DataNova 2026 · Open Innovation · Sustainability

Which medicinal plants of South Asia are we losing, and which have never even been assessed?

RedRoot joins five open datasets with a model validated on plant families it never saw, to show which of 1,319 medicinal plants most urgently need an IUCN Red List assessment.

Explore the priority list Read the findings

Findings

Two things the data revealed

Finding 1 · data quality

Synonyms hide threatened plants

GBIF attaches IUCN categories to the name IUCN used. When GBIF calls that name a synonym, the plant shows up as “Not Evaluated”. Kuth is Critically Endangered as Aucklandia costus, but GBIF’s accepted Dolomiaea costus returns nothing.

Status is transferred only from homotypic synonyms. A naive rule would make ashwagandha Critically Endangered via Withania chevalieri, a Cape Verde endemic. That candidate and 21 others are rejected and logged.

Finding 2 · the blind spot

Range-based screening misses the plants we harvest most

Range size vs medicinal demand

Each dot is a medicinal species. Threatened species that the model catches are mostly narrow-range endemics with few recorded uses. The ones it misses are widespread and heavily used.

Where to look

Hotspots of unassessed, likely-threatened plants

Half-degree cells with native-range records of the unassessed species the model screens as likely threatened. Larger circles hold more species. Click a cell to see which.

For assessors

Priority list

Species without a usable IUCN assessment that the model screens as likely threatened. They are candidates for expert Red List assessment and field surveys, not conclusions.

Native range Min. P(threatened) Weight on medicinal importance 0.5 Family

Hide widely cultivated Hide unknown native range

For trade monitoring

Harvest watchlist

Species explorer

Look up any of the 1,319 plants

One page per species: status, score and its drivers, range figures and the occurrence map.

How far to trust it

Validated on plant families it never saw

Model vs the standard rule

Out-of-fold predictions

What the model relies on

Permutation importance: the drop in ROC-AUC on held-out families when each feature is shuffled.

How many threatened species would assessors find?

Can the score be read as a probability?

Methods, limitations & responsible use