Automated bird acoustic detection at Las Arrieras Nature Reserve in Sarapiquí, Costa Rica

Ecological characteristics favor high biodiversity on the Caribbean slope of Costa Rica, but this piedmont zone is poorly studied. In birds, the use of automated song and call recognition has progressed to support bird studies about ecology and behavior. We used a Pattern Matching method to label the presence of the Cinnamon Woodpecker, Great-Green Macaw, Red-capped Manakin and Chestnut-backed Antbird to create a random forest model and detect the species’ vocalizations to characterize their vocal activity in the reserve. We used Audiomoth recorders. For all acoustic detection models, accuracy, and precision values above 91% were obtained despite the imbalance of positive and negative classes, the value of Unweighted Average Recall was high for each model. Three sites showed the highest number of detections and the number of detections per site varied among species. A preference of use for some sites within the reserve was identified for A. ambiguus and C. mentalis and a more generalist use for C. loricatus and P. exsul. For each species, greater acoustic activity was found in the morning hours with less activity in the afternoon hours for some of the species with a peak of activity between February to May. The acoustic detection patterns found agreed with the literature for each species when analyzing the ecological behaviors inside and outside the breeding season of birds in Costa Rica. This acoustic information will improve conservation decision making for the species involved and other species that develop in these ecosystems.