Multiple CRS across assets in Sentinel-2-L2A

[BnJam]

Is having multiple coordinate reference systems across different band Assets in a single granule and date expected?

# poly
POLYGON ((-137.8227033237725 59.43881852787066, -137.85202575601502 59.52747473469491, -138.02588212273236 59.51238228891469, -137.99614605332758 59.4237653545415, -137.8227033237725 59.43881852787066))

tgt_bands =  ['B02', 'B03', 'B04']
date_range = f'2020-04-01/2020-08-30'

this_items = api.search(
            collections = ['sentinel-2-l2a'],
            bbox=poly['geometry'].bounds,
            query={"eo:cloud_cover": {"lt": 70}},
            datetime = date_range,
        ).get_all_items()
signed_items = [planetary_computer.sign(item).to_dict() for item in this_items]

data = (
    stackstac.stack(
        signed_items,
        assets=tgt_bands + ['SCL'],
        chunksize=4096, # Set chunksize
        bounds_latlon=poly['geometry'].bounds, # clip to AOI bounds
    ).where(lambda x: x > 0, other=np.nan)
)

ValueError: Cannot pick a common CRS, since assets have multiple CRSs: asset 'B02' of item 2 'S2B_MSIL2A_20200402T204019_R014_T07VFG_20200924T184821' is in EPSG:32607, but assets before it were in EPSG:32608.

Please specify a CRS with the `epsg=` argument.

[TomAugspurger]

I would need to double check the source, but I think that the CRS of an item is tied to the s2:mgrs_tile. These differ across the items you've found in your search:

In [1]: import shapely, pystac_client, stackstac, planetary_computer

In [2]: poly = shapely.from_wkt("POLYGON ((-137.8227033237725 59.43881852787066, -137.85202575601502 59.52747473469491, -138.02588212273236 59.51238228891469, -137.99614605332758 59.4237653545415, -137.8227033237725 59.43881852787066))")

In [3]: api = pystac_client.Client.open("https://planetarycomputer.microsoft.com/api/stac/v1")

In [4]: date_range = f'2020-04-01/2020-08-30'
   ...:
   ...: this_items = api.search(
   ...:             collections = ['sentinel-2-l2a'],
   ...:             bbox=poly.bounds,
   ...:             query={"eo:cloud_cover": {"lt": 70}},
   ...:             datetime = date_range,
   ...:         ).get_all_items()

In [5]: set(x.properties["s2:mgrs_tile"] for x in this_items)
Out[5]: {'07VFF', '07VFG', '08VLL', '08VLM'}

In your post, you say

Is having multiple coordinate reference systems across different band Assets in a single granule and date expected?

That would be unexpected, but I don't think it's happening here. The message from stackstac is saying that some of the assets have a different CRS from some other assets (across all items, not within a single item).