A couple of days ago, I received the Almagre measurement data from the excellent dendrochronological lab at the Univesity of Guelph. I’ve posted the measurement data online in two files (as I received them): crossdated and uncrossdated. There are 41 cores that have been crossdated and 20 cores are not crossdated. These files are as I received them – in “Tucson format” – which is in a punch card format. I’ve re-collated them into an alternate format crossdated uncrossdated which will be easier to use for anyone other than dendros.
As all of you well know, I’m just working my way through this process to see what happens. Given that our cores worked from bark in, I would have thought that there would have been no particular difficulty in crossdating once you had done the measurements. It’s not as though these are timbers from a medieval pueblo that one is trying to place in a sequence. But dendros have their own ways of doing things. There’s a reason for why they do this – the methods are ones that have been used for a long time and are quite stable. These aren’t Mannian innovations to multivariate analysis. However, I don’t know whether there are any statistical implications in this process and I’m interested in seeing what the difference is between the crossdated and uncrossdated trees. I need to visit with the lab and get some more education on this. My guess is that they use a program called COFECHA which tests for a common signal.
From a statistical point of view, the step does seem intriguing and I’ll need to crosscheck with Pete Holzmann on the quality of each of the non-crossdated cores. If some or all of these were “good” cores i.e. there was at least a decent section before heartrot was encountered (and heartrot was more frequent in the cores than one would ever guess from the literature) and a priori there was no reason why the core could not be crossdated, one wonders whether the COFECHA process is introducing a false unanimity into the “signal”.
Information on the core locations in a Excel file in the relevant data directory http://climateaudit.info/data/colorado .
After I look through the data, I’ll forward the data to the ITRDB data bank, where I’m pretty sure that I’ll set a record for promptness of placing the data into a public archive.
However as we stand presently, the measurement data is now online here. IPCC AR4 Box 6.4 illustrates 8 proxies going back to the MWP of which 6 are tree ring based. Three of the data sets are completely unarchived: Taymyr, Yamal and the Luckman-Wilson Jasper measurements (over which Rob Wilson does not have jurisdiction); updates for two more are unarchived (Tornetrask, Sheep Mountain – the active ingredient of the Mann and Jones 2003 PC1). Only one (Jacoby’s Mongolia) is archived. Our meta-data is considerably more complete than for any of these sites. We have also posted pictures online. I will be receiving a DVD with detailed photos of the cores used for digitizing (which will be over 1 GB in size and will send a copy of the DVD, together with a DVD of tree photos, to ITRDB in case this sort of data ever becomes of interest to anyone.)
I need to do some analysis of these cores. After reading Mannian RegEM dreck for the past few days, I feel like I need to be cleansed.
Climate Audit 
191 Comments
Cleaner proxies will tell us the temperatures, dirtier ones the teleconnections. I hope I’m not belaboring the obvious.
======================================
Beware, tree bark may not be as pleasant of a substitute for soap as one might think.
However as we stand presently, the measurement data is now online here.
Steve, “missing a link” ?
Next time you wade through “dreck”, use one of these
Well I went looking for cherries (hockey sticks), not that anyone would do that with this sort of data, in crossdated:
1. Core 20A, Core 20B, and Core 21 show a bit of a hockey stick shape, starting up in 1930’s.
2. Core 33A and Core 33B show a rise in 1950 – 1970, but come back down in 2000
Otherwise not much hockey represented in crossdated.
John Norris, could you share with the rest of us a way to open these files?
How many climate scientists does it take to buy a latte and pull Steve’s data from ITRDB?
All joking aside … my hats off to you Steve! I find your blog fascinating from multiple angles (least of which as a comment on human nature). I have an honours Physics and Math degree … but … have spent the last 18 years building a software business. In order to provide quality product and service to our clients I demand “TVP” within my organization … Transparency, adherence to our corporate Values (two of which are honesty, and diligence), and Process. Successful science and business intersect on this.
Believe it or not this is my first post to any blog and this is new to me … so … please snip as you see fit.
Perhaps this has been answered elsewhere before, but I would really like to understand this better:
– What is the assumed model for average annual temperature vs tree ring thicknesses?
(Presumably there is an assumed function f so that Tavg_i = f(D_i)?)
– How can one test/validate such a model?
(I mean it would take quite some time if you had to wait…)
– What is the inaccuracy of the measured tree ring thicknesses?
– How many tree ring samples are typically taken from each tree? Each site?
– Is there a method for how these samples should be taken?
(Like “always 90 degrees angle toward the tree itself”, “always from the north/south/whatever side”, etc)
(And, how does one select suitable trees to take samples from?)
(How does one know that the sample is a good reprasentative of the individual tree?)
– How does one try eliminate the impact of other variables (such as precipitation)?
Is there a site or document that outlines these things?
I would appreciate if someone could point me in the right direction.
Thanks,
/Johan
ShauneS, Welcome! It’s always good when a lurker comes out from the shadows to make a contribution, especially one who is new to the idea of an interactive internet 😉
Does Steve have that bad of a snipping reputation? lol.
Steve, I just took a look at your location map: http://www.climateaudit.info/data/colorado/Almagre_Location.jpg
“Almagre” sounds so exotic for the south side of Pike’s Peak!
Looks like access is off the back road to Cripple? — the old RR grade, ims, which I always mean to drive sometime. I spent part of a season in the late 60’s drilling a uranium prospect in the CC basin — long before the casinos, when Cripple Crk was still something of hardship duty. I recall spending quite some time finding a way around the old hwy tunnel for an oversize drill rig…. Huh, guess I was younger than JEG then.
If you ever need 4×4 support or an assistant in northern AZ (winters) or northern NM (summers), just whistle. For that matter, Colo Spgs is just a half-day drive from our summer place…
Cheers — Pete Tillman
Johan i Kanada
http://www.climateaudit.org/?p=2183
Th referenced post gave me a considerable amount of basic information about this topic as did the subsequent entries and posts. By the way, if you do try to follow along, be prepared to spend some time, for our host and his commentators are an indefatigable lot with a prodigious (and illuminating) output.
Anthony Watts
Mann filter! I nearly lost my lunch, laughing. It’s even better than Mannomatic.
This is fascinating work Steve. I only wish there was more exchange between RealClimate and here so we could get more feedback about how this data, the Craig Loehle study, and your statistical assumptions appear to contradict many of RealClimate’s assumptions and their 2004 defense of the hockey stick. Unless I’m mistaken they are simply not addressing these new challenges to prevailing assumptions, and not addressing the scathing critiques by Wegman. Howe much is politics trumping science?
I remain very confused about how relevant tree ring studies are to the IPCC assumptions about AGW. Isn’t it correct that tree rings represent only a very small part of the body of evidence suggesting AGW is most of GW?
RE: snarky comments – these diminish the quality of this approach as “collaborate science” where experts can use online community to review, critique, exchange ideas. Calling people out on their bad assumptions is great – in fact it is a cornerstone of good science. Calling people names is not worthy of what is becoming a *really, really* interesting, science rich blog.
re #5 Tony Edwards
The .dat files are just ascii text files. I used notepad on Windows XP, but I think wordpad works as well. Once you get the data in a text editor, I cut and paste the whole lot of text into an excel spreadsheet (don’t tell Steve I did this as he will start preaching again about “r”). You can use the “Text to Columns” function under the “Data” pull down menu in excel to sort of get the data in reasonable columns.
RE #7
Cook, E.R. and Kairiukstis, L.A., (1990) Methods of dendrochronology: applications in the environmental sciences, Dordrecht: Kluwer Academic Publishers, 123-132
is an edited book that covers several of those questions.
I’ll put the files into an easier format. This is Tucson format left over from punch cards and how dendros deliver things.
Kim, thank you for your input. Please “belabor” away. I for one am constantly missing the obvious or otherwise need to be reminded of it.
#10: Tree rings are 100% essential to the hockey stick shaped results of any multi-proxy study that contains them, regardless of non-tree proxy series that are included. Given the numerous identified and potential problems with tree rings as temperature proxies (detailed in at least 30 posts on this site), it seems unwise to continue to use them.
Other proxies exist, but their utility as temperature proxies also remains to be seen. In addition, studies including them have been shown to cherry pick (including ice cores that show warming, excluding those that do not) and contain at least some questionable statistical methods. Steve has focused most of his time on tree ring proxies because they dominate the majority of temperature reconstructions.
I think you overestimate RC’s dedication to the science over advocacy. They are not interested in anything that weakens their advocacy, and certainly not if it is derived from here.
True, but after so many years of Team nonsense, it can no longer be helped. Perhaps you are new to the “debate” that has actually been raging for years (though really only recently getting noticed) so you are likewise unaware of what has transpired in the past. It is an ugly, deep, dark Rabbett-hole if you choose to enter it.
Mark
There are several posters in here from the Springs as well, myself included, and obviously MrPete (as well as Steve’s sister!).
Mark
Steve@13 – Could you also place column headers when you do that.
BTW #5 and #11:
You can import these files directly into excel by opening a blank excel spreadsheet and then opening the DAT files (make sure you specify the file type you are opening in the drop down menu for the open command). #11’s method is a bit convoluted, but still gets you there (just more slowly).
I apologize for my slowness in providing the rest of my end of the data — some detailed photographic evidence to help folks actually *see* the trees we’re talking about, including cores. (My laptop is currently being pressed into duty for showing the CU/Nebraska game 😉 )
Anyway, here are a few answers to questions raised:
# cores per tree:
We did our best to obtain at least two. All cores are identified by tree number and core letter. So 028A, 028B, 028C are three cores from a single tree.
Best practices are that more than core per tree should be obtained. Unfortunately, our experience so far is that many important series contain only one core per tree.
Almagre and Pikes Peak:
Almagre is most certainly its own mountain, with a nice valley between it and Pikes Peak. From the PP observation deck, look to the south east. You’ll see some nice lakes/reservoirs, and a mountain beyond — that mountain is Almagre, aka Baldy.
Pictures online:
Don’t tell anyone, but I’m in the process of integrating some panoramic images with an experimental new system out of CMU called GigaPan.org — search there for Almagre and you’ll find one or more of my panoramas. Over the next few days, there will be more. The first one is fully integrated into Google Earth, which is rather nice. We’ve also been asked whether we’d be willing to have the Almagre tree photos available as part of the mainstream Google Earth layer set — a rather nice idea, I think!
