KEY
URLS
Ross
McKitrick
Abstract
for AGU, Oct. 18, 2004
realclimate.organ al.,
Dec.
4, 2004a
Dec
4, 2004b
Dec
4, 2004c
Jan
6, 2005
Richard
Muller, Technology Review, Oct. 15, 2004
M&M,
Revised Nature submission, (Apr. 2004)
Mann
et al.,
Reply to M&M Jan. 2004 (Mar. 2004)
M&M,
Nature submission, Jan. 2004
Comment
by McIntyre and McKitrick on Corrigendum, July 1, 2004
July
1, 2004
Corrigendum by Mann, Bradley and Hughes
Internet
response by MBH (Nov. 2003)
Supplementary
Information to McIntyre and McKitrick (2003)
McIntyre
and McKitrick (2003)
MBH98
|
The
purpose of this webpage is to provide supplementary information and
comments to ongoing research on the use of proxy information to develop
temperature reconstructions. To contact
me:
Email: Stephen McIntyre. |
Jan.
5, 2005 Postscript Graphic
Here is
a postscript version and pdf
version of the graphic
showing simulated PC1s with a hockeystick shape and the MBH98 Northern
Hemisphere reconstruction. Can you find the "real" MBH98
hockeystick?
|
Dec.
22, 2004 AGU Meeting
The
AGU conference is unbelievably big. I'm told there were over 10,000 people
there. The printed program is 512 pages long and for individual papers
only lists authors and titles. I was surprised how little attention was
paid to climate in the last 1000 years. Climate in the past 3 million or
30 million years was a big topic, with many presentations from ODP
drilling. There were several presentations on the climate effect of the
closure of the Panama Seaway (due to plate tectonics).
Impressionistically, the plate tectonics people tend to date this about 4
MM years ago and the ODP climate people about 2.8-3 MM years ago. The
closure ended Pacific waters moving into the North Atlantic and bringing
heat to the Arctic and initiated a long-term cooling through the Pliocene
and Pleistocene. The warming in the Holocene (10000-12000 years),
including modern warming, is still well short of the Pliocene warm period.
The Pliocene is not dinosaur vintage, but is getting to a recognizably
modern configuration. But once the Panama Seaway closed, ocean currents
had to re-arrange themselves and ultimately the modern Conveyor Belt was
established, in which warm waters now come around the Cape of Good Hope.
The re-arrangement is a big job, when you think of it. One of Nof's
students presented a poster arguing that one of the big changes in the
Holocene has been the opening of the Bering Strait, which even though is
very shallow, de-bottlenecks Arctic water.
Bob
Carter showed me some very interesting material from his ODP drilling on
the east New Zealand offshore. This is the main entry point for cold
Antarctic waters into the Pacific - there are surprisingly few entry
points. Bob has found that there are fluctuations at all scales - in fact,
his data would indicate that changes of 0.8 deg C in a century are pretty
much the norm for millions of years and lesser changes (the premise of the
hockey stick) would seem to be the exception.
My
poster session was on the Friday afternoon and I was reasonably busy. I
talked to about 40-50 people. I had visited the poster session for
Nonlinear Methods and quite a few of these mathematically oriented people
reciprocated. They got the point of the principal components argument
almost immediately and you could almost see them laughing at the punchline.
I had printed off the computer program from Mann's FTP site and
highlighted the 4 lines that contain the programming error.
People
who were not mathematically inclined were intrigued by a graphic showing 8
hockeysticks - 7 simulated and 1 MBH (the same sort of graphic as the one
put up here a while ago, but just showing 1 simulation.) Quantity seems to
matter in the demonstration. No one could tell the difference without
being told. I'll insert this graphic here in a day or two.
Again,
I was surprised how little was on recent climate. There was very little
representation from tree ring people. Casper Amman had a presentation
describing his emulation of MBH98; he is planning to web up R code, which
should be interesting. He outlined many issues pertaining to problems in
emulating MBH98 - most of which will be familiar to any followers of our
work, but conspicuously made no reference to our efforts. He said that he
could emulate MBH98 results, but made no reference to principal components
calculations or such esoterica as the MBH98 editing of the Gaspe series.
(I'm still trying to get the new data from Jacoby et al., who say that the
old data is more "temperature sensitive" and should be used in
preference to the new data, which they refuse to archive. LOL)
|
Dec.
10, 2004 www.realclimate.org
Mann
and some of his colleagues have set up a blog at the above address. A
couple of Mann's first postings have been arguments against our papers.
I'll post up a two quick comments below. I'm going to be in San Francisco
at the AFU conference all next week, where I'm presenting a paper [see abstract]
and will post some more when I get back.
|
Dec. 10, 2004 Robustness
If
2 PCs are used in the AD1400 North American network along with
conventional (centered) PC calculations, we argued in our Nature
submissions that MM-type results are obtained. This is now effectively
acknowledged by MBH. To try to salvage MBH98, they now argue that they
should be entitled to increase the number of PCs in the AD1400 North
American network from 2 to 5 and that our not doing so is
"incorrect". They point out that, using centered PC
methods, the PC4 (instead of the PC1) has a hockey stick shape (from the
bristlecone pines) and, as long as they can use the PC4, the PC4 now
drives world climate history. Doesn't this just seem silly? Now
we're not dealing with a "dominant" pattern of world climate,
but a PC4. ROTFLOL.
Secondly,
I defy anyone to show me how the actual retention of PC series in MBH98
can be derived from the Preisendorfer criteria now said to be used in
MBH98 for tree ring networks (although MBH98 itself only talked about
spatial distributions for tree ring PC retention). Below are two plots
made on the same basis as the plot shown at Mann's blog for the AD1400
North American network - only here for the AD1600 Vaganov and AD Stahle/SWM
networks. In the first case, MBH98 retained 2 PCS and in the second case,
MBH98 retained 9 PC series. I do not believe that there is any rational
policy here. I sure can't see how the actual retention can be linked to
Preisendorfer. It would be helpful to see some source code here. Maybe
there's something weird and inconceivable like their centering
method.
Thirdly,
what does this do to their claims of robustness? A robust reconstruction
obviously should not stand or fall on whether 2 or 5 PCs are used in the
AD1400 North American network - but this is exactly what Mann et al. are
saying. Remember all the grandiose claims about MBH98 being robust to the
presence or absence of dendroclimatic indicators altogether (see both
MBH98 and Mann et al.[2000]). Now it seems that MBH98 is not even robust
to the presence or absence of a PC4. Also remember that Mann et al. have
known about the lack of robustness to the bristlecones for a long time -
look at the PC1 in the BACKTO_1400-CENSORED directory. It's almost exactly
the same as ours. Maybe someone can explain to me how you can claim
robustness after doing the CENSORED calculations.
|
In
the figures above, lines are from Preisendorfer-type simulations
using AR1 coefficients. Red is using centered calculations; black
is MBH98 method, showing both the archived value and our
emulation. (I did these calculations a few months ago; I haven't
reconciled why the emulation differs from the archived value in
the Vaganov AD1600 network, but the discrepancy is not large and
is non-existent in the Stahle/SWM network. Again riddle me this:
why does the AD1600 Vaganov network have 2 PCs and the AD1700
Stahle/SWM network have 9 PCs?
|
Dec. 10, 2004 Rutherford et al [Journal of Climate
2004]
The
Dec. 4 post "False Claims" refers to an article by Mann and his
associates [Rutherford et al. 2004], supposedly discrediting our work.
There is nothing in this article about the two main points in our Nature
submissions: 1) the modification of the PC algorithm such that it produces
hockey sticks in the PC1 from red noise; 2) the ad hoc and unique
extrapolation of the Gaspe series. It regurgitates the comments made in
the Nov. 2003 Internet response in which Mann et al. attributed the
difference in PC results to a stepwise procedure used in MBH98, but never
previously reported. However, the calculations in our 2004 submissions
implemented the stepwise procedures and isolated the difference to a
completely different cause: the weird centering method introduced at line
168 of their Fortran program. Rutherford et al [2004] do not discuss this
matter, although Mann et al. are obviously aware of the issue, as it is
referred to in the first paragraph of the web posting.
It
is interesting that this matter is being discussed in a Journal of Climate
article. Jones and Mann [Rev Geophysics 2004] refers to a submission by
Mann et al. to Climatic Change, which was said to have refuted our first
submission. Does it strike anyone as odd that this submission mentioned in
print over 6 months ago has not seen the light of day? Is it
possible that something happened to the article at Climatic
Change?
|
Dec.
4, 2004 Hasn't Mann's data "always" been available?
There's
a difference between the underlying proxy data, source code and supporting
calculations, and the situation is different for each category.
Proxy
data: There is an archive of proxy data located at Mann's FTP site at
the University of Virginia and this is not a current problem area. This
data has not "always" been available. The FTP site was started
on or about July 30, 2002, about 4 years after publication of MBH98, so it
was not available prior to then. The proxy data is presently located at
the url <ftp://holocene.evsc.virginia.edu/pub/MBH98>
and bears a date stamp from 2002. The FTP site has private areas. For
example, the directory presently located at <ftp://holocene.evsc.virginia.edu/pub/MANNETAL98>
was formerly located at <ftp://holocene.evsc.virginia.edu/pub/sdr/temp/nature/MANNETAL98>
and was not indexed. If you knew the exact url, you could retrieve it, but
not otherwise. When the url was changed to its present location, the date
stamp for the directory did not change, so it looks like this directory
has been public since December 2003, but it hasn't been. (BTW my access to
Mann's FTP site from my computer has been blocked, although I can still
get to it from computers at the University of Toronto. This seems a little
petty.) Prior to publication of our first article, there had been no
references to this url, even at Mann's FTP site. I believe that it is
quite possible that the directory was re-located - similar to the
relocation of the MANNETAL98 directory and was only accessible since Nov
2003. Be that as it may, the proxy data is currently available.
Source
code: There is source code at Mann's FTP site for the calculation of
tree ring principal components only. There is no source code for the
calculation of reconstructed principal components, for the calculation of
NH average temperature, for the Preisendorfer-type simulations, etc. etc.
We have requested source code and been refused; we have sought
intervention from Nature and the U.S. National Science Foundation without
success. For the calculations where code is available and a full
reconciliation is possible, there were obviously major discrepancies
between the procedures as described in Nature and the procedures as
actually used. Who would ever have thought that they used an uncentered
algorithm on de-centered data? There were also material discrepancies
between the series listed as used and the series actually used. Who would
have expected this? Mann et al. have provided a vast new Supplementary
Information attached to the July 2004 Corrigendum, which Nature has stated
to us finally provides a complete and accurate description of their
procedures. Since Nature has not obtained the source code and reconciled
it, how can they possibly know that the SI is an accurate description. (By
the way, the Supplementary Information was not edited by Nature and almost
certainly not peer reviewed.)
Supporting
calculations: The supporting calculations that I most want to
see are the calculations for the AD1400 step which is in controversy. The
only information available on this step is an RE statistic (of 0.51).
Nature has refused to provide supporting calculations for the RE
statistics. It is significant that the R2 and other verification
statistics have not been provided. My calculations indicate that the R2
and other verification statistics are embarrassingly low, which is
probably why no one wants to disclose them or to provide the supporting
calculations from which they can be conclusively calculated. The attitude
of Nature is that an interested party can calculate their own values. This
is hardly an adequate response both given the widespread reliance on this
study and the prior track record of inaccurate disclosure of both data and
methods. I'd like to see the exact calculation and the refusals make me
all the more interested.
So
while quite a bit of new information has been provided, there's a lot of
material which has not. It should be easy to simply archive the programs.
You'd think that it would be easier to archive the source code than to
fight about it.
|
Dec.
3, 2004 Other Multiproxy Studies
We
are sometimes asked about other multiproxy studies which are held to
somehow support Mann. A couple of comments. First, if Mann's calculations
are wrong, the fact that other studies get similar results is neither here
nor there. Equally, a critique of MBH98 doesn't refute these other
studies, nor have we claimed this. Second, I'm not convinced that these
studies are anywhere near as mutually supporting as claimed. When I get to
it, I'm going to try to quantify exactly what is supposedly being shown by
the spaghetti diagrams and see if they rise statistically above spaghetti
diagrams from our simulated hockey sticks (see the Oct. 25 comment for an
example). Third, the record for other multiproxy studies is, in all but
one case, worse than MBH98. Here is a brief summary:
Crowley
and Lowery (2000) |
After
nearly a year and over 25 emails, Crowley said in mid-October that
he has misplaced the original data and could only find transformed
and smoothed versions. This makes proper data checking impossible,
but I'm planning to do what I can with what he sent. Do I need to
comment on my attitude to the original data being
"misplaced"? |
Briffa
et al. (2001) |
There
is no listing of sites in the article or SI (despite JGR policies
requiring citations be limited to publicly archived data).
Briffa has refused to respond to any requests for data. None of
these guys have the least interest in some one going through their
data and seem to hoping that the demands wither away. I don't see
how any policy reliance can be made on this paper with no
available data. |
Esper
et al. (2002) |
This
paper is usually thought to show much more variation than the
hockey stick. Esper has listed the sites used, but most of them
are not archived. Esper has not responded to any requests for
data. |
Jones
and Mann (2003); Mann and Jones (2004) |
Phil
Jones sent me data for these studies in July 2004, but did not
have the weights used in the calculations, which Mann had. Jones
thought that the weights did not matter, but I have found
differently. I've tried a few times to get the weights, but so far
have been unsuccessful. My surmise is that the weighting in these
papers is based on correlations to local temperature, as opposed
to MBH98-MBH99 where the weightings are based on correlations to
the temperature PC1 (but this is just speculation right
now.) The papers do not describe the methods in sufficient
detail to permit replication. |
Jacoby
and d'Arrigo (northern treeline) |
I've
got something quite interesting in progress here. If you look at
the original 1989 paper, you will see that Jacoby
"cherry-picked" the 10 "most
temperature-sensitive" sites from 36 studied. I've done
simulations to emulate cherry-picking from persistent red noise
and consistently get hockey stick shaped series, with the Jacoby
northern treeline reconstruction being indistinguishable from
simulated hockey sticks. The other 26 sites have not been
archived. I've written to Climatic Change to get them to intervene
in getting the data. Jacoby has refused to provide the data. He
says that his research is "mission-oriented" and, as an
ex-marine, he is only interested in a "few good" series.
Need I comment?
Jacoby
has also carried out updated studies on the Gaspé series, so
essential to MBH98. I've seen a chronology using the new data,
which looks completely different from the old data (which is a
hockey stick). I've asked for the new data, but Jacoby-d'Arrigo
have refused it saying that the old data is "better" for
showing temperature increases. Need I comment? I've
repeatedly asked for the exact location of the Gaspé site for
nearly 9 months now (I was going to privately fund a re-sampling
program, but Jacoby, Cook and others have refused to disclose the
location.) Need I comment? |
Jones
et al (1998) |
Phil
Jones stands alone among paleoclimate authors, as a diligent
correspondent. I have data and methods from Jones et al 1998. I
have a couple of concerns here, which I'm working on. I remain
concerned about the basis of series selection - there is an
obvious risk of "cherrypicking" data and I'm very
unclear what steps, if any, were taken to avoid this. The results
for the middle ages don't look robust to me. I have particular
concerns with Briffa's Polar Urals series, which takes the 11th
century results down (Briffa arguing that 1032 was the coldest
year of the millennium). It looks to me like the 11th century data
for this series does not meet quality control criteria and Briffa
was over-reaching. Without this series, Jones et al. 1998 is high
in the 11th century. |
These
studies are less "independent" than they appear. Many proxies
recur in nearly all studies (e.g. Tornetrask, Polar Urals, Tasmania). If
you look at all the authors, there is much overlap. Mann is in 4 of the
studies; in addition to Jones et al 1998 and the two articles with Mann,
Jones is a co-author in Briffa et al. 2001 and supplied much of the data
to Crowley and Lowery. Bradley and Jones have been frequent
co-authors.
|
Nov. 8, 2004 MBH Reply to our first Nature submission
online
On
July 16, 2004, Stephen Schneider posted up the MBH reply to our first
Nature submission here: Reply to MM04a, Mar. 2004
which
may interest readers. This was presumably sent to Schneider in support of
a still unpublished submission by MBH to Climatic Change. Some of
the comments in our re-submission
addressed points raised in this reply, especially in connection with
bristlecone pines. M pointed out that 14 sites made a material
contribution to their PC1 - not just Sheep Mountain. When we investigated
these 14 sites, we found that they were all bristlecone and foxtail pine
sites, collected by Donald Graybill and reported in Graybill and Idso
(1993). There is a rich irony in these sites turning up in their new alter
ego as the MBH98 PC1. These sites also enabled an explanation of the
BACKTO_1400-CENSORED subdirectory at Mann's FTP site. There is no
explanation of this directory, but we found that this contained PCs
calculated without 20 (nearly all) bristlecone pine sites, getting a PC1
virtually identical to ours. MBH98 did not report these calculations.
|
Nov. 7, 2004 A Comment on the MBH98 North American PC1
Responding
to a recent challenge, I did the following experiment: I added 0.5 for the
period 1400-1450 to 50 North American tree ring series (normalized in
MBH98 to mean of 1) (the 50 which were in the BACKTO_1400-CENSORED
subdirectory) without adding 0.5 to the 20 (mostly) bristlecone pine
series. The graphic below shows the impact on the PC1 for a centered PC
calculation and for the MBH98 de-centered calculation (all smoothed to 21
years). Above: PC1 using centered calculation: solid - base case; dashed -
with arbitrary addition of 0.5 to 50 non-Graybill sites. This behaves as
expected. Below: PC1 using MBH98 decentered method: solid -as archived;
dashed - with arbitrary addition of 0.5 to 50 non-Graybill sites as above.
This unexpectedly and perversely reduces the PC1 in the 1400-1450 period.
Carried through to the NH temperature index, the NH temperature index is
reduced in the 1400-1450 period by this increase of ring width in 50
series. Pretty strange.
Graphic
updated on Dec. 5, 2004
|
Nov. 4, 2004 SCRIPTS FOR HOCKEYSTICK SIMULATIONS
The
script to do the simulations showing that MBH98 short-segment
standardization consistently yields hockey-sticks from red noise with the
persistence properties of the North American tree ring network is here.
This loads and uses functions where are located here
and a matrix
of North American tree ring site chronologies, which was previously
collated from versions at Dr. Mann's FTP site. The figure below shows the
distribution of the difference between the 503:581 mean and the 1:581 mean
divided by the 1:581 standard deviation using arfima modeling. A 1-sigma
hockey stick is produced over 99% of the time and a 1.5 sigma hockey stick
over 72% of the time using arfima modeling. With AR1 modeling, a 1 sigma
hockey stick is produced 80% of the time, but a 1.5 sigma hockey stick is
almost never produced.
|
Nov.2, 2004 CROWLEY'S DATA
I've
been trying to get the data for Crowley and Lowery (2000) for nearly a
year. The majority of the series are not archived at WDCP, but, even if
they were, one needs to look at the version used in the multi-proxy study
since, as seen with MBH98, the versions can differ and the differences can
be material. I've sent Crowley over 20 emails. Last June, he promised to
archive the data at WDCP by the end of August. This didn't happen. Then he
promised to send me the data by the end of September. This didn't happen.
On Oct. 15, 2004, he finally sent smoothed and scaled versions of his
data. When I asked for the actual data, he said that he "wasn't sure
where they are, or even if [he] had them". I'll post some comments on
data sources for some of the other multi-proxy studies. The time
series from the composite in the email looks like it's different from the
time series which he archived in 2000. It will be interesting to see
why.
|
Oct. 26, 2004 "SPAGHETTI DIAGRAMS"
Maybe
I'll start blogging some odds and ends that I'm working on. I'm
going to post up some more observations on some of the blog
criticisms.
One
of the most common arguments against our criticism of MBH98 is
that it is supported by other multi-proxy studies. This
"support" is usually shown by "spaghetti
diagrams", usually showing the results on an almost
unintelligible scale. Here's an oddity from spaghetti diagrams
from Briffa et al. (2001) and Jones and Mann (2004). In the Briffa
et al. spaghetti diagram, the Crowley-Lowery reconstruction is the
"coldest" in the 17th and 18th centuries and does not
intersect the MBH99 reconstruction. In the Jones and Mann
spaghetti diagram, the Crowley-Lowery reconstruction intersects
the MBH99 reconstruction in the late 17th century.
Figure
1. Spaghetti Diagram, Briffa et al. (2001)). <http://www.ngdc.noaa.gov/paleo/pubs/briffa2001/plate3.gif>.
The Crowley-:Lowery (2000) reconstruction is in a orange-brown and
is the lowest strand in the 17th and 18th centuries, lower
than MBH98/99 (purple) or Briffa et al (2000) (green).
Figure
2. Spaghetti Diagram, Jones and Mann (2004). Here Crowley-Lowery
(2000) is in black In this diagram, it intersects MBH99 in the
late 17th century and is consistently above Briffa et al. (2001).
In
the Jones and Mann spaghetti diagram, the reconstructions were
"scaled by linear regression against the smoothed
instrumental NH series over the common interval 1856-1980, with
the exception of Briffa et al (2001), which has been scaled over
the shorter 1856-1940 interval owing to a late 20th century
decline in temperature response in some of the underlying
data". This supposed "decline in temperature
response" is an interesting story in itself. The Briffa
spaghetti diagram states that all values are re-expressed relative
to a 1961-1990 mean (see y-axis label). I'm not expressing any
views on this right now, merely noting it.
|
Oct.
25. 2004 A COMMENT ON CONNOLLEY
A
sci.environment posting by William Connolley has been circulated
by Tim Lambert,
Brad DeLong and David Appell, as somehow contradicting the point cited by
Muller.
In
our description of the idiosyncratic MBH98 data transformation
prior to principal components calculation (in a submission
constrained by word length), we said that the series:
“were
first scaled to the 1902-1980 mean and standard deviation, then
the PCs were computed using singular value decomposition (SVD) on
the transformed data…”
Connolley's
posting said:
So we
need to look at MBH’s software, not M&M’s description of
it. MBH’s software is
here,
and you can of course read it yourself… Fortran is so easy to
read…What they do is (search down over the reading in data till
you get to 9999 continue):
1. remove the 1902-1980 mean
2. calc the SD over this period
3. divide the whole series by this SD, point by point
At
this point, the new data are in the situation I described above:
datasets that trend upwards at the end have had their variance
reduced not increased. But there is more…
4.
remove the linear trend from the new 1902-1980 series
5. compute the SD again for 1902-1980 of the detrended data
6. divide the whole series by this SD.
This
was exactly what I was expecting to see: remove the linear trend
before computing the SD.
Then the SVD type stuff begins. So… what does that all mean? It
certainly looks a bit odd, because steps 1–3 appear redundant.
The scaling done in 4–6 is all you need. Is the scaling of 1–3
harmful? Not obviously.
Perhaps
someone would care to go through and check this. If I haven’t
made a mistake then I think M&M’s complaints are unjustified
Connolley
makes it seem that we did not cite the location of MBH's PC
software; in fact, he used our citation. He also makes it seem
that we did not describe the method accurately. We were writing
for 800 words and the description was terse, but I fail to see any
inconsistency between our description of the method and
Connolley's description of the method. In fact, our description of
the previously unreported transformation is almost exactly echoed
by Mann et al. in the Corrigendum
SI as follows:
All predictors (proxy and long instrumental and historical/instrumental
records) and predictand (20th century instrumental record) were standardized,
prior to the analysis, through removal of the calibration period (1902-1980)
mean and normalization by the calibration period standard deviation...
The results are not sensitive to this step (Mann et al, in review).
Obviously
we disagree with the last sentence of this paragraph. More on this
later. Nature's policies
say that SI's are peer reviewed. However, I was informed by a
Nature editor that the Corrigendum SI by Mann et al. was not edited
by Nature.
In
Connolley's terms, our claim is that the scaling in step 1
(removal of the 1902-1980 mean) is harmful. Connolley does not
disprove this; he merely says that this is not
"obvious". Given the tentativeness of Connolley's
observations, it is curious to see how rapidly it has circulated
in the blogosphere as supposedly disproving our findings. In fact,
one of the Nature referees - the one who was an expert in
principal components - took a different view than Connolley. In
his first review, he expressed his concerns about
"short segment standardization" (which is what MBH98
does) and in his second review said that he was very
"unimpressed" with Mann et al. "shouting louder and
longer so they must be right".
The
effect may not be "obvious", but Rich Muller understood
it exactly. Just because it is not "obvious" to
Connolley also does not mean that the effect is not real. Below
are two graphs: one is a simulated PC1 using the MBH98 method on a
network of red noise with the persistence properties of the North
American tree ring network (modeled with fractional differencing
not AR1). Can you tell which is the "real" hockey
stick?
In
10,000 simulations, we obtained a PC1 with a 1902-1980 mean over 1
sigma away from the 1400-1980 mean in over 99% of the cases (about
half the time upside-up and half the time upside-down, but this
doesn't matter for regressions). We have described this in detail
in a submission to another journal. Our claims are easily verified
and rather than arguing about it, I will put up scripts (in R) in
a few days, showing the simulations and enabling anyone to test
these results.
Connolley's
argument was picked up and added to by Brad
DeLong as follows:
But Connolley argues--I think correctly--that McKitrick and McIntyre are simply confused: the normalizations diminish the influence of series that show a recent uptrend
I
don't see how Connolley's argument supports DeLong's final clause.
However, be that as it may, the supposed conclusion is incorrect
and it is DeLong who is confused. The short-segment
standardization enhances the influence of series with a strong
20th century trend (it can be either upward or downward, with the
PC weighting changing the sign).
In
fact, in unpublished responses submitted to Nature, Mann et al.
have acknowledged two ways in which short-segment standardization
affected the North American PC1. First, they have acknowledged
that, without the MBH98 data transformation, the controversial
bristlecone pine series (which account for over 99% of the
variance in the MBH98 PC1) are not overweighted in the PC1, but
are overweighted in the PC4 rather
than the PC1. Secondly, they acknowledged that the eigenvalues for
the North American PC calculations are completely different and
submitted a figure to Nature (shown below) showing this. So
the blogger argument that short-segment standardization doesn't
matter to the PC calculations themselves is not one that was
presented by Mann et al. themselves.
Figure 2.
"Comparison of eigenvalue spectrum for the 70 North American
ITRDB data based on MBH98 centering convention (blue circles) and
MM04 centering convention (red crosses). Shown is the null
distribution based on simulations with 70 independent red noise
series of the same length with the same lag-one autocorrelation
structure as the actual ITRDB data using the centering
convention of MBH98 (blue curve) and MM04 (red curve). In the
former case, 2 (or perhaps 3) eigenvalues are distinct from the
noise floor. In the latter case, 5 (or perhaps 6) eigenvalues are
distinct from the noise floor." Figure and caption from Mann
et al. (2004, unpublished)
Mann
et al. themselves made a different argument when they realized the
effect of their short-segment standardization. In effect,
they acknowledged that our calculations were correct using 2 PC
series for the AD1400 network, if the series were standardized
over the 1400-1980 period. So instead, they argued that, if they
increased the number of PC series in the AD1400 network from 2 to
5, they could still salvage MBH98-like results. They justified
this increase by claiming that the number of retained tree ring PC
series was determined through application of Preisendorfer's Rule
N. We did not have an opportunity to reply to this argument and
one of the referees was impressed by this apparent defence and
"wavered" in his support for publication.
However,
there are a couple of replies to this argument. First, MBH98 says
that the number of retained tree ring PCs was determined through
an effort to achieve spatial balance; it does not say that
Preisendorfer's Rule N was used to determine the number of
retained tree ring PCs (although it does say this in connection
with the number of retained temperature PCs.) If you look at the
actual pattern of retained PCs in MBH98, it appears impossible to
me to actually verify the actual use of the above policy: for
example, MBH98 retains 2 PCs for the AD1600 Vaganov network and 9
PCs for the AD1700 Stahle/SWM network. I defy anyone to obtain
these selections using a variation of Preisendorfer's Rule N.
(Actually, I defy anyone to obtain these selections using any
objective method.) Applying the stated MBH98 criterion of spatial
balance, the MBH98 reconstruction is already overloaded with
American tree ring series; adding 3 more PC series from the North
American network (as now proposed as a means of salvaging MBH98)
is hardly justified under spatial balance considerations. Second,
if the bristlecone pines are now relegated to a PC4, it gets
harder to argue that they supply the "dominant
low-frequency" signal. Since Mann et al. made very grandiose
claims about the "robustness" of their results to the
exclusion of all dendroclimatic indicators (especially in Mann et
al. (2000)), the isolation of the presence or absence of the PC4
as a factor dramatically affecting early 15th century values
directly refutes these grandiose "robustness"
claims, regardless of one's position on whether the PC4 (and the
bristlecone pines) should be in or out.
While
much attention is being paid to the hockey stick PC1s, no one
seems to have picked up on the issues involving bristlecone pines.
There are real issues about the validity of these growth rates as
temperature proxies. We also recommend that bloggers pay some
attention to the BACKTO_1400-CENSORED directory at Professor
Mann's FTP site. The PC1 in this directory is virtually identical
to our much-criticized PC1 and NH temperatures calculated from the
PCs in this directory yield results virtually identical to
ours. |
On
Oct. 18, 2004, the AGU accepted McIntyre and McKitrick, Verification of
multi-proxy paleoclimatic studies: a case study, as Paper PP53A-1380
for
presentation at the Fall Meeting in San Francisco Dec. 13-17 (which I will
attend). The Abstract is here. I will
post up some of the computer scripts used in the simulations referred to
here in a few days. |
On
Oct. 15, 2004, Richard Muller of the University of California - Berkeley
published an article on our work in the MIT Technology Review entitled
Global Warming
Bombshell:
A prime piece of evidence linking human activity to climate change turns out to be an artifact of poor
mathematics.
|
BACKGROUND
The
principal focus to date has been on analyzing the reconstruction by Mann
et al. in Mann,
Bradley and Hughes (1998) and 1999,
because this reconstruction was used by the International
Panel on Climate Change, especially in its Third
Assessment Report (2001) to support claims that the 1990s were
the warmest decade and 1998 the warmest year of the millennium, and was
applied in the main promotional graphics used by IPCC, as shown below.
However, we
are in the process of examining the other major multi-proxy studies (Jones
et al. (1998), Crowley and Lowery (2000), Briffa et al. (2001), Esper et
al. (2002), Mann and Jones (2003)).
FIGURE
1. NH Temperature Reconstruction - IPCC Version of MBH98, MBH99
Despite
the extensive policy reliance on this presentation, IPCC did not itself
audit or verify the data and methods of MBH98 and MBH99. To our knowledge,
other than our efforts, no other person or institution has attempted any
replication or verification. (See here
for some comments on replication standards.)
When
we attempted to replicate the results of MBH98, we encountered many data
errors and defects in the underlying data set to which we had been
directed, which we reported on in McIntyre and McKitrick (October 2003).
These errors and defects appeared to include collation errors, use of
obsolete data, incorrect principal components calculations for tree ring
networks, unjustified truncation or data, geographical mislocations etc.
We re-collated tree ring data from original sources and obtained updated
versions wherever possible. (See SI.)
When we carried out our own re-construction, we obtained quite different
results - with high early 15th century values, which prevented the MBH98
claim of 20th century uniqueness, as shown in the diagram below:
FIGURE 2. NH Temperature Reconstructions
from MM03.
MM03
occasioned considerable controversy. First, Mann et al. argued that they
had sent us the wrong data and this compromised the MM03 calculations.
Mann then deleted the "wrong" data from his FTP site. Since we
re-collated the compromised data, this criticism had no bearing on our own
calculations, although it did have a bearing on a couple of our criticisms
of MBH98. The matter is discussed here,
but has little bearing on the scientific outcome of the dispute. Professor
Mann did however provide a different FTP
location said to be the location of MBH98 proxy data. Since
November 2003, we have examined this FTP location and found many
additional data errors and defects, additional to those described in MM03.
Some of these issues have been addressed in the recent Nature Corrigendum;
there are many data errors in MBH98 which remain unacknowledged.
Second,
Mann et al. argued
that the differing results could be attributed to different handling
of tree ring principal component calculations - in particular, that MM03
had not carried out a "stepwise" calculation - and to our use of
a different version of the TTHH tree ring series. They produced the
following emulation of our results by changing 3 indicators. They asserted
that the use of a "stepwise" procedure for tree ring principal
components resulted in the use of 159 series, rather than the 112 series
referred to in MBH98 and listed in pcproxy.txt.
FIGURE 3. NH Reconstructions from Mann et
al. (2003)
Since
the principal components calculations in MBH98 were described as
"conventional" and there was no mention of 159 series, we
carried out a conventional principal components calculation using the
maximum period in which all sites were available. As it turned out, this
procedure resulted in the NOAMER PC1 and STAHLE/SWM PC1 being unavailable
in the AD1400 step. We used a version of the TTHH series which began later
(and ended later), because this was the later version archived by Jacoby
(which was calculated well before the publication of MBH98). We make no
apology for not using obsolete data; however, since the TTHH series
doesn't start until 1459 in the MBH98 version, it is irrelevant
anyway. As it turns out, the presence or absence of the Stahle/SWM
PC1 has virtually no effect on early 15th century values.
The
differences between results depend entirely on the calculation of the
NOAMER PC1. We have closely examined this matter and have identified an
error in the MBH98 procedure for calculating principal components, which
we have submitted to a journal. This one error accounts for most of the
difference between our results. We have also been able to identify one
other data misrepresentation, which accounts for the balance of the
difference between our results. We anticipate that these results will soon
become public.
We
have also endeavoured to ensure a complete public record with respect both
to MBH98 and other multi-proxy studies. It seems scandalous to us that
there should be anything other than copious public disclosure of all data,
methods and source code in papers relied upon for major public policy. We
have had some partial success in this endeavour with the new SI for MBH98,
which provides considerable new information on this study. Nonetheless,
the public record remains incomplete and multi-proxy authors have refused
to provide key information. The public disclosure record for the studies
which supposedly confirm MBH98 (e.g. Crowley and Lowery (2000), Briffa et
al (2001) etc.) is actually worse than MBH98. We have endeavoured here to
provide a complete record of all data and methods used in MM03, including
all scripts. (See SI to MM03)
As
noted above, our studies have progressed significantly since MM03 and we
hope to be able to comment more fully our submission in review in the near
future.
|
|
|