filter_spatial: clarify behaviour

[clausmichele]

The filter_spatial process definition https://processes.openeo.org/#filter_spatial is not clear, since it mentions keeping only the pixels intersected by the geometry and later, under the geometries parameter, mentions that no masking is applied.

We need to decide what's the expected output.
Is it only a filter_bbox process that automatically extracts the AOI from the bounds of the geometries?
Is it the same as mask_polygon?

@jdries said that on their side masking is applied, code here: Open-EO/openeo-geopyspark-driver@4e3ef04

We are waiting for a clarification to proceed with this PR: Open-EO/openeo-processes-dask#170

Tagging some other people for opinions:
@ValentinaHutter @dthiex @soxofaan @SerRichard

[m-mohr]

I just looked at the git history and it's relatively clear that masking should be applied. The confusing part was added with the datacube changes, i.e. only in 2.0.0 rc1 and thus we should clean that up again to align with the 1.x defintion.

[m-mohr]

PR is here: #470

[soxofaan]

Indeed I found this discussion where it is clear that intent is to have masking included:
#131 (comment)

[clausmichele]

So the main difference with mask_polygon is that filter_spatial may reduce the spatial dimensions and mask_polygon not?

@masawdah At the implementation side I think that filter_spatial becomes filter_bbox + mask_polygon internally.

[soxofaan]

So the main difference with mask_polygon is that filter_spatial may reduce the spatial dimensions extent and mask_polygon not?

Indeed.
And note that in the VITO backend implementation we even infer the spatial extent from mask_polygon if no other spatial extent filtering is present in the process graph.

[m-mohr]

Yes, and that mask_polygon has additional options, of course.