To: Timothy Carter <firstname.lastname@example.org>,email@example.com
Subject: Re: Pattern scaling document for the TGCIA
Date: Sat, 18 Jan 2003 21:05:49 +0000
As promised some comments on the paper.
General: It is very good, just what is needed and puts the last 4 years of
debate into the right context.
General: why consistently 'climate changes' rather than the more usual
Abstract, line 10: why only quote as high as 0.99 and not the lowest
correlation (which actually is more to the point - it is still very good
after the 2020s, even for precip).
Abstract, lines 12-13: as worded this does not quite follow, although I see
from later that the ellipses used are at 95% confidence. Just because they
fall outside natural variability does not *in itself* prove they are stat. sig.
p.2, lines 17-19 (and also several places on p.4): impacts are mentioned,
but nothing said about adaptation. It is really adaptation
actions/decisions that are crucial, impacts are only one way to get
there. Alter the focus.
p.2, line -10: add 'necessarily' between 'not' and 'be'. AOGCMs may
actually do not so bad a job on occasions about climate change (relative
changes for example), so don't completely dismiss this one.
p.5, section 2: general point: there is no list or table or statement about
exactly what these 17 experiments are. The models are listed, but not the
experiments. e.g. which SRES scenarios did which modelling group and how
many ensembles? For the lay person this is not obvious.
p.7, top line: you should perhaps make the point that simple bias indices
such as these may partly be explained by elevation offsets (model height
vs. real height). It is to my mind a mitigating factor than can work in a
model's favour (not always). It should be mentioned, because the biases
may not be due to just dumb models, but due to simple resolution issues
that can be adjusted easily. A similar point perhaps applies in the next
para. about ocean/land boundaries. OK, you could say this just shows how
bad models are, but it perhaps gives people a poorer view of the model
physics and credibility than is truly needed. Another point to mention in
this para about precip. is the obvious point about decadal natural
variability. It's a tall order to expect the models to get the 1961-90
monthly mean precip. exactly right, owing to internal variability. Indeed,
give such variability can be plus/minus 10-20% or more it would be
astonishing if they matched. Be generous to models I say.
p.9, middle - interesting point about ECHAM4 and NCAR masks!!
p.15, para 2: didn't you have A1FI available from Hadley? Surely it could
have been used to test this? Last sentence in this para: why 'evidently
p.16, last line: interesting point here: if you claim the pattern-scaling
didn't work for the 2020s because of nat var (S/N ratios) then why actually
should we go with the raw model results anyway - certainly if it is the
signal we are interested in (and not the noise), it suggests the raw 2020s
models results are misleading us! This is a rather circular argument I
realise but the bottom line point again comes back to S/N ratios and the
role of nat decadal variabiliy, esp. for precip. Are we going to recommend
adaptations to noise or to signals - and why?
p.17, middle para: what about mentioning climate sensitivity here? I know
its out of vogue now, but PCM and NIES differences are explained by overall
model sensitivity aren't they.
p.17, para 4: this point about where agreement occurs between models is
important. Some people - I heard Wigley do it recently - write models off
at regional scales re. precip changes because they all disagree. They do
for some regions, but not all and where we think we have physical grounds
to accept agreement as legit. (e.g. UK; cf. UKCIP02 scenario metholody)
then we should be confident to say so.
p.17, line -7: why use 'forecasting' here? Could confuse some people. The
old argument about terms I guess. And again top line on p.18 is dangerous
- we can "predict" nat. variability in a stochastic sense using
ensembles. Change the wording.
p.18, line 9: not only are they difficult to forsee, they are simply
unforseeable to a significant extent because it is we who determine them; I
prefer to make the distinction between different types of prediction
problem more explicit.
p.18, lines 19-20: I don't like the use of 'truth' and 'precise' here. It
implies a strong natural science view prediction and the competence of
science (modellers!) which I think should be softened.
p.18, para 4: the inter-model differences bit being as large as the
inter-scenario differences. Again at least mention the role of nat var
here - some of these inter-model differences *must* be due to nat var, not
simply models not able to agree with each other.
p.19, para 1: I think the stabilisation case should be mentioned
here. What about pattern-scaling stab scenarios? As I hear it from DEFRA
and Hadley here in UK this was a big issue at the TGCIA meeting. Make a
comment at least; I think in principle p-scaling is probably OK (within
some limits) even here. I think you should make reference to some of Tim
Mitchell's work here (and/or elsewhere) since he has looked at some of
these things too. His thesis or his CC paper perhaps.
And finally, w/o sounding as self-serving as Tom Wigley, it would be nice
if you could reference (perhaps in section 3.3) the Hulme/Brown (1998)
paper in CR which was the first time I published scatter plots in this form
for GCMs results - and possible the first time this form of presentation
had been used anywhere (but I stand corrected of course; maybe I simply
picked it up from someone else).
So there it is: a great piece of work and a good write up. I don't know
Kimmo but pass on my congratulations to him. I'll look out for it on the
At 13:42 13/01/03 +0200, Timothy Carter wrote:
>Dear Mike and Tim,
>I know that you are not now involved in the TGCIA, but there is still some
>old baggage from the days of Mike's tenure that you may have some interest
>to comment on concerning regional pattern-scaling work.
>I attach a paper that we have prepared and distributed at the latest TGCIA
>meeting for comment (last week). If you have any comments, I would be very
>appreciative. I need comments if possible by the end of this week.
>The 96 pages of scatter plots are currently enormous files, and I can't
>possibly attach these for you to see. I am working on a way to get these
>substantially reduced in size. I have attached one example so you can see
>what to expect.
>Any feedback would be much appreciated. We intend to post this document,
>or something like it, on the DDC.
>Tim - have you published any of your Ph.D. results yet?
>Best regards and Happy New Year,
>Research Programme for Global Change
>Finnish Environment Institute (SYKE)
>Box 140, Mechelininkatu 13a, FIN-00251 Helsinki, FINLAND
>Tel: +358-9-40300-315; GSM +358-40-740-5403; Fax: +358-9-40300-390