Edzer Pebesma
> in [m] for projected data, and possibly in [km] for unprojected? Yes, that is right, and I agree it is confusing. Today I would do this differently; suporting `units`...
From the top of my head the distances are evaluated from the point to the centroid of the target polygon, not as the distance to the polygon (boundary). Would that...
> there is an existing simple implementation by @edzer https://github.com/ngort01/fuzzyMM not by me, but by Nikolai Gorte (at that time a student I supervised)
In case of a four-dimensional `stars` object, it would make sense to associate dimension 3 with facet rows, and dimension 4 with faced columns, or vice versa. I think that...
> I use aggregate from terra which is very convenient and quick. Have you tried on a 10000 x 10000 image, or larger, and compared to down sampling when the...
> Exporting st_downsample - yes, that makes sense. still need to think this through; the more common pattern is to use `st_as_stars()` on a `stars_proxy` object and use the `downscale`...
@rhijmans it seems that e.g. ```r spatSample(landsat_terra, 1e4, as.raster=TRUE) ``` doesn't return a very useful raster; when plotted: 
Some work on converting bands to rgb, including cutoffs and stretching, was done in `stars::st_rgb()` - maybe some of it overlaps this.
Sounds good!