Sources

Every value in Trefle comes from somewhere else. This page is the register: what each source is, what it contributes, how much of the database it accounts for, and how to credit it.

If you want to know where one particular value came from, ask the API directly: see data provenance. This page tells you what the answer means once you have it.

How to read the table

Feeding the database means records are linked to that source today, with the count measured against the production database. Ranked but not yet connected means the source has a place in the arbitration order and nothing is importing from it yet. The rank is reserved, not active.

The rank decides who wins when two sources claim the same field. It runs strongest first, and the reasoning behind the order is in understanding the data.

Feeding the database

Rank	Source	What it contributes	Records linked
1	Community corrections	Values reviewed and accepted by a maintainer, submitted with a reference	—
3	TRY	Plant traits from published measurements: life form, lifespan, flower and fruit characters, leaf texture, root depth, leaf C/N, Ellenberg indicators	Import built, not yet run against production
8	POWO (Kew)	Accepted names, authorship, synonymy, native range. The nomenclatural backbone	1 242 348
9	World Flora Online	Accepted names, including groups Kew does not cover	360 940
10	IPNI	Where and when a name was first published	316 686
11	GBIF	Occurrence records, common names, how often a species is observed	360 318
12	USDA PLANTS	Agronomic ratings for the North American flora	Historical — see below
15	Baseflor / Catminat (Philippe Julve)	Ecological indicator values, flowering period, flower colour, fruit type	~3 500 species per indicator
18	Pl@ntNet	Photographs, common names	31 547
21	Wikipedia	Descriptions and common names — never numbers	Text only

WCVP (Kew's World Checklist of Vascular Plants) is used as a bulk dataset rather than a ranked source: it supplies the genus-to-family table and the synonymy that names are reconciled against, upstream of arbitration.

Why Wikipedia never contributes a measurement

A Wikipedia article giving a plant's height is quoting a source. We would rather cite that source. So Wikipedia contributes prose and vernacular names, and nothing that lands in a numeric field.

Ranked but not yet connected

These hold a place in the arbitration order so that connecting them later does not require re-arguing precedence. Nothing imports from them today, and a species will not show them as a source.

Rank	Source	Intended contribution
2	IUCN	Conservation status
4–7	Flora Iberica, Flora of North America, Flora of China, Flora Europaea	Measured, peer-reviewed descriptions, authoritative for the flora they cover
13	Tropicos	Nomenclature
14	Tela Botanica	French flora
16	Trees and Shrubs Online	Dendrological descriptions
17	EOL	Aggregated descriptions
19	Plants For A Future	Edibility and uses
20	OpenFarm	Cultivation guidance

The four regional floras sit above every global aggregator on purpose: within the flora they cover, a specialist description outranks a database that mostly relays.

The USDA inheritance

A large part of the trait data predates the provenance system. It came from the USDA PLANTS database, it only ever applied to the North American flora, and most of those columns are no longer exposed by the API. Where one survives, treat it as historical.

This matters for arbitration. A value written before provenance recording existed carries no fact, so nothing defends it on the record. Rather than let the weakest source overwrite good legacy data simply because no competitor was on file, an unbacked column value is treated as though claimed by a source at rank 7, the level of a regional flora. Only the six ranks above it may replace such a value; anything weaker can still fill a gap, and its claim is recorded as a competing fact either way.

Crediting a source

Several sources are published under licences that require attribution. If you redistribute Trefle data, the obligation travels with it.

Source	How to credit
TRY	Kattge, J. et al. (2020) TRY plant trait database — enhanced coverage and open access. Global Change Biology 26, 119–188. CC BY: credit the TRY initiative and its contributing datasets
Baseflor / Catminat	Julve, Ph., 1998 ff. — baseflor. Index botanique, écologique et chorologique de la flore de France. Programme Catminat
POWO / IPNI	IPNI, published on the Internet at ipni.org; and the World Checklist of Selected Plant Families, Royal Botanic Gardens, Kew
Wikipedia	Wikipedia, CC BY-SA

The register is incomplete

Trefle stores a citation template per source, and several are still empty: GBIF, USDA, Tropicos, IPNI, World Flora Online, Tela Botanica, Pl@ntNet and OpenFarm currently have none recorded. Their own sites state their terms, and you should follow those rather than assume our silence means no obligation. Filling these in is open work.

Where a field's shape is decided

A source hands us a value in its own units and on its own scale. What Trefle stores is a separate decision, and it is written down in two places:

• config/traits.yml in the API repository is the contract: for every field, its unit, the range a value must fall in to be accepted, and any closed vocabulary. It is the file the ingester, the plausibility checks and the completeness ratio all read, so it cannot drift from behaviour.
• The mapping tables used by the import pipeline record, per source and per field, the conversion applied and the published reference for it.

Where a conversion would misrepresent a value, we decline it and say so. Take the Ellenberg reaction indicator: it ranks soil acidity on an ordinal scale, our ph_minimum and ph_maximum hold pH readings, and a published table converting one to the other exists. We do not apply it, because it would double the pH columns' population with derived values in a field that could no longer tell you which of its numbers were measured. The indicator is kept as evidence and promoted to nothing.

See the ecological indicators for what that looks like on the fields you actually receive.