Nature Trace’s global forest-type probability maps at 10 m resolution, covering the v1 baseline for 2020 and annual v2 beta maps for 2017–2025. Epoch hosts unchanged GeoTIFFs alongside spatially indexed GeoParquet catalogs for discovery and filtering by location and year. The two versions use different forest classifications; v2 remains a beta release.
Forest Typology (ForTy) maps forest types at 10 m resolution. Nature Trace produced the data. Epoch hosts the files and adds STAC metadata for discovery.
Availability: The v1 baseline and the complete v2 beta annual series are mirrored byte-for-byte. The version-based STAC catalog and GeoParquet indexes describe these files. The existing v1 index remains available for existing readers.
The 2020 baseline has five stored forest-type score bands. The score for Other Land is implicit.
Data · Existing STAC index · New collection · New GeoParquet index
The beta series has one collection and one combined GeoParquet index, with an integer year column. Each annual raster has seven stored score bands, including Other Land and separate mature and regrowing primary-forest bands. It uses the same tile grid as v1.
Data · Collection · Combined 2017–2025 index
gs://nature-trace/export_rasters/forest_typology/v2_beta/<year>/. The mirror uses the object generations and MD5 hashes in the pinned inventories.CC-BY 4.0. Attribution is required. Cite the publisher's dataset and paper. The Figshare version DOI above identifies the v1 baseline. Identify v2 beta and its year separately when you use the annual series. This mirror does not imply endorsement by the authors.
v2_beta/inventory/.Each band has a valid range of 0 to 250. The NoData value is 255. Multiply a valid score by 1/250 to read it on a scale from 0 to 1. A stored value of 0 is valid data.
A categorical view is the argmax over the five stored scores and the implicit Other Land score.
Other Land is stored as band 6 in v2 beta. Do not calculate it with the v1 complement formula. The stored bands and class definitions differ between versions.
Each collection also has README.md and AGENTS.md. The compact catalog's items.parquet asset has the metadata role. It publishes no standalone Item JSONs and does not claim Portolan conformance. Use the GeoParquet rows for full-index queries and TIFF discovery; use PMTiles to view all tile footprints. Browser GeoParquet previews may show only part of the index.
Item footprints and collection extents use EPSG:4326. Each data asset has its own proj:code, proj:shape, and signed proj:transform. The CRS changes with the UTM zone. Filter by forty:cell or forty:utm_zone in the new indexes.
This example uses the available v1 index. Its raster asset key is forty.
The new indexes use the raster asset key data. Use v1/items.parquet for the baseline. The combined v2 index has an integer year column for predicate pushdown, unique IDs such as 2020_raster_50N_20_2_9, and the original tile ID in forty:tile_id.
Use v1/items.parquet for the new baseline collection. The old v1/stac/forty_2020_stac.parquet URL remains available for existing readers.
proj:transform records that orientation.This product provides a compact STAC discovery catalog: STAC 1.1 metadata over the same files, which a STAC client or an agent can read without a server.
Parquet-only discovery: this compact STAC catalog publishes no standalone Item JSONs and does not claim Portolan conformance. The items.parquet asset has role metadata. Query the GeoParquet rows for TIFF assets; complete STAC Item traversal is unavailable. The v2_beta/items.parquet index combines 2017–2025; filter the integer year column. IDs include the year and forty:tile_id preserves the original ID. For spatial filters use geometry_bbox, the non-wrapping geometry covering; bbox preserves wrapped STAC bounds.