Format: ORAL

Pierre Bonnet1, Antoine Affouard2, Jean-Christophe Lombardo2, Mathias Chouet1, Hugo Gresse2, Vanessa Hequet1, Remi Palard1, Maxime Fromholtz2, Vincent Espitalier1, Herv Goau1, Thomas Paillot2, Benjamin Deneu2, Christophe Botella2, Joaquim Estopinan1,2, Csar Leblanc1,2, Maximilien Servajean3, Franois Munoz4, Alexis Joly2

1 AMAP, Univ Montpellier, CIRAD, CNRS, INRAE, IRD, Montpellier, France 2 Inria, LIRMM, Université de Montpellier, CNRS, Montpellier, France 3 LIRMM, AMIS, Université Paul-Valéry - Montpellier 3, CNRS, Montpellier, France 4 LIPHY, Univ. Grenoble Alpes, CNRS, Grenoble, France

Human activities are exerting an increasingly profound impact on global biodiversity. There is a pressing need for a substantial improvement in our understanding of biodiversity to assist land managers, and decision-makers in formulating effective day-to-day strategies for preserving species and habitats. At the European level, significant efforts are underway to harness and leverage the vast corpus of floristic, environmental, and remote sensing data to refine our understanding of vegetation. The Pl@ntNet platform has been set up to meet this challenge, based on a workflow at the intersection of citizen science, botany, and data science. It now enables the automated identification of over 45,000 plant species using state-of-the-art deep learning algorithms. The millions of produced plant observations have contributed to numerous research and conservation projects, positioning Pl@ntNet in 2022 as a top-5 data provider to the GBIF international platform. In this presentation, we will introduce the innovative methods currently being developed by the Pl@ntNet team to assist field botanists. Specifically, we will discuss how todays AI technologies allow us to establish contextualized summaries of plant biodiversity in specific areas. We will discuss advancements in the analysis of multi-species images commonly generated during floristic inventories. Additionally, we will showcase results obtained through DeepSDMs (Species Distribution Models based on deep learning) for predicting the distribution of species and indicators characterizing plant biodiversity at very high resolution. These results are disseminated through free online maps and services for managers, assisting in the identification of key conservation areas. Notably, this work has enabled map production from local to regional scale, at 50 m resolution. We will provide a brief overview of identified and implemented uses of these results, demonstrating the various contexts in which these research findings can be impactful, whether for education, management, monitoring, or promotion of natural heritage.