A cloud-native mirror of Forest Typology 2020 (v1) (Nature Trace / Neumann et al.), a global 10 m map of forest typology for the 2020 baseline. This repository holds the per-class probability tiles as Cloud-Optimized GeoTIFFs, with a geoparquet STAC index.
ℹ️ This is an unmodified, byte-for-byte mirror of the Figshare release — see Provenance & modifications. Figshare does not speak the bucket/range protocols cloud-native tooling expects, so we re-host the unzipped tiles unchanged and address them by UTM cell.
ForTy classifies each 10 m pixel of global forest into a typology for the 2020 baseline. Six classes are represented:
The five planted/natural classes (0–4) ship as per-class probability bands;
the sixth, Other land, is implicit (250 − sum of the five bands). The
categorical typology map for a pixel is the argmax over all six classes.
See the publication for full class definitions, methods, accuracy, and limitations (preprint DOI 10.31223/X58R27).
Tiles are mirrored unchanged, addressed by MGRS grid-zone cell (UTM zone + latitude-band letter), each cell holding many 10 m sub-tiles:
Bands (5 × uint8, valid 0–250, NoData 255). Each band is the quantised
probability of one class; multiply by 1/250 to recover a [0, 1] probability.
Band N carries class index N−1 (the class-table value above); class 5
(Other land) is implicit and has no band:
Public, anonymous read via the source.coop data proxy.
Find tiles via the STAC index (one Item per sub-tile; geometry in EPSG:4326,
native UTM EPSG in the projection extension, the six-class table in the
classification extension):
Read a tile (rioxarray, COG range reads):
AWS CLI (anonymous):
ForTy is projected per UTM zone and its sub-tiles don't share a single regular
grid, so the tiles are not unioned into one contiguous array here. For a
cloud-native, GDAL-free workflow, treat the STAC index as the spatial index:
select the tiles intersecting your area of interest (above), open each COG by
byte-range read (e.g. with rioxarray, or virtualize each tile with virtual_tiff
for a GDAL-free path), and reproject/mosaic onto your own target grid at load time.
Tiles are stored bottom-up (ascending y); call da.sortby("y") for a north-up
view. 255 is NoData; 0 is a real (zero-probability) value, so mask on 255,
not 0.
<cell>/<filename>.tif); the geoparquet STAC index is added by Epoch.
No tile bytes are altered.License: CC-BY 4.0. When using this data you must attribute the original authors. Please cite the dataset and its companion preprint:
Neumann, M., et al. Forest Typology 2020 (v1) [Data set]. Figshare+. https://doi.org/10.25452/figshare.plus.32288616.v1
Neumann, M., et al. Forest Typology 2020 (preprint). EarthArXiv. https://doi.org/10.31223/X58R27
This mirror is provided by Epoch and does not imply endorsement by the original authors.