the Air Vent

Because the world needs another opinion

A Different Yamal – Corrections and Signals

Posted by Jeff Id on October 8, 2009

This post investigates the effects of RCS on the Yamal proxy data. It started out with a thought that it might be possible to create a simple method for creating series from tree ring widths that would act as a sanity check for the other methods. However, it turned into a learning experience with the tree ring data. Here’s what I found.

Yamal mean ring width per year

Figure 1

Figure 1 is an average of the entire Yamal dataset. The change in variance from the zero age to the 400 age is due entirely to the number of trees available in the series. The far left is from 252 trees while the far right is from one. This data is fit to a function of Corrected= a + b * e^-(c*age). However, all the individual trees are fit which of course weights the fit very heavily toward the younger trees. So the question becomes what do the 252 Yamal trees look like when plotted. The red curve is the mean overlaid.

Yamal all series

The next plot depicts the RCS correction factors fit to the orignal Yamal in Blue on top of the mean data.

Yamal mean ring width per year with correction factors
The blue line is an excellent fit up to about 200 years and then it diverges from the data. This occurs because the series is very heavily weighted to younger trees in this form of RCS. My next thought was to try and fit the curve to the mean of the data. This weights all years equally even if less data is available in older trees – Green line. Consider that in collecting tree chronologies it would be much easier to find older trees which are still alive. The fossilized or sub fossil trees are more rare and far more difficult to come by. This would create a problem where the most recent data in typical RCS reconstructions is the least corrected for.

The next plot is the 12 trees which make up the most recent years of the Yamal hockey stick. The red lines are the corrected data while the black are the original ring widths divided by 50 (mean of the correction factor).

Yamal 12 corrected and not
This plot was interesting to me because the red lines in the longer series are almost universally amplified at their endpoints in relation to the original black lines, while shorter series are a decent match. Yad 06 starts out with the historic black ring widths being higher than the recent ones and the opposite in the most recent portion of the curve.

Basically everything makes sense that the living trees are under corrected for in relation to the rest of the series resulting in some bias in the final trend. So the next plan was to do the reconstructions using the two different corrections calculated above.

Yamal different reconstructionsThe black curve is the emulated original per modified code from Steve McIntyre and the red line is the version which uses RCS calculated from the mean of the annual data allowing each year to have equal weight in the RCS curve fit. The end of the graph has dropped by about 0.5 C while the historic portion remains nearly unchanged. Not terribly unexpected considering the fact that the beginnings and ends of fossilized trees overlap prior to 1800 allowing for an averaging of defects in the RCS method. However, the endpoints are strongly affected by different fits simply due to the inhomogeneity of tree age in the record.

While this method makes more sense to me, the net result of this curve is still dependent on very few series.

UPDATE: RomanM’s plot of tree age and specemin count clarifies the point some.

averageage[1]

61 Responses to “A Different Yamal – Corrections and Signals”

  1. P Gosselin said

    I have found rare footage of Steve and Gavin going at it…
    http://www.thesun.co.uk/sol/homepage/news/2670719/Thug-humiliated-on-internet-video.html
    Just a little humour I thought…
    Gee, which one do you think is gavin?

  2. treyg said

    This probably violates some fundamental “law” of RCS, but the upward trend around age 300 would fit better with a positive exponent:

    Corrected= a + b * e^-(c*age) + b’ * e^+(c’*age)

    The younger age tree’s RWs would probably not fit as well, but maybe the overall RMS would be lower.

  3. Jeff Id said

    I want to add on some thoughts to this post because I think the casual reader might miss the point. RCS only affects the ends of the plot. The whole center portion will remain basically unchanged because the positive and negative corrections applied get canceled out in the midpoint of the graph.

    This means that if your series are not densely enough populated and evenly enough populated at all age levels the trees will be underfit the calibration equation and an artificial signal will be added. Since we’re concerned with the most recent years on comparatively old trees. In the case of Yamal, there isn’t any reason that the mean of the series isn’t as accurate as RCS.

    My red curve has substantially lower signal in recent times than the black in the last figure but if you look at figure 2 the tree rings on the right side which represent the long series are part of the most recent 12 that make the hockey stick. You can see their tree ring widths aren’t at all unprecedented as compared to the rest of the group.

    My point is not that the red line is correct but rather as we too often see in climatology the correction is the signal.

  4. Steve Fitzpatrick said

    Jeff –

    I don’t see that the function a + b * e^-(c*age)is a reasonable fit at all for the oldest trees.

    The increase in variance is understandable, but the trend in the oldest trees is clearly upward after 300 years age. Could this upward trend just be due to something simple like the trees that live the longest just happen to be in a favorable place (solar exposure, protection from wind, etc.) and so just grow faster, are healthier, and in better shape to survive periods of colder than normal summer temperatures? In any case, would not a correction function that tracks the trend through the entire range of ages (it looks like a parabolic function or a pair of exponentials would do it)not make more sense?

