Inconsistent Sentinel-2 L2A Raster Footprint Polygons

[scottyhq]

It seems like some STAC Items for Sentinel-2 L2A Use the MGRS Square as the STAC geometry and others use a valid data footprint. Maybe the metadata generation changed at some point in time?

https://planetarycomputer.microsoft.com/api/data/v1/item/map?collection=sentinel-2-l2a&item=S2A_MSIL2A_20151228T050222_R119_T45RVM_20210526T203938

https://planetarycomputer.microsoft.com/api/data/v1/item/map?collection=sentinel-2-l2a&item=S2B_MSIL2A_20230505T045659_R119_T45RVM_20230505T104315

The stactools library suggests using a valid data footprint https://stactools.readthedocs.io/en/stable/api.html#module-stactools.core.utils.raster_footprint

[ZirconCode]

I've also run into this, but I've found correct polygons before and after a broken one, so haven't been able to pin it down.

[TomAugspurger]

I think that the version of https://github.com/stactools-packages/sentinel2 just gets the geometry from the metadata provided by ESA. It'd be good to confirm that first.

We do have the option to generate geometries from the actual rasters themselves, using the raster_footprint utility. That said, I am a bit worried about the computation cost of doing that for a dataset of sentinel-2's size.

[cmosig]

Is there any update on this?

I just posted this but then found this repo later.

[TomAugspurger]

No updates from me. Might be worth reaching out to ESA to report the issue, in case they're able to fix it at the source.

[cmosig]

@TomAugspurger happy to reach out to ESA. What would be their best channel to report the issue? Do you know by chance?