Dear Mark

I wonder what you think about this (from an email I sent on Fri, but it seems
that it might not have got posted until yesterday):

> Thanks for clarifying the use-case. I agree that we don't have an existing way
> to provide this information, and it would be fine to give it a new standard
> name. I suppose you could attach this information to the data variable using a
> scalar coordinate variable - is that what you think?

I've thought a bit more about this. Since there is only *one* member from
the original ensemble (rather than a subset of more than one) could we not
regard this as a collapse by a point cell_method? The point method represents
variation within the uncollapsed dimension by selecting the value at a
particular location. That is exactly what you are doing by selecting one
member from the ensemble.

If it were a cell_method, then as discussed in ticket 108, we could record
the original ensemble size as the original dimension e.g.

cell_methods="realization: point (dimension: ensemble)"

where realization is an existing standard name, and ensemble is a netCDF
dimension containing the original ensemble size. No standard name would be
needed.

Best wishes

Jonathan

> ----- Forwarded message from "Hedley, Mark" <[email protected]> 
> -----
> 
> > Date: Fri, 31 Oct 2014 14:50:53 +0000
> > From: "Hedley, Mark" <[email protected]>
> > To: John Graybeal <[email protected]>
> > CC: CF Metadata List <[email protected]>
> > Subject: Re: [CF-metadata] realization | x of n
> > 
> > I'm happy to be more specific and stick with
> > 'original ensemble'
> > as it meets my use cases just fine.
> > 
> > So, I think that the proposal stands as:
> > 
> > standard_name:
> > number of realizations
> > 
> > units:
> > ''
> > 
> > description:
> > In a model or operational forecast, the number of member realizations 
> > within a given ensemble. This provides context for any specific 
> > realization, for example orienting a member relative to its original group 
> > (even if the group is no longer intact).
> > 
> > many thanks
> > mark
> > 
> > ________________________________
> > From: John Graybeal [[email protected]]
> > Sent: 30 October 2014 23:14
> > To: Hedley, Mark
> > Cc: CF Metadata List
> > Subject: Re: [CF-metadata] realization | x of n
> > 
> > Glad you liked the text!
> > 
> > Regarding 'given ensemble' vs 'original ensemble', how can we resolve the 
> > ambiguity? That is, if you use this attribute, how will the user know what 
> > ensemble the attribute is in reference to?
> > 
> > If the 'common practice among forecasters' (and required capability) is 
> > exclusively describing the originating ensemble, I propose the name and 
> > text should reflect that narrower definition, to avoid misuse. (I'm hoping 
> > for this case.)
> > 
> > If the common practice includes both use cases, somehow the user needs to 
> > derive which meaning applies -- either we need to define two standard 
> > names, or suggest in the definition that the variable name or long_name 
> > should resolve it, or something. (We could be deliberately vague as well, 
> > but a sentence like "This could refer to either the original ensemble for 
> > this realization, or a more recent collection in which the realization 
> > occurs." would help make that explicit.)
> > 
> > John
> > 
> > On Oct 30, 2014, at 10:44, Hedley, Mark 
> > <[email protected]<mailto:[email protected]>> wrote:
> > 
> > Thank you for the feeedback
> > 
> > John:
> > I like the text
> >   In a model or operational forecast, the number of member realizations 
> > within a given ensemble. This provides context for any specific 
> > realization, for example orienting a member relative to its original group 
> > (even if the group is no longer intact).
> > I would like to use this as is in the proposal.
> > 
> > > Reviewing this and going back to your original request, there is still a 
> > > likely point of confusion for users -- it isn't obvious that "given 
> > > ensemble" refers not to the currently constituted collection, but to the 
> > > one originally created with this realization.
> > 
> > > If you want that to be the use case for this standard_name (for 
> > > everyone), I think 'within a given ensemble' needs to explicitly say 
> > > something like 'within its originally created ensemble'. And perhaps the 
> > > standard name itself should follow that thought, something like 
> > > 'initial_number_of_realizations'.
> > 
> > I had thought about this, but my consideration was that there are ensembles 
> > which are created after the fact, not necessarily in the 'originally 
> > created' set; e.g. multi-model ensembles.  I considered leaving the name so 
> > that it could be used in this context as well.  This is not a strong use 
> > case for me, so I would be content to be more specific if that is 
> > preferred, but I didn't see the need to, so I left it more general.  I'm 
> > happy to be guided on this aspect.
> > 
> > 
> > Jonathan:
> > > Maybe you are dealing with an intermediate case, having a subset of the 
> > > ensemble members, and you want to record how many there originally were 
> > > in total. Is this a common use case? It seems rather surprising to me. 
> > > But I'm not sure that's what you mean.
> > 
> > Yes, this is what I mean.  I have one of the ensemble members, I have 
> > chosen it from the collection and passed it to a friend, for reasons best 
> > known to myself; I want to label it as member x from emsemble of size y.  I 
> > am confidently assured this is common practice amongst forecasters and the 
> > capability is required.  It has been an explicit part of the GRIB 
> > specification for years.
> > 
> > >> seven of nine
> > > But this seems different. It's not the number of members there are, but 
