Survivorship Bias

bender has sent in the following interesting graphic showing how drought-caused dieback could bias a tree ring chronology and interfere with interpretation of past records, emphasizing the continued need to keep U-shaped temperature response functions firmly in one’s mind when considering these series.

[image being re-sized]

(2012 note) For reference, here are some related postings discussing upside-down quadratic response and potential evidence of this in recent northern Canadian and Alaskan tree ring series (an explanation for the Divergence Problem not properly canvassed by the NAS panel).

Upside Down Quadratic
Twisted Tree Heartrot Hill
Wilmking in Alaska
Positive and Negative Responders
bender on Gaspé

Other potential biases to tree ring chronologies neglected by the NAS panel (and discussed here from time to time) are altitude changes, "modern sample bias", standardization bias, pith offset bias.


102 Comments

  1. Michael Lenaghan
    Posted Aug 4, 2006 at 8:05 AM | Permalink | Reply

    Climate Science has posted this: “Most of What a Climatologist Should Know About Correlations” by Robert Livezey and Marina Timofeyeva.
    Not quite on topic, but probably of interest.

  2. TCO
    Posted Aug 4, 2006 at 8:43 AM | Permalink | Reply

    1. Is there a good ref that shows that higher temps correlate to more droughts (for the kinds of tree locations that we study)?

    2. My impression is that the added temp of AGW will be concentrated at night and in the winter, so will have the effect of mildening the climate rather then making the unbearable summers in DC which both warmers and skeptics seem to unconsiously associate with AGW. EVen if the effect of AGW is average over the year, one still needs to consider how overtemp summer (or daily) times are compensated by other parts of the year. Would think that for trees at the “cold limit” of altitude/lattitude that the impact of increased temps would not lead to quadraticy until we get to temps like those at the “warm limit” of extent of the trees.

    3. I guess in practice changes in precip pattern or precip needs that correlate with temp are the important things for a proxy. Still the tree itself may have some botanical reaction to overly high temps even in the prescence of always adequate water. So for logical clarity, we should carefully label the concepts.

    4. My remarks are just critical in an exploratory manner. I like the point about survivorship bias and agree that we need to keep it in mind. There may also be some age-survivorship interactions.

  3. TCO
    Posted Aug 4, 2006 at 8:45 AM | Permalink | Reply

    Also in the spirit of moving things forward, what sort of field study (or greenhouse) stody would best test bender’s hypothesis. Not asking anyone to do it or why hasn’t it been done. But since talk is cheap, what would be an efficient, effective way to assess the issue?

  4. JerryB
    Posted Aug 4, 2006 at 9:20 AM | Permalink | Reply

    What was the source from which bender got the graphic?

  5. JP
    Posted Aug 4, 2006 at 9:23 AM | Permalink | Reply

    Here’s a link to a study done by David Stahle in 1998. He was working with anthropologists covering the severe drought in colonial Virginia. Stahle used the bald cypress on the Blackwater and Nottoway Rivers as proxies.

    http://tinyurl.com/f9dz6

    What’s interesting here is that this drought occured during the coldest period of the LIA. I don’t know if Stahle’s study was audited or peer reviewed. Can certain species of trees reveal both accurate precipitation history as well as temperature history?

  6. John Creighton
    Posted Aug 4, 2006 at 9:37 AM | Permalink | Reply

    I like the graphic a lot and I think it illustrates the point well. However, it does seems to imply that drought only occurs at high temperatures. Is there anyway to show the survivorship bias with a more realistic graphic?

  7. mark
    Posted Aug 4, 2006 at 9:38 AM | Permalink | Reply

    Hey, another tinyurl convert! :)

    Mark

  8. Tim Ball
    Posted Aug 4, 2006 at 9:41 AM | Permalink | Reply

    Accusations that acid rain was destroying maple trees and negatively impacting the mayple syrup industry in eastern Canada were disproved when it was shown that the cause of decline of yield was due to dieback from two major causes. One was a pattern of very early warm spells triggering spring growth followed by a severe cold spell that damaged the tree. The other was drought. When these conditions ameliorated maximum yields were achieved.

  9. bender
    Posted Aug 4, 2006 at 9:57 AM | Permalink | Reply

    Note this is a ‘cartoon’. It is for demonstrative purposes only, to illustrate how survivorship bias could work, nothing more. It is not a statement of fact. It is a hypothesis. Anyone wanting to improve the cartoon so that it is more applicable to their system may do so.

    Recall the comments of Willis Essenbach – trees can exhibit instantaneous responses to severe perturbations (fire, insects, drought) – such that a 3y drought could easily be embedded in the LIA, and could thus influence the tree-ring record.

    Were there more frequent droughts during the MWP? That’s a good question. That’s why I wanted to post the graphic, to show why this is a question worth thinking about.

  10. John Creighton
    Posted Aug 4, 2006 at 10:40 AM | Permalink | Reply

    #9 I understand that the graphic was just to illustrate a point. I just thought I’d bring the obvious to the fore font. It may be the case though that in warmer times that there is more drought in some regions. I’ve seen historical sketches of what the past looks like showing more deserts in warmer times. (At least at the lower latitudes) I am not completely sure why. In the short term perhaps it is a response to cloud cover. In periods where there is more cloud cover it has a cooling effect. Just a guess.

  11. bender
    Posted Aug 4, 2006 at 11:02 AM | Permalink | Reply

    I follow, JC. Every figure needs a caption, lest it be misunderstood. My comment was more a pre-emptive strike than a response to any one criticism.

    #3 TCO, you are right: talk IS cheap. I wonder what the tree-ring signatures look like in those drought-stricken pinyon pines. I wonder if the dead have narrower rings than the dying than the living.

  12. Ogie Oglethorpe
    Posted Aug 4, 2006 at 11:40 AM | Permalink | Reply

    ref #2,3

    In addition to TCO’s comments, what correlation could one draw regarding temperature and (predatory or neign) insect infestations? Is there any relationship?

  13. Ogie Oglethorpe
    Posted Aug 4, 2006 at 11:42 AM | Permalink | Reply

    It seems as if one should do a fishbone diagram to list out all the variables and then look to 1st and 2nd order effects therof. But thats just how I do industrial problem solving.

  14. beng
    Posted Aug 4, 2006 at 2:03 PM | Permalink | Reply

    RE 5:

    Interesting link since I’m somewhat familiar w/alot of these US mid-atlantic areas. Also interesting in that the period early 1500s to early 1600s are shown by many (but not all) proxies as a sharp downturn (~1550) in solar (leading to Maunder min), global temps, tree ringwidths, etc.

    Drought’s a tricky thing to speculate on in the mid-atlantic states — it can obviously occur w/both higher or lower annual temps. Dry/hot summers would correlate well, tho winters in the same yr could be very cold & either dry or wet (the 1970s & early 80s).

  15. Armand MacMurray
    Posted Aug 4, 2006 at 2:07 PM | Permalink | Reply

    Regarding droughts and temperature, in areas where substantial winter precip is stored as snow/ice and released throughout the year as melt, temperatures can directly cause “drought” by changing the distribution of water availability during the year. For example, an abnormally warm spring can melt most of the snow that normally would have provided water during the summer.

  16. bender
    Posted Aug 4, 2006 at 2:13 PM | Permalink | Reply

    It’s not just precipitation that’s the issue here. It’s evapotranspiration rates that determine whether you get loss of leaf turgor, wilt, stem cavitation, twig & whole-tree dieback or not – and these are as much determined by temperature as by precipitation. I’m not sure how much dieback you would get when it’s dry, but cool.

  17. JP
    Posted Aug 4, 2006 at 2:15 PM | Permalink | Reply

    #14

    This drought is of particular interest because it was smack in the middle of some of the coolest hemispheric temps for the last 500 years. Stahle’s analysis of the bald cedar proxies led him to say (according to the article) that the East Coast drought was one of the driest and longest in 800 years for North America. It would be nice if some other paleoclimotologists to look at his data and see if they concur.

    There seems to be a lot of questions the accuracy of some tree ring proxies regarding surface temperatures. I wonder if some issues also exist with tree rings and precipiation proxies.

