aggregate_spatial: use 'id' property of input geojson to name output features

[jdries]

when writing netcdf output for aggregate_spatial, we now name the features like['feature_0','feature_1',....]
Our users would like us to use the feature id of the input geojson to name these features, to more easily link back a timeseries to the input.

This relates to support for vector cubes/feature collections, which also requires us to preserve feature properties throughout the processing.

[jdries]

Loading geometry through specific process should trigger usage of new-style vector cube, which retains these features id's.

[soxofaan]

FYI: see EP-3981 and Open-EO/openeo-python-driver#114 for my initial implementation of VectorCube support (geopandas based)

basic test that illustrates workflow: https://github.com/Open-EO/openeo-python-driver/blob/master/tests/test_views_execute.py#L754-L810

• load new "VectorCube" from geojson with (experimental) load_uploaded_files process
• use vector cube as geometry in aggregate_spatial
• export result as geojson again (note preservation of original geojson properties and addition of aggregation values as new properties)

[bossie]

Test case:

polygon_1 = Polygon([(10.4566, 51.3747), (10.4335, 51.3732), (10.4527, 51.3615), (10.4566, 51.3747)])
polygon_2 = Polygon([(10.4566, 51.3747), (10.4334, 51.3732), (10.4528, 51.3614), (10.4566, 51.3747)])

def as_feature(geometry, id) -> dict:
    return {
        'type': 'Feature',
        'id': id,
        'properties': {},
        'geometry': mapping(geometry)
    }

feature_collection = {
    'type': 'FeatureCollection',
    'properties': {},
    'features': [
        as_feature(polygon_1, id="apples"),
        as_feature(polygon_2, id="oranges")
    ]
}

im = (conn
      .load_collection("SENTINEL2_L2A",
                       bands=["B04", "B03", "B02"],
                       spatial_extent={"west": 10.4005, "south": 51.3371, "east": 10.5152, "north": 51.3856},
                       temporal_extent=["2021-07-08T00:00:00Z", "2021-07-08T00:00:00Z"])
      .aggregate_spatial(feature_collection, "mean"))

im.download("/tmp/test_aggregate_spatial_feature_ids.nc")

bossie@rastapopoulos:~$ ncdump /tmp/test_aggregate_spatial_feature_ids.nc | grep 'feature_names ='
 feature_names = "feature_0", "feature_1" ;

[bossie]

Example client code:

feature_collection = conn.datacube_from_process("load_uploaded_files",
                                                paths=["/data/users/Public/vdboschj/FeatureCollection.geojson"],
                                                format="GeoJSON")

im = (conn
      .load_collection("SENTINEL2_L2A",
                       bands=["B04", "B03", "B02"],
                       spatial_extent={"west": 10.4005, "south": 51.3371, "east": 10.5152, "north": 51.3856},
                       temporal_extent=["2021-07-08T00:00:00Z", "2021-07-08T00:00:00Z"])
      .aggregate_spatial(feature_collection, "mean"))

im.download("means.nc")

@soxofaan : is there a more elegant way to write this?

[bossie]

Alternative:

from openeo.processes import load_uploaded_files

feature_collection = load_uploaded_files(paths=["/data/users/Public/vdboschj/FeatureCollection.geojson"],
                                         format="GeoJSON")

# ...

[bossie]

The input GeoJSON file has to be accessible from the OpenEO back-end and its Features should carry an "id", either:

• as a child of the Feature (like below) or;
• as part of its "properties": "properties": {"id": "apples"}

{
  "type": "FeatureCollection",
  "properties": {},
  "features": [
    {
      "type": "Feature",
      "id": "apples",
      "properties": {},
      "geometry": {
        "type": "Polygon",
        "coordinates": [
          [
            [
              10.4566,
              51.3747
            ],
            [
              10.4335,
              51.3732
            ],
            [
              10.4527,
              51.3615
            ],
            [
              10.4566,
              51.3747
            ]
          ]
        ]
      }
    },
    {
      "type": "Feature",
      "id": "oranges",
      "properties": {},
      "geometry": {
        "type": "Polygon",
        "coordinates": [
          [
            [
              10.4566,
              51.3747
            ],
            [
              10.4334,
              51.3732
            ],
            [
              10.4528,
              51.3614
            ],
            [
              10.4566,
              51.3747
            ]
          ]
        ]
      }
    }
  ]
}

[bossie]

@lru_cache load_collection is now able to cope with DriverVectorCube being passed as an argument and part of the cache key. Added some tests to make sure that this caching actually works (I don't think it did).