  5. Steve Fitzpatrick said

    I see that treyg said basically the same thing as me, he posted while I wrote.

  6. Jeff Id said

    #4 I’m no dendro but if I had more experience fitting long lived trees perhaps the typical pattern is exponential. It seems possible and even likely from a competition aspect that the typical pattern might be a U shape so at this point IMO you and #2 could be perfectly right.

    The only way this creates a hockey stick is if the correction factor runs substantially closer to zero than the signal and that IS entirely arbitrary. Yamal in this form is dead.

    Did you get a chance to translate some of the russian’s new paper? I ran most of it through google translator. They have a higher core count and older trees and came to the same result. If we don’t get that data, the paleo’s will simply claim verification of this obviously flawed result and move on.

    Right now, my opinion is that the mean of the trees is by far the best representation of the signal from Yamal.

    And trees still make lousy thermometers.

  7. Kenneth Fritsch said

    Thanks, Jeff ID, you have provided some more food for thought and perhaps more importantly a better understanding of the methods used. I think you are on the right track in the direction of your analysis.

    As to the attention paid to RC and Gavin (and their amen choirs) here and at CA, it really smacks too much of wanting an authority figure to pay you some attention and some how showing some approval. Sorry if that sounds too harsh – and maybe it was just a bad day babysitting grandkids, but I needed to vent.

  8. Steve Fitzpatrick said

    I’m no dendo either, but I sure have seen vast differences in growth rate for the same species of tree in the same area (but I don’t want to get into nature vs nurture for trees!).

    Do I have it right that if a “U” shaped curve is the proper correction function, but you use the RCS exponential function, then this would tend to give a hockey stick? Or do I not understand the calculation of temperature correctly?

  9. Jeff Id said

    #8 I agree with the comments about the U shape.

    #7, Don’t worry about it, you’ve made your opinion clear several times. If people can’t handle criticism, I would recommend not blogging. Of those that do blog or comment on blogs, I can’t think of a single one who didn’t tend to form their own opinions.

    WRT RC, my opinion of Gavin is that he is one of the most pompous people I’ve ever run across. He holds himself out to be an expert on almost everything climate and completely non-political. Guys like you have been doing this for a while and know the reality.

    Currently, I’m even suspicious of the real funding behind real climate – it seems Romm like. I’ve noticed a lack of mention at both RC and Tamino starting about two months ago and I suspect there is some snipping going on. After all, their commenters love to target me. So I would say, there is and was no need for recognition of myself but rather the purpose is to point out the charade that real climate is.

    Honestly, when I started blogging I didn’t know who these guys were and didn’t really care other than they seemed to be experts. Now that they are transparent to me, it seems important to point out that not only do they hold strong political biases, they are not behaving as scientists.

  10. Kenneth Fritsch said

    What has been ignored in most of this discussion goes back to Craig Loehle link to the Ecology paper that noted the changing growth patterns in larches and apparently concluding that older larch trees are required to proxy the climate. No one has been sufficiently knowledgable to comment on this paper or its conclusions.

    The results from that paper do, however, appear to agree with Steve M’s and Tom P’s sensitivity tests that in my judgment show that we should be looking at the response of trees that are older than a given age. Also we have had no discussion on how the Ecology paper findings fit the RCS and corridor adjustment methods.

  11. Kenneth Fritsch said

    We have also been preoccupied with the modern period part of the series when I see changes in all parts of the series when older trees are used exclusively. Overall I thik the analysis of Yamal needs to expand its horizons. Otherwise we get trapped in the dendro box and playing by their rules.

  12. Kenneth Fritsch said

    Jeff ID I disagree that Gavin is that exceptional when it comes to a scientist’s demeanor and personality. In my former life I dealt with many scientists and was required to take “different” approaches for different individuals to obtain the most value from their works and comments. Engineers tended to need less coddling and filtering. Production management where I worked tended to have more colorful and explosive personalities.

    Scientists could be just as emotional and biased about a topic as their non scientist brethern. The image and expectation of the stoic and always rational scientist is a myth.

  13. RomanM said

    Jeff:

    Not terribly unexpected considering the fact that the beginnings and ends of fossilized trees overlap prior to 1800 allowing for an averaging of defects in the RCS method. However, the endpoints are strongly affected by different fits simply due to the inhomogeneity of tree age in the record.

    Here is a bit more information in that direction.

    The plot shows the average age of all the proxies which contribute to a given year along with the number of proxies being used.

  14. Jeff Id said

    Roman,

    I added it to the post. Thanks.

  15. RomanM said

    #14 Jeff

    Looks like a hockey stick, doesn’t it?😉

  16. Steve Fitzpatrick said

    “Looks like a hockey stick, doesn’t it?” What the heck?!? Sure does.

