Climate Audit

Pat Keating

565
WOW! Thanks for the great post, Craig, and congratulations on putting together a much-need temperature reconstruction. That is big.
What does RealClimate say?

jae

Craig: GREAT, I’ve been waiting, as you know. Congrats.

Craig Loehle

Re: Steven Mosher throwing tomatoes at my reconstruction (see toward end of Unthreaded #24): Thanks to CA I was warned about certain data sets, and warned also about the variety of tomatoes that can be thrown, so I’m not too worried about that. I can imagine that RC will turn up their noses at the proxies I used, since they are addicted to bristlecones and I didn’t use any fancy stats. By the way, GRL sent it back unreviewed because they were tired of seeing reconstructions.

Joel McDade

From Craig’s paper

Many other series could not be used because they had too few dates within the 2000-year span or were not calibrated to temperature. In a few cases, data that were appropriate could not be obtained from authors.

Just wondering, who wouldn’t give up their data?

John V.

Typo above: Series A has an average of 0.625 (not 0.3125).

reid simpson

JohnV:

Not sure I understand your point and do not want to answer for Craig, but the abstract states: “In this study, eighteen 2000-year-long series were obtained that were not based on tree ring data.” Since each of the series was 2000 years long, doesn’t this allay your concerns?

scp

Well, I think RC will say “It’s Energy & Environment don’t you know.
No need to take that seriously”. Then, they’ll run for the ice. ; -)

bender

My review:

I would accept this paper for publication subject to major revision.

The writing style is clear and concise, and the introduction is useful, especially the summary of the problem with tree ring based based proxies. For that reason the approach is novel and the results worthy of publication.

There are several problems with the paper as submitted, but none of them are fatal.

(1) The locations of the proxies are not mapped. The audience needs to know two things: (i) what’s the largest area not covered by the network; (ii) what’s the average distance between sample points. i.e. What is the extent & resolution of the proxy network.
(2) The locations of the proxies that could not be obtained are not mapped. Consequently there is now way to evaluate the potential spatial importance of these missing data.
(3) There is no indication as to how spatially variable the proxies are in their degree of coherence to the global mean. Readers will want to see a global map of the correlation field.
(4) The individual proxy series should absolutely not be smoothed prior to the analysis of among-proxy variability. Through the Slutsky effect, this has the effect of exaggerating mid-frequency coherence, and attenuating between-proxy variability. The proxies are likely not as reliable as the analysis supposes. Although the attempt to quantify this variability is laudable.
(5) Although a permutation approach to sensitivity analysis is acceptable, there are boostrap methods available for robust significance testing. Generelized cross-validation should also be explored. Uncertainty analysis is one of the most critical areas of proxy-based temperature reconstruction and this paper has an opportunity to make a very useful contribution in this area. Exploration here will boost the value of the paper and attract the interest of a wider audience.
(6) The author should consider plotting 95% confidence intervals on the mean chronology, although it is recognized that there is error in both the x (date) and y (proxy value) co-ordinates, making this task more difficult than with accurately-dated tree-ring based proxies.
(7) Limitations in the proxies used should be discussed more fully. e.g. In the same level of depth that the tree-ring based reconstrucitons are discussed in the introduction.
(8) Individual proxies should be archived and the analysis script made available to verify correctness.

The author should address each of these concerns in a rebuttal or make the necessary changes. I would leave it to the Editor to determine if re-review is required. I would be pleased to serve in that role if called upon.

bender

Attacking EE is ad hom. Well done, Dr. Loehle.

Timo Viitanen

#9

It looks like all the series are clamped to the same time frame, 2000 years ago to now, so that shouldn’t be a problem?

JS

How does the computed mean compare with, say, the first few principal component?

Think of it as just another robustness test.

The implicit hypothesis behind the averaging of the data is that each series contains a signal plus some noise that cancels out in the averaging. Principal components analysis just takes that process a little further by presuming a signal plus other components that can be common or idosyncratic. If you use PCA to find the common components (not necessarily confined to one particular principal component) its just a slightly different way of cancelling out the idosyncratic noise than averaging. You would probably want to check the loadings of the series into the principal components to make a determinantion about common or idiosyncratic.

bender

Re #24
Usually the first PC in PCA IS the mean vector.

bender

I don’t know under what condiitons it is true or not – just that it happens a lot of the time with the data I use. Steve M is an algebra-nut though. He might know.

bender

When I said PC1 is usually the mean, I should have said ‘absolute value of the mean’. The issue here is not the sign of PC1, but its loadings. i.e. its shape.

Allencic

#19
Yes, there are things worse than being wrong. You could be Al Gore, which means being not only wrong, but scientifically and mathematically dumb too.

steven mosher

Craig,

the approach of selecting non “tree ring series” is methodologically interesting.

I’m interested in your selection criteria after that. The total list of non tree ring
stuff, why did you picked 2000 years?

Throwing out tree rings can also be a subtle form of cherry picking.

More to follow.

Bernie

This is a compelling article. However, the general tone here is on target, in that those committed to tree rings and HS will be fierce in how they pull this apart. I suspect it is already being dissected and torn apart.
Has the raw proxy data been archived? Can Craig’s work be replicated?

Geoff Sherrington

Re # 32 Steven Mosher,

Are you not being a little unkind in saying that throwing out tree rings can be a form of cherry picking? I thought the logic behind the paper went that there was doubt about the utility of tree rings, so let’s see what other proxies look like without tree rings. I can also pick at points after a first reading of the paper, but to me the important factor was that the signal:noise ratio looks favourable. Sure there can be better description of noise, and we might be treated to more such data.

My main philosophical problem is that some arguments used to doubt tree ring thermometry can often be levelled at the remaining methods. While it is encouraging to see other methods marching in step, one wonders if part of this is due to a dominant effect, such as the scientific truth of oxygen isotope work upon which a lot of reliance is placed. But not all, which is good.

Bender #16 has written a short thesis of his own which covers other comments I could volunteer. But overall, Dr Loehle has done good work to answer a question that many of us must have asked. Take away tree rings, then what do you find?

Steve Moore

Bob in #37:

Webster gives the variant spelling, but I used the definitions I was taught — which I still believe to be the primary ones.

The King’s English just ain’t what it used to be:
Perimeter/parameter, compose/comprise, gantlet/gauntlet, etc…

tetris

Re: 16 and 36
Bender and Geoff Sherrington
Gauntlets, tomatoes and other trivia aside, it would seem to me that it might be a tad more useful all around to focus on Bender’s detailed comments in 16 and Geoff’s pertinent question in 36 [“take away tree rings, then what do you find”?]

DeWitt Payne

I would be interested in seeing a frequency domain analysis of the series.

Nasif Nahle

# 41

Jonde,

I would suggest that Dr. Loehle would only focus on pointing the globe wide phenomenons of MWP and LIA.

You’re right, but first I would demonstrate that MWP and LIA were real.

Focus on only to the point you can clearly prove and leave speculation to minimum.

Yeah, you’re right.

UC

I’ll just copy Bender’s (4), (6), (8), and add

To date, only two long-term reconstructions have been published (low-frequency series of Moberg et al., 2005; Viau et al., 2006) that do not incorporate tree ring data.

Note http://www.ncdc.noaa.gov/paleo/ei/ei_nodendro.html (not very long-term).

I’d avoid repetition of word ‘believe’ in this kind of publications. Extra points for not using term ‘remarkably similar’.

OT-ish, if proxy response has a limited bandwidth wrt. target, the commonly used model P=scale*T+noise won’t work well.

Hans Erren

re 13

The series used were:
1) GRIP borehole 18O temperature (Dahl-Jensen et al., 1998);
2) Conroy Lake pollen (Gajewski, 1988);
3) Chesapeake Bay Mg/Ca (Cronin et al., 2003);
4) Sargasso Sea 18O (Keigwin, 1996);
5) Caribbean Sea 18O (Nyberg et al., 2002);
6) Lake Tsuolbmajavri diatoms (Korhola et al., 2000);
7) Shihua Cave layer thickness (Tan et al., 2003);
8) China composite (Yang et al., 2002) which does use tree ring width for two out of the eight series that are averaged to get the composite, or 1.4% of the total data input to the mean computed below;
9) speleothem data from a South African cave (Holmgren et al., 1999);
10) SST variations (warm season) off West Africa (deMenocal et al., 2000);
11) SST from the southeast Atlantic (Farmer et al., 2005);
12) SST reconstruction in the Norwegian Sea (Calvo et al., 2002);
13) SST from two cores in the western tropical Pacific (Stott et al., 2004);
14) mean temperature for North America based on pollen profiles (Viau et al.,2006);
15) a phenology-based reconstruction from China (Ge et al., 2003);
16) annual mean SST for northern Pacific site SSDP-102 (Latitude 34.9530, Longitude 128.8810) from
Kim et al. (2004); and
17) Spannagel Cave (Central Alps) stalagmite oxygen isotope data (Mangini et al., 2005).

This gave a total of eighteen (?) series with quite wide geographic coverage (including tropical) and based on multiple proxies.

richardT

From the manuscript: “all data were used that had at least 20 dates over the 2000-year period.”
This is simply wrong. Few, if any, of the sediment proxies have 20 dates in the last 2000 years. Sargasso Sea 18O (Keigwin, 1996) certainly does not.

Hans Erren

Before I give my verdict, I’d like to see the input data. I remember well Steve Mc’s apple picking exercise to “get” the MWP.
So, links to the archives, anybody?

MarkR

Robinson. The Hockey Team are school bullies who gang up in the Peer Review process to prevent publication if they can. This one slipped through, but they have Nature etc. under lock and key with the co-operation of the Editorial Staff. Be interesting to see how hostile the Peer Review was on this one.

Hans Erren

Please don’t get carried away.

kim

Time machine insight, Douglas; thank you.
=========================

Robinson

MarkR, maybe I have too much faith in Science. Theories become “fashionable” (string theory for example) but I don’t believe there is a systemic attempt at deception going on in any branch of science. A collective delusion I could understand however. I do believe the truth will win through eventually, whatever that may turn out to be.

Jason

What were the comments from GRL when they rejected this?

MattN

I suppose the question has to be asked: Craig, which oil company are you funded by?

😛

Don Keiller

Rob re#59. I would take your 10 pence worth more seriously had you not spliced instrumental NH temperature measurements (which are subject to question in their own right) to proxy reconstructions. This is a classic “apples and oranges” comparison which has been systematically discredited on CA.
PS any comments about Steve’s own attempts at dendroclimatology, or on Ababneh’s failure to discuss the Graybill BCP series in her thesis? Comments from a “dendro” would be appreciated.

Gary

Agreeing with #42 DeWitt Payne, it might be useful to see spectral analyses of all of the proxies. If the frequencies between so many differently obtained series lined up, one could have a lot more confidence in the embedded signal being common to them all. Then for an interesting test, see how the tree-ring series compare.

MattN

So. How are the boys a Broke Stick Mountain going to attack and discredit this? Because you know they are.

Jason L

It sounds like each series was somehow calibrated locally. Is it correct to say that many other studies are calibrated to global mean temperatures?

bender

Rob Wilson predicts the work will be ignored, but I don’t think it can be. Not completely anyways. If the major achievement of Loehle (2007) is to become part of a Wilson spaghetti-o-gram that stretches the envelope of possibilities vis a vis MWP vs CWP, that’s not a bad outcome.

Regarding a putative apples-oranges splice in Wilson’s graph. He has smoothed both instrumental and proxy series in the same way and is therefore not making an unfair comparison between time-periods. There is a big difference between overlaying multiple time-series on a single panel and blending them to blur the distinction. The only thing that is apples-oranges is the fact that there is error in the x direction on the non tree-ring proxies, whereas the tree-ring proxies are accurately dated, possibly 100% error-free. Don’t forget that when trying to determine how well the proxies match each other. Loehle’s graph does not honestly portray his error in the x-direction. The TR-based proxies do.

Nasif Nahle, a list is not a map. Not only do readers need a map of the sites, they need a map of the correlation field of individual proxy to mean.

steven mosher

RE 36. Geoff. I said I think not using tree rings was methodoligcaly interesting. My last comment
about ” subtle form of cherry picking” wasnt meant in a mean way. I’m thinking others may see it that
way. The justification for not using trees is clear.

bender

If you want a fruit salad, you gotta have cherries.

UC

#59,

In Panel B, I have scaled (same mean and variance for the period of common overlap – 1850-1979) all the series to NH temperatures (averaged from 0-90 degrees latitude using the CRU3 and/ocean data-set). This is likely not an optimal target for all the data-sets, but it is a compromise for this exercise. Panel C shows the same graph, but for the last 250 years.

i.e. variance matching. Can you tell how noise will affect these reconstructions, i.e. how to estimate confidence intervals ? Thinking about scaling error, maybe Craig has the most accurate reconstruction, and that’s why is has the largest variation 😉

bender

Let us suppose for a moment that Loehle’s MWP is correct. Let us suppose further that the Anasazi people in the SW US did indeed go into decline as a result of a series of MWP “megadroughts”. What do you think bristlecone pine ring widths would look like under those circumstances, given they grow in a subalpine desert where “water is the essence of life”? Fat? Or narrow? Would this be consistent or inconsistent with the Sheep Mtn bcp update? Is it possible that Loehle’s data and Ababneh’s data are wholly consistent?

Craig Loehle

Re: Rob Wilson’s second plot. He scaled all series including mine to have common mean (ok) and variance (not ok). Scaling variance to be the same prevents one from seeing that one reconstruction has more amplitude than others. Also Rob: I did not do a calibration. each series had been independently calibrated on temperature inthe original pubs, so you need to look them up. this in a sense is a double blind study–I did not do the calibrations to temperature so I couldn’t have gotten the answer I wanted.

bender

I agree with rebuttal #83. My point was directed at others who might try mentally “wiggle-matching” two recons that ought not be “wiggle-matched”. (Wiggle-matching only works if there is no dating uncertainty.)

But to get back to this paper: knowing the direction of the bias (a good thing) does not excuse the obligation to plot robust confidence intervals. In this case that would include error in both x (date) and y (magnitude), and the confidence envelope around your MWP would be correspondingly wider. The only way to have a fair MWP vs CWP comparison is to have a robustly estimated confidence interval around each.

Rob Wilson

Craig,

Re: Rob Wilson’s second plot. He scaled all series including mine to have common mean (ok) and variance (not ok). Scaling variance to be the same prevents one from seeing that one reconstruction has more amplitude than others. Also Rob: I did not do a calibration. each series had been independently calibrated on temperature inthe original pubs, so you need to look them up. this in a sense is a double blind study–I did not do the calibrations to temperature so I couldn’t have gotten the answer I wanted.

the scaling makes little difference to your series except depressing the cooler period (e.g. the LIA) a little. The MWP levels are the same. I only did this to ‘fit’ the series on to the same scale. Personally, I think the TR NH series should be calibrated against NH summer temperatures. With multi-proxy records like yours and Moberg, the optimal season is not so easy I think.

Still – it would be very informative to see a stack plot of all 17/18 records

R

Rob Wilson

Craig,
can you list the sampling resolution of your 17/18 proxy series?

I am troubled with the use of a 30 year moving average on a time-series that only has 20 observations over 2000 years.

What are the potential biases from aliasing and how could such effects impact the final mean series?

Rob

Jean S

However, by “giving” up after rejection from GRL and going for E+E, it is quite likely that your series will never be taken seriously.

That would tell a lot about the quality of science by the people doing multiproxy reconstructions. I don’t think the key issue in Craig’s paper is the exclusion of the tree rings, IMO the key issue is the local vs. the global (hemispheric) calibration. The Mannian school of calibration on large scale temperature trends has not been, to my best knowledge, justified anywhere. Now we have a reconstruction, which takes a different approach and shows significantly different results as you graphs show. So the ball is now with the global calibration people as the local calibration seems much more sensible approach (at least to most people, I think).

UC

I don’t think the key issue in Craig’s paper is the exclusion of the tree rings, IMO the key issue is the local vs. the global (hemispheric) calibration.

Silly assumption, proxy responds to local changes instead of global.. How can you claim small enough CIs then 😉

sod

I am troubled with the use of a 30 year moving average on a time-series that only has 20 observations over 2000 years.

so am i.

ending the series in 1995 of course is rather a CONVENIENT thing.

it removes about 7 of the 10 hottest years EVER from the record.

http://tamino.wordpress.com/2007/11/10/temperature-2007/#more-476

with the 30 years interval, that centers the last “warming” in 1980.

anyone got access to the data, to calculate a Mann hockey stick mean for the 1965 to 1995 period?

i think even the Mann data will look rather “cool” then..

ps: it would be really great, if someone with some time at hand could provide a link list to even just the abstracts of the proxy papers.
the most important data of course would be the year of the last data point used to calculate that means up to 1995.

pps: had to laugh on this one:

What’s a bit confusing is that this is immediately followed by the list of proxies used. But the sentence above doesn’t state that “only” proxies with 20 dates were used. A better sentence would have been, “… all proxies were used that had at least 20 dates over the 2000-year period plus a few that had fewer dates.”

sounds slightly like “islands had to be evacuated” to me.

Rob Wilson

JeanS,

Now we have a reconstruction, which takes a different approach and shows significantly different results as you graphs show.

lets not forget that the differences could also just simply be related to the use of a different set of proxies rather than the method.

This is of course easy to test. Normalise (same mean and standard deviation) the 17/18 series to a common period, average them together and then calibrate (regression or scaling) the final mean series to a NH target temperature time-series (Let’s call this the Esper/D’Arrigo approach as it is methodologically quite different to the Mann methods). Then compare with the Loehle E+E version.

Rob

steven mosher

RE 85. Dr. Loehle. Thanks for sharing the paper and subjecting your self to questions. I won’t draw
comparisions between yourself and others in this regard. But, you can figure who they are. Well done.

sod

honest question:

is 900-1100 considered to be the current “official” time frame of the MWP?

it is somewhat CONVENIENT to declare a 500 years period MWP.
http://en.wikipedia.org/wiki/Medieval_Warm_Period

(i m quite good at guessing lottery numbers. this week my bet is on the first 5 million numbers)

Craig Loehle

Bender wants to know what the x and y errors are. So would I. Dating error is rarely quantified and combining dating error and temperature estimation error have never been done to my knowledge. Feel free to publish a method. I used Jacknife and Bootstrap methods as the best I could do under the circumstances.
Re: 17 vs 18 records. Stott had 2 cores that were not close together that were used. Rob Wilson wants to calibrate the series against NH temp data, but the individual series are already calibrated against temperature. Please read some of them to see what they did. Sod implies that 1995 was “convenient” as an end point. That was as far as the data went. Sorry. Feel free to check the refs out.

bender

sod,
Look up “megadrought” while you’re at it.

SteveSadlov

RE: #47 – Hey punk, I am a scientifically trained engineer, at least in terms of my uni. I have a bachelors in geophysics and a masters in EE. If my masters was in a “science” and not “engineering” I’d properly label myself a scientist. (Some might even argue my bachelors alone allows that label).

(Folks – this is so typical of the ongoing smear campaign. They try to characterize us toothless hicks, without educations, sitting in red state double wides. That’s always how it is. The reality of well educated blue staters rising up against the high priests breaks the stereotypes. How dare we.)

Jean S

Rob(#98)

lets not forget that the differences could also just simply be related to the use of a different set of proxies rather than the method

Sure, it could be also an artifact of double low pass filtering, for example. I don’t know, I haven’t tried. Whatever the reason the local vs. the global thing (and the paper) can not be just ignored.

The test you propose is definitely worth trying. If you use the calibration time for the standard period and flip the series with negative correlation, then it is also Juckes’ CVM. That is such an excellent method as it recently passed flawlessly through an unbiased review as we all know.

UC

with that illegal variance scaling

without, I meant. I don’t want to break the law.

Jean S

SteveSadlov: Please send me e-mail (jean_sbls@yahoo.com), I have something off-topic for you.

SteveSadlov

Re: #75 – The bulk of the MWP, plus the RWP, plus the Minoan WP …. wouldn’t want people to make any sorts of connections in their psyches between civilizational golden ages and warm periods (or any sorts of connections between dark ages, barbarian conquests, war, famine, pestilance, poverty and death, with cold periods).

SteveSadlov

Re: #77 – To be fair, there is something to be said for at least scaling back on the rise in GHG emissions. Until we better understand what the real sweet spot for CO2 concentration actually is (we don’t know it now) some caution is warranted. That, of course, does not mean I agree with the IPCC / Gore program. It does mean that “business as usual” is probably not currently defensible either. I would speed studies regarding what the CO2 sweet spot is (based on the paleo record). I would also aggressively pursue an uptick in scientifically driven global hazard analysis, employing Sigma methodology and advanced statistics and physics.

SteveSadlov

RE: #82 – Yes.

SteveSadlov

RE: #114 – Got it, will do later today.

Bill

Doctor Loehle,

Well-done, and thank you for submitting to the gauntlett / gantlet.

I’ve seen several of these studies appear here and there, one at a time, over the last few years. But it is impressive to see them coalesce in a single graph. Mann’s hockeystick was a sort of “lamination” of other researchers’ data; this seems an appropriate and necessary reply, maybe even a tour-de-force. Other technically-savy writers here can advise you about the statistical pitfalls therein. I was blown away by seeing the “line-up” under your “Methods” section.

A good argument might start by making “points” one at a time, and establishing the validity of each. A commenter earlier asks that you detach a few of the individual components to show what they look like. I think this makes good sense, as a sort of “disclosure” of the merits of each research, and adds weight to the final combined graph. It might be worth the time and extra paper to overview some of these.

You’ve devoted a page to explaining why tree rings don’t work for temperature analysis, and you take pains to point out that “no tree rings were used…” but neglected to inform the readers how the methods you’ve chosen are efficacious or “proven” as accepted proxies. Perhaps this is a redundancy for some researchers, but then, I think your paper may be forging a path into new territory. You could bolster your argument by demonstrating how other processes of layer deposition provide a more readable, reliable result. As a non-techie, I’d be interested in knowing how general categories of this research (18 O, cave speleothems, SST, etc) have been validated.

Your first “reviewer” caught the business about uncovering the numbers on the timeline. This would help those of us looking for particular historical periods.

Now for the rottenest tomato of all: you use the word “quite” about 4-5 times, enough to be a little annoying.

Thanks for giving me the opportunity to line up with the real experts. It was fun.

Michael Jankowski

ending the series in 1995 of course is rather a CONVENIENT thing.

it removes about 7 of the 10 hottest years EVER from the record.

http://tamino.wordpress.com/2007/11/10/temperature-2007/#more-476

Welcome to the “bring the proxies up to date” crowd, where we strive to actually see if the proxies are capable of identifying the warm extremes they are portrayed as having done so hundreds of years ago! Why don’t you invite your buddies at Tamino to join in the cheer?

http://www.climateaudit.org/index.php?p=89

The hockey team has been making excuses for years now. It’s nice to see such an argent proponent of the hockey stick such as you to join the crusade. I am sure that in no time, we’ll see a massive push from you and your cronies to update the proxies so that not only does this reconstruction come up close to present, but all the others do, too! Then none of them will “conveniently” end too soon.

mccall

Congratulations! Great effort.

===== future supplement info candidates ===

1) I’ll echo the stacked 17/18 proxy request.

2) In the Wegman Report (in testing an M&M conclusion), the major proxy contributors to the HS were identified (~19 BCP + 1 other), and there contribution to HS was determined — conclusion: no BCP ==> not much of a HS (validating M&M conclusion)! I would like to see similar effort, on the big contributors from 1 — I suspect Keigwin’96 Sargasso will be one.

===== re: 47 Webster Hubble Telescope =====
WHT is an historic (and hysterical) TROLL … major worshipper of Deltoid with a minor in Rabett worship … a virulent left-wing idealogue of weak physical science.

A sample WHT post from his blog, when a Professor Dan Epp posted regarding “Taken By Storm” review thread:

Professor Dan Epp said… I would love to witness a debate between you and the authors. Your refutations do not seem to have much technical or scientific substance. We all need a cause, but it seems to me that you are a little too shrill in your comments. What is your agenda. I do not pretend to be well versed in math and the sciences, but these guys seem to make sense. Dan Epp 7:56 PM
[snip]

mccall

Note: I deliberately omitted posting a link to his post(s). Use a search engine to track, if you must; but I don’t think he’s earned the hits.

steven mosher

I know I asked this question before and have forgotten the answer ( Gunnar syndrome) Why would
you use a centered Smooth? I can see averaging the past into the future ( a trailing MA)
kinda carrying information content forward… but what is the physical justification
for a centered Smooth…. ah wait..hmmm. Anyway, Explain away. I’m wearing golf shoes
and don’t want to tread on my tender parts.

Craig Loehle

Mccall said “2) In the Wegman Report (in testing an M&M conclusion), the major proxy contributors to the HS were identified (~19 BCP + 1 other), and there contribution to HS was determined — conclusion: no BCP ==> not much of a HS (validating M&M conclusion)! I would like to see similar effort, on the big contributors from 1 — I suspect Keigwin’96 Sargasso will be one.”
This is why I did the 2 sensitivity plots. Dropping any 1 series (fig 2) or any 4 series (fig 3) does not change the conclusion. No series were weighted, they all had equal weight of 1.

steven mosher

RE 129.

Bender I have a Gunnar question. ( GQ for short)

I’m puzzled on this Proxy and teleconection question.

A tree in colorado is teleconnected to the hemisphere.

A thermometer in Denver only correlates with thermometers 600 miles away.

For grins I did a little study to see how accurately thermometers in the US could serve as proxies
for NH temp. I should prolly go do that again.

I think that led me to conclude that we should core trees every year and get rid of NOAA.

MattN

Wait wait wait….Lemme get this straight!

One of “us” creates a reconstruction that uses proxies ending in 1995, and that’s bad. But one of “them” creates a reconstruction that uses proxies from the 1980s (Graybill) and that’s just dandy?!?!?

L.M.A.O!

Craig Loehle

Mosher: why use a centered smooth? That is the mean temp of that 30 year interval. I am not doing forecasting, just trying to deal with data that are not annual.

SteveSadlov

RE: #116 – BTW, the late first millenneum was an underreported civilizational golden age in many parts of Europe. Think Charlemagne, the earliest seeds of the university system, the proto French state, the first serious public buildings built since the late 400s, etc.

MattN

Mccall:

Not sure if I my point was vague. If you look at the reconstruction in #111, you’ll see that by 1000AD, we were well on our way down from the peak of the MWP, in fact there appears to be a very quick drop right at Y1K. By 1000, the bulk of the MWP was done. I submit that Mann knew (or strongly suspected) this to be the case and stopped at Y1K on purpose to cut it out. I suspected this when I saw the GRIP data. I also submit that if the years 700-1000 in any way strengthened Mann’s case, they would have been included in his reconstruction. But they weren’t. He conveniently stopped at Y1K, and served up a giant cherry pie.

Craig Loehle

I want to thank UC for the plots in 111. Note that tree rings may be valid for COOLER periods, thus Mann’s results match from 1400 AD to present, but not before, when my results are WARMER.

steven mosher

RE 140. Dr. Loehle. The choice of a smoothing approach ( so I’ve been lectured here) can have some
consequence. As Long as you supply the data and other folks can smooth away with their favorite filter
I have no Issue. The issue of centered versus trailing isnt one of forecasting in my mind.

Pat Keating mentioned an approach to allow your curves to not toss away data at the end of the time series.
So, I’m just curious and ignorant as to why people would even use a centered approach as that averaging
doesnt reflect any underlying physical reality. Acausal was the word people used. This is an open question
not a critcism.

Again, As long as the underlying data is there and people can filter to their hearts content, whatever method you
choose to choose is your choice.

Craig Loehle

For TEG in 143. My my, aren’t we upset! If I take the instrumental record for a state over the 20th century and let’s say the records are widely dispersed. I can compute the anomalies and get a mean for each year. Right? Is this not what GISS does except they have unequal number of stations in any given region, so they weight them (more weight per station in sparse areas). If there is something wrong with this operation, you should tell Hansen. For the error bars, I assume that either the reader knows about the problems with the various proxies or can go read my 17 citations. But: what is known about the various types of error is not readily converted into error bars (go ahead, tell me how to do it), so I did fig 2 and 3 as the best that can be done with this type of data. 1970 to 1999 are not AWOL. The data only go to 1995 and 1980 is the last 30 year smooth that can be computed. Try reading the paper after you cool off. Cross-validation and RE and R2 are only relevant if you have a MODEL and I did not compute a regression or any model. Are there RE or crossvalidation stats or R2 on the GISS 126 year temperature history? At each point what I have is the mean of the 18 smoothed proxies. That’s all. I do not claim that it is a valid global mean and of course it is not. But that is all the data that existed as of 1 year ago in public archives that I could find. We will never have 1000 datasets to get a valid global 2000 year mean. I said that for these 18 points scattered across the globe, the MWP appears real.

bender

JEG,
To what degree was your #143 review written independently of my #16? Thanks in advance.

sod

Wikipedia? Heh!

i will very gratefully accept any link to an official definition of the MWP time frame. please lead me there!

remarkably similar during 1400-present.

yes, very similar up to about 1980. just missing the hokeystick part.

Sod,

The span of the MWP depends entirely upon how long the temperature anomaly is above zero circa 1000. It is ‘the name of the bump’ – so whatever size the bump is -> that’s the size of the MWP.

this would be fine, when looking at ONE study. if you use multiple studies, it is getting weird:

lets put some random noise between 0 and 2000. if we get a spike between 800 and 1300 it is a point for MWP, if the spike is between 1960 and 2000 it goes to Mann. now guess what..

Welcome to the “bring the proxies up to date” crowd, where we strive to actually see if the proxies are capable of identifying the warm extremes they are portrayed as having done so hundreds of years ago! Why don’t you invite your buddies at Tamino to join in the cheer?

i would prefer proxies to be up to date! but i understand, if people can t manage to do it. but doing a hockey stick debunk, while EXCLUDING the hockey stick period is slightly stupid.

this paper has tree ring data up to 2003 and produces a hockey stick:

Click to access Buentgen_2005_CD.pdf

Doesn’t a centered smooth also (largely) avoid endpoint issues?

doesn t a centered snooth also (largely) avoid results that have HOCKEY STICK form?

Pat Keating

140
The recursive averaging approach is also computationally very efficient (although that matters less nowadays). It would reduce the JEG-type criticism re sinc function sidelobes from the sharp edges of the simple averaging window.

bender

Re #111
The divergence period from AD1000-1100 looks something like the modern divergence pattern.

trevor

Re #143: JEG, you did note, didn’t you, that Steve McIntyre, in the header for this post stated:

I’ll try to comment later but have some other obligations right now.

And so far as I can see, he hasn’t yet turned his attention to Craig’s paper.

Can I suggest that you might take that into account in your comments.

Sam Urbinto

I too am curious about the RE CE and R etc I was also wondering about the lack of weighting, assuming all the proxies being used aren’t at the same levels. I’m guessing that’s the “each series had been independently calibrated on temperature in the original pubs”

Why don’t you wait until Steve has a chance to tear into it before asking for the information, JEG? I’m not qualified to do any kind of analysis, myself. And while you’re waiting, perhaps, if you know them, you can ask some of the dendro guys with unarchived data to archive it too.

Since you’ve stopped by and thrown yourself into the heart of us “ignorant global-warming-denialists” perhaps you can explain to us how you get 2.5 C of warming from doubling carbon dioxide in the atmosphere and increase our knowledge levels. I’d also like to know your take on the relationship of the GHG and how they react with airborne and ground level particulates, clouds, cities, freeways, farmland, forests and deforestation, and the ocean on the entire cooling/heating system.

And why an extra 100 ppmv over 170 years of a single GHG that absorbs IR at 4.3 and 15 micrometers, bringing it up to ~.04% of the atmosphere by volume plays such a huge roll in the 125 year +.7 C trend rise in the global mean GHCN-ERSST anomaly.

Or why land-use change isn’t spoken of more. Certainly the IPCC counts both.

Oh, and then explain this graph to me, I don’t understand it.

I must have something wrong, it shows temperature, O18, solar variation, carbon dioxide and methane levels are all in synch. That can’t be correct. Also strange is the 10 degree centigrade temperature swings 4 times in the last 420,000 years.

A sentence or two on each would be helpful to all of us I think.

Craig Loehle

Even though I posted this above, people keep asking about validation stats. The temperature series were all derived from regressions of proxies against LOCAL temperature over recent periods in the original pubs. All of these have R2 stats etc. I did not compute a model or a regression, so it is not possible to compute goodness of fit. Cross-validation and RE and R2 are only relevant if you have a MODEL and I did not compute a regression or any model. Are there RE or crossvalidation stats or R2 on the GISS 126 year temperature history? Also, I did not do a formal hypothesis test, as this is not supported by the data.

bender

Remember: that the study has “flaws” doesn’t mean the recon is inaccurate. Be careful not to be too dismissive.

pk

#157,
I’m not Nasif and this isn’t the prettiest overlay, but Loehle and TSI reconstruction.

bender

Cross-validation and RE and R2 are only relevant if you have a MODEL

You have a model. Your model is that the mean of the sample is a meaningful descriptor of the population that was sampled. GCV can be used in conjunction with resampling, which is sort of what you tried to achieve with your leave-some-out sensitivity analysis. The concept is great. Maybe you are not familiar with these kinds of validation techniques, but that doesn’t mean they don’t exist. They do. [Too bad the dendros are guilty of the same thing, however.] Also, remember each proxy itself is subject to its own error (measurement sampling, calibration), and this is something that can be included in a resampling approach. Something the dendros systematically choose to ignore in what appears to be a concerted effort to downplay uncertainty.

Reid

Jeremy Friesen says: “I hope to see JEG come back for a rebuttal.”

I hope JEG doesn’t beclown himself by calling CA denialist. Steve M. has never denied AGW. Let’s be precise here. CA and Steve M. has denied Mann’s Hockey Stick not AGW. Overturning the dominant paradigm is an unintended side effect.

bender

Drs. Curry (Judith) and Liljegren (lucia),
You share a common interest in trying to describe time-series with harmonic structure to it (LJ: temperature, JC: hurricanes). We were working on this last year when we got buried by an avalanche of papers by Emanuel et al. on hurricane pattern analysis. JEG is maybe even interested in this topic too.

I suggested that John Creighton’s MATLAB code for orthogonal filtering could be converted to R to make some better hurricane prediction models. (Timely, given David Smith’s recent observations.) I have yet to relocate that code, but it is probably worth revisiting. (Such a model would have predicted a low hurricane count for 2007.)

bender

I think I found it: http://www.climateaudit.org/?p=692#comment-79684

Craig Loehle

It may interest all of you that I decided on my analysis method and wrote the code for it when I only had 8 data sets, so I did not change the method to get the answer I wanted. Re: Bender’s comments about error bars. It seemed like the big issue was whether particular series were unduely unfluencing the outcome. To test this I did the jacknife and bootstrap tests–which by the way are also standard stat methods. To compute error bars I would need to consider measurement error for the proxy (usually unclear from the pubs), model error for the temperature conversion for that proxy from their regression, the possible bias due to the proxy (suspected but often unknown), dating error (also often unknown), and then propagate all four of these for each of the 18 series (using a monte carlo propagation?), and then combine them somehow. In a perfect world, perhaps it could be done. Anyone have an example pub where someone did this? Since I did not do specific hypothesis tests, it is up to the reader to consider that there is of course error in my reconstruction that probably looks something like fig. 3 if you could compute it.

Sam Urbinto

Sorry, I messed up that middle paragraph, the last two sentences are switched or should have been a separate paragraph. I wasn’t trying to link Phary and JEG together! (He was much nicer in his post….)

steven mosher

RE 165.

I like what Geoff has to say. As I noted the approach of tossing out all tree rings has a certain appeal.

As for what follows after that methodological choice, I’m going to have to listen to folks like UC, JeanS, bender, Willis.
IN the past they havent
been shy about pointing out errors. The point of the Gauntlett idea was to allow the paper to undergo
a harsh but fair.. (no motive hunting ) beating. maybe I’m just used to the Redteam/blue team paradigm
where you expected to beat an idea senseless and se if it it can still stand.

JEG kinda let his emotions get the best of him. That’s ok. We all see it.

Hi Dr. Curry!!! Did you read Lucia’s response to you about your hamming filter question? (he/she) gave
a nice explaination to your question. I think it was on the hurricane thread. I can check if you like.

bender

Re #178

To test this I did the jacknife and bootstrap tests–which by the way are also standard stat methods.

Could you point to the paragraph in the methods that describes this procedure? I’m afraid I missed it.

captdallas2

My comment, prior to reading all the others, was the lack of error bars and the 1970 ending should be addressed. JEG makes valid points in an edgy manner, but valid, that should be addressed.

One point I think should be expanded is the validity of SST proxies in determining long term temperature reconstructions. A good paper would be the comparison of the SST proxies and tree ring proxies with the results through 1970. This would sidestep the 1970 ending period question and the error bars should be significantly lower for the sst proxies.

A good statistical evaluation comparing the skill of sea sediment versus tree ring proxies, I think would be amusing.

Craig Loehle

re: 184 Bender. It is last para before discussion:
“A test of sensitivity to individual series was performed. Individual series were dropped and the mean recomputed, with all other steps the same. These eighteen series plotted in overlay (Fig. 2) show the same pattern as the original (Fig. 1). This plot demonstrates that no single series has undue influence on the result.”
That is jacknife.
“Random subsets of fourteen series were constructed and still show the same basic pattern (Fig. 3).”
That is bootstrap.

Craig Loehle

Thanks to Jean S #183. Exactly.

lucia

Judy–
I could, by the way, explain the rule of thumb of “30”. It’s a rule of thumb, and so can be modified. The reasoning has to do with beta error. If you’d like other discussions I guess the place to do that is unthreaded.

Bender– thanks for mentioning the first name or Judy’s jaw might truly have dropped at the thought that “Dr. Liljegren” was here in comments. 🙂

JEG

#152
JEG,
To what degree was your #143 review written independently of my #16? Thanks in advance.

almost entirely. started writing last night and was happy to see we had common points. I would argue that most of yours are “fatal”, though, but that is a subjective judgement.

#164
Even though I posted this above, people keep asking about validation stats. The temperature series were all derived from regressions of proxies against LOCAL temperature over recent periods in the original pubs. All of these have R2 stats etc. I did not compute a model or a regression, so it is not possible to compute goodness of fit. Cross-validation and RE and R2 are only relevant if you have a MODEL and I did not compute a regression or any model. Are there RE or crossvalidation stats or R2 on the GISS 126 year temperature history? Also, I did not do a formal hypothesis test, as this is not supported by the data.

There’s a good reason why people keep asking about it, Craig : while you may not realize it, you *are* using a statistical model called a multivariate regression. Indeed, each of the proxies was calibrated against temperature using such regression methods, usually linear, but sometimes non-linear (Mg/Ca ratios depend exponentially on seawater temperature, for instance). So by doing even a simple average, you are effectively using a statistical model, and are not exempt from the verification exercise.

It is quite ironical to be calling d18O “temperature” without considering how much temperature variance it describes, and downplaying the GISS NH temperature curve altogether. While the latter involves (minor) statistical assumptions, it is a compilation of physical *measurements*. Thermodynamics is not sedimentology. But to a statistician that may be all the same, who knows ?

For the error bars, I assume that either the reader knows about the problems with the various proxies or can go read my 17 citations.

How convenient ! That way we can just take your word for it and stop questioning the choice of proxies. Brilliant.

But: what is known about the various types of error is not readily converted into error bars (go ahead, tell me how to do it), so I did fig 2 and 3 as the best that can be done with this type of data.

Here’s how to do it, for example : http://www.image.ucar.edu/~boli/manuscripts/2007-LNA-TeA.pdf
If that is the best that *you* can do, then you lose every credibility as a challenger of the “Team”.

1970 to 1999 are not AWOL. The data only go to 1995 and 1980 is the last 30 year smooth that can be computed. Try reading the paper after you cool off.

OK OK, my bad. The late twentieth century just looks a little unusual. I acknowledge i had read this part a little quickly, but I still think you should pay closer attention to smoothing potentially non-stationary timeseries.
(warning : epileptics and people allergic to the name “Mann” should not open this link)

[…] At each point what I have is the mean of the 18 smoothed proxies. That’s all. I do not claim that it is a valid global mean and of course it is not. But that is all the data that existed as of 1 year ago in public archives that I could find. We will never have 1000 datasets to get a valid global 2000 year mean. I said that for these 18 points scattered across the globe, the MWP appears real.

Then isn’t it staggeringly misleading to call this a ‘global temperature reconstruction’ ? (did i misread, or is it the title of the article ?). The adjective ‘Global’ implies representativeness of an area-weighted average, which it is not. Now, it may turn out that your spatial sampling does a reasonably job at representing tropics, mid and high-latitudes , but it seems a little careless not to ensure it, and even more unforgivable no neglect mentioning it.

I would have no problem if the paper said : “Picking 18 series that obey a set of objective criteria, i find that on average they display a warmer than present MWP”. One can argue about the criteria, but at least the conclusion is falsifiable.

What you did is VERY different thing : you pick 18 timeseries on criteria that do not guarantee a proper spatial and temporal resolution of the temperature field. You smooth them using rudimentary methods, aggregate them together without any other form of consideration, and sell it to reader as a “global reconstruction”. Worse, you conclude that the “it would indeed seem to show the MWP to be warmer than the late 20th century” by 0.3 deg on average, implying that it is representative of a global average. Or did i misread ?

FYI, you can test that result for significance : i heard it was called a Z test. In truth, it would have to be a lot more sophisticated than that (cf Li et al 2007). But hey, if CA readers and participants are so much smarter than us dumb climatologist, they can do it ! Bring it on, my friend !
If, as you claim, ” formal hypothesis test, as this is not supported by the data”, then perhaps you are saying more than what your data warrants ; which is pretty much nothing. The bottom line is : if yours was a serious study, you would build upon the vast array of statistical techniques that have been developed for that in the past 9 years that elapsed since MBH98 (see the NAS report of a catalogue). Since you do not bother to do that, do not expect to be taken seriously.

If on the other hand, you are after a bona fide tree-ring-independent reconstruction, i’m sure you will find support in the climate community.

Re #167 : thanks for your support Judy, but you’ll find that i’m hard to scare.

I knew when i entered this debate that i would face Nobel laureates like steve mosher calling me “puppy” or even bird names. I knew some people who jump out of their chair at the mere mention of “denialist”. I knew people like Sam Urbinto would deflect comments from the actual question, and make other artful uses of ignoratio elenchi. As i said, in spite of all that junk, i still read CA (and i am obviously not alone) because when you want to know what happens, you don’t read party’s paper, you read the opposition’s. It is now up to this opposition to make a clear statement about the goal of this blog :”Climate Audit” or “Climate Trash-talk” ?

JEG

hmmm, please forgive the typos in my previous post. Ah, the heat of the battle !

Re #143: JEG, you did note, didn’t you, that Steve McIntyre, in the header for this post stated:

I’ll try to comment later but have some other obligations right now.

And so far as I can see, he hasn’t yet turned his attention to Craig’s paper.

Can I suggest that you might take that into account in your comments.

Yes, i did note that, and i find it very entertaining that when Caspar Amman is struggling with publication deadlines, it deserves a “Where’s Caspar ?” entire thread, but that when Steve McIntyre is busy, his army stands still to hear what he has to say because they can’t form a judgment by themselves.

Julien

Mike

JEG seems angry

Spence_UK

JEG,

I note you make a few valid comments – particularly the lack of confidence intervals (although I wouldn’t hold Mann’s attempts in MBH98 as a gold standard for this – poorly described, incorrect confidence intervals presented as valid are not an improvement over no confidence intervals).

With regard to proxy selection, calibration and dating, I would suggest that relying on others to achieve this is an entirely fair way of doing it. I would argue it reduces the opportunity for researcher bias to enter into the study. That said, I would personally like to see a definition for what justifies a proxy selection with sufficient accuracy to enable out-of-sample data to be added in the future, as more records become available. Practically, this probably needs some account of signal-to-noise (or confidence) of the original study to local temps, plus some ability to weight geographically to prevent well-studied areas becoming overweighted. I’m not claiming this study is a panacea in this regard, but it is an interesting and (I think) valid alternate approach.

– SMOOTHING : a running mean is essentially a multiplication by a gate function , which – as the author must be well aware of – amounts to a convolution with a sine cardinal in frequency domain. This introduces sidelobes that distort the signal, and would presumably mess up the edges of the timeseries. Oddly enough, the 1970-1999 period has gone AWOL…. One would like to see other smoothing methods applied to remove that doubt.

This interests me. I am well aware of the preference for climate studies to want to run their reconstructions right up to the end point after smoothing. But why would you necessarily want to do this? We’re generating 2000 year record, is a loss of 15 years at either end a big deal? “Clever” end-point processing clearly involves either some form of extrapolation or increase in uncertainty (the uncertainty, I might add, is rarely properly accounted for). There is a valid reason to do this as an interim step (to prevent your data shrinking away), but there is really no valid reason to do it for presentational purposes. The running mean has pretty grotty spectral qualities, agreed, but it is unbiased, it has the benefit that it is about as short a filter you can get in the time domain for a given amoount of smoothing, requires no assumptions about end point handling / creates no end point issues if you simply truncate the series by half of the filter width at both ends. Comparatively, the other methods inevitably impose researcher assumptions into the study, creating an opportunity for researcher bias to influence the results (the end-point assumptions).

Finally, on your criticism on the effect of sinc function spectral leakage on the results, you may or may not be aware that Mann and Jones 2003 includes filtering of the MBH PC1 and a Jacoby series. The filtering applied is an interim result and the end points are retained. Guess what? Yep, the sinc function is there, clear as day, spectral leakage, an artefact of the processing. Perhaps you might want to have a word with Mann about what you have regarded here as a serious error? Note: I’m not arguing that the presence of the sinc function in the spectrum is a big problem, I’m just curious that you think it is, how that applies to MJ03; I’m happy to explore the point in more detail, especially as MJ03 actually use the end-points in their analysis.

Nasif Nahle

Yeah, JEG’s angry. BTW, he has not answered my question, is the universe static?

Please, could someone here instruct me on how I could load a scheme on this page?

Hans Erren

Craig:

Nag nag, url’s to your used proxies please!

Spence_UK

Re #200

Did I write Mann and Jone 2003 there? Oops! Where I wrote Mann and Jones 03, replace with MBH99 CO2 correction. Mae culpa.

Dave Dardinger

re: #143 JEG,

You did notice that your posting was comment #143 which had been posted after less than 24hrs, didn’t you? And while there were some ataboys, there were a number of very pointed questions aimed at Dr. Loehle. There really wasn’t any call for a remark from you like,

It’s hard enough sifting through the comments of ignorant global-warming-denialists on this site

Of course I also agree that some of our allies here leave a bit to be desired at times, but I’d hope you’ve noticed that you have similar problems. Do you perhaps have a sample post at RC which illustrates how you folks take on people who generally agree with you? We’re always anxious to learn.

DocMartyn

” bender says:
November 16th, 2007 at 8:53 am

Let us suppose for a moment that Loehle’s MWP is correct. Let us suppose further that the Anasazi people in the SW US did indeed go into decline as a result of a series of MWP “megadroughts”. What do you think bristlecone pine ring widths would look like under those circumstances, given they grow in a subalpine desert where “water is the essence of life”? Fat? Or narrow? Would this be consistent or inconsistent with the Sheep Mtn bcp update? Is it possible that Loehle’s data and Ababneh’s data are wholly consistent?”

I think you are asking exactly the wrong question. Indeed, the point you miss is what are you looking at in the tree ring data; survivors. Although a single Bristlecone Pine can have a 2,435-year tree lifespan, they rarely live over 1,500 years. The one’s that do have survived extremes of heat, cold and water stress.
the living trees that are sampled MAY live in places where due to some querk, are shielded from extremes of temperature/water stress.

JS

I reiterate my suggestion from #22 to look at a PC version of the series. This is a way of checking robustness as well as being informative in its own right.

PC is atheoretical as it does not impose any interpretation other than a mathematical one on the data. Nonetheless, it should serve to illuminate the implicit assumption underlying your ‘reconstruction’. Because you have quite a strong implicit assumption underlying the validity of a straight average of the data.

Averaging assumes that there is a common signal in each series and then random noise added to them. If there is any other quality to the data then averaging will lead to biases in the result. Averaging works best for more or less homogenous populations (e.g. the height of white males). However, your proxies are not obviously homogenous in the way they capture any temperature signal. They are from different processes and are from different locations around the globe. Yes, they have been calibrated to local temperature, but there are still idiosyncratic factors there.

I should note that PC is just a tool. It can be used well and it can be used badly. I would encourage you to consider if it can be used well in this case.

Craig Loehle

I have been accused of exaggerating my claims. The last sentence in my paper is “It must be emphasized, of course, that this result is based on limited data.”

Bender would like a formal significance test. How do we test for “the MWP existed (or didn’t)”? I was trying to show qualitatively that the MWP is plausible if you don’t use tree rings, and I think this graph shows that. I did not feel the data supported formal hypothesis tests so I did not do them.

JEG: check out how bender is twisting my arm without being angry, without insulting me. very professional. try it.

Dave B

As a relative layperson, I’d like to point out how refreshing it is to see a proxy study with clear inclusion criteria (2000 year, 20 points) and exclusion criteria (tree rings). The consistent unwillingness of the “Team” to assess and admit what happens to a proxy analysis in the absense of Cherry-picked bristlecones arouses suspicion.

SteveSadlov

This is so cool …La Sorbonne en web ….

bender

Doc Martyn,
You are too funny. Please check this post. Have a great weekend.

bender

JEG, could you redo Li et al. (2007) analysis without bcp/foxtails & other drought-limited species?

Bernie

Most of your ideas sound okay to me and well worth considering. Here is how one CA’er thinks about what you said.

Relationship of each proxy to *local* temperature is not even discussed
Agreed. It would be helpful to provide a synopsis of the +/- of the series as a temperature proxy
Error bars on the proxies
Agreed. These would help understand how much range there is in the curve and they should be in the table that you requested above. However, isn’t the sensitivity analysis itself is a form of addressing this issue?
“Borehole d180 temperature” is a climate neologism .
Not sure I understand this – unless it is simply a dig.
Relationship to global temperature Agree but you are asking for something that I do not recall seeing elsewhere. Moreover, when the sample is so small given the overall population then weighting these things is false precision to say the least. A meagre sample is a meagre sample.
TimingPerhaps this is a separate paper, but seems to be an additional point to the one being focused on here.
SmoothingSeems like this can be applied to all proxy studies. The extending of the proxies to take in the last few years is a great idea.
Validation R2 – Good idea but the sensitivity analysis again does this at least at visual level.

Overall, it seems that you have captured and applied many of the issues that Steve and others wrestle with here. I suspect Steve will be as pointed as you, but probably with a slightly more constructive tone.

Will J. Richardson

Dear Dr. Emile-Gaey,

Since you are presently working with Dr. Mann on a paper, could you please ask him to provide some assistance in resolving the issues presented on this Climate Audit thread?

http://www.climateaudit.org/?p=2380#comments

Thank you.

Steve McIntyre

I’m on a little holiday trip with my wife and only have time to check in briefly today and don’t have time to respond to many interesting points. Our “apple picking reconstruction” to the NAS panel is here . Rosanne D’Arrigo told the NAS panel that you have to pick cherries to make cherry pie; our point as that if you picked apples you could make apple pie.

JEG raised review points that would surely apply to virtually every multiproxy reconstruction. JEG observ$se:

– Relationship of each proxy to *local* temperature is not even discussed. We are just shoved a list of references (hey Craig , have you heard of tables ? They are a great means that scientists use to convey information clearly). How is it that tree-rings are seen as the penultimate antichrist but that, to take but one example, d18O of speleothem calcite (e.g. Holmgren et al., 1999) is a flawless paleothermometer ? Shouldn’t one discuss one by one, and with great care, the pros and cons of each proxy as a temperature recorder ? d180 in forams is a notoriously flawed temperature proxy, is being replaced by Mg/Ca where possible, and one can only surmise why the Keigwin[1996] series is here.. Is it just a form of screening for proxies with a low ‘hockey-stick index’ (in McIntyre parlance) ?

I agree that a detailed list of exact data sources should be a prerequisite of this sort of calculation. Having said that, Hegerl et al Nature 2006 for example contains no information on the proxy provenance. As an IPCC reviewer (where the paper was cited), I requested information on the series on the series used, which the authors then refuse to provide and IPCC did not require them to provide. Partial information was provided in a later publication J Clim), but after nearly a year and repeated emails, I have still been unable to determine the identity of several series versions. Crowley at one point said that he had a hard drive failure but that was nearly a year ago.

JEG, I am unaware of the type of discussion that you request in any of the usual papers. Where is the calibration to local temperature (for example) of the Arabian Sea G Bulloides series used in Moberg and Juckes? Why is this a defect in Loehle and not for Moberg?

JEG, you say:

Say what you want about MBH98 : in spite of its many flaws, it made extremely explicit attempts at estimating error bars. One would expect this work to have paid attention to past work and up the ante where it can.

Perhaps you can explain to us how MBH99 confidence intervals are calculated. We’ve tried pretty hard to figure it out and haven’t succeeded. Requests both to Mann and to theNAS panel to clarify this have gone unanswered. See http://www.climateaudit.org/?p=647 http://www.climateaudit.org/?p=648

Merely using 2-sigma errors in an overfitted calibration period (as in MBH98 but not MBH99) is not an improvement over no calculation. The calculation is incorrect and very misleading. The verification r2 of Mann’s reconstruction is more or less zero and CIs calculated from the verification period would be approximately the same as natural variability – whatever that is (and it’s not an easy question to answer) See illustrations of overfitting through Mannian-Ammannian inverse regression here http://www.climateaudit.org/?p=949

I agree that “borehole dO18” is meaningless. Dahl-Jensen is a borehole reconstruction.

You cite Hughes and Diaz 1994 as an authority. Without going into detail, the proxies selected in Hughes and Diaz are very problematic and I do not think that you can draw any conclusions from this study. Even Jan Esper in his IPCC review comments recommended that it not be cited.

JEG

Bernie says :

Error bars on the proxies
Agreed. These would help understand how much range there is in the curve and they should be in the table that you requested above. However, isn’t the sensitivity analysis itself is a form of addressing this issue?

The sensivity analysis only addresses “red-herring effects” : how much one given series influences the reconstruction as a whole. By construction, since they have equal weights, it’s only 1/18th of the result, so it can’t do much unless it has a vastly different value.
It is a very useful analysis in itself, but gives a false sense of security in regards to the accuracy of the mean. Imagine all of the 18 proxies were accurate within +/2.5 degrees. Then what to say of a 0.5 degree change ?
I agree that error bars are not easy to truthfully establish, but say what you will about MBH98 ; they tried, and that’s a BIG difference. Sure, in an “ideal world” (in Loehle parlance), all the proxy 2-\sigma errors would be know perfectly over the past 2000 years. It is not the case, but that is a poor excuse for ignoring the issue altogether.

“Borehole d180 temperature” is a climate neologism .
Not sure I understand this – unless it is simply a dig.

OK, let me try again. There is no such thing as a “Borehole d180 temperature” (methods section, L4). Borehole temperatures are one direct (although blurred) measurement of past temperature down the ice cap. d18O is an isotope ratio that has a complex (though interpretable) relationship to temperature in the adjacent region. “Borehole d180 temperature” = nonsense.

Relationship to global temperature Agree but you are asking for something that I do not recall seeing elsewhere.

Hmmmm… let me see : MBH98, MBH99, Esper et al 2002,Rutherford et al 05, Hegerl et al 2006, etc. etc etc.

The point is : either you care about LOCAL temperatures and you use something known as a Composite Plus Scale approach. Or you only care about the ability of a proxy series to record *some* climate information via teleconnections : that is the heart of climate-field reconstruction techniques, like MBH98 or the more sophisticated RegEM-based versions. (Rob, would you agree with this dichotomy ?)

In either case you explicitly account for how proxies describe variance in the mean temperature ; either by regressing them against the mean, or against local temperature (and then average them), or against a subset of principal components that describe the large-scale features of the temperature field, then use that to reconstruct this linear subspace of the T field and then average it globally (the Mannian approach).

No such thing is done here, and it should. Again, it comes down to whether Craig wants to convince scientists that his reconstruction says anything about climate, or whether he only wants to cater to a crowd (which is some, but crucially not ALL of the CA readership, thank goodness) that goes off in loud cheers anytime someone throws dirt at Mann, Hansen, Al Gore, and others.

Validation R2 – Good idea but the sensitivity analysis again does this at least at visual level.

If a visual fit is enough for validattion, then why are there so many complaints about Mann et al not dutifully using the Rsquared statistic for validation ?

Again, i’m simply making the point that if you have high standards for “Team”, then you must have high standards for everyone.

Overall, it seems that you have captured and applied many of the issues that Steve and others wrestle with here. I suspect Steve will be as pointed as you, but probably with a slightly more constructive tone.

Thank you. But i think I’m damn constructive ! Look at how much progress we’ve made so far ! 😉

I will surely read Steve’s insightful analysis with much interest.

Craig Loehle says:
“Hans: I will try to get the links to the raw data when I can, but JEG is keeping me pretty busy.”

I love it !!! Now it’s my fault ! “The reconstruction is meaningless and i don’t make raw data available on time, but blame it on the Frenchie !”
Well, perhaps now you understand what it is to be a climate scientist, have many obligations for your job, and also have to satisfy the demands of CA’s inquisitive minds. Overall i think everyone will learn valuable lessons here.
I certainly am.
BTW, thanks to everyone who is actively engaging in the scientific part of this discussion ; it is refreshing.

Bender says :
“JEG, could you redo Li et al. (2007) analysis without bcp/foxtails & other drought-limited species?”

Why don’t you do it yourself and tell me the result ? 😉 If it’s worthwhile i’ll listen.

Wishing you all a pleasant week-end… see you monday.

Sorry again if anyone got offended at my tone but it did push just the buttons i wanted to push, and perhaps now we can talk about science.

John Norris

Had I known JEG was teaching a hockey stick class right here in Atlanta, I could have saved some significant CA bandwidth.

http://eas8001.eas.gatech.edu/syllabus.htm

Michael Jankowski

sod,

this paper has tree ring data up to 2003 and produces a hockey stick:
http://www.wsl.ch/staff/jan.esper/publications/Buentgen_2005_CD.pdf

My 10-minute glance, since you’ve posted this paper many times and deserve some sort of response.

The only thing close to hockey stick I see may largely occur due to the “padding” from 1980-onward. One would have to see the raw data.

I noticed the pine series shows no hockey stick in recent years, despite the author noting that the pine picked-up the MWP better than the larch ones (148, 4.1, 2nd para). And the pine series does not seem to be responding to the supposedly anomalous 20th century warmth, even though it seemed more sensitive in the past to temp changes than the larch series. So I’m curious if something else possibly (CO2 fertilization) might be driving the two strongest larch series (or larch series in general?).

I noticed the author in multiple locations (third para of 4.1, second para of 4.3, last para overall) notes the “insecurity” of the reconstruction near the endpoint. With the flat pine series ending in ’97, a relatively flat larch series ending in ’94, and one other series ending in 1915 that would’ve contributed who-knows-what to recent decades, that’s another caveat. The recent decades are dominated by the two larch series which show explosive growth, without the dampending one would expect to see if the other series continued.

But maybe that’s all getting nit-picky. After all, there are instrumental records in the area which show common trends, correct? Good. So what you may have in the end is tree proxies which suggest the summer temps in the alps are higher than they’ve been in over 1,000 yrs. The thing is, there are any number of regional proxies which suggest a MWP warmer than today. These are dismissed as “regional phenomenon.” So maybe that card has to be played here, too?

Nasif Nahle

Evidently, it did hurt them… 😉

steven mosher

RE 235.

That’s a cool idea. Upon request which writer JEG or Dr. Loehlm will post the data and methods

Dr. Loehlm will post his and JEG will post hers.

Maybe this one?

http://www.agu.org/cgi-bin/SFgate/SFgate?&listenv=table&multiple=1&range=1&directget=1&application=sm07&database=%2Fdata%2Fepubs%2Fwais%2Findexes%2Fsm07%2Fsm07&maxhits=200&=“A43D-07

Or this one:

http://digitalcommons.libraries.columbia.edu/dissertations/AAI3249076/

She can choose which ever. I think Ssadlov would love some of the ENSO stuff.

Andrey Levin

As a layman reader of CA, I would like to thank Dr. Loehle for this clear and straightforward paper. It IS very impressive.

No doubts it will get this kind of reception from the establishment:

http://www.snappedshot.com/archives/964-Professional-Protester,-Jihadi-style.html

BTW, it would be interesting to hear opinion of Dr. Loehle on the question of Steve Sadlov “…what the real sweet spot for CO2 concentration actually is…”.
Maybe later, on Unthreaded?

Re#59, Rob Wilson:

Thanks for inclusion of NH instrumental reconstruction in the graph. Once again it suggests that last 30 years instrumental global temperature reconstruction is a junk.

Nasif Nahle

“appreciate” and “plotted”, I’m really sleepy… sorry.

Louis Hissink

General comment on “peer review” – in the New Concepts in Global Tectonics Newsletter, No 44 Oct 2007

18 New Concepts in Global Tectonics Newsletter, no. 44, September, 2007
PRECURSORY LUMINOUS PHENOMENA USED FOR EARTHQUAKE PREDICTION –
THE TARO VALLEY, NORTHWESTERN APENNINES, ITALY
Valentino STRASER
94, Località Casarola – 43040 Terenzo PR, Italy
fifurnio@tiscali.it

ABSTRACT
Earthquakes do not occur randomly. They are the manifestation of natural phenomena ruled by definite physical laws. Prevention of seismic risk lies in the study and interpretation of areas under dynamic evolution. Earthquakes are associated with energetic manifestations that are not only mechanical, but also electric, electromagnetic and electrochemical in nature. Comparison of their recorded data as well as on-field research and interviews has enabled the building of a specific model to establish the frequency of events that might act as a guide to a more tailored definition of seismic risk for any area in the world. The results of this study refer to the Taro River Valley geo-structural complex and highlight a close link among the studied geophysical events, clearly showing them to be enforced by a ‘Code’. Interpreting such a Code allowed the successful forecast of superficial and low-average magnitude seismic events from 2005 to 2007. Forecasts were made several weeks in advance, and included the indication of the epicentre zone and the time lapse of the event.

So if the paper is scientifically sound, we in the renegade heretical camp of plate tectonic denialists, do make sure things get published. I suspect Craig’s paper would have been given the all clear flag in NCGT as well.

JS

Another comment:

Why are you smoothing the data at all? Particularly when you are taking an average which will tend to cancel out the high-frequency idiosyncratic noise anyway.

Smoothing is just one more step that introduces possible distortions into the results. By all means smooth the final result for expositional and presentational purposes. But there is no apparent need to smooth the individual series prior to the final result. Regardless, I think you need to be clear why you are smoothing and what you hope to accomplish by doing so. You then need to balance what you hope to gain by the smoothing (I would argue not much) against the costs in terms of possible artifacts and general confusion about what is going on (see, for example, all the discussion about ending in 1980).

Geoff Sherrington

Re # 194 JEG

(Mg/Ca ratios depend exponentially on seawater temperature, for instance).

Scientifically, this is an inexact statement. The movement of calcium (in whatever form) and magnesium (ditto) in water does not of necessity follow the same pattern of movement of temperature. It is not a sufficient test to sample a dead shell population and match Ca/Mg with temperature. It is required as well to show that any association persisted before the measurements and ideally, persists after the measurements.

I know of no inherent reason why a Ca/Mg ratio comprehends an exponential function. I do suspect that some methematicians or modellers sometimes jump to conclude that a non-linear response must by default be exponential, because it is the next most convenient of the relationship curves; but this objection is secondary to the danger of relating Ca/Mg to temperature – especially when each can exist in water in small biologic forms as well as inorganic ions and other forms of likely different behaviour re their uptake into shells.

So, JEG, where are your error calculations for the relation between Ca/Mg and temperature in your dogmatic assertion? What coverage of the globe has been tested with this assertion? etc etc. If you seek to criticise using examples, your examples must also withstand similar examination.

And yes, I did note that one of the many series cited was 3) Chesapeake Bay Mg/Ca (Cronin et al., 2003). I have read it.

UC

#88

the scaling makes little difference to your series except depressing the cooler period (e.g. the LIA) a little. The MWP levels are the same.

And this kind of distortion of (possible) signal doesn’t bother you?

Paul

Sorry,

Third line should of couse read

Craig has not mirrored…

carry on.

Will J. Richardson

(Trying one more time)

Dear Dr. Emile-Gaey,

Since you are presently working with Dr. Mann on a paper, could you please ask him to provide some assistance in resolving the issues presented on this Climate Audit thread? Code would be nice.

http://www.climateaudit.org/?p=2355#more-2355>

Regards,

WJR

Douglas Foss

Geoff #237

Two points continually recur: (1) climate currently warmer than last [supply interval], and (2) rate of change in climate temperature is greater than ever. Your comment is directed at x-axis considerations; that is that fewer data points measured on the x-axis means that the investigator may miss events occurring within intervals recorded on the x-axis. As you point out, one might miss a spiked fever and therefore not be aware of both the highest temperature and the fastest rise in temperature.

In RC comments, Gavin takes the position that considerable information exists for the current interval (forget the source for now) and that one can derive the relative rate of warming in the current interval. I thought that the entire point of paleoclimate research was to obtain information concerning past fluctuations to draw comparison with current fluctuation. Generally speaking, such analysis would afford information concerning ranges of temperature, particularly in the Holocene and other interglacials and relative rates of change. The first would allow some analysis of, among other considerations, whether the current warming is atypical and capacity of organisms and ecologic niches to adapt to changes. The second would allow, at a minimum (and in addition to the issues mentioned for the first) analysis of whether drivers other than CO2 have more or less dramatic impact on temperature and information from which to understand better how these drivers actually function (as opposed to how in theory they should operate).

What I wondered about Craig’s analysis (and I don’t know anything about the actual processes of smoothing) was whether his 18 sets of data points allowed an intelligible assessment of rates of change within the interval. Tree rings generate lots of x-axis information. For a variety of reasons Craig has less confidence in the accuracy of their y-axis data. So, he eliminated tree rings as proxies. I had thought that diatoms in sediments had considerable x-axis information (though bioturbation might generate doubt that we can find individual years in the sediments). And Craig was specific that other series were scant in x-axis data points (every hundred years potentially in some). But, even with smoothing and other operations performed upon the raw data, points on the graph where there are strong slopes should be reacting to y-axis and x-axis information (unless I completely misunderstand these procedures for teasing signal from data). This was the point of the question – given the way Craig proceeded, and the data he had, are the sharp slopes appearing between 800 A.D. and 900A.D. and between 1700 A.D. and about 1750 A.D. informative? It would seem that they would be on the reasoning that they must have reference to the underlying data and should not be merely an artefact of the statistical procedures employed, but I lack the sophistication to determine this.

Craig Loehle

the greenland data is ice core not borehole. typo. Why did I smooth before averaging? The data series have different degrees of resolution. Some are annual, but others have only points decades apart. If I did not smooth, the annual series would dominate the response. Because of the data gaps it is not possible to answer the question of Foss about rapid rates of change.

In constructing the GISS or CRU data, sites are averaged at each date. There is not much different about what I am doing than that, except that the underlying temperatures at each date are estimates from proxies instead of from a thermometer.

Douglas J. Keenan

Craig, nice paper. I have four suggestions (partially repeating some others above).

1. A table that lists, for each site, the following: lat, long, elevation, proxy, freq, refs, notes, data-url (if feasible).

2. An equal-area map of the world (e.g.), showing the locations of the sites.

3. Some consideration of the Slutsky effect.

4. An explanation of why you cannot provide confidence/likelihood intervals.

Regarding #4, it is obviously better to acknowledge that you do not know how to calculate a well-founded answer than to present an ill-founded answer. (I recognize that in the hurly-burly world of scientific publishing, that is sometimes forgotten, but that does not change the veracity.)

Regarding the bootstrap, what your paper does seems like a reasonable approach to me. It is not a correct bootstrap, though. If you are interested in bootstrap, a very good review is by Young G.A. et al. [Statistical Science, 9: 382–415 (1994)].

Proper bootstrapping requires modeling the structural interrelationships in the data. Our current understanding of the structure is poor, however; so I suspect that we cannot adequately do the modeling. In other words, I am skeptical that bootstrap can be reliably done with paleoclimate data.

I had some discussion about using bootstrap in dendroclimatology last year, on the denrdo email list:
http://listserv.arizona.edu/cgi-bin/wa?A2=ind0605&L=itrdbfor&T=0&P=2593
This was regarding the Precon dendroclimatological program, which relies on bootstrapping for its confidence intervals. In that discussion, the authors of Precon (dendrochronologist Hal Fritts and statistician Joel Guiot) came to agree with my comments.

bender

snip

Steve McIntyre

Here are perhaps relevant urls:

http://www.climate2003.com/data/moberg/djgrip.txt #email version
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/pollen/recons/liadata.txt
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/cronin2003/cronin2003.txt
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/keigwin1996/fig4bdata
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/nyberg2002/nyberg2002.txt
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/speleothem/china/shihua_tan2003.txt
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/yang2002/china_temp.txt
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/speleothem/africa/cold_air_cave.txt
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/demenocal2000/658_sst.txt
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/farmer2005/farmer2005.txt
http://doi.pangaea.de/10.1594/PANGAEA.438810
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/stott2004/stott2004.txt
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/pollen/recons/northamerica/viau2006namerica-temp.txt
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/historical/china/china_winter_temp.txt
http://doi.pangaea.de/10.1594/PANGAEA.438838

rhodeymark

Re #232 – Thanks John, that was illuminating. Just as Mr. Gore has held workshops to raise up a different sort of proxy to go forth and evangelize, here we see what purports to be a critical examination of the HS presented by an associate of Dr. Mann. It’s cute how young JEG and Cobb are, I hope they haven’t been coddled in their ascent. I found it surprising when it was revealed how fresh Mann’s doctorate was when he was placed in such a position of prominence, but I suppose that in such a rapidly evolving field it shouldn’t be surprising to have new experts baked fresh daily.
I look forward to the continuation and if need be, evolution of Dr. Loehle’s work. I’m grateful that he saw fit to roll up his sleeves in an area where there was an acknowledged need, and trust that it will spur on more beneficial work. One thing I really enjoyed about this thread was Douglas Foss in #57 offering historical insights along the timeline. CA The Movie will absolutely need historians rebutting the current consensus concerning the “global” prosperity of warm vs cold periods. It would seem the coolings inspire a much nastier sort of diaspora, historically. I certainly don’t want the Canadians showing up at my door with knives in their teeth 🙂

Nasif Nahle

# 111

UC,

Introduce a polynomial trend for each plot (Craig’s and Mann’s) and the similitudes will vanish before your astounding eyes. 🙂

Bob Meyer

Re 253:

JS said “It also means you can plot the data all the way to 1995.”

True, but I suspect that the averaging was done first because the x-axis points were not the same for all 18 data sets. Using a moving average filter first can serve as a kind of interpolation to make it possible to average the data sets later.

Unless I completely misunderstood the section on method, this is what Loehle did. In his paper he states

Data in each series had different degrees of temporal coverage. For example, the
pollen-based reconstruction of Viau et al. (2006) has data at 100-year intervals. Other
sites had data at irregular intervals. This data was taken as is without interpolation.
Data in each series were smoothed with a 30-year running mean. This should help
remove noise due to dating and temperature estimation error. If data occurred every
100 years, each point would be stretched by the smoothing to cover 30 years.

I don’t have any experience with comparing relatively sparse, non uniformly spaced data sets so I don’t know what the pitfalls are, if any. I do know that if you use higher order interpolation, (like quadratic or spline) it is very easy to introduce weird spurious high frequency components.

Kenneth Fritsch

With the paper under discussed here being organized and constructed such that the reader can readily follow what Craig has done, does make it a target for criticism and criticism that can be rather quickly formulated and without a deep understanding of statistical methods.

While it may show what some posters here want to see, I would not personally want to suspend my skepticism of the lack of effort in determining the uncertainty in the results – anymore than I would do this for those reconstructions arguing for AGW. I do like Craig’s bootstrap and jackknife methods of sensitivity testing that seems to go further than those methods used by Mann in his and his progenitors’ reconstructions. Beyond this processing one must have yet some testing to do on the appropriateness of using any or all of the reconstructions. It is those individual series used in the reconstruction that should then become the focus of the discussions.

If the purpose of the paper was to show that a series of temperature reconstructions could be averaged together with the singular a prior for selection (of not being tree rings based) I would say that Craig has presented a counterpoint for those series that have been put together to show a trend more readily attributable to AGW. Rather obviously he has not gone beyond that purpose to do any detailed statistical analysis (beyond what in general other temperature re-constructers have effectively done given that after the fact rationalizations are not judged acceptable and stating a confidence limit does make it a valid one) to show that his reconstruction is more correct than lets us say the HS.

It should be an easy exercise for others here, or Craig, to use the data and manipulate it in manners as they see fit for more detailed analyses; something that has been much more difficult when dealing with other reconstructions discussed here. A good start as some have suggested already is to average the data on yearly basis before doing any filtering.

The criticism of not determining how well the series gives a measure of global coverage and thus global average temperatures is valid and can be made of other reconstructions –given that an edict that they are global in character or regionally representative must have some evidence to back it up. I am almost to the point of doing a study of extracting sets of 18 randomly selected stations from the 1221 USCHN stations and determining how much the averages will vary over various time periods from a year to decades of years.

Once again I am disappointed that a number of those who come to visit this blog spend too much time on describing, in not always polite terms, what they see as a generalized POV here and too little enlightening us with their counterpoints.

Craig Loehle

re: Bob Meyer #170 you are correct

Costard

Kudos Craig for doing this.

I echo the suggestions about showing the global distribution of the sites. You might show (or list) the proxy locations, and introduce a fourth graph, weighted by hemisphere, or hemi-hemisphere, or whatever seems most rational to the firing squad. I suppose would be a model and would require some space given to discussion of confidence/verification/etc., but then it’s really beginning to seem that you’ll need to address this topic anyway. Maybe bender has some ideas for arriving at comprehensive error bars. They may not be “necessary,” but if they’re at all possible, they would be wise.

P.S. Welcome to pandora’s box.

Larry

Susann, 279:

1. How does this article stack up against other articles in the same area? Is it good science?

2. What are the flaws/limitations in the article?

3. What is new, innovative?

4. How does it add to the literature?

5. As a policy analyst…

You ask some excellent questions. I hope they get answers. Here’s what you’re up against:

274, Kenneth Fritsch:

With the paper under discussed here being organized and constructed such that the reader can readily follow what Craig has done, does make it a target for criticism and criticism that can be rather quickly formulated and without a deep understanding of statistical methods.

In other words, the scientist who doesn’t want the GP to be looking and questioning will write in the most opaque manner possible. Furthermore, the reviewers will also use codespeak for the same reason. I hope what happens here on this blog isn’t that, but this is a work in progress. We don’t know how this is going to turn out. Stay tuned. I hope you’re pleasantly surprised.

richardT

#275
Craig
Caribbean Sea 18O (Nyberg et al., 2002) is not calibrated as temperature. This core also has a foram-based sst reconstuction. Did you use this? If so which season?

Indur Goklany

Off-topic but in response to Steven Mosher #121: Steven, more importantly, despite any warming, richer-but-warmer worlds may well be better off than poorer-but-cooler ones even if one agrees with the Stern Review’s estimates that warming would/might reduce GDP by 5-20% “now and forever” — especially for developing countries who are deemed to be most-at-risk of climate change because they lack the fiscal and human resources to cope with any climate change.
See: Is a Richer-but-warmer World Better than Poorer-but-cooler Worlds?” forthcoming in Energy & Environment, vol. 18, nos. 7 and 8 (2007). Abstract follows:

Greater economic growth could lead to greater greenhouse gas emissions, while simultaneously enhancing various aspects of human well-being and the capacity to adapt to climate change. This begs the question as to whether and, if so, for how long would a richer-but-warmer world be better for well-being than poorer-but-cooler worlds. To shed light on this issue, this paper draws upon results of the “Fast Track” assessment (FTA) reported in a special issue of Global Environmental Change: Part A 14(1): 1-99 (2004), which employed the IPCC’s emissions scenarios to project future climate change and its global impacts on various determinants of human and environmental well-being. Results suggest that notwithstanding climate change, through much of this century, human wellbeing is likely to be highest in the richest-but-warmest (A1FI) world and lower in poorer-but-cooler worlds. With respect to environmental well-being, matters may be best under the A1FI world for some critical environmental indicators through 2085-2100, but not necessarily for others. An alternative analysis using the Stern Review’s worst-case results for potential welfare losses due to climate change indicates that welfare, adjusted for market and non-market impacts of climate change and the risk of catastrophe, should be highest under the A1FI scenario, at least through 2100.

Craig Loehle

There are several requests for a map of the locations. The coordinates of all sites are either at the links or in the publications. Can someone map this please? I don’t have the software. Note that one site represents North America and one China, so not all are points.

Indur Goklany

Re #286: Link to paper: http://members.cox.net/goklany/Richer-but-warmer%20RV.pdf

david abrams

“As a relative layperson, I’d like to point out how refreshing it is to see a proxy study with clear inclusion criteria (2000 year, 20 points) and exclusion criteria (tree rings).”

I totally agree with this. I’m not a climatologist but I’m an attorney and people lie to me all day long. You could say that I have PhD in BSology. It’s troubling the way that some of the research in these areas seems very open to the possibility of cherry picking and other kinds of data manipulation.

rhodeymark

Susann – I would think that #5 is the crux of the biscuit. The way I distilled this paper is that if a reconstruction that shows great paleoclimatic variation can withstand scrutiny, then there is a) definitely one forcing agent that can’t be blamed, and b) other forces that are currently being given an entirely too short shrift.

Costard

sod:

ending the series in 1995 of course is rather a CONVENIENT thing.

it removes about 7 of the 10 hottest years EVER from the record.

if i did a 2000 year of computer technology analysis, using your methodology and ending around 1960, i might end up with the conclusion that some medieval computing machine was the high point of achievement!

the misrepresentation at the end is already quite bad, by using 1995 as last year. it gets worse by averaging it to 1980. but if a significant number of your proxies have their last temperature data BEFORE 1995 (and i doubt that Conroy Lake pollen (Gajewski, 1988) includes 1995 data..), the “smoothing” gets even worse.

First of all, Craig has made a compilation of proxies – not modern measurement data. Please reference any proxy which has shown “7 of the 10 hottest years ever” since 1995. Presumably, this would have a 2000-year span, ending in, say, 2006.

Furthermore, since most proxies, especially non-treerings, don’t (accurately) show individual years, they’re probably not going to show anomalies like 1998. As you point out, even the length of the MWP is highly ambiguous. So the proxies may not even show a short peak like 1980-present. Smoothing is a reality of the data.

Since what you seem to be saying is that global warming requires the measured warming of the last two-and-a-half decades for much of its impact, as well as the high frequency of modern data – i.e. the “hottest years” – doesn’t that seem to imply that modern data can be misleading when we’re talking about long-term, millenial temperature reconstructions? Or, similarly, when we’re talking about a climate system that largely follows even-longer-term trends? It seems clear to me that the closer one is able to look, the more the noise will dominate, and so differentiating between noise and a “rapid, unprecedented” rise in temperatures will be difficult.

Perhaps non-centered smoothing, and so extending the graph to 1995, wouldn’t be a bad idea. Others would know better than me.

richardT

#292
Craig
In your paper,you write that
“After an extensive search, all data were used that had at least 20 dates over the 2000-year period”
If you scroll slightly past the data for Conroy Lake on your link, there is a data for another lake, Hell’s Kitchen, that fits your inclusion criteria.
While the reconstruction for Conroy Lake is much warmer in the MWP than present, Hell’s Kitchen’s is not. Cherry picking?

Susann

Steve, I agree that I don’t care where the data leads, as long as it leads to the facts. I wouldn’t agree that climate reasearch is junk although there is and always will be junk amidst the gems. It’s inevitable, but it can be minimized. We humans are political animals, our training in the scientific method notwithstanding. Scientists are expected to take into account errors and bias in their research design and questions but very often, they are completely unaware of their personal political biases and how those affect every step of the research process from choice of topic to forumlating scientific problems to evaluating findings. And, as in every walk of life, some are outright frauds.

Of course, that said, the scientific method is really all we have, isn’t it? I certainly don’t want to chuck it over. I am all for rigorous oversight and review with as little political interference as possible. We can’t expect issues as potentially politically explosive as climate change and global warming to escape politicization, and so being aware of it and trying to counter it is all the more important when it comes to developing policy. Policy development is hobbled by bad science — like computers, it’s GIGO.

Steve Huntwork

Once upon a time, in a land far away:

In 10th grade High School, my science teacher had us do an experiment. He threw a weight attached to a long stream of paper off the school roof, and an electrical buzzer would make marks upon the paper as it moved. Our requirement as students was to measure the change in the buzzer marks on the paper and deduce that gravity caused the weight to fall at 9.8 meter per second squared.

I got an F for that experiment, because the fall of the weight was NOT 9.8 meters per second and I refused to give the expected answer.

However, my science teacher absolutly loved me for calculating the coefficent of drag for a weighted strip of paper, as a function of it’s length. We both learned something that day.

Why does anyone AVERAGE data? The real science is in the details and why it does not conform to conventional theory. When I read any report that only presents averaged or smoothed data, I throw it in the trash can.

I want the details, and want to know WHY this data set is not conforming to the conventional theories.

That is what honest science is all about!

Nasif Nahle

# 300

DocMartyn,

I have the series for ISI based on 10Be and Ca from NOAA. They published them here. Plot it along the delta TT series and you’ll get a nice surprise.

JS

Re #258 and #270

Craig, I think you need to be clear about two things then – interpolation and smoothing. These fulfil different functions and combining things can muddy the waters. As Bob noted, with linear filters the order of application doesn’t matter. You need to interpolate data to annual frequency before you average it – but that is only necessary for some series and to different extents. Indeed, how are you applying a 30-year centred moving average to data with less than annual frequency? While you state that the data were not interpolated, you also state “If data occurred every 100 years, each point would be stretched by the smoothing to cover 30 years.” This comment seems to skip a few steps and isn’t entirely clear to me. How have you gone from one point every century to something that you can a apply a 30-period moving average to? You must be doing some interpolation prior to that step. More exposition of how you applied a 30-year smooth to decadal or centennial data would be beneficial. The key here will be the interpolation, not the 30y MA.

I’m not sure why you are concerned that “If I did not smooth, the annual series would dominate the response.” – any high frequency idiosyncratic noise will be attenuated by the averaging. And, as Bob noted, it will be attenuated to exactly the same extent if you only smooth (linearly) after computing the average. What process do you imagine you would apply to average the series if you didn’t smooth (but obviously interpolated)?

Gary

#279 Susann, here’s an attempt to answer your questions.

1. How does this article stack up against other articles in the same area? Is it good science?

It’s a ‘methods’ paper addressing criticisms of paleo-temperatures reconstructions from proxies using tree-rings. It doesn’t use new data; however, unlike many recent reconstructions, it sets out selection criteria a priori to minimize ‘cherry-picking’ bias. That makes it ‘better’ science. It’s clearly written, concise, and doesn’t speculate beyond the data.

2. What are the flaws/limitations in the article?

Lacks metadata on the proxies, especially a map or table of the locations. There may be other statistical methods that could be used (but this is beyond my competency to comment on).

3. What is new, innovative?

Not particularly new in a general sense, except that this sort of analysis seems to be regarded as unnecessary in dendroclimatology unlike in many other disciplines where methods are scrutinized more severely.

4. How does it add to the literature?

It’s additional evidence against using treerings as temperature proxies. This sort of thing should have been done long ago to evaluate the validity of treering data. It adds to the literature because it comes so late in the game and questions some assumptions that have a higher likelihood of being wrong than once thought.

Nasif Nahle

# 308

Steve Huntwork,

Yeah! That’s the reason I don’t trust GCMs. Oh, dear! GCMs are so vulnerable… Heh! 😉

Oops! A thunderstorm here in November? Wow! That’s a deep “climate change”…!

bender

Dr. Loehle,
Do you have a volunteer yet to do the mapping of the proxy co-ordinates? I’ll do it if no one else wants to. Let me know ASAP.

Steve Huntwork

Any GCM model that does now show a warming of the global Earth’s temperature between January and March, is false.

Who can tell us why?

Steve Huntwork

Is the Earth warmer in January or July?

Is the Earth closer to the Sun in January or July?

If the proxy or instrumental data does not show such an obvious fact, why not?

There is only one possible way that “global temperatures” could be warmer in July. Why is something so obvious being ignored?

Steve Huntwork

For 30 years, I have always asked some rather simple questions:

1) Acid rain: Why where the trees dying at the top of the hills, instead of in the vallies where human produced pollusion would be concentrated? Today, we know that they were dying of old age.

2) Ozone hole: Why does nobody realize that ozone is created from solar radiation, and for six months out of the year, the polar regions do not get any sunlight. In the spring, when solar radiation begins to illuminate the polar regions, you get a VERY HIGH concentration of ozone around the poles? I call this the ozone donut, but everyone focused upon the hole in the middle of the donut and ignored what was right around it. Today, we now understand that the Sun creates ozone.

3) Global warming: Why does nobody ask why the historical records show a maximum global temperature in July instead of January? That is not possible, because of the Earth’s orbit.

Oh well, the obvious is always rather difficult for some people to understand….

bender

Gaping holes in the network in South America, Russia, India, Antarctica. Still, nicely dispersed.

richardT

#327
That site is Conroy Pond. I shaded N. America and China to represent the three regional reconstructions.

bender

Re #331 Fundamentally flawed? Prove it.

Steve Huntwork

Avid Reader:

http://epod.usra.edu/archive/epodviewer.php3?oid=196160

Did you actually view that? Do your own astronomy search on the subject of Earths orbit around the Sun.

Ok, now I will shut up….

Steve Huntwork

@333

Not a bad “global” sampling of data….

bender

Hans Erren #333. Not quite. Check your quotes. China, US, Canada, should all be gray. Please fix & repost? Thanks.

Craig Loehle

Bender, yes please do plot the data I have no volunteer yet.

Smoothing question. For data with equal intervals (all annual, for example) it does not matter if you smooth and then average or converse. But when one data point in one series is only every 100 years, an annual data series will completely dominate the signal. Try it. There is not perfect way to deal with this type of data. Interpolate? Linear filter? Gaussian filter? Each has pitfalls and will create distortions of the underlying data. Create some data where different series are sampled at different intervals and try diff methods– will get diff results.

steven mosher

RE 331.

Hmm.

“#73 Craig: “E&E review was as rigorous as for any other journal I’ve submitted to (of my 107 pubs).”
There’s a real lesson here. Some CA regulars are quick to defend E&E,
but the fact that E&E accepted Craig’s paper when JEG so quickly showed it to be fundamentally flawed is
damning for the journal”

I think that would be damning for ALL journals. If you accept Craig at his word,

Further, JEG didnt show it to be fundamentality flawed. JEGHEAD reraised issues raised by
everyone else, notably bender, a full day later. He added the twist of being snide.

By your logic then if bender or JEGHEAD or Anyone can criticize a paper in a day, then the journal is bankrupt?
Seems fair. I like that logic. Let’s see who survives:
HEY BENDER take a wack at Nature and Michael Mann.. or Willis take a wack at Juckes and climate of the Past.

richardT

retry
library(maps)
map()
map(regions=c("Canada","USA","China"),col="grey70",fill=T, add=T)
pos=list(GRIP=c(73, -38),Conroy=c(46, -68),Chesapeake=c(38, -76),Sargasso=c(33, -57), Caribbean=c(18, -67), Tsuolbmajavri=c(68, 22), Shihua=c(40,116),Cold_Air_Cave=c(-24, 29.2),ODP_658C=c(20.75, -18.58),ODP_1084B=c(-25.5, 13),MD952011=c(66.967,7.633),MD98_2181=c(6.3, 125.83),MD98_2176=c(-5, 133.44),SSDP_102=c(34.9530, 128.8810), Spannagel=c(47.0882, 11.6715))
pos2=t(sapply(pos,function(p)p))
points(pos2[,2],pos2[,1], col=2, pch=16)
text(pos2[,2],pos2[,1], col=2, label=rownames(pos2), pos=c(2,2,2,2,2,4,2,4,2,2,2,2,2,2,2), cex=.7)


DaveR

#332, 345 etc. re: “fundamentally flawed”, JEG put it neatly, if rather unkindly, thus in post #194:

…isn’t it staggeringly misleading to call this a ‘global temperature reconstruction’ ? (did i misread, or is it the title of the article ?). The adjective ‘Global’ implies representativeness of an area-weighted average, which it is not. Now, it may turn out that your spatial sampling does a reasonably job at representing tropics, mid and high-latitudes , but it seems a little careless not to ensure it, and even more unforgivable no neglect mentioning it.

Craig – it would be very interesting to see the reviewers comments from E&E. Did any of them pick up on this?

sod

First of all, Craig has made a compilation of proxies – not modern measurement data. Please reference any proxy which has shown “7 of the 10 hottest years ever” since 1995. Presumably, this would have a 2000-year span, ending in, say, 2006.

NASA-GISS.

read it over at Tamino:
http://tamino.wordpress.com/2007/11/10/temperature-2007/#more-476

Furthermore, since most proxies, especially non-treerings, don’t (accurately) show individual years, they’re probably not going to show anomalies like 1998. As you point out, even the length of the MWP is highly ambiguous. So the proxies may not even show a short peak like 1980-present.

this is obvious: using several different papers, all of which have some “MWP” peak somewhere between 800 and 1200 will create some sort of a plateau there.
his approach is HELPING the MWP thesis, which so far fails to give a period that is exact up to a century even!

on the other hand, his methodology REMOVES the last 25 years, basically enforcing that the current spike will not show up.

do you think it makes sense that he calls it a “2000 years temperature reconstruction” and leaves out the last 20?

those years that are so important, that they fill countless blog pages?

Smoothing is a reality of the data.

i am not 100% sure of this, but others here know much more about statistics than i do.

Since what you seem to be saying is that global warming requires the measured warming of the last two-and-a-half decades for much of its impact, as well as the high frequency of modern data – i.e. the “hottest years” – doesn’t that seem to imply that modern data can be misleading when we’re talking about long-term, millenial temperature reconstructions? Or, similarly, when we’re talking about a climate system that largely follows even-longer-term trends? It seems clear to me that the closer one is able to look, the more the noise will dominate, and so differentiating between noise and a “rapid, unprecedented” rise in temperatures will be difficult.

Craig decided to make a comparison between the MWP and present, not me.

One persistent question is whether the MWP was “really” warmer than the end of the 20th century.

it is hard to assume, that this paper was meant as anything but ammunition against the hockey stick. it should at least try to include the hockey stick period.

Perhaps non-centered smoothing, and so extending the graph to 1995, wouldn’t be a bad idea. Others would know better than me.

good first step, yes.

“If data occurred every 100 years, each point would be stretched by the smoothing to cover 30 years.” This comment seems to skip a few steps and isn’t entirely clear to me. How have you gone from one point every century to something that you can a apply a 30-period moving average to? You must be doing some interpolation prior to that step. More exposition of how you applied a 30-year smooth to decadal or centennial data would be beneficial.

well said. another important information missing. why did you chose 30 year period anyway?

It’s a ‘methods’ paper addressing criticisms of paleo-temperatures reconstructions from proxies using tree-rings. It doesn’t use new data; however, unlike many recent reconstructions, it sets out selection criteria a priori to minimize ‘cherry-picking’ bias. That makes it ‘better’ science. It’s clearly written, concise, and doesn’t speculate beyond the data.
..
It’s additional evidence against using treerings as temperature proxies. This sort of thing should have been done long ago to evaluate the validity of treering data. It adds to the literature because it comes so late in the game and questions some assumptions that have a higher likelihood of being wrong than once thought.

sorry, but this is pure fiction!
1. we have nothing but the word of the author on the non-bias criterias.
(end date 1980 sounds like a LOT of bias to me..)

2. if it was “better science” it would not be published in E&E

3. it speculates a LOT beyond its data, mainly about a period that it has none on.

4. you prefer to speculate beyond the paper as well? or is the assumption you are speaking about not the hockey stick?

Steve Huntwork

Craig:

Like it or not, the JEG comments have helped you create a better data analysis. I for one, have been very impressed with how rapidly you have attempted to answer those questions.

That is what honest science is all about.

bender

Whether the paper is “fundamentally flawed” or not (and it’s not) is irrelevant because the author has ‘moved on’. Seriously. One by one the criticisms are being addressed (mapping, documentation, turnkey scripting, etc.). Eventually all of them will be addressed and then we will see whether CWP is “unprecedented”.

As for statistically robust confidence intervals – the one real sticky wicket – let’s just use MBH99 method. Oh, yeah, I forgot. We don’t actually know what that method was. They’re unaccountable for their methods.

Now, Li et al. (2007) claim to have a methodological solution, but guess what: JEG says they’re not interested in trying it on the Loehle (2007) proxies. That’s one hell of a solution.

JEG, you can’t have it both ways. You can’t own a monopoly on the methods and expect your work to run the public policy agenda. It’s undemocratic. Take your choice: monopoly on methods OR sharing data to help drive policy.

steven mosher

333.

since thermometers are teleconnected at a range of 1200KM ( hansen) It would be kinda
neat to see 1200km rings around the sites..

bender

Re #354
Sure, why not … with the correlation field.

kim

Did you know you have a thread almost 1300 comments long at Pharyngula named after you, Stan?
=============================================================

Kenneth Fritsch

Re: #282

You ask some excellent questions. I hope they get answers. Here’s what you’re up against:

274, Kenneth Fritsch (said):

With the paper under discussed here being organized and constructed such that the reader can readily follow what Craig has done, does make it a target for criticism and criticism that can be rather quickly formulated and without a deep understanding of statistical methods.

(and you, Larry, responded)

In other words, the scientist who doesn’t want the GP to be looking and questioning will write in the most opaque manner possible. Furthermore, the reviewers will also use codespeak for the same reason. I hope what happens here on this blog isn’t that, but this is a work in progress. We don’t know how this is going to turn out. Stay tuned. I hope you’re pleasantly surprised.

I am not certain how you have interpreted what I said, but what I meant to say was that Craig has written an article sufficiently clear in descriptions of what he has done that there will be no delay for many interested parties in the GP to comment on it and critique it. In fact he has made it possible to rather easily repeat what he has done and do it with other approaches. I am in effect applauding what Craig has done here and contrasting it to what was done with Mann and other Mannian like reconstructions.

At the same time I am reserving any judgements on the validity of the series that went into the reconstruction, its global coverage and the appropriateness of the methods used for averaging until more analyses are completed and questions asked. As you and Craig have stated, the article is a work in progress and the approach he has taken will allow active participation and analyses at this blog. If we can avoid the inclination to get personal, Craig has initiated something that can be really fun and a great learning experience at the same time.

Craig Loehle

Sod: the proxies available do not go past 1995. If you want to account for that, add a few tenths at the end. I wasn’t playing with the data. Where did I get 30 years? It is a convention in climate studies, but people use other periods also. If there wasn’t bias by reviewers, I wouldn’t have to publish in E&E. You seem to misunderstand the question of bias. Mann’s method picked out proxies (tree rings) that had a hockey stick shape and weighted them very very heavily in the averaging of series. THAT is bias. Other people seem to pick tree series or ice core data that they like and ignore others. THAT is bias.

Larry

360, actually, what I was really trying to get across with the “work in progress” comment is that the process of a pre-review by blog is a new concept, and while I have high hopes, we really won’t know how this works out until it’s tried a few times. What I’m hoping is that it’s not only going to be effective in working a lot of the bugs out prior to submittal, but it’ll also subject the paper to a lot of eyes that would otherwise not see them; i.e. statisticians.

A side benefit is the general public gets to watch the process from the bleachers, and can even interact (though that’s the part that could get out of hand).

Nasif Nahle

# 350

Steve Huntwork,

I agree. Craig has answered each questioning and uncertainty with that professionalism that it’s hard to find “somewhere else”. Craig Loehle has not hidden his data and methodology. I would like that others “somewhere else” followed the example of real science given by Dr. Loehle and the aperture demonstrated by Dr. Steve McIntyre to all kind of arguments. Steve McIntyre gives us the opportunity to explain our arguments and to debate the arguments from others.

I know that he is odious to make comparisons, but as I passed through the experience of being banned and I couldn’t defend my work, I think that I have the right to make the comparison. One of my papers was strongly criticized in Real Climate. I did try to support my paper scientifically, but they did not allow me to express a word to make the pertinent explanations on the paper. They even made fun from quantum mechanics, and I could not refute their odious mockery because they did not allow me to do it.

Thus, open your eyes, Julien. You demanded and pushed Craig to reveal even the minimum detail of his investigation, and soon you ended here to express badly from all those that displeased you; nevertheless, Steve McIntyre allowed you to speak and write your drolleries. Can you see the difference?

Larry

363, which raises a question in my mind: if the MWP and LIA are visible in the reconstructions, and the current warming isn’t, could we be underestimating the magnitudes of the MWP and LIA? It seems like either that has to be true, or the current surface data are exaggerated.

Craig Loehle

According to my 2005 paper, the magnitude of the MWP may be underestimated due to dating error.

rhodeymark

Dr Loehle – RichardT asked about Conroy Lake vice Hell’s Kitchen in #299. Since it seemed relevant to the above, was there an answer that I missed?

rhodeymark

Darn, bad timing. I hope you have some well deserved relaxation.

Henry

In terms of “warmest ever” the final paragraph

One persistent question is whether the MWP was “really” warmer than the end of the 20th century. Even keeping in mind that Figure 1 shows 30-year running means, it would indeed seem to show the MWP to be warmer than the late 20th century. The eighteen series used here show a mean difference of about 0.3°C between the MWP and the 20th century (range of 0 to 0.6°C difference over the periods). It must be emphasized, of course, that this result is based on limited data.

seems to use some weasel words; there is a difference between the 20th century as a whole and the late 20th century.

Steve Huntwork

#366 Nasif Nahle:

“One of my papers was strongly criticized in Real Climate. I did try to support my paper scientifically, but they did not allow me to express a word to make the pertinent explanations on the paper.”

My profesional expierence of writing scientific publications, was for the Atmopheric Sciences Labratory of the U.S Army. My specialty was in Lidar measurements of obscurants. Global warming was able to fund some of the research that I did, but not once did my data support this concept.

Oh, I was smart enough to word my formal data reports with the proper language to make the people who were paying for the project happy, but the raw data was rather neutral.

During the first Gulf War, I used the DMSP satellite data to predict what was happening in Iraq. For some strange reason Saddam was not willing to share accurate weather data with the rest of the world, so I had to rely upon our satellites.

Ok, the DMSP satellite data was fantastic in a time of war, but would I ever use it as an absolute calibration for everything else? Heck no!

The MSI sensor on the DMSP satellites measures the illumination of multiple CO2 bands. If you know the pressure and temperature (using radiosond measurements from a balloon), then you can convert the MSI CO2 data into a temperature or pressure profile.

During the first Gulf War with Iraq, this is all we had to work with, and the MSI data was absolutly invaluable.

But to use the MSI or SSI satellite data as a “standard” for today’s climate analysis, does make me want to cry.

Sometimes, people need to step back and take a deep breath. What exactly are you actually measuring?

REALITY CHECK!

Hans Erren

I think the PC title of the paper should have been:

An almost 2000-year global temperature reconstruction based on mostly non-treering proxies.

😀

Bob Meyer

Nasif # 366 said

“One of my papers was strongly criticized in Real Climate. I did try to support my paper scientifically, but they did not allow me to express a word to make the pertinent explanations on the paper.”

Sometimes I think that the “Real” in “Real Climate” is used in the Old English sense of “royal”. RC sure does act as though they were royalty dealing with serfs. Theirs is to expound, ours is to acquiesce and obey.

It’s gotten too hard for me to wade through their stuff. I let others do that and then I follow their lead if they find something that sounds interesting. I wish that their site was completely without value, then I wouldn’t have to go there at all.

kim

‘paucity’, perhaps?
===========

Jaye

Regarding the JEG and others of the “RC” ilk, I just wish they would be honest with themselves by using the word that they are really thinking…heretic. Just say it guys…it’ll make you feel better.

bender

Doesn’t this assume a linear relationship between the temp and the proxy even if your just using anomalies?

Yes, good point. But that is what the Team typically is willing to assume. That doesn’t mean it’s right. My point is that unless the nonlinearity is severe the autocorrelation structure is not going to change much. With an upside downquadratic tree-ring response, however, you have a problem.

This is precisely why JEG and I both suggested that some significant discussion is required of underlying proxy accuracy.

JS

#340
“But when one data point in one series is only every 100 years, an annual data series will completely dominate the signal. ”

I’m not sure what you really mean by this. A one in a hundred year observation is unlike an annual series and to treat it like an annual series with 99 missing values is wrong. The most obvious interpretation is that the one in a hundred year observation represents the average of the century. To convert to annual frequency you just put that average in for every year in the century. If the underlying data generation process is different you might do something different.

You seem to be suggesting that you give 1/100th the weight to the centennial series when you average. That is obviously silly. You have to interpolate in some way. The simplest approach based on the approach alluded to above is to use the same data point for all annual observations +/- 50 years from the data point you have (or the 100 years before depending on the way the data are collected). You then have something that looks like a step function that changes every hundred years. As you mention, there are alternatives.

But regardless, you must have done some sort of implicit interpolation to convert one observation every 100 years into an annual series (smoothing or no smoothing). So what interpolation have you used for these series? If you set out the process you used to convert 100-year data to annual data our conversation will be easier and the apparent misunderstanding will likely be resolved.

Jay Currie

Susann, like you I come to CA from a public policy perspective. If you hunt around CA you will find more analysis of tree ring series than you could possibly ever need. The take away: treerings are potential proxies for a variety of variables: drought, temperature and actual CO2 levels. But, as proxies they have significant and salient problems.

Given that tree ring proxies lie at the heart of the IPCC analysis and the Gorian alarmism and that these views are driving public policy vis a vis climate change it makes a lot of sense to look at them closely. Craig’s article is an attempt to use another, ringless, set of proxies to see where the climate may have been. As such it is a very valuable baseline in assessing the value of the tree ring mavens’ work and, with that, the plausibility of the IPCC’s proposals. (There is no point in looking at Gore as his work is entirely derivative and is, after all, show biz.)

Susann

Jay, I will poke around a bit more to find the tree ring posts. My inclination would be to conclude that excluding an entire proxy type might be a problem, esp. if in doing so, the results are more “acceptable” for one side — just as including it might lead to more acceptable results for the “other” side. In other words, not including tree ring data results in a MWP while including them eliminates or minimies MWP. Why is the MWP important? (I think I know the answer, but want to work through the logic.) This seems to be one of if not “the” focal point of this paper and much of the debate. Should it be?

Steve, sorry to appear too sensitive.

Hu McCulloch

I must agree with Bender (#16, 87), and Julien Emile-Geay (JEG #143) that Craig Loehle’s paper would have been greatly improved with some standard errors and confidence intervals. We can’t say that his MWP is really warmer than his LIA, eg, until we see some error estimates.
Craig (#73, 178, 214) questions how to do this when his proxies have divergent or even unreported measurement errors themselves. This is a good question, but turns out not to be a big problem.
Suppose we have n observations y_i = mu + e_i, but var(e_i) = v_i, which is different for every observation but unobserved. Craig is estimating mu with the simple average y-bar. Assuming independence, var(y-bar) = (1/n^2)Sum(v_i). As long as the v_i’s are not too wildly distributed, Sum(v_i) is consistently estimated by the Sum of Squared Residuals SSR = Sum(y_i – y-bar)^2. Therefore the traditional formula sqrt(SSR/(n*(n-1)) will consistently estimate the standard deviation of y-bar about mu, and will provide asymptotically N(0,1) t-tests (or “z-tests” if you prefer) for hypotheses.

These t-stats will not, however, exactly have the classical small-sample t-distribution in this heteroskedastic (unequal-variance) case, because SSR will not have the required chi-squared distribution, even if the e_i’s are Gaussian. Nor will the traditional degrees of freedom adjustment (n-1) be precise in small samples, since this also requires homoskedasticity (equal variances), if not Gaussianity. So Craig should go ahead and report the traditional standard errors, with the equalification that they are only asymptotically valid because of the heteroskedasticity introduced by the unequal (and unnown) measurement errors of his sources. His webpage cited in the paper (www.ncasi.org/programs/areas/climate/Loehle#&#2007.csv) would be a better place for this suplementary material than here, however, since the reader of his paper would not know about this website, let alone this thread.

If Craig had good estimates of the errors of his source series, he could quantify the total error variance and then use Weighted Least Squares (WLS) to efficiently estimate mu and compute precise t-stats. Meanwhile, the above is good enough.

Craig (#150, 164, 178) is correct, however, to reject Julien’s demand (#143) for indicators of fit like R^2. R^2 measures how much better one’s regression model fits than a model with just a mean and no regressors. Since, as Craig points out, he (unlike his sources) is not fitting a regression, his R^2 is identically 0 and not worth reporting. The fact that Julien insists (#194) anyway demonstrates either that he does not understand basic statistics, or else that he is just trying to flimflam his audience. Perhaps this works for him over at RC.

[snip]

nanny_govt_sucks

To get “Global” coverage for the proxies in this study, I think the way to do it is to simply claim that they are teleconnected to global average temperatures. Actually, then you only need to include one proxy to claim you have a “global” reconstruction. Teleconnections make things a lot easier.

Jim Clarke

Sod,

You seem to be fixated on the last 20 years of the data, ignoring the fact that the main problem with ‘hockey stick’ reconstructions is not the blade, but the stick. No one is denying that global temperatures have generally warmed over the last 100 years. Still, in order to discredit the statement that this work indicates that the MWP was warmer than late 20th century temperatures, you would have to assume that those last 15-20 years would reveal a rise of 0.40 degrees C. Since no temperture measurement of any kind shows such a rapid warm-up over the years in question, your protestations lack substance.

theduke

Susann, #388:

Very often, the sense of caution and uncertainty that exists in scientific studies gets lost when the findings are taken into the public domain.

Yes. Exactly. That is why this blog exists and has become so popular. Furthermore, many people here believe that the sense of caution and uncertainty in scientific studies that you speak of was lost before the information was submitted to governments and other related bodies like the IPCC that constitute the so-called “public domain.” And that it was further massaged and compromised once the bureacrats got their mits on it.

Self-serving agendas abound.

Steve McIntyre

If you google Why is the hockey stick important? you can find attempts by both me and McKitrick to answer this.

As an IPCC reviewer for AR4, I said that, if the HS was not relevant to policy issues, then the entire paleoclimate chapter should be deleted from the assessment report. The “consensus” was for its inclusion, so I presume that climate scientists still consider it relevant. Long before the document was drafted, I suggested to an organizer that AR4 include a careful exposition and articulation of the physics underlying the AGW effect and all feedbacks. But they decided not to do that either.

jaye

392…excellent post.

Glen Raphael

#383 >> Why is the MWP important?

It’s more political than scientific. The unfamiliar is scary; the more alarmist politicians can paint recent changes as unfamiliar, the easier it is to scare the public. Some on the AGW alarmist side would like to claim not just that it’s been gradually getting warmer since the little ice age, but that it’s getting dangerously warm or warming at a dangerous rate. A few even postulate an imminent “tipping point” such that once the global temperature gets above X degrees, we’re all doooooomed.

But if we know the temperature has been above X within recent human history, that should remove some of our sense of urgency. Maybe we don’t need to be quite so alarmed. Maybe we have time to wait a few more decades, let the science solidify and just see what happens rather than panicking and shutting down the engines of economic progress.

As an attempt to rewrite history, the hockey stick gives the whole AGW-assessment project a bad name. Skeptics reasonably figure that good data and good science shouldn’t need to be embellished, exaggerated, hidden, or protected from opposing analysis. The fact that the hockey team tries so hard to protect their bad science makes it hard to credit even some of the good science being done in their area. We need more papers that positively add to the good science, rather than only negatively tear down the bad science. With any luck, Craig’s paper will move things in that direction.

In any case, at least it’ll make for more interesting “spaghetti” charts! 🙂

pliny

Re #383
I agree about the irrelevance of the MWP to the validity of AGW. But I disagree about the war analogy. AGW is either happening or it isn’t. If it is, the Earth will go on getting hotter until there is no possible doubt. And if it isn’t, AGW will fade away. It won’t be decided by a battle over the MWP.

Ivan

SOd says

do you think it makes sense that he calls it a “2000 years temperature reconstruction” and leaves out the last 20?

Ask Michaels Mann the same question.

Stan Palmer

re 403

we know that in the past, other factors have lead to significant warming.

Not according to the Mann/IPCC hockey stick

rhodeymark

from what Craig gives us, we don t even know whether there are any proxies past 1980. without checking his literature, we can t be sure if any data point in the 20th century was used at all. using his methodology with the right parameters, you could come to the conclusion that a REAL hockey stick doesn t have a blade either..

Still banging that kettle at 3 am? Craig has stated openly above that the data ended in 1995. The rest is bad faith, and considering that he has welcomed co-authors to refine the work product, you may as well smash that badly skipping record already.

MrPete

Susann:
The MWP question does have scientific significance, from what I’ve seen.
The AGW concept describes, by definition, an unprecedented process.
Suppose that today’s warming is unprecedented. If so, that in itself is not proof of AGW but provides significant room for AGW hypotheses.
However, suppose that today’s warming, whether in value or rate of increase, is nothing special in the last few thousand years. If so, that would falsify most AGW hypotheses.
Thus, the size and “shape” of the MWP is quite significant for the AGW question.
Yes, it appears true that nobody has proven that A has caused today’s GW, nor has anyone analyzed historical climate well enough to prove what drove the MWP and LIA. The fact that we don’t really understand our climate all that well appears not to matter to most real climate scientists, nor to policymakers. Find a big-picture “match” and they’re confident enough to recommend drastic “fixes” with unknown actual impact.
And that’s why Craig’s paper is important.

Sod: there are several problems with the river scenario:
* We must remember that the reservoir was quite low 400 years ago (LIA)
* Major disagreement on where the recent water is coming from (CO2, solar, etc etc)
* Major disagreement on whether we can control the water sources (can we reduce solar? does planting
trees help or hurt? etc)
* Major disagreement about what is the reasonable/helpful range of levels in the reservoir
(was MWP good? LIA bad? Is the status quo really “best”?)
* Many ignore that the reservoir is so large it has its own mechanisms beyond river inlets/outlets (clouds
apparently are a negative feedback, not positive — by analogy, reservoir evaporates, forming clouds that
carry moisture away… I know, it’s a ridiculous comparison ;)… think of the analogy here, not the mechanism)
* Major disagreement over whether the reservoir is about to overflow, or perhaps was more full than today
only 1000 years ago, or perhaps has FAR more capacity than we imagine. (MWP, ancient warming periods.)
If we don’t know where the water is coming from, don’t know whether we can actually control the water level, don’t know if there’s a problem with the reservoir… to me it adds up to a lot of unknowns much in need of careful scientific work, and a lot less hyper-confident declarations that we know exactly what is going on. This really does look like the Y2K fiasco.
I for one much appreciate Craig’s paper as a good attempt to find some data that at least is consistent in measuring something about climate… just what we’re seeing is an interesting question.

Hu McCulloch

Craig —
Your Figure 1 series at http://www.ncasi.org/programs/areas/climate/LoehleE&E2007.csv is greater than 0.3 during 1249-1255, but barely 0.1, if that, in the surrounding years. Did several series kick out in unison during that period?
If so, se’s would presumably show that uncertainty was just temporarily high during that episode, because of the small sample size. As is, your graph seems to say there was a clearcut 7-year heatwave at that time.

Hu McCulloch

Re 322: Great map, Hans! But I only count 15 points. Do some overlay?
Again, I’d like to see these and other supplmentary materials on a webpage of Craig’s creation that would have more permanence than this little discussion. Speaking of which, Craig’s page should also provide a link to this thread (plus any extended at RC etc.)
A graph of each series, with its smoother as construted by Craig, would also be informative.

Hans Erren

re 412:
Thanks, This is just the map from RichardT’s R-script. I haven’t found time to look at the original proxy list. I will make a better looking map when I find the time.

EW

And what about the Brazilian speleothem? It wasn’t suitable?

bender

#416
Loehle will probably respond in a day or so. Meanwhile, it would be useful to make a full list of proxies that were not used but you think should have been used and get him to explain why in one shot. Because the obvious methodological question is “how were the proxies chosen?” “Were they cherry-picked?” He has given his answer, twice now. If you find his answer to be incomplete or imprecise then be patient. Rather than ask him to answer the same question 50 times, give him a chance to prepare a full rebuttal to the whole question, rather than proxy by proxy.

Craig Loehle

Not being an insomniac, I am only now available to respond. Bender #366 thanks. Why did I pick conroy alke and not Hell’s kitchen? I do not remember finding the Hell’s kitchen data. Either I missed it (over a year ago when collecting the data), or it wasn’t posted at that time, or I thought there was something wrong with it. Since it does not ring a bell, I think I missed it.
The proxies are mostly calibrated to linear regressions (except the pollen, borehole, and phenology data). Are they subject to parabolic nonlinearities like we suspect tree rings are? Maybe. What I suggest is that since I had many proxies they would not all have the same flaws. This certainly needs investigation.
The MWP is important because it is implied by the hockey stick that the “normal” pattern of climate is flat, without major ups and downs. Therefore any big jump must be due to GHG.
How did I handle viau data every 100 years? One way is to connect the dots (interpolate) or use step functions. I decided that if he estimated the data every 100 years it is giving too much credit to spread it out over 100 years. I will do some plots later showing things change if you interpolate instead.

Susann

Thanks for the tolerance I’ve experienced here so far. Sorry about my off-topic posts.

The Lohele reconstruction seems to be important because it addresses the question about the value of tree ring proxies in paleoclimate temperature reconstructions. I understand what taking the tree rings out of the data has shown — tree ring data does not show a MWP. The proxy measures used in Loehle do. I would think it preferable to develop a better understanding of tree rings and climate rather than ignoring the tree ring data because that has the appearance of rejecting data that fails to give the results one expects (just as including it gives the results one expects). As well, updating the existing tree ring data would seem important so that the effects of current warming and CO2 levels/ moisture etc. on tree growth are better appreciated (I understand Mr. McIntyre has already done this!)

The other contribution of Loehle seems to be the existene of a MWP, especially one possibly warmer than today. The existence of a MWP suggests that climate variances on a global scale are more common than was once imagined. That still leaves the question of the cause of today’s warming. The fact that warming occurred in the past is a given. The cause of today’s warming is the $64,000 question. The back and forth between camps on the paleoclimate data, the “hockey stick” and MWP seems to be useful to a policy maker only to the extent that it induces a sense of caution about using paleoclimate temperature reconstructions to make policy (and to rouse the rabble.) It doesn’t answer the question of warming today or what, if anything, should be done.

Stan

GHG are not often asserted by climate scientists as the only possible cause. You are painting with too broad of a brush.

bender

A warm MWP implies:
-an elusive “something else” may account for current warming “trend”
-current warming is not “unprecedented” (which proves there is an alarmist camp masquerading as scientists, unwilling to apply proper statistics to time-series inferences)
-the magnitude of the CO2 sensitivity coefficient (which almost everyone at CA agrees is non-zero) may be lower than what the IPCC “consensus” suggests
-the internal variability of terawatt heat engine Earth is not necessarily miniscule compared ot the external forcing, implying that the GCMs tunings may be off somewhat

i.e. Dr Loehle’s paper address the fundamental question of “attribution”.

If the Team truly wanted to address this question, they would re-run Li et al. (2007) without the active ingredients to which they are addicted – which is exactly what Loehle (2007) has tried to do.

bender

cross-post consensus!

Bob Meyer

#425

Stan said:

“GHG are not often asserted by climate scientists as the only possible cause. You are painting with too broad of a brush.”

Yes, but they call the scientists who don’t assert this “denialists” or “shills for the oil industry”.

Larry

Couldn’t anthropogenic GHG magnify a natural forcer-induced change and lead to a crisis?

No. Feedbacks act as a consequence of temperature, and can magnify a forcing, natural or otherwise. Anthropogenic forcing can add to natural forcing. No one has claimed that GHGs change the climate sensitivity. The closest thing to that is Hansen’s conjecture on non-linear feedback, which is pure blue-sky speculation.

jae

433: Judith, I think that is pure snobbery. There’s as much junk in the “mainstream” journals as there is in EE, and this is well-proven by this blog.

rhodeymark

GHG are not often asserted by climate scientists as the only possible cause.

No – but they are happy to ignore that conclusion-without-qualification by their proxy in the media. Or are you unaware of the absolute hysterical tone of the current reports coming out of Valencia? Amazingly enough, GHG reduction is asserted to be the only “remedy”, and time is running out (and they really really mean it this time!).

windansea

Craig, my understanding of the Mannian perspective is that they do not deny the existence of the MWP as an influence on North Atlantic climate (and western Europe, eastern U.S.), but rather deny that there is a global signal from the MWP. Most of your data samples come from the region that is directly influenced by the North Atlantic Ocean, with few points outside this region in the southern hemisphere or in the Pacific region.

1. perhaps you missed the “where’s waldo” threads, most of the GISS warming appears to be regional (NE hemisphere)

2. AGW proponents either minimize temps of MWP or dismiss it as regional, there are proxy studies from all 7 continents listed at CO2 science that show a strong MWP signal equal or greater than the CWP. I don’t know how many of these studies meet Craig’s paramaters.

http://www.co2science.org/scripts/CO2ScienceB2C/data/mwp/mwpp.jsp

http://www.google.com/custom?domains=www.climateaudit.org&q=waldo&sitesearch=www.climateaudit.org&sa=Google+Search&client=pub-3495138952800993&forid=1&channel=7265431255&ie=ISO-8859-1&oe=ISO-8859-1&safe=active&cof=GALT%3A%23008000%3BGL%3A1%3BDIV%3A%23336699%3BVLC%3A663399%3BAH%3Acenter%3BBGC%3AFFFFFF%3BLBGC%3A336699%3BALC%3A0000FF%3BLC%3A0000FF%3BT%3A000000%3BGFNT%3A0000FF%3BGIMP%3A0000FF%3BFORID%3A1&hl=en

2.

S. Hales

Just some general comments: Craig’s data show that global temperature anomalies switch between apx. +/- .6C and his reconstruction anomalies are evenly distributed above and below 0. Mann’s reconstruction is skewed to the negative. I am unsure what the meaning of this is but if climate has regular cycles then a reconstruction should show regular patterns when the time series is sorted from high to low values. If it does not then perhaps the reconstruction is not accurately capturing variability.

Also, 40% of his sites are not influenced by the NAO.

Craig Loehle

While some hockey stick builders don’t deny that the MWP exists (some do deny it) they say it either was a minor blip or strictly regional. Mann argued that peaks in different places don’t line up in time, hence it is not real. this ignores both sample error and dating error, of course.
At this point I don’t care much about reputation vis a vis E&E status. It was important to get it out there. Now I plan to redo it in more detail.
Also, please note the dozens of basic geology papers that drilled a core or something, and just plot the data over time. No hypothesis tests, just a plot. Isn’t that science too? I would have thought so.

richardT

The curse of the < strikes again!
#1) GRIP borehole 18O temperature (Dahl-Jensen et al., 1998);
grip=read.table("http://www.climate2003.com/data/moberg/djgrip.txt", header=T)#BP2000?
#2) Conroy Lake pollen (Gajewski, 1988);
conroy=read.table("ftp://ftp.ncdc.noaa.gov/pub/data/paleo/pollen/recons/liadata.txt", skip=50, nrow=68)[,1:2]#BP1975!
hells=read.table("ftp://ftp.ncdc.noaa.gov/pub/data/paleo/pollen/recons/liadata.txt",skip=225,nrow=75)[,1:2]
clear=read.table("ftp://ftp.ncdc.noaa.gov/pub/data/paleo/pollen/recons/liadata.txt",skip=176,nrow=48)[,1:2]
plot(hells, col=2, type="l", ylim=range(c(hells[,2],conroy[,2], clear[,2])))
lines(conroy, col=3)
lines(clear, col=4)
#3) Chesapeake Bay Mg/Ca (Cronin et al., 2003);
ches1=readLines("ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/cronin2003/cronin2003.txt")[78:527] #year AD
ches=t(sapply(ches1,function(r)as.numeric(strsplit(r,"[[:space:]]+")[[1]][c(3,5)])))
rownames(ches)

Steve McIntyre

Yes. I’ve collated and plotted probable versions of the series. I’ll post them up right away.

I’ve obviously spent a lot of time on individual series in different proxy reconstructions. The first 8 series are all used in Moberg and constitute the majority of his low-frequency series. May I observe the irony of Richard T (whose comments I value) taking an interest in Loehle proxies purporting to show an MWP, while seemingly not taking a similar interest in prior studies of a similar nature.

Bob Meyer

#446

Stan – I was just teasing. Maybe I need to start using smileys.

UC

Hu,

Craig (#73, 178, 214) questions how to do this when his proxies have divergent or even unreported measurement errors themselves. This is a good question, but turns out not to be a big problem. Suppose we have n observations y_i = mu + e_i, but var(e_i) = v_i, which is different for every observation but unobserved. Craig is estimating mu with the simple average y-bar. Assuming independence, var(y-bar) = (1/n^2)Sum(v_i). As long as the v_i’s are not too wildly distributed, Sum(v_i) is consistently estimated by the Sum of Squared Residuals SSR = Sum(y_i – y-bar)^2. Therefore the traditional formula sqrt(SSR/(n*(n-1)) will consistently estimate the standard deviation of y-bar about mu, and will provide
asymptotically N(0,1) t-tests (or “z-tests” if you prefer) for hypotheses.

If I got it correctly, this time we have local temperature recons, and from RC I read that only a small number of stations is needed to calculate global temp anomaly. Let’s say 18, and assume that all those series used in Craig’s paper are exact (perfect local temp). In this case, your method would overestimate uncertainties (negatively correlated errors). Overestimation of uncertainties is unprecedented in climate reconstructions. Spatial sampling comes into play with local reconstructions, teleconnections make things easier.

Now, where did JEG go? Went to check how MBH99 CIs were calculated, and will never come back?

richardT

#447
I am only requesting data that should have been in the supplementary info of the paper. Without seeing the data, their resolution and their last observation, it is not possible to evaluate the reconstruction.

Craig Loehle

Steve M has posted most of the data I will be updating this as quickly as possible.

Susann

Trevor

It seems that you are new here, otherwise you would not make such a statement.

I’ve read here on and off, but I haven’t posted, in part because climate science is new to me (I’m more of an environmental health policy type) and I’ve only just read up on the debates in prep for a course on policy analysis starting after Christmas. (I plan on looking at this issue – wish me luck!) I haven’t read every post, but is there really such consensus here on all the issues that no one would make such a statement except a neube? What happens to them if they do? I thought this place welcomed open-minded discussions.

The issue with the Hockeystick is not that tree ring studies do not show the MWP – quite a few do in fact. The issue is that it appears that the tree series have been ‘cherry-picked’ such that series inconveniently showing a MWP have been excluded.

I didn’t know that — was a reason given why? Or was this an after-the-fact admission? I thought all the tree ring data was used in MBH.

As well, the novel procedures used have been shown to extract a ‘Hockeystick’ shape from pretty much any data series evaluated. Finally, the proponents seem to have knowingly ignored standard statistical tests that show that the Hockeystick produced has little or no skill.

Of that I was aware, which I see as the main contribution of Mr. McIntyre’s work. I have read up on the basic “hockey stick” debate, but I am only now digging into the specifics, like proxy use.

The more I do, the more I feel that paleoclimate global temp. reconstrutions are quite unreliable as policy tools other than to make the point that climate has changed in the past. What proxies should be used? How many are enough to produce reliable results? What statistical methods should be employed to analyse them? And most importantly, what can they say about current warming?

Timo

Craig,

I have been reading your paper during the weekend. Very interesting and convincing, despite some questions being raised in this thread. I am of the opinion that not the number of proxies is relevant, but the reliability of the proxy and methods used. A few realiable data covering the Earth (surface) can tell a clear story.

I am a layman in this field of science, so I will not interfer in the technical discussions. However, I am interested in the background of a particular statement in your paper. You state: “In a few cases, data that were appropriate could not be obtained from authors”.

Can you elaborate why this information was not available and if these data were or could be of importance for your paper?

Thanks in advance.

Judith Curry

A few further comments;

Re E&E. The reputation of a journal rests in the number of citations it receives, the scientific stature of its editorial board, ts reputation for fair and critical reviews of papers, and its selectivity. E&E scores very low on all of these (with possible exception of the review process, i simply don’t know). Further, based upon their public statements, many of the editors of E&E have an overt agenda with respect to energy and environmental politics. Hence choosing to publish a scientific paper in this journal taints the paper and the author as being driven by a political rather than a scientific agenda.

Peer review is no panacea, and no guarantee of anything. Peer review supervised by an editorial board without strong scientific reputations and an overt political agenda (as reflected by public statements by the editors) means little more than that the paper passed some sort of “litmus” test. Peer review supervised by an editorial board with strong scientific reputations and no overt agenda (as reflected by their public statements or lack thereof) guarantees nothing more than that the paper is original, moves the science forward, and is free of obvious methodological errors.

Personally, if i had a paper that I was unable to publish in one of the mainstream journals, I would select a lower impact, online journal that is not associated with a political agenda, something like ArXiv. IMO, it is better to simply publish scientific paper on your own website (or the journal of climateaudit) than E&E. My issue with E&E isn’t snobbery, it is the overt political agenda of much of the editorial board and the journal’s reputation as a refuge for people who can’t get published anywhere esle that results in papers (and authors) being ignored or dismissed by other scientists simply because of the journal in which it was published. IMO, serious scientific papers on the subject of climate change should seek another publication venue besides E&E. I hope that SteveM’s future publications end up somewhere else.

With regards to the MWP. There are two broad issues: a scientific one (regarding how temperature across the globe has changed and to what extend the MWP was a global phenomena) and a political one (associated with debunking the hockey stick owing to its importance in the IPCC reports and its policy implications). The strategies for addressing these two MWP issues is somewhat different. Attempts to falsify the hockey stick by pointing out methodological errors can be useful in two ways: sowing uncertainty about AGW, and in pointing the way to a better understanding of the global temperature record on milennial time scales. At this point, the challenge is to move the science forward. This can only be accomplished by interpreting the complete collection of proxy time series based upon the individual proxy sensitivity to climate variations and the regional coherency of different proxy types. Based upon the extensive data assembly efforts being undertaken by SteveM and others on this site, you have the potential to make an important contribution to moving the science forward (which requires some coherency and understanding of the regional climate variations, no way around it).

Susann

Sincere apologies. I have been lurking here pretty much since the earliest days, and had thought that the issues relating to the Hockeystick had been pretty much settled and indeed, that there was ‘consensus’. My bad.

The issue as I see it is not the hockey stick, which I accept has been settled. The hockey stick graph, from what I can see, has become a lumpy snake, with a MWP and LIA, plus late 20th c. warming.

The issue is the MWP and what it means to the debate about current warming. Loehle does not use tree ring data, and the result is a clear MWP that appears global. The existence of a MWP suggests that such a warming on a global scale has happened at least once since the start of the last interglacial and is not unprecedented. It seems that debate remains on the implications of this for our understanding of current warming.

This is where I am not clear. If we are in a similar period of natural warming, how will the anthropgenic GHG we have produced since the late 19th c. affect the degree of warming we will experience? What is at work in the current warming? How much of it is due to natural forcers (and what are they?) and how much to anthropogenic GHG? How will the doubling of GHG affect any natural forcing?

This is a critical question for policy analysis — we can’t make good policy decisions if we have bad science.

Susann

That was the primary point of the McIntyre-McKintrick critique of the hockey stick.

Sorry to go over already-tilled ground, and I will stop after this, but mea culpa for I understood the hockey stick to be about an improper use of PCA, leading to a hockey stick shape no mater what the data. I also thought it was the use of tree ring data, which does not accurately reflect temperature change — the Bristlecone Pine data. I hadn’t realized that there was cherry picking involved in the data itself. I thought that was an accusation launched back at MM.

shakes head

It’s very complex. Will there be a test? 🙂

Larry

465, someone more up-to-speed on the details is going to have to explain the several issues with the HS, but the “cherry picking”, as I understand it, was automatic, and quite possibly inadvertent. I don’t believe that the issue was ever explicit manual selection. But I’d like someone here with more familiarity to confirm/deny/fill in the holes.

John F. Pittman

Dr. Loehle, I was wondering if the data could be decompressed to annual resolution using appropriate method(s)/assumption(s) (audio and graphics have some interesting routines), and recomputed as an average for the series; and then plot an end of your composite, and “splice” (in a different weight/color) temperature anomalies (deviations) to 2000+? When I look at your data from century to century, I agree with other posters, I am not sure where/how the interpolation and the averaging have taken place. I understand that this would worsen the request of confidence intervals that some have requested. However, I believe these are two separate issues.

I would refine using the ~40% not influenced by NAO to a series, 0% NAO to 100% NAO. First would be the ~40% not influenced NAO, then increase the non-NAO by 1,3,…, all NAO. Compare and determine if there is an appreciable statistical difference in the series.

I just caught up after only 1 1/2 days and it was a bit of reading. However, I was impressed by the number and reasoning of those who wish to understand how the interpolation/averaging/selection of proxies could effect the MWP. I think the decompression and series would help a lot on this.

Finally, thank you for an approach that questions obvious “cherry picking”. One of my degrees is in biology; and I never got educated on “teleconnections”. Several other biologists have complained in other threads of this particular assumption. I found LEG’s comments particularly uninformed/biased.

How is each proxy representative of a particular area ? What is its geographical weight in the final “mean” series ?

I know that this may be an old argument for you and all, but I do not recall a verification/validation that the Mann’s (or any other’s) proxies were representative to a typical area with respect to LEG’s statement about “representative of a particular area” much less “its geographical weight in the final “mean” series”. Perhaps all are atypical, although they are particular; likewise, they may not be geographical. It was shown that one area of the US was a sweet spot for the entire world, according to Mann, is there some intrinsic value/relationship for your paper as well?

Perhaps LEG, rather than you, should validate/verify, AND POST, what this intrinsic value/relationship is? That way the rest of us could examine it as well. I was satisfied with your reasoning, but I am sure any details of quandries you considered while your wrote your paper wrt area vs geographic interpretations would be appreciated.

you are effectively using a statistical model, and are not exempt from the verification exercise

As pointed out above, I find it strange, somewhat, that someone familair with Mann, such as LEG, would be talking of verification with the “censored” file of Mann, part of the historical record. But I wonder, that when you get the time, could you examine this verification/validation question in light of your works. What verification/validation do you think is appropriate and why? I realize that the assumption of no, or little tree-rings tends to preclude this type of question, or should, but still I believe that examining what you consider an appropriate verification/validation sequence would be an enlightening conversation.

Jean S

Judith:

My issue with E&E isn’t snobbery, it is the overt political agenda of much of the editorial board and the journal’s reputation as a refuge for people who can’t get published anywhere esle

Do you have a similar issue with Climatic Change? For an outsider, it seems to be the E&E equivalent in the other end of the spectrum.

Gerald Machnee

RE #463 – **This is where I am not clear. If we are in a similar period of natural warming, how will the anthropgenic GHG we have produced since the late 19th c. affect the degree of warming we will experience? What is at work in the current warming? How much of it is due to natural forcers (and what are they?) and how much to anthropogenic GHG? How will the doubling of GHG affect any natural forcing?**
Those are the questions to ask when someone tells you the “science is in”. Steve M is waiting for an answer to the detailed calculation of climate sensitivity as Larry stated it in #464.
This blog is about verifying or duplicating results. That is difficult when researchers do not archive data.
Similarly there are no direct measurements of the amount or percent of the temperature change caused by GHG or CO2.

Larry

475, it’s broader than science. The only thing that keeps the NYT in business is the brand recognition that gives them a hold on certain people. If it were about merit of content, they’d have gone under years ago.

Craig Loehle

Re Judith Curry and E&E reputation. All that Judith says is correct to a point. How do you tell bias? After an extensive publishing career (go ahead, google me) I can tell. I have submitted mainstream papers that got a few nitpicks or were accepted with no revisions. I have submitted novel papers (but not contrarian) and I got some review comments that didn’t understand it. I have submitted papers where I didn’t explain myself very well and had to work on it more. On the other hand, with contrarian papers I have always gotten nasty reviews, ad hominim attacks, and reviews that in their totality consisted of “this is the worst **** I have ever seen” or “you can’t do this type of analysis” with no explanation or justification, as well as the journal simply refusing to review it. The idea that E&E is biased but Nature and Climatic Change are not is simply not correct. The bias on this issue is right out in the open, and is mostly to totally prevent contrarian work from coming out. Remember the controversy over the Soon and Baliunas work a few years ago and people trying to get “denialists” fired. So, let us just say that this paper opens the door to further work, and speed was important rather than impact factor.

Timo asked why data were not provided when I asked authors for it. Either the authors did not yet have it put together, or didn’t know who I was, or were too busy, or like to keep the data to themselves, or whatever. I don’t know. The importance is that more data would make a better analysis.

John Pittman asked about data decompression to annual—I don’t understand this but John the data will all be available shortly and you can try it. He also asked about representativeness of the data. I have no idea how to assess the extent to which a particular site is representative or a region. The best solution to this question is to increase the sample size and show that no particular samples unduly influence the result.

Craig Loehle

Roman M: Excellent! thank you. Some very clever people here!

steven mosher

re 478. HA. good one romanM.

You know when Hansen correlates one station wth another ( at 1200km) the corelation is .6 in the NH
and .5 in the SH.

Bob Meyer

Re 478

Wow! That is really impressive. Thanks.

steven mosher

RE 483. Ffreddy. Hence my qualifier “most people”. we should not cluuter this thread with policy. MY bad.
unthreaded if you like.

steven mosher

RE 487. very nice nasif. Someday on unthreaded we should pass the time of day and chat.

Joel McDade

Oops, I got it, sorry

Larry

Gunnar, let me completely clear about something. I hold out the possibility that feedback could be high, low, negligible, or even negative. But its simpleminded to just outright declare that it doesn’t exist. Same for the greenhouse effect. The net effect may turn out to be negligible when all is known, but that’s not the same thing as it not existing. Anyone who claims that something that basic is a myth just adds a lot of noise to the discussion. The net effect may turn out to be negligible, but these are not myths.

John F. Pittman

Dr. Loehle, not necessary now. RomanM #478 has done it for me. His approach rids us (or at least my concerns) of the teleconnection issue, as well as provides what I wanted to see. For the series part, his correlation and error estimates provide similar information. The only thing left is to take the plot and splice it on to yours. The colors are nice as well, as long as you change yours to green (lol). I note that his ~1940 is ~0.0 as is yours. Perhaps your graph, the spliced graph, and RomanM’s graph as an insert would be illuminating.

RomanM what a clever way to defuse the teleconnection issue.

Kenneth Fritsch

Re #461

Peer review is no panacea, and no guarantee of anything. Peer review supervised by an editorial board without strong scientific reputations and an overt political agenda (as reflected by public statements by the editors) means little more than that the paper passed some sort of “litmus” test. Peer review supervised by an editorial board with strong scientific reputations and no overt agenda (as reflected by their public statements or lack thereof) guarantees nothing more than that the paper is original, moves the science forward, and is free of obvious methodological errors.

The first sentence in the paragraph above would appear to me to limit the remainder of the observations to minor differences in what may or may not be applied by the less knowledgeable and/or tenaciously analytical reader, as a very much less than full proof guide, based on a trust in the review editors’ definitions of what is original, moves science forward and is free of obvious methodological errors. That trust will evidently depend on how the reader views the motivations of the editors.

My observation would be that peer review is no guarantee of anything — period. It can be applied, with guidelines that are not uniformly or consistently applied, to limit the number of publications to those that that specific scientific community has explicitly or implicitly agreed have value. The quality and utility of peer review, in my view, varies widely from one scientific community to another and makes generalizations about it rather meaningless. Judgments of peer reviewed papers are ultimately in the hands of the interested reader with peer review practically used to avoid having to sort through too much junk to get to the valid papers and the rare gems.

Nasif Nahle

# 488,

Thanks, Steven. It would be great! 🙂 It could be on Friday or Saturday, although I have some other activities non related with biocab.

# 492

Larry,

You’re right; however, we have to admit that the term “greenhouse” is not semantically adecuate to describe scientifically the process given in the atmosphere.

# 479

Craig,

And expect more coincidences with your plots. That’s precisely what the theory of truth on correspondence says, a theory must be consistent with other theories with which such theory is related. It refers to the descriptions of real phenomena. A description of an observed phenomenon should always be coherent with other descriptions of the observed phenomenon.

Seventeen or eighteen studies from about ten proxies reveals that something odd occured with climate through the MP. One study based on one proxy reveals that nothing occurred… mmmh… 😉

jae

One of the problematical primary “outcomes” of the “elite journal” syndrome is that, once an article in one of these journals is published, there is a tendency for the “elite” scientists to reference it without reservation (“after all, it must be accurate, since it was published in this respected publication”). Thus, we have a situation where certain publications (such as Held and Soden for the negative feedback idea), are referenced to almost “prove” some concept, when it may not be true. We tend to build this upside down pyramid, which is as good as the bottom stone. Same with Ahrennius, who was NOT accepted as the Mesiah by his contemporaries.

Nasif Nahle

# 499

Craig Loehle,

I know it’s true. Have you considered the coincidences when collating ISI on your graphs? I didn’t either think on the final coincidence… but they matched effortlessly. 🙂

Kenneth Fritsch

Re: #497

Peer review is only as good as the peers.

That would be the short version of my relatively lengthy post. Maybe I would have said: A particular area of peer review is only as good as that particular community of peers.

bender

Maybe you’ve been getting them mixed up all along? That would explain things. 😉

jae

Brilliant, bender. To use your vernacular, prove it.

steven mosher

RE 499. Dr. Loehle I hope you enjoyed this as much as we have.

Pat Keating

512 Jaye

Amen. However, duplication and prediction checking take longer…….

mccall

re: 144 — I understand your position, but AR4 chose to go the other way. Mann-Jones’03 (and to a lesser extent Jones-Mann’04) dominate the push of AR4 1300 years-BP focus on lengthening the HS handle, instead of explaining the divergence problem.

Generous portions of precooked/canned spaghetti made this easier for AR4 SPM writers: en.wikipedia.org/wiki/Image:2000_Year_Temperature_Comparison.png

Steve McIntyre

There is a serious problem with JEG’s insistence that Loehle’s proxies be weighted by their calibration statistics.

This is no more than Partial Least Squares regression (about which I’ve posted in the past). Mannian inverse regression is equivalent to this underneath the obscure linear algebra as shown in the past. It’s a multivariate method but there are many multivariate methods. This method lends itself to calibration overfitting in high noise situations as I’ve particularly discussed in connection with Ammann.

John A

Jules shows that fence-sitting can be extremely uncomfortable:

I am a regular reader of RealClimate.org which gathers distinguished scientists of the likes of David Archer, Stefan Rahmstorf, Ray Pierrehumbert, Gavin Schmidt and Michael Mann. These people are personally ersponsible for some of the msot exciting work in climatology, their blog has won a few awards and it is increasingly taken as authority by a number of newspapers, including my own “Le Monde” from Frogland. Blog posts on RealClimate are often entertaining, always informative, but it arguably leaves little room for heterodox climate views and it sometimes leaves the strange aftertaste that its authors are defending and selling their own research, while not explicitly acknowledging this partisanship. In particular, Michael Mann is the lead author behind the illustrious Hockey Stick graph, which is by any means a *very* controversial piece of research, and RealClimate.org (RC) has proven somewhat closed to discussion on the topic. Though i have not personally experienced censorship on RC, i have heard many accounts thereof. It is perfectly sensible to block global-warming-denialist trolls from spewing out insults on the forum, but not if they politely ask inconvenient questions.

Obviously, the difference between “global-warming-denialist trolls from spewing out insults” and “global-warming-denialist trolls … politely ask[ing] inconvenient questions” is a difference entirely in the eye of the beholder. Since Mann, Schmidt and Connelley delete the latter while claiming them to be the former, the result is the same, isn’t it Jules?

Craig Loehle

Mosher asked if I’m having fun. yes, very much, althouh not so fun to be insulted and implied that I cheated. I have so many ideas now for a second ms. What I really need is collaborators who can help evaluate the proxies. People with geological training. I can handle the stats and smoothing and writing. Anyone like that please contact me for round 2.
By the way, I have submitted 2 follow-up papers on divergence and temperature trends. I will keep you posted.

Hans Erren

Craig,
Have you prepared a Supplemental Information package?
Raw proxies, processed proxies, method, weighting factors, results
Detailed meta-information about the used proxies (time span, number of entries, location, authors, url)

John A

I think the problem here is intrinsic to the multiproxy paradigm and not simply to Mannian-type statistical manipulations.
If there was such a thing as a global mean temperature, and if each proxy had a linear response to it, then it would be reasonable to be able to construct such a model using linear transformations and Fourier analysis.
But those are big “ifs”. The problem is that there is no such thing as a global mean temperature, and no reason to expect that different proxy sets in different locations with different local microclimates can be meaningfully synthesized without problems with spurious regressions, white noise, microclimatic influences overwhelming the “global signal”.
This is the fundamental problem of climate science and Loehle’s model is no better than the others because the assumptions underlying the multiproxy approach appear to be invalid.
I would also add that Loehle’s data has not been tested for autocorrelation, the assumptions of stationarity, linear response, normality and IID behavior being given as so axiomatic as to be unnecessary even to state them.

steven mosher

RE 520. Dr. L, you kept your cool and gave good answers unlike the laconic one who’s name shall
not be mentioned but with whom Willis is well aquainted.

and now a musical interlude

Carrick

If there was such a thing as a global mean temperature, and if each proxy had a linear response to it, then it would be reasonable to be able to construct such a model using linear transformations and Fourier analysis.

But that’s the real problem. The system isn’t even linear in its time dependence on the forcing terms, and neither are the proxies being used. I would even go as far to say that the nonlinearities associated with the various feedback loops are poorly understood at best, and beyond the scope of what a current global climate model is capable of.

So you can’t even construct a reliable nonlinear model that you could use to perform the inversion from the relevant proxies. GIven the absence of a well understood physical model, I’m frankly concerned that any real policy weight is being given to these temperature reconstructions, especially when the temperature differences in global mean temperature between e.g. the Mann Hockey Stick and Loehle’s reconstruction are so small.

These differences may lie well within the systematic uncertainties associated with the differences between a linear system, and a nonlinear system with positive-feedback forcing terms, like this one.

John A

I think you’ve missed the rest of the URL there, Craig