16:00 〜 16:15
[SSS06-09] Unsupervised exploration of seismic activity at Mount Fuji, Japan
キーワード：Mount Fuji, Seismic activity, Low-Frequency Earthquakes (LFEs), Automated event detection, CovSeisNet, UMAP
Mount Fuji, located approximately 100 km from Tokyo, poses a direct threat to over 30 million people. Its last eruption occurred in 1707, and it has remained dormant ever since. Today, seismic activity—particularly Low-Frequency Earthquakes (LFEs)—serves as the primary indicator of subsurface processes, often linked to fluid movement. However, these signals are still difficult to detect and to understand.
Our objective is to develop an automatic method for detecting and classifying LFEs, along with other seismic events at Mount Fuji, using continuous seismic records from 2008 across 11 stations. To achieve this and effectively analyze this large dataset, we use the CovSeisNet software to detect events by analyzing wavefield coherence, derived from the network covariance matrix width. Over a year of continuous data, wavefield coherence reveals distinct patterns corresponding to different event types, including LFEs and tectonic earthquakes. To aid interpretation, we apply the UMAP manifold learning algorithm to reduce the dimensionality of coherence patterns into a two-dimensional space. We refer to this low-dimensional representation as a "coherence atlas," where each point represents a time window of seismic data, grouped by similarity.
This automated approach enables not only the detection and classification of seismic events—validated against the Japan Meteorological Agency catalog—but also the identification of previously unrecorded events and the definition of new event classes. By autonomously mapping and classifying seismic activity beneath Mount Fuji, this method provides unprecedented insights into its activity and reveals events that had remained hidden in manually curated catalogs.
Our objective is to develop an automatic method for detecting and classifying LFEs, along with other seismic events at Mount Fuji, using continuous seismic records from 2008 across 11 stations. To achieve this and effectively analyze this large dataset, we use the CovSeisNet software to detect events by analyzing wavefield coherence, derived from the network covariance matrix width. Over a year of continuous data, wavefield coherence reveals distinct patterns corresponding to different event types, including LFEs and tectonic earthquakes. To aid interpretation, we apply the UMAP manifold learning algorithm to reduce the dimensionality of coherence patterns into a two-dimensional space. We refer to this low-dimensional representation as a "coherence atlas," where each point represents a time window of seismic data, grouped by similarity.
This automated approach enables not only the detection and classification of seismic events—validated against the Japan Meteorological Agency catalog—but also the identification of previously unrecorded events and the definition of new event classes. By autonomously mapping and classifying seismic activity beneath Mount Fuji, this method provides unprecedented insights into its activity and reveals events that had remained hidden in manually curated catalogs.