[CF-metadata] Standardizing how ensemble (realization) axes are encoded

From: Jonathan Gregory <j.m.gregory>
Date: Wed, 20 Nov 2013 10:22:17 +0000

Dear Steve

Sorry not to be clear.

> Mine would be this: We add "ensemble" as a standard_name. Then
> normal CF guidelines support its use in cell_methods, as in
> "ensemble: mean". Doing so also restores compatibility with CDM --
> an issue that effects real-world interoperability. It seems a clean
> and simple solution.

That is not my conclusion. I do not think "ensemble" makes sense as a standard
name. An synonym for "realization" is something with the sense of "ensemble
member". "Ensemble" means all the members together. I don't think it would be
meaningful to have a variable with "ensemble" as a standard name.

The existing conventions and standard_name table allow us to put e.g.
"realization: mean" in cell_methods to indicate a mean over the members of
an ensemble, in the case where there is no dimension named "realization".
In that case, the name "realization" is interpreted as a standard name, and
the cell_methods entry implies that the mean has been calculated over all the
members of an unspecified ensemble.

I feel that this convention is OK and probably adequate. However you and others
have suggested, I think, that it is not clear enough. Therefore my conclusion
is that we could (if you think it worthwhile) define "ensemble:" as a special
keyword that could be used in cell_methods to indicate that the statistic has
been calculated over the members of an unspecified ensemble. That would be
defined to mean exactly the same as "realization:" in the sense of the last
paragraph, but might be clearer.

Best wishes

Jonathan
>
> >As I said, permitting axis='E' would also be a convention change.
> >
> >If anyone would like to propose these additions in a trac ticket, I'd
> >support it.
> >
> >Best wishes
> >
> >Jonathan
> >
> >
> >On Fri, Nov 15, 2013 at 11:27:59AM -0800, Steve Hankin wrote:
> >>On second though it is all thoroughly ambiguous, whether CF names
> >>its axis standard_names for the collective or for the individual
> >>member. "ensemble" seems immediately clear and short, compared to
> >>"ensemble_member", which seems a bit labored. But I understand your
> >>concern over consistency with "realization". Either choice is fine
> >>with me.
> >>
> >> - Steve
> >>
> >>============================================
> >>
> >>On 11/15/2013 11:15 AM, Steve Hankin wrote:
> >>>On 11/15/2013 10:30 AM, Jonathan Gregory wrote:
> >>>>Dear Steve et al.
> >>>>
> >>>>>I support the idea that the term "ensemble" be allowed (by whatever
> >>>>>machinery) as an alias for "realization".
> >>>>It'd be fine to have an standard_name alias, I agree, but I
> >>>>think it should be
> >>>>"ensemble_member", not "ensemble". The ensemble is the
> >>>>collection of members
> >>>>(aka realizations).
> >>>It feels a bit backwards, doesn't it? By this logic shouldn't the
> >>>standard_name "time" be "snapshot" instead -- named for its
> >>>individual members, rather than for its collective?
> >>>
> >>>"time" "depth", "longitude" and "latitude" axes all represent the
> >>>collection of many individual points. By analogy an "ensemble"
> >>>axis would be the collective of many individual members. It's
> >>>true that, the choice of "realization" as a standard_name took the
> >>>opposite outlook. hmmm ... Are we better off to maintain
> >>>consistency with CF's well known geo-spatial axis standard_names?
> >>>or consistency with "realization"?
> >>>
> >>> - Steve
> >>>>>axis="E" also seems like an appropriate step to maintain consistency
> >>>>>with other well known axis types, given the high likelihood that
> >>>>>ensemble axes will become commonplace in the future.
> >>>>That would require a change to the convention to be proposed on
> >>>>a trac ticket
> >>>>by someone.
> >>>>
> >>>>Best wishes
> >>>>
> >>>>Jonathan
>

----- End forwarded message -----
Received on Wed Nov 20 2013 - 03:22:17 GMT