    I can’t see any reasonable sampling rational that is consistent with that kind of age distribution. Living trees of an average age similar to the historical average ought to be the most plentiful in the region, so why the big increase in average age for the last 200 years? From what publication does this data come from, Briffa 2000?

  17. Steve Fitzpatrick said

    Kenneth Fritsch,

    “Scientists could be just as emotional and biased about a topic as their non scientist brethern. The image and expectation of the stoic and always rational scientist is a myth.”

    Agreed. But I think there are some differences between scientists in different fields. More established fields (with a long history and a foundation of firmly “settled science”) are generally not receptive to wild-eyed speculation and are populated by people primarily focused on the science itself. Newer fields like climate science, and especially climate modeling, don’t have a long history and solid foundation, and so the scientists can be and are more free-wheeling (and often very wrong!). In the specific case of climate science, the field is both new and appears to attract mostly people who already have very strong “green/environmental protection” inclination and a liberal/green political agenda. It is a toxic combination that leads to a lot of very bad science.

  18. RomanM said

    #16 Steve Fitzpatrick

    This data is the Briffa data which has been released after ten years of captivity (see the multiple threads at ClimateAudit.org). It has been used in multiple climate reconstructions in that time period.

    Having dealt over a fairly long time with scientists planning biological anfd forestry related research projects, the selection of these long-lived trees for coring does not surprise me.

    From a practical (but not statistical aspect) the choice to select the larger ones would simply be based on a desire to overlap as much of the time period as possible. After all, why would you do shorter ones and get less “information” for the same amount of work?

    Secondly, the spurt of more recent growth (late in the trees life) would also not surprise me since a justification could be made to choose healthy (dare I say, robust) trees as “carrying a good temperature signal”) would also be a natural choice.

    Statistically, one would realize that this leads to selection trees that are not representative of the broader population which is selected in a different sampling fashion. Add to this the post hoc choices of subselecting from the these trees based on “climate response” and you have a problem. Without possible adjustments to account for the selection process, the bias is problematical in the analysis. Add to that the fact that the analysis method exaggerates the differences and the whole result is under suspicion.

  19. Kenneth Fritsch said

    RomanM at #13 I would like to take bets on the response from a Team member or team cheerleader when presented with your graph.

    But in my mind the question becomes, knowning what we know about larch growth: Should we use older trees or younger ones or can we compensate for a mix of ages? I am betting on a restriction to a uniform older tree age.

    But it is easy for a layperson with no professional stake in the matter to speculate.

  20. Kenneth Fritsch said

    Steve Fitzpatrick, I am in essential agreement with your views. I do, however, think that regardless of the general leanings of climate scientists as a group that that bias can be applied when evaluating an individual scientist’s work. I would not want my comments writen off as those of a skeptic, denier or cheerleader for the fossil fuel industry.

  21. Steve Fitzpatrick said

    #18 RomanM

    “Secondly, the spurt of more recent growth (late in the trees life) would also not surprise me since a justification could be made to choose healthy (dare I say, robust) trees as “carrying a good temperature signal”) would also be a natural choice.”

    This seems contrary to the stated fundamentals of dendro-chlimatology: only stressed trees show climate effects. The rational seems to be that all trees of a single species grow at rates almost independent of environmental factors if they are not stressed (if they live the good life, they are always fat and happy), so those closest to the limit of their temperature range make the best “thermometers”. Selecting fat and happy trees to detect temperature changes just doesn’t make sense, putting aside the statistical issue of a having a modern sample with an average age that is representative of the historical population. The more I learn about dendro-climatology the more closely Briffa 2000 resembles garbage.

  22. Steve Fitzpatrick said

    #20 Kenneth Fritsch

    Fair enough, nobodies scientific work should be discounted based only on their personal beliefs or politics. But it is only reasonable to take those factors into account when deciding what work warrants a more careful evaluation. If a fundamentalist Christian scientist claims to have found evidence of dinosaurs living 20,000 years ago, that might warrant a more critical evaluation than the same person claiming evidence of human habitation of the Americas 20,000 years ago, even though both claims might be scientifically suspect. Agendas do matter, and they do lead to expectation biases and confirmation biases in scientific work.

    That Jim Hansen spends his weekends protesting the mining of coal with greens (and sometimes getting arrested for it!) is by for me good enough reason to look very closely at his scientific work for evidence of bias.

  23. hswiseman said

    Hi Jeff,
    What happens if you run the RCS for a shorter period, say through 1800? Would the end point bias heighten the apparent temperature spike around that time versus the same data run through 2000? Is this a viable method of measuring/demonstrating the impact of endpoint weighting inherent in the technique?

    Does the use of RCS preclude the calculation of error bars because RCS takes the place of other forms of regression analysis?

