Develop best practices for handling non-spatiotemporal dimensions

[chris-little]

This issue is to discuss how to support custom extra dimensions, such as "wavelength". The custom dimensions may not even be ordered or implicilty continuous, like (x, y, z, t). A collection of ensembles of forecast data is one example of a discrete and unordered dimension. (but see Issue #172 for support of ensembles specifically)

[jonblower]

Just to bring the key info over from the previous discussion - CovJSON already supports arbitrary numbers of dimensions. The use of domainTypes can restrict this to reduce the multiplicity of options that clients have to deal with. So there isn't a problem with the schema (as far as I know).

So I think this is more a case of working out what is "best practice" in different situations. Is it best to define new domainTypes? Or model the dimensions as variables (e.g. in a multispectral image, the different wavelengths could be modelled as dimensions or variables).

I'd suggest renaming this issue "Develop best practices for handling non-spatiotemporal dimensions" or something like that.