← Back to Sessions

Deep Clustering of Ambient Volcanic Seismicity: An Example at Erebus Volcano, Antarctica

Description: 

We present a novel approach for unsupervised deep clustering of different families of ambient seismic behaviors at Erebus volcano, Antarctica. As part of an NSF-CAIG initiative, we explore several multimodal and array flexible architectures with the goal of utilizing continuous multi-station seismic datasets to sort transient behaviors on multiple time scales into families in a clustering space. Variational autoencoder based architectures, in particular, show promise in terms of clustering robustness and family reproducibility when paired with convolutional layers and spectrogram inputs. For Erebus volcano, where multiple cataloged datasets exist including icequakes, Strombolian eruptions, teleseisms, iceberg tremor, and various flavors of ambient noise, such deep learning setups have already shown for example the ability to cleanly separate Strombolian eruptions from icequake activity, a task that typically requires a multi-step process involving STA/LTA detectors and multi-station matched filtering. We further show that inter-family separability, such as in the case of repeating icequake activity, can be seen in the subspace of clustered families, showcasing the flexibility and depth of these approaches. We anticipate eventual training and implementation of real-time networks with the goal of adding to a pool of next generation tools aiming to improve monitoring efforts of dynamic media.

Session: Data-driven and Computational Characterization of Non-earthquake Seismoacoustic Sources [Poster]

Type: Poster

Date: 4/16/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: Julien

Student Presenter: No

Invited Presentation: 

Poster Number: 70

Authors