Arctic | Mammals
Leveraging deep learning to monitor Pacific walrus from satellite imagery Presenter: Casey Youngflesh , ccy@msu.edu, Michigan State University Christian Che-Castaldo , christian.che-castaldo@stonybrook.edu, Stony Brook University David Douglas , ddouglas@usgs.gov, U.S. Geological Survey Alaska Science Center
The Pacific walrus ( Odobenus rosmarus divergens ) is an important benthic predator in Arctic marine ecosystems and is essential for the cultural and economic survival of Indigenous coastal communities in the Pacific Arctic. As a result of increasing pressures associated with global climate change, including rising temperatures and rapidly declining sea ice, the future status of this species is uncertain. Robust, range-wide monitoring efforts are needed to quantify responses to these rapid changes and to enact appropriate management action. Monitoring Pacific walrus at large temporal scales across their range in the Bering and Chukchi Seas, however, is a challenging task. Satellite-based sensors provide a means by which to monitor walrus haulouts while avoiding costly and potentially dangerous field data collection efforts. Given the large size and gregarious nature of walrus, coastal walrus haulouts can be identified by human observers in publicly- available satellite imagery. Manual classification of this imagery at a range-wide scale, over multiple years, however, is impractical. Using advances from the field of computer vision, we developed a framework to automatically detect walrus haulouts in 10-m resolution multi-spectral Sentinel-2 satellite imagery. To accomplish this, we trained a deep learning- based convolutional neural network using manually annotated satellite imagery from across the Pacific walrus’ range. Our model successfully identified walrus in a held-out testing dataset with robust performance, particularly for larger haulouts. This approach will be applied to satellite imagery from across the Pacific walrus’ range to characterize how the spatiotemporal distribution of walrus haulouts is responding to changing conditions in the Pacific Arctic. Ultimately, these efforts will serve as the groundwork for a range-wide, automated, satellite-enabled monitoring effort, providing stakeholders with the information necessary to implement timely management action.
Alaska Marine Science Symposium 2023 272
Made with FlippingBook flipbook maker