Machine learning models are increasingly applied across scientific disciplines, with deep-learning based pose estimators revolutionizing the fields of neuroscience and marine biology, allowing researchers to automate and enhance accuracy of behavioral analysis. While markerless pose estimators have transformed behavioral neuroscience, their effectiveness is limited by a lack of species- and domain-specific data, especially for marine invertebrates such as cephalopods and starfish. Due to their highly flexible body structures, starfish cannot be effectively represented by the rigid skeletal models commonly used for terrestrial vertebrates, making existing pose estimation techniques unreliable for tracking their movements. This project addresses this by developing a deep learning-based pose estimation model and archive database specific to cephalopods and starfish. Using DeepLabCut, we train a supervised machine learning model to track movement patterns in both naturalistic and laboratory settings. Our dataset, sourced from the Hodin lab in Friday Harbor, undergoes preprocessing with embedding and clustering algorithms to identify representative frames for model training. By establishing a reliable, quantitative framework for cephalopod behavior analysis, this product can enhance reproducibility and contribute to the development of standardized methodologies and definitions of behaviors in marine and neuroscience research. This tool would ease cross-lab collaboration and eliminate ambiguities when investigating cephalopod and starfish behavior.

Eelgrass is a foundational biome that provides critical habitat for numerous species, making its conservation vital. Specifically, sunflower stars (Pycnopodia helianthoides) use eelgrass as a nursery. In 2013, the sunflower star population crashed due to an unprecedented disease event creating a need to determine where the stars were historically to inform efforts in both eelgrass and sunflower star recovery. The Washington State Department of Natural Resources (WDNR) monitors eelgrass trends in the Salish Sea through the Submerged Vegetation Monitoring Project (SVMP). The SVMP video archive is roughly 6000 hours of footage spanning the Salish Sea in Washington state and dates back to 2000, providing a resource to observe the correlations between the stars and eelgrass. I created this research project centered around this connection to gain insight into the abundance of sunflower stars before and after the disease outbreak. To identify stars within the video archive, I sorted the footage into high-quality clips for sunflower star detection and discarded lower-quality ones due to the difficulty of confirming sightings. A computer vision model using hierarchical criteria was developed to assist in my annotations of video clips based on quality. In the high-quality clips, I also identified and annotated various organisms to understand if there are any further correlations with the sunflower star abundance. When sunflower stars were detected, I recorded their location and timestamp, creating a historical dataset. Once the annotations were completed, I made a comprehensive map of the detected sunflower star abundance and location over the SVMP video archive's time span. This project showcases the value of cross-year pattern analysis and camera quality normalization techniques. My annotations will eventually support the development of an automated video-cleaning system and a sunflower star detection model, enhancing the SVMP archive’s effectiveness in future conservation efforts.

Sunflower stars (Pycnopodia helianthoides) are the world’s largest sea stars and critical predators for habitat health. Sunflower stars historically dominated west coast benthic ecosystems, but in the last decade lost over 90% of its global population due to an epidemic of wasting disease. The complete extirpation of Sunflower Stars in many regions of the west – notably Northern California – has exposed kelp forests to overgrazing by urchins, leading to a loss in critical habitats for many marine organisms, increased coastline erosion due to wave action, and decreased atmospheric carbon sequestration. The beginning of restoration efforts are underway to restore populations of these endangered stars, including the first-ever sunflower star captive breeding program at Friday Harbor Labs, where our work was conducted. Despite their clear ecological importance, the surprisingly complex behaviors of sunflower stars has very little documentation in literature. In this experiment, we used an emerging technique called Motion Sequencing to measure juvenile stars’ responses to basic abiotic factors of light and temperature. We found that Sunflower Stars exhibit the most movement during periods of changing light, supporting the dominant theory. We also found they move more in higher temperatures, potentially hinting at resilience to climate change. In doing so, we hope to expand our understanding of sunflower star behaviors – such as their diurnal activity levels, and how they respond to shifts in temperature and other stressors, thus informing both ongoing and future conservation efforts.