Identifying individuals and estimating population density : the case of the boreal lynx ; Identification d’individus et estimation de densité de population : le cas du lynx boréal

We are currently experiencing a biodiversity crisis, with species becoming extinct 10 to 1000 times faster than the natural rate (OFB, 2022; IPBES, 2022). It is therefore becoming essential to monitor biodiversity. To do this, we need ecological indicators such as abundance and diversity. New technologies have enabled the emergence of non invasive monitoring methods such as phototraps. This is the case here for the boreal lynx, whose detection rate is very low and requires automatic monitoring. At the same time, Artificial Intelligence has emerged to automate the prediction of large volumes of data using deep learning. On the other hand, the spatialized method of capture recapture has emerged, making it possible to estimate population size and density. Can w e therefore consider deep learning and spatial capture recapture models as a solution for lynx population monitoring using photo traps? To this end, an open population deep learning model was developed, comparing individual photos with each other to predic t the class of a new image. Also, two spatial capture recapture models, using human and AI processed data, are compared to see the impact of AI on population density predictions. We obtained a model which, for a given new image, has a 67% chance of finding the same individual among the 5 closest images. This l ed to an estimate of lynx density by AI that was three times higher than that obtained by humans. ; Nous vivons actuellement une crise de la biodiversité, les espèces s’éteignant 10 à 1000 fois plus rapidement que le rythme naturel (OFB, 2022 ; IPBES, 2022). Il devient donc essentiel de suivre la biodiversité. Pour ce faire nous avons besoin d’indicateurs écologiques tels que l’abondance et la diversité. Ainsi, de nouvelles technologies ont vu le jour et ont permis l’émergence de méthodes non-invasives pour le suivi comme les pièges photos. C’est le cas ici pour le lynx boréal dont le taux de détection est très faible et nécessite un suivi automatique. En parallèle, l’Intelligence Artificielle a ...