Clément Violet @clementviolet.bsky.social · 12/05/2026Thank you @aec13.bsky.social for this highligth! Check out the paper, the data and the app! 🤩 010
Clément Violet @clementviolet.bsky.social · 19/03/2026Sure thing! All the data are on Zenodo: zenodo.org/records/1097... If you need any help navigating the different files, feel free to DM mezenodo.orgLeveraging Citizen Science to Classify and track Benthic Habitat States: an Unsupervised UMAP-HDBSCAN Pipeline applied to the Global Reef Life Survey Dataset 000
Clément Violet @clementviolet.bsky.social · 13/02/2025A big thank also to you, all the Reef Life Survey volunteers, without whom this work would not have been possible! 000
Clément Violet @clementviolet.bsky.social · 13/02/2025I would like to warmly thank all the co-authors for their invaluable help with this @aurelien-boye.bsky.social, Rick Stuart-Smith, Graham J. Edgar, Elizabeth Oh, Stanislas Dubois and Martin Marzloff. 110
Clément Violet @clementviolet.bsky.social · 13/02/2025🌍 Understanding habitat states & transitions is key for conservation! 🏝️ Helps track ecosystem shifts from climate change 🛑 Detects early warning signs of degradation (e.g., kelp loss → urchin barrens). 112
Clément Violet @clementviolet.bsky.social · 13/02/2025🌍 With our data-driven standardized way to classify & monitor marine habitats, what did we find? 🏝️ 17 distinct habitat states, including kelp forests, coral reefs & transitional zones 📉 Our pipeline captures fine-scale ecological changes over time 132
Clément Violet @clementviolet.bsky.social · 13/02/2025This study wouldn’t be possible without citizen scientists! 🌍🌊 👩🔬 Reef Life Survey divers collected 6554 transects worldwide 📸 Standardized photoquadrat sampling ensured consistency 📊 It enabled us to track habitat transitions over time 141
Clément Violet @clementviolet.bsky.social · 13/02/2025Machine learning models like UMAP & HDBSCAN are powerful, but they’re often hard to interpret. So, we used SHAP to: 📌 Identify which features drive cluster formation 📌 Uncover non-linear interactions in benthic ecosystems 📌 Improve ecological interpretability of results 111
Clément Violet @clementviolet.bsky.social · 13/02/2025We used better methods: UMAP (for dimension reduction) & HDBSCAN (for clustering). ✅ Captures complex ecological structures ✅ Finds clusters of varying shapes & sizes ✅ Filters noise instead of forcing data into clusters 111
Clément Violet @clementviolet.bsky.social · 13/02/2025Clustering is widely used in ecology to define community types, habitat states & biodiversity patterns. But common methods used in ecology have major drawbacks. (See this excellent video of @JohnDataHealy youtu.be/dGsxd67IFiU?...)youtu.beHDBSCAN, Fast Density Based Clustering, the How and the Why - John HealyYouTube video by PyData 111
Clément Violet @clementviolet.bsky.social · 13/02/2025🚀 New paper alert! We used a UMAP-HDBSCAN pipeline to classify & track benthic habitat states at a global scale using citizen science data from Reef Life Survey ! 🌍🏝️ 🔗 Read it here: doi.org/10.1016/j.ec... #MarineScience #Ecology #CitizenScience #MachineLearningdoi.orgRedirecting 161
Reposted by Clément VioletAmelia Curd @aec13.bsky.social · 11/02/2025Great talk @clementviolet.bsky.social given during the #BIOcean5D general assembly @ICM-CSIC! So much work went into looking up all those marine NIS native origins and building the ✨ Shiny app.Can’t wait to see the future #Rpackage #marineecology #ultraviolet 🧪🦀🐚🌱🪸 @ifremer.bsky.social 051
