date: Fri, 09 Jul 1999 12:03:50 -0400 (EDT)
Dear Chapter 13 authors
I attach some comments on your first draft- I hope they are useful.
With best wishes
Review of first draft of chapter 13 of the IPCC TAR
This is a good first draft- it reasonably short, well laid out, mostly easy
to read and generally to the point. It also sticks well to the assessment of
methodology of producing scenarios, not scenarios per se. I would urge the
authors to resist embellishing the chapter. I had some difficulty with the
approach in section 220.127.116.11
I found this section hard to read- long convoluted sentences, unnecessary and
numerous qualifiers, and a slightly odd choice of words.
eg lines 128-132
"for example, most future climate change scenarios include the characteristic
of increased tropospheric temperature (except in some isolated regional and
physical circumstances) and that particular condition is one in which most
climatologists have most confidence (I.e. believe to be the most
plausible)(Schneider et al,1990)"
which means (I think)
"For example, most scenarios include increased surface temperature as this
the change climatologists think most plausible"
line 381 could reference Cubasch et al, 1994, Mitchell et al, 1999
Another advantage of the using the latest period is that impacts are
evaluated with respect to present (i.e. the present status of the parameter
being impacted) and not to a period some period in the past.
A disadvantage is that every ten years or so, the baseline moves, and new and
old studies become less compatible. If the rate of climate change
accelerates, then this will become more of a problem.
There is a philosophical problem here- is one really gaining any skill by
It is not just that the model's level of skill is restricted to much larger
scale than that of a grid point. Even with a perfect model, the scale at
which the model is skilful is quite limited, (see Stott and Tett, 1998(?)),
because of internal variability.
Regional models and MOS methods may help to some extent by allowing for
mesoscale effects, particularly those due to orography and changes in surface
type. However, even if the GCM is skilful at its gridscale, the impacts
modeller must remember that downscaled fields are plausible rather than
necessarily more skilful.
Statistical downscaling relies on the availability of long, good quality
timeseries of the appropriate climatological data. Hence there are regions of
the globe where one would struggle to implement this approach.
Lines 858-861 Leave the assessment of changes in ENSO to chapter 9. For the
purpose of this discussion it is sufficient to state that we have little
confidence in the predicted changes for ENSO- some models do predict an
increase in frequency, and this could be incorporated in scenario development.
871-874 Again, this should be based on chapter 9. If chapter 9 doesn't cover
this adequately, then help them.
There is also an uncertainty form getting from concentrations to forcing (eg
the indirect aerosol effects)
I presume signal to noise ratios means sampling uncertainty.
931-5 I think this repeats what was said earlier in the chapter.
I wasn't quite sure where this was leading- may be I misread the sign posts-
I think the issues are
I think for impacts the things of interest are Tf and changes in the
distribution of T" which together are needed to evaluate extreme events.
The control distribution of T" (orTf/T") is of interest from the point of
view "will we notice the any differences?"
Models give Tf +T" - (or Tf +T"/(n)**0.5 from an ensemble of n) and so do not
give a clean estimate of Tf.
One point is that many studies to date have used data where T" is comparable
to Tf because of the short meaning period. Giving the standard deviation of
T" can help impact scientists to assess the signal to noise ratio of their
changes, and hence ignore insignificant changes.
Another problem is that even with a long control run, it is not simply
possible to add the distribution of T" estimated from the long control to the
estimated Tf from a model run to get the mean and spread, because Tf will be
contaminated by T".
Another issue is using ensembles for a non-linear impact- the mean of the
impacts for each ensemble is a better measure than the impact of the mean
1034 scale dependant, AND dependant on what variable is used.
1042- 1044 This is more to do with uncertainty in the response to forcing
and, if included, should got there. Averaging across models as a way of
improving accuracy has no scientific basis and is highly controversial. If
models all have the same sort of error, then this will not work. If the model
errors are smaller than the level of internal variability on the timescale of
interest, then this will work as one is effectively increasing the "noise"
1203ff There is no easy answer to this. There is some comfort in having a
reasonable simulation of present day climate, particularly in and around the
region of interest. As noted , having neighbouring sea-ice/snow cover in the
right place, and the correct positioning and seasonal movement of the
rainbelts affecting the area of interest is obviously a good idea. But as
noted in chapter 9, one can vary parameters in a model which have little
effect on present day climate, but alter the sensitivity to climate change.
This implies a good simulation of present day climate is not a sufficient
condition for accurate simulation of climate change- the chapter should note
this. (It is even possible that a model with a poor simulation of present day
climate could provide a more accurate simulation than one which has a good
simulation of present climate- if it contains a better representation of the
dominant feedbacks) However, it is difficult to validate the feedbacks.
1294 albedo (or does snow have a high albino?)
1301 -1313 update and harmonise with sea-level chapter-let them do the
Attachment Converted: "c:\eudora\attach\CHPT13~1.DOC"