  24. Jeff Id said

    #19, I think the solution is actually pretty easy. Take a lot of samples and average. Chop off the earliest 100 years of the result and be aware that the shape of the large trees will give a very slight upward bias to the recent end.

  25. Jeff Id said

    #23 I did something like that here:

    https://noconsensus.wordpress.com/2009/10/05/yamal-the-dirty-dozen/

    I have a natural aversion to error bars in proxy studies as they only explain the variance of the data and not the consideration that we’re looking at a whole pile of different signals lumped together. You might know more than me on this but I can’t think of any reason why confidence intervals couldn’t be calculated.

  26. RomanM said

    #21 SF

    This seems contrary to the stated fundamentals of dendro-chlimatology: only stressed trees show climate effects. The rational seems to be that all trees of a single species grow at rates almost independent of environmental factors if they are not stressed (if they live the good life, they are always fat and happy), so those closest to the limit of their temperature range make the best “thermometers”.

    It seems to me that “stressed” is a relative thing here. Stress can be caused by many things: local conditions due to soil, moisture, crowding, etc. I’m guessing that dendros would avoid such trees. What you are saying is that trees are chosen because of a specific climatic environment. It is within this limited environment that I was indicating that a choice might be made for a “relatively” fat and “relatively” happy tree to guarantee a good set of rings for their sample.

    If I was consulting, I would ask such questions to determine what characteristics a sample might have. I am happy to be corrected on this if someone has specific inside information.

  27. Jeff Id said

    Here’s a cool reference from JNorv which was co-authored by Briffa that discusses some of the problems.

    http://www.cru.uea.ac.uk/cru/people/briffa/Briffa_HB_2008.pdf

    Roman, You may have already read it but it discusses some of the problems with tree sensitivity and RCS corrections.

  28. Jeff Id said

    Here’s a quote that’s interesting:

    When the index series are realigned by calendar year, each series systematically underestimates the
    magnitude of the ideal forcing in its early section and overestimates the signal later, a potential
    medium-frequency bias. In the average chronology (Fig. 5.2f), the original overall signal trend is
    captured by the differences in the means of the index series. In our simplified example, the bias in the trends of individual index series cancel to some extent by virtue of the compensating biases in overlaps between early sections of some index series and late sections of others. In situations where there is a good overlap in many series, this potential bias could be averaged out. However, this cannot happen at the start and end of the chronology. In the case of a long-term declining signal, the Dendroclimatology chronology will, respectively, under- and overestimate the ideal chronology at the beginning and end. With a long-term positive forcing trend, the signs of the biases will be reversed.

    If they’re so aware of the problem you would think they would be more careful about publishing ridiculous results like Yamal.

  29. Steve Fitzpatrick said

    #26 RomanM

    “If I was consulting, I would ask such questions to determine what characteristics a sample might have. I am happy to be corrected on this if someone has specific inside information.”

    Me too. I would specifically ask why younger trees would not show temperature effects as clearly in the last 100 years as they are expected to 1000 years ago. I would also ask why a selected population of 12 tree cores would be chosen instead of a larger population of all different ages.

  30. Steve Fitzpatrick said

    The cynic in me says that they (the Russians and/or Briffa) tried a representative age selection, but it did not show the temperature signal “clearly”, so they narrowed the population to old trees, and got what they expected.

  31. Layman Lurker said

    Sorry for the OT comment. I attended a conference several years ago and one of the speakers was Peter Leavitt. He was co-author of a paper which studied lake sediments from the Canadian Praries and the northern midwest U.S.. The project analysed the diatom fossil record contained in the sediments. Lakes in this region are prone to salinity during droughts, and diatom species indentification enables inferences about the level of drought and persistence over time. Drought patterns were found to persist on decadal and centennial time scales. Evidence of climate shifts were discussed at the onset of the MWP and the LIA.

    I know this is not quite the polar region, but it is another climate factor which would obviously have an impact on tree rings.

    http://www.pnas.org/content/100/5/2483.full.pdf

  32. RomanM said

    #27:
    Thanks jeff. I had it, but hadn’t read it. Scanning what was there, I found lots of generalities with little meat or theoretical content, but there some useful things to think about.

    #29:
    Consultants usually learn that there are lots of questions to be asked.😉

  33. Kenneth Fritsch said

    Steve F. I think the link that Jeff ID gives is a good illustration of my point. The Briffa provisios on RCS adjustments contradicts the conclusions taken from his paper on Yamal. I think much can be learned from what the scientist says and when that scientist becomes an advocate – not so much.

  34. hswiseman said

    Jeff,

    My knowledge of this stuff is pretty dismal, but I thought you need to perform regression analysis to determine CI. If RCS represents a substitute form of regression, that would make the CI calculations a regression based on data that had already been processed to drive out the outliers. I am only guessing that this is a no-no. If you could calculate a CI, I would suspect that the error bars would be so broad over the 12 tree portion of the sample that no meaningful conclusion could be drawn from the 12 tree trend.

