return type of `cell_centers`

[keewis]

Right now, Dataset.dggs.cell_centers returns a Dataset containing the geographical coordinates. However, since we return a Dataset we have to use the lengthy ds.pipe(lambda ds: ds.merge(ds.dggs.cell_centers())) to assign the result to the original object (and this won't even work for DataArray).

Another option would be ds.pipe(lambda ds: ds.assign_coords(ds.dggs.cell_centers().coords)), which at least works for DataArray objects, but otherwise is slightly longer than the code with merge.

Ideally, we'd allow something like this:

ds.assign_coords(lambda ds: ds.dggs.cell_centers())

For this to work, ds.dggs.cell_centers() would have to return a Coordinates object (or a dict, or we make the geographic coordinates data variables), and assign_coords would have to accept a Callable[[Dataset | DataArray], dict | Coordinates].

What do you think, @benbovy?

[benbovy]

My 2 cents:

I think it would be nice if we can have a consistent API for all cell geometry extraction methods, i.e., something like this:

# return cell centers as (shapely) point geometries
ds.dggs.cell_centers() -> xr.DataArray

# return cell centers as lat/lon coordinates 
ds.dggs.cell_centers_latlon() -> tuple[xr.DataArray, xr.DataArray]

# return cell boundaries as (shapely) polygon geometries
ds.dggs.cell_boundaries() -> xr.DataArray

• always use DataArray as the xarray object return type (maybe within tuple or dict)
• cell_xxx() would always return shapely geometries
• cell_centers_latlon() as an additional but common special case
  • _latlon is also used in other xdggs method names
  • a tuple of DataArrays seems the simplest and the most intuitive here to me. This is what I would expect from most Python scientific libraries. A dictionary repeats the information that we already have in the DataArrays (i.e., their names) and we may not want to be too opinionated about the names of the lat-lon coordinates.

Personally I wouldn't mind doing an extra step to assign lat-lon coordinates:

cell_lat, cell_lon = ds.dggs.cell_centers_latlon()
ds.assign_coords(latitude=cell_lat, longitude=cell_lon)

If we want a convenient, functional-style friendly way to achieve this, maybe assign_coords could also accept Iterable[DataArray]?

# reuse the names of the two DataArrays returned by ds.dggs.cell_centers_latlon()
ds.assign_coords(ds.dggs.cell_centers_latlon())

Actually, if we can assign coordinates like above the dggs.assign_latlon_coords() method may not add a lot of value so I'd be +1 to remove it.