cc: Barrie Pittock <pittockatXYZxyzironmental-change.oxford.ac.uk>, Luis Jose Mata <l.mataatXYZxyz-bonn.de>, Stephen H Schneider <shsatXYZxyznford.edu>, "Kozak, LeeAnn" <LAKOZAKatXYZxyzthernco.com>, email@example.com, firstname.lastname@example.org, email@example.com, firstname.lastname@example.org, John Houghton <jthoughtonatXYZxyzcwg1.demon.co.uk>, email@example.com, firstname.lastname@example.org, Neil Leary <nlearyatXYZxyzcrp.gov>, email@example.com, firstname.lastname@example.org, email@example.com, firstname.lastname@example.org, Jerry Meehl <meehlatXYZxyzker.ucar.edu>, email@example.com, firstname.lastname@example.org, email@example.com, firstname.lastname@example.org, Barrie Pittock <barrie.pittockatXYZxyz.csiro.au>
date: Thu, 07 Sep 2000 18:29:53 +1200
from: Martin R Manning <m.manningatXYZxyza.cri.nz>
subject: RE: Extremes table
Thanks for your interesting comments on the uncertainty issue. Some
Being a physical scientist I have no problems with the frequentist approach
and this would be my preference where it can be applied. I also agree that
it is confusing, and possibly illogical, to associate uncertainties with
probabilistic statements - I have already argued against doing that in the
WG II SPM.
Your experience in forecasting must provide a strong basis for considering
measures of uncertainty or probability for WG I statements - but I would
argue that you can only apply this approach where the basis for predictions
is well established (e.g. the models being used are well validated and
The TAR also has to deal with some highly policy relevant issues that
depend on models which are still being developed. In these cases we have no
guidance from experience with previous forecasts and I do not see how the
frequentist approach to uncertainty can be used. Your comments about not
having a large "sample" echo the same concern.
But your response to this problem is....
>"Without a large sample, it seems to me that it is not possible to assess
>our confidence in our probabilistic statements. Our confidence is already
>built in --- for low probability events we don't expect them to occur very
>often. This is why I believe in working group 1 we changed from
>confidence statements to likelihood statements."
... and I am afraid I do not follow this argument at all.
If you can not apply a probabilistic approach then it seems to me that you
have to look for another measure of confidence in model results. To say
that the confidence is "built in" sounds like giving up rather than
addressing the problem. It does not recognise inherent limitations in
making assessments where views on the key driving processes are still
evolving, model structure still changing, and model validation still
sparse, as in the case of the THC. Estimates of confidence, based on expert
judgement as to how comprehensive the present models are, may not meet the
philosophical rigour of the frequentist approach, but do provide important
information in these cases.
The Uncertainties Guidance paper gave a number of ways in which you might
develop such confidence levels from expert judgement on an objective basis.
Your approach seems to limit the definition of uncertainty to the range of
results from current models and in doing so may be rejecting an important
source of information, viz any expert consensus on how good the models
Thus my bottom line is that we should be able to make statements about our
"confidence" in a result even when defining "uncertainty" in that result
may be problematic. And I really think that WG I need to consider and state
their confidence in some key statements - such as the THC issue.
There are of course other problems with producing simple summary statements
about THC changes. It is hard to define exactly what one is talking about -
most coupled ocean atmosphere models show a weakening of ocean circulation
patterns in response to GH warming - so the < 1% probability by 2100
statement being made by WG I presumably refers to some level of major
collapse in the THC. How do you set that level? How non-linear is the
dependence of the probability estimate on the chosen degree of THC
collapse? Then there are the time-scale issues - what is the probability
that GH warming by 2100 causes significant change in THC after 2100?
For these reasons I would prefer to see WG II deal with singular climate
change separately from the table being discussed for Ch 3.
Thanks very much for your comments - I am still learning from this
Martin R Manning
National Institute of Water & Atmospheric Research
301 Evans Bay Parade, Greta Pt
PO Box 14-901, Kilbirnie Phone: +64 +4 386 0535
Wellington Fax : +64 +4 386 2153
New Zealand m.manningatXYZxyza.cri.nz