[CF-metadata] Encoding Errors on variables in CF

[Russ Rew]

Ag,

> I think that Russ's suggestion about prefixing "for_variable_" or
> "for_variables_" is fine. As long as this becomes a standard it probably
> doesn't matter too much what is decided on as long as the link is clear. The
> one-to-one and one-to-many mappings will also be potentially useful in
> making it clear to the user what variables are related in a file(s?).

Just to be clear, I didn't actually propose (or intend to propose) any
prefixes such as "for_variable_" or "for_variables_".  Instead I
proposed just using a variable attribute named "for_variable" or
"for_variables" to capture the intent that a variable provides
ancillary information for another variable.

Typically, attributes are used for ancillary information about a
variable.  However, in the netCDF data model, attributes can't have
attributes (such as units) and can't be multidimensional.  So when the
ancillary information is sufficiently complex, you need to store it in
another variable instead of in an attribute.  In such cases, I'm
recommending identifying the variable or variables to which the
ancillary information applies with an attribute named "for_variable"
or "for_variables" which has as its value a string that names the
variable(s) to which the ancillary information applies.

--Russ