Abstract:

Ice seals (ribbon, ringed, spotted, & bearded) use spring sea ice as a platform for pupping, resting, and their annual molt. Seals play vital roles in Arctic and subarctic marine ecosystems & are a resource for Alaska Native communities. To manage & conserve these species, reliable population estimates and distribution maps are needed for management and understanding how they respond to climate change and other human impacts. The Polar Ecosystems Program at NOAA’s Alaska Fisheries Science Center conducts aerial surveys to estimate the abundance and distribution of ice-associated seals and polar bears in the Bering, Chukchi, and Beaufort seas. Millions of images are collected using color (RGB), thermal infrared (IR), & ultraviolet (UV) cameras. Current IR machine learning (ML)models struggle to detect rare animals, including unattended white-coat seal pups, because of their size, and polar bears because of their variable thermal signatures. UV imagery has been introduced to address these challenges. This project seeks to enhance ML detection models by integrating UV, RGB, and IR imagery, using annotated datasets developed by NOAA researchers for training and validation. The goal is to create a robust, open-source model capable of detecting 80% known animals with fewer than 40%false positives. This system will improve survey efficiency & accuracy, providing reliable population estimates & supporting conservation efforts. With the expected outcome of more precise abundance estimates for decision-making, this project supports NOAA’s mission to understand and predict species distribution under changing sea ice conditions. The results are from the NOAA EPP/MSI CSC NERTO graduate internship project conducted with NOAA mentor Ms. Erin Moreland of the Alaska Fisheries Science Center, Marine Mammal Laboratory, Seattle, WA. The NERTO aligns with NOAA CSC CESSRST-II’s goal for research on Coastal and Marine Habitat and Ecosystem Goods & Services. The research conducted supports NOAA’s mission by utilizing multidisciplinary tools to enhance the monitoring, understanding, and conservation of coastal and marine resources and habitats that are especially vulnerable to both natural and human-induced stressors. Through the NERTO Detection of Seals and Polar Bears in Multispectral Imagery project, the intern advanced mission-aligned research skills at NOAA. She developed new competencies in artificial intelligence/machine learning for object detection and classification, and employed remote sensing and computer vision methods to interrogate and validate multispectral datasets.