cc: "Sarah Raper" <S.RaperatXYZxyz.ac.uk>
date: Fri Feb 10 13:59:10 2006
from: Tim Osborn <t.osbornatXYZxyz.ac.uk>
subject: Re: a thought on deltaQ2x
to: Jonathan Gregory <j.m.gregoryatXYZxyzding.ac.uk>
Thanks for data. Here are results for all 4xco2 runs. No time to think about it further
today, nor time to (e.g.) exclude the flat part of the forcing, as you suggested. Some
results *very* weird: ECHAM5 and NCAR PCM with NEGATIVE lambda - what's going on there? I
will look into this next week. Will also estimate regression uncertainty.
At 13:32 10/02/2006, you wrote:
Aha! I had wondered if it might be something assuming the time-profile, but
had guessed that might require some more elaborate minimisation, not having
worked through the eqns. Ingenious.
I suppose a problem is that F/Qt tends to be rather small, since F tends to
a small part of Q (heat uptake is less important than climate sensitivity).
Also the part you want is the part with the increasing forcing, not the
constant forcing. Maybe the 1pctto4x runs would work better. If I were you
I would stop soon after stabilisation so that the regression isn't dominated
by the cloud of values near the destination.
I attach a new tar file in which I have used straight-line fits to the control
segments instead of the control annual means themselves. In almost all cases
this reduces the uncertainty on my fitted slopes, also attached.