Discuss STAC extension dependencies in load_stac for assets with time dimension

[VincentVerelst]

Currently the load_stac process requires both the item_assets and datacube STAC extensions in order to be able to detect whether or not an asset has a time dimension.

When these extensions are not present in the STAC collection, the job will still run successfully, but the end result will not have a time dimension anymore.

Some questions:

• Why do we rely on two different extensions? Would for example only relying on item_assets not be enough?
• Could it be an option to throw a warning if one or two of these STAC extensions are not detected in the collection metadata?
• As of STAC 1.1 item_assets will be deprecated (https://github.com/stac-extensions/item-assets) and become part of the core spec. I guess in this case we would not need to rely on optional STAC extensions anymore to detect a time dimension in the assets, correct?

[jdries]

The item_assets 'extension' is now indeed part of the core spec, and I expect that we will use it more and more in general.
It will also allow us to detect that the media type is netcdf, which is an important clue for us to select the right data loading code path.

Then the only question remaining is how to detect that we are loading a collection of netcdf samples, where each netcdf covers only a limited area. Until now, the datacube extension is the only one I know that does this, and is already commonly used in the openEO ecosystem. It's not impossible that alternatives exist, but quite likely that it would still require an extension somehow. Swapping out one extension for the other will of course not solve things.

To throw a warning, we could try and inspect the netcdf metadata when we read data from it. If metadata indicates a specific dimensionality which is unexpected, we can perhaps throw an error or log a warning.