Passive acoustic monitoring of an endangered lemur species: conservation and sustainable development

The accelerating loss of biodiversity calls for practical, scalable, and automated systems to monitor wildlife populations. Passive Acoustic Monitoring (PAM) is emerging as a critical tool to achieve this, particularly for a singing species like indri (Indri indri), the largest extant lemur endemic to Madagascar and, according to IUCN, critically threatened with extinction. Indri groups produce species-specific loud calls, or songs, which are key in mediating intra- and interspecific interactions. These complex vocalisations can travel considerable distances from the point of emission and are easily distinguishable within passively recorded data, making them particularly well-suited for PAM. My thesis explores the potential of PAM in advancing our understanding of these species’ behaviour, ecology, and conservation. Each chapter focuses on distinct yet interconnected aspects of this research. In the first chapter, I show the potential of an automated algorithm in detecting the presence of indris in the Maromizaha New Protected Area in Madagascar. By processing 66,443 tenminute recordings collected between 2019 and 2021, I developed a convolutional neural network that achieved high accuracy (>90%) and recall (>80%) in detecting indri songs. This workflow significantly reduces the manual effort required to classify acoustic data and provides critical insights into indris's daily and annual vocal patterns, informing optimised field data collection. The second chapter examines the relationship between weather variables and indri vocal behaviour. Using PAM, I analysed how temperature and precipitation influence indri singing patterns. Results showed that rainfall constraints vocal activity, while warmer temperatures correlate with increased song emission. These findings underscore the value of PAM in studying species’ biology and behavioural adaptations to environmental conditions, offering essential insights for conservation in the context of climate change. In the third chapter, I evaluate the utility of PAM for estimating indri density in Maromizaha challenging tropical forest environment. By adapting methods from marine acoustics and incorporating Generalized Linear Mixed Models, I estimated indri density with remarkable accuracy (predicted vs real: 5.55 vs. 6.15 groups per km²). This approach highlights PAM’s potential to overcome limitations of traditional survey methods, providing reliable data on population density and habitat usage to guide evidence-based conservation strategies. In the final chapter, I integrate a novel application of BirdNET, a deep neural network algorithm for sound recognition. I trained this algorithm on Maromizaha-specific data to detect vocalisations of Indri indri and Varecia variegata, another critically endangered and loud calling lemur species living in the Maromizaha New Protected Area. I used the resulting data to investigate interspecific vocal interactions and overlap in their temporal and spatial vocal behaviour. This chapter synthesises ecological, behavioural, and acoustic insights, highlighting the complementary role of automated detection tools and PAM in studying vocal animals. Together, these studies illustrate PAM's transformative potential in advancing our understanding of vocal primate ecology and developing robust, data-driven conservation strategies for these endangered species and their habitats.