[CF-metadata] CF data model

From: Jonathan Gregory <j.m.gregory>
Date: Wed, 12 Jan 2011 14:33:52 +0000

Dear Bryan (and Steve)

Thanks for your diagram and comments. Yes, it would be good to sit down with
you and see if we can reconcile our views expressed in words and pictures. :-)
We can talk about that in other emails so that CF doesn't have to join in the
discussion of our diaries.

I may be disagreeing somewhat with you and Steve about whether the netCDF
model should restrict us. It is where we started, but we may have constructed
a logical model which has some more abstract concepts than netCDF does. For
example, in netCDF we associate a dimension and a (Unidata) coord var by the
equality of their names. In the CF model, the dimension and its coord var are
part of the same thing, not two things with the same name. That is implied by
your diagram too, which doesn't show "dimensions" on it. Also, in my model I
have lumped together some things which CF does in different ways. For
instance, I suggest that non-scalar formula terms are really auxiliary coord
vars, and scalar coord vars and size-one coord vars are the same sort of thing.
These are just my suggestions, which I am making because I think it might help
how we think about further developments, including the representation of CF
in other file formats (as you say), and the use of CF as a data model in an
OO language.

> - I don't like the concept of a space construct. It's a new term to me
> which doesn't carry any useful attributes. I much prefer the use of
> object, which *is* a language neutral term (in data modelling).

You don't like the term or you don't like the concept? If "object" is a
language-neutral term, I'm happy to call it a "space object" instead. It
corresponds to a data variable, but I don't call it that because I think
everything in the logical model - all the metadata and the coordinates - are
part of it. So it is data and the space in which the data resides. But if
you haven't yet filled in the data, it is just the space. It is possible in
memory that you might have a space object with no data array in it, but by
the time you write it out to a CF-netCDF file, you must have a data variable.

> -I don't like central assumptions that are violated. Either we make it
> or we don't. You can make that assumption property of some attributes of
> space constructs.

That is a good point. I suppose it means there is a higher level, perhaps
a fragile and optional one, which relates some of the space objects. A higher
level would also be needed to relate staggered grids and construct supergrids,
as suggested by Balaji, but this hasn't been done in CF-netCDF yet.

> - I think text without pictures is just as unhelpful as pictures without
> text (my version :-).

Let's see what we can do about that together.

Best wishes

Jonathan
Received on Wed Jan 12 2011 - 07:33:52 GMT