Sentinel-2 L2A scenes (item index)

Every Sentinel-2 L2A scene that AWS Earth Search publishes, as one year-partitioned GeoParquet table you can query in place. A row is one scene. It states the footprint, the acquisition time, the MGRS tile, the cloud cover, the scene-classification percentages, and the complete upstream STAC assets object.

For most work, prefer sentinel-2-c1-l2a, ESA's uniform reprocessing of the archive. This collection mirrors Earth Search's original index, which reaches back only to November 2016 and repeats many early scenes.

This catalog publishes no imagery. The Cloud-Optimized GeoTIFFs remain in the public sentinel-cogs bucket on AWS, and the table already reproduces each of their URLs.

Query it

Every asset href is in the assets JSON-string column — no URL templates, no API. Swap $.visual.href for $.red.href, $.scl.href or any other key; the agent guide lists them all.

That query is cheap for three reasons together. A tile id and a year identify the part, so 31UFU is zone 31 and 2021 is z21-31.parquet, and the Layout section below gives every range. Rows inside each part sort by month first, as (_month, _hilbert) through 2025 and (_month, s2:mgrs_tile, _hilbert) from 2026, so a month filter and a spatial filter both prune row groups. The answer comes entirely from HTTP range requests against the Parquet files, with no API in front to queue behind.

This collection sets partition:glob to year=*/*.parquet, a Hive layout. Setting hive_partitioning = true exposes year as a column the files themselves omit, and a filter on it skips whole files. DuckDB expands the glob when it can list the store, which the anonymous s3:// door allows. Over plain https:// it reports "Globs (*) for generic HTTP file are not supported", so name the part as above. A year-pruned count through the glob:

That opens two parts out of sixty and reads their footers. A scan of every part takes minutes rather than seconds, and this collection's table:row_count and temporal extent give the whole-archive answer without one.

Layout

One directory per year Earth Search actually holds items for, so the listing starts where its record does — see Coverage below.

A year's archive is one file through 2018, and from 2019 it is split by the UTM zone of s2:mgrs_tile (the leading digits of the tile id). 2019 and 2020 are four files: z01-20.parquet holds zones 1–20, z21-35.parquet zones 21–35, z36-46.parquet zones 36–46, z47-60.parquet zones 47–60. From 2021 a year is eight, nested inside those four: z01-15.parquet (zones 1–15), z16-20.parquet (16–20), z21-31.parquet (21–31), z32-35.parquet (32–35), z36-40.parquet (36–40), z41-46.parquet (41–46), z47-52.parquet (47–52), z53-60.parquet (53–60). The split is what keeps an 8-million-scene year buildable, and it is also a spatial index for free: a tile id and a year name the part, so a query for 31UFU in 2021 can open year=2021/z21-31.parquet alone and skip the other seven eighths of the year. The current year also carries live.parquet, rebuilt daily from the Earth Search API and folded into the archive parts once a month. The daily rebuild drops every id the year's archive parts already hold, so no two parts of a year overlap. A glob over year=*/*.parquet therefore reads each scene exactly once and includes yesterday, whichever shape the year has. A client can always dedupe on id anyway; it is safe and it removes nothing.

Each year=YYYY/YYYY.json item states that year's measured row count, time range, footprint bounds and platforms, so a client can choose a year without opening a byte of Parquet. Each data asset in the item states the same for the one part it names.

Coverage

The record starts in November 2016, when Earth Search produced its first L2A Cloud-Optimized GeoTIFFs: 2015 and most of 2016 have no COG products, and 2017-2018 are partial. That is what Earth Search serves. This mirror adds no rows and drops none, so the gap is upstream, not here.

sat:orbit_state and s2:granule_id are NULL on newer items, because Earth Search stopped populating them. Read the agent guide before you write a query that depends on either.

Provenance and license

Items come from the Earth Search STAC API, collection sentinel-2-l2a, which Element 84 runs over the Sentinel-2 L2A COGs on the AWS Registry of Open Data. No step here filters, reclassifies or interpolates the upstream record. The columns added are two sort helpers, _month and _hilbert, documented in the agent guide.

Contains modified Copernicus Sentinel data. The Copernicus Sentinel Data Terms and Conditions grant free, full and open access for any use, and this collection states them by their SPDX identifier, CC-BY-SA-3.0-IGO. The citation for the upstream COGs, from their AWS Registry of Open Data entry:

Sentinel-2 Cloud-Optimized GeoTIFFs, accessed on [DATE] from https://registry.opendata.aws/sentinel-2-l2a-cogs.