  18. bender
    Posted Aug 4, 2006 at 3:03 PM | Permalink | Reply

    I wonder if some issues also exist with tree rings and precipiation proxies.

    Why wouldn’t they?

  19. TCO
    Posted Aug 4, 2006 at 4:15 PM | Permalink | Reply

    Q: “Why wouldn’t they”

    A: I would expect that some similar problems would occur given that RW is inherently at danger of confounding factors and the methods of dealing with that confounding are imperfect and even the degree of their imperfection not completely understood. That said, it might still be interesting to compare flaws in precip studies and in temp studies as a revealing compare and contrast study. Are there insights from what we learn about flaws in one that make us look and detect similar in the other (and this can go in either direction). Are there differences in accuracy because of proxy detecability versus parameter variation? ARe there differences in work practices in the field or in data handling?

  20. bender
    Posted Aug 4, 2006 at 4:20 PM | Permalink | Reply

    You want to turn over another rock before you’re finished with this one?

  21. TCO
    Posted Aug 4, 2006 at 4:28 PM | Permalink | Reply

    Huh? Did I do something wrong? The question of “do some errors exist” is trivial as you showed. I don’t really see much point in discussing further the “yes or no” question as to whether “some issues also exist with tree rings and precipiation proxies?”

    It’s like me asking the group if anyone has a pen and your replying, “yes.” (Someone, in the world, and maybe not willing to share it DOES have a pen!) A more intersesting question is “who has a pen and will loan it to me, bla bla”.

    If I’m reading you wrong and you think that we need to keep going on some other issue, sorry…what did you want to keep discussing?

  22. Dano
    Posted Aug 4, 2006 at 4:31 PM | Permalink | Reply

    RE 8 (Ball):

    Accusations that acid rain was destroying maple trees and negatively impacting the maple syrup industry in eastern Canada were disproved when it was shown that the cause of decline of yield was due to dieback from two major [climate-related] causes

    No.

    It is basic biophysical science and botany that your implicit assertion is, at best, nàƒ⣩vely incorrect.

    Acidification alters Cation Exchange Capacity and limits Ca++ uptake which weakens the plant and exacerbates climatic stressors (such as the ones Tim said in #8).

    1.
    http://intl-jeq.scijournals.org/cgi/content/abstract/31/6/1993
    http://intl-jeq.scijournals.org/cgi/content/abstract/31/5/1676
    http://www.esf.edu/efb/mitchell/Myron%20Ref/Evn.Pol.123.327.336.pdf
    http://www.blackwell-synergy.com/links/doi/10.1046/j.1526-0992.2001.01046.x

    Certainly there may be other reasons [e.g. pp 23-25] for reduced production in some areas, but Tim’s tout isn’t one of ‘em (it’s a stressor to an underlying condition).

    HTH,

    D

    [I'm not sure why the tags are screwed up in this comment...]

  23. bender
    Posted Aug 4, 2006 at 4:40 PM | Permalink | Reply

    Dano,
    Do a search of the more recent primary literature. It will surprise many people, and there are still a few deniers, but Ball is correct. The work on yellow birch is most convincing, but sugar maple appears to be the same. I’ve posted extensively on this in another thread. I don’t want to argue. Just to say, there has been a shift in the tide on that debate.

  24. bender
    Posted Aug 4, 2006 at 4:42 PM | Permalink | Reply

    #21 TCO, I forgot the humour symbol. Sorry. Good rock. Keep turning. Lift. :)

  25. TCO
    Posted Aug 4, 2006 at 5:03 PM | Permalink | Reply

    Humor. The last refuge of a botanist! :)

  26. beng
    Posted Aug 4, 2006 at 6:08 PM | Permalink | Reply

    RE 22:

    Acidification alters Cation Exchange Capacity and limits Ca++ uptake which weakens the plant and exacerbates climatic stressors (such as the ones Tim said in #8).

    It depends on what tree & under what conditions you’re talking about, Dano. Weakens which plants? Some plants are stressed by a lack of soil acidity. Many of my acid-deprived plants look visually greener after an electrical storm, presumably due to nitric oxides/acids produced & rained out during the storm (and those small amounts from anthro sources). I can produce the same result in these plants from watering/spraying w/acid fertilizer. Point is, acid helps some plants.

    But you already knew that, didn’t you, Dano? :)

  27. Dano
    Posted Aug 4, 2006 at 6:09 PM | Permalink | Reply

    23:

    Thank you bender.

    I’ll make some phone calls/e-mails, as my work is in western US urban ecosystems & my last forest ecol class was probably 6-7 years ago, so it’s possible I haven’t heard of these developments. I doubt the basic soil chemistry has changed since I was in uni, tho, so I’m interested in what’s changed.

    If it’s not too much trouble, if you come across your previous comments I’d like to read them, thx…

    Best,

    D

  28. bender
    Posted Aug 4, 2006 at 6:15 PM | Permalink | Reply

    Dano, found it here.

    It’s not the soil chemistry science that’s changed, it’s the soil structure science.

  29. Dano
    Posted Aug 4, 2006 at 6:26 PM | Permalink | Reply

    26:

    Weakens which plants? Some plants are stressed by a lack of soil acidity.

    The typical NE forest, beng, is stressed by exceedences [1] of acidity [2] due in large part to underlying soil chemistry’s inability to absorb added acid via deposition.

    You can acidify your garden soil with S or dark organic matter too, to avoid fertilizer application.

    Best,

    D

    [1] http://www.springerlink.com/(ssm42a55vxmdmjzmgrqa0jnw)/app/home/contribution.asp?referrer=parent&backto=issue,5,21;journal,4,325;linkingpublicationresults,1:100344,1

    [2] http://www.springerlink.com/(1shtx3vybhqluy45qo4jau45)/app/home/contribution.asp?referrer=parent&backto=issue,9,24;journal,16,325;linkingpublicationresults,1:100344,1

    [sigh...what's up with the tags today?!? ]

  30. bender
    Posted Aug 4, 2006 at 6:28 PM | Permalink | Reply

    Re #28
    P.S. Do a text search on “sugar maple” to get to the correct post in that thread.

  31. bruce
    Posted Aug 4, 2006 at 7:05 PM | Permalink | Reply

    Hey Dano. Love your work! While you’re here, perhaps you might comment, from your deep experience as a dendro guy, on the main topic of this thread. Explain to us plebs how it is that the HS corpus is based on assumption of a linear relationship between tree ring thickness and temperature, and why bender’s schematic of a more complex relationship is wrong.

    best

    b

  32. Dano
    Posted Aug 4, 2006 at 7:20 PM | Permalink | Reply

    28:

    Sheesh, all these cross-posts!

    Thank you bender.

    Yes, your McGill link says what I said in 22 [Acidification alters Cation Exchange Capacity and limits Ca++ uptake which weakens the plant and exacerbates climatic stressors]. Thus we have two places that refute the implication in #8 that there’s no problem at all.

    In an interesting indicator (not of the quality of your work but of others’), the link you provided to your post has a comment #8 that provides a paper that purports to disprove an acid rain-mortality link. I see nowhere in the paper do they test the soil, leaves, or leaf litter for increased acidification; rather it was a paper about gap dynamics in one forest. Just pointing out how wrong ideas get…er…planted.

    Which leads me to your (2) in #9. I don’t see a test of a hypothesis containing anything about increased acidification or loss of Ca++ (rather, it tested a hypothesis of xylem conductivity/cavitation to various freezing treatments). So I’m a little confused as to your assertion and I’m sorry I missed commenting on your post to draw this out (but the topic was tree rings anyway, so we’d be OT). Surely we can agree that reduced base CEC in a pot can affect a plants’ ability to refill embolisms.

    YMMV.

    BTW, my last research was in urban ecosystems and the earlier forestry work I did was on watershed-level ecosytem dynamics/NPP wrt changed Fire Return Intervals under modern forestry management, so I’m not the last word on the subject of your bumped comment. We’d core, get the FRIs, and give the core to the dendro ppl for further analysis & that’s the extent of my ring analysis. And I liked to core because my future wife looked very fine when it was her turn to sweat. ;o)

    Best,

    D

  33. bender
    Posted Aug 4, 2006 at 7:27 PM | Permalink | Reply

    I do want to stress that this is only a hypothesis. It’s sensible, but it is only a hypothesis. That’s why there’s an “Ho” and a “?” in the title of that graphic. TCO in #3 has got the right idea: someone experienced in BCP ecology should test this.

