Passive acoustic monitoring and automated detection of the American bullfrog

Abstract

Biological invasions pose significant threats to biodiversity and ecosystem functioning. Removal of introduced species is most successful when detected early. We evaluate the effectiveness of passive acoustics combined with automated recognition in detecting the invasive American bullfrog (Lithobates catesbeianus). We applied this technique to two real-world monitoring programs aimed at determining the optimal time of day for monitoring the species in Europe, for which we recorded the species in Belgium and Italy; and for evaluating the effectiveness of BirdNET (a free and user-friendly automated recognizer) in analyzing a large dataset collected in Spain. BirdNET was highly effective in automatically detecting the bullfrog presence, with a detection rate (compared to visual inspection of sonograms) of 89.5% using default settings (85 of 95 recordings with known presence), and 95.8% with user-specific settings (91 of 95 recordings detected). The system showed remarkable precision, correctly identifying 99.7% (612 out of 614) of the verified predictions, and with only one mislabelled recording (predicted to be present when it was absent). The species’ vocal activity in Belgium and Italy was higher during the night compared to crepuscular periods. Recording analyses and output verification of the dataset collected in Spain was carried out in 3.8% of the recorded time, and resulted in significantly reduced effort compared to visual inspection. Our study highlights the effectiveness of this technique for remotely surveying the American bullfrog, making it a significantly potential tool for informing management decisions, particularly for the early detection of the species’ arrival in new areas.