If anyone knows a service that integrates normal/flat photos directly into the new Google Earth photo system, I’m all ears.
Mr. Pete (#22)
There is a great way to geolocate your pictures. Go to http://www.panoramio.com and sign up for an account. You can then upload images, and then map them into Google Earth. There is a way, although I have not tried it yet, to set layers to where you could actually include the dendro data as an additional layer.
With the images in Google Earth, you can just provide the location or coordinates and people can zoom to them and probably get a nice look at the mountain as well. I just checked and the resolution is such that you can see individual trees.
I know about Panoramio. Unfortunately, it uses the old (html) embedding — which puts photos in a separate box. In the current Google Earth, photos literally can integrate into the landscape, facing a particular direction, at a particular height, etc. 🙂
#9
Thanks, I took a look at that entry and it mentions:
“the Team hypothesis of a positive linear relationship between temperature and ring widths”.
Is this really the model all these guys use? Ti = kDi+l?
Are there any biological or physical evidence supporting this model?
What will Steve et al use when analyzing the core data discussed here?
Thanks,
/Johan
RE 22. Mr Pete. Glad you like panoramics!!! 4 years ago I made a product that accelerated the stitching
in panoramics. I guess we were ahead of our time.
http://www.dpreview.com/news/0305/03052001creativecard.asp
The panoramic algorithms were very cool and it was fun to see them sped up by graphics Hardware.
Fun didnt make too much money though. Opps.
ha, maybe today, 4 years later it would be a better idea.
Addendum to #4, wrt crossdated
11A is a bit of a HS as well, although mostly blade, very little handle.
Panoramas are MUCH better today. (Next version of hugin.sourceforge.net automagically corrects exposure and vignetting so photos taken with a simple camera easily can be formed into a detailed panorama… that’s what I’m using.)
John – HockeyStickness is meaningless in these cores without reference to other cores intra- and inter- tree. I think what you are seeing is additional evidence that “hockey sticks” have little relationship to climate.
Am I pushing my lack of credentials to far to suggest that:
* a single tree typically encounters a single climate history
* when a tree’s cores are widely varying, it is plausible that variations are not related to climate
?
This is just fascinating. It’s a wonderful look into an area of science I know next to nothing about. I’d imagine that very few scientists would permit “looking over their shoulder” in this fashion. Very much looking forward to the final results.
Probably someplace the units of measurement should be recorded in the data file.
Do you mean cores from “trees,” plural, and there is wide variance in nearby treed?
BTW, weeks back Steve suggested you had a 19th century Colorado strip bark theory. Have you posted it?
re 28: MrPete
I wouldn’t at all be surprised. I am very interested to see what Steve and others here do with this data.
By the way, thanks for your efforts collecting this data.
ftm asked “do you mean cores from “trees,” plural?
That would be reasonable, as we do see wide variance there.
However, we also are seeing wide variance in multiple cores from a single tree.
Certainly, such results should give pause. From the outside I’ve been curious about how much consistency is found in cores. So we took lots of cores. Supposedly a waste of time. Now not so certain our curiousity was wasted.
As to theories… will comment later. I think some photos will be helpful illustrations for several discussions about the value or lack thereof of strip bark. I really do need to avoid spending time commenting… I got my laptop back (CU-Neb 65-51!!) so I’m back to photo processing… 😉
See y’all later…
Remembering that I am new here and new to this blog and debate, I am interested in what Steve M is doing. Is there a post elsewhere that gives an overview of this project? I thought Steve and CA folks were against the use of tree ring proxies, or is it just the BCP proxies?
I just finished re-reading the NAS Panel report, esp the section on tree rings. From what I understood, there are a number of uncertainties in tree ring proxies, and a few are more likely to show CO2 fertilization and moisture conditions rather than surface temperature and so should not be used in reconstructions. In particular, “strip-bark” trees — the bristlecone pine. Otherwise, the NAS Panel report seems to support the use of tree ring proxies for temperature reconstruction. My question is this, and apologies if I should have RTFR, but are there any proxy reconstructions that use tree rings but not the BCP?
Steve, please let us know how your Starbucks Hypothesis P&L is looking. The tip jar rang off the hook to help you and MrPete defray the costs when you first announced the project, and if you’re still in the red, I’m sure I speak for many here who will gladly drop in more funds to close the gap.
Susann #33, see A Little Secret for the first post on Steve & MrPete’s summer vacation. A CA search for “Almagre” will find a couple of subsequent ones, but this will get you started.
Welcome aboard.
Susann, try the FAQ as a starting point…
Steve: The FAQ is very obsolete as it was written for MM2005 before the blog started,
Do a search on CENSORED, which is the original Mann recon without BCPs.
Personally, until someone can identify all non-linearities and work out how to deal with correlated inputs, I will not believe anyone that says tree-rings are suitable as a proxy for temperature, at least not using linear methods such as PCA or RegEM.
Mark
#33. Bristlecones/foxtails are used in MBH, Esper et al 2002 (twice), Mann and Jones 2003, Crowley and Lowery 2000 (twice), Moberg (but not in a version or in a way that makes a lot of difference), OSborn and Briffa 2006 (twice), Hegerl et al (2006) (twice), Juckes et al 2007 (twice). Rutherford et al 2005 is the same network as MBH98 as is Mann et al 2007.
There is another variation in which only 6 proxies are used, where there’s another piece of sleight of hand. Briffa et al 1995 argued that the 1032 was the coldest year of the millennium in the Polar Urals based on only 3 poorly dated cores. In 1998, more medieval cores became available showing a very warm MWP (as Shiyatov had observed in 1995). Briffa did not report the updated Polar Urals results. Instead he did his own calibration of Yamal data which differed from Hantemirov’s and which had a huge HS. Briffa 2000 had elevated 20th relative to MWP (but doesn’t with Polar Urals Update.) D’Arrigo et al 2006 has virtually the identical medieval network and the same sensitivity.
Briffa et al 2001 does not go back to the MWP and has a divergence problem where Briffa truncates the series in 1960 rather than showing its post-1960 failure. This truncation was done for the first time in IPCC TAR in a highly deceptive way that was first commented on here.
Almagre Data thread, Re: #41, M.Jeff, November 23rd, 2007 at 7:41 pm
Agreed and your pasting was much better than mine.
#39 — yikes! I just read about “censored”. It gives me the shivers.
I’m now scared to ask, but is the whole issue of “teleconnections” (which I think is probably a valid concept, although perhaps invalidly applied) used to explain or justify keeping certain tree rings in the mix? I have not read closely enough or understood well enough the most recent Mann and other recon papers — there is just too much material here to go through. I am still working through the issues raised in MBH99 and MM and the two committee reviews. I’ve read them once or twice but I haven’t done a close study yet, as in taking notes and reflecting. At this point, I’m trying to slide into this whole new world and get my sea legs.
Well … a first look at the Almagre Treemometer is not encouraging.
I took a look at the crossdated data. Here’s the two cores from the very first tree in the record:
I must admit … not what I expected. It looks like two different trees, rather than two cores from one tree. Some random thoughts:
1. Temperatures dropped considerably from 1865 to 1915 … or not.
2, Temperatures rose strongly from 1915 to 1945 … or not.
3. Temperatures rose markedly from 1950 to 1960, then dropped a similar amount to 1970 … or not.
4. I can certainly see why the treemometricians want to choose between trees, selecting those that are ‘responsive to temperature’. Heck, I want to do it just based on two cores, much less two trees.
5. R^2 of the two cores from Tree 04 is only 0.30 … not encouraging.
It gives an appearance as though one side of the tree is subject to some influences that don’t affect the other side. Or perhaps one side is just more ‘responsive to temperature’ in general. Now, I could live with that. It would mean that one set of tree ring widths is a perhaps complex but sign-preserving transformation of the other. In other words, instead of one being a linear transformation of the other of the form:
Y = m x + b + error,
the constant “m” is replaced with a variable V(t) which varies over time, so that:
Y = V(t) x + b + error
This means that sometimes the wider ring could be twice the width of the thinner ring, while in other times it might be say 1.1 times the thinner.
Now, there is a way to test whether this is the case. If both of them are responding to the same stimulus, even if the size of the response changes over time, both of them should increase or decrease together. They will increase or decrease by different amounts, but they will move in the same direction. One will get a little wider, and the other will get a lot wider but if they are responding to the same stimulus, they should both either increase or decrease over the preceding year.
And this is easy to test. We merely take the first difference of the data, and we see how often the signs of the data disagree. In the case of Almagre Tree 04, one width increased when the other decreased an amazing 44% of the time … which means the directions of indicated change (warmer or cooler than the previous year) only agree a little over half the time (56%)
And assuming a binomial distribution with a probability of 0.5 as our null hypothesis, 56% right out of 147 trials doesn’t even attain statistical significance, so we can’t exclude the null hypothisis of no connection between the widths of the rings. There may be a common signal in there … but it is so buried in the noise that even the direction of indicated change, not the size but the direction, cannot be established on an annual basis.
CONCLUSIONS
1. Core B in this treemometer obviously has a manufacturing flaw. Core A reveals the human-influenced temperature rise from the time of the industrial revolution. Since it is ‘responsive to temperature’ and core B is not, disregard Core B in future reconstructions.
No, but seriously, folks … if a tree doesn’t even agree with itself half the time, will averaging a thousand such trees reveal a signal?
My best to everyone,
w.
A stupid question, perhaps, but what sides of the tree did you take the cores? I mean with respect to facing north, south, etc. Isn’t it the case that tree rings will be thicker on one side vs. the other based on orientation to the sun and other environmental factors. I’ve read thicker on north side, thinner on south.
I believe that you need to toss in prevailing wind direction, slope gradient and obvious disparity of nutrient availability (rockiness) as well. Not that any of those factors are discernible in published data but they might turn up in field notes – perhaps as indicators for selecting the “right” type of treemometer.
Directions, angles, etc are all recorded in the provenance data spreadsheet.
MrPete,
That would be this spreadsheet, correct? What does “Tree Cover” mean?
A comment on correlation between cores. All one has to do is look at the cross section of a log. A log from any tree. How many perfect exactly concentric circles do you see? None? Gee whiz! Then compare the cross sections every 2 feet or so, as it is cut into firewood. Sure, there is some similarity, but not enough to base useful predictions or analysis on. They have to declare the debate settled, as it is all smoke and mirrors anyway, and obviously none of them ever expected outside critical review of their work. They certainly were not prepared for it in any event. All these splinter (unintended pun) professional groups are incestuous by their very nature.