    My other question is whether RCS imposes endpoint bias at any selected endpoint, or only when you have massive shifts in sample size as in the Yamal series. I see what occurs in the Dirty Dozen when you sequester the 12 from the series entirety. To isolate the impact of endpoint selection on RCS calculated trend, one would have to rerun the RCS only to alternative chosen endpoints and compare the trend at alternative endpoints to the trend at the same point in time when a longer time series has been used. My hypothesis is that RCS selection criteria (systematic under-weighting of older trees having slower childhood growth rates) will generate amplified endpoint trends at all intervals, but that the risk of amplified endpoint trends is much higher when the sample size at the endpoint is much smaller than the average sample size over the entire interval.

    See,

    Low frequency bias in proxy paleoclimate drought
    reconstructions using regional curve standardization
    K.J. Anchukaitis

    “The Malpais RCS chronology reproduces the low frequency variability and the underlying trend quite well (Figure 3a,c), diverging from the target trending pseudoclimate time series only when sample size is low and entirely dominated by trees with rapidly changing growth rates.” (footnotes omitted)

    http://www.ldeo.columbia.edu/~kja/rcs/pubs/rcs2009pp.pdf

  35. curious said

    bender flagged this up at CA – I think it is worth a read:

    “Recent climate warming forces contrasting growth
    responses of white spruce at treeline in Alaska through
    temperature thresholds”
    MARTIN WILMKING*, GLENN P. JUDAY*, VALERIE A . BARBER* and HAROLD S . J . ZALD
    Global Change Biology (2004) 10, 1–13 doi: 10.1111/j.1365-2486.2004.00826.x

    http://biogeo.botanik.uni-greifswald.de/fileadmin/user_upload/_temp_/Martin/Wilmking%20GCB%202004.pdf

  36. Kenneth Fritsch said

    The Jeff ID excerpt states that with an increasing trend the RCS adjustments will underestimate a finishing HS and not over estimate it – lets be perfectly clear on this matter.

  37. Jeff Id said

    #36 It’s written in a clever way but actually if you read the last bit carefully.

    will, respectively, under- and overestimate the ideal chronology at the beginning and end. With a long-term positive forcing trend, the signs of the biases will be reversed.

  38. kuhnkat said

    Curious,

    son of a gun. Sounds like they discovered that biologicals have a RANGE of physical parameters built into the genome!! What will they discover next!! ;>)

  39. Kenneth Fritsch said

    #37

    I read it several times and what I get out of it is that if we had a tree series in shape of a Kaufman like Arctic reconstruction or the classical Mann reconstruction with the shaft of the HS declining to the right, the blade would be biased upward, i.e. yielding the classic Mann HS – agreed.

  40. Jeff Id said

    Kenneth,

    In the center of the series, the result is basically unaffected due to cancellation of the bias. Where the problems occur is at both ends of the series. If we consider that

    1 – Older trees are going to tend to occur at the most recent end of the series due to ease of location
    2 – The recent years in any reconstruction are biased towards existing in the flat handle of the exponential function
    3 – The flat handle will naturally have less data as it is typcially from the oldest trees and will have the least effect or need of correction.

    My view is that the whole series of a properly fit RCS should match the mean of the data if done properly. Data at the flat handle portion which rises above the flat handle will receive substantial amplification by RCS. Which only affects recent years as it is canceled by overlapping data in historic years.

    Consider what that means if trees don’t grow in an exponential shape but rather a slight U shape. We would always get a hockey stick from RCS where the mean would give a near-proper answer. — Always.

  41. Layman Lurker said

    Jeff, not sure if I have been keeping up with everything so apologies in advance if this is redundant.

    Since the RCS curve is supposed to represent the climatic response of a given tree population – minus the non-climatic noise, the implicit assumption in Yamal is that the CRU 12 is representative of the same population, and been subjected to the same signal. Have you considered comparing the RCS curve of the CRU 12 vs. the rest of the population (and perhaps the sensitivity of CRU 12 RCS to outlier removal)?

    Also, what is the correlation of the mean vs. RCS corrected data to the JJA (or whatever months are appropriate) insturmental temps that Roman showed at CA?

