Advancing software capacity for the statistical analysis of AI-processed ecological data

Artificial intelligence (AI) and remote sensing technologies, such as camera traps and acoustic recording units, are transforming ecological monitoring by enabling large-scale assessments of species distributions and abundance. However, realizing the full potential of AI-derived monitoring data requires statistical tools that account for AI classification errors and other challenges unique to remotely sensed observations. This project will expand the widely used open-source Ecoverse software ecosystem with new AI-focused functionality and workflows that allow researchers to more effectively leverage AI-generated data. Specifically, we will develop three new classes of statistical models that estimate species and community distributions while explicitly accounting for false positive and false negative errors produced by AI classification algorithms. Comprehensive workflows, from data preparation through interpretation, will make these advanced methods accessible to both novice and experienced users. By integrating state-of-the-art AI applications with rigorous ecological inference, the enhanced Ecoverse will enable researchers, and natural resource managers to make more informed decisions from rapidly growing AI-based monitoring datasets. Two training workshops and a graduate seminar at North Carolina State University will further build capacity by training users and fostering the next generation of Ecoverse contributors and maintainers. All software developments will be publicly available and open source, broadening access to advanced AI-enabled analytical tools across the environmental science and natural resource management community while advancing a strategic priority of accelerating the responsible development and adoption of artificial intelligence. These updates to the Ecoverse software packages will advance the fields of ecology and ecological statistics by providing open-source software explicitly designed to analyze remote sensing data processed with AI. New false positive occupancy and abundance models will account for inherent false positive errors that arise from AI classification algorithms, improving the reliability of ecological inferences for individual species and multi-species communities. New coupled classification models will allow users to more efficiently estimate species distributions and abundance by directly incorporating the continuous output from AI classification algorithms. Functionality for continuous-time occupancy models will provide the ability to answer questions about species interactions and diel activity patterns that are not possible with standard models implemented in discrete time. Altogether, these updates will allow researchers across the world to gain unprecedented insights into myriad ecological processes such as species distributions, populations, and species interactions. Importantly, the proposed updates can be used to generate ecological insights on any taxonomic group that can be monitored with remote sensing and AI classification. 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.