Land occupation and transformation

A land flow says five things. Until this taxonomy the list had somewhere to put two of them.

The last three had nowhere to be, so they lived in a comma-separated flow name — and a name is a string that three source lists spell three ways. arable, non-irrigated, intensive, annual crop, non-irrigated, intensive and BAFU’s spelling of the second were three flow objects for one piece of land. Worse, a balanced pair of factors — from forest, primary at −396.7 against to forest, primary at +396.7 — was held apart by nothing but the words from and to, in a field the deduplication does not read.

Now each is a field, and two names that decompose to the same fields are the same class with no crosswalk row required. The direction being a field is what keeps the balanced pair apart: they are different values, so they mint different objects.

Every axis is an independent question about one piece of land, so the filters are read together: Cover: Forest with Intensity: Intensive is intensively managed forest, not everything that is either. Only values this build reaches are offered — a dropdown naming a regime no flow states would invite a filter that selects nothing and look no different from one that is broken.

24 of 296 classes.

Nothing here is invented. Each axis value cites a published class once — not each of the 296 classes that mention it — and a class’s citations are computed from its fields, so two classes that share a field share that field’s citation exactly and cannot drift apart.

Three authorities, split by what they can actually type. ENVO types the terrestrial cover, because the whole built environment is a single class in IUCN GET and the source lists characterise a traffic area and a dump site differently. IUCN GET types the aquatic cover, because ENVO’s terrestrial branch does not reach the sea floor or a canal. AGROVOC types the regime — irrigated against rainfed, intensive against organic, clear-felled against coppiced — because neither of the other two says anything about how land is worked. Which authority a cover cites follows its realm and is a build error rather than a preference.

exactMatch is earned, not assumed. Most covers are honestly a close or broad match, because our value is narrower or coarser than the published class; most regimes are a related match, because the AGROVOC concept is a practice and our field says the land is under it. And where the hierarchy says one class sits under another, that edge is generated by dropping a field — so the whole hierarchy is this project’s, and no edge can be published as ENVO’s or GET’s that they do not assert.