NetNews Usenet Archive 1992 #27/NN_1992_27.iso/spool/sci/geo/meteorol/3382

Internet Message Format | 1992-11-17 | 4.5 KB

Path: sparky!uunet!know!mips2!news.bbn.com!noc.near.net!news.Brown.EDU!qt.cs.utexas.edu!yale.edu!spool.mu.edu!darwin.sura.net!zaphod.mps.ohio-state.edu!uwm.edu!rutgers!uwvax!meteor!stvjas
From: [e-mail] (Stephen Jascourt)
Newsgroups: sci.geo.meteorology
Subject: Re: STOP Re: mesoscale forecast model
Message-ID: <[e-mail]>
Date: 17 Nov 92 01:07:50 GMT
References: <[e-mail]>
Distribution: na
Organization: University of Wisconsin, Meteorology and Space Science
Lines: 79

In article <[e-mail]> [e-mail] (Doug Westphal) writes:
>Oh boy, here we go again: uninformed people making broad generalizations
>about topics or models which they aren't sufficiently familar with,

Uninformed? I am using one of the models I talked about and have colleagues
who have used the others, I have seen model output, been to conferences and
colloquia at which model results and problems were discussed. Also, I made
specific comments about the models rather than just a vague assertion.
If I recall correctly, Harold has also done some modeling and I'm sure at OU
he has had many discussions with more experienced modelers.

>causing the completely uninformed public to become confused. Great.
>Below are two examples of this kind of subjective commentary.
>None of their arguments are substantiated, nor can they be in this forum.
>
>>>
>>>From [e-mail]:
>>>
>>>The best mesoscale model around now is the The Penn State/NCAR model,
>>>    ^^^^^^^^^^^^^^^^^^^^
>>>MM4.  It beats the life out of the second most common model, CSU-RAMS.
>>>
>>>Harold Brooks                    [e-mail]
>>>National Severe Storms Laboratory/CIMMS (Norman, OK)
>
>
>>>From: [e-mail] (Stephen Jascourt)
>>>
>>>I have heard various rave reviews about the MM4, but my experience seeing the
>>>output of people using it (I haven't used it myself) is that it is filled with
>>>                           ^^^^^^^^^^^^^^^^^^^^^^^^
>>>problems, worst being that it smooths the heck out of *everything* in the
>>>horizontal, so you are left only with features that are strongly forced,
>>> < blah, blah, blah >
>>>Stephen Jascourt   [e-mail]
>
>The original posting and responses were reasonable and I would hope that
>the poster will now thoroughly research the different models before
>choosing which is best for him or her; but don't do it over Usenet.
>
>Dave Blanchard then says:
>
>>>Harold,
>>>
>>>That's a pretty strong statement.  Could you elaborate on why you believe
>>>PSU/NCAR (MM4?) outperforms RAMS?
>
>Noooooooo!!! The question of which is the 'best' model cannot be
>decided with a Gallup opinion poll or on Usenet. Let's not continue
>this, okay? The whole thing should be decided in an entirely different way:
>IN THE PEER-REVIEWED LITERATURE.

Dave asked a perfectly reasonable question-- he wanted Harold to provide
some information on what the MM4 does so well.

There is no "best" model. Each has certain things it does well and certain
things that give the model problems. There is a "best" model for a particular
application, but more likely the difference from one to another may be no
worse than their errors. And, worst of all, model intercomparisons are next to
impossible for a multitude of reasons. A particular model can perform very
differently by just changing a few parameters, and the parameter values that
work best for particular situations for one model may be different than the
values best for another model. For generic, artificial tests like dropping a
cold bubble in a box, reasonable comparisons can be made but the test itself
is rather artificial. For real data cases or realistic idealized experiments,
the initial numerically balanced fields will be different from one model to
another, so you're not really even starting off the same! And, of course,
most of these sophisticated models have many options for various
parameterizations such as soil models, radiation, and sub-grid scale turbulence,
by choosing different options for these you really have many different models
in one program.

As for whether these things should be "decided" on usenet, I think discussion
of the strengths and weaknesses of different models, what machines they can
run on and how much computational resources they require, etc. are good topics
for discussion-- they are part of an exchange of scientific information that,
after all, is supposed to be the function of usenet.

Stephen Jascourt   [e-mail]