[CF-metadata] How to handle a forecast model with non-monotonic coordinate variables

From: Jonathan Gregory <j.m.gregory>
Date: Fri, 21 Feb 2014 22:35:50 +0000

Dear Jim

> For Rich?s case it may be that the complex grid network is stretched out somehow into a rectangular array space, so there are no coordinates to have for the ?missing? elements.

But that seem to imply there are no (non-auxiliary) coordinates either. Perhaps
that was the case. I've forgotten. Is that right, Rich?

> As to whether or not missing values have ever actually been forbidden in auxiliary coordinates (except in people?s minds), here?s the definition from the CF conventions document.

Yes, aux coord vars are not coord vars, but they contain coordinate data.
Table A indicates vars which indicates coord data with C. _FillValue and
missing_value did not have C until it was added for the sake of chapter 9.
I am certain we intentionally did not previously allow aux coord vars to
include missing data.

What I am wondering is if anyone can recall or think of something which would
go wrong if we allowed this. Obviously one thing that probably would go wrong
is that existing software, not expecting to check for missing data in aux
coord vars, would wrongly interpret missing data if it was actually included.
If a change would break existing software, there has to be a good reason for
making the change.

Cheers

Jonathan
Received on Fri Feb 21 2014 - 15:35:50 GMT