  42. Layman Lurker said

    #41

    WRT to my previous point, here is a reference which supports the notion of RCS subsample comparison, from: Esper, Cook, Krusic, Peters, and Schweingruber, “Tests of the RCS Method for Preserving Low Frequency Variability in Long Tree Ring Chronologies; Tree Ring Research; Vol. 59(2), 2003

    RCS is clearly sensitive to the effects of different subsample populations entering into the calculation of single RC. Including samples from different “biological growth” populations in one RCS run could bias the resulting chronologies (e.g. TRW in Figure 8C) thus affecting interpretations of climate made from resulting chronologies. However, opportunities to test the data for existence of different populations are limited. This dilemma originates from the condition that the RCS requires only one FC for all the series then calculates anomalies form this one function for each single series. This approach works like a black box, making latent defects during the standardization process difficult to detect. Such defects can be studied and corrected much more easily when each single series is standardized individually. We recommend separating the data into possible subsamples then analyzing (i) the raw chronologies, (ii) the mean curves after age realignment, and (iii) the relationship of the mean versus the age of individual series. The classification of the population subsamples might follow the meta-information of a collection, and should certainly consider such differences as dead vs. living trees, site ecology and species composition.

  43. Layman Lurker said

    Correction to above (my transcription error): “RCS requires only one FC” should read “RCS requires only one RC”.

  44. Layman Lurker said

    BTW, the above quote was on pg 96.

    There is another interesting quote from this paper (pg. 82) which I interpreted to mean that RCS method cannot rule out persistent, non-climatic variation from the climate signal (ie the RC) in the reconstruction – perhaps as in the blade of the hockey stick?

    Every tree ring series, whether it is based on TRW or MXD, contains some non-climatic variations. This “noise” can be caused by site related effects (e.g. competition and disturbance) or biological effects (e.g. aging). As a consequence of integrating biological and ecological impacts within the measured parameter, TRW or MXD, variability in the resulting chronology will often depart from variability associated with climate. This departure from climate is often systematic and persistent over time.

  45. Stevo said

    I realise that you’re trying to replicate RCS and are therefore probably not interested in discussion of alternatives, but why on Earth are they dividing by the mean? Unless perhaps the idea is that the mean is proportional to the standard deviation, but easier to estimate?

    Since the distribution is so obviously skewed (which is causing the estimate of the mean to get biased upwards with the low sample sizes after 300 years) I would have instead converted it to percentiles. Plot lines exceeding 10%, 20%, etc. of the data at each age, fit your curves (preferably weighted according to the evidence available), then interpolate to get an estimate of where each ring width lies on the distribution. Or if the data fits, assume a lognormal distribution at each age.

    Your second figure I found the most interesting. It would be even better if you could see how the density varied across the solid black region. If you took this data and applied the right correction, it ought to come out with a constant distribution over time. (Assuming that factors affecting growth other than age have an independent distribution.) Does the RCS correction do that? And what distribution do you get?

    It’s also said several places elsewhere that the change in growth is related to tree diameter, so it might be interesting to plot ring width against diameter rather than age, to see if that gives a narrower spread. The other thing I’ve seen mentioned is that fast growing trees tend to be always fast growing, and slow ones always slow, and the fast ones slow down faster. (Probably competition related, I guess.) So doing something like estimating the ‘speed of the tree’ with a curve fit and then finding percentiles separately for each tree speed might be interesting to investigate.

    But the more I learn about tree rings, the more I am reminded of von Neumann saying “With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.” It’s hard to take seriously.

  46. Layman Lurker said

    #41

    Jeff, apologies that I did I did not go back and refresh myself on what you had done in this post: https://noconsensus.wordpress.com/2009/10/05/yamal-the-dirty-dozen/.

  47. Jeff Id said

    #46, Sorry I didn’t reply sooner. I didn’t pay much attention to blogs this weekend and instead spent my time pouring over code by Ryan and Nic. Really too much time IMO for not getting paid. When blogging turns into work rather than entertainment it’s too much.

    This issue is so stick a fork in it dead that I get pissed off when I read blogs like deep climate. He’s almost earned himself a vent. They should be praising the crowd for not screaming Fraud across the headlines – it’s tempting. This SOB is as bad as Mann 08 but it’s worse because it’s even simpler. The more it’s sat in the back of my mind the more ticked I get. I’m not surprised that Briffa refused to disclose which data was used.

    Deep is over their claiming libel against McIntyre (something RC was careful to avoid). Oddly, if he’s claiming libel and can’t provide a quote isn’t he libelous in doing so? He also claimed it against me previously when I responded (kindly) to his idiotic polynomial projection graph in his post on end point filtering so it’s a personality disorder or something. Anyway, I think there’s probably one more Yamal post which could happen.

  48. Mark T said

    Oddly, if he’s claiming libel and can’t provide a quote isn’t he libelous in doing so?

    Technically, yes, realistically, no. At least, in the US, pursuing libel/slander claims is very difficult (obviously libel would be simpler because it is written down) and very easy to defend against, so to point out that his claims are libelous may be true at face value, but there’s not really anything you can do other than point out his (her?) hypocrisy.

    Mark

  49. Jeff Id said

    And thus we blog…😉

  50. Layman Lurker said

    Your post on the RCS subsample is the “stick a fork in it” moment AFAIC. According to Esper et al, if the subsample does not produce the same RC curve then this is indicative of two populations having different responses to the same climate signal.

