Bat Detection and Species Determination Around Wind Turbines using AI

Problem Statement The increased use of wind energy has negatively affected resident and migratory bat species. Innovative and cost-effective technologies, such as passive acoustic monitoring, are needed to refine our understanding of the risks of wind turbine interactions. However, challenges remain in using acoustic data to identify call signatures of bat species, such as the quality of annotated data, the lack of advanced models, and the unavailability of do-it-yourself AI tools. Addressing these challenges is critical to minimize wildlife impacts and supporting sustainable wind energy development in the United States. How This Problem or Situation is Being Addressed To better identify and reduce the effect of wind turbines on resident and migratory bat species, an open source system is proposed to automate bat detection and species determination around wind turbines. The system will allow experts to curate acoustic data, perform clustering for auto-discovery, and use AI models to accurately predict bat species from echolocation bat calls. Interactive visualizations of data and AI model outputs will also be available in a web browser to facilitate data discovery. This system is expected to advance the state of the art in the field and support the environmentally sustainable development of wind energy in the United States. SBIR Phase I Activities The initial phase involves collecting and processing data to develop AI models and user interfaces for bat detection and species identification around wind turbines. The aim is to collect a representative dataset in collaboration with NREL and USGS, which will be used to train deep neural networks to develop an initial prototype classifier. The dataset's high degree of annotation uncertainty will be addressed by clustering and unsupervised methods. A web interface will enable users to run algorithms on the server side and visualize spectrograms and data clusters. Commercial Applications and Other Benefits Our proposed system will create curated data and an ensemble of AI models that can be used and retrained by bat experts – such as those in the North American Bat Monitoring Program – to improve data collection, storage, and sharing accuracy and speed. The open source model will help other communities working on acoustic data by providing a system that integrates data, models, and visualization in an accessible web application. Our approach can also be applied in underwater passive acoustics and related terrestrial wildlife or urban monitoring domains.