simplify specifying collection properties

[jdries]

Currently, in load_collection, property filtering only seems to work with callbacks:

    s2 = connection.load_collection(s["SENTINEL2"],                                 
                                temporal_extent=temp_ext,
                                spatial_extent=spat_ext,
                                bands=["B03","B04","B05","B06","B07","B08","B11","B12","SCL"],
                                properties={"cloudcover":lambda x: x.eq(90)}

This increase the learning curve somewhat, and a lot of property filters are based on equality. So something like:

    properties={"cloudcover":90}

could be translated into a correct callback by the client?

[m-mohr]

What would properties={"cloudcover":90, "platform": "S1a"} translate to? Are they connected with "and" or "or"?

Long-term it might make sense to adopt something like CQL2 text (see OGC APIs) as a simplified version instead of objects, which are more human-readable and flexible (e.g. cloudcover < 90 AND platform = 's1a'). Assuming that a CQL2 text parser will be implemented soon.

PS: cloud cover equality is probably not the best example...

[jdries]

yes, multiple properties would imply an AND condition in my interpretation
cloudcover is indeed a special case: most search catalogs somehow don't seem to have a problem with specifying this as an equals and interpreting it as smaller than, but we may need to check the details on that!

[soxofaan]

or introduce a minimal (SQL-like) query language

properties="cloudcover<90 and platform=s1a"

that is converted by python client appropriately

[m-mohr]

CQL2 text would be exactly that, which then you can also re-use for STAC API queries etc. And we could hope that we get a Python library from OGC API implementors so that we don't need to implement it ourself.

[soxofaan]

related API issue: Open-EO/openeo-api#396 (what properties can be filtered on?)

[soxofaan]

or introduce a minimal (SQL-like) query language

such a mini language could even be useful in other places, e.g.:

cube.reduce_dimension("bands", reducer=" (B08 - B05) / (B08 - B05)")

mask = cube.apply("x == 4")

that's actually a generalization of what's under construction under #255

[soxofaan]

In #331 I experimented with this approach:

from openeo import collection_property

cube = con100.load_collection(
    "S2",
    properties=[
        collection_property("eo:cloud_cover") <= 75,
        collection_property("platform") == "Sentinel-2B",
    ]
)