Research Seminar: Mapping biodiversity at very-high resolution in Europe

Talk Abstract

High-resolution biodiversity monitoring is important for research, conservation, and land management, yet fine-scale, up-to-date maps of species and habitats are still largely missing at the continental scale.  To address this gap, we developed GeoPl@ntNet, an open-access web platform that makes European plant biodiversity explorable and actionable at a 50×50 meter resolution. The GeoPl@ntNet is built on top of the new GeoPlant dataset, the largest resource of its kind, including around 5M presence-only (PO) records and 100K standardized presence-absence (PA) surveys, together with multi-year satellite imagery, climate time series, and detailed environmental layers covering more than 10,000 plant species across Europe.

To transform this heterogeneous data into actionable information, GeoPl@ntNet uses a multimodal deep learning pipeline that combines remote sensing, climate, and observation data. A specialized deep species distribution model (deep-SDM) predicts species assemblages for each location, while a transformer-based model infers EUNIS habitat types from predicted species lists. The platform dynamically generates high-resolution maps and region-specific biodiversity "reports", including threatened and invasive species, habitat types, and other key indicators. Overall, GeoPl@ntNet bridges the gap between modern ecological modeling and practical conservation, providing an interactive, user-friendly gateway for researchers, decision-makers, and land managers to access, explore, and utilize biodiversity insights at a continental scale.