[CF-metadata] Getting back to ensembles

[John Caron]

Simon Wood wrote:
> Dear Paco, Jonathan and others,
> 

> conceptually the dataset is surely dimensioned 
>   (ensemble, reftime, time, level, latitude, longitude).  

Hi Simon:

This is the way I have been thinking of these kinds of datasets. It is especially natural when there are always the same number of forecast/valid times for each reftime. 

Flattening those into a single dimension that has two auxilary coordinates, as you do, seems natural also, and solves the problem when there are a different number of forecast/valid times for each reftime. But if we had ragged arrays, I would prefer those.

Im wondering which form really fits your use case from your POV: (ensemble, reftime, time, level, latitude, longitude), or (ensemble, i, level, latitude, longitude) ?

John