  51. Kenneth Fritsch said

    Jeff ID, I am posting this information here as I have had trouble getting it posted at CA. I wanted to post there since that is where the discussion of Tom P’s sensitivity test was initiated. Perhaps here at The Air Vent is a better place to analyze in another post the before and after Yamal series going through thje RCS algorithm.
    .

    I have finally gotten around to doing a sensitivity test for the Yamal tree ring ages in the form that I preferred. I did this by removing tree ring ages, not individual trees by age as Tom P did, in increments of tree ring ages over 75, 100, 125 and 150 years. In order to do make this work with the Steve M’s RCS chronology algorithm, I had to reset the ages in each group by off setting the tree ring ages by 74, 99, 124 and 149 years respectively. The algorithm evidently does not work without an age 1 starting point.
    .

    I have posted links to the graphs of the modified RCS series below with each including the archived Briffa chronology for reference. Additionally I have posted links to the tree ring counts per year over the series for all ages, over 74, over 99, over 124, and over 149.
    .

    What one sees as obvious is that as one goes to older tree ring ages the HS appearance of the series with all ages begins to erode and finally vanishes all together. At the same time one sees that coverage for some years becomes rather thin as the younger tree rings are excluded from the series. However, with the trending, I would judge that this sensitivity test is indicating what Craig Loehle’s excerpt at: Post #202 above
    http://www.climateaudit.org/?p=7241#comments in that larches are better indicators of climate as they become older. I do not consider the responses in the series with the older tree rings ages noise as Tom P has indicated.
    .

    Since I have not studied the Steve M RSC chronology method, I am not certain whether the modification I used is legitimate. I think it is, since the higher frequency responses appear to be similar to the Briffa RCS chronology. I borrowed much code from Steve M and list below the code I used for the graphs. I also looked at the tree ring widths time series before and after putting it through the Steve M RCS chronology and found (not shown in this post) that the RCS methods greatly change the shape of the tree ring responses. That might be fodder for later discussion of that method and perhaps alternative methods.
    .

    What ever I accomplished by the analysis reported here and whether it is legitimate, I found by going through the process I learned a great deal more about the RCS process and R coding than by simply reading the post introductions and responses.
    .









    Code for tree ring age graphs:
    .

    #Load functions and utilities
    source(“http://www.climateaudit.org/scripts/utilities.txt”)
    source(“http://www.climateaudit.org/scripts/tree/utilities.treering.txt”)
    f=function(x) filter.combine.pad(x,truncated.gauss.weights(21))[,2]
    #Yamal measurement data
    loc=”http://www.cru.uea.ac.uk/cru/people/melvin/PhilTrans2008/YamalADring.raw”
    download.file(loc,”temp.dat”)
    tree=make.rwl_new(“temp.dat”)
    tree$id=factor(tree$id) #252
    tree=agef(tree)
    #save(tree,file=”d:/climate/data/yamal/yamal_cru.rwl.tab”)
    #load(“d:/climate/data/yamal/yamal_cru.rwl.tab”) #tree
    range(tree$year) #202 1996
    yamal=tree
    dim(yamal) # [1] 40892 4
    yamal$rw=yamal$rw/10 # Sep 28

    yamal125=subset(yamal,age>124,select=c(id,year,age,rw))
    yamal125[,3]=yamal125[,3]-124
    chron.yamal=RCS.chronology(yamal125,method=”nls”)
    loc=”http://www.cru.uea.ac.uk/cru/people/melvin/PhilTrans2008/Column.prn”
    briffa=read.table(loc,skip=1,fill=TRUE)
    name0=scan(loc,n=8,what=””)
    name0=outer(name0,c(“”,”count”),function(x,y) paste(x,y,sep=”.”) )
    n=nchar(name0[,1])
    name0[,1]=substr(name0[,1],1,n-1)
    names(briffa)=c(“year”, c(t(name0) ) )
    briffa[briffa== -9999]=NA
    briffa=briffa[,c(“year”,”Yamal.RCS”,”Yamal.RCS.count”)]
    briffa=briffa[!is.na(briffa[,2]),]
    briffa=window(ts(briffa[,2:3],start=briffa[1,1]),start=-202)
    yamal.crn=briffa[,1]/1000
    par(mar=c(3,3,2,1))
    delta=mean(yamal.crn)-mean(chron.yamal$series);delta # -0.04820995
    ts.plot(f(yamal.crn))
    lines(f(chron.yamal$series)+delta,col=2)
    legend(“topleft”,fill=1:2,legend=c(“Archived Briffa”,”Emulated for Tree Ring Age 124 Years”)
    title(“Yamal RCS Chronology TR Age>124 Years vs Briffa (CRU)”)

    Code for tree ring counts by year graphs:
    .

