BINOCULARS TO BYTES: DEVELOPMENT AND FIELD VALIDATION OF AN AI-DRIVEN SYSTEM FOR AVIAN MONITORING

Author

Christian J. Dupree, University of Montana, MissoulaFollow

Year of Award

2025

Document Type

Thesis

Degree Type

Master of Science (MS)

Degree Name

Wildlife Biology

Department or School/College

W.A. Franke College of Forestry and Conservation

Committee Chair

Victoria Dreitz

Commitee Members

Zachary Cheviron, Doug Brinkerhoff

Keywords

Camera trap, Waterfowl, wetland monitoring, object detection, distance estimation, Artificial intelligence

Subject Categories

Artificial Intelligence and Robotics | Ornithology | Population Biology | Terrestrial and Aquatic Ecology

Abstract

Autonomous camera-trap arrays coupled with artificial-intelligence (AI) vision can lift bird monitoring beyond the spatial, temporal, and labor limits of traditional field surveys. We present Binoculars to Bytes (B2B), an open-source pipeline that turns 180–360° time-lapse imagery into analysis-ready avian data. At Freezout Lake Wildlife Management Area (Montana, USA) the system ran four-hour morning deployments during spring and fall migrations (2023–2024). A YOLO-NAS detector, incrementally refined with a “Specialized Localized Iterative Model” workflow, quadrupled local accuracy and, after confidence-based species-binning, cut false-positive rates in half. Daily AI species lists were benchmarked against contemporaneous eBird citizen-science checklists and recovered ≥ 70 % of the birds reported each day at lenient probability thresholds (≤ 0.25). Camera-derived counts—extracted from the peak image in each 30-s window—were compared with time-matched tallies by professional biologists at six ponds; mean absolute error was 18 birds, with undercounts concentrated in large flocks. Continuous sampling captured first-arrival dates up to two weeks earlier than field observers and tracked departure within days. A proof-of-concept monocular-depth routine yielded half-normal detection curves consistent with conventional distance-sampling theory, paving the way for density estimation. By integrating adaptive AI training, confidence-aware post-processing, and optional distance correction, B2B provides high-resolution information on waterfowl presence, migration timing, and relative abundance while sharply reducing observer bias and manual workload. The pipeline bridges computer vision and population ecology and offers a transferable template for large-scale, near-real-time avian conservation monitoring.

Recommended Citation

Dupree, Christian J., "BINOCULARS TO BYTES: DEVELOPMENT AND FIELD VALIDATION OF AN AI-DRIVEN SYSTEM FOR AVIAN MONITORING" (2025). Graduate Student Theses, Dissertations, & Professional Papers. 12443.
https://scholarworks.umt.edu/etd/12443