://www.theeapress.com/harvestdata.htm
http://www.inmanfamilywines.com/wines/pinot_noir.html
Click to access 2004RussianRiverPinot_000.pdf
http://www.wynboer.co.za/recentarticles/0403phenol2.php3
Roger Pielke Sr. discusses a paper today Was the 2003 European Summer Heat Wave Unusual in a Global Context? today that has a direct bearing on Chuine. Two of their conclusions are
This paper only looks at data back to 1979.
RE: #85 – Here on the W. Coast of North America some of the more dramatic instances, under even normal conditions, of extreme heat relatively near the ocean are found between about 38, 30 N and 42 N. That is in the transition zone between the Mediterranean and Marine West Coast climate zones (some have actually classified those particular California Counties as Marine West Coast, but I personally disagree with that). I digress. By extreme heat, I am alluding to summer day time temps in the 110 Deg F plus arena – places like Ukiah and Calistoga are notorious for this. Only a fairly minor perturbation in the persistent summer time Subtropical Marine High poleward, coupled with a series of semi stagnant Highs in rapid succession to the northeast of the big oceanic High (Bermuda – Azores High in the Atlantic, the Hawaiian High in the Pacific) and extreme heat and low humidity fall into place to latitudes as high as the low 50s. Contrast this with the ongoing damp, cool and often dreary weather from 55N to the Marine Subarctic climate zone boundary. Being on the Southeastern edge of the Marine West Coast climate zone means that at times, Mediterranean type weather will intrude.
RE: #86 – I meant Southwestern edge, but to be antipodally correct, what I really should have written was the Equatorward, seaward edge. 😉
I was somewhat unclear as to exactly what the grape thermometer was supposed to be measuring so I went back and attempted to reconstruct some phenological periods for the Dijon Pinot Noir grape. Here is what I found using what I have assumed to be average temperatures for that area and the equations provided in the SI for the Chuine Nature article:
Flowering model starts at April 2 and flowering starts June 15.
Cell elongation evidently starts after flowering and finishes July 15.
Photosynthate accumulation starts after cell elongation and finishes August 15.
A period of approximately a month occurs between the end of photosynthate and the beginning of harvest where maturation is effectively independent of temperature.
Harvest begins September 15.
The Nature article implies that the temperature reconstruction works with the temperatures in the spring/summer months of April, May June, July and August. On closer look, I come away with the distinct view that the authors used two different thermometers, one based on the temperatures leading up to flowering April, May and ½ of June, while the second used the temperatures from mid month June to mid month August. The article discusses these two thermometers separately and gives validation test results for each separately. In the end it notes that it uses the second thermometer for the reconstruction back over 600 years as it is more precisely correlated to the historically recorded harvest dates.
What would one conclude from this information? It would appear that the temperatures reconstructed must then necessarily be limited to the period from approximately June 15 to August 15, or only 2 months of the year.
If one looks further into the verasion maturation process used in the second grape thermometer, the article explains that it is divided into two separate phases that depend on temperature differently. The first phase is cell elongation and it follows a function of temperature as in:
Re = 1/(1+e^(a(t-b))).
For that process the sensitivity for cell elongation to average mean temperature, t, goes from 0% to 100% on the t going from 10 degrees C to 30 degrees C. From 15 degrees C to 25 degrees C, the sensitivity goes from approximately 10% to 95%.
The second phase of verasion is called the photosynthate accumulation and it depends on temperature, as was previously describe by Willis E :
Rp = mt^3 + nt^2 +ot + p.
Here the sensitivity goes from 80% to 100% and back to 80% on the temperature going from 10 degrees C to 30 degrees C and from approximately 95% to 100% back to 95% on going from 15 degrees C to 25 degrees C.
From the forgoing, one sees that the verasion grape thermometer is most sensitive to temperature changes in the month of the cell elongation period and the thermometer becomes in essence a one month of the year thermometer. One must also question how an average daily temperature for the entire verasion period is fit to these differing temperature sensitivities.
If my analysis is incorrect, perhaps D. Keenan is sufficiently familiar with the subject to point out where I might have gone wrong.