    yamal125=subset(yamal,age>124,select=c(id,year,age,rw))
    Y125_year=reshape(yamal125, direction=”wide”,timevar =”year”,drop=c(“rw”,”age”),idvar=”id”,v.names=”year”)
    dim(Y125_year)
    x=ncol(Y125_year)
    Yseries=colMeans(Y125_year[,2:x],na.rm=TRUE)
    YS=as.vector (Yseries, mode=”integer”)
    Ysums=colSums(Y125_year[,2:x],na.rm=TRUE)
    YSums=as.vector (Ysums, mode=”integer”)
    Ycount=YSums/YS
    Yct=rbind(YS,Ycount)
    YctT=t(Yct)
    YctO=order(YctT[,1])
    Series_Count125=YctT[YctO,]
    #Year =0 creates division by zero so add in here
    Series_Count125[79,2]=5
    plot(Series_Count125[,1],Series_Count125[,2],type=”h”,main=”TR Counts by Year for Tree Ages Older Than 124 Years”, xlab=”Years”,ylab=”Counts”)

  52. Jeff Id said

    #51, There seems to be a problem in the chronology dating. As the series get older the red graph is offset to the left. Something is wrong.

  53. Kenneth Fritsch said

    Jeff ID, as I remove more tree rings by excluding more of the younger rings there are fewer rings to cover the earliest part of the time series. Remember that a 150 year limit excludes all trees and rings younger than 150 years of age.

    This also implies that the early part of the series is dominated by younger trees, but let me check this out and get back to you. In the R code I do use the ncol function to get the number of years for each time series for the TR counts and it appears to agree with the TR width series. If this is incorrect, I think it will only affect the earliest part of the series.

  54. Jeff Id said

    #53 I actually wrote a whole guest post on this before I noticed the problem, had it worked people wouldn’t have been looking at a comment about Fox news. hehehe. The blade of the red stick is shifted by a few years from the black, some other features seem to follow suit. I’m not terribly motivated right now because I looked at AMSR-E sea ice data all night but perhaps tomorrow I’ll take a look at what’s going on.

  55. Kenneth Fritsch said

    Jeff ID, here is some R code that I ran that will help you understand the apparent year shift when more and more younger tree rings are excluded from the individual series. The minimum years agree with the graphs.

    min(yamal[,2])
    [1] -202

    min(subset(yamal,age>75,select=c(id,year,age,rw))[,2])
    [1] -127

    min(subset(yamal,age>99,select=c(id,year,age,rw))[,2])
    [1] -103

    min(subset(yamal,age>124,select=c(id,year,age,rw))[,2])
    [1] -78

    min(subset(yamal,age>149,select=c(id,year,age,rw))[,2])
    [1] -53

  56. Kenneth Fritsch said

    Sorry Jeff ID, I just realized you were looking at the other end of the series (modern times). My R code for maximum does not agree with the time series. Thanks and let me look into this in more detail.

    max(yamal[,2])
    [1] 1996
    max(subset(yamal,age>75,select=c(id,year,age,rw))[,2])
    [1] 1996
    max(subset(yamal,age>99,select=c(id,year,age,rw))[,2])
    [1] 1996
    max(subset(yamal,age>124,select=c(id,year,age,rw))[,2])
    [1] 1996
    max(subset(yamal,age>149,select=c(id,year,age,rw))[,2])
    [1] 1996

  57. Kenneth Fritsch said

    Jeff ID, I traced the problem of the premature ending of the tree ring age series to looking at the R code: chron.yamal$series. For example, the RCS chronology for greater than 124 years for tree ring ages starts at the expected time but ends at 1977. If I print out the series it has data from 1977 through 1996. I can overcome this and obtain a plot through 1996, but in the meantime I want to understand the premature cutoff.

  58. Kenneth Fritsch said

    Jeff ID noticed that my series graphs were prematurely cut off for the latest years. In order to correct that I did the following:

    From Steve M’s RCSchronology function I changed limit for min to max as in:
    series124″,xlab=”Years”,ylab=”Tree Ring Width”,type=”l”)

    The series graphs are presented below and the observations/conclusions are the same as those from my previous post.





  59. John F. Pittman said

    JeffID, KFritsch, the RCS methos actually uses the fitted growth curve for calculating anomolies, and then uses the historical temperature to “tune” the anomolies for a historical reconstruction correct?

  60. Jeff Id said

    #59, I think you’ve got it. The ‘tuning’ is a calibration to temp.

  61. […] I’ve got to say again Briffa’s original Yamal is a disgusting piece of garbage work and the sooner paleo’s drop the P.O.S. the better. It’s got an unreasonable blade created from RCS with NO science or verification to prefer the ‘accidentally’ chosen exponential curve that is ENTIRELY RESPONSIBLE for the big evil bullcrap blade.  See one of my posts on this HERE. […]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

 
%d bloggers like this: