Towards automated animal density estimation with acoustic spatial capture-recapture.

Area of Science:

• Ecology
• Bioacoustics
• Computational Biology

Background:

• Passive acoustic monitoring (PAM) is crucial for surveying elusive wildlife.
• Machine learning (ML) aids call detection but struggles with false positives.
• Existing methods inadequately address ML-derived false positives in acoustic surveys.

Purpose of the Study:

• To introduce a novel acoustic spatial capture-recapture (ASCR) method.
• To address the under-investigated issue of false positives in ML-based acoustic monitoring.
• To improve wildlife population density estimation from acoustic data.

Main Methods:

• Developed an ASCR model treating species identity as a latent variable.
• Incorporated ML detection outputs as random variables dependent on latent identity.
• Utilized a mixture model likelihood for estimating call density.

Main Results:

• The proposed ASCR method demonstrated improved accuracy compared to a standard false positive correction factor.
• Estimates from the new method closely aligned with analyses excluding false positives.
• Simulations confirmed near-zero bias and accurate coverage probabilities.

Conclusions:

• The novel ASCR method effectively mitigates false positives in acoustic monitoring.
• This approach enhances the reliability of wildlife density estimations from acoustic data.
• The method offers a significant advancement for bioacoustic surveys and ecological research.