Clément Violet
@clementviolet.bsky.social
PhD Marine Ecology | Postdoc on Marine Non Indigenous Species | IFREMER 🐬🇫🇷
A big thank also to you, all the Reef Life Survey
volunteers, without whom this work would not have been possible!
volunteers, without whom this work would not have been possible!
February 13, 2025 at 10:00 AM
A big thank also to you, all the Reef Life Survey
volunteers, without whom this work would not have been possible!
volunteers, without whom this work would not have been possible!
I 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.
February 13, 2025 at 10:00 AM
I 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.
🌍 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).
🏝️ Helps track ecosystem shifts from climate change
🛑 Detects early warning signs of degradation (e.g., kelp loss → urchin barrens).
February 13, 2025 at 10:00 AM
🌍 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).
🏝️ Helps track ecosystem shifts from climate change
🛑 Detects early warning signs of degradation (e.g., kelp loss → urchin barrens).
🌍 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
🏝️ 17 distinct habitat states, including kelp forests, coral reefs & transitional zones
📉 Our pipeline captures fine-scale ecological changes over time
February 13, 2025 at 10:00 AM
🌍 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
🏝️ 17 distinct habitat states, including kelp forests, coral reefs & transitional zones
📉 Our pipeline captures fine-scale ecological changes over time
This 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
👩🔬 Reef Life Survey divers collected 6554 transects worldwide
📸 Standardized photoquadrat sampling ensured consistency
📊 It enabled us to track habitat transitions over time
February 13, 2025 at 10:00 AM
This 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
👩🔬 Reef Life Survey divers collected 6554 transects worldwide
📸 Standardized photoquadrat sampling ensured consistency
📊 It enabled us to track habitat transitions over time
Machine 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
So, we used SHAP to:
📌 Identify which features drive cluster formation
📌 Uncover non-linear interactions in benthic ecosystems
📌 Improve ecological interpretability of results
February 13, 2025 at 10:00 AM
Machine 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
So, we used SHAP to:
📌 Identify which features drive cluster formation
📌 Uncover non-linear interactions in benthic ecosystems
📌 Improve ecological interpretability of results
We 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
✅ Captures complex ecological structures
✅ Finds clusters of varying shapes & sizes
✅ Filters noise instead of forcing data into clusters
February 13, 2025 at 10:00 AM
We 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
✅ Captures complex ecological structures
✅ Finds clusters of varying shapes & sizes
✅ Filters noise instead of forcing data into clusters
Clustering 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.be/dGsxd67IFiU?...)
HDBSCAN, Fast Density Based Clustering, the How and the Why - John Healy
YouTube video by PyData
youtu.be
February 13, 2025 at 10:00 AM
Clustering 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.be/dGsxd67IFiU?...)