    But from a pure skeptic’s point of view, the onus is not on the auditer to PROVE the hypothesis. It is up to the paleoclimatologists wanting to make the assumption (linearity, no bias) to DISPROVE it. Have they done an adequate job? You be the judge.

  34. bender
    Posted Aug 4, 2006 at 7:38 PM | Permalink | Reply

    Hi Dano,
    Like I said – on acid rain, I’m not here to argue. Just to say that there is a vigorous debate going on about that right now, and it is neither black nor white. The link I provided was not intended to be authoritative, just something I pulled off the web real quick. If you consult the primary literature, you’ll get a much richer and more nuanced view of the issue. Note: the question in that debate isn’t whether acid rain exists or if it is bad or not. No expert I know disputes that it exists and is bad. The question is whether it was a necessary condition causing sugar maple decline. If you want some references, try the ones below. And yes, I know, they’re cherry-picked to favour one side of the debate.

    Alder, N.N., Sperry, J.S., and Pockman, W.T. 1996. Root and stem xylem embolism, stomatal conductance, and leaf turgor in Acer grandidentatum populations along a soil moisture gradient. Oecologia 105: 293-301.

    Auclair, A.N.D. 1993. Extreme climatic fluctuations as a cause of forest dieback in the Pacific rim. Water, Air, and Soil Pollution 66: 207-229.

    Auclair, A.N.D., Eglinton, P.D., and Minnemeyer, S.L. 1997. Principal forest dieback episodes in northern hardwoods: development of numeric indices of areal extent and severity. Water, Air, and Soil Pollution 93: 175-198.

    Auclair, A.N.D., Lill, J.T., and Revenga, C. 1996. The role of climate variability and global warming in the dieback of northern hardwoods. Water, Air, and Soil Pollution 91: 163-186.

    Auclair, A.N.D., Martin, H.C., and Walker, S.L. 1990. A case study of forest decline in western Canada and the adjacent United States. Water, Air, and Soil Pollution 53: 13-31.

    Auclair, A.N.D., Worrest, R.C., Lachance, D., and Martin, H.C. 1992. Climatic perturbation as a general dieback mechanism. In Forest Decline Concepts. APS Press, St. Paul, MN. pp. 38-57.

    Braathe, P. Birch dieback – caused by prolonged early spring thaws and subsequent frost. Norwegian Journal of Agricultural Sciences Supplement 20.

    Cox, R.M. and Zhu, X.B. 2003. Effects of simulated thaw on xylem cavitation, residual embolism, spring dieback and shoot growth in yellow birch. Tree Physiology 23: 615-624.

    Davis, S.D., Ewers, F.W., Sperry, J.S., Portwood, K.A., Crocker, M.C., and Adams, G.C. 2002. Shoot dieback during prolonged drought in Ceanothus (Rhamnaceae) chaparral of California: a possible case of hydraulic failure. American Journal of Botany 89: 820-828.

    Frey, B.F., Lieffers, V.J., Hogg, E.H., and Landhausser, S.M. 2004. Predicting landscape patterns of aspen dieback: mechanisms and knowledge gaps. Can. J. For. Res. 34: 1379-1390.

    Gross, H.L. 1991. Dieback and growth loss of sugar maple associated with defoliation by forest tent caterpillar. Forestry Chronicle 67: 33-42.

    Hinrichson, D. 1987. The forest decline enigma: What underlies extensive dieback on two continents? BioScience 37: 542-546.

    Hogg, E.H., Brandt, J.P., and Kochtubajda, B. 2002. Growth and dieback of aspen forests in northwestern Alberta, Canada, in relation to climate and insects. Canadian Journal of Forest Research 32: 823-832.

    Robitaille, G., Boutin, R., and Lachance, D. 1995. Effects of soil freezing stress on sap flow and gugar content of mature sugar maples (Acer saccharum). Can. J. For. Res. 25: 577-587.

    Siemens, J.A. and Zwiazek, J.J. 2003. Effects of water deficit stress and recovery on the root water relations of trembling aspen (Populus tremuloides) seedlings. Plant Science 165: 113-120.

    Sperry, J.S., Donnelly, J.R., and Tyree, M.T. 1988. Seasonal occurrence of xylem embolism in sugar maple (Acer saccharum). Amer. J. Bot. 75: 1212-1218.

    Sperry, J.S., Nichols, K.L., Sullivan, J.E.M., and Eastlack, S.E. 1994. Xylem embolism in ring-porous, diffuse-porous, and coniferous trees of northern Utah and Alaska. Ecology 75: 1736-1752.

    Wargo, P.M. and Auclair, A.N.D. 2000. Forest declines in response to environmental change. In Responses of Northern U.S. Forests to Environmental Change. Springer, New York.

    Zhu, X.B., Cox, R.M., and Arp, P.A. 2000. Effects of xylem caviation and freezing injury on dieback of yellow birch (Betula alleghaniensis) in relation to a simulated winter thaw. Tree Physiology 20: 541-547.

  35. bender
    Posted Aug 4, 2006 at 7:45 PM | Permalink | Reply

    Re #34,
    Hi Peter K. Anderson,
    You say “there’s not any direct relation to “drought’ and temperature”. Can you cite some authoritative references? I don’t doubt what you’re saying, but I would like a little more quantitative proof. Realize too, that for a plant “drought’ means excessively high evapotranspiration rates, where temperature is directly implicated. So you can see why your post would interest me. Thanks!

  36. TAC
    Posted Aug 4, 2006 at 7:45 PM | Permalink | Reply

    The original post addresses something I’ve wondered about for a while: Even if we accept that trees, on average, respond to climate, how do we know that the trees we see today are representative of the trees that were present in the past? On its face, it seems a doubtful proposition: In fact, we know the survivors are different from the others — they survived. The ones that didn’t make it would likely tell a different story. I had assumed that dendrochronologists had found a way to correct for this.

    The basic problem amounts to “sampling bias” (where the “target population” (the entire population of trees that were alive in the MWP, for example) differs from the “sampling population” (trees that were alive then AND survived to the present)). Statisticians are taught to worry about sampling bias, in part because of a shameful history. One famous example: In the early years of Statistics, a magazine called Literary Digest (LD) decided to “call” the 1936 U.S. presidential election. It polled its readers and — because it was easy to do — people with telephones. The LD predicted (correctly, as it turns out) that the 1936 election would be the greatest landslide in American history. The part they got wrong was that Roosevelt, not Landon, was the winner. Subsequent studies established that subscribers to LD and people with telephones did, in fact, vote overwhelmingly for the Republican Landon. Everyone else — including voters who could not afford telephones or subscriptions to LD — voted overwhelmingly for Roosevelt. The experience left a black mark on Statistics. It also marked the demise of LD.

    Despite all of the other problems that SteveM has uncovered, I still find it hard to believe — really incredible — that the analysis of proxy records might also have this problem. Whatever. Wegman was likely right when he said that climate scientists need to work more closely with statisticians.

  37. bender
    Posted Aug 4, 2006 at 7:48 PM | Permalink | Reply

    Re #37 I think you are right: Wegman was probably right.

  38. bender
    Posted Aug 4, 2006 at 7:59 PM | Permalink | Reply

    TAC,
    That’s why I decided to post the graphic – because I felt a lot of posters were dancing around the issue, without hitting it right squarely on the head. True or false, its purpose is to encapsulate in a single picture all the disparate things I’ve been hearing from people like Willis, John A, Steve, TCO, and many others. Maybe a dendrochronoloigst can refute it. Maybe they’ll just dismiss it. It would be interesting to know.

  39. JP
    Posted Aug 4, 2006 at 8:06 PM | Permalink | Reply

    “The basic problem amounts to “sampling bias” (where the “target population” (the entire population of trees that were alive in the MWP, for example) differs from the “sampling population” (trees that were alive then AND survived to the present”

    TAC, that was what I was driving at in my orginal question concerning tree proxies and thier accuracy. From my #5 post, I stated Stahle’s assertion that the 15th century drought was one of the worst in N. American History based on bald cedar tree rings. Has anyone disputed his claim based on biological evidence with trees when there is so little rainfall?

