the Air Vent

Because the world needs another opinion

The Unstoppable Dirty Dozen

Posted by Jeff Id on November 8, 2009

Science is good fun. Lately we’ve been working with an unnamed paleoclimatologist who goes by the handle of Delayed.Oscillator. He’s been good enough to answer some questions on dendroclimatology temperature reconstructions. Recently I’ve made comments about the Yamal series data and Briffa’s corrections about not allowing for trees to grow faster as they age. Briffa uses an exponential decay for his corrections to Yamal tree growth which looks like this:

yamalex

Figure 1

The data is divided by the red line in top pane in the above plots. It has been my contention that since the red line drops well below the rest of the data it amplifies the older data creating an artificial uptick at the end of the reconstruciton. DO has made the following statement at his blog:

Let me be as clear as I can be, there is no sign that I can detect that it is old trees that increase their growth at Yamal (even if identified, this phenomenon would require some hypothesis as to the cause), At Yamal, a portion of the old trees are those that were growing together during a period of climate warming. If you examine the raw ring width, there are a few fossil series that have rapid increases toward the end. If Jeff’s hypothesis were correct, we’d expect these to be the oldest, right? In fact, the seven subfossil samples I identified as having rapidly increasing growth in their later years, six had a wide range of ages from 90 to 180 years (this comes with the caveat that we don’t know the exact pith age).

Now DO beat me to an analysis suggested by Steve Fitzpatrick to separate the recent old trees from the historic old trees and verify whether the U shape in pane 1 above (the average of all old trees) is a result of recent climate or is a standard shape in older trees. There are plenty of explanations as to why trees can grow faster in older age that we can be certain DO is aware of so the parenthetic portion is a bit grumpy of him. In DO’s conclusion his plot looked like this:

non20c[1]

Figure 2

This plot represents all trees with a last measured ring width dated before 1950. My own contention was trees greater than 200 years could show the increased growth though so it wasn’t enough to show what happens. We need to see the old trees separate from the record as Steve Mosher pointed out. So here’s a plot of the pre-1950 trees similar to that shown by DO.

There are only 19 trees to work with but here they are with an Esperesque spline fit.

Briffa trees spline  pre1950

Figure 3

You can see the curve in the top pane has almost the exact same shape as DO. Not bad really. In the bottom pane there is no apparent upslope in pre 1900 Yamal area trees with ages greater than 250 years. Now when this is compared to 250 year old trees with last RW after 1950 the plot looks like this.

Briffa trees spline post1950

Figure 4

The trees (first pane) show a clear upslope in later years. On the surface DO seems to be correct on this point. The trees ending before 1950 look like this.

Briffa trees older than 250 pre 1950

Figure 5

Figure 5 just left more questions for me. Many of the trees seem to reach a minimum at 200 years of age and peak upward trickling back down before ending their lives. Is it possible that the difference between the post 1950 Yamal data is that they haven’t slowed down in their last 20 years before dying yet? Would that make the downslope at the end of pane 1 Figure 3. I don’t think it’s unreasonable because you can see the mid life minimums in L13371, PO9281 L21081 L11631, L00861,L01181, L01041, L13181 even though they have different climatological conditions. They started growing faster after about 200 years and then trickled off at the end. I’m still not certain that most trees don’t show an uptick as they age keep in mind that this data is the extent of my experience.

The last graph P09281(last graph above) starts increasing after 200 years and then drops off and dies. It getting warmer from 1100 to 1150 according to the last graph but not according to L09301 (third one down on the right) which was dying at that timeframe or L12641 which in 1100 AD was before its pre-200 year mimimum and still dropping (these graphs are uncorrected though so perhaps reasonable growth standardization corrections can turn it into warming).

Next is the same kind of plot for some of the Post 1950 trees. These trees haven’t shown their end of life dropoff cause most if not all were alive at the time of sampling.

