Integrating ecoacoustic monitoring with machine learning to survey black-and-white ruffed lemurs (Varecia variegata) in Madagascar's Corridor Forestier d'Ambositra Vondrozo (COFAV)

Black-and-white ruffed lemurs (Varecia variegata) are a Critically Endangered species limited to the fragmented eastern Malagasy rainforests. The importance of their conservation is underscored by their vital seed dispersal and pollinator niches; they are a keystone species and indicator of forest health. Little is known of their distribution in the Corridor Forestier d'Ambositra Vondrozo (COFAV), an important rainforest corridor connecting multiple protected areas. Because of their fragmented and dispersed distribution, we used passive acoustic monitoring to survey ruffed lemurs throughout the COFAV. We deployed autonomous recorders at 30 sites across the COFAV from December 2021 to June 2022. We then implemented a machine learning pipeline by compiling an annotated training data subset of ruffed lemur call presence and absence clips and then training a convolutional neural network to automatically detect these calls in the full dataset (550,000 recording minutes). After validating top model predictions, we used acoustic detections and non-detections with geospatial variables in spatial occupancy models to assess ruffed lemur occurrence across the COFAV. Ruffed lemurs were only detected at 10 of 30 sites, all of which were in the northern COFAV. There appears to be an abrupt delineation in their distribution around a band of deforested land near the middle of the corridor, which may be preventing their southward dispersal. We found that ruffed lemur occupancy was positively influenced by the Normalized Difference Vegetation Index (NDVI), a proxy for vegetation health. This study demonstrates the power of combining different conservation technologies to generate meaningful ecological and conservation insights, and can be generalized to other species.