    We do know from human records that this 15th century drought was severe; Stahle went beyond that based on just tree ring samples from Virginia. The article piqued my interest primairily because it occured during a period of what was once considered one of the coldest periods since AD800.

  40. TCO
    Posted Aug 4, 2006 at 8:10 PM | Permalink | Reply

    I thought it was a good graphic. We have talked about some of the elements at differnt times, but this was additive (especially wrt survivorship correlation to heat) and the graphic was a good framework. I like it. Now we do the thought construciton of how to test it (my comment about talk being cheap was mean positively).

    I do think that there are a lot of different ways to cut things, so we should be careful to differntiate them: survivorship of other sorts, the covariance of precip with temp, the timing of heat, pure temp quadraticy (unrelated to water availability), etc. variation of need for precip with temp, etc.

  41. TAC
    Posted Aug 4, 2006 at 8:11 PM | Permalink | Reply

    #39 Bender, thanks for bringing up this topic and shedding light on it. I, too, would be interested in the answer.

  42. bender
    Posted Aug 4, 2006 at 8:15 PM | Permalink | Reply

    The basic problem amounts to “sampling bias”

    Yes, survivorship bias is a specific kind of sampling bias – one resulting from differential survivorship of, in this case, healthy vs drought-stricken trees. If the drought-stricken trees have a different signal in them than the healthy trees, and they die, then that signal is lost from the record. Thus it is a sampling bias that is very difficult to correct for.

    It would suggest that the “Divergence Problem” might not be unique to the 20th century, but that evidence for it from previous times is simply lost from the record. This in turn might explain why a linear approximation might *seem* to work, when it could actually be failing quite badly.

    Devil of a problem.

  43. bender
    Posted Aug 4, 2006 at 8:28 PM | Permalink | Reply

    Re #41
    I know pinyon pine are vulnerable to drought and that it’s the moisture limitation information that would be lost from the record. That would be a good test case. I don’t know if BCP are vulnerable to drought (or winter dessication, which is the same thing, different seasonal process). That would be a good thing to know. The cedars are an interesting case because you have one single species exhibiting different limitations depending on if it’s treeline (temperature) or cliffside (moisture).

    The question of covariance in temperature/precipitation is a good one too. The reason I framed all this in terms of hot+dry summer drought is simply because of what we see today with pinyon pine: hot+dry summer=death.

  44. Tim Ball
    Posted Aug 4, 2006 at 10:20 PM | Permalink | Reply

    The discussion about droughts is confused. First, there are different definitions of drought. Second, do not confuse droughts
    and desert regions. Third, droughts can occur in all regions even in deserts but the mechanisms are usually different.
    Deserts are simply defined by the Koppen system of climate classification as B climates. These can be hot deserts (Bw) or cold deserts (Bk). The polar regions are cold deserts. The Bh hot deserts occur under the descending air on the poleward side of the Hadley cells and generally occur between 15 and 30° of latitude. This is distorted by the land water pattern, just as the heat equator is not coincident with the geographical equator.
    As global temperature changes the polar regions expand and contract. The desert regions also expand and contract by becoming more dry (known as interpluvials) or less dry (pluvials) These are documented amongst other evidence by changing compositon of sediments off the coast of Africa. With warmer temperatures the desert regions expand mostly toward the poles. In North America during warmer periods the Great American Desert expanded north and east. A similar expansion occurred in the Medieval Warm Period as partly evidenced by the ultimate collpase and migration of the Anasazi people.
    The other significant drought cycle is found in the middle latitudes. This most clearly defined in the northern hemisphere (there is very little land mass in the middle latititudes of the southern hemisphere). The pattern of these droughts follow a 22 year cycle that is coincident with the 22 year solar cycle.

  45. bender
    Posted Aug 4, 2006 at 10:29 PM | Permalink | Reply

    A similar expansion occurred in the Medieval Warm Period

    Tim, do you have any references handy for that?

  46. John Creighton
    Posted Aug 4, 2006 at 10:32 PM | Permalink | Reply

    I’m having trouble with the exact nature of this survivor bias. If it was warmer then today in the MWP but only the trees survived that were in regions comparable to days temperatures then wouldn’t the tree rings observed from MWP have rings the same size as today’s trees? It would seem to me you would get a greater bias if all the tree’s lived just as a result of the nonlinearity.

    I agree that if there was drought in the MWP then that would explain why the tree rings were smaller. That seems more like an issue of on modeled or unobserved effects rather then survivor bias.

  47. TAC
    Posted Aug 4, 2006 at 10:35 PM | Permalink | Reply

    #42 and #43 “Devil of a problem” describes the situation well, at least if I understand it correctly.

    This is vaguely reminiscent of a hugely controversial study that came out a few years ago — I think it was in Nature or NEJM — asserting that life expectancy of left-handed people was 9 years shorter than that of right-handed people (as a lefty, I pay attention to such news). The researchers apparently had regressed age-at-death vs. handedness, and also had plotted up the age distribution of lefties in the society. There it was, plain as day: A 9-year difference!

    What they failed to recognize, however, was that until recently left-handedness was not well tolerated in many parts of our society. As a result, many lefties were “taught” to be right handed. So, there aren’t many old lefties. Consequently, the average age of death of lefties is reduced (for now).

    In that case, there was a way to get around the sampling problem by using “longitudinal” datasets. Later studies, using these datasets, confirmed the problem with the original study and seem to show that life expectancies are about equal — within a year, anyway — which is good news for us southpaws.

    However, I don’t see an obvious way to get around the sampling bias in tree studies. Perhaps it is small enough that it can be ignored?

  48. bender
    Posted Aug 4, 2006 at 10:45 PM | Permalink | Reply

    Remember JC, we are talking about treeline trees that are, in general, limited by cold temperature. That means you’re generally on the increasing, left hand part of the parabola. Today it is warm. So we are at the peak, and maybe even starting down the right hand side of the downslope on the parabola. Now what if the drought of the MWP that forced out the Anasazi was so bad that the California desert expanded (as posited by Tim Ball), those trees at treeline were now moisture limited, and growth was at the low end of the right hand side of the parabola. But what if the drought was so prolonged that those tres actually died, and only the ones that were not moisture limited survived? (e.g. The ones growing in slightly wetter depressions, were more drought-adapted, etc.). Then that lower right hand side of the parabola would be selected out of the modern-day record. No dendrochronologist could ever sample those trees. Thus the modern record is a biased sample compared to the population that existed 1000 years ago. And what would that record look like, supposing you could reconstruct a nonlinear temperature response (which you can’t)? It would have a MWP much warmer than we see evidenced in today’s BCPs.

    Only a hypothesis. But think about it before replying – it’s logical.

  49. bender
    Posted Aug 4, 2006 at 10:49 PM | Permalink | Reply

    Perhaps it is small enough that it can be ignored?

    Maybe it is and maybe it isn’t. Wihtout data, how do you know? You can guess what the dendroclimatologists would prefer to believe, however.

  50. bender
    Posted Aug 4, 2006 at 10:59 PM | Permalink | Reply

    It would seem to me you would get a greater bias if all the tree’s lived just as a result of the nonlinearity.

    No, that would merely exaggerate the poor fit of the linear model. This is only 1/2 the problem. The survivorship issue adds a second element to it.

    In fact, the full hypothesis argues that, ironically, it is the higher rate of mortality of the drought-stricken trees which makes the linear approximation work as well as it does! i.e. When you’re in the early stages of a warming trend, as in today, what you get is the increasingly noted “loss of sensitivity”. The nonlinearity you refer to shows up. Ten years of drought later, those trees are long since withered away. Then the measured response looks alot more linear … but only because you’ve truncated the high end of the response curve.

  51. John Creighton
    Posted Aug 5, 2006 at 12:14 AM | Permalink | Reply

    Maybe the “cartoon’ is backwards and the trees that showed the greatest response in the medieval warm period all died out because it was too cold in the little ice age.