Briffa trees older than 250 post 1950

Figure 6

There are a lot of trees which show increases in growth post 1900 but there are also quite a few which show less growth post 1900. Really lousy thermometers these. From Figure 4 pane 1 we would expect a steadily warming climate for the last 200 years (that’s the average) and that’s pretty well what we see in pane 3 of Figure 4.

So then I decided to calculate least squares slope fit’s to the ring widths since 1920 instead of by tree age. Since the slopes are from trees over 250 years that end post 1950, the EXP correction from Briffa or the spline corrections would have a negligible effect on slope. In a huge surprise, there are a lot of negative slopes in these trees during the greatest warming in the Yamal area. Remember these are 250 year old trees from 1920-1996 so it’s 174 years and older in pane 1 of Figure 4. – Look at pane 1 of the figure and you’ll see why negative slopes after 174 years are a surprise.

[1] 0.09525731 -1.67765546 0.08194925 -1.57939710 1.56477900 -1.08070478
[7] 0.10164856 -1.66012718 0.01617010 0.79523364 0.29315124 -4.15916120
[13] -0.15019269 -4.84568356 -0.31551278 0.02163996 -1.24181240 -1.88490060
[19] 0.10123042 0.38330849 -0.05190331 2.60531409

There are 11 negative slopes of 22 series! That was unexpected! What’s more the average slope is -.572 (negative!) with a standard deviation of 1.66 – Not exactly precision agreement. Now 1920 was picked at random with consideration of when we started producing lots of CO2 around then. Be careful and don’t overconclude form these slopes – with this noise level different years will give different results. IMO the main conclusion from this is that the slope is not strongly positive – no indication of warming!!!! and not clearly defined! Since there is no evidence that trees are thermometers, rather than conclude anything about warming lets say tree growth is apparently not unprecedentedly large in recent years. Below is a histogram of the slopes. How’s that for a bell curve!

Briffa trees older than 250 post 1950 hist

Figure 7

So the next thing was to look at uncorrected tree ring data for trees at least 250 Years old which have a last ring after 1950. These are probably 100% living trees but we don’t know.

Briffa trees MEAN post 1950

Figure 8 - Yamal Area Data Simple Mean

This does not look like Figure 1. The EXP corrected version (Briffa’s style) results in the plot in Figure 9.

Briffa trees EXP post 1950

Figure 9

The exponential hockystickization correction of Figure 9 is just enough to push the blade into unprecedented territory but not by much. This doesn’t look like Figure 1 either. The spline corrected version of the data is Figure 4. Figure 4 looks a lot like the simple mean of Figure 8, this represents reasonable standardization of old trees across the time series. This is my #1 criticism of Yamal that the original version didn’t look like the mean, indicating the blade was created by the standardization.

Ok, so let’s get into some new stuff. The total dataset in the Yamal area includes the following groups:

Yamal, Live, Jah, Por, Yad, Khad, Russ035

The russ035 is from SteveM’s sensitivity test. You learn so much more playing with the data than you can reading about it. I discovered that in several cases the ‘new’ briffa sensitivity test has the same core ID’s as the old briffa data.

The total unique series names available is 302, Yamal itself had 252, Live -17, jah – 25, por-12, yad – 10, khad – 18 (Russ035 – 35 – SteveM’s not used by Briffa). In Briffa’s sensitivity test, there were 82 series which should have been unique to prove that Yamal is not sensitive to different datasets. However, the sum of the non-Yamal series = 17+25+12+10+18 = 82 series added to Yamal for sensitivity. However there are only 302 – 252 = 50 unique new series from the original Yamal. Thirty two series were used at least twice in the Briffa sensitivity test.

JAH141 – *
JAH162 – *
M021
M202
M331
POR011 – *
POR031 – *
POR051 – *
POR081 – *
POR111 -*
X02S
X13
YAD041 – *
YAD061 – *
YAD071 – *
YAD081 – *
YAD121 – *
878031
878081

Stars are the original dozen. You need to be following along for this but if the dirty dozen are re-included in the sensitivity test to determine the effect of the dirty dozen? Is that a fair test?

