[CF-metadata] CF and a representation of probalistic forecasts

[Jonathan Gregory]

Dear Jamie and Karl

I agree with Karl that it is quite natural to regard realization as a
dimension. The statistical processing of realizations is something which
cell_methods could record. Hence I agree with both of you that we could regard
the realization_weight as like cell_measures. On the other hand, since it
isn't related to the grid or computed in any way from coordinates, but depends
more on data variables, it isn't really a cell_measure, I think. The
realization weights are probably closer to being information about the values
of the data variables, which is the purpose of the ancillary variables.

Best wishes

Jonathan