[CF-metadata] "positive" attribute

From: Jonathan Gregory <j.m.gregory>
Date: Thu, 28 Jun 2007 15:28:29 +0100

Dear Bryan

> Your original email started me worrying about a number of things. I'm aware
> that we're busy building standard names with embedded sign conventions, and
> I'm not entirely comfortable about that.

This is not a new convention. The standard names have had explicitly stated
sign conventions in them from the inception of CF i.e. for more than seven
years. I think the strongest argument for this is that it makes it impossible
to omit the sign convention! If we did not state the sign convention in the
standard name but had it in a separate attribute, it is 100% certain that it
would sometimes be omitted, with the result that the data would not be self-
describing. Not knowing the sign convention is a serious problem for analysis
of data. The fact that the data would not be CF-compliant would be no
consolation for its being useless. It is better to prevent problems happening.

> What happens if I build (or have) a model which happens to have a different
> sign convention than the one chosen thus far? Do I have to reprocess all my
> data, or introduce a duplicate set of standard names with the opposite
> convention embedded. The latter doesn't seem right!

You have to introduce other standard names. We already have several upward/
downward pairs for this reason, but it turns out that in practice there is not
lot of such duplication, because people tend to use common conventions,
so it's never been suggested to present a significant problem. Hence I think
the evidence is that this pragmatic decision was reasonable, and although
other decisions could have been taken there is at present no reason to change
the guidelines.

Best wishes

Jonathan
Received on Thu Jun 28 2007 - 08:28:29 BST