Dynafor présent à la 4ième édition de la conférence JJBA

Léa Mariton est intervenue le mercredi 17 juin dernier à Besançon pour parler des travaux de post-doctorat qu'elle mène à Dynafor dans le cadre de la conférence internationale des Journées des Jeunes BioAcousticien·nes (en anglais Emerging Bioacousticians' Days). Sa communication orale 15 min s'intitulait : "Assessing the impact of agriculture practices on bird communities using ecoacoustics" ("Evaluation des impacts des pratiques agricoles sur les communautés d'oiseaux grâce à l'écoacoustique").

Mariton, L.; Laroche, F.; Barbaro L (2026) Assessing the impact of agriculture practices on bird communities using ecoacoustics. Conférence internationale des Journées des Jeunes BioAcousticien·nes, 16 au 18 juin à Besançon (France)

Abstract:

Since 1962 and the publication of Rachel Carson’s book "Silent Spring", the threat posed by intensive agriculture to the abundance and population trends of birds has been extensively documented. Practices that are more respectful of biodiversity are needed, and it is essential to evaluate their effectiveness. Traditionally, to do so, data collection has been carried out by human observers in the field, who conducted point counts. However, this method is time-consuming and allows birds to be observed only for a limited time at each site. In comparison, passive acoustic monitoring (PAM), combined with deep learning tools, could enable monitoring on much larger temporal and spatial scales, although this approach remains to be tested in agricultural landscapes. The objective of this study is to use ecoacoustics to determine whether bird acoustic communities are influenced by agricultural practices at the landscape scale. To this end, we collected data from 20 agricultural areas across France during the spring, between 2023 and 2025. In each area, recordings were made over several days at multiple sites located in landscapes with varied agricultural practices. The collected acoustic data were processed using BirdNET, an automatic classifier of bird vocalisations (convolutional neural network). Since validating a subset of detections is crucial to avoid any bias, this step must be carried out carefully by experts to allow for the adjustment of BirdNET’s confidence thresholds according to each species. Because acoustic data differ from occurrence data, further reflections are needed to determine how to define an acoustic community and how to perform statistical analyses on this data without knowing the detectability of the target species. Through this study, we aim to demonstrate that ecoacoustics is a promising monitoring method in agricultural landscapes, provided that the methodological challenges associated with collecting vocalisation data are taken into account.