  52. TAC
    Posted Aug 5, 2006 at 5:08 AM | Permalink | Reply

    #51 It seems “plausible” that these tree studies amount to sampling from truncated distributions, although the details of the truncation are not very clear. In any case, it makes for very difficult statistics — much harder than the relatively trivial problem of computing population statistics from, for example, censored data (where the magnitudes of small (large) observations are unknown, but we know how many values were censored). Perhaps there are empirical likelihood methods that can be employed. However, first one would need to understand the truncation mechanism.

    I think I agree with #50:

    Maybe it is and maybe it isn’t. Without data, how do you know?

    What a mess!

    Re: #48 Some prominent statisticians discuss the “life expectancy/left-handedness” issue here.

  53. Steve McIntyre
    Posted Aug 5, 2006 at 5:41 AM | Permalink | Reply

    #52. don’t forget about positive and negative responders here

  54. TAC
    Posted Aug 5, 2006 at 6:44 AM | Permalink | Reply

    #53 Thanks for the reminder. I just took a look at here, and all I could do was laugh! “Ridiculous” does not begin to describe these proxy studies. Makes me wonder. How many of the signs (i.e. +/- in front of the matrix elements) in the first row of the PC transformation matrix are “wrong” (i.e. inconsistent with hypothesized processes)?

    I suppose I should not laugh. These studies, and their manifold errors, are not just of academic interest.

  55. Geoff
    Posted Aug 5, 2006 at 7:45 AM | Permalink | Reply

    Perhaps some of the knowledgeable readers can comment on this article by Zweifel.

    He seems to stress that moisture is the most important factor in tree ring growth:

    “There is a missing link between tree physiological and wood-anatomical knowledge which makes it impossible mechanistically to explain and predict the radial growth of individual trees from climate data”

    “The results further suggest a strong dependence of radial growth on the current tree water relations and only secondarily on the C-balance. A concept is discussed which links radial growth over a feedback loop to actual tree water-relations and long-term affected C-storage to microclimate”.

    The study covers a recent period in Europe during which one year was hotter that the rest, 2003. “For much of Europe including Switzerland, the year 2003 was extremely hot and dry (Beniston, 2004). As a consequence the gas exchange of Q. pubescens and P. sylvestris at Salgesch and also P. abies at Jeizinen ceased completely from mid July onwards and returned to a low level only in September (Fig. 4). At the cooler and wetter site Jeizinen (300 m higher in altitude), T PET_1 was also drastically reduced, but Q. pubescens and P. sylvestris were able to maintain a low but continuous activity throughout the summer. These climatic conditions led to narrow tree rings in all species”.

    They do comment on a tempature effect, but find it will sometimes be negative:

    “These results further support experimental findings that the current tree ring seems to be determined by the current and the preceding climatic conditions (Breda and Granier, 1996; Rigling et al., 2002; Fonti and Garcia-Gonzalez, 2004). At a dry and hot site such as the Wallis, a temperature effect is potentially negative and expected to be less pronounced since Q. pubescens has been shown to be drought- and heat tolerant for a wide range of locally occurring conditions”.

    “This balance between a positive and a negative response to temperature is expected to reduce a clear temperature-related growth trend in tree rings of P. sylvestris at these sites. In a dendroecological investigation of P. sylvestris in the same area, Rigling et al. (2002) reported a significant negative correlation between temperature and radial growth for the month of June and a positive but non-significant relationship for the month of November of the past year. These results were based on thetotal annual increments and have therefore a limited power of explanation for a growth mechanism. They represent more the indirect effects on growth over a longer time-scale”.

    Are these trees very different that other trees used in dendroclimatology? Seems they are more suited to measuring rainfall than temperature.

    Ref:

    Roman Zweifel, Lukas Zimmermann, Fabienne Zeugin and David M. Newbery, (2006) Intra-annual radial growth and water relations of trees: implications towards a growth mechanism; Journal of Experimental Botany, Vol. 57, No. 6, pp. 1445–1459, 2006
    doi:10.1093/jxb/erj125, available here

  56. bender
    Posted Aug 5, 2006 at 8:30 AM | Permalink | Reply

    Re #55: Dendro people more or less know all this. They would argue that that’s exactly why they cherry-pick: their botanical insight gives them the foresight needed to sample from microsites where the positive responders are likely to grow. i.e. These may be rare, but you can find them if you know how.

    But is it “botanical foresight” giving you the positive responders or statistical hindsight?

    What Steve has opened my eyes to is how much sifting is done of the output after-the-fact. I had no idea. I read methods, results, and conclusions, not appendices.

  57. bender
    Posted Aug 5, 2006 at 8:43 AM | Permalink | Reply

    intra-annual vs. inter-annual growth:
    Geoff: note the title on that Zweifel paper refers to INTRA-annual growth: growth within a single year. Tree-ring proxy studies are interested in INTER-annual growth: growth among years. Not that the paper’s irrelevant. Just that you have to be careful about extrapolation of its conclusions to the inter-annual situation. (Which is Willis’s big, good argument.)

  58. bender
    Posted Aug 5, 2006 at 8:51 AM | Permalink | Reply

    Re #51
    “Maybe the cartoon is backwards”

    Maybe it is. But I’ve never heard of massive dieback during cold episodes. Whereas there is currently a massive dieback of pinyon pine going on as a direct response to drought during a warming trend. But maybe you’re right, maybe things are different in the alpine situation. I wouldn’t know. As I say, I am not a BCP ecologist.

  59. Dave Dardinger
    Posted Aug 5, 2006 at 8:52 AM | Permalink | Reply

    re:#55

    I think the whole point of using treeline stands of trees is that they’re much more likely to be at their lower temperature limits rather than water limits. The trouble is that A) the normal ups and downs of temperature through the year isn’t going to keep them at the low temperature range and thus they may spend part of the time on the right side of the drawing. But the theory is that such trees will spend far more time on the left side and on average there will be a strong temperature signal.

    Where I see a problem is that because the treelines move to different heights in different periods of history, the likelihood is that a given stand of trees isn’t going to be in the theoretical low temperature regime throughout its life and thus the limiting factor will be something else. Now this would be fine, in a sense, since the low frequency signal would still probably be temperature related (though it might be confused with a long-period wet-dry signal. But it certainly makes trying to compare decades and years a theoretical impossibility since the high-frequency signal riding on top of the temperature signal is probably going to be something other than temperature. (There can be separate temperature signals, of course, with different frequencies and different causes).

  60. TCO
    Posted Aug 5, 2006 at 8:53 AM | Permalink | Reply

    The way to push the debate forward is to publish field studies that are explicitly written down with methods before hand, where all decisions, good/bad are recorded and where chips are allowed to fall as they may. You will still get screams from experts that “of course you didn’t get the right answer, you didn’t use the secret decoder ring, etc.” But the whole campaign of publishing such reports will start to raise doubt about post-sifting versus pre-identifying and will at a minimum drive the field to really quantify and record their pre-identifying traits more quantitatively.

  61. Dave Dardinger
    Posted Aug 5, 2006 at 9:26 AM | Permalink | Reply

    re:60

    Methods are both important and good for one’s citation index. Publishing a protocol which is adopted in a field will get you cited every time someone uses it for years, even decades. The one Journal Article which has my name as an author (though a couple other articles cite my MS thesis) is still cited because the method of assaying for the enzyme activity we developed is still useful 30 years later.

  62. TCO
    Posted Aug 5, 2006 at 9:50 AM | Permalink | Reply

    unpublished science is wasted resources…

  63. John Creighton
    Posted Aug 5, 2006 at 9:54 AM | Permalink | Reply

    #62 You have quite a publishing obsession TCO. While publishing is the most important way someone can contribute to science it takes a lot of time and energy. If that time was spent researching instead of studying perhaps the results could be much more significant.

    It is interesting that the biggest pries in mathematics the fields award does not depend at all on the quantity of publications. Rather it depends entirely on the significance of the work.

  64. TCO
    Posted Aug 5, 2006 at 10:06 AM | Permalink | Reply

    But the work here is unfinished, not in context, and (at times) skewed.

