[CF-metadata] Multiple file datasets

From: Jonathan Gregory <j.m.gregory>
Date: Tue, 24 Nov 2009 08:23:27 +0000

Dear John

> Do we use "cross referencing variables" just in the "coordinates"
> attribute or are there other places?

There are several of them apart from coordinates: variables are named by
cell_measures (as Balaji mentioned), formula_terms, ancillary_variables,
bounds, climatology, grid_mapping.

If the variable named exists within the file that refers to it, or if there
is only one variable of that name anywhere in the dataset, it is easy to use
it. If we could agree that this is allowed, that would be a useful first step
for handling datasets comprising several files.

A more complex step is the one you mention, of how to aggregate files when
they contain duplicate variables. We could define CF rules for that too.

> NcML does this, as apparently does the cdms
> cdscan utilility, and im sure lots more. These external systems that
> define collections may have semantics about variable names. NcML, for
> example, has a "union" aggregation that logically combines all variables
> with distinct names into a single dataset, ignoring duplicates. It also
> has a "join" aggregation that logically combines all variables with the
> same name along an "aggregation dimension", essentially concatenating
> them, eg for time slices.

It's interesting to know what NcML does. The CF convention states that nothing
depends on variable names. A CF-compliant aggregator, I think, would use only
the metadata to decide whether variables should be aggregated, not their names.

Best wishes

Jonathan
Received on Tue Nov 24 2009 - 01:23:27 GMT