NSF Award
2426762
Humans are part of a rich global ecosystem containing vast numbers of different species of plants and animals. These highly diverse species interact in complex ways to fundamentally impact human life, health, and economy. Remarkably, most species on earth remain unknown to science. This is particularly true for small organisms, such as insects and fungi, that operate behind the scenes but have a fundamental impact on breaking down and recycling waste and on food production. Particularly given these critical roles, there is growing concern that global environmental change may lead to a reduction in biodiversity and impact which species occur where and how they interact. To reliably study such processes and obtain critical data for making policy decisions, it is imperative that automated methods are available for biodiversity monitoring. This project develops transformative new wireless communication and artificial intelligence (AI) tools for automating biodiversity data collection, processing, and analysis. These tools can be used to infer which bird, insect and bat species are present at any given location and time based on their vocalizations, while assessing relationships with factors ranging from climate to urbanization. The AI tools developed by this project will impact not just biodiversity research but many other fields across the sciences and industry. Young researchers involved in the project will receive training in developing and applying new AI methods in challenging settings. <br/><br/>This project develops novel AI methods and applies wireless communication technologies for biomonitoring. Under the ongoing biodiversity challenges, methods are needed for autonomously inferring which species are present in different spatial locations; such monitoring provides critical data for studying impacts of climate change and environmental disruption on biological community dynamics. Current data on biodiversity are heavily taxonomically and spatially biased. Scientists still know remarkably little about the dynamics of species communities, driving which species are present, the interactions among these species, and the impact of disruptions. There have been recent improvements in labor-intensive methods for biodiversity monitoring across broad groups of taxa, including birds, mammals, insects, and fungi. However, data collection remains very costly and requires substantial human intervention and management. Contemporary tools for species classification are error-prone; failing to account for these errors can lead to inaccurate scientific conclusions and flawed policy recommendations. Next-generation biomonitoring requires cost-efficient technologies for automatic, adaptive sampling and flexible inference to characterize and learn from ecological conditions in real time. This project develops transformative new tools for fundamentally improving biodiversity monitoring having immense societal and scientific impact. Fundamental innovations include: (1) probabilistic paradigms to account for errors in inferring species composition from audio, imaging and DNA-barcoding data; (2) new classes of interpretable and identifiable AI-Joint Species Distribution models to characterize how community composition is driven by biotic and abiotic factors; (3) improvements in wireless communication technologies for remote monitoring; and (4) adaptive designs to optimize allocation of limited resources to maximize learning of ecological community dynamics.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
