A portolan/stac catalog of every known available non-AI generated field boundary.
One rule survives every edit to this file. Every claim here is quoted from a source or measured from the data; every query below was run before it was written down and its output follows as comments.
51 collections, one per source dataset, all in the fiboa schema (id, geometry, bbox, optional metrics:area in m², determination:datetime, crop columns where the source has them). Public root: https://data.source.coop/ftw/harmonized-field-data/catalog.json. Each collection is hive-partitioned by edition: <collection>/year=<Y>/<collection>-<Y>.parquet, with the newest edition copied to <collection>/latest/<collection>.parquet.
Single files: plain https under https://data.source.coop/ftw/harmonized-field-data/. Globs: the S3 form of the same prefix, s3://ftw/harmonized-field-data/, through the Source Cooperative proxy (endpoint https://data.source.coop, path-style, no credentials) — * needs a listing and https has none. Newest edition of every collection in one query (schemas differ per source, hence union_by_name):
Every edition of every collection:
There are none. id is unique within one edition of one collection only; collections do not share identifiers and editions are not tracked across years. Spatial joins are the only bridge, and each collection is in its own CRS (proj:code on the collection and the data asset), so transform before joining.
summaries.proj:code per collection.metrics:area is square metres; year is the edition (publication) year, not an observation date.crop:code/crop:name are the source's own code list; hcat:code/hcat:name (where present) are the harmonized EuroCrops HCAT taxonomy, hierarchical by digit prefix.Assets and structural links resolve relative to the object that carries them; catalogs carry no self link. Generated by tools/catalogize.py; fix documentation there.