The sound of hope: searching for critically endangered species using acoustic template matching

Abstract

Passive acoustic monitoring (PAM) has become increasingly popular in monitoring biodiversity. It produces large amounts of dataand can provide a foundation for understanding the long-termconsequences of environmental degradation. However, extractingbiological information from such extensive datasets can be challenging and requires advanced computational skills. Herein, we introduce a streamlined workflow for detecting acoustic signals of threecritically endangered birds: Cherry-throated Tanager (Nemosiarourei), Alagoas Antwren (Myrmotherula snowi), and Blue-eyedGround-dove (Columbina cyanopis). As these species are amongthe world’s most endangered birds, locating new populations isa top priority. We chose potential templates based on the acousticparameters of the vocal repertoire and evaluated their performanceusing soundscapes with known composition (gold standard data).To evaluate the efficiency of the templates, we used precision andrecall metrics and found that achieving high precision rates comesat the cost of recall rates. Although we used gold standard data tocalibrate our algorithm, large-scale validations have revealed thelimitations as some templates have exhibited significantly lowerprecision values. The use of binomial models helped reset precisionvalues to 90%. Our workflow can process large amounts of dataefficiently, helping to monitor populations of these critically endangered species, locate new populations and evaluate populationdynamics.