>Date: Wed, 21 Oct 1992 10:53:33 -0700 (PDT)
>From: HANKIN@xxxxxxxxxxxx
>To: netcdfgroup@xxxxxxxxxxxxxxxx
>Subject: RE: Suggestion for Coordinate Mapping convention
In the above message Steve Hankin wrote:
>I suggest the use of "dummy" variables for the purpose of binding together
>components:
>
> dimensions:
> lat=20,lon=20,depth=5,vector_def=1;
> variables:
> float u(lat,lon,depth);
> u:long_name = "zonal speed";
> float v(lat,lon,depth);
> v:long_name = "meridional speed";
> float w(lat,lon,depth);
> w:long_name = "upwelling computed by divergence ...";
> char velocity(vector_def); // ** dummy variable **//
> velocity:long_name = "3-component velocity";
> velocity:components = "u v w";
Because we might wish to generalize this virtual variable mechanism at some
later date, I would argue for the following (slight) variation:
char velocity(vector_def); // ** dummy variable **//
velocity:long_name = "3-component velocity";
velocity:definition = "(u,v,w)";
That is: 1) the use of the attribute `definition' rather than `components';
and 2) the use of a mathematial expression for the value of the `definition'
attribute.
I'm not happy about the `char velocity(vector_def)' expression because it
appears too artificial. But, without adding, for example, a `virtual'
keyword to the CDL files and supporting the creation of virtual variables
in the netCDF API, this is probably as good as we can do for the moment.
--------
Steve Emmerson <steve@xxxxxxxxxxxxxxxx>