#88, Ken. Chuine et al. did indeed use different thermometers. They get away with it in their model calibration, because the calibration inputs daily temperatures, rather than seasonal temperatures. This is the fourth problem in the list (on my site) of things that I thought the peer reviewers should have caught.
The flowering-date model is used by the authors, to obtain the start date for cell elongation (tf).
Your point about the model being much more sensitive to temperature during the cell-elongation phase than during the photosynthate-accumulation phase seems right. I had noticed that the temperatures during the pre-flowering phase were less important, but not the difference between the importance of the two véraison phases. Perhaps Chuine et al. would say that their model performs a weighted average of the temperatures.
I just checked the squared-correlation of AMJJA temperatures with mid-June/mid-July temperatures: over the 77 years for which data was available, it is 0.43; over the 17 hottest years, it is 0.36; over the 9 hottest years, it is 0.17. (These are of course estimates with error bars.) The year with the hottest mid-June/mid-July temperature, by far, was 1976.
And of course the 64k question for Chuine is:
Where is the archive of the Dijon Temperature data?
In a thread about Hegerl et (2006), I commented that the authors’ lack of statistical competence should be evident to anyone with an appropriate statistical background without even seeing the data. This appeared to be “further evidence that the peer reviewers didn’t have an appropriate statistical background to review the paper, or they didn’t read the paper, or they didn’t care that it was wrong”.
Tim Ball responded, suggesting that the problem arose because the 43 people that Wegman identified as co-authoring together were also peer reviewing each others papers. I thought that was a good suggestion.
The work of Chuine et al., however, has even more obvious statistical incompetence than that of Hegerl et al. (again, even without the data). Yet Isabelle Chuine is not one of the 43, nor even close to them (it seems). So Tim’s suggestion doesn’t seem as good anymore, unfortunately. And the statistical incompetence is more pervasive.
Hans (#90), Chuine didn’t archive any of the temperature data, but Keenan has put up links to portions of it:
http://www.informath.org/apprise/a3200/b2.htm
#91. Sara, your comment is bang-on. Given that Nature has explicit policies on statistics that I’ve posted up in the past, it is also interesting that Nature makes no efforts to ensure that its policies are complied with in climate science publications – which is particularly objectionable given that readers could reasonably expect that reviewers would, at a minimum, ensure compliance with journal policies. I doubt that climate reviewers even know about Nature’s statistical policies. Maybe someone who’s acted as a Nature reviewer could tell us about whether Nature told him as a reviewer about such policies and asked reviewers to attend to them.
Here is a comparison of Luterbacher Chuine and De Bilt
De bilt is available as monthly observed dataset (1706-2006)
Chuine as AMJJA seasonal modeled temperature (1370-2003)
Luterbacher as MAM and JJA seasonal modeled temperature (1500-2004)
Data sources:
A van Engelen and Nellestijn, JW, 1996: Monthly, seasonal and annual means of air temperature in tenths of centigrades in De Bilt, Netherlands, 1706-1995. KNMI report from the Climatological Services Branch
http://home.casema.nl/errenwijlens/co2/LABRIJN.xls
extended with giss data (2001-2006)
http://data.giss.nasa.gov/gistemp/station_data/
Chuine:
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/historical/france/burgundy2004.txt
Luterbacher:
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/historical/europe-seasonal.txt
Not bad.
#94, Hans. I calculated the squared-correlation between the AMJJA observations at De Bilt and the modelled data of Chuine et al. For the available 297 years, it is 0.44; for the 149 warmest years, it is 0.30; for the 74 warmest years, it is 0.18; and for the 37 warmest years, it is 0.04.
So, grape harvest dates in Dijon are poor indicators of extreme AMJJA warmth in De Bilt.
Like I said, plants are not thermometers … they read high temperatures as low.
w.
A couple of interesting points I just noticed.
First, the r^2 of De Bilt temperatures with Dijon temperatures is 0.58 (1922-2003.
Second, the r^2 of grape thermometer temperatues with Dijon temperatures is 0.65.
