[CF-metadata] subgrid variation

[Jonathan Gregory]

Dear Brian

Thanks for considering this and for your suggestions. 

> 1. I think the cell_methods values should be single tokens with blank spaces
>    replaced by underscores, e.g., mean_over_sea instead of "mean over sea".

That's possible but would mean listing them as cell methods. I was thinking
that "over sea" etc. was a kind of qualifier like we have "within days" and
so on for climatological time, whereas the method itself (mean) is essentially
unaltered.

> 2. It may be convenient to use the same qualifiers as are used in the
>    standard names.  So replace over_TYPE by where_TYPE.

Yes, that would be nice, but the reason I suggested "over" was that the same
qualifiers could be used for time axes in cell_methods. In that case you would
want to say "when" rather than "where". I thought that "over" could be used
for both "where" and "when".

> 3. It may be convenient to allow a generic qualifier like "where_defined"
>    which could be used in most (all?) instances instead of the more
>    specific versions "where_TYPE".

To do that would mean defining the "defined" region for each standard_name,
I think, which we can avoid by allowing the data-writer to specify the TYPE
explicitly. The explicit TYPE is also more self-describing, I would argue.

Best wishes

Jonathan