davidw@xxxxxxxx writes....
>After following much of this discussion on attribute conventions I am
>left wondering about something.... Many of these conventions would
>be a great idea if standardized. However, what good will they be if noone
>uses them?
>In my experience with netCDF data sets, I have NEVER seen a netCDF
>file created by a scientist type with Fortran code that has ANY
>attributes defined not even "UNITS" (computer scientist types working
>on netCDF packages excluded). There's just too many extra subroutine
>calls to do this to be bothered with by most users. (I'm not faulting
>netCDF here, its just that many people usually don't bother with
>attributes.)
David, I hate to prove you wrong, but we have been very careful to define
all necessary attributes for the NetCDF data structure we have designed here.
This includes units for all variables which have units, scaling factors for
variables where reals are packed into integers, fully descriptive variable
names (with title attributes where extra information is necessary), and global
attributes which list information such as a dataset description, time of
creation, and file naming information. I don't claim that it is perfect,
but it makes an attempt to be self-describing and complete.
The reason we did this is that we wanted the datasets (which contain images
from a NASA multispectral scanner) to contain all the information that anyone
would ever need to use the data. We designed the dataset to be used by
people other than ourselves, which is an important consideration. You can
use NetCDF much like a programming language. That is, you can write code
which you are going to use once and then forget about, or you can write
carefully documented and structured code which will be used again and again
by many differnt people on many different platforms.
So in general, it certainly is possible to design a NetCDF data structure
which is bare bones only, but it is also possible to carefully design a
structure which is self-explanatory and detailed. Suggesting that scientist
types (myself included) can't and/or won't use care in designing such
structures is not very helpful. Certainly, everyone who designs NetCDF
data structures ought to take care in doing so, just as they should when
developing source code.
Cheers,
Liam.
--
Liam E. Gumley | Phone : (301) 982-3700
NASA/GSFC MODIS Science Data Support Team | Fax : (301) 982-3749
Research and Data Systems Corporation | Internet : gumley@xxxxxxxxxxxxxxxxx
Greenbelt MD, USA |
