cc: peter.stott@metoffice.gov.uk, myles <m.allen1@physics.ox.ac.uk>, Tim Barnett <tbarnett-ul@ucsd.edu>, Nathan Gillett <n.gillett@uea.ac.uk>, Phil Jones <p.jones@uea.ac.uk>, David Karoly <dkaroly@rossby.metr.ou.edu>, Jesse Kenyon <kenyon@duke.edu>, Reto Knutti <knutti@climate.unibe.ch>, Tom Knutson <Tom.Knutson@noaa.gov>, Toru Nozawa <nozawa@nies.go.jp>, Doug Nychka <nychka@cgd.ucar.edu>, Claudia Tebaldi <tebaldi@rap.ucar.edu>, Ben Santer <santer1@llnl.gov>, Richard Smith <rls@email.unc.edu>, Daithi Stone <stoned@atm.ox.ac.uk>, Michael Wehner <MFWehner@lbl.gov>, Xuebin Zhang <Xuebin.Zhang@ec.gc.ca>, francis <francis.zwiers@ec.gc.ca>, Hans von Storch <hvonstorch@web.de>
date: Fri, 18 May 2007 17:26:15 -0400
from: Gabi Hegerl <hegerl@duke.edu>
subject: Re: 5AR runs next iteration- reply by 26th
to: Karl Taylor <taylor13@llnl.gov>

<x-flowed>
Hi all,

I see this. On the other hand, when some republicans did a grilling 
about attribution in some house subcommittee,
I was very happy to be able to resort to Tim's argument that the model 
runs were older than the heat uptake data
and therefore, there was no secret tuning in the 2001 ocean attribution 
results..

So using the 20th c for tuning is just doing what some people have long 
suspected us of doing...
and what the nonpublished diagram from NCAR showing correlation between 
aerosol forcing and
sensitivity also suggested.
Slippery slope... I suspect Karl is right and our clout is not enough to 
prevent the modellers from doing this
if they can. We do loose the ability, though, to use the tuning variable 
for attribution studies.

Should we ask to admit in their submission what variables were 
considered when tuning, and if any climate
change data were considered and at what temporal and spatial 
representation (global mean trend?),
and advise that we will not be able to use those models for any future
attribution diagrams? That would at least lay it in the open...

Gabi

Karl Taylor wrote:

> Hi Peter and all,
>
> There will clearly be different perspectives on this.  A model 
> developer will want to make use of all available observational 
> information to help decide whether his model is realistic or not.
>
> We can envision two candidate models that appear equivalent in most 
> respects, but one fails to produce ENSO's.  The developer would choose 
> the one that simulated ENSO.
>
> Likewise, suppose two candidate models were identical in most 
> respects, but one could accurately simulate the climate of the 20th 
> century (when all forcings were included), whereas the second had a 
> very low global sensitivity and produced too little warming.  The 
> developer would again want to choose the model that reproduced the 
> observed trends.  In fact this model would probably produce a better 
> estimate when forced by future emissions scenarios too (because, 
> presumably, its sensitivity is closer to the truth).
>
> It would be hard to argue that information about 20th century trends 
> shouldn't be used in model development.
>
> I agree that this may rule out attribution studies (following the 
> established approaches), but wouldn't we have to argue that 
> attribution studies are more important that model projections to 
> convince the groups not to consider trends in the model development 
> cycle?
>
> cheers,
> Karl
>
>
> peter.stott@metoffice.gov.uk wrote:
>
>> Hello everybody,
>>
>> We're having a lively debate in the Hadley Centre about whether climate
>> change experiments should be run as part of the model development
>> process, ie whether model developers should test their model against
>> climate change as they are developing their model. I think it might be
>> worthwhile us developing and expressing a view on this as we don't want
>> to risk getting into a position where attribution results in AR5 are
>> undermined by the development and model tuning procedure adopted by
>> modelling centres.
>>  
>> Also I don't think you quite captured the point that another reason for
>> separating out the ghg response from the response to other forcings is
>> to aid understanding, as we are finding out in trying to understand the
>> precipitation response. I think that requesting ALL, GHG, and NAT
>> ensembles would be the basic set.
>>
>> Best wishes,
>> Peter
>>
>> On Fri, 2007-05-18 at 10:33 -0400, Gabi Hegerl wrote:
>>
>>> Hi all.
>>>
>>>  From your comments, I assembled a word file with our suggestions on 
>>> the 5AR run
>>> proposal, but I am not sure
>>> I caught it all completely. Also, I had a chat with Jerry yesterday, 
>>> and he said getting
>>> suggestions of what should be stored will be useful at this point.
>>> My plan is to communicate this with Jerry when we are done with it, 
>>> and then propose
>>> it at the WGCM meeting.
>>>
>>> I drew a strawman list of what I could think of in 3 minutes, and am 
>>> asking you to
>>> add to it. Its all in track changes, so dont hesitate to go wild 
>>> (but please keep in mind that
>>> we need to restrict data requests to something you think you will 
>>> work with in the next
>>> years, since it is a fair amount of effort from the modelling 
>>> centres to haul the data over
>>> etc, and the more we request, the more likely it is that only few 
>>> ensemble members etc
>>> get sent...)
>>>
>>> Karl, I am cc;ing you since your perspective would be useful
>>>
>>> Gabi
>>>
>>

-- 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Gabriele Hegerl 
Division of Earth and Ocean Sciences, 
Nicholas School for the Environment and Earth Sciences,
Box 90227
Duke University, Durham NC 27708
Ph: 919 684 6167, fax 684 5833  
email: hegerl@duke.edu, http://www.env.duke.edu/faculty/bios/hegerl.html


</x-flowed>
