Soundscapes and deep learning enable tracking biodiversity recovery in tropical forests

Area of Science:

• Ecology
• Conservation Biology
• Bioacoustics

Background:

• Tropical forest degradation exacerbates climate change and biodiversity loss.
• Assessing the pace of biodiversity recovery in regenerating forests is crucial but challenging.
• Effective monitoring tools are needed to guide tropical forest restoration efforts.

Purpose of the Study:

• To evaluate the efficacy of bioacoustic and metabarcoding techniques for measuring forest recovery.
• To determine if automated acoustic measures can reflect the restoration gradient of vertebrate and invertebrate communities.
• To demonstrate the utility of new technologies for monitoring tropical forest restoration success.

Main Methods:

• Utilized bioacoustics and DNA metabarcoding to assess forest recovery in Ecuador.
• Analyzed vocalizing vertebrate community composition and expert-identified species.
• Developed and applied an acoustic index model and a Convolutional Neural Network (CNN) for automated analysis.

Main Results:

• Community composition, not species richness, of vocalizing vertebrates indicated forest restoration.
• Automated acoustic measures strongly correlated with the restoration gradient (adj-R² = 0.62 and 0.69).
• Automated acoustic measures also reflected the composition of nocturnal insects detected via metabarcoding.

Conclusions:

• Automated bioacoustic monitoring is a reliable tool for assessing tropical forest recovery.
• New technologies offer robust and reproducible data for conservation and restoration initiatives.
• Effective monitoring supports strategies to address climate and biodiversity crises.