date: Thu, 3 Feb 2005 11:47:59 -0800 (PST)
from: "David M. Ritson"
subject: Your Science perspective letter.
to: t.osborn@uea.ac.uk, k.briffa@uea.ac.uk
Dear Tim Osborn and Keith Briffa,
I read with interest (and some puzzlement) your perspective on
"The Real Color .." relative to the work of Storch et al.
If one was dealing with the effects of noise on a single proxy record
your `perspective would make perfect sense. However the MBH98 results
derive from an analysis of large families of proxies, as of course you are
well aware. For instance the ITRB North American set of proxy records exceeds a
number N in excess of seventy. Whatever the method of analysis, SVD or
simple averaging, the extracted signal is independent of N whereas
the noise contamination decreases as 1/N**.5. With seventy or so proxies, and I
have veriified this with the ITRB series, low fequency noise distortion
effects become small.
The problems, (if there are problems), appear to lie elsewhere, for instance:
1) MBH normalize, ?calibrate, each proxy record with its detrended variance.
Absent proxy-specific noise (disease, incect infestations etc)
this would be problem free. However in the presence of proxy specific
noise the variance is inflated by the canonical factor
(1+m)**.5 where m is the ratio of proxy noise to the temperature noise,
and hence their temperature scaling factors will be very significantly
modified by the presence of such noies.
MBH do not appear to make any mention of this? It is however calculable
intrinsically from the data. Maybe MBH98, as a last
anlysis step, scaled their results into agreement with the observed
temperature anomaly record? If they failed to do this, analysis of
their Northern tree data, shows that their temperature anomalies would be
underestimated by a factor around 2.5.
2) For w5atever reason the ITRB proxy family shows qualitatively vey large
differences in sensitivities and noise contamination. This of courses
makes weighting and selection of data highly subjective? This, not
short-segment standardization (a red-herring), lies at the heart of the M&M
debate where M&M want to eliminate whole classes of tree-proxies.
3). MBH98 assume that `growth', X_i, is linearly related to real temperature
anomalies for each proxy i via sensitivity factors K_i
X_i=K_i*T + e
where T is the temperature anomaly at time t and e is random noise.
or
X_i=K_i*f(t)+e where T=f(t)
of course if (as seems likely) the sensitivity factor
varies over centuries so that in terms of a slowly varying function F
K_i=K0_i*(1+F(t))
or
X_i=K0_i*(1+F(t)*f(t) +e
This could equally as well be interpreted as
X_i=K0_i*F'(T)+e
where
F'(t)=f(t)*(1+F(t)
Without careful independent considerations, the presence or absence of low
frequency temperature components is degenerate with growth sensitivity and, on
multi-century scales, is indeterminate?
I would certainly appreciate any clarifications you may have for the above.
Frankly I am apalled by an apparent poverty of mechanisms in the climate field
to resolve such problems, or alternatively to classify them as irresolvable?
Sincerely
Dave Ritson
================================================================================
David Ritson, Emeritus Prof of Physics
Physics Dept
Varian Physics Buiding
382 Via Pueblo Mall
Stanford University
Stanford, CA 94305-4060, USA
e-mail: ritson@slac.stanford.edu
Telephone number: 650/723-2685
FAX Number: 650/725/6544
================================================================================