date: Wed, 19 Jan 2005 10:51:33 -0500
from: Edward Cook <drdendroatXYZxyzo.columbia.edu>
subject: Re: Re:
to: Gerard van der Schrier <g.schrieratXYZxyz.ac.uk>
I wouldn't contact Dai until you are certain about the problem in his
calculations. Perhaps simply showing him a comparison of your PDSIs for
Europe and his at the same grid point location with freak values would
be enough to stimulate a useful initial discussion, or do as Keith
suggests with maps. I don't know Dai at all to say how he handles
possible criticism. I do know that Trenberth, his co-author, is
extremely defensive and combative when ever criticized about anything
because he figures that he is smarter than everyone else and virtually
infallible. So at the start, if you do contact Dai, do not include
Trenberth in the communication. Let Dai sort that issue out himself.
I am glad to hear that the SC-PDSI also appears to improve things for
Europe as well. It appears to be the way to go. With respect to the
program, have you written your SC-PDSI program in Fortran? And if so,
is it possible to get a copy of the source code?
I would also appreciate seeing a copy of your European SC-PDSI paper
when you feel comfortable with sending it to me.
On Jan 19, 2005, at 10:09 AM, Gerard van der Schrier wrote:
> Dear Ed,
> Thanks for sending me the pdfs of those articles. And the
> recommendation to read the supplementary material.
> I've computed the self-calibrating PDSI for Europe only (but it is
> trivial to do that for other regions in the world). The SC-PDSI is a
> real improvement over the 'ordinary' PDSI: many thanks for suggesting
> to look into this! About two weeks ago, I gave a first draft of a
> paper on the European SC-PDSI to Keith and Phil (haven't heard
> anything since). Initially the idea was to make an update to a
> publication of Keith on European PDSI from 1992.
> About the Dai et al. PDSI: they used the code available from the NGDC.
> This code is a modified version from the 1980s, when internal memory
> of the computers was very limited. I'm certainly not a computer-gizmo,
> but my guess about the mistake in Dai et al. is that they simply took
> this code, put some do-loops around it (to compute PDSI for every
> gridbox), but then failed to properly declare and initialize the
> arrays. Fortran might get internal memory places mixed up when
> declarations & initializations are not properly sorted out, especially
> when you are working with a code that it written to be as economical
> as possible with internal memory. At least: this all happened to me
> when I had my first try on computing PDSI. I also got freak values of
> 20 or so. The PDSI algorith then makes sure that these freak values
> are quickly damped away, which makes their impact relatively short.
> I haven't told Dai et al. about this. Should I? Keith urged me to
> explicitly show one of their most horrible PDSI maps and compare it to
> ours. They also have freak values in Europe.....
> Thanks for the comments about the preferred format. I'll stick to that.
> I'll be in touch.
> Cheers, Gerard
>> Hi Gerard,
>> I had a nice talk with Keith yesterday about a number of things
>> including asking about your progress in generating monthly gridded
>> PDSIs over global land areas. Keith indicated that you had
>> successfully programmed the self-calibrating PDSI algorithm and had
>> calculated gridded PDSIs for at least Europe (if I understood him
>> correctly). That is great.
>> Do you have any timetable for producing gridded PDSIs over global
>> land areas? I am very interested in comparing what your PDSIs with
>> those of Dai first of all. (Maybe you have already done that) Keith
>> indicated that you discovered a computational error in Dai's program
>> that produced unusually extreme PDSI values in some cases. (Has Dai
>> and that miserable prat Trenberth been told about this? As you might
>> tell, I have genuine dislike for that arrogant Kiwi) That is not
>> surprising to me because I found PDSIs >20 in some of Dai's Mexican
>> grid point estimates. I also suspect that the quality of the monthly
>> precipitation and temperature data used by you is better than that
>> used by Dai. So, between not having the computational error and
>> having what I suspect to be better quality input data, your PDSI data
>> will be a significant improvement over Dai's data. Added to that, the
>> self-calibrating PDSI algorithm would appear to add even more
>> improvement. All this being the case, the next step would be for me
>> to re-do my North American PDSI reconstructions with your gridded
>> PDSI data when they are available. Please let me know when this might
>> be the case. I also have a Chinese graduate student here from Lanzhou
>> who will work on producing gridded PDSI reconstructions for China and
>> surrounding land areas (e.g. Mongolia) from long tree-ring records.
>> Your gridded PDSI data over that region would also be extremely
>> useful for that project.
>> I also talked with Keith about the eventual format of the global land
>> area PDSI file. As you know, Paul Krusic has written a very handy
>> Fortran program for extracting sub-regions of monthly climate data
>> from the Mitchell et al. (2003) 0.5� global data sets. It would be
>> great if your eventual global PDSI data, derived from the same
>> Mitchell et al. data (I believe), were organized in the same format
>> as Mitchell's data so that Paul's program could be used with minimal
>> Finally, I have attached pdfs of my recent drought paper in Science
>> and its Supporting Online Materials for you interest (the latter is
>> very important to read in my humble opinion). Also, my recent QSR
>> paper on the Esper et al. estimates of past temperature is attached.
>> Dr. Edward R. Cook
>> Doherty Senior Scholar and
>> Director, Tree-Ring Laboratory
>> Lamont-Doherty Earth Observatory
>> Palisades, New York 10964 USA
>> Email: drdendroatXYZxyzo.columbia.edu
>> Phone: 845-365-8618
>> Fax: 845-365-8152
Dr. Edward R. Cook
Doherty Senior Scholar and
Director, Tree-Ring Laboratory
Lamont-Doherty Earth Observatory
Palisades, New York 10964 USA