Go Steve! I wish I had been a better student in statistics, but back in my day we still used Roman numerals.
#45 Willis Eschenbach
Isn’t it a wee bit early to reach a conclusion based on a sample of one tree?
I have no real conclusion from one tree, other than … it was an inauspicious beginning, and illustrative of the difficulties in extracting a temperature signal from ring widths.
w.
Dear Steve, you can check a typical COFECHA output here
http://www.ncdc.noaa.gov/paleo/treering/cofecha/speciesdata.html
best
Willis’s point in #45 is valid. The (+) deviations in the red curve that you see in the 60s and 80s are the kind of growth spurt that led to the HS in Graybill chronologies. So, shall we hypothesize that one half of the tree is responding to a global temperature teleconnection, while the other half of the tree is not? #54 chides #45, however I’m willing to bet that additional samples will show similar within-tree disparity. Worth noting that dendros are aware of this within-tree variation problem to some degree. That’s why they take more than one core. But the assumption is that the average is of the two is representative. The questions is: representative of what? Averaging one non-stationary series with one stationary series gives you a half-stationary series. Terrific. What does it signify?
#58, Bender:
Well, what if the underlying root structure is anisotropic ? Perhaps the roots provide H2O to the cambium during very wet or dry years better between 20-45 degrees E of N than between 45-70 degrees E of N. It could be that as you sample around the tree, certain portions of the tree could be more moisture-limited than others and correspondingly unable to respond fully to favorable growing temperatures.
If this devil’s hypothesis were true, wouldn’t averaging a ‘good’ and ‘bad’ core represent half (?) of the (theoretical) effect of underwatering on the best climate signal the tree was able to capture ?
It would be interesting to compare a series of graphs as Willis plotted above sorted by degrees away from the nearest major root.
Rick Ballard – yes, that spreadsheet. Look in the first (“Data Dictionary”) tab for definitions of all columns.
Hopefully Steve will have time soon to upload the current update of the file, which has some more columns and updated info based on trip #4.
With respect to hypotheses about variant growth in a tree, this is where Leslie’s expertise comes in. I shared Willis’ note with her. Her eye-twinkling response: there are an incredible number and variety of possible confounding factors. Resolving them historically is likely quite difficult, yet perhaps there are some patterns that could be discerned. Here are a few from Leslie that might be interesting:
* Anisotropic antler-rubbing. An animal rubs its antlers on part of the bark, compressing cambium and slowing growth.
[Important background notes for the rest:
1) Conifers tend to grow in a spiral, observed to be anywhere from 0 to (almost 360?) degrees per meter. Does that extend into the root system? Perhaps a reader knows.
2) Root systems pull nutrients primarily from the approximate “drip line” radius out to twice the drip line radius; roots are known to extend as far as 5x the drip line radius. (Inside the drip line, roots tend to be more structural in nature.)]
* Anisotropic competition aka tree crowding: trees growing and dying, particularly within 2x drip line radius. We note that over time, trees tend to thin out due to various depredations. (The surviving trees also sustain damage from falling tree trunks.)
* Anisotropic soil structure (one source of Jim Edward’s comment): roots grow differently in various kinds of soils
* Anisotropic nutrient availability (another source of Jim’s comment): nutrient pockets can exist in various places for various reasons. This can affect tree growth both chronologically and “angularly.” (Must be an appropriate vocabulary word for that ;))
* Of course, there are prevailing winds, slope, and more.
* And then the whole thing is not-quite-randomized by spiral growth. (Very careful measurement could resolve this.)
Finally (for now), remember that overall tree growth tends to be precipitation-limited more than anything else.
“best climate signal the tree was able to capture” — I wonder if the trees have read the instruction book so they’d know their purpose, eh? 😉 [Reminds me of going scuba diving and being told “don’t worry about the Barracudas, this species won’t hurt you” — and wondering if Uncle Barry is following his instructions!]
Very interesting plot from # 45 Willis, but what would also be very interesting would be for a plot of average spring/summer temperature for the nearest weather station, as well as average precipitation for the same period. Surely this would really show whether there is any temperature connection without needing any statistical manipulations. And, given the relatively low visual connection between these two cores, I’m not holding my breath for a successful match to temperature alone
I archived the files as I received them – in “Tucson format” – which is in a punch card format. I’ve also re-collated them into an alternate format crossdated uncrossdated which will be easier to use for anyone other than dendros.
#60,
One of the most distinctive features of the strip bark – observed in a post about a month ago – was the big growth in the center and micro growth on the edges. If you had some wobble in where the “center” was, you could have things look like growth pulses in an individual core, when what you’re doing is sampling a wobble from an edge to a center and back to an edge. IF conifers grow in a spiral, if you confound this with strip bark, would this be possible.
Regardless, looking at this data with fresh and hopefully statistical eyes, the BIG problem is how do you rationally model these growth pulses. I’m taking the position that the FORM of the model is a longitudinal growth model, such as you in health or drug trials. But the problem here is hugely confounded by these growth pulses that are wildly disproportionate between tree and tree and seemingly even between cores within a tree.
If the pulses are not climatically generated and you don’t have a model to adjust for them, the size of the pulses is so enormous that any slight extra population of pulsed cores will dominate the chronology. What a bizarre way to measure world temperature.
re: #60
Does Leslie know if conifers have a prefered handedness?
53: Steve:
I’m not sure about bristlecone pine, but while spiral growth is sometimes found in commercially used conifers, it is not normal.
Following on from #45, I thought maybe the ratio of the two cores might reflect the changes in the local environment of this tree, so I graphed the ratio, and tried to fit a series of step changes to it, with reasonable success. Green is the 3 year mean of this average, I assume these small variations to be noise induced by the measurement process. Red is my ‘direction of preferred growth’ metric and the step changes are at the bottom. The changes do seem to average out over time, but it does appear that whilst the ratio can change suddenly, any change is likely to be persistent.
Quite how this varying ratio can be successfully be applied to both series together to resolve the underlying whole tree signal is beyond me at the moment. Maybe the 3 core trees would shed some light, or the presumed correlation between the whole sample could be used. Still not sure this proves anything other than trees grow.
#52
Is there a problem with the alternate format data set? Some of the values for core 4A seem to be different (out by 100) in the 1990s and 2000s.
Try #6.. Am I reading the table correctly and that tree has 4 cores b,c,d,w?
if so, it’s wild
Steve: The tree had 3 stems and we took one core from each stem. It also had a very large exposed root and just for fun we took a sample of that too. Core 06W is actually Core 06A in Pete’s table – its mistransliterated at the lab.
RE 58. the reformat does not square with Willis’s chart. Maybe this explains why I got very weird
results for tree 6. I wont post till it’s cleared up
Steve: In my collation, I relabeled 06W to be 06A.
Let me check. I did the re-format using a one-off script and I may have introduced an error.
Re #53
The tree is trying to close the wound that is the “strip” of the stripbark. When the wound is small this happens quickly, with no growth distortion on the side of the tree opposite the wound. Maybe when the wound is very large growth hormone concentrations are higher in the center of the intact bark opposite the wound than on either edge. Sort of a perverse allocation of effort in growth.
Re #56
Episodic dieback of cambium and/or roots, probably in specific drought years, that’s my guess.
#60. Checked and did a slight edit. Left over from Fortran days, dendros put a trailer on each tree – 999 in this case. I had to slightly modify my collation program to keep track of segments and in this one-off run, I hadn’t NAed the trailers; I’ve fixed and it should be clean now.
bender, if one is trying to model the growth of a tree, do you have any idea how to specify these sort of wound effects? One has
RW = a+b*Temperature+ c* Precipitation+ d*Wound Effect+….+ interactions,… where Wound Effect seems to vary depending on the position of the core and how far through the pulse it is.
#59. Tree 06 should look something like this. The very strong increase in the root (blue) is quite intriguing. I’ve never seen any discussion of root ring widths before. It just seemed like an interesting experiment.
I don’t know if this has been posted somewhere else, but I found an R package that includes a function to convert the Tuscon formatted files into a more useful format.
http://cged.genes.nig.ac.jp/RGM2/pkg.php?p=dplR
The specific function is here and one can preview the source code at that page:
http://cged.genes.nig.ac.jp/RGM2/R_current/library/dplR/man/read.rwl.html
It seems to successfully process the original SMCD.dat and SMIM.dat files as is and returns a data.frame with the years for rows and cores for columns. I spot checked a few cores/years to see if the data was in the right place and it seems to be.
And for those (like me) who didn’t know how to read those files, this was a good primer: ftp://ftp.ncdc.noaa.gov/pub/data/paleo/treering/treeinfo.txt
re #65
Something is not quite right here. The 6A (6W) data that I downloaded started in 1638.
6W 1638:2000
6B 1578:2000
6C 1821:2000
6D 1771:2000
I think your 6A is not really 6A/6W.
re #67
Oops, I meant –
6W 1638:2007
6B 1578:2007
6C 1821:2007
6D 1771:2007
re #65
Note that there is a 5B and a 5b in the data file. I think the 6A/6W in the chart above is really 5B.
Steve: I did a manual change on the levels and maybe I changed the wrong id. I’ll check. Arrgh. You’re right. When I was allocating segments, I altered the orders. Fixed.
wood volume.
The tree knows where to put its wood to survive and thrive. thick on one side. thin on the other.
Reverses sometimes.
The trees reasoning was sound. it survived. Ring width is the wrong measure. Total wood added.
how to get from one to the other?
Tree #6 is a hoot.
wood volume.
The tree knows where to put its wood to survive and thrive. thick on one side. thin on the other.
Reverses sometimes.
The trees reasoning was sound. it survived. Ring width is the wrong measure. Total wood added.
how to get from one to the other?
Tree #6 is a hoot.
RE 65. yes that is what I got. reran with the ammended data.
Made me think about Total wood formed in a given year.
Steve: Run it again. the file is amended.
Re 44, George M on correlation between cores,
Just before Steve let us into his little secret, I took advantage of a Wellingtonia that a neighbour had taken out to look at how uniform the rings are and was astonished at how poor they are. I have a slice from some way up and photographed the stump. In numerous places the rings vary by 2 to 1 over just a few degrees of rotation. One reason for this that I can see is that branches can get removed and over time the tree grows over leaving no visible sign on the outside but considerable disturbance above and below them.