Probably the main criticism of Yamal by some dendro’s was the low core count in recent years. If we don’t double up the trees and properly sort the data, what is the core count for all the data in the region and how does that compare to the original Yamal? Core counts are shown per year in table 1. All series include SteveM’s schweingruber set.

Yamal Briffa Series All Series
1989 10 45 61
1990 10 45 61
1991 10 34 34
1992 10 22 22
1993 10 22 22
1994 10 22 22
1995 5 10 10
1996 4 9 9

What does the full RCS style hockeysticization reconstruction look like now with all trees included and the original Briffa style corrections.

Briffa trees EXP all

Figure 10

Core counts in 1996 have doubled now to an anemic 9 but the blade was cut from a Figure 1 peak amplitude of 2.75 (red line) down to 2.25. A good size jump downward from the new trees but it’s still an excellent quality hockey stick. If you cut off the last 6 years the plot doesn’t look the same but that’s not the point. RCS corrections to the tree ring widths rely completely on the homogeneous nature of the data. We are applying the same correction to all the data. Differences in homogeneity will result in unexpected results. Since RCS is an ad-hoc style correction, great care needs to be used to determine if the results make sense.

Briffa’s correction used for Yamal was an exponential decay. The exponential curve plot (Figure 10 pane 1) tapers off to a flat horizontal line becoming pretty horizontal after 100 years. To be clear, there isn’t much slope in the correction after that time. This means that for the average older trees in the recent 100 years, a correct standardization would reproduce results similar to the mean of the data. There just shouldn’t be a huge difference because there isn’t a huge slope correction. My contention is that there is a big difference created entirely by RCS reacting with the inhomogeneity and has resulted in coining the phrase hockeystickization.

I must mention that Roman M is the first to point out this effect on a CA thread. Roman does a lot of things first. Anyway, this is what the mean of the data looks like. In this form the oldest data is distorted but the recent data should be reasonably similar to Figure 10.

Briffa trees MEAN all

Figure 11

Now that looks a little different wouldn’t you say?!! This means that there is nothing unique or unusual about the 20th century tree rings. There is an upslope in the last few years of the plot represented by very few trees but the ring widths don’t exceed 1930.

Conclusions:

– Briffa’s sensitivity test used the original Yamal data and got the original result!!

– RCS standardization is reacting poorly to the typical data and creating an artificial hockey stick on the end of the data!! – Homogeneity must be questioned.

– There is nothing unique or unprecedented in recent tree ring widths in Yamal area.

– Average slopes of living trees in Yamal area after 1920 for older trees are actually negative. I picked 1920 for industrialization, I did not try other years or pick the best one on purpose. Results are going to be dependent on the choice of start year.

– The standard deviation of the slope variance is huge, selection of different years will result in huge differences. – The data is noisy!!

– The math is creating most of the variance (hockeystick) in Yamal’s new version in recent years and must be thoroughly (and honestly) examined to accept any sort of corrections.

Somehow the pro’s don’t agree!

——-

Now the next step will be to redo the whole sensitivity test by Briffa without the dirty dozen and see what we get! I’m tired though and these posts take a lot of work.

Does anyone want to bet what all the data looks like with the dirty dozen removed?

15 Responses to “The Unstoppable Dirty Dozen”

  1. Kondealer said

    Fantastic work, Jeff. It is amazing what you can produce with low replicates, high variance and an uncritical audience.

    Seriously this sort of work should “slay the stick” once and for all.

    Why not team up wuth Steve M. and get this published?

  2. Jeff Id said

    #1 Thanks, publishing is a pain though and I never wanted to be a climatologist. It’s interesting to see how they’ve concluded such strong things with so little information.

    I mean we have non-linearity, non-stationarity, multiple correlated signals, core variance. The list goes on and on.

  3. Layman Lurker said

    Jeff have you looked at the Devi et al 2008 paper yet which is being discussed at CA?

    http://www.wsl.ch/news/081119_GCB_Ural.pdf

    According to this paper, a large proportion of old, living larch trees in this area are multi stem trees which grew horizontally until early 1900’s. After this they began to grow vertically and experienced an exponential increase in growth rate – growing at a much higher rate than single stem larches of comparable age and year.

