Data sources and licences¶
Status: initial survey, October 2026. Always confirm current licence and terms before ingesting.
Rule of thumb: taxonomy and facts about species may come from open datasets; traditional knowledge comes only from consenting people/communities or from published sources whose licence and ethics permit reuse. Just because something is online does not mean it is ours to ingest.
Candidate sources¶
| Source | What it gives | Licence / terms notes | Status |
|---|---|---|---|
| SANBI National Plant Checklist | ~21.5k South African plant taxa (indigenous, endemic, naturalised); annual release; 2025 version on Zenodo | Confirm dataset licence on the Zenodo record | Primary taxonomic backbone |
| e-Flora of South Africa / PlantZAfrica | Descriptions for ~22k taxa | (verify) reuse terms with SANBI | Candidate; likely needs permission |
| GBIF | Occurrence records, species backbone | Per-dataset licences (CC0, CC BY, CC BY-NC) | Range and phenology; respect licences |
| iNaturalist | Observations + photos | Per-observer licences incl. CC BY-NC and all rights reserved; API for apps, not bulk scraping (verify) | Validation and ranges; ML training use needs care |
| Pl@ntNet API | Image identification | Free tier (reported 500/day), attribution required; non-profit terms available on request (verify) | Candidate ID provider |
| Wikidata / Wikipedia | Names, identifiers, links | CC0 / CC BY-SA | Identifier linking |
| Mozilla Common Voice | Open speech data | CC0 for many datasets | ASR research and baselines |
| SADiLaR catalogue | Text, speech, annotation resources | Varies per resource; many free for research | Language research |
| Published ethnobotany literature | Documented traditional uses | Copyright applies to text; facts are not copyrightable but community interests still matter | Cite, never copy wholesale |
| Community contributions | The heart of the project | Governed by per-record consent and agreement | Core |
Ingestion rules¶
- Record the licence and source URL on every imported row.
- Never ingest CC BY-NC data into a dataset or model that will be redistributed or used commercially without checking compatibility; keep it in a separate, labelled partition.
- Do not scrape. Use official APIs/bulk exports; respect rate limits and terms.
- Published ethnobotanical claims are "documented, not endorsed". They enter as claims attributed to the publication, with a safety review status of
unrevieweduntil a qualified reviewer signs off. - No ingestion of community knowledge from archives without the originating community's involvement. Even where an archive permits, we ask.
Data we generate¶
- Species and claim records — governed by their contributors' consent.
- Open reference data (taxonomy, safety flags, our own curation) — intended for open licence (licensing).
- Language data (audio, text) — only released as datasets for records with
ai_use = training_allowed, with community approval.
Gaps to research¶
- Exact licence on the SANBI checklist and e-Flora
- Voucher/herbarium data integration (SANBI collections)
- Toxicology databases for safety flags (authoritative and open?)
- South African Poison Information Centre data-sharing possibilities