Thanks for sharing @martinfleis. I see that “long table form” is mentioned at Vector Data Cubes, which is how I would have normally thought to model this data. It’ll be interesting to see which form is most convenient for which operations (and hopefully the various DataFrame.to_xarray and Dataset.to_pandas methods can efficiently move between these forms, but I’m not quite sure which operations can be zero-copy of data).
And just to be completely clear, since it confused me at first, this is different from Tables, (x)arrays, and rasters¶, where rasters might be stored as columns in a tabular structure. IIUC, the unique thing about vector data cubes is having a coordinate dimension that’s vector labels? Again, I think all these representations might be valuable depending on the use case.