Life sciences
Analyzing and mapping plant communities using AI and the Pl@nBERT tool
Date:
Changed on 03/12/2025
A study by CIRAD, INRIA, the University of Montpellier and several French and European partners has demonstrated the merits of a language model applied to plants for understanding plant communities and mapping natural habitats in Europe. It was one of the first studies to use an artificial intelligence model to characterise plant communities on a large scale.
The study was conducted as part of a thesis, with financial support from the EU GUARDEN and MAMBO projects, and paves the way for a better understanding of biodiversity and better preservation, thanks to artificial intelligence (AI). According to the researchers, while this work centred on understanding assemblages of plant species and mapping European natural habitats, the AI model developed during the study has numerous possible applications.
Pl@ntBERT, a language model applied to plants
To build their model, the scientists based themselves on the BERT tool, a form of artificial intelligence that serves to understand text, which people use daily without even realising when doing Google searches. To make it work better on plants, the researchers trained it by feeding it with more than a million European vegetation records. This plant-specific language model, called Pl@ntBERT, is capable of deciphering the "syntax" of the phrases formed by cohabiting plant species, classified in order of abundance.
César Leblanc, Lead author of the study, PhD student with the INRIA-University of Montpellier IROKO project team
Just as a conventional language model is capable of finding a missing word in a sentence, Pl@ntBERT can predict the plant species likely to be found in an assemblage of species and thus supplement field records.
In practical terms, Pl@ntBERT facilitates vegetation inventories by suggesting species that are probably present but not recorded. The tool even goes further, by predicting coherent, previously unseen species assemblages.
Mapping habitats to protect them better
The study also demonstrated Pl@ntBERT's capacity to classify types of habitats, out of the 250 recorded European natural habitats, very accurately and based on lists of species. Habitat maps are available to the public on the geo.plantnet.org website.
Pierre Bonnet, biodiversity informatics researcher with CIRAD, AMAP research unit, and co-author of the study
The EU Nature Restoration Law, adopted in 2024, relies on the notion of habitat to identify restoration strategies. This is why it is important to inventory habitats on a European scale, to be able to monitor their future evolution.
Differences between the expected species assemblages for a given type of habitat and those observed on the ground may therefore reveal ecosystem disruptions and prompt early, appropriate conservation actions.
Outside Europe, this type of approach offers considerable prospects for tropical regions whose exceptional biodiversity is still largely unknown, could be pinpointed better, notably in order to steer conservation and restoration efforts within these ecosystems, which are among the most vulnerable on the planet.
Lifting the veil on dark diversity: another application is planned
Predicting the species missing from an assemblage and mapping habitats are not the only things Pl@ntBERT can do! As the researchers stress, its use goes well beyond those two objectives, and they hope to see the scientific community take this open-access tool on board. The field of play is huge, for instance assessing dark diversity more effectively. This fashionable concept refers to the species that could live in a given habitat but are not there, for as yet unknown reasons. Pl@ntBERT could soon provide clues.
Reference:
Leblanc, C., Bonnet, P., Servajean, M., Thuiller, W., Chytrý, M., Acic, S., Argagnon, O., Biurrun, I., Bonari, G., Bruelheide, H., Campos, J.A., Carni, A., Custerevska, R., De Sanctis, M., Dengler, J., Dziuba, T., Erenskjold Moeslund, J., Garbolino, E., Jandt, U., Jansen, F., Lenoir, J., Perez Haase, A., Pielech, R., Sibik, J., Stancic, Z., Uogintas, D., Wohlgemuth, T. & Joly, A. Learning the syntax of plant assemblages. Nat. Plants 11, 2026–2040 (2025). https://doi.org/10.1038/s41477-025-02105-7.