Bat Detection and Species Determination Around Wind Turbines using AI

The growth of wind energy as a key component of the global renewable energy portfolio has been accompanied by environmental challenges, notably impacts on wildlife, such as bats. Bats are critical to many North American ecosystems and provide several ecosystem services for humans (e.g., agricultural pest suppression), yet their populations are threatened by interactions with wind turbines. Ensuring wind energy build-out without incurring biodiversity losses presents a complex dilemma. Addressing the adverse effects on bats is not only a matter of regulatory compliance but also pivotal for the sustainable advancement of wind energy technologies. To navigate these challenges, it is imperative to innovate and implement cost-effective, non-invasive monitoring technologies that can mitigate wildlife impacts without hindering the growth of wind energy. To this end, Kitware, in collaboration with the National Renewable Energy Laboratory (NREL), has been developing a solution to detect bats and identify their species from acoustic data. This solution leverages artificial intelligence (AI) algorithms enriched with expert verification to ensure precision and reliability. In Phase I, we developed a web-based platform to facilitate a range of functionalities for stakeholders, including the ability to upload and disseminate bat echolocation calls, visualize call spectrograms, and manually annotate these spectrograms at both the pulse and sequence levels. Concurrently, we have engineered advanced AI models capable of analyzing spectrograms and audio classifications to accurately identify bat species. The integration of these AI models into our platform is currently underway. Domain experts from the U.S. Geological Survey (USGS), U.S. Forest Service (USFS), and other organizations have engaged with the initial application, hosted at batdetectai.kitware.com, contributing valuable feedback which has significantly enhanced the platform. In the proposed Phase II effort, we will continue to develop our AI system and perform dataset compilation in collaboration with domain experts, including subcontractors at NREL and USFS alongside USGS and other collaborators. We will advance data management and control mechanisms in preparation for using our app at the enterprise level, and develop a desktop version of the software to accommodate individual user deployments. We aim to productize the machine learning models developed in Phase I with an automated continuous model training and testing infrastructure (MLOps), which will evaluate model performance as new data is added to the system. Additionally, we will invest in methods that expedite annotation in novel domains, such as unsupervised clustering or multi-modal annotation. The software's utility extends beyond its initial application, offering potential for broad deployment in detecting migratory bat speciesŚa critical concern for mitigating bat fatalities at existing and future wind turbine sites. Moreover, we anticipate the applicability of our technology in a wider array of fields, including underwater passive acoustics and urban monitoring. This initiative will be released open-source, promoting both early-stage engagement. Our approach leverages Kitwareĺs extensive experience with open-source projects, as evidenced by our successful deployment of platforms such as VIAME (https://www.viametoolkit.org), ParaView (paraview.org), and CMake (cmake.org), further underscoring our commitment to innovation and collaboration within the open-source ecosystem.