A preliminary matlab interface for FAN

• To: netcdfgroup <netcdfgroup@xxxxxxxxxxxxxxxx>
• Subject: A preliminary matlab interface for FAN
• From: "Philip W. Miller" <pwmiller@xxxxxxxxxxxxxxx>
• Date: Fri, 13 Sep 1996 10:47:57 -0400

I have written a preliminary version of a matlab/fan interface.  The two
utilities, nc2mat and mat2nc, are mex-file analogs to the fan utilities
nc2text and text2nc.  I would like to submit it for inclusion in the
netcdf contrib library if there is interest.  These utilities supplement
the capabilities of the existing mexcdf in much the same way that fan
supplements the netcdf library.

These utilities are still quite crude, but they seem to work for me.
For the most part, I simply wrote a mex file interface to nc2text and
text2nc, so they should have the same capabilities and limitations of
the fan utilities.  Of course, these are preliminary versions, so they
have additional limitations imposed on them due to a lack of time and
programming skill on my part.  

The read utility, nc2mat, is fairly complete---it is the one I use most
often.  It was also the simple to write since both fan and matlab store
data internally as double precision elements.  There are some
enhancements that I would like to make to nc2mat.  Primarily I would
like to modify how the internal fan function replaces bad data (missing
values, values < valid_min, values > valid_max).  I would like to
replace these with NaN, -Inf, +Inf, but I do not know how to get the
representations of them in C code. I guess I could try nan = 0./0.; minf
= -1./0.; pinf = 1./0., but that is not appealing to me.  Any
suggestions?  I hope that there is a common IEEE arithmetic way to do
this, but I am pretty sure that it will be system dependent.  (I am
using a sun with the Solaris operating system.)

For completeness (and as a programming exercise for myself) I also wrote
a write utility, mat2nc, which may need further improvement before it
will be very useful to matlab users.  Part of the problem is a
limitation in my implementation, and part of it is a limitation in the
underlying fan utilities.  For instance one cannot use the fan utility
text2nc to do the following:

echo '[1 2 3; 4 5 6]' | text2nc 't.nc newVariable'

This makes more sense in the context of matlab where one might want to
write new data to a netCDF file; i.e., 

x = [1 2 3; 4 5 6];
mat2nc('t.nc newVariable', x);

But this is almost an ill-defined case, as netCDF is not really aimed at
such dynamic creation of data.  Of course, in matlab I can use the
mexcdf utility.  

The issue brings up a generic netCDF question on what is a good
convention for creating arbitrary dimensions.  For instance, if I have a
5 by 10 matrix, netCDF requires that I have a dimension of length five,
and a dimension of length ten before I can store the matrix.  I know one
can do this, (by having dimensions d5 = 5, d10 = 10), but what is a good
way to do it?  Is there a netCDF convention for it?

Thanks and let me know if there is interest in nc2mat and mat2nc.

Phil

------------------------------------------------------------------
Philip W. Miller
pwmiller@xxxxxxxxxxxxxxx

Lockheed Martin/GES, MS 137-232
199 Borton Landing Rd
PO Box 1027
Moorestown, NJ 08057-0927

voice: (609) 722-6979
fax:   (609) 231-5100
------------------------------------------------------------------