Dear Jonathan, ----- Original Message ----- > > Thus, although I am unsure we should impose it from the convention, > > e.g. by allowing to define some attributes from the "umbrella" level > > (such as :coordinates), I see little use for defining vector objects > > that do not share all their dimensions. > > In your umbrella proposal, all the data variables still describe > themselves > independently, including their grid, and I see that as an advantage. > Actually I do think there is a point in permitting one umbrella to shelter > components > which are on different grids. The Arakawa C-grid is a real use-case. > Many models use this arrangement, and CF currently offers no way of > grouping the > components, although it does of course contain them independently. I > think it could be convenient to some software to have an easy way to find the > components of a vector on an Arakawa C-grid. It is not essential, but it > could be > useful, just as when the components are at the same gridpoints. However, I > don't feel strongly about it.
The umbrella proposal indeed let room for a lot of flexibility, since all the data variables (components) still describe themselves. The example of the Arakawa C-grid might indeed be a legal vector entity. However, I am still looking for a way to impose some level of conformance between the components. At the current stage, it would be valid to define an umbrella vector variable from two component variables, one being dX(time,xc,yc) and dY(time,height,xc,yc). What would be the meaning of such a construct? And if invalid, how can we enforce that the component variables have to somehow share the same dimensions? Would a sentence in the convention be enough to avoid these constructs? Regards, Thomas _______________________________________________ CF-metadata mailing list [email protected] http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
