date: Tue May  3 11:04:19 2005
from: Tim Osborn <t.osborn@uea.ac.uk>
subject: Fwd: 2005JD005799 Decision Letter
to: Craig Wallace <craig.wallace@uea.ac.uk>

   Not good news on our JGR paper.  I've only scanned them so far - rev2 is good, rev3 is
   biased because he/she doesn't see the point of pattern scaling, but rev1 makes some valid
   criticisms.  We need to read these in detail and decide what can be done.
   Tim

     X-Mailer: MIME::Lite 3.01 (F2.6; B2.11; Q2.03)
     Date: Mon, 2 May 2005 18:56:53 UT
     To: t.osborn@uea.ac.uk
     Subject: 2005JD005799 Decision Letter
     From: jgr-atmospheres@agu.org
     Reply-To: jgr-atmospheres@agu.org
     X-Spam-Score: 0.2
     X-Spam-Level: /
     Content-Disposition: inline
     Content-Length: 6317
     Content-Transfer-Encoding: binary
     Content-Type: text/plain
     Dear Dr. Osborn:

     Enclosed please find 3 evaluations on your manuscript entitled "Linear and exponential
     relationships between global temperature change and patterns of precipitation change"
     [Paper #2005JD005799].  After careful consideration,
     based on the reviewers' recommendations of your manuscript, I am sorry to inform you
     that I have decided to reject it for publication in the Journal of Geophysical Research
     - Atmospheres.
     If you wish to resubmit this paper, please note that you will need to send a
     point-by-point response to all of the reviewer's criticisms.  In this case, your
     manuscript will then be treated as a new submission.

     I am sorry I cannot be more positive.

     Sincerely,

     Ruth Lieberman
     Editor, JGR-Atmospheres

     ------------------------------------------------------------------------------

     Reviewer #1 Evaluations:
     Assessment: Category 5
     Ranking: Select one
     Annotated Manuscript: No
     Reviewer #1(Comments):
     Though this paper is one if the best-written papers I have received for review lately, I
     must recommend that this paper NOT be accepted for the following reasons:  The data in
     the paper do not support the authors' conclusion that the mixed scaling "replicates GCM
     data better", there is no attempt at cross-validation, there is no physical basis for
     the curve-fit, and in fact it is logically inconsistent, and finally, the metrics of
     goodness-of-fit given here are weak.  I addresss each of these points in more detail
     below, and finish with some techincal points.  However, the authors did convince me that
     an improvement on linear extrapolation would be useful, as well as of the need to
     develop better metrics.  Their method, and this paper fall short, however.
     Everything presented in this paper is a fit to data, with no cross-validation.  That is,
     the authors do not do not address the problem of extrapolation.  Yet this is critical
     for the applications of this technique.  For example, they  could have trained the fit
     on delta-T up to 1K and then "predicted" the higher values.
     This is purely an exercise in curve fitting.  There is no physical basis other than that
     p>0.   To get an idea of the pitfalls of this approach, consider the following thought
     experiment:  Suppose we had run the GCM experiments in reverse -- starting with high
     values of global temperatures and decreasing GHGs, and suppose that the GCM's followed
     the same trajectory down the GHG curve as they did up.  Where precipitation increased in
     the "normal" set of runs it would decrease in the "reversed" set.  According to the
     logic of this paper, you would fit the curve with a linear function in the first case
     and with an exponential in the second case.  Yet it would be the same data that you are
     fitting in two different ways!
     Their conclusion that the mixed linear-exponential model can " replicate GCM data
     better" is not supported by the evidence in this paper.  The performance of the mixed
     model was at best...mixed.  It depended on the GCM, and even then it depended in an
     unpredictable way on the actual temperature change.  I quote: "Given the clear curvature
     of local precipitation changes (with respect to global  temperature) for HadCM3, it is
     surprising that the global-mean precipitation is quite linearly related to delta-T, and
     the exponential and mixed functions are not as food fits as the linear function. For the
     NCAR PCM model, the slight curvature of the global precipitation-temperature
     relationship is reasonably captured by the mixed function for delta-T < 1.4K, and by the
     exponential function for delta-T > 1.6K, but the linear sclaing is a poor fit
     throughout."  Clearly the fit doesn't work for one model, and for another it is
     dependent on the temperature change in an unpredictable way.
     No overall metric of goodness of fit - such as mean-square error -  is presented. Only
     "extremes" of drying and wetting -- a small percentage of the globe-- and global-mean
     precipitation (a weak metric that hits method doesn't really do well at) are looked
     at.   Since the mixed functional fit is "worse" for the HadCM3 globally, then there must
     be many local regions where their fit is not good at all.
     Technical points:
     The method is not documented clearly enough to reproduce.
     In the title, abstract and in the text clearly state where this temperature is being
     measured.  Is it surface air temperature, surface temperature, or some other measure of
     global mean temperature?
     p<6mm/mon  (0.2mm/day) ignored.  What is the impact of this?  Why was this done?  If the
     attempt is to get better scaling for extremes then why exclude the dryest areas?
     50 year sliding boxcar(?)  window means that the samples used in the fits are not
     independent.  The SRES A2 scenario extends out to  the year  2100 -- so there are 3
     (1950 - 2100)  independent samples.  I am not sure whether they have taken this into
     account in their statistical test if Figure 1. The authors do not mention the length of
     the runs that they used.
     Reviewer #2 Evaluations:
     Assessment: Category 2
     Ranking: Very Good
     Annotated Manuscript: No
     Reviewer #2(Comments):
     Review of "Linear and exponential relationships between global temperature change and
     patterns of precipitation changes" by Osborn and Wallace.
     This is a well-written, succinct paper that makes some useful points relevant to scaling
     patterns of precipitation change, and I recommend its publication almost without change.
     The only changes that I suggest are:
     1.      ON page 1,line 4, I find it stylistically better to say, "the rate of
     precipitation decrease tends to decrease .... "
     2.      On page 2, lines 11-12, I would write, "or WHEN the regional pattern of sulphate
     aerosol forcing changes through time [  ], although for some models the regional
     temperature response to aerosol forcing is governed by the pattern of climate feedbacks
     triggered by overall warming, rather than by the forcing pattern [Harvey, 2004]
     3.      Page 4, 2nd last line, I find it stylistically better to say, "with a decreasing
     rate of change ...."
     Reference:
     Harvey, L.D.D. 2004. Characterizing the annual-mean climatic effect of anthropogenic CO2
     and aerosol emissions in eight coupled atmosphere-ocean GCMs. Climate Dynamics 23:
     569-599.
     Reviewer #3 Evaluations:
     Assessment: Category 5
     Ranking: Poor
     Annotated Manuscript: No