  65. Steve McIntyre
    Posted Aug 5, 2006 at 10:14 AM | Permalink | Reply

    #63. When I went to my university reunion last spring, I ended up in the line for dinner chatting to John Polanyi, the famous scientists, and Jim Arthur, a Fields Award winner. Pretty heady company. Arthur was extremely charming. I sat beside him at dinner. He was a couple of years ahead of me at Trinity College, University of Toronto and either remembered me or convincingly pretended to remember me from 35 years ago, thus giving me a very favorable impression of the acuity of Fields Award winners.

    For non-math people, a Fields Award is like a Nobel Prize in math – although the quality of winners is probably more consistently high.

  66. TCO
    Posted Aug 5, 2006 at 11:37 AM | Permalink | Reply

    dude was he drawing on the windows like Russel Crowe?

  67. David Smith
    Posted Aug 5, 2006 at 12:53 PM | Permalink | Reply

    Since there are many knowledgeable, talented people visitng this site, I’d like to ask a question here:

    Does anyone know of a source for current (and past) global cloud cover, probably expressed as a % cover? The various satellite agencies generate a lot of data, but I’ve been unable to find this particular one.

    Thanks, and my apology for being off-topic.

    And TCO, your position on publishing is appreciated, as is your offer to pause your reminders to Steve about publishing.

    I find that I just quick-scan your comments because they have been like a broken record. I have probably missed some important content posted by you because of that.

    David

  68. Armand MacMurray
    Posted Aug 5, 2006 at 1:13 PM | Permalink | Reply

    I remember that in some of the (Siberian? Uralic?) published examples of nice fieldwork, involving latitudinal and altitudinal transects, the researchers sampled “sub-fossil” material (stumps?) in addition to living trees. It seems that this sampling method would reduce the survivorship bias.

  69. bender
    Posted Aug 5, 2006 at 2:20 PM | Permalink | Reply

    Re #68
    Any time you can get time-series data from samples of the past and compare it to “comparable” time-series data from samples from the present, it is a good thing – because it allows you to test* if the statistical distribution of ring widths has changed over time (e.g. due to truncation effects). It would be a great idea to examine this question with any hockey-stick species, such as BCP or Jacoby’s cedars. I don’t know if any older material exists for BCP (e.g. those withered trunks?), but Kelly et al. 1994 do talk about dead old cedars that are still available for sampling.

    *Since I know TCO will ask – because (s?)he always does – the actual name of the test is the Kolmogorov-Smirnov test.

    Louise Filion has done some good work on sub-fossil logs from paludified deposits in Québec, abstract here. She does a nice job showing it was hemlock looper that cleaned out hemlock from the pollen record ca. ~5000BC.

  70. TAC
    Posted Aug 5, 2006 at 3:54 PM | Permalink | Reply

    #62 TCO, I know you mean well when you state

    unpublished science is wasted resources

    but I don’t think you’re right in this case, and because you repeat this message so often, I feel a need to speak out.

    Under normal circumstances, in a healthy scientific discipline, I would agree with you: Publications are good. However, in this particular area of climate science, for whatever reason (Wegman’s social network analysis hints at the problem, I think), things are sadly different. While other disciplines welcome and celebrate skeptics and critics (the sharper and more vocal the better), this area of climate science seems committed to silencing dissent. Defending the “consensus” seems to take precedence over ensuring the validity of research. Skeptics are not welcome, and neither are their manuscripts. Under the circumstances, I’m not convinced that publishing is either realistic or sensible.

    Having said that, obviously SteveM can do whatever he wants; my thoughts on the matter are irrelevant. However, I feel compelled to point out that SteveM has achieved far more with his single publication (and this blog) than most of us have achieved with our dozens or hundreds of academic journal articles. He has gotten the full attention of three Congressional Committees, an NAS panel, and the international media. He sits alongside — as an equal — the world’s most respected scientists in this field. In addition, he and Ross have managed to discredit a prominent and influential piece of work. Not bad for a couple of outsiders.

    SteveM isn’t following the usual academic path, but it seems that neither are the people whose work he is auditing. However, SteveM has found his own way forward. Who can argue with this?

  71. TCO
    Posted Aug 5, 2006 at 4:31 PM | Permalink | Reply

    Steve’s not getting published is because he has not tried. I’m sick of people here saying that Steve can’t get into print when he does not say so and saying that he is getting papers rejected when he doesn’t even write them.

  72. Ken Fritsch
    Posted Aug 5, 2006 at 4:57 PM | Permalink | Reply

    re: #60

    The way to push the debate forward is to publish field studies that are explicitly written down with methods before hand, where all decisions, good/bad are recorded and where chips are allowed to fall as they may. You will still get screams from experts that “of course you didn’t get the right answer, you didn’t use the secret decoder ring, etc.” But the whole campaign of publishing such reports will start to raise doubt about post-sifting versus pre-identifying and will at a minimum drive the field to really quantify and record their pre-identifying traits more quantitatively.

    TCO, I have tried, I have really tried to get the gist of some of your posts with the above example being a prime example. I read it before my martini and nothing registered. Than I tried it with 3 oz. of Bombay Sapphire (shaken with ice and not stirred, poured into chilled martini glass and topped with 3 large olives). The result is that I can now put my own interpretation on what you are saying and of course it makes perfect sense.

    Please follow my observations on this particular thread and tell me if it connects with what you said in the above post.

    I must first interject that I have problems with your advice to Steve M, the way it is delivered and the frequency with which it is delivered. You might to what you do if you where Steve’s father, but I have two grown sons to whom I would not deliver your kind of advice. Then again they inherited some surliness from their father that I do not see in Steve M.

    This thread allows some use of posters’ intellect in looking at how data in climate reconstructions could easily be misinterpreted. I thought, however, that we already had concluded that the tree ring widths, even with adjustments for age and density, were subject to such a large number of potential variables that their use in temperature proxy reconstructions was suspect — unless someone could demonstrate how the these potential effects could be handled/modeled in the calibration period where the effects and temperatures where known or could at least be estimated. I would rather we review that literature, if it exists, that deals with measuring these effects. I think in that way I am more like you in that I want/need someone to explain these details of climatology to me and in a way I can understand it. I would also want to know if that literature does not exist or if it does but is beyond my comprehension.

  73. Posted Aug 5, 2006 at 5:20 PM | Permalink | Reply

    Ken – you bring up an interesting point.

    If we analyze various kinds of trees, to find what their response to various changing factors are (e.g. temperature, moisture, CO2), then perhaps what it might be possible to do is build a computer model of those trees, which takes those things as inputs and outputs width and density series.

    Then what one could do is populate a virtual map with these trees, and “replay” historical records of (recently) climate – temperature, rainfall, CO2, etc. – and then do “virtual” multiproxy studies. The goal would be to have the output of those studies closely match actual studies.

    If this could be done from first principals – i.e. with greenhouse studies of various types of trees, and not using random “fudge factors” – then I think it would be a helpful exercise, for the following reason.

    If this model could be shown to be accurate, then the output of multiproxy studies for dates before which we have instrumental data, could be fed into this model, along with the locations and types of trees used in the study. We would expect the output of this study to then match the inputs to the multiproxy studies. If they do not, it means that something – such as the problem being discussed in this thread – has thrown the reconstructions out, and it would allow us investigate exactly what caused it and whether it might be possible to do accurate reconstructions anyway, by adjusting the way the study processes data and re-running it through the model to see if that improves the situation.

    The model could even take into account the death of trees, and decide to take them out of the final samples if it determines they would not be preserved to the present day.

    I realize we’re all highly suspicious of models but I think this type could be constructive and interesting. Think of it as a verification model, if you will. One of the things things about it is, if it just plain doesn’t work, we’d know, because we’d never be able to get it to match the expected chronologies derived from intrumental data in the first place.

  74. John Creighton
    Posted Aug 5, 2006 at 6:47 PM | Permalink | Reply

    I think using a genetic algorithm to try and fit climate data would be a really cool idea. It certainly would be challenging and there would be numerous statistical, physical and biological issues that would have to be combined to decide what is the best fit to past climate data. If we could track, rivers, deserts, soil moisture, insect patter, tree growth, co2, etc. etc. etc.. back though time with any kind of statistical certainly it would be quite a hoot.

