Suggestion by Freek:
We currently have functions to read an STM from Zarr and CSV. It would also be easier to have function (from_dict(), for example) to initiate an STM directly in memory, make assumtions on dimensions (space, time) and pre-fill some parts.
This can be a wrapper creating an xr.Dataset.