Finally, the r^2 of grape thermometer temperatues with Dijon temperatures is 0.66.
In other words, after all of the calculations of veraison dates and flowering dates, after all the analysis of photosynthetic rates, after all of that … the raw harvest dates are a better predictor of the Dijon temperatures than the Chuine calculated temperature … go figure.
The same is true in the longer term. The raw harvest dates are a better predictor of De Bilt 1706-2003 temperatures (r^2 = 0.47) than the Chuine calculated temperatures (0.44).
w.
Re #96
Like I said: not bad.
Re #98
Does this surprise you? Is it somehow damning?
Willis E, I agree with your observations that, in the end, a correlation with harvest dates is more impressive than the model that Chuine et al. were able to produce and rationalize. I also agree with Bender that the correlations are impressive — taken at face value. Finally, I agree with D. Keenan that the reconstruction model will not be able to accurately handle extreme temperature events based on his analyses and the limitations that derive from the truncating nature of the effect of temperature on the grape maturation as indicated by Chuine et al. and particularly so for the verasion process.
I worked backwards to derive the following equation from Chuine et al that relates temperature anomalies (t) to harvest dates (d):
t = -1E-05d^3 +0.0021d^2 -0.1693d +3.1509. Fit with R^2 = 0.9968.
Based on my analysis of the rationale for the Chuine model, I found that the most sensitive part of their grape thermometer would derive from the months June and July (JJ), but was puzzled why the model on inverse regression called on the average temperatures for April, May, June July, August (AMJJA). The verasion period that the model claims to use (as a second grape thermometer with other one being one based on April, May and part of June grape flowering) would only be connected with the months JJA.
In order to analyze this situation further I looked at harvest dates versus various the average temperatures from consecutive months of the year. I used the Dijon monthly temperature data to which Hans Erren provided a link for the period from 1951 to 2003 and the harvest dates provide through the Chuine link. I looked at the R^2 values for various monthly combinations for a fit to a linear and 3rd order polynomial regression:
JJ(0.49, 0.61); JA(0.44, 0.50); JJA(0.62, 0.66); MJJ(0.61, 0.64); AMJ(0.52,0.57); MJJA(0.71,0.72); AMJJA(0.73, 0.75); MAMJJA(0.69,0.70); MAMJJAS(0.67, 0.67); FMAMJJAS( 0.56, 0.62); FMAMJJASO(0.52, 0.56); JFMAMJJASON(0.47, 0.54); AllMonths(0.42, 0.48)
To me the model rationalization by Chuine et al. is misleading and in the end it would appear that a model based on some biological first principles that can be sufficiently parameterized will approximate what could have been obtained by simply regressing the average growing season temperatures against harvest times.
Now if Hans Erren or D. Keenan or anyone else can provide some daily/hourly Dijon temperatures over an extended period, the extreme temperature limitations of the model could be more thoroughly and empirically tested.
Re #99, bender, you say (regarding the fact that a plain vanilla linear regression of the harvest dates does better than Chuines temperature reconstruction):
While it is not damning in the slightest, it surprised me. I would have guessed that a more sophisticated analysis would give getter results, and that if it didn’t, it would be discarded. However, I guess that getting the result that way is more “scientific” … I haven’t yet analyzed the differences in the results.
w.
you are asking the wrong person:
Now if Dr. Chuine or anyone else can provide the Dijon temperatures that were used in the study over an extended period.
Here is Metar data for Dijon (online available starting 1996)
http://www.weatherunderground.com/history/station/07280/2003/8/12/MonthlyHistory.html
Re: #102
Hans Erren, that link was just what the doctor ordered. I can look at the maturation models of Chuine and apply them on an hourly basis to determine just how limiting biological first principles are on the use of a grape thermometer. I think that they may be more limiting than indicated by the parameterized models — that, in turn, appear closer to straight regressions of temperatures on harvest dates.
Besides when She Who Must Be Obeyed comes with the next remodeling (house that is) assignment, she seems more forgiving of seeing spreadsheets on the computer than web pages.
LOL – I have found the same thing! R scripts good; conspiracy theories bad.
