[CF-metadata] size-one axes

From: Brian Eaton <eaton>
Date: Wed, 16 Jul 2003 10:01:06 -0600

Hi Jonathan,

On Fri, Jul 04, 2003 at 05:36:19PM +0100, Jonathan Gregory wrote:
> My proposal is that, for coordinate variables having a dimension of one,
> except for the basic xyzt (the ones which might have an axis attribute), it
> should be permissible to name the coordinate variable in the coordinates
> attribute *without* the dimension being a dimension of the data variable.
> That would allow many of such size-one coordinate variables to be associated
> with a data variable without insisting that it has many dimensions of size
> one. Since such dimensions do not affect the ordering of the elements of the
> array, they are not really needed as dimensions.
>
> The reason for excluding xyzt is that COARDS applications would expect the
> data variable to have these dimensions, even if they do have size one.

If we associate size one axes with data variables via the coordinates
attribute then it makes sense to allow them to be scalars which are easier
to implement than 1D arrays. The dimension is no longer needed as the
thing which associates the coordinate with the data variable.

Why exclude xyzt when those are the most common uses of size one axes? The
reason for introducing a "convenience" feature is to encourage data writers
to provide this metadata. If we only make it easier to add a small
subclass of the metadata of interest, then I don't think the additional
complexity of a new feature is justifiable. I don't see that COARDS
compliance is an issue. We're trying to make it easy for people to add
metadata that they might not otherwise add because including size one
dimensions in data variables is somewhat confusing, and more work than
associating a scalar via the coordinates attribute. Ag observed that
people rarely use size one axes and instead generally encode this
information in attributes and comments, and that when they do provide size
one axes they don't associate them with the data variables by adding the
appropriate size one dimensions. This reinforces my view that the current
method of using size one axes is too confusing to be widely adopted.

I prefer using the existing coordinates attribute to introducing a new
"size-one-coordinates" attribute.

Brian
Received on Wed Jul 16 2003 - 10:01:06 BST