    The paper also shows the effect of increased precipitation (particularly on multi stem) in the 20th century.

  4. Geoff Sherrington said

    Maybe it is not logical to use tree ring width. Maybe better to use tree ring volume on a given section. Should be able to estimate trunk diameter from core, depending on how complete it is.

    The Devi paper just blows larch dendroclimatology in Siberan larches out of the Tundra.

  5. Kondealer said

    Layman, as soon as they start growing vertically, they get more light=more growth.
    A small increase in temperature/increased precipitation is probably all that is needed to get growth to change from horizontal to vertical (unless growth mode is under genetic control- as it is in Arctic Willow)- then it will take off.

  6. delayed.oscillator said

    Hi Jeff,

    You’ve clearly done a lot of work here (I wish I had your free time!). A few observations and conclusions of my own:

    [1] I’m glad we now agree that old trees do not show a ‘u shape’ uptick in growth at the end of their lives.

    [2] 1920 seems like an odd selection for a starting point for industrialization or the anthropocene — between 1800 to 1900 would be more typical. I’m not sure what your analysis of slopes since 1920 is intended to establish, but if you calculate the least squares slope for the living ‘dozen’ trees, you find that while 50% dp have negative slopes since 1920 (coincidentally a year of warmer JJA temperature in the CRUTEM3 gridcell corresponding to Yamal), since 1900 only 25% have negative slopes, 17% since 1850, and 0% since 1800. So, your trend analysis is not robust to your selection of starting years.

    [3] Your figure 8 and 11 (once again, as I’ve indicated a few times to you) mix age and climate related effects, as does conceptually your comment about the ‘homogeneous nature’ of the growth curves. The mean regional curve is calculated from lots of trees growing at lots of different times — this is the very definition of homogeneous when applied to RCS.

    [4] The math is not creating most of the variance, rather, you’re probably removing some portion of the climate signal when you subtract the simple mean from truncated raw data, which again I’ve shown previously in my posts.

  7. Jeff Id said

    #1, It seems that many of the trees show a U shape at one point but the end years tailed off. I’m happy to say you are right on this point. Is this a standard shape? Would that be enough not to use an exponential fit?

    2, The fact that the RCS Yamal has a huge upslope from 1920 on are what lead to the choice. It’s surprising that the actual data which is in a flatter portion of the EXP curve would have a negative slope.

    This seems like a good opportunity to mention that it’s unfortunate that scientists trust CRUTEM without verification. We don’t actually know if its comprised of thermometers or spaghetti noodles. While Giss temp corrections may be perfect the methods require a small leap of faith, whatever methods and data HadCRUT use should receive reasoned scrutiny. I wish more would demand release of the data and methods for this important metric.

  8. Steve Fitzpatrick said

    Hi Jeff,

    Did both Steve Mosher and I make the same suggestion about the standardization curve based on old trees that died prior to 1900?

    I exchanged a few comments with DO at his blog. I think the real weakness in tree ring temperature dating is how “divergent series” are excluded. Briffa 2000 and Briffa 2008 are both based on exclusion of data from geographically close sites when that data does not match up with (that is, diverges from) the sites that ended up being used in the reconstruction. Steve McIntyre showed that substitution of one geographically close Yamal data set (with more 20th century cores!) for the smaller “dirty dozen” data set makes the Briffa 2000 reconstruction results null.

