load_stac + resample_spatial incompatibility

[soxofaan]

minimal reproduction example:

import openeo
connection = openeo.connect(url="openeo.dataspace.copernicus.eu")

cube = connection.load_stac("https://stac.openeo.vito.be/collections/wenr_features")
cube.resample_spatial(projection=3035, resolution=10.0)

gives

MetadataException: Expected two spatial dimensions but found spatial_indices=[]

Current load_stac implementation only detects:

>>> cube.metadata.dimension_names()
['bands', 't']

and resample_spatial expects spatial dimensions to work on

[soxofaan]

Solution would be to make sure that DataCube.load_stac adds the spatial dimensions to the client side metadata when necessary (probably just always, or when the raster/eo extensions are present)

[VictorVerhaert]

@soxofaan I think the python clinent should be as flexible as possible with incoming stac collections, and always result in datacube with spatial, temporal and band dimensions unless the stac item contains a datacube extension that states otherwise.
If a stac collection cannot be reached the results should be some default (empty) metadata as well.

@VincentVerelst this might be relevant for the datacube dimensions question you had

[soxofaan]

quickfixed this with 05b527f:
indeed always assume presence of spatial dimensions, which should resolve the problem for common use cases

[soxofaan]

Reopening to align spatial dimension detection with the approach for temporal dim:

openeo-python-client/openeo/metadata.py

Lines 750 to 764 in 05b527f

	def get_temporal_dimension(self, stac_obj: pystac.STACObject) -> Union[TemporalDimension, None]:
	"""
	Extract the temporal dimension from a STAC Collection/Item (if any)
	"""
	# TODO: also extract temporal dimension from assets?
	if _PYSTAC_1_9_EXTENSION_INTERFACE:
	if stac_obj.ext.has("cube") and hasattr(stac_obj.ext, "cube"):
	temporal_dims = [
	(n, d.extent or [None, None])
	for (n, d) in stac_obj.ext.cube.dimensions.items()
	if d.dim_type == pystac.extensions.datacube.DimensionType.TEMPORAL
	]
	if len(temporal_dims) == 1:
	name, extent = temporal_dims[0]
	return TemporalDimension(name=name, extent=extent)

[soxofaan]

started PR #739 to leverage pystac for better spatial dimension detection

[soxofaan]

I'm going to close this ticket, as the remaining work is now actually topic of

• Simpler or more predictable dimension detection with load_stac? #743