This dataset contains the Sentinel-2 Level 3 Quarterly Cloudless Mosaics from CDSE rehosted by Taylor Geospatial.
This is a mirror of the Copernicus Sentinel-2 Global Mosaics (product S2MSI_L3__MCQ).
The European Space Agency produces the mosaics and the Copernicus Data Space Ecosystem (CDSE) publishes them.
Upstream access needs a CDSE account, while this copy is open over plain HTTPS or S3 with no key, account or request quota.
Taylor Geospatial Engine hosts it on Source Cooperative and adds a STAC catalog and per-quarter manifests.
The data is one cloud-free mosaic per quarter at 10 m, for 36 quarters from 2017 Q1 through 2025 Q4. That is four bands, 1,017,575 mosaic tiles and 4,070,300 Cloud-Optimized GeoTIFFs, 607.5 TB in all.
Each mosaic is the first quartile of the cloud-free Sentinel-2 Level-2A surface reflectance observations in the three-month window, taken per pixel and band. A pixel with no valid observation is nodata. The upstream mosaic algorithm note has the details.
Values are int16 scaled by 10000, so multiply by 0.0001 to get surface reflectance.
The scale factor is not stored in the files.
Nodata is -32768.
Each tile is 10008 × 10008 pixels, about 100 km square, in the UTM zone of its MGRS grid cell. Every file is a COG with DEFLATE compression and five overview levels. The internal tile size is 256 × 256 or 1024 × 1024, depending on the quarter.
All 36 quarters from 2017 Q1 through 2025 Q4 are complete, and the totals above are summed from the quarterly manifests. A quarter holds between 25,172 tiles (2017 Q3) and 30,668 (2019 Q1), because the mosaic footprint grew over the record.
The bucket also holds 2026/Q1/ and 2026/Q2/, which are not part of the catalog yet.
Their manifests list 30,360 tiles (121,440 objects, 19.10 TB) for Q1 and 28,893 tiles (115,572 objects, 17.25 TB) for Q2.
A glob over manifest/manifest_*.parquet matches these two quarters as well, so filter on year <= 2025 to get the cataloged quarters.
The manifest/ prefix holds one Parquet per quarter, manifest_{year}_{Qn}.parquet, listing every transferred object with its source key, destination key and size.
Keys follow a fixed pattern, so tiles can be read without a lookup:
{year}/{Qn}/{MGRS}_{i}_{j}/{band}.tif
For example, the red band of tile 31UFU_0_0 for 2024 Q2 is:
https://data.source.coop/tge-labs/sentinel-2-quarterly-cloudless-mosaics/2024/Q2/31UFU_0_0/B04.tif
The same objects are readable over S3 without credentials at s3://tge-labs/sentinel-2-quarterly-cloudless-mosaics/.
Use the endpoint https://data.source.coop with path-style addressing.
To read a window from one tile without downloading it:
AGENTS.md has more queries, including DuckDB examples over the manifests.
The imagery is covered by the Legal Notice on the Use of Copernicus Sentinel Data and Service Information, which allows free, full and open use, including commercial use and redistribution.
There is no SPDX identifier for those terms, so the collections declare license: "other" with a rel: license link to the notice.
Attribute it as "Copernicus Sentinel data 2017–2025". Derived products should say they contain modified Copernicus Sentinel data, so they are not read as ESA products. The mirroring, catalog metadata and coverage statistics are Taylor Geospatial Engine's work and add no further restriction.
The European Space Agency produces the Sentinel-2 Global Mosaics, Sinergise Solutions runs the mosaic production, and CloudFerro operates the CDSE infrastructure this mirror copied from.
The upstream collection is sentinel-2-global-mosaics in the CDSE STAC API, and each collection here has a canonical link to it.
The four bands are byte-identical to the upstream files.
The copy does not include the upstream observations.tif band, the userdata.json metadata or the zipped Product asset, so each tile has four assets here and seven upstream.
Item ids are unchanged, so a tile can be matched to its upstream record by id.
Upstream reprocesses old quarters, so a finished quarter here is a snapshot of the product at the time it was copied.
mosaics holds the imagery, with bands, tile geometry and per-quarter partitions (README, agent guide).coverage holds tile footprints and per-tile, per-quarter statistics (README, agent guide).