    DO offered (IMO) no rational for the exclusion of divergent data. Since the cause(s) for divergence are not known, it is impossible to say that similar divergences did not happen in the past. Maybe Briffa’s Yamal data is “normal” for 1850 to 1990, but terribly “divergent” at one or more sites for the 900 AD to 1040 AD period, and the actual temperature for 900 to 1040 was much higher (or much lower) than the reconstruction suggests…. it is impossible to know. Since the whole point of a tree-ring re-construction is to glean believable climate data from before the instrument temperature record, lack of understanding of what causes “divergence” makes any climate reconstruction based on selected sites highly doubtful, and IMO, essentially useless. The only way to draw meaningful climate inferences from tree rings is to 1) randomly select trees of all ages from a wide range of geographically similar sites, 2) collect the cores from those trees, 3) do the analysis, 4) calculate realistic confidence limits, 5) see how the reconstruction correlates with the instrument temperature record, and 6) calculate a pre-instrument re-construction, including appropriate confidence limits.

    And NEVER reject data from sites you “don’t like” nor hide what data you have used.

    Neither Briffa nor other dendros are under any obligation to explain which sites/data sets were considered for a study (AKA “data snooped”), but not included in the final analysis, so there is no limit to the potential distortion in a reconstruction. I see the entire exercise as a sophisticated form of cherry picking, and so terribly subject to expectation bias. Obvious cherry picking of data would lead to loud laughter in most other scientific fields; that it is the accepted norm in dendro climate work forces me to conclude that the field has little or no technical validity.

    DO censors any posts at his blog that make these types of comments (which really pissed me off), but fortunately, the Air Vent is a more open forum. I suggest that DO put on his big-boy pants and stop censoring comments at his blog; good science doesn’t need the protection of censors.

    And DO: Assuming you read this, people care about tree ring studies because they are part-and-parcel of the justification for proposed draconian changes in future energy production and use. You should accept that many people who want to comment at your blog are going to be skeptical of the scientific validity of denro work. Your blog would be a lot more valuable if you would just defend the methods and results of dendro work, and admit that any exclusion of “divergent” data leads to the very real possibility that reconstructions can be influenced by personal/political agendas. You seem to justify censoring comments at your blog by saying that you would not be well received at a nanotechnology conference if you expressed extreme doubt about the validity of nanotechnology science. I do not know if that is true, but I do know that the situation is completely different: the results of nanotechnology research are not being used to justify many trillions of dollars of politically imposed costs and major changes in the world’s economy, while the results of Briffa 2000 and similar studies are. You compare apples with elephants.

  9. Jeff Id said

    DO, the following questions may be more difficult but they are not gotcha type and I don’t plan to argue them although there will probably be other questions. I’m curious about your opinion on their answers.

    A question I would like to understand is what is considered enough cores for a site? It seems like that is of primary concern in Yamal.

    Also, CO2 fertilization is a known effect. Since CO2 is assumed or stated to be the cause of temp rise in most climate papers these days, how can CO2 growth be separated from the temperature growth and do you know of any attempts to perform the separation?

  10. Jeff Id said

    #8 I apologize for the name mixup and have read all of your comments on this matter at DO and here.

    Regarding divergence exclusion, I completely agree with your point. Dendro’s eliminate data far too often which makes Briffa’s complaint about SteveM insinuating he may have used data that ‘wasn’t divergent’ all the more odd. The noise level in this data does not allow for elimination due to divergence over a hundred year timeframe.

    There are multiple non-stationary effects intermixed and the best you can do is to take an average of all growth corrected data. Whether there is correlation or not, has little meaning to me but that could be used to sell the curve as temperature. I’m not anywhere near convinced that it is though.

    Regarding censoring, I agree that it’s far to common in AGW blogs. I snipped one yesterday, but it was the first in probably 6 months. Snipping reasonable comments is a result of a thought process which has a high correlation to a certain political viewpoint. He doesn’t need to worry that much on a climate blog, if you saw Tom Fuller’s latest poll results (link on the right) 90% of skeptics at WUWT (long and incorrectly reviled as he mosh pit of delialism) have a bachelors degree or more.

    By email, I’ve suggested that DO could do a great service to the field by having an open moderation policy. He’s been honest in my limited experience so all of a sudden there’s an avenue to get a lot of questions answered. Whether we like the answers or not at least there could be some answers. If dendro develops good answers regarding the separation and linearization of temperature response and reasoned elimination of data chopping methods we’ll all be a lot happier.