> > > the ordinal number (7) of this particular member. Why can't that be 
> > > recorded in a variable with the existing standard_name of realization?
> > 
> > there are two pieces of information here, in CF terms this is:
> > realization = 7
> > number_of_realizations = 9
> > I just unpacked this into a single label, to illustrate the information 
> > wanted (but I seem to have reduced clarity again; never mind).
> > 
> > mark
> > 
> > ________________________________
> > From: John Graybeal 
> > [[email protected]<mailto:[email protected]>]
> > Sent: 30 October 2014 17:10
> > To: Hedley, Mark
> > Cc: CF Metadata List
> > Subject: Re: [CF-metadata] FW: realization | x of n
> > 
> > Hi Mark,
> > 
> > It is a worry if the definition is a repetition or variant of the words in 
> > the name. In particular, the word 'realization' will be meaningful to 
> > modelers/forecasters but not universally.
> > 
> > My first desire was to generalize the term (e.g., 'how many entities are in 
> > a collection of that type of entity'), but I suspect that will be annoying 
> > to the primary users. So can we make it specific and say
> >   In a model or operational forecast, the number of member realizations 
> > within a given ensemble. This provides context for any specific 
> > realization, for example orienting a member relative to its original group 
> > (even if the group is no longer intact).
> > Or else, define what we mean by 'realization' and 'ensemble'.
> > 
> > Reviewing this and going back to your original request, there is still a 
> > likely point of confusion for users -- it isn't obvious that "given 
> > ensemble" refers not to the currently constituted collection, but to the 
> > one originally created with this realization.
> > 
> > In my use case, the whole ensemble is not present, I only have a subset of 
> > the members. I have a metadata element telling me how many members there 
> > were at the time the ensemble was created, which I would like to encode.
> > 
> > If you want that to be the use case for this standard_name (for everyone), 
> > I think 'within a given ensemble' needs to explicitly say something like 
> > 'within its originally created ensemble'. And perhaps the standard name 
> > itself should follow that thought, something like 
> > 'initial_number_of_realizations'.
> > 
> > John
> > 
> > 
> > ______________________________________
> > From: CF-metadata 
> > [[email protected]<mailto:[email protected]>] 
> > on behalf of Jonathan Gregory 
> > [[email protected]<mailto:[email protected]>]
> > Sent: 30 October 2014 16:40
> > To: [email protected]<mailto:[email protected]>
> > Subject: [CF-metadata]  FW:   realization | x of n
> > 
> > Dear Mark
> > 
> > > Please may people raise any further concerns about a new standard name:
> > >   number_of_realizations
> > > with a canonical unit of
> > >   ''
> > > and a description of
> > >   The number of member realizations within a given ensemble.
> > 
> > My concern is probably the same one as before. Sorry about that. Does this
> > mean the number of members the ensemble has got? If it does, why does it 
> > differ
> > from the ensemble dimension? If the ensemble dimension has been collapsed to
> > size 1, we could record this in cell_methods. Maybe you are dealing with an
> > intermediate case, having a subset of the ensemble members, and you want to
> > record how many there originally were in total. Is this a common use case?
> > It seems rather surprising to me. But I'm not sure that's what you mean.
> > 
> > > This name enables a single member from an ensemble to explicitly be 
> > > labelled, e.g.
> > >   seven_of_nine
> > > which is often required in operational forecasting.
> > 
> > But this seems different. It's not the number of members there are, but the
> > ordinal number (7) of this particular member. Why can't that be recorded in 
> > a
> > variable with the existing standard_name of realization?
> > 
> > Cheers
> > 
> > Jonathan
> > _______________________________________________
> > CF-metadata mailing list
> > [email protected]<mailto:[email protected]>
> > http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
> > 
> > 
> > On Oct 30, 2014, at 01:40, Hedley, Mark 
> > <[email protected]<mailto:[email protected]>> wrote:
> > 
> > Thank you for the discussion on the number of realizations in an ensemble.
> > 
> > Please may people raise any further concerns about a new standard name:
> >   number_of_realizations
> > with a canonical unit of
> >   ''
> > and a description of
> >   The number of member realizations within a given ensemble.
> > 
> > This name enables a single member from an ensemble to explicitly be 
> > labelled, e.g.
> >   seven_of_nine
> > which is often required in operational forecasting.
> > 
> > I would like this to be added to the standard name list.
> > 
> > thank you
> > mark
> > _______________________________________________
> > CF-metadata mailing list
> > [email protected]<mailto:[email protected]>
> > http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
> > 
> > _______________________________________________
> > CF-metadata mailing list
> > [email protected]<mailto:[email protected]>
> > http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
> > 
> 
> > _______________________________________________
> > CF-metadata mailing list
> > [email protected]
> > http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
> 
> 
> ----- End forwarded message -----
> _______________________________________________
> CF-metadata mailing list
> [email protected]
> http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata

----- End forwarded message -----
_______________________________________________
CF-metadata mailing list
[email protected]
http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata

Reply via email to