    I am always wary of computer models and think that perhaps they are a good tool but must be used in context with other evidence and techniques.

  75. bender
    Posted Aug 5, 2006 at 7:02 PM | Permalink | Reply

    Re #74
    JC: Genetic algorithms, you realize, have the same risk of over-fitting as methods such as the Mannomatic & RegEM & neural networks, right? This is not fatal to the approach, mind you, but you do have to somehow contend responsibly with that pesky overfitting problem. (You can use cross-validation methods to prune back your regression trees, but my sense is the statistics here get pretty dicey.)

    The urge to try to get “something for nothing” is universal. Hence the tendency in all fields where data is expensive to acquire (=models hard to validate) to ignore those pesky little gray-area problems like: serial autocorrelation, distinguishing nominal vs. effective degrees of freedom, adjusting confidence levels for multiple, simultaneous, a posteriori hypothesis tests, etc.

    If these problems were easy to solve, they’d be solved already.

  76. John Creighton
    Posted Aug 5, 2006 at 7:35 PM | Permalink | Reply

    I totally agree there is a risk of over fitting that is why I said:

    ” It certainly would be challenging and there would be numerous statistical, physical and biological issues that would have to be combined to decide what is the best fit to past climate data.”

    You’re cross validation technique could work. I was thinking of perhaps collecting data to show the distribution of climate variables. That is tree ring widths in one area and one time wouldn’t all be the same thickness. So we could fit the data to the distribution we expect and then try to use a genetic algorithm to model the conditions that would give that distribution of data. Actually I think it should go a step further, the genetic algorithm should model the distribution of possible conditions that might produce that data.

    The genetic algorithm should be multi objective. One objective is how close each model of the past approximates the distribution of past tree ring widths. The other is how different the models are. The idea is to keep a diverse population of possibilities.

  77. Paul Linsay
    Posted Aug 5, 2006 at 9:24 PM | Permalink | Reply

    RE: modeling tree growth. Are trees any different from any other living organism in that there is great variability between individuals and their response to identical conditions? Everyone in a family is a different height, weight, strength. Even with a careful greenhouse study of growth relationships doesn’t this make any model very problematic?

    RE: tree survivorship. Prior to Europeans arriving in North America, the environment was managed by the Indians using fire. In the late fall they would set large fires to burn out the underbrush and provide forage for their favored game in the spring. This had to have had a major impact on which trees survived and which didn’t. The population of trees in existence today hasn’t been culled that way so it’s not clear that it would be representative of the pre-Columbian trees used for dendroclimatology.

  78. John Creighton
    Posted Aug 5, 2006 at 9:33 PM | Permalink | Reply

    Hmmmm…fire, yet another complication. Jack pine growth could indicate previous fires. I’m glad I’m not a dendroclimatology.

  79. Steve McIntyre
    Posted Aug 5, 2006 at 10:52 PM | Permalink | Reply

    #76. If you go to the Esper category, I plotted up some ring width distributions last August. HEre here for example.

    BTW I’ve made R-tables of about 3000 tree ring data sets at ITRDB organized as. The formats at ITRDB are inconsistent with a few major types, but you have to watch for little ad hoc format changes.

    Tree-id, Year, Age, Ring Width

    So if anyone wants a data set for a particular site in R-format, I’ll upload it.

  80. Posted Aug 5, 2006 at 11:09 PM | Permalink | Reply

    Paul Linsay : Yes, you’d want to throw random variation of individual trees in there. Hopefully with enough samples that averages out, both in any model and in the real world. Isn’t that one of the reasons why reconstructions involve as many individual trees in a given location as possible?

  81. John Creighton
    Posted Aug 5, 2006 at 11:14 PM | Permalink | Reply

    #79 Steve it is an intriguing proposition but I am not sure I want to take on a task that big by myself. Even with course approximations to frosts based on assuming homogenous distributions over large areas I think it will require distributed computing. Given I don’t think any complex numeric routines will be required, I think java might be a suitable language because it fairly easy to program and well suited for the web.

    Actually, a distributed computing approach would be fascinating because the more people running the software, the greater the quantity and diversity of population in the genetic algorithm. This would give a much clearer indication of what the confidence intervals actually are.

    Still before any code would ever be written I think it would take considerable planning. I am usually more interested in theory then hard core coding because coding can sometimes be too much grunt work. Also I know how easy it is to introduce bugs into code. How difficult would it be to test, validate and debug such a program.

  82. John Creighton
    Posted Aug 5, 2006 at 11:32 PM | Permalink | Reply

    #80 I think this whole idea that you can just average out tree widths to get the relationship between temperature and tree widths is just naiveté. Tree widths are roughly related to temperature by:
    W=k1- k2 T^2
    The relationship between the mean width, the standard deviation of the temperature and the mean of the temperature is given by:

    W_bar=k1 — k2 T_bar^2 — k2 sigma_T^2 + 2 sigma_T T_bar

    Where:
    T bar -is the mean of the temperature
    sigma_T -is the standard deviation of the temperature.
    W_bar -is the mean tree width

    The variance of the tree widths actually depends on the third and forth order statistics. Perhaps if you had a bunch of trees that responded differently to temperature in a close enough area that you new the temperature was the same then you could solve some kind of optimization problem to get the mean temperature and the standard deviation of the temperature back. This is a far cry from averaging though.

  83. fFreddy
    Posted Aug 6, 2006 at 1:51 AM | Permalink | Reply

    Re John Creighton
    John, a while ago John A installed tex on this blog, if you are familiar with it.

  84. Posted Aug 6, 2006 at 3:05 AM | Permalink | Reply

    John Creighton : Well, you certainly are much more qualified than me to be able to comment about this issue. But what I have to say is this:

    If what you say is true, doesn’t that cast doubt on any simple analysis of dendrochronological data?

    How many of the dendro. studies take that into account? I don’t know the answer, but I’ll hazard a guess: 0

    Can anyone enlighten me? Mr. McIntyre?

  85. Ken Fritsch
    Posted Aug 6, 2006 at 9:55 AM | Permalink | Reply

    I have reviewed some of Steve M’s linked blog articles and their subsequent discussions at this blog about published analyses of RW vs T and found that evidently the background information from these literature published articles about which I am most interested in learning is either missing, incomplete, contradictory and not explained in detail. For example, the apparent lack of use of MXD in Esper et al., 2002 and confusion on how a linear and nonlinear division is used in adjusting the raw RW versus T relationship.

    One would think with the potential impact of these studies that all of this information would be gathered together and explained in detail in a book or review article for the likes of the comprehension levels of a TCO or even me — well at least a TCO as he has confessed to an ability to comprehend the general gist and quality of highly technical papers.

    TCO, perhaps with your innate abilities in this area you could do a review for this blog along these lines. I’ve got my gin bottle ready.

  86. bender
    Posted Aug 6, 2006 at 10:10 AM | Permalink | Reply

    One would think with the potential impact of these studies that all of this information would be gathered together and explained in detail in a book or review article for the likes of the comprehension levels of a TCO or even me.

    I think the closest thing you get to this is the Cook & Kariukstis 1990 monograph:

    Methods of Dendrochronology: Applications in the Environmental Sciences. International Institute for Applied Systems Analysis, Kluwer Academic Publishers, Boston, MA.

    The reality is that dendroclimatologists were largely left alone to do their experimental thing, isolated and “chronically underfunded”. Most were pretty aware of the speculative nature of the art (and sometimes science) of “reconstruction”, which is why they were publishing in fairly obscure journals. Trying to get the nuggets of science right before making a big deal of it all. That culture of modesty and austerity and cautious interpretation (the only true safeguard against statistically intractable problems) changed very quickly when an amibitious certain someone arrived on the scene in the late 1990s.

    So I do not believe the review that you seek has been written yet. It’s only really been a short time since the weaknesses in the science of dendroclimatology have really come to light.

  87. TCO
    Posted Aug 6, 2006 at 10:31 AM | Permalink | Reply

    I blame Douglas.

  88. bender
    Posted Aug 6, 2006 at 10:31 AM | Permalink | Reply

    As a side note, a search of the dendro forum list serv indicates that there has been systemic resistance to the idea of statistically robust methods of cross-dating (or wiggle-matching) tree-rings. The gatekeepers of the “science” suggest that visual methods of ring-matching are superior.