What is needed is for someone to section some typical whole trees and collect all the ring width data at a range of heights and directions and determine to what extent a single core is likely to contain meaningful data. Looks like a good undergrad project.
I guess you mean a core like the one below. It looks like the tree equivalent of i.i.d.
Craig Brunstein has devoted years to studying the forms of bristlecone pine near Almagre. See here (12.8 MB). Pete actually located a couple of Brunstein tags, but we don’t know which trees in his book that they refer to. Contortions and distortions of bristlecone are legendary. They are about as irregular as they get. Here are some of Brunstein’s section diagrams.
Ok, I am still trying to figure this out. The cores form 04a and 04B from SMCD.dat are labelled the opposite way in SMCD_ascii.dat. Which file is correct?
re 65, 69, 72
Steve,
Ok, your tree number 06 chart now looks like the chart I made yesterday from the original, unfriendly ascii file. 6A/W looks like it starts about 1638 and the first two values are 43 and 66.
However, I think the link at the top of the thread is not to your updated alternate format file. The 6A/W data looks like it is marked as 8B. The first two values for 8B, in the alternate format file, start at 1638 and are 43 and 66. In the original file I have 8B starting in 1602 and the first two values are 47 and 37.
Steve: It looks OK to me. I refreshed just in case I copied a stale file from my console. CAn you refresh and see if you still have the problem.
re 77
Looks good now. Sorry for the pestering.
Steve: I appreciate it. I used a one off script for this and am glad to fix things.
The ring width data from the file SMCD.dat has been plotted separately for each tree (with multiple cores from the same tree overlayed in the same graph). The result can be found here.
RE 79.
RomanM. have a look at the dustbowl era data for all cores. Say 1930-1940.
A cursory glance at 15+ cores ( 1932-1935) showed some interesting consistency
A nearby weather station is Canon I believe… more later.
I was thinking of the following. Briffa and other look for signals of volcanoes in
Tree rings.
Can we also look for big signal swings in tree rings. That is, in a given year ( say 1934)
the surfac temp swung by 1.5C can a tree detect such a swing? what’s its time lag as a sensor..
Is it a temp bandpass filter..
Is a tree arguably like the ocean in it’s filtering of high freq temp variations..
Might be neat to take at look at the local temp record and figure the time constant of
tree response..
Detrending question.
St. Mac. My undertstanding is that the rings ALSO hav to go through a detrending.
Since in their youth they put on wood like crazy. Some kinda negative exponential
detrending.. How does that step work or should I just RTFM
David, Steve:
Thanks for the confirmation (73, 74, 75). I’ve never tried to do a scientific analysis, but when you cut a cord or so of firewood (Texas Live Oaks) every year, you tend to notice things like lack of uniformity in the rings. I’ve always assumed these dendrochronologists or dendroclimatologists or whatever they style themselves as knew what they were doing. Brings me back to a set of advice from a friend:
Assume Nothing, Trust No One, Don’t Do Dumb Things
Advice they seemingly ignore.
#80 stevenmosher
I replotted all the cores during the time period 1900 to present. This makes it easier to view the 30s (looked like a time of somewhat increasing ring widths) and it puts all of the plots on the same time scale. The plot is avaiable here. If I get a chance I’ll maybe look at the local temperature record tomorrow. Gotta make my other half happy this evening.
#83
If you pick the right cores you can confirm any theory you want to prove.
# 81
That dplR package I mentioned (#66) also contains a detrending function that allows various options, including a negative exponential. And I think the chron function in the same package calculates a summary series of all the detrended series, but I haven’t gone through the source code for the functions to see what’s actually happening.
Here’s the CRAN link as well, which has some documentation: http://cran.r-project.org/src/contrib/Descriptions/dplR.html
A few pointers about tree-ring data:
First thing is to check the crossdating and measuring quality using program COFECHA (available from http://web.utk.edu/~grissino/software.htm). Crossdating is a critical step in the effort; one wants to make sure that one is evaluating annual rings with each other when developing a chronology. Five-needled pine spp in western US tend to be very complacent, i.e., not a lot of variability in total ring widths, which tends to reduce the in-common patterns. Run the Almagre data set with 100 yr segments overlapped by 50 yrs (option 2 in the menu) which will tend to see more in-common patterns in the data than the default 50 yrs overlapped by 25. By doing so, one will end up with less flags, low correlations that suggest there may be dating or measuring errors in the segments. (By the way, the measurement series in these files are given with a precision of 0.01 mm.) Overall interseries correlation of 0.52 in the Almagre series is actually very good compared to many 5-needle pine collections. There is only one segment with a flag, core segment 23Ab, which since it is only 23 yrs long should be dropped from the chronology anyway.
One can also run the undated series through COFECHA at the same time as the dated series. Doing so suggests that some of the series in the undated could have been included in the dated series: for example, core 47A has an overall series correlation of 0.65 with the master series.
There is, of course, readily available routines to convert ring-width measurements to vertical lists (year, value) for importing into spreadsheets or other plotting programs (try the Google sometime). Two of these are FMT and YUX, both available through the link above.
Once one has absolute confidence in the crossdating of ones ring-width measurements (which, by the way, are only one type of digital data available from tree rings), the next is to understand what climate and environmental conditions affected growth. Obviously, as the discussion here has already concluded, there are variations both within and between trees. Ring width for any one year is a function of multiple environmental, biophysical, and climatic factors. Environmental factors include disturbances (both to individual trees – e.g., lightning strikes, branch loss by wind, endemic insect defoliation – and those that may have affected an entire stand – e.g., fire, insect outbreaks). Biophysical factors may include numerous non-stationary growth processes (variations in above-ground and below-ground carbon allocation during the life of a tree, masting, cambial dieback [strip barking], etc) but the major one is the simple fact that radial growth rings decline in at least full-bark trees due to geometric constraints as the tree grows larger. Consider that once a tree has reached some sort of canopy dominance (i.e., mature), leaf area remains more-or-less the same from year to year, which produces more-or-less the same volume of wood per year. However, since this same volume is being added to an ever-increasing circumference (again, at least on a full-barked tree), radial ring width must decline. Hence, in open-grown stands of trees, often this looks like a deterministic negative-exponential function. And there can be great variations in circuit uniformity both within and between species, undoubtedly related both to genotypes as well as variations in growing conditions between individuals (e.g., a trees growing on slopes, or bent by earth movements during a trees lifetime).
Dendroclimatologists often mention signal vs. noise; climate is the signal of interest and is assumed to be the most commonly expressed patterns in growth, both annually and over longer time scales, but these noises must first be defined and removed. This step is accomplished through standardization – also called detrending – of the ring-width (or other digital) data. The common method for this is the program ARSTAN (again available through the above link), which stands for auto-regressive standardization. In general, there is a great effort to retain as much of the low-frequency (multi-annual) variation in the individual ring-width series as possible, so conservative detrending methods tend to be used. One common method is use of cubic smoothing splines, which can be tuned to retain or remove low-frequency information in the series. ARSTAN detrends the individual ring-width data and makes an average chronology from all.
Once one has a chronology, the next step is to understand what climate variations the tree has been responding to at least during the period of instrumental records. I have discussed this step before here: http://www.climateaudit.org/?p=1310. Please refer there for further thoughts on the issue.
Bottom line is the ring-width data are only the first step in the process.
BTW – and this is a most important point for perhaps every bit of discussion on this entire blog – separation of signal vs noise is very much dependent on one of the most fundamental statistical principles, the law of large numbers. One doesn’t begin by looking for a climate response in only one core, or on tree, or even only one chronology. To look for an inter-hemispheric signal, one looks at inter-hemispheric data. One looks for the consilience of multiple lines of evidence from multiple disciplines that all point to a single, logical conclusion; geez, kind of like global warming in general, eh?
PMB
Steve:I’ve spent quite a bit of time looking at the ARSTAN algorithm a couple of years ago and may post on it some time. I wrote an emulation of ARSTAN, which I’ll post up on sometime and have calculated a number of chronologies. PB says:
This may be a “common method” but I know of no occasions in which such a method is used in the series in common use in multiproxy reconstructions. Briffa’s recons use “RCS” which is a fit of one curve to all trees. Jacoby uses what he calls “conservative” detrending, which means using the ARSTAN option in which a negative exponential is fit or a straight line. The following routine emulates ARSTAN coefficients for this procedure. In the ARSTAN version that I downloaded 3 years ago now (and I haven’t verified this), the coefficients of the first tree were always calculated wrong in ARSTAN. I don’t suppose most dendro consumers checked this, but it’s so. Some differences also occur depending on the settings for the number of iterations permitted in the nonlinear solver. The differences don’t “matter” much other than being of numerical analysis interest. The code here is 3 years old and I’d probably write it a bit better now.
PB cites the statistical principle that “separation of signal vs noise is very much dependent on one of the most fundamental statistical principles, the law of large numbers”. I dare say that I’m as familiar with statistical principles as PB and don’t particularly require a first-year lecture on the central limit theorem. I’ll stick these measurements into ARSTAN and into my own emulation in the next few days. I know that an answer will come out of the meatgrinder. But what is the statistical meaning? The noise model is very puzzling. The pulses in these bristlecones are highly autocorrelated and the size of the pulses is very irregular. ARSTAN assumes that it falls out in the wash, but, if you’re going to do proper statistics, and show that the conditions of the CLT apply, then one needs to examine what ARSTAN does and doesn’t do and perhaps that will be a theme over the next few weeks.
Steve McIntyre;
My wife and I were more than happy to help out with the costs for your tree core analysis.
Please continue the actual science and sample the mud in the lakes closest to the trees that you actually cored.
If there was an increase of water available to the trees, then this would be reflected in the depth of the historical deposits of the local lakes.
If there was an actual alteration of neutrients that could promote tree growth, then a simple chemical analysis of the lake mud should also show this.
I want to know why we have peaks in the data, that nobody can explain!
RE 86. Like Moshpit said (#70). volume of wood.
Ill bet a dollar to a doughnut that growth rates for the bcps, as well as many other series in the West, are really controlled by precipitation.
B I N G O
Why do I have to ask the most obvous of questions? If the tree growth rings are be influenced by available water or neutrients, then the nearest lake will show the exact same thing in it’s mud deposits.
Sometimes, I want to ring the necks of some scientists. THINK!
As noted in an earlier posting, the local Ag-Extension Master Gardeners all got a real laugh out of the idea that dendro growth might be temp-limited. To them, it’s pretty obvious, bordering on “why bother checking” that growth is precip limited — here and in most places. Ah well…
BTW, Steve H, sampling lake mud around here is not a simple task. We don’t have much in the way of natural lakes, period. What remains are reservoirs — mostly off limits 😦
It’s an interesting idea however! Perhaps geologist-readers would have insights on sampling of other layered-buildups…
The problem is there is no way to accurately guess which jumps in rw are a result of post-wounding internal physiology (“noise”) vs. T*P*N*C*etc (proxy “signal”). What do you use as the independent variable in your model? What I would recommend is using a valid physiological model of drought-driven cambium dieback and use that to correct the rw signal. But that could take years to develop. Unless one exists already? Maybe read up on “xylogenesis”.
One approach that would be fun as a first cut would be to use the dendro program JOLTS to detect sudden non-stationarities in bcp growth. The results could prove to be very ironic. If one core in a tree shows the odd “JOLT” and the other core tends not to, what would that tell you? Actually, such an analysis might even be sufficiently novel to be publishable.
Steve McIntyre,
22 and 23 are junk…
# 19
Steve H,
Besides solar irradiance, fog, terrain gradient, winds, soil composition and temperature, I would consider also phytopathogens, plagues and physiological aging processes, especially through drought periods. We can find tracks of them in the samples. If a sample shows tracks or signals having been affected, that sample must be rejected because the growth of that tree could be affected by hormones and protective chemicals that the affected tree produced during the attack.
In dendrochronology and dendroclimatology the things are not as easy as they seem to be because the researcher works with the assumption that the variations of the conditions at present must have been the same or nearly the same in the past. We cannot use the same calibrations for the trees of the NH and the trees of the SH because there are substantial differences in the composition of soils, the kind of biomes, the subspecies, etc. Perhaps, Manniacs used the same weighs for US Western, US Southern, Europe, China, etc. individuals of Pinus sp?
I think the last conundrum could be revealed on this audit.
RE 88
Wound injury repair is allometric growth,The rate of metabolic change is greater so energy has to be transformed from “reserves”
At least I got you to think about all of the other ways that tree growth can be altered, besides only temperature.
I honestly wanted people to attack my simplistic statements, for a very valid reason.
Monthly rainfall records for about 100 years should be available from data from the notorious USHCN sites. For the previous 1900 years, you are on your own.
Comment on referencing post numbers. I suppose Steve is deleting occasional posts, and it makes some of the hanging references to post numbers nonsensical. Thus, it is probably a good idea to also include the name. The number will get you close, and the name the remainder. Sure, occasionally there are multiple sequential posts from the same author, but anyone who can’t get over that obstacle, well…………..
We cannot use the same calibrations for the trees of the NH and the trees of the SH because there are substantial differences in the composition of soils, the kind of biomes, the subspecies, etc.
Then it is about time that we start to compare Apples with Apples for a change.
The last time I looked at the area where Steve and his friends sampled some trees, there was a very large lake.
Sample the damn lake mud!
Steve: It’s the Colorado Springs reservoir. Talk to them.
How do I say this without insulting anyone?
I strongly suspect that there was a natural forest fire aaround 150 years ago at that location.
The lake mud deposits will show an increase of carbon.
Steve, the key is in the mud deposits of the lake, which is located only a few killometers away from the trees that you sampled.
Nasif Nahle:
True or false?
Would lake mud deposits reflect the chemical and physical variabilities of the local environment, where the trees were growing?
# 27
Steve H,
I think… Well, I think that it wasn’t the primordial objective of Steve McIntyre and colleagues on sampling the site. I believe that sampling the mud could be a second step in the process of investigation.
Steve: The main objective was to prove the Starbucks Hypothesis and show that the proxies could be updated despite the fatuous excuses of Mann and Eil Rabett. And secondly to show that the data could be made available on a prompt basis without peekingto see whether it was advantageous or disadvantageous to any point of view.
I would think yes, but you aren’t going to get in to do this sort of testing in a reservoir. Steve noted that above.
Mark
Steve: Its the Colorado Springs reservoir. Talk to them.
OK, I will try to do that. It may actually happen, when you know who to contact.
# 101
Mark T,
I know, but Steve H insists on that. Additionally, the investigators must count on advanced techniques for the measurement of environmental stable isotopes, ionic mass spectrometry, etc.
Believe me, it’s all an odyssey; starting by the special equipment required to take the samples. It is not an impossible task, if it is made in Antarctica, it could be made anywhere on the world, but it is a very expensive research.
Thanks, Steve McIntyre, for claryfing the point (# 100).
Continuing on from RomanM’s charts at #79 and 83, here is a set of charts showing the same data, but in a different way.
i) The X axis starts at 1200 AD, so that the chronological extent of each series is transparent
ii) No point markers
iii) The plots are of the core ID, not of the core segment – this reduces the number of series on some charts, but the data points remain the same
iv) There are three spaghetti plots at sheets 24, 25 and 26 – no smoothing, 11 year MA, and 22 year MA. The moving average plots strongly suggest that something out of character happened around 1850.
v) There is a frequency distribution of the ring widths at sheet 27. This is a bit more spikey than I expected – spikes in a frequency distribution usually suggest some sort of rounding has been applied somewhere.
RE#105 DaleC
These trees seem to agree that 1860 was a very bad year (although not about much else). What happened in 1860? My first thought was Krakatoa, but that happened in 1883 – temporal teleconnection? 🙂 Might be a clue here about a key BC growth factor.
Cliff
Re #106 Cliff,
The trailing moving average algorithm pushes peaks and dips fowards. The average on the unrolled chart Spaghetti-orama MA0 shows that the bottom of the dip was around 1843.
Something has severely constrained growth from 1780 onwards, accelerating through the 1830s. So I would be looking for growth-sensitive factors pre 1840.
1843 climate related events:
June 1: It snows in Buffalo and Rochester New York and Cleveland Ohio.
July 2: An alligator falls from sky during a Charleston South Carolina thunderstorm.
November 13: Mt. Rainier in Washington State erupts.
Think it was the alligator? 🙂
BTW, DaleC thanks for the plots!
Cliff
re 105 DaleC – nice collection of graphs.
I’m sure somebody here must be searching through USHCN or somesuch place for precip data. I found the following site interesting. It doesn’t measure precip, but uses a tree ring proxy to illustrate Lake Mead water capacity. They show the same dip in ring width (lake level?) in the 1830’s. They pretty clearly label “flood” years and “drought” years.
I don’t know how to import images, but the image is in the link above. There was no link at the bottom of the page that showed assumptions in the tree ring graph.
DWhite
I can’t seem to get the “link” feature to work. Anyway, the url for 109 Dan White is:
http://www.lakepowell.org/drought.htm
See Figure 1, Page 2. Some of the dates seem to match up with some of the dips in ring widths.
Click to access Ma-01Kaufmannetal.pdf
Drought of 1845 to 1856:
http://www.ncdc.noaa.gov/paleo/pubs/woodhouse2002/woodhouse2002.html
So we can find a drought. Still looking for a teleconnection.
Folks, I don’t want to discourage you, but it is already well-known that bcps respond to moisture limitation. Further, you should not be jumping to conclusions about the nature of a signal based on the analysis of one event. The problem with proxy calibration/reconstruction is a deep one: how do you reconstruct one of many interacting inputs (i.e. T out of T, P, C, N, disturbances, etc.) given only one (nonlinearly integrating) output (i.e. the tree signal)? The answer is that it is not possible, not using a statistical approach. Full stop. All that is possible is a “reasonable approximation”. Hence the question: how reasonable an approximation is a linear univariate reconstruction model?
RE92 JOLTS.
Shocks to the system got me thinking a while back up the thread. 1934 is the hottest year on record.
How do BCPs record this? Can they record this? What does 1933 look like, and what about 1934, how long
dos it take a tree to respond?
Nearly without exception, tree rings were smaller in 1934 than in 1933. in only 1 case were they bigger.
( thats like 1 core out of 30 or so.)
In over 95% of the cases the wood put on in 1934 was LESS than the wood put on in 1933. Odd. Perhaps
it might be a drought? Dustbowl?
Anyway, the other interesting thing to see was the way the different trees reacted depending on the wood put on
the previous year.
Y= Ring1934-1933
X= Ring1933.
Y = 10.182Ln(X)-28.874
R^2= .41
Basically faster growing trees were hurt more. Slower growing trees showed the least effect in year or year growth.
So, for this shock to the system the tree is a good indicator. Only, it’s not measuring temp.
#75
Brunstein mentions the impact of porcupines on bcps. I recall seeing a tree-ring paper showing that porcupines in the SW US have a 20-year population cycle synchronized with the bidecadal solar and drought cycles. Is it possible bcp growth tends to rebound in a jolt following porcupine grazing? (Porcupines eat tree phloem, and in do doing partially girdle the tree by severing the cambium.) Are there porcupines in this part of California?
Porcupine Population Fluctuations in Past Centuries Revealed by Dendrochronology
Porcupine Feeding Scars and Climatic Data Show Ecosystem Effects of the Solar Cycle
Re #102, Steve H., Colo Spgs reservoir
A few obvious questions, if you proceed:
1) When was the dam built? First filled to capacity? Lake-level records?
2) Has it been drained for repair? Has it ever gone dry?
3) Has anyone else taken lake-sediment cores? {G}
Good hunting! — PT
By golly, maybe bender has something with the porcupine idea. Their range includes all of the western states and western Canada. Maybe we can construct a porcupine proxy.
jae, in the second paper in #115, note the HS in the Fig 1a porcupine proxy.
A porcupine says what? http://www.junglewalk.com/popup.asp?type=a&AnimalAudioID=10999
RE 114. And 1935 saw some record rainfall in colorado, so guess what happens to tree rings
in 1935 versus 1934? Guess. Kinda funny.
The 1840s stuff, is quite dramatic compared to the 1933,34,35 stuff..
Fun stuff about old old climate data
Click to access Mock_REVISED.pdf
BCPs and 1882, slide 11, frost ring studies
And Anthony will like the photos of the pikes peak observatory… and the grass valley
comments
The porcupines in Maine are teleconnected to solar-driven precipitation in the Seine.
RE 122.
Seriously bender, I was pretty shocked by seeing 95+% of the cores respond in the same fashion
from 1933-1934. Given the length of records I suppose I could always find a year where 95%
of the coin flips came up heads.
I think the jolts approach or what I would call the shot noise approach has some merit.
How does the sensor respond to a shock.
The other thing I was thinking was to look for “runs” runs in temp. for example, where
you have a 10 years or so of steady increase or steady decline..
118: Very interesting study. Influence (teleconnections of a sort) by the St. Lawrence is claimed, but not the Seine :).
Drought. colorado 1840s.
http://www.ncdc.noaa.gov/paleo/pubs/woodhouse2002/woodhouse2002.html
Click to access drought.pdf
http://www.sciencedaily.com/releases/2005/02/050218140645.htm
I would start with Colorado Springs Utilities at http://www.csu.org.
Mark
Actually it is probably possible, but only if a few conditions are met (which your explanation sort of rules out… hehe).
1) The inputs must be uncorrelated. They aren’t however, so strike one.
2) The output must be a linear combination of the inputs. It is not, so strike two.
3) You have a-priori knowledge of ALL inputs, and general knowledge of their shape/statistical properties. Nobody can even agree on the noise covariance, let alone what all the inputs are. Strike three.
Number 1, btw, makes it impossible to tell whether your solution is unique, and/or what portion of your result is due to which of several correlated inputs. Number 2 makes it impossible to tell whether or not your solution settled on a spurious local maxima/minima. Number 3 is a result of the fact that most of these methods don’t come with a little flag that identifies the results. There are multiple PCs, which is which? Temperature and CO2 look similar over the past 100 years, how do we know the answer is one and not the other? In other signal processing applications, you often have some knowledge of your desired result, a unique user sequence, for example, or some specific frequency that should not otherwise be present in your received signal. Not so here… Well, the Team does have an a-priori answer as we surely all know.
Mark
Want detailed photos of cored trees? I’ve finished a major chunk of work:
The Almagre Bristlecone Gallery now contains several new albums with detailed photos on all the trees that we cored and that were revisited in trip #4 (undertaken to obtain detailed angle and bark measurements and photos.)
Tree numbers 024, 026, 028, 030, 031, 033, 037, 047 and 048 all have detailed photo folders now.
Many un-cored trees also have new photos; I’ll do those later.
Also, I’ve uploaded many of the new panoramas that provide a way to see this bristlecone forest in context. They will soon be integrated into Google Earth. Visit http://www.GigaPan.org and search for Almagre to see this work in progress.
re # 58 Steve Mosher
Tree root ring sampling.
We all know that root direction is constrained by surrounding rock and soil and the quest for water. If a growing root meets an impenetrable barrier, it throws out lateral root(s). The direction of these roots is partly controlled by gravitropism, whose mechanisms have been reasonably well studied, so they do not often grow upwards. If they struggle to bypass a barrier they can put on girth at a rate different to that in free growth. Sometimes they can be eroded or bulge enough to be exposed to air, where a faster growth rate is theoretically feasible. Roots would seem to have more problems than trunks in interpretation. Do porcupines burrow?
I have seen examples of gargen trees started in 2 inch pots and replanted in 6 inch pots, with the operator being too lazy to take off the small pot first. The roots inside the small pot grew huge, those than made it threough the drain holes at the bottom of the small pot were tiny. A graphic example of comparing rings taken from two parts of a tree root. In fact, the whole tree was badly rooted before it died.
I agree with Bender that the many other factors that confuse temperature reconstructions are dauntingly large. It is easy to think of even more to add to the list. A theme I see more frequently from dendros is that the best tree ring trees are from trees struggling to survive. As I’ve quoted from other authors, with the Huon pine in Tasmania there is a gradation of perceived utility from higher altitudes (good for temp reconstruction) to lower. At lower altitudes other weather seems to effect each year’s progess more than temp. I do not know of a magic way to tell where to draw the line. Also, for old Huon Pine logs that have been underwater for centuries, one might not be sure if they originally grew above or below the magic line before being taken by flood and current to where they are now.
This has little or nothing to do with temperature measurement. I’d try the thermometer records of the Mt Lyell Mining Company as a closer source than those used from East Tasmania, if they exist.
BTW, I’m still sticking to my odd-ball “Something happened in 1888” theory, especially after seeing Dale’s and RomanM’s charts.
Finally, # 55 jae, trees in the SH spiral in the opposite handedness to those in the NH. At the Equator, ….. That’s my (light) theory and I’m sticking to it.
Late thought – In the Aust tropics and elsewhere there are trees like the Moreton Bay Fig that throw down aerial roots from the branches, so an old tree would have dozens of stems to sample, each from N, E, S and W as you might choose. Might be a good candidate for within-tree variability of rings.
re 129. well 1888 started off with historic blizzard record low temps. winds ar 60mph in denver.
not sure about the rest of the year.
#127 Mark T,
In the case of many-inputs-to-one-output calibration during instrumental period may be possible, but reconstruction of the past is not. Full stop. It’s not the nonlinearity that is the problem, it’s the multiple determination. There’s more than one way to get a particular output, so a given output (e.g. narrow ring) provides ambiguous information.
Yes, I should have taken it a step further: the “output” is not one-to-one (and maybe not even onto) and hence, not invertible. But yes, I agree with that wholeheartedly. 🙂
Mark
DaleC, #105:
Thanks for your graphs. It must be rewarding for Pete and Steve to see these initial interpretations of the data. It’s cerainly interesting to me.
Re:
Would you help me understand what this graph shows about ring width distribution? What unit appears on the x-axis?
Thank you.
Mr. Pete:
I haven’t had a chance to read all the comments above. Sorry if this question is redundant.
Have you been able to reach any general conclusions about which sides of the bristlecone trees were stripped?
I tried to reconstruct this in a quick scroll-through of pictures, keeping a second window open with your spreadsheet, in order to get directions. In many cases, the side of the stripping is not provided, leaving this question unresolvable as far as outside observers are concerned. So, your comments / general observations on this would be most helpful.
Thanks (And thanks for your work so far).
Reconstruction of the past from dendrochronology for climate proxies is possible if treerings are calibrated with 14C and vice versa. If the last case is given, biomarkers as proxies can be calibrated with other proxies from isotopes.
OTOH, Bristlecone pines grow better and live longer if they are growing on limestone soils because the water from rainfall remains available in that kind of soils for more extended periods than on sandy soils. There are several subspecies of bristlecone pines, so we have to take into account also the subspecies since some are more sensible to changes of environmental conditions than other.
I consider that one of the most accurate methods to reconstruct the environmental changes is iron-stained rocks, although all methodologies have pros and cons.
Re Bill: November 26th, 2007 at 12:27 pm
“What unit appears on the x-axis?”
The x axis is the ring width value item. The chart shows that the most common ring width value is 32 (x axis), at 240 occurances (y axis).
To verify this naively
i) open SMCD.txt in Excel
ii) select column E
iii) search/replace 32 with xxx in the selected column, Match entire cell contents ON
Excel will report that 240 replacements were made.
The clever way to do this would be as a pivot table of counts on column E.
The leading ‘c’ in my x axis values is an artefact of the charting software – ignore it.
Looking at the spikiness issue, the most extreme instance is 44. Width 43 occurs 128 times, width 44 occurs 203 times, and width 45 occurs 139 times.
For this many samples over such a small range (9778 measurements between 4 and 173) I expected a much smoother curve. Why should 44 be so much more popular than 43 or 45?
re DaleC #138
Looks like a preference for reporting even number ring counts. Also looks like if you averaged even and odd pairs, you would get a very smooth curve. Does that give an indication of error with respect to the accuracy of ring counting process?
Interesting chart.
Bill,
First, I apologize for the lack of labels. As noted elsewhere, I hadn’t realized that the file names would be invisible upon uploading 😦 … labels are now in place for all the new tree folders.
Second, while looking at the photos will provide confirmation, you can get measured angles on strip/whole segments from the provenance data file. There’s a column containing that info for every tree revisited on trip #4.
Finally, no I’ve had no time to think about such implications. I could guess-without-looking but my general sense is:
* In this location, weather tends to come from the westerly and northerly directions
* Therefore trees tend to fall south and east — and will bash other trees along the way
* And trees will tend to have snowbearing branches to the south and east (protected a bit from the wind)
* Therefore trees will sustain more physical (giant sledgehammer) strip-damage on their west and northerly sides
* And snow will cause more torn-off branches on the south and east sides
Now that I think about it, in general, I think I saw more trees looking sickly to the north and west, and looking a bit greener to the south and east.
Now you can look at the measurements and see if my hunches are even close to reality 😉
By the way, all this will vary a LOT based on microclimate: many little valleys and knolls guide the wind and weather. So everything I just said can be adjusted for microclimate 🙂
Dale: Thanks, I just had to look at it cross-eyed a few times.
And, since you put the graphs together, you are now prepared to wager fame and fortune on the causes of the 1850 event(s), right?
Pete:
Thanks. I appreciate your reply. I need to leave the house this evening, so I’ll look at the strip-bark issue again tomorrow.
That even-preference anomaly is rather interesting. I’d guess it has to do with how the size algorithm does its rounding and error accumulation.
Since most likely it is an off-by-one error, it would have little impact on the overall outcome. AFAIK, in the good old days, these measurements were more generalities than exact numbers anyway — more on the order of large/medium/small bins (well, more than that but you get the idea.)
How interesting! As we shed light on each step of the dendro data collection and analysis process, we’re discovering a variety of process limitations and sources of Uncertainty (mathematically speaking). I wonder how many of these have been written up in the past.
John Norris, November 26th, 2007 at 6:22 pm
Odds and evens was a fruitful notion. I have added an annotated zoom of the frequency distribution as Chart 28 here.
Interestingly, for widths less than 26 and greater than 52, the spikes are all at odd widths. For the widths between 26 and 52, the spikes are all at even widths.
Certainly looks like a processing bias of some sort somewhere.
I have also added the data itself (from SMCD.txt) at the Citations sheet as a convenience.
Bill, as to the pre 1850 event(s), I will happily leave the pursuit of fame and fortune to others. It’s all just a bunch of numbers to me.
So here’s a nice puzzle for someone to work out:
a) For each multi-core tree, what’s a measure of “higher” and “lower” growth curves (if any) and can you identify which cores (in a single tree) have more/less growth?
b) If the above exercise produces any fruit (so to speak), is there any correlation between high/low growth and:
1) core taken parallel to, on the upslope or downslope side of the tree? (and the amount of slope)
2) tree lean
3) direction of core period. One possibility: toward/away from prevailing winds (generally westerly/northwesterly)
Or… is high/low growth independent of “externally imposed” angles?
I must be missing something here. A plot of cumulative annnual increment over time indicates which core exhibited more growth. But that’s too simple an answer, so please clarify what it is you’re after.
DaleC:
Re:
I’ll second that. I tend towards the “Something happened / Something’s happening” hypotheses about climate change.
As you say, “Something out of character” occurred in the period around 1840 – 1850, at least in this area of Colorado. I’m from Colorado, so this seems like a mystery worth investigating. Do you think any other period in the graphs should be considered as equally anomalous?
If you go to the link in #111 you will find that there were regional fires in this area in 1841 and 1851.
#147. In one of Craig Brunstein’s articles – I think the 1992 AAR one – he talks about exceptionally severe winters in the 1840s.
Bender,
I’m sure it would be a simple value — cumulative growth might be it.
However, if two cores “cross over” (as in, one has spurious growth in the 1800’s, the other has spurious growth in the 1900’s), I would want a relative “growth index” value that is greatly reduced to the extent such crossovers exist.
My bottom line question here is: is it true that trees exhibit consistent spurious/reaction growth due on the upslope/downslope/whatever side?
So the first part is to find core sets with consistent, differentiated growth betweeen the various cores in a single tree.
The second part is then see how such divergent growth (or lack thereof) compares to the slope/lean of trees and which side of the tree is cored.
I follow.
1. You want to know to what extent radical departures are synchronized between cores within trees. If they are synchronized, they are more likely to be caused by external environmental perturbations. If they are asynchronous, they are likely to be heavily filtered by internal physiological processes occurring at a very fine scale. You don’t like correlation analysis because it’s global, not local, and also it gives as much weight to normal fluctuations as to radical departures.
2. Also at issue is not just the timing of these departures, but their magnitude. If one side of the tree always exhibits the larger anomaly, that may help to narrow down the set of possible causes.
For the latter problem, use JOLTS to filter the series to calculate the anomalies, and then a paired t-test on the set of inferred jolts to see if one core always has larger jolts then the other core. (Not that JOLTS is exceptional at this. It is just traditional.) Once you have determined the jolts, you want to do a one-tailed t-test on the difference in jolt timing between cores to determine if it is significantly different from zero. (You may also want to do this for all pairs of trees, but we can discuss that later.)
That’s extremely well put. And proves that I have a LOT to learn, to go beyond being a good data collector and data integrity watchdog 😉 …
hmmm…
re DaleC #144
Thanks for another interesting chart. There probably is little use to chasing down the source of the problem. However, I find it very annoying. I guess only because it seems as it should be easy enough to figure out.
You are most welcome MrPete. You & Steve M did some nice work obtaining those samples. Let me know if you need any additional clarifications.
Steve: For the big brains that are doing the statistical review and mapping of the ring data- are the updates of the nine (9) Graybill trees (out of the 17 you and Mr. Pete located) cross dated, and have you in any way identified them. It would be interesting to see what the extension over the life of those that were cross dated look like appended to the original Graybill data. I hadnt seen anyone mention these yet .
Back to lurking ..
CWells
All Graybill trees sampled are 100% identified – see the provenance data for the tags.
Detailed photos available for most (see the photo gallery) — we’re supplying improved provenance data to ASU, which has very little contextual data in their own database.
As for crossdating… Steve M can answer.
RE 147. REcord droughts in the 1840-1850 period. Just google it.
pointed this out a while back on thread.
http://www.ncdc.noaa.gov/paleo/pubs/woodhouse2002/woodhouse2002.html
with supporting tree ring data. As Bender has pointed out BCP respond to drought
You see that in the 1845 era, you “see” it in the 1930s.
Re #157 Better to cite Salzer & Kipfmueller publication than bender’s big mouth.
Well, I finally got around to arranging the Almagre tree ring width data into a table. So this is a story of mathematical exploration. First order of business, of course, is our favorite meal, the spaghetti diagram. Here ’tis …
Figure 1. All of the raw data. Ring widths are shown in mm times ten … I think.
At first glance, junk. But a closer scrutiny is fruitful.
First, what happened ~ 1840? I guess the rumors of drought are true.
Second, there are a number of discontinuous jumps, both upwards and downwards. The prime example is the jump around 1835 shown by the yellow line in the upper right.
Third, there is a common signal. This is most clearly visible say 1600 – 1700 (The blue line at the bottom ~ 1600 appears to be another discontinuous jump). But how to find the signal?
Here’s another way to look at this same data. We can look at how the ring widths are distributed. Are they normally distributed about the mean? As it turns out … no.
Figure 2. Violin and Box Plots, N = 9,585. These show the same thing in slightly different ways. The violinplot contains within it a simplified boxplot.
To those not familiar with boxplots, the box encloses half the data, from the lower quartile to the upper quartile. The “whiskers” that extend above and below the box extend out 1.5 times the interquartile range. Data outside that are called “outliers”, and are shown by circles. In the violinplot, the total amount of data of a particular value is shown by the width of the purple area. Each side of the “violin” is like a smoothed histogram turned on its side.
To me, this is not too pretty. It looks like a Poisson distribution, which is ugly because it’s not symmetrical. This means that one standard deviation above the mean is not the same as one standard distribution below the mean. So, because of the discontinuous jumps, and because the distribution is not favorable, I looked for an appropriate transformation.
With this kind of single-year discontinuous jump, one possibility is first differencing. The “first difference” or “Delta” transformation is the first derivative of the original data. Each value X(t) for a given year “t” is replaced by the difference between that year and the previous year, X(t-1). It has a curious property. If there is a 1 year jump, it only affects the value that year. Here is the first difference of the Almagre dataset.
Figure 3. First difference transformation of the Almagre dataset. In English, this is the difference, for each year, between that year’s ring width W(t), and the previous year’s ring width W(t-1). Or when speaking in Math, it is D(W) = W(t) – W(t-1), where W(t) is ring width in year t.
Now, this also looks kind of hopeless, there’s less evidence of a visible signal. But there are some things to be learned here. A large double spike, up and down, with both spikes the same height, is likely a single piece of bad data. I’m sure you can see it. I haven’t looked it up, but what happens is it jumps up to the erroneous value, then drops back down the next year by about the same amount, leaving an equal height spike up and down. Or, of course, it could be one and only one really great or really bad year for that tree … but in this particular case I don’t think so.
However, there are also single spikes evident in the record, both upwards and downwards. These record catastrophic events of some kind. The tree that had shaded another for 426 years was blown over in a storm. Insects ate all the leaves off one side. It went from being full-bark to strip-bark for the first time. Lots of possibilities leading to discontinuities in the record of any given tree. After that traumatic year, however, the tree goes back to it’s normal annual variation … but way out of place.
Here’s another look at the same data:
Figure 4. Violin and Box Plots of the First Difference (Delta) of the Almagre dataset, N = 9,544.
This is also curious, although it’s much better behaved. By that I mean it is symmetrical, with the shape above and below the mean being very similar. The oddity is the distribution. Look how short the box is in the boxplot. Half of the changes were within a ±16% change in ring width. The distribution has very fat tails, and is far from normal. Although the standard deviation is 36%, there’s lots of data points more than three standard deviations away from the mean.
How to make sense of that record? The median of the first differences offers a possibility. Because the median is not an average, sudden one-year jumps have little effect on it. So I looked at the median of the first difference, to see if it contained a signal. Here’s that graph, this time from 1900 on:
Figure 5. Median of First Differences.
This looks more promising. Again, some interesting issues. Is that a bad data point at 1924, or just one really bad year for that one tree with no lasting effect? And why is the median so variable in the last two years? One possibility is that the wood is not completely dried at the outer edge … dunno. The reconstructed ring width in 2005 was 5.2 mm wide, where it had been running at around 3 mm.
I was encouraged, in any case, to invert the median of the first differences. This should reconstruct the common signal. Here’s what I get from that inversion:
Figure 6. Effect of environmental factors on the growth rates of bristlecone pines in Almagre
Further notes to follow, it’s late. Lot’s more work to do, I have to put error bars on this, which is an unknown quantity for me. Neither distribution is normal, the interquartile range on the first difference is tiny, not sure how to figure the errors. I’m pretty happy with this signal detection, however. I think it will prove robust, and that it does a good job of capturing whatever common environmental signals are affecting all of the stand.
I would not say, however, that this is a temperature proxy. Nor is it a precipitation proxy. I would say that it is an integrated record of of the stand’s response to the instant by instant changes of temperature, moisture, sunlight, soil moisture, CO2, humidity, and other factors. At the end of the day, all we can say is that the trees grew better in 1550 than in 1650, recovered their growth quickly by 1700, went into a tailspin until 1850, and since then, things have been improving. Good news in Almagre.
More to follow, best to everyone …
w.
PS – as a final late night inquiry, I just looked at the difference between my reconstruction and a simple median of the raw data:
The median of the raw data (yellow) is influenced by addition of trees with wider rings, and by the discontinuous jumps in ring width. I’d say my reconstruction is more accurate than the raw median.
Next, I’ll go take a look at how the dendro folks combine a batch of ring widths … that’s the beauty of not knowing a lot, I can investigate without preconceived ideas of how to do it. Or perhaps someone can tell me how the dendro people combine a bunch of cores into a single record. Me, I’m going to bed.
Willis, we’re missing http://homepage.mac.com/williseschenbach/.Pictures/Spaghetti_al_Magre.jpg
Very interesting analyses, Willis!
How hard would it be to run the un-crossdated data through the same set?
RE 159.
Willis I looked at a very isolated case ( 1933-1934 )
basically:
Y= Ring1934-1933
X= Ring1933.
Y = 10.182Ln(X)-28.874
R^2= .41
Basically faster growing trees were hurt more. Slower growing trees showed the least effect in year or year growth.
If that helps any fine, otherwise ignore. It was, as I said, 1 isolated case, but if generalized you might find
that the first differences would still carry info from the previous year
Mr. Pete, I noticed the missing graphic when I came to work today. Unfortunately, the graphic is on my computer at home, so I can’t fix it for another couple of hours.
I don’t understand the difference between the cross-dated and the un crossdated data … nor is it clear to me, since the cores are all from living trees, why any dating (cross or otherwise) should be necessary. Could you explain that to me?
Steve Mosher, I didn’t use x(t) minus x(t-1) for that reason. Instead, I used percentage change. That way all of the faster and slower growing trees receive equal weight.
I’ll post the picture when I get home like I said, it was late when I posted.
w.
#148, 149, and 157 – Thanks for the links. I’m reading – and watching the developments here, too.
Steve: I’ve ordered a copy of Craig Brunstein’s article through our interlibrary loan system. From the PDF document about BC pine growth patterns (you linked to it above, in #75), it appears that he does good work, drawing a range of inferences about these trees from a relatively small sampling, I think about 35 trees. Amazing pictures.
Steven M: These are great articles. I skimmed the longer and read the shorter, Woodhouse / Brown. It seems inescapable that several of the downturns in the graphs above represent historically verified, widespread droughts in the West, including Colorado. I believe it was you, also, who suggest a concordance of local tempertures with the significant downturns. Tomorrow, I’d like to see how many of these line up with negatives on the Palmer Drought Index, and other BC Pine series that have been linked.
Home again, some random thoughts on re-reading.
1. I just noticed the original reconstruction in Figure 6 shows the reconstruction shifted upwards by about ten units (tenths of a mm). I didn’t realize it was off until I compared it to the raw data in Figure 7. I’ll correct that graphic.
2. I corrected the first link.
3. Steve Mosher, further reflection on your question sparks off the thought that a log transform of the first difference data might be closer to normal. It doesn’t matter to a median-based reconstruction, except that if it is normal, the error statistics would be better behaved. Thanks for the idea, I’ll take a look.
4. Where is the record for the nearest temperature station?
More later …
w.
willis 3 stations..
http://data.giss.nasa.gov/cgi-bin/gistemp/gistemp_station.py?id=425724660010&data_set=1&num_neighbors=1
http://data.giss.nasa.gov/cgi-bin/gistemp/gistemp_station.py?id=425724680010&data_set=1&num_neighbors=1
http://data.giss.nasa.gov/cgi-bin/gistemp/gistemp_station.py?id=425724640001&data_set=1&num_neighbors=1
Can someone download http://www.ltrr.arizona.edu/pub/dpl/DOCUMENT.ZIP and email it to me. The Tree Ring Lab at the University of Arizona, as noted before, continues to block my access to their site. Buncha jerks.
It will save me going to the university to download it.
re 167:
done
Thanks. It didn’t have what I was looking for – I’m looking for a pdf of Cook’s 1985 thesis. I saw it in a zip file recently but can’t relocate it.
Your hockey stick?
From the same paper in #170:
What would happen to bcp ring widths during an unprecedented medieval megadrought? Would the trees even survive? How would this influence temperature reconstructions that assume (1) temperature and precip are independent and non-synergistic, and (2) responses to each are linear? Would you recover Lamb’s MWP, or would you get a hockey stick?
not this cook?
http://hol.sagepub.com/cgi/content/abstract/7/3/361
yes that cook
RE 165. Willis.. I tried ratios but I got nervous about the varience ( Assymetry etc) and
I recalled that months ago we had dscussed this ( it was either you or RomanM) and a log transform
was suggested…
My sense ( wild ass guess) is that when the tree undergoes a jolt or a period of climatic runs
that the first difference is going to be related to the absolute size of the preceding year..
Vaguely.. if you are putting on “small” rings yearly, then jolts and runs will not have that big
an impact ( ratio wise) . If you are putting on Big rings then jolts and runs will have bigger impacts..
Perhaps it is related to tree maturity and resilence…
Rank speculation.
See http://www.climateaudit.org/?p=956 for interesting pictures of submerged medieval trees. I think tha tMiller et al is good on this topic. See the left frame Proxies – Medieval for some prior posts.
It gets better.
That explains why there are trees growing at the bottom of today’s lakes in the Sierra Nevada. They were puddles back then.
Now what would all this imply for the “temperature reconstructions” that use drought-sensitive bcps? And as a Michael Moore might ask: why is the paleoclimate research community NOT asking this question? Is it too mundane? Too difficult to fund? Involve heavy equipment to be hauled to the ass end of nowhere?
Or maybe the ecologists are afraid to confront the global climate modelers and express their skepticism? Or maybe there is more grant money to be found in following the consensus than in questioning it? Maybe they are afraid they will be branded as planet-haters? Maybe there are multiple reasons why it’s better to not think about it or keep your mouth shut.
RE 173. You know, if you could magically take away the political pressures…. This
Paleo recon science would be fun as hell.
Some of Cook’s most recent PDF’s are here:
http://www.ldeo.columbia.edu/res/div/ocp/pub/cook/
MrPete, we are nothing if not a full service blog. Here is your requested graph of the un-crossdated trees:
Can’t say I understand the differences. There are less trees, so that would have an effect, but we’re still seeing a very different record.
Two questions, might have been answered but if so I missed it:
1) How do the dendro folks do their averaging of the tree data to get a single line from a spaghetti diagram?
2) What’s the difference between the crossdated trees and the un-crossdated trees?
Best to everyone, more when I get time.
w.
PS – Steven Mosher, your idea about using logarithms is very good. Instead of using W(t)/W(t-1), I’m using ln(W(t)) – ln(W(t-1)). Makes no difference to the reconstruction, but it gives a distribution that is much nearer to normal.
RE 179… Thanks Willis… I have a couple other ideas I am playing around with.
Someday we should set the SETI software loose on tree rings… ha.
THANKS, Willis!
We’re all quite confused about the un-crossdated trees. Some of them were (physically) very very nice samples. Hopefully we’ll learn something from the dendro lab about this sooner than later.
re 166 – only the first one, Colorado Springs, is really even close. Pueblo is very far away with huge differences; Canon City as well. Even COS is not very comparable… several thousand feet altitude difference.
Just for example: the last few days I’ve noted a 15+ degree (F) morning temp difference over just a few miles of relatively flat territory, between Black Forest CO and the nearby Air Force Academy stations.
Re: non-crossdated trees.
I can guess, if you like, what these represent. I wasn’t going to, but now the questions are popping up.
Program COFECHA is typically used to ensure samples correlate well with one another (interannual fluctuations). If the correlation of a sample with a set of others or a chronology is very, very poor, there may be missing rings in the sample. This is a common problem for a highly responsive species like bcp. Rather than insert placeholders for missing rings into the problematic samples, it is better to set those non-crossdateable samples aside and start your analysis with the data with which you are confident. At some point the non-crossdated samples may become crossdatable (as you obtain more and more samples), so you don’t pitch them out; you keep them aside.
That is why the spikes don’t line up with the blue & red curves – those blue series have been shifted one year to the right for each sample for which there is a missing ring. The variance on the blue curve should be higher than on the red, because there is a lot less accurate dating (i.e. no cross-dating) on the blue. Cross-dating is basically “wiggle-matching”.
Just have patience. The lab will explain everything to you.
Thanks, bender.
Another point of interest. I used the median of the sample first differences for my reconstruction. I was interested to see the difference if I used an average. Here’s that comparison:
This is encouraging, because it indicates that the reconstruction is fairly robust with respect to the choice of measure of central tendency.
Finally, the correlation with the temperature in Colorado Springs is abysmal, R^2 = .004 …
w.
Interesting about missing years, Bender. “Crazy” thoughts…
When there are obvious physical breaks/gaps in a core, they seem to simply treat the core segments as multiple cores, with separate crossdating for each segment.
So… seems like an interesting puzzle, to find the gaps in the non-crossdated samples, and then perhaps to treat them as separate segments. If that’s possible, is there any statistical/scientific reason not to do so?
It is possible. The segments are referred to as “floating” chronologies. The goal is to “anchor” them (i.e. provide an absolute date for one of the annual rings) by using a variety of techniques, from visual inspection – paying attention to ring features and so-called “marker years” – to statistical wiggle-matching. It takes an experienced eye to match rings based on physical appearance, so usually both are used in combination. But before acting on my advice, be sure to ask Douglas J. Keenan his take on the reliability of statistical cross-dating.
MrPete is there a closer station? the three I picked were all I could find
Nearby stations: USHCN prob not. Station of record: try
K4BM (Wilkerson Pass) – it’s at least at altitude and in the same general mountain range
KAFF (Air Force Academy) – right at the base of the front range, reasonably nearby.
KCOHOLLI1 (Holliday Hills) – personal weather station (PWS) at 9600′, within a few miles
KCOMANIT2 (Crystal Park) – PWS approximately on the road to Almagre, but faces COS so perhaps influenced? I dunno
By the way, here’s another source that potentially could be a gold mine. I use links from this page to obtain weather estimates for the top of Pike’s Peak and Almagre (which itself is in the 12000 foot range)
http://inte099018.halls.colostate.edu/~vigh/weather/pikes/pikes_peak_weather.htm
Some links from there go to…
– Zone 82 (Pikes Peak above 11,000 feet — the Almagre sites are in this range, on the SE edge
http://forecast.weather.gov/MapClick.php?zoneid=COZ082 (and from there, a point forecast for the AlmEnd BCP site.)
– 48 hour upcoming weather graph — this is the best forecast I can get for a specific place up there
– Temp/Wind by altitude: http://inte099018.halls.colostate.edu/~vigh/weather/pikes/loop/
is climate history data there? I dunno…
Looking in SMIM.dat,- reload today, I can see the following lines.
Core.12B1620 59 65 80 85 94 76 101 92 109 74
Core.12B1630 78 95 49 999
Core.12B1583 35 49 17 25 27 23 33
Core.12B1590 34 36 35 54 42 58 47 41 49 50
I read this to mean there are two cores called 12B. Both start in 1583. One first one finishing in 1629, and the other finished 2007.
Should the second core be 12C, that is another core from tree 12, or have I misunderstood the file format?
Sean
What happened with the results?
And what are the statistical chances of getting a truly representative core from a heep Mountain type Pine on the only randomly selected occasion?