Integrating acoustic monitoring and machine learning for orthoptera biodiversity assessment

Post restoration monitoring is crucial for evaluating the success of forest restoration efforts. However, monitoring how forest ecosystems respond to restoration is challenging due to the complexity of forest ecosystems, logistical constraints, limited funding and labor-intensive sampling methods. Orthoptera may be a useful insect group to focus on due to their species-specific acoustic signals and ecological significance. Assessing restoration success with acoustic surveys for Orthoptera presents an opportunity for cost-effective biodiversity monitoring.

Here, we aim to compare traditional sampling techniques with passive acoustic monitoring using AudioMoth recorders. We collected data in a forest restoration site in Cuyahoga Valley National Park (Ohio, USA). To streamline species identification from acoustic data, we developed a convolutional neural network (CNN) model (ResNet34) utilizing the OpenSoundscape Python library. The model was trained on annotated Orthoptera calls and validated against manually identified recordings.

Preliminary results indicate notable differences in Orthoptera detection rates between sampling methods, suggesting that no single approach is sufficient to capture the entire Orthoptera community. Integrating acoustic monitoring with traditional sampling provides a more complete understanding of Orthoptera diversity, particularly valuable in resource-limited contexts. This combined approach offers a promising strategy for sustainable, long-term biodiversity monitoring in forest restoration projects, facilitating efficient allocation of resources and supporting informed ecological management decisions.