[CF-metadata] CF and multi-forecast system ensemble data

From: Bryan Lawrence <b.n.lawrence>
Date: Tue, 31 Oct 2006 15:54:46 +0000

Hi Jonathan

> > a) we should create some standard names which exactly correspond to
> > some recommended global variables, and model integrations should use
> > global variables where they contain exactly one realization to indicate
> > these.
>
> In netCDF terms that should be "global attributes" rather than "global
> variables". I think that if a file contains data variables which have come
> from various sources, but the variables are not aggregated across sources,
> the global attributes could simply be converted into data variable attributes.
> There is no need to make them into scalar or size-one coordinate variables.

Agreed, Sorry, it was sloppy language on my part. I meant global
attributes!

> > Issue C: should we have a standard name modifier associated with a
> > variable on the aggregation dimension to show that it was produced by
> > aggregating file attributes?
>
> Maybe I have misunderstood, but I don't see a need for a standard name
> modifier here. Ancillary data, labelled by standard name modifiers, supplies
> metadata about the *individual* values of another data variable (CF 3.4), not
> about a particular dimension. If you wish to supply extra information along
> a particular dimension, this can be done by including another auxiliary
> coordinate variable (CF 6.1), as you also mention.

In the suggestion I've just made, I allowed ancillary data dimensioned
with different dimensionality. Reading your email makes me realise I
should have used auxiliary coordinate variables as Jamie was actually
suggesting, in which case his example was exactly right.

(Mind you, rereading 6.1.1 reminded me that an external vocab option
would make handling geographic regions cleaner too - while we can define
things all we like in CF, to actually use these names in software we
need regional boundaries coordinate datasets etc which we can only get
if we use external vocabs and their information systems).

> I'd say what is currently being considered is quite a modest expansion of
> the kind of approach we already have with global attributes.

The trouble with modest extensions is that after a lot of them one has
spaghetti. I'm also suggesting named containers, I just don't want them
to be confused with standard names of variable content.

Cheers
Bryan
Received on Tue Oct 31 2006 - 08:54:46 GMT