  11. Steve Fitzpatrick said

    #10 Thanks for your reply.

    I agree that it would be very constructive if people like DO would offer reasoned explanations for the handling of dendro data. But based on what DO has said so far about excluding divergent data, there is no reasonable justification for any such exclusion.

    It would also be constructive if people in the dendro field would demand that data collection and handling be robust. Why is DO not screaming about the lack of sensitivity analysis for substituted data sets in Briffa 2000? Why does it take someone like Steve McIntyre to show how weak the Briffa 2000 study is, when people in the field should all be doing that? Why do people in the field not insist on simple things like disclosure of data snooping, detailed rational for excluding data sets, and complete archiving of raw data and meta-data for both included and excluded data sets?

    I agree that convolution of temperature effects with CO2 fertilization is a real problem for the post 1850 period. Short of multi-year, multi-site studies of free-air CO2 enrichment with trees growing near the tree line, I don’t see how this problem could ever be resolved. The large majority of CO2 growth studies with a wide range of plants show a strong CO2 effect; this makes CO2 driven increases in growth in the 20th century likely. A host of other factors (insects, disease, precipitation, etc.) are all convoluted with the temperature signal, and without access to a time machine, it is difficult to see how these other factors could ever be accounted for.

    I don’t have much hope that the dendro field can ever produce accurate pre-instrument temperature histories.

  12. Layman Lurker said

    DO, at CA commenter JS separated the “age effect” from the “year effect” (plotted, complete with confidence intervals) with a random effects model. The net result for the “age effect” was a curve with a very similar shape to Roman M’s loess fit. I liked this approach (including confidence intervals) because it highlights the fact that the blade of Yamal can be summed up in one word – uncertainty. You argue that Jeff’s analysis likely intermingles climate signal with the growth curve. I would argue the reverse, Briffa has not demonstrated that the negative exponential RCS is the correct method to standardize with this data. The random effects model (according to JS) rejects the hypothesis that there is not a statistically separate “age effect”.

    The Devi et al paper shows that multi stem larches vs. single stem larches in this area can constitute a bimodal growth population for old, living trees. Briffa offers no metadata analysis to verify that such heterogeneity does not contaminate the 20th century growth signal.

    The Devi et al paper clearly demonstrates the growth sensitivity of Siberian larch to precipatation, which has centennial scale variability. Considering the endpoint issues of RCS, Briffa has not demonstrated that the climate signal of the 20th century is not due to the considerable increase in precipitation.

    Lastly, you mention the need for a hypothesis test of a proposed physical cause of a separated growth function, does this argument not apply to those series of Yamal/Urals excluded in chronologies for so called “divergence” as well? I am not aware of any attempt at an empirical, let alone physical, reason for excluding any such series. The statistical issues of selection bias in climate reconstructions need no demonstration here.

  13. Kenneth Fritsch said

    Jeff ID, take a look at the post I made at One Size Fits All thread at CA. I think a lot of what we see in the Yamal series is noise. I looked at the same tree in the same year with replicate samples and I see considerable differences. I also did a graph of the Yamal series showing calculated CIs and that shows much uncertainty also. In addition, the in-tree differences I show appear to trend upward with tree age. I think we need to do the basic analysis first.

    That tree ring growth curve with the scatter and upward trend at the end that you show above is a little misleading if you are not aware of the scatter of points at over the entire range of tree ring ages. I posted a scatter plot with tiny red dots at CA that shows what I am talking about. Your curve is looking at means of deltas at each trre ring age.

    What surprises me is the under utilization of data by dendros that could be applied to doing better and more basic analyses. It appears there is more interest in making a connection with AGW – at least somewhere in a published paper.

  14. Jeff Id said

    Very nice, would you like me to repost it?

  15. […] Climate Audit and here whereby the non-linear response to temperature(as described by Dr. Loehle), or the regularization methods of tree ring data and noise have all contaminated the data to the point of non-usability.  Here I discovered that the […]

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: