Saturday, April 28, 2012


date: Tue, 13 Apr 1999 15:05:02 -0400 (EDT)

Dear Keith

(Tim, please get this to Keith by FAX or other means, if
he is unlikely to have received this at his own email while traveling).

It's a good piece overall. As you might suspect, I do have several
comments. Ray and Malcolm may send along a few of their own. Malcolm
in particular may want to comment on some of your points regarding
dendroclmiatic series and our ITRDB PC#1 series which figures
so prominently in our millennial reconstruction.

1) page 2, top paragraph:

It's is very misleading to make it sound as if we are strictly
reconstructing northern hemisphere mean temperature, and then
say "4 of the records are actually from the southern hemisphere
locations". This is misleading for a number of reasons. First of
all, if one is going after true northern hemisphere areally-weighted
mean temperature 0-90 degrees (as we are), then the southern hemisphere
tropics are actually more relevant than the high-latitudes of the
Northern Hemisphere. Careful diagnostics of surface temperature
covariances by Alexey Kaplan, Mark Cane and others have shown this
clearly to be true. BUt more than that, we are reconstructing the full
20th century surface temperature domain shown in Figure 1 of our '98
Nature paper. This is a GLOBAL domain, albeit sparse outside the southern
hemisphere tropics/subtropics, particularly the southern oceans,
for obvious regions. THe proxy network roughly overlaps the spatial
domain of surface temperature we are reconstructing (ie, compare
Nature '98 figure 1a and figure 1b). We choose to diagnose from
this spatial domain the northern hemisphere mean only because that
is the hemisphere for which we can meaningfully talk about a
true hemispheric mean. But both the predictor and predictand have
a global distribution. Without going on and on, I think its clear
why your comments here are a bit unfair in how they represent why
we use southern hemisphere data. This is probably the most important
point that needs to be revised here.

2) page 2, 2nd paragraph

A minor point, but an important one: It is incorrect to say the
our uncertainties are based only on "a consideration ...goodnest
of fit...over the calibration period"! This is not correct. A
key point is that the verification period (1854-1901) diagnostics
(though based on a somewhat sparser distribution of gridpoint
data from which NH mean temp can be estimated) give very nearly
identifical diagnostics in terms of unresolved reconstructed
NH mean temp variance. So our uncertainties are based both
on 20th century calibration and independent confirmation from
19th century data. PLEASE MAKE SURE this is clear.

On the bigger point being made here, I agree w/ you in principle,
and this is a point that Phil has raised too: what we *DONT*
take into account (though I challenge anyone to really ever
be able to take this into account!) is the unknown potential
bias due to degradation from diminishing quality of the underlying
proxy data back in time. However, on some of the specific
points in that regard, it is very likely not a significant
concern in our reconstructions. We closely examined the spectra
of the underlying proxy data to insure that those upon which
our reconstruction ultimately relies have the amount of
millennial scale trend/variability that would be expected for
a climatic series for at least the null hypothesis of red noise.
Malcolm independently examined the tree ring chronologies underlying
our ITRDB PC #1 to verify that the standardization was appropriately
conservative for a millennial-scale reconstruction. Furthermore,
Malcolm verified that the ITRDB PC #1 is made up of heavily replicated
chronologies as far back as we use them. So neither of the
points you raise appear to be all that relevant to our reconstruction.

With regard to this point, I have some issues with your Figure
that accompanies the piece. It is quite ironic
given your comments about the potential impacts of
standardization on the long-timescale veriations. For our
millennial reconstruction we have verified as carefully as
has ever been verified, that the millennial scale trend is
likely to be meaningful. I don't think you have done so for
the 2000 year-long trend in the series you show, and if you
have not verified that it is likely to have retained 2000 year
long trends, it is VERY misleading to show this series along
with the others. I don't believe that it is likely to accurately
represent the 2000 year long trend in NOrthern Hemisphere mean
temperature, as you imply by showing it here. I think this series
needs to be removed from the plot. I have a related comment
below (point #5).

3) page 3, 1st paragraph:

Remove "this is a moot point" and replace with more appropriate
language. It is not "a moot point" because the problem you point
out has largely been shown to apply to tree ring density data
(which you have largely been using), and much less so tree
ring width data (which we are using). Furthermore, the comparison
only goes through 1980 at which point there is little evidence
that there is a significant declinde in tree ring width response,
although more evidence that there is already a problem at that
point with density data. Your criticism is not quite fair here,
and the statements should be revised to reflect more accurately
on what we have done.

4) page 3, 2nd paragraph:

When you talk about proxy-based ENSO reconstructions, you
should mention our NINO3 reconstruction! This is complementary
to Stahle's SOI reconstruction in a number of ways. The appropriate
references here are both our Nature '98 papers, and the chapter
in Henry Diaz's latest book (in the press):

Mann, M.E., Bradley, R.S., and Hughes, M.K.,
Long-term variability in the El Nino
Southern Oscillation and associated teleconnections, Diaz, H.F. & Markgraf,V.,
(eds) El Nino and the Southern Oscillation: Multiscale Variability and
its Impacts on
Natural Ecosystems and Society, Cambridge University Press, 321-372, Cambridge,
UK, in press, 1999.

if you care to, you can download the galley version here:

in either pdf format (chapter-diaz.pdf) or postscript (

5) accompanying figure (see also my point #3):

There are problems with the 2000 year series in terms
of your definition of the baseline for comparing with the
other series, and this differs quite a bit from what
we are likely to be showing in IPCC. It appears that both
the density NH reconstruction and your 2000 year long
series fall at least 0.1C below the other series during
the 20th century, and are probably running at least that
much too cold the whole way through.

Also, correct "global temperature and non-temperature proxies"
in your description of our series to "global climate proxies"
which is a more honest way of describing them given our
methodological approach, and make sure it is clear to the
readers which series are extratropical and warm season, and
which are full northern hemisphere/annual mean estimates (ours).
Such discussion will, again, figure prominently in IPCC, and
it would be a shame for Science to be publishing something
that is misleading in that respect. In part, it was this issue
that forced the publication of a followup to Phil's perspective
by me, Ray, Malcolm, and Phil a year ago, and it would be nice
to avoid that scenario this time around...

Thanks for your consideration of the above comments. I believe
your piece will make an excellent "Perspectives" article for
Science, once these comments are appropriately taken into account.
I'll leave it to the Science editor in charge to determine if
that is the case.

best regards,


Michael E. Mann
________Current_____________________________Starting Fall 1999_________
Adjunct Assistant Professor | Assistant Professor
Department of Geosciences | Dept. of Environmental Sciences
Morrill Science Center | Clark Hall
University of Massachusetts | University of Virginia
Amherst, MA 01003 | Charlottesville, VA 22903
e-mail:; (attachments)
Phone: (413) 545-9573 FAX: (413) 545-1200

No comments:

Post a Comment