DATA ARK BACKLOG
Datasets for the next chapter.
Explore datasets being considered for future archiving. Review their sources, licenses, storage needs and the regional work already underway.
Explore:
embeddings
forests
land cover
Dataset backlog
5 datasets
Satellite · Global
Google Satellite Embedding V1
Annual geospatial representations for mapping, classification and change detection. Each 10-metre pixel contains a 64-dimensional vector summarising surface conditions across a calendar year.
satellite
embeddings
AlphaEarth
Regional work in progress
- Publisher
- Google and Google DeepMind
- License
- CC-BY-4.0
- Original hosting
- Google Earth Engine and the public alphaearth_foundations bucket on Google Cloud Storage
- Coverage
- Global terrestrial surfaces and shallow waters; annual products
- Formats
- Cloud Optimized GeoTIFF · GeoParquet index
- Planned scope
- Candidate for broader geographic and annual coverage. The full global collection is not archived here.
- Existing work
- Separate 2025 selections for Malawi and mainland Norway are being transferred to Hetzner and AWS. These regional selections do not represent the complete global dataset.
- Storage
- Full global collection: not measured. Selected 2025 original files: Malawi 99.00 GB and mainland Norway 227.93 GB per copy, including complete border-crossing tiles.
- Attribution
- The AlphaEarth Foundations Satellite Embedding dataset is produced by Google and Google DeepMind.
Satellite · Global
High Resolution Canopy Height Maps
Global and regional estimates of tree canopy height, generated with machine learning from high-resolution Maxar satellite imagery. The distribution includes height rasters and observation metadata.
forests
canopy height
Meta
Under consideration
- Publisher
- Meta and World Resources Institute (WRI)
- License
- CC-BY-4.0
- Original hosting
- AWS Open Data, dataforgood-fb-data bucket, forests/v1 prefix, us-east-1
- Coverage
- Global height maps, regional products and a downsampled global variant
- Formats
- Cloud Optimized GeoTIFF · GeoJSON · Raster masks
- Planned scope
- Choose a country or regional subset before archiving. No transfer has been started.
- Existing work
- Storage inventory completed; no verified archive copies are available here yet.
- Storage
- Measured 21 October 2026: global main product 17.92 TB per copy; all forests/v1 products 27.06 TB per copy (54.12 TB for two copies). Regional and downsampled products overlap geographically.
- Attribution
- Meta and World Resources Institute (WRI), 2024. High Resolution Canopy Height Maps (CHM). Source imagery for CHM © 2016 Maxar. Include the source URL and access date when citing.
Agriculture · Global
Fields of The World (FTW) — Global
Global 10-metre estimates of agricultural field boundaries for 2024 and 2025, derived from Sentinel-2 composites with per-field confidence scores. Related products include prediction-confidence rasters, model-input composites and probability mosaics.
agriculture
field boundaries
GeoParquet
Under consideration
- Publisher
- Taylor Geospatial and collaborators
- License
- CC-BY-4.0
- Original hosting
- Source Cooperative; published by Fields of The World (Taylor Geospatial and collaborators)
- Coverage
- 195 countries and territories; 2024 and 2025 predictions in 574 administrative partitions
- Formats
- GeoParquet · PMTiles · Cloud Optimized GeoTIFF · Zarr v3
- Planned scope
- Choose which products to archive: annual GeoParquet boundaries, confidence/quality rasters, Sentinel-2 model-input composites, probability mosaics, and/or PMTiles. These products overlap in content and have different storage and use cases.
- Existing work
- Source and product structure reviewed; no transfer has been started and no verified archive copy is available here.
- Storage
- Source Cooperative reports approximately 210 GB for the GeoParquet vector collection. Full storage for all products is not measured; input composites include approximately 22,700 tiles per year plus global Zarr mosaics.
- Attribution
- Credit Taylor Geospatial and collaborators, including Microsoft AI for Good, Arizona State University, Washington University in St. Louis, Oregon State University and Clark University. Cite Robinson et al. (2026) and retain CC-BY-4.0 terms.
Population · Global
WorldPop Population Data API
REST API for discovering WorldPop datasets and their metadata, including population estimates and related health data. API entries can point to downloadable GeoTIFF source files.
population
WorldPop
REST API
Under consideration
- Publisher
- WorldPop, University of Southampton
- License
- Varies by dataset; review source metadata
- Original hosting
- WorldPop REST API at https://www.worldpop.org/rest/data
- Coverage
- Global catalog, with project-specific geographic and temporal coverage
- Formats
- JSON metadata · GeoTIFF datasets
- Planned scope
- Inventory available projects, countries, years, source files, licenses and storage needs; then select an appropriate dataset or regional subset before archiving.
- Existing work
- API documentation reviewed; no data transfer started and no verified archive copies available here.
- Storage
- Not measured. The API exposes multiple datasets and years; estimate storage only after selecting project, geography, period and resolution.
- Attribution
- Preserve the citation and attribution supplied in each dataset's API metadata. WorldPop cautions against discriminatory, exploitative or surveillance uses, especially when combining population data with other datasets.
Buildings · Global
Google–Microsoft–OSM Open Buildings
A combined collection of approximately 2.7 billion building footprints assembled by VIDA from Google Open Buildings V3, Microsoft GlobalMLFootprints and OpenStreetMap. Individual footprints retain their source attribution.
buildings
building footprints
Google
Hetzner transfer in progress
- Publisher
- VIDA; source data from Google, Microsoft and OpenStreetMap
- License
- ODbL-1.0
- Original hosting
- Source Cooperative; published by VIDA
- Coverage
- Global collection with country and S2-grid partitions; source snapshots span different dates.
- Formats
- GeoParquet 1.1.0 · FlatGeobuf · PMTiles
- Planned scope
- Full GeoParquet 1.1.0 by-country S2 distribution, preserved byte-for-byte. Phase one stores the payload in Hetzner only; the Source Cooperative original is recorded separately as trusted secondary storage. Country-wide GeoParquet, FlatGeobuf and PMTiles are overlapping alternatives and are excluded.
- Existing work
- Frozen inventory: 1,306 files across 198 country partitions, 365,667,633,580 source bytes. Hetzner payload transfer is running; no AWS payload copy is planned for phase one. Source Cooperative ETag and byte-length checks passed for all 1,306 source objects.
- Storage
- Measured 23 October 2026: 365.67 GB source payload (340.6 GiB), plus small metadata and manifest files. Hetzner bucket currently uses 32.82 GB; projected use after import is about 398.49 GB, within the stated 1 TB capacity.
- Attribution
- Credit VIDA for the combined collection and preserve attribution to Google, Microsoft and OpenStreetMap contributors. Retain the ODbL license and source provenance.