Sentinel-2 Level-2A Bottom of Atmosphere Imagery

Hello 👋! ESA has globally rolled out bottom of atmosphere “Level-2A” (aka “L2A”) corrected imagery as of December 2018. Are there any known technical limitations for supporting this imagery? If not, can we extend this project to support L2A?

At first glance, it looks like L1C and L2A datasets adhere to a similar (although not identical) S3 schema:

$ aws s3api list-objects --bucket sentinel-s2-l1c --prefix tiles/1/C/EM/2019/11/20/0 --request-payer requester
{
    "Contents": [
        {
            "Key": "tiles/1/C/EM/2019/11/20/0/B01.jp2",
            "LastModified": "2019-11-20T19:55:25.000Z",
            "ETag": "\"96effcc32e2c0f0d6cc45bf18efbb8ce\"",
            "Size": 297678,
            "StorageClass": "INTELLIGENT_TIERING",
            "Owner": {
                "ID": "91d380b3cead28df927c824731b0173701336cd8d67b0679d7166288f3850f38"
            }
        },
        ...

$ aws s3api list-objects --bucket sentinel-s2-l2a --prefix tiles/1/C/EM/2019/11/20/0 --request-payer requester
{
    "Contents": [
        {
            "Key": "tiles/1/C/EM/2019/11/20/0/R60m/B01.jp2",
            "LastModified": "2019-11-20T21:10:52.000Z",
            "ETag": "\"620a1f03d96f1df8bdfdd32b64c90608\"",
            "Size": 314419,
            "StorageClass": "INTELLIGENT_TIERING",
            "Owner": {
                "ID": "91d380b3cead28df927c824731b0173701336cd8d67b0679d7166288f3850f38"
            }
        },
        ...

Notice the addition of the “R60m” to the prefix of the L2A dataset. This seems to indicate the ground sample distance (GSD) of the dataset. Bands with finer resolution have a variety of GSD available. For example, B02 has R10m/B02.jp2, R20m/B02.jp2, andR60m/B02.jp2 variants.

It looks like rio-tiler hard-codes top of atmosphere (“L1C”) support: https://github.com/cogeotiff/rio-tiler/blob/6b3b57e4d837cfa855d9b4a74175812339ac433e/rio_tiler/sentinel2.py#L19 I would imagine L2A support being added behind an optional parameter to bounds / metadata / tiles functions?

[vincentsarago]

just pushed https://github.com/cogeotiff/rio-tiler/pull/143

This PR introduce the new Sentinel 2 sceneid (and still support the old one for now).

The old sceneid fallback to level l1c.

@DanSchoppe would you be able to review the PR ?

[matthewhanson]

I also don’t understand why they have multiple resolutions - JPEG2K theoretically allows decompression at any arbitrary resolution…but in practice the available decoders just aren’t very efficient so maybe that’s why they have multiple files.

At any rate, I am currently working on the COG version of the L2A data for Africa, it will be in us-west-2 and will be indexed in a STAC API. The first data should start appearing in the bucket in February.