    However I saw a piece recently on television where they described how “hair-matching” has been disallowed in some jurisdictions as a basis for forensic evidence in criminal trials. In part this is because we now have statistically robust DNA-based screening methods. But the fact is: hair-matching (multiple pairwise comparison based on physical charateristics of hair strands) has proven to be junk science: sometimes it works, but the error rate is very high.

    Interestingly, the statistical problem of finding a matching pair given a huge volume of possibilites, using the human mind as processor, is not unlike over-fitting using Mannomatic/RegEm/neural net/genetic algorithms. Both are prone to generating false positives, and it is very hard to devise methods to guard against these errors.

    Woodhouse, C.A. 1999. Artificial neural networks and dendroclimatic reconstructions: An example from the Front Range, Colorado, USA. The Holocene 9(5): 521-529.

    Zhang, Q.B., Hebda, R.J., Zhang, Q.-J., Alfaro, R.I. 2000. Modeling tree-ring growth responses to climatic variables using artificial neural networks. Forest Science 46(2): 229-239.

    A final point. Note how much of the new methodological stuff is published in the non-peer reviewed “gray literature”:

    Guiot, J., Keller, T., Tessier, L. 1995. Relational databases in dendroclimatology and new non-linear methods to analyse the tree response to climate and pollution. In: S. Ohta, T. Fujii, N. Okada, M.K. Hughes, and D. Eckstein, eds., Tree Rings: From the Past to the Future. Proceedings of the International Workshop on Asian and Pacific Dendrochronology. Forestry and Forest Products Research Institute Scientific Meeting Report 1: 17-23.

    Keller, T., Edouard, J-L., Guibal, F., Guiot, J., Tessier, L. 1998. Atmospheric CO2 doubling impact on tree-growth in French Alps and French Mediterranean area. In: C. Urbinati and M. Carrer, eds., Dendrocronologia: una scienza per l’ambiente tra passato e presente. Atti del XXXIV Corso di Cultura in Ecologia, San Vito di Cadore, Italy, September 1-5, 1997. Dipartimento Territorio e Sistemi Agroforestali, Universita degli Studi di Padova, Padova, Italy: 55-64.

    Keller, T., Guiot, J., Tessier, L. 1998. The artificial neural networks: A new advance in response function calculation. In: C. Urbinati and M. Carrer, eds., Dendrocronologia: una scienza per l’ambiente tra passato e presente. Atti del XXXIV Corso di Cultura in Ecologia, San Vito di Cadore, Italy, September 1-5, 1997. Dipartimento Territorio e Sistemi Agroforestali, Universita degli Studi di Padova, Padova, Italy: 43-53.

  89. bender
    Posted Aug 6, 2006 at 10:34 AM | Permalink | Reply

    Dendrochronology: a terrific method in search of a hypothesis.

  90. TCO
    Posted Aug 6, 2006 at 10:41 AM | Permalink | Reply

    EE is also unacceptable gray literature.

  91. Steve McIntyre
    Posted Aug 6, 2006 at 11:20 AM | Permalink | Reply

    bender – last year I checked on the dating of the 11th century cores used in Briffa et al 1995 on the Polar Urals, purporting to show that 1032 was the coldest year of the millennium. The number of cores was below usual standards for replication and IMO I showed that they were misdated. I submitted a short note on this to the originating journal (Nature) on the basis that they were responsible and the note was rejected. This was one of the few notes that I submitted last year. TCO said that it was a waste to submit to Nature and that it should have been submitted to a dendro journal. Perhaps so, but I strongly feel that originating journals should have responsibility for their products and that it was necessary to submit the note to Nature.

    Subsequent to this, I got information on the Polar Urals Update and haven’t updated the note for re-submission. I should do so – among the 20 things I need to do.

  92. bender
    Posted Aug 6, 2006 at 11:38 AM | Permalink | Reply

    Re #91:
    Very interesting. Some day I will have some stories to tell you.

    P.S. It is impossible to keep up with the rate at which you read & dissect all these papers! I am hoping you are paid twice as much as me and have a large team of technicians & secretaries to support you. Lord knows how you do it.

  93. Steve McIntyre
    Posted Aug 6, 2006 at 12:41 PM | Permalink | Reply

    If you are paid half as much as me, you’d be making nothing as well.

  94. bender
    Posted Aug 6, 2006 at 12:44 PM | Permalink | Reply

    Dude, you need a research chair.

  95. fFreddy
    Posted Aug 6, 2006 at 3:13 PM | Permalink | Reply

    Ah, but the film rights should be worth a bob or two …

  96. James Lane
    Posted Aug 6, 2006 at 5:56 PM | Permalink | Reply

    Bristlecones in the news:

    http://tinyurl.com/ewpac

  97. bender
    Posted Nov 9, 2006 at 10:55 PM | Permalink | Reply

    Google: “elevated aridity and megadroughts” and search for the abstract by Ed Cook.

    Surprised no one has mentioned “megadroughts” here yet. Sounds like the kind of thing that could burn up bristlecone pines, thus forcing them out of the tree-ring record.

  98. bender
    Posted Nov 9, 2006 at 11:02 PM | Permalink | Reply

    The characteristics and likely causes of the Medieval megadroughts in North America

  99. bender
    Posted Nov 20, 2006 at 6:06 PM | Permalink | Reply

    Muse on “storage carbohydrates”.

    Further to #34. See:

    Horsley, S.B., Long, R.P., Bailey, S.W., Hallett, R.A., and Hall, T.J. 2000. Factors associated with the decline disease of sugar maple on the Allegheny Plateau. Can. J. For. Res. 30: 1365–1378.

    Horsley, S.B., Long, R.P., Bailey, S.W., Hallett, R.A., and Wargo, P.M. 2002. Health of Eastern North American Sugar Maple Forests and Factors Affecting Decline. Northern Journal of Applied Forestry 19: 34-44.

    Bourque, C., Cox, R., Allen, D., Arp, P., Meng, F. 2005. Spatial extent of winter thaw events in eastern North America: historical weather records in relation to yellow birch decline. Global Change Biology 11 (9): 1477-1492.

    Each finger “root storage carbohydrates” as the mechanism by which Mg/Mn nutrition, soil thaw-freeze events, and insect defoliation cause hardwood decline.

    One wonders whether bcp “storage carbohydrates” (roots or foliage) might be part of:
    -the 20th c. uptick in annual growth of strip-bark forms
    -the divergence problem (positive and negative responders)
    -drought-caused decline of high-elevation pines

  100. bender
    Posted Nov 22, 2006 at 11:12 PM | Permalink | Reply

    From Lloyd (1997, CJFR: 27, p. 940) on foxtails:

    “Past episodes of unfavorable climate led to the death of the most marginal individuals (those within approximately 100 m of tree line; Lloyd & Graumlich 1997), while leaving individuals short distances from the edge (100-200m) unaffected (this study).”

    Lloyd (Fig 2) shows peak per capita mortality occurring during the megadrought of AD1150, at a rate of 0.09/50 years. I note that death rates for “marginal” and less marginal trees would not need to differ by much to produce strong survivorship bias in the tree-ring record.

  101. Steve McIntyre
    Posted Nov 22, 2006 at 11:33 PM | Permalink | Reply

    bender, what was the original source of the Ho graphic shown above?

  102. bender
    Posted Nov 23, 2006 at 6:09 AM | Permalink | Reply

    I made it up myself, Steve.

One Trackback

  1. [...] CA was the first blog to discuss the recent papers by Wilmking and his associates on “positive and negative responders” – the opposite response of trees to recent warming in which – at the same site – some trees responded positively and some negatively. This is an important observation in trying to provide an explanation for the “divergence” problem – ring widths going down while temperatures go up. Previous discussions included:Wilmking in Alaska and Positive and Negative Responders. Other related posts include Upside Down Quadratic, Twisted Tree Heartrot Hill , bender on Gaspé Survivorship Bias . [...]

Post a Comment

Required fields are marked *

*
*

Follow

Get every new post delivered to your Inbox.

Join 2,897 other followers

%d bloggers like this: