Climategate 2.0: Hughes — hokey stick should be treated with ‘considerable caution’

Hokey stick co-author Malcolm Hughes could have been a skeptic.

From the Climategate 2.0 collection, Hughes writes to hokey stick co-author Michael Mann and Ed Cook:

All of our attempts, so far, to estimate hemisphere-scale
temperatures for the period around 1000 years ago are
based on far fewer data than any of us would like.
None
of the datasets used so far has anything like the
geographical distribution that experience with recent
centuries indicates we need, and no-one has yet found a
convincing way of validating the lower-frequency
components of them against independent data. As Ed
wrote, in the tree-ring records that form the backbone of
most of the published estimates, the problem of poor
replication near the beginnings of records is particularly
acute, and ubiquitous. I would suggest that this problem
probably cuts in closer to 1600 than 1400 in the several
published series. Therefore, I accept that everything we
are doing is preliminary, and should be treated with
considerable caution.
[Emphasis added]

Read the e-mail below.
cc: “Michael E. Mann” , Malcolm Hughes
, esper@wsl.ch, k.briffa@uea.ac.uk, t.osborn@uea.ac.uk,
p.jones@uea.ac.uk, tcrowley@duke.edu, rbradley@geo.umass.edu, jto@u.arizona.edu,
srutherford@virginia.edu
date: Thu, 11 Apr 2002 19:57:12 -0700
from: “Malcolm Hughes”
subject: Re: Your letter to Science
to: Edward Cook , “Michael E. Mann”

Dear Ed and Mike and others,
All of our attempts, so far, to estimate hemisphere-scale
temperatures for the period around 1000 years ago are
based on far fewer data than any of us would like. None
of the datasets used so far has anything like the
geographical distribution that experience with recent
centuries indicates we need, and no-one has yet found a
convincing way of validating the lower-frequency
components of them against independent data. As Ed
wrote, in the tree-ring records that form the backbone of
most of the published estimates, the problem of poor
replication near the beginnings of records is particularly
acute, and ubiquitous. I would suggest that this problem
probably cuts in closer to 1600 than 1400 in the several
published series. Therefore, I accept that everything we
are doing is preliminary, and should be treated with
considerable caution. I differ from Ed, and his co-authors,
in believing that these problems have a special
significance for the particular implementation of RCS
they used, in the light of one of their conclusions that
depends heavily on that implementation.
As I understand what Ed, Keith and Hal Fritts have
written at various times about RCS, and from my own
limited experience with the method, it is extremely
important to have strong replication, and I don’t see 50-70
samples probably from 25-35 trees as a big sample. For
reference, most chronologies used in dendroclimatology
are based on 10-40 trees, that is 20-80 samples at 2 cores
per tree for a single “site”, usually a few hectares.
Here are two passages from Briffa et al., 1992:
page 114, column 1, last paragraph, “For a chronology
composed of the same number of samples, one would
therefore expect a larger statistical uncertainty using this
approach than in a chronology produced using
standardization curves fitted to the data from individual
trees……………The RCS method therefore requires greater
chronology depth (i.e. greater sample replication) to
provide the same level of confidence in its representation
of the hypothetical “true” chronology.” ECS mention this
issue.
page 114, column 1, third paragraph, there is a discussion
of the problems arising from applying RCS when pith age
is not known, “In the ring-width data, the final
standardization curve probably slightly underestimates
the width of young trees and could therefore impart a
small positive bias to the standardized ring-width indices
for young rings in a number of series. However, this
effect will be insignificant when the biased indices are
realigned according to calendar growth years and
averaged with many other series.” The problem here is
that this latter condition is not met (in my view), and the
“small positive bias” that may be retained could turn out
to be important to the most controversial conclusion of
ECS (the Medieval question). I also suspect that Keith
and colleagues underestimated both the size and
variability of the loss of years at the beginning of records,
but the point stands even if this is not so. So far as I can
see, ECS do not mention this issue, at least in the context
of a possible positive bias. The discussion of RCS in the
supplementary materials seems to assume good
replication.
ECS, as Ed rightly points out, clearly indicate, in both
words and diagrams at several points in their paper and in
the supplementary materials, that the number of sites and
number of samples they used decreases sharply before
1200. Even so, ECS gives prominence (second sentence
of the abstract, for example) to the reconstruction in that
very period, and makes a comparison with the magnitude
of 20th-century warming. All the methods, and their
realizations so far, have significant problems. In our letter
(Mike and I) we draw attention to a specific problem with
this implementation of RCS that has a special bearing on
the reconstruction of a period to which ECS have drawn
attention. Hence the strong note of caution about the ECS
conclusion on the comparison between the 10th/11th and
late 20th centuries.
I hope it’s clear from this that I don’t disagree with the
general proposition that all existing reconstructions of
hemipsphere-scale temperatures 1000 years ago (or even
for all the first half of the second millennium AD) should
be viewed as very preliminary. If anyone is interested I
attach a short note on the replication in the year AD 1000
of records used in MBH99 to give an idea of what we are
up against.
We all have a lot to do. I see four important tasks – 1)
more investigation of the strengths and limitations of
methods like RCS and age-banding – for example, how
many samples would have been enough in this case, does
the RC change through time? and so on; 2) use of treering
records where the loss of low-frequency information
is least – those with long segments from open stands; 3)
the search for tree-ring parameters without age/size
related trend; 4) the development of completely
independent proxies with intrinsically better lowfrequency
fidelity.
Cheers, Malcolm
The Briffa et al reference is to the 1992 paper, Climate
Dynamics, 7:111-119
> Hi Ed,
>
> OK–thanks for your response. I’ll let Malcolm respond to the
> technical issues regarding RC. I’m not really qualified to do so
> myself anyway. Your other points are well taken…
>
> Cheers,
>
> Mike
>
> At 12:09 PM 4/11/02 -0400, Edward Cook wrote:
> Hi Mike,
>
> Thanks for the reply. I too do not want to see anything
> personal in our disagreements. It would be a shame if it got to
> that and it shouldn’t. I don’t think that the science we are
> talking about is sufficiently known yet to claim the “truth”,
> which is why we are having some of our disagreements. I mainly
> wanted to clarify some issues relating to some criticisms of the
> ECS results that I thought were not totally fair. My biggest
> complaint is with Malcolm’s contribution to your letter because it
> really isn’t fair to use such words as “perilous”. ECS did not
> hide anything and the uncertainties are clearly indicated in EGS
> Figs. 2 and 3. So, you can make your own judgement. However,
> Malcolm’s opinion does not invalidate the ECS record. If Malcolm’s
> statement is correct, than ALL previous estimates of NH
> temperature over the past 1000 years are “perilous”, especially
> before AD 1400 when the number of series available declines
> significantly in most records.
Ed
>
> Ed,
>
> It will take some time to digest these comments, but my
> initial response is one of some disappointment. I will
> resist the temptation to make the letter to Science
> available to the others on this list, because of my fears of
> violating the embargo policy (I know examples of where doing so
> has led to Science retracting a piece form publication). So thanks
> for also resisting the temptation to do so…
>
> But I must point out that the piece by Malcolm and me
> is very similar in its content to the letter of clarification that
> you and I originally crafted to send to Science some weeks ago,
> before your co-author objected to your involvement! If there is no
> objection on your part, I’d be happy to send that to everyone,
> because it is not under consideration in Science (a quite
> unfortunate development, as far as I’m concerned). The only real
> change from that version is the discussion of the use of RCS. That
> is in large part Malcolm’s contribution, but I stand behind what
> Malcolm says. I think there are some real sins of omission with
> regard to the use of RCS too, and it would be an oversight on our
> part now to comment on these.
>
> Finally, with regard to the scaling issues, let me simply
> attach a plot which speaks more loudly than several
> pages possibly could The plot takes Epser et al (not
> smoothed, but the annual values) and scales it against the
> full Northern Hemisphere instrumental record 1856-1990
> annual mean record, and compares against the entire 20th
> century instrumental record (1856-1999), as well as with
> MBH99 and its uncertainties.
>
> Suppose that Esper et al is indeed representative of the
> fullNorthern Hemisphere annual mean, as MBH99
> purports to be. To the extent that differences emerge
> between the two in assuming such a scaling, I interpret
> them as differences which exist due to the fact that the
> extratropical Northern Hemisphere series and full
> Northern Hemisphere series likely did not co-vary in the
> past the same way they co-vary in the 20th century (when
> both are driven predominantly, in a relative sense, by
> anthropogenic forcing, rather than natural forcing and
> internal variability). What the plot shows is quite
> remarkable. Scaled in this way, there is remarkably little
> difference between Esper et al and MBH99 in the first
> place (the two reconstructions are largely within the error
> estimates of MBH99!)!, but moreover, where they do
> differ, this could be explainable in terms of patterns of
> enhanced mid-latitude continental response that were
> discussed, for example, in Shindell et al (2001) in
> Science last December. So I think this plot says a lot. Its
> say that there are some statistically significant
> differences, but certainly no grounds to use Esper et al to
> contradict MBH99 or IPCC ‘2001 as, sadly, I believe at
> least one of the published pieces tacitly appears to want
> to do.
>
> It is shame that such a plot, which I think is a far more
> meaningful comparison of the two records, was not
> shown in either Esper et al or the Briffa & Osborn
> commentary. I’ve always given the group of you adequate
> opportunity for commentary on anything we’re about to
> publish in Nature or Science. I am saddened that many of
> my colleagues (and, I have always liked to think friends)
> didn’t affort me the same opportunity before this all
> erupted in our face. It could have been easily avoided.
> But that’s water under the bridge.
>
> Finally, before any more back-and-forths on this, I want
> to make sure that everyone involved understands that
> none of this was in any way ever meant to be personal, at
> least not on my part (and if it ever has, at least on my
> part, seemed that way, than I offer my apologies–it was
> never intended that way). This is completely about the
> “science”. To the extent that I (and/or others) feel that the
> science has been mis-represented in places, however, I personally
> will work very hard to make sure that a more balanced view is
> available to the community. Especially because the implications
> are so great in this case. This is what I sought to do w/ the NYT
> piece and my NPR interview, and that is what I’ve sought to do
> (and Malcolm to, as far as I’m concerned) with the letter to
> Science. Being a bit sloppy w/ wording, and omission, etc. is
> something we’re all guilty of at times. But I do consider it
> somewhat unforgivable when it is obvious how that sloppiness can
> be exploited. And you all know exactly what I’m talking about!
>
> So, in short, I think are some fundamental issues over
> which we’re in disagreement, and where those exist, I will
> not shy away from pointing them out. But I hope that is
> not mis-interpreted as in any way personal.
>
> I hope that suffices,
>
> Mike
>
> p.s. It seemed like an omission to not cc in Peck and
> Scott Rutherford on this exchange, so I’ve done that. I
> hope nobody minds this addition…
> At 10:57 AM 4/11/02 -0400, Edward Cook wrote:
> Hi Mike and Malcolm,
>
> I have received the letter that you sent to Science
> and will respond to it here first in some detail and
> later in edited and condensed form in Science.
> Since much of what you comment and criticize on
> has been disseminated to a number of people in
> your (Mike’s) somewhat inflammatory earlier
> emails, I am also sending this lengthy reply out to
> everyone on that same email list, save those at
> Science. I hadn’t responded in detail before, but
> do so now because your criticisms will soon be in
> the public domain. However, I am not attaching
> your letter to Science to this email since that is
> not yet in the public domain. It is up to you to
> send out your submitted letter to everyone if you
> wish.
>
> I must say at the beginning that some parts of
> your letter to Science are as “flawed” as your
> claims about Esper et al. (hereafter ECS). The
> Briffa/Osborn perspectives piece points out an
> important scaling issue that indeed needs further
> examination. However, to claim as you do that
> they show that the ECS 40-year low-pass
> temperature reconstruction is “flawed” begs the
> question: “flawed” by how much? It is not at all
> clear that scaling the annually resolved RCS
> chronology to annually resolved instrumental
> temperatures first before smoothing is the correct
> way to do it. The ECS series was never created to
> examine annual, or even decadal, time-scale
> temperature variability. Rather, as was clearly
> indicated in the paper, it was created to show how
> one can preserve multi-centennial climate
> variability in certain long tree-ring records, as a
> refutation of Broecker’s truly “flawed” essay. As
> ECS showed in their paper (Table 1), the high-
> frequency correlations with NH mean annual
> temperatures after 20-year high-pass filtering is
> only 0.15. That result was expected and it makes
> no meaningful difference if one uses only extra-
> tropical NH temperature data. So, while the
> amplitude of the temperature-scaled 40-year low-
> pass ECS series might be on the high end (but
> still plausible given the gridded borehole
> temperature record shown in Briffa/Osborn),
> scaling on the annually resolved data first would
> probably have the opposite effect of excessively
> reducing the amplitude. I am willing to accept an
> intermediate value, but probably not low enough
> to satisfy you. Really, the more important result
> from ECS is the enhanced pattern of multi-
> centennial variability in the NH extra-tropics over
> the past 1100 years. We can argue about the
> amplitude later, but the enhanced multi-centennial
> variability can not be easily dismissed. I should
> also point out, again, that you saw Fig. 3 in ECS
> BEFORE it was even submitted to Science and
> never pointed out the putative scaling “flaw” to
> me at that time.
>
> With regards to the issue of the late 20th century
> warming, the fact that I did not include some
> reference to or plot of the up-to-date instrumental
> temperature data (cf. Briffa/Osborn) is what I
> regard as a “sin of omission”. What I said was
> that the estimated temperatures during the MWP
> in ECS “approached” those in the 20th century
> portion of that record up to 1990. I don’t consider
> the use of “approached” as an egregious
> overstatement. But I do agree with you that I
> should have been a bit more careful in my
> wording there. As you know, I have publicly
> stated that I never intended to imply that the
> MWP was as warm as the late 20th century (e.g.,
> my New York Times interview). However, it is a
> bit of overkill to state twice in the closing
> sentences of the first two paragraphs of your
> letter that the ECS results do not refute the
> unprecedented late 20th century warming. I
> would suggest that once is enough.
>
> ECS were also very clear about the extra-tropical
> nature of their data. So, what you say in your
> letter about the reduced amplitude in your series
> coming from the tropics, while perhaps worth
> pointing out again, is beating a dead horse.
> However, I must say that the “sin of omission” in
> the Briffa/Osborn piece concerning the series
> shown in their plot is a bit worrying. As they say
> in the data file of series used in their plot (and in
> Keith’s April 5 email response to you),
> Briffa/Osborn only used your land temperature
> estimates north of 20 degrees and recalibrated the
> mean of those estimates to the same domain of
> land-only instrumental temperatures using the
> same calibration period for all of the other non-
> borehole series in the same way. I would have
> preferred it if they had used your data north of
> 30N to make the comparisons a bit more one-to-
> one. However, I still think that their results are
> interesting. In particular, they reproduce much of
> the reduced multi-centennial temperature
> variability seen in your complete NH
> reconstruction. So, if the amplitude of scaled
> ECS multi-centennial variability is far too high
> (as you would apparently suggest), it appears that
> it is also too low in your estimates for the NH
> extra-tropics north of 20N. I think that we have
> to stop being so aggressive in defending our
> series and try to understand the strengths and
> weaknesses of each in order to improve them.
> That is the way that science is supposed to work.
> I must admit to being really irritated over the
> criticism of the ECS tree-ring data standardized
> using the RCS method. First of all, ECS
> acknowledged up front the declining available
> data prior to 1200 and its possible effect on
> interpreting an MWP in the mean record. ECS
> also showed bootstrap confidence intervals for
> the mean of the RCS chronologies and showed
> where the chronologies drop out. Even allowing
> for the reduction in the number of represented
> sites before 1400 (ECS Fig. 2d), and the
> reduction in overall sample size (ECS Fig. 2b),
> there is still some evidence for significantly
> above average growth during two intervals that
> can be plausibly assigned to the MWP. Of course
> we would like to have had all 14 series cover the
> past 1000-1200 years. This doesn’t mean that we
> can’t usefully examine the data in the more
> weakly replicated intervals. In any case, the
> replication in the MWP of the ECS chronology is
> at least as good as in other published tree-ring
> estimates of large-scale temperatures (e.g., NH
> extra-tropical) covering the past 1000+ years. It
> also includes more long tree-ring records from the
> NH temperate latitudes than ever before. So to
> state that “this is a perilous basis for an estimate
> of temperature on such a large geographic scale”
> is disingenuous, especially when it is unclear how
> many millennia-long series are contributing the
> majority of the temperature information in the
> Mann/Bradley/Hughes (MBH) reconstruction
> prior to AD 1400. Let’s be balanced here.
>
> I basically agree with the closing paragraph of
> your letter. The ECS record was NEVER
> intended to refute MBH. It was intended, first
> and foremost, to refute Broecker’s essay in
> Science that unfairly attacked tree rings. To this
> extent, ECS succeeded very well. The
> comparison of ECS with MBH was a logical
> thing to do given that it has been accepted by the
> IPCC as the benchmark reconstruction of NH
> annual temperature variability and change over
> the past millennium. Several other papers have
> made similar comparisons between MBH and
> other even more geographically restricted
> estimates of past temperature. So, I don’t
> apologize in the slightest for doing so in ECS.
> The correlations in Table 2 between ECS and
> MBH were primarily intended to demonstrate the
> probable large-scale, low-frequency temperature
> signal in ECS independent of explicitly
> calibrating the individual RCS chronologies
> before aggregating them. The results should
> actually have pleased you because, for the 20-200
> year band, ECS and MBH have correlations of
> 0.60 to 0.68, depending on the period used.
> Given that ECS is based on a great deal of new
> data not used in MBH, this result validates to a
> reasonable degree the temperature signal in MBH
> in the 20-200 year band over the past 1000 years.
>
> Given the incendiary and sometimes quite rude
> emails that came out at the time when ECS and
> Briffa/Osborn were published, I could also go
> into the whole complaint about how the review
> process at Science was “flawed”. I will only say
> that this is a very dangerous game to get into and
> complaints of this kind can easily cut both ways.
> I will submit an appropriately edited and
> condensed version of this reply to Science.
>
> Regards,
>
> Ed
> —
> =================================
> Dr. Edward R. Cook
> Doherty Senior Scholar
> Tree-Ring Laboratory
> Lamont-Doherty Earth Observatory
> Palisades, New York 10964 USA
> Phone: 1-845-365-8618
> Fax: 1-845-365-8152
> Email: drdendro@ldeo.columbia.edu
> =================================
> _____________________________________________
> __________________________
> Professor Michael E. Mann
> Department of Environmental Sciences, Clark Hall
> University of Virginia
> Charlottesville, VA 22903
> _____________________________________________
> __________________________
> e-mail: mann@virginia.edu Phone: (434) 924-
> 7770FAX: (434) 982-2137
> http://www.evsc.virginia.edu/faculty/people/mann.sht