Missing connection in the CF data model: Cell Method -> Coordinate

[davidhassell]

Topic for discussion

The recent change in the conventions for data anomalies created a new connection in the CF data model, between the Field Ancillary and Cell Method constructs (see the current UML diagram below). During this work it came to light that there is another connection between data model constructs that has been missing all along.

This new connection is between the Cell Method and Generic Coordinate constructs.

The connection is needed to account for the case when a where or over cell method qualifier provides an auxiliary coordinate variable name, rather than a value from the area type controlled vocabulary. See https://cfconventions.org/Data/cf-conventions/cf-conventions-1.13/cf-conventions.html#statistics-applying-portions for details. In such a case, the different parts of the data array use different area types, each of which is defined by a data element of the referenced auxiliary coordinate variable.

The possibility of a connection between a Coordinate construct and a Cell method construct needs to be be explicit in the data model because the data model can not use inspection to make such links (such as looking for a coordinate construct with an appropriate standard name), which in turn is because inspection can not be robust in all cases (e.g. there could be two coordinate constructs with that standard name).

The current UML diagram is:

I envisage the new corrected UML diagram looking like:

The only substantive difference in the new diagram is the one-way arrow connecting Cell Method with Generic Coordinate Construct.

All other differences (mainly the moving of the white box towards to the right hand side from the centre) are just a rearrangement of the boxes to hopefully make the diagram less cluttered.

This new image needs to be accompanied by a small amount of new text (not yet written) in Appendix I to describe this connection, and to indicate how this is encoded in CF-netCDF.

Any thoughts are welcome on this topic.

Many thanks,
David