[CF-metadata] files with no data variables

From: Jonathan Gregory <jonathan.gregory>
Date: Tue, 25 Feb 2003 13:25:30 +0000

Dear Karl

PRISM also needs to store "grid" information in files separate from data
variables, for the same reason as OCMIP i.e. to save space. Perhaps CDMS can
already handle this when it aggregates files into a dataset?

For coordinate variables (in the Unidata sense) and their bounds, there is
little problem in CF with putting them in a file with no data variables. In that
file they are identifiable as coordinate variables because of their names being
the same as the names of their dimensions. The files with data variables and no
coordinate variables are defective by themselves, but OK when aggregated with
the coordinate variable file.

This is not the case for CF auxiliary coordinate variables. If they are put in
a file without data variables, they look like data variables, since they are
normally distinguished by being listed in the coordinates attribute of a data
variable. But again I don't think there is any problem when they are aggregated
with the files containing the data variables.

It is therefore possible to store the grid information in a separate file
without introducing any more concepts in CF, though I agree we could comment
and provide guidance on this possibility. Up to now we have taken the view that
CF deals with the correct construction of a single netCDF file by itself, not a
collection of files. It isn't possible, however, to create an instance of a
"grid", defined by a combination of coordinate variables (and perhaps a grid
mapping), but containing no data. I don't know whether we need to be able to do
this, but that's what my abstract variable proposal is about.

Best wishes

Jonathan
Received on Tue Feb 25 2003 - 06:25:30 GMT