August 21st - Professor Fidel Costa and MSc student Matthieu Courcelles (visiting researchers from IPGP, France)
They will each speak for approximately 15 minutes on two different topics, followed by time for discussion.
Fidel Costa - Title: A quantitative framework for exploring geochemical relationships between volcanic systems at the global scale
Matthieu Courcelles - Title: Textural analysis and unsupervised classification of zircon crystals
Time: 12:30
Place: Askja, 3rd floor meeting room (room 367) and streamed virtually via Zoom:
Fidel Costa (Universite Paris Cite - IPGP)
Title: A quantitative framework for exploring geochemical relationships between volcanic systems at the global scale
Abstract
Volcanoes show a wide variety of magma compositions, tectonic settings, and eruptive behaviours, making it difficult to quantify the relationships among volcanic systems at a global scale. To address this challenge, we introduce DashVolcano, a web-based platform (https://dashvolcano.ipgp.fr) aimed to develop a quantitative framework for exploring relationships between volcanoes using whole-rock major oxides compositions. DashVolcano includes > 450,000 samples and 1,229 volcanoes worldwide. To quantify the relationship between volcanoes the compositional distributions of each volcano are computed using the Aitchison distance, and comparison between volcanoes is done using the Energy Distance. With this approach we are able to identify the volcanoes that are compositionally closer to any given chosen volcano of interest. Preliminary results will be illustrated with examples of volcanoes in Iceland.
Matthieu Courcelles (Universite Paris Cite - IPGP)
Title: Textural analysis and unsupervised classification of zircon crystals
Abstract
Zircon crystals record the timing and dynamics of magmatic processes, but their complex internal zoning remains difficult to classify visually. We introduce a reproducible workflow combining superpixel segmentation, radial grayscale-intensity representations, Wasserstein distances, and hierarchical clustering. Tests on a controlled synthetic dataset correctly recover predefined crystal families and distinguish similar grayscale distributions with different zoning patterns. The workflow is then applied to 82 natural zircons from three main eruptive periods of Mount Pinatubo. We found that the early Inararo (81 ka) and Crow (5 ka) VEI 6-7 eruptions have their own distinct zircon populations, whereas the latest 1991 VEI6 eruption recycled the crystals from the previous events and lack it’s own distinct zircon population. This approach enables rapid quantitative comparison of large crystal populations and supports geological hypothesis generation from complex zoning patterns.
All are welcome.