Ok, bender, name the “chateau” of Rumpole’s plonk (I almost said Rumsfeld’s plonk).
I’ve added an Addendum to the page describing more technical issues with Chuine et al. This tells how the model performs no better than linear regression, and thanks Willis for pointing that out.
Following on from #88, by Ken, I also tested correlations between mid-June/mid-July temperatures and harvest dates. The correlations, though, were much smaller than with AMJJA temperatures.
#99, bender: I don’t want to play sophistry.
Re: #105
http://en.wikipedia.org/wiki/Horace_Rumpole
I loved that character, Rumpole, (and SWMBO. It was characters like him that made work in my former life more fun than work.
I had some grandiose plans to do an in depth analysis of the Chuine model using the more detailed Dijon temperature data sourced by Hans Erren. Alas, I settled for something simpler, but if I did my calculations correctly, probably almost as revealing, albeit limited in general conclusions to be derived.
I used the daily Dijon temperatures for the years 2002 and 2003 to calculate the flowering, cell elongation completion and photosynthate completion dates using the Chuine pinot noir grape maturation equations published in Nature. These are the equations that use daily average temperatures and then on inverse regression use the AMJJA mean temperature to predict a harvest date. Remember that 2003 was the year of the very hot summer in Dijon, and Europe generally, and was emphasized in the Chuine et al. Nature article.
The calculated flowering, elongation completion, photosynthate completion and harvest dates for the years 2002 vs 2003 were June 19 vs June 9; July 5 vs June 25; August 11 vs August 1 and September 12 vs September 2, respectively.
The harvest dates predicted, using the Chuine et al. inverse regression model and the AMJJA mean temperature, for 2002 vs 2003 were September 18 vs August 18, respectively.
The AMJJA mean temperatures for 2002 vs 2003 were 22.0 degrees C vs 25.0 degrees C, respectively. The mean temperatures for the entire year for 2002 vs 2003 were 17.2 degrees C vs 18.1 degrees C, respectively. The monthly mean temperatures in degrees C for the grape growing season of AMJJA for 2002 were 16.7, 18.9, 25.6, 25.6 23.3, respectively and for 2003 were 17.8, 21.7, 27.8 26.7 31.1, respectively.
It turns out that the Chuine equations (not the inverse regression) differentiate between 2002 and 2003 only in the flowering period and the differences depended on the temperatures in only a 2 months time period, i.e. the latter part of April, May and the first part of June. Although the accumulation of the flowering factor started in early April for 2002 and 2003, it really does not contribute significantly until the middle of that month. On closer examination with actual temperatures, it turns out, as Willis E noted earlier, the photosynthate process is relatively insensitive to the normal temperatures at which it operates in Dijon and even slows when an optimum temperature is exceeded. I had earlier noted that the cell elongation process had a steeper slope with temperature than that of the photosynthate process and suggested that because of this one might expect cell elongation period temperatures to correlate better with the harvest dates than the photosynthate period. For the years 2002 and 2003, the time period for cell elongation had mean daily temperatures that all exceeded the sloping portion of the curve and put them into a temperature independent range. Both years had a calculated 16 days for this process to complete. I would expect that only a significantly cooler period than normal would have an effect on the length of the cell elongation process.
If one makes correlations with 2 month periods of mean temperatures (1951-2003) to harvest dates, the best correlation is for the 2 months of May and June which gives a linear regression fit with R^2 = 0.54 versus the second best 2 month combination of June and July which yields R^2 = 0.48. This would be in agreement with what I found for the years 2002 and 2003 using the Chuine equations, but the Chuine equations do not appear, at last from a precursory analysis, to explain why the combinations of MJJ with R^2 = 0.61 and MJJA with R^2 = 0.72 and AMJJA with R^2 = 0.73. These results lead me to suspect (from the 2002 and 2003 data) that while the Chuine equations/processes play a role in the pinot grape maturation process, and thus the harvest dates, there are other considerations which put the latter part of the growing season temperatures back in play in determining the harvest timing.
Since I consider my analysis as more “playing around with the data” then anything of a serious nature, I did not bother with tabling my calculations. But I do have a question to John A on whether I can copy tables and graphs directly from my computer (Excel/Word) or must they first be transferred to an online site?
Before relying on computer algorithms, Dr Chuine should have read things such as the 1844 story from abbot Cochet who wrote about the history of grappe growing in Normandy.
http://www.bmlisieux.com/normandie/cochet01.htm
Some excerpts:
-It is unrefutable that Normandy was a production site of excellent wines and its hills, now covered with apple trees were once the place of abundant vineyards.
-The evidences of this once prosperous activity are numerous, be it written testimonies, architecture, monuments, traditions, chronicles, objects, town names…
-Testimonies go back to 12th century but “recent” events are also plenty, for example the Battle of Arques on 21th september 1589, recorded by the duke of Angouleme where the cavalery had difficulties to move because of the vigourous vineyard.
-In the 13th century, the first harvests were obtained on the 6th of AUGUST!
To give an idea, the feast of the Beaujolais Nouveau (fine wine… for the Japanese) officially starts at 0:00h the 15th of November.
-Around the end of the 17h century, long and harsh winters began to set in. 1684 winter was “horrible” and lasted 5 months. Even the ocean was freezed from Tréport to the Havre and the port of Fecamp was totally blocked by ice. 1709 was even worse.
Apparently, there are plenty historical elements describing mediaval wine activities in the North of France. Maybe the next publication of Dr Chuine would be about Normandy’s grappes harvest dates as a temperature proxy?
What is evident from 300 years of observations is that summer and winter temperature in Western Europe is decoupled.
Note that in 2006 for the first time since records began in Holland the boundary between Ca and Cb (hottest month 22 degrees) was passed. The thirty year moving average ( Kàƒ⵰pens definition of climate !), however, is still well within the limits of a Cb climate.
image:
John A, I answered my question in comment #108 concerning HTML codes for tables (and bolding for that matter) with a little experimenting here, but I remain unsure how I would copy a graph or picture to a comment space.
There is an update on the technical comment here:
http://www.informath.org/apprise/a3200/b2.htm#Addenda
The update explains that the model of Chuine et al. is essentially just a cubic polynomial, despite initially appearing much fancier.
Also, I was contacted about Chuine et al. last year by an American TV producer, who was going to do a documentary on climate change. It seems, though, that the producer had misunderstood my paper: he thought that my paper argued against substantial correlation between temperatures and grape harvest dates. A recent post on ClimateAudit suggested a similar misunderstanding.
In fact, the data of Chuine et al. make it clear that temperature explains over half the variance in the Burgundy harvest dates. There is no dispute about that, and the figure on my site (at the above link) illustrates the correlation.
The primary problem with the publication of Chuine et al. is that the authors’ model is inadequate for estimating the temperature in unusually warm years (as detailed in my paper and on my site). An important additional problem is that the model was designed to give results on average, and does not necessarily work for individual years (this is clear and it is also treated on my site). Following is a partial illustration of the latter problem, which I sent to the producer.
Re: Douglas J. Keenan (#113),
Thank you for the update. It is much appreciated.
Doug – We know each other from our days at Lehman Brothers, back in the 90’s. I would like to get back in touch with you, but have found no other way to do this prior to discovering your comments here. Likewise, I no longer am active in trading, but we again share a common topic of interest regarding climate change science.
Please give me an email at mepolhill@yahoo.com, it would be great to catch up with you.
Best Regards,
Mark
James Hansen has just given a presentation to the UK’s Royal Society. Apparently, one of his three main pieces of evidence for an increase in “extreme events” due to global warming occurred in northern Europe during 2003, for which Hansen cites “shrivelled French grapes at the end of Europe’s hottest summer on record, in 2003”.
http://www.independent.co.uk/environment/climate-change/global-warning-climate-sceptics-are-winning-the-battle-2368617.html
One Trackback
[…] spat and Doug appears to think that Nature is being rather disparaging about him being praised on Climate Audit. I can’t quite see the slight myself but then I’m not an academic trying to protect my […]