cc: Tim Osborn <t.osbornatXYZxyz.ac.uk>, firstname.lastname@example.org, email@example.com, cubaschatXYZxyzz.de
date: Fri, 04 Feb 2000 12:42:51 +0100
from: Reiner Schnur <schnuratXYZxyzz.de>
subject: Re: INTEGRATE -- comments on draft of part B.
to: Simon Tett <sfbtettatXYZxyzo.gov.uk>
just a few comments on Simon's comments, sorry that I haven't got to
this earlier ..
> Methodology/work description
> line 1/2 solar output -> total solar irradiance.
> line 3 forced -> force
> line 4 ". Both models" -> which [HadCM3 has 1200 years, what does
> ECHAM4 have??]
ECHAM4/HOPE (the one we're going to use) has a 1000-yr control run
completed. You should use this acronym since ECHAM4 is only the
atmospheric component, and there is also another coupled version,
ECHAM4/OPYC3, which uses a different ocean model.
We should stress that the other project providing the natural run from
gkss is already in progress, as opposed to submitted, so the reviewers
will not think the success of INTEGRATE depends on acceptance of another
> add to end of methods.
> The Natural simulations will be used to estimate the
> contribution of natural forcing to climate variability while the
> All experiments will be used to compare the simulations with proxy
> data in order to evaluate the veracity of the simulations. [Think
> we may need to work a bit harder to justify the Natural only
I see it more the other way round: the natural-only is easily justified
because we want to get better estimates of natural variability (I think
also this was what we started out with) to be used e.g. in detection
studies. And this is according to the introduction (B4), page 4) the
ultimate thrust of the project. The natural+anthro ("All") runs wouldn't
be needed for this. Also, in WP5, if we want to answer the question "How
does external forcing alter the climate and its variability" by
comparing unforced with forced simulation, we would only need the
naturally forced runs because there are already many experiments which
answer the question how anthropogenic forcings alter the climate (this
is not meant to mean that we couldn't use more :-).
The justification for performing the "All" runs since 1750 is harder ..
Of course, as Simon mentions, they should be verified against the proxy
data but this doesn't really give a justification for performing the
runs in the first place. Maybe we have to work a little harder to
describe the actual uses for the "All" simulations.
The proxy data after about 1750 will be contaminated by a possible
anthropogenic signal (although week in the beginning). If we only had
the natural runs then we could only compare the model with proxy data in
the 250 years 1500-1750. So the "all" run is the one corresponding to
the proxy data over the full time period. In comparing all three runs
(control, natural, all) we can better
investigate periods where the forced model runs differ from the unforced
control runs, and why (natural vs. anthropogenic).
There can be periods which could be explained by natural
forcings alone, but there might also be periods which can be best
explained by natural+anthropogenic, e.g. the warming around the 1940's.
The combination of all forcings has been considered e.g. in Tett et al.,
1999 to see if the "all" forcing can explain certain periods better than
anthro alone, but they didn't use a model simulation that used all
forcings at the same time but a linear combinations of the different
signals (you don't have experiments with all forcings together, Simon,
do you?). We could check if the G+s+Sol+Vol experiments is in better
agreement with the obs than any linear combination of the single
> Workpackage 6
> #1 Replace "statistical" with "quantitative".
> #4 Add "from workpackage 5". [Um interesting idea to
> synthesize model and proxy data. Do we think we'd end up with
> NH coverage then using proxy data to tie down some kind of
> model based interpolations]
This was what I wrote some time ago ... One idea I have on this would be
to use Bayesian analysis to merge the proxy and model estimates of
natural variability (there is variant of the classical Bayes analysis
which only deals with priors and posteriors in terms of expectation,
variance and covariance, instead of full probability distributions .
this might be interesting to explore). Another possibility (but
computationally more demanding) would be do some kind of optimal
interpolation between model and proxy data (something done in NWP to
reduce analysis errors by incorporating teh error structure of
observations) .. this would essentially be some kind assimilation of
proxy data into the model data to reduce errors.
> Page 12 -- WP 4. Some comments. Why 2000 years? I would think given the
> nature of the model/proxy comparison we should be talking about 500
> years. [Reiner -- what is your take on this?] -- this would of course
> make life easier in getting the stuff out quickly enough. Why
> additional coverage for the last 1000 -- why not the last 500 ??.
Well, I could imagine that there is some added value in having long
proxy time series e.g. when we compare statistics like return periods
(for the models we have shorter time series, but two 500-year runs). we
have 2x500 years of model estimates of natural variability, but if we
only had 500 years of proxy data since 1500 we'd have only 250 years of
proxy data representative of natural variability (until 1750). If we
have more proxy data back in time we can compare better, I think, how
realistic the estimates since 1500 are or, at least, how they relate to
a longer time period.
We can also use a relationship we find between proxy data and naturally
forced climate models to reconstruct climate variables prior to 1500.
This is usually done by finding this relationship between proxy data and
recently observed instrumental climate variables, but since these
already contain the anthropogenic signal there might be some added value
in using the natural model runs for this (between 1500 and 1750).
Dr. Reiner Schnur <schnuratXYZxyzz.de>, phone: +49 40 41173-379, fax: +49
Max-Planck-Institut fuer Meteorologie, Bundesstr. 55, 20146 Hamburg,
courier service only: DKRZ, 1. Stock, Beim Schlump 58, 20146 Hamburg,