Six legal regimes hide behind "open": at 10 m global embeddings, CC0 TESSERA and CC BY AlphaEarth are the two to know
Asked:
“What are the major open satellite imagery and geospatial embedding datasets, their global coverage, resolution, and licensing terms?”
A practitioner shortlist of 10 datasets — 6 imagery or corpus products and 4 precomputed embedding rasters — consolidated from 18 screened source rows spanning 12 independent hosts in a broad live-web scan. “Major” has no formal universe: this is a curated comparison, not an exhaustive registry. Resolutions run from NAIP's 1 m aerial imagery to Clay's 2.5 km embedding patches.
Resolution versus reach, coloured by what the licence really lets you do
Each bar spans a dataset's published pixel sizes (log scale, finer at left). Top band: raw and harmonized imagery. Bottom band: precomputed embedding rasters. Hover or tap a mark for coverage, cadence, dimensions and exact terms.
Public domain / CC0 — use freely, even commercially
Free, full and open data terms
CC BY 4.0 — attribution mandatory
Custom non-commercial — legal review
Terms not clearly stated
AlphaEarth and TESSERA are the only globally dense 10 m embedding layers here, but their terms differ sharply: AlphaEarth embeddings require attribution under CC BY 4.0 and its model weights stay proprietary, while TESSERA ships embeddings and weights under CC0 with MIT software —
arxiv.org,
github.comLandsat's 50-plus-year, 30 m global archive is US public-domain data — yet open copyright is not free compute: users still pay their own AWS processing and egress costs —
ntrs.nasa.govThe sharpest pixels are the least global: NAIP's 1 m imagery covers only the United States, and NICFI's 4.77 m basemaps cover tropical forests between 30°N and 30°S under a custom non-commercial licence —
fsa.usda.gov,
assets.planet.comMajor TOM is an open near-global Sentinel-2 source corpus (~225 M km²) for training, not a precomputed embedding raster; ESD offers a 25-year 30 m embedding record whose formal licence is not stated —
arxiv.org,
arxiv.org
The shortlist, row by row
| Dataset | Coverage | Resolution | Temporal | Dims | Licence | Source |
|---|
Shortlist of 10 consolidated datasets from 18 screened source rows across 12 independent hosts (no single URL backs more than 5.6% of rows), gathered in a live-web scan; duplicate Sentinel-2 and Landsat product rows and mixed overview rows were folded into their parent datasets for presentation, with their catalog pages kept as corroborating links. Resolution is ground sample distance in metres; coverage and licence text follow each source's own wording. Some sources are primary publishers, others corroborating catalogs or papers; custom-licence datasets merit legal review. Access cost (such as cloud egress) is separate from copyright permissions.