Datasets:

links-ads
/

fabdem-v12

tags:
  - earth
  - geospatial
  - gis
size_categories:
  - 10K<n<100K
license: other
license_name: non-commercial-government-licence-v2
license_link: https://huggingface.co/datasets/links-ads/fabdem-v12/blob/main/license.txt

FABDEM v1.2 STAC Catalog

The rasters have moved. This repository now holds the STAC metadata only. The GeoTIFFs live in the public bucket links-ads/fabdem, and every STAC item here points at them, so nothing changes for STAC clients. What does change: hf download links-ads/fabdem-v12 no longer gives you the tiles — see Accessing the data.

FABDEM (Forest And Buildings removed Copernicus DEM) is a global 30 m elevation dataset derived from Copernicus GLO-30, with forest and building height biases systematically removed by the University of Bristol. It represents bare-earth topography, which matters for hydrological modelling, flood risk, and any application where vegetation and built structures corrupt the terrain signal.

Dataset summary


Resolution	30 m (~1 arcsecond)
Coverage	Global, 19,011 tiles of ~1° × 1°
Format	Cloud-Optimized GeoTIFF, 3600 × 3600 px
Compression	DEFLATE, predictor 2, 512 × 512 internal tiles
Overviews	levels 2/4/8, average resampling
Data type	float32 elevation in metres, nodata −9999
CRS	WGS84 (EPSG:4326)
Version	1.2
Total size	438 GiB

The tiles were rewritten from the University of Bristol distribution into valid COGs with internal overview pyramids. Pixel values are unchanged — the rewrite touched only the container, so low-zoom and windowed reads no longer require pulling whole files.

Structure

catalog.json                   root catalog -> collection.json
collection.json                collection "fabdem", 19011 item links
stac_catalog/<id>/<id>.json    one STAC item per tile

Item links hang off the collection, not the catalog, so a client walking rel: item links must read collection.json.

Links inside the catalog are relative, which makes this tree a relative published catalog: the identical documents are also published in the bucket at https://huggingface.co/buckets/links-ads/fabdem/resolve/. Use whichever you prefer — this repo's CDN is faster for small documents, the bucket has no per-IP rate ceiling and is the better choice for automated traversal.

Accessing the data

Assets are plain public HTTPS objects, no authentication, readable by GDAL through /vsicurl/:

import rasterio

url = "https://huggingface.co/buckets/links-ads/fabdem/resolve/tiles/N40E000-N50E010_FABDEM_V1-2/N44E007_FABDEM_V1-2.tif"
with rasterio.open("/vsicurl/" + url) as src:
    print(src.profile["dtype"], src.shape, src.overviews(1))
    preview = src.read(1, out_shape=(1, 450, 450))  # reads an overview, not the full grid

stac_catalog_query.ipynb in this repo shows the full path: read the collection, filter tiles by area of interest, load only the matching items, and stack them with stackstac.

Limitations

Temporal snapshot: represents conditions circa 2020.
Processing artifacts: some remain in complex terrain.
Polar regions: coverage follows Copernicus DEM constraints.

Licence and attribution

FABDEM v1.2 is distributed under the Non-Commercial Government Licence v2.0 — see license.txt. Non-commercial use only; keep the licence alongside any redistribution.

Produced by the University of Bristol (https://data.bris.ac.uk/data/dataset/s5hqmjcdj8yo2ibzi9b4ew3sn), from Copernicus GLO-30 © ESA. Reprocessed and hosted by Fondazione LINKS — AI, Data & Space.

The original STAC cataloguing approach was inspired by https://github.com/cordmaur/fabdem-brazil-south.

Citation

Hawker, L., Uhe, P., Paulo, L., Sosa, J., Savage, J., Sampson, C., and Neal, J. (2022). "A 30 m Global Map of Elevation with Forests and Buildings Removed." Environmental Research Letters 17(2), 024016. https://doi.org/10.1088/1748-9326/ac4d4f