Hidden pressurized fluids prior to the 2014 phreatic eruption at Mt Ontake.

A large fraction of volcanic eruptions does not expel magma at the surface. Such an eruption occurred at Mt Ontake in 2014, claiming the life of at least 58 hikers in what became the worst volcanic disaster in Japan in almost a century. Tens of scientific studies attempted to identify a precursor and to unravel the processes at work but overall remain inconclusive. By taking advantage of continuous seismic recordings, we uncover an intriguing sequence of correlated seismic velocity and volumetric strain changes starting 5 months before the eruption; a period previously considered as completely quiescent. We use various novel approaches such as covariance matrix eigenvalues distribution, cutting-edge deep-learning models, and ascribe such velocity pattern as reflecting critically stressed conditions in the upper portions of the volcano. These, in turn, later triggered detectable deformation and earthquakes. Our results shed light onto previously undetected pressurized fluids using stations located above the volcano-hydrothermal system and hold great potential for monitoring.

AI-Based Unmixing of Medium and Source Signatures From Seismograms: Ground Freezing Patterns.

Seismograms always result from mixing many sources and medium changes that are complex to disentangle, witnessing many physical phenomena within the Earth. With artificial intelligence (AI), we isolate the signature of surface freezing and thawing in continuous seismograms recorded in a noisy urban environment. We perform a hierarchical clustering of the seismograms and identify a pattern that correlates with ground frost periods. We further investigate the fingerprint of this pattern and use it to track the continuous medium change with