date: Sat, 11 Sep 2004 17:27:23 -0700
from: Milind Kandlikar <mkandlikaratXYZxyzi.ubc.ca>
subject: Re: Brief communications arising from Murphy et al. Nature paper
to: James Risbey <james.risbeyatXYZxyz.monash.edu.au>
Dear Suraje et al.,
The modified draft with Roger's and Jeroen's comments looks tight.
There is a broader confusion in my mind on the issue on the role and
utility of "tuning" in GCMs.
In engineering it is routine to fit your models to observations - this
assumes that the model has the right physics, but model parameters need
to be determined using some form of parameter estimation. Once
estimated models can be used a prediction devices - in fact much of
online process control uses mathematical representations of the actual
system being controlled. Online process control is now fairly routine.
With GCMs the issue is different. Tuning may be a way to fudge the
physics. For example, understanding of clouds or aerosols is far from
complete - so (ideally) researchers build the "best" model they can
within the constraints of physical understanding and computational
capacity. Then they tweak parameters to provide a good approximation to
observations. It is this context that all the talk about "detuning" is
confusing. How does one speak of "detuning" using the same physical
models as before? A "detuned" model merely uses a different set of
parameters that match observations - it not hard to find multiple
combinations of parameters that give the similar model outputs (in
complex models with many parameters/degrees of freedom) So how useful
is a detuned model that uses old physics? Why is this being seen as
some sort of a breakthrough?
On 10-Sep-04, at 11:27 PM, James Risbey wrote:
> Hi Suraje et al.,
> Suraje's draft and the edits by Jeroen and Roger look great. I think
> that covers the main points pretty well. I like Jeroen's point that
> in the commentaries the limitations of the work have disappeared.
> There is some confusion in Murphy et al. over whether the CPI weights
> according to fit with observations or according to fit with other
> ensemble members. The text on their page 770 first suggests the
> former, and then the latter. We more or less imply the former in our
> draft(s). I don't think it is too critical as the point stands either
> way. Whether they weight to observations fit or to something like an
> ensemble mean, they are tending to bias the sensitivity estimates back
> toward the value obtained with the tuned model. This just fyi, as I
> don't think we need to amend the text.
> Thanks Suraje, and looking forward to the integrated draft.
> Suraje Dessai <s.dessaiatXYZxyz.ac.uk> writes:
>> Dear all,
>> Please find attached our latest draft for the Brief communications
>> arising from the Murphy et al. Nature paper.
>> Let me know what you think and any changes you'd like to make. We're
>> 50 words over the top so any suggestions for deletion would be most
>> welcome. Please get back to me soon as I'd like to have this completed
>> by next week.
Dr. Milind Kandlikar
Institute for Resources, Environment and Sustainability &
The Liu institute for Global Issues
213, 1924 West Mall
The University of British Columbia
Vancouver BC V6T 1Z2
Phone: 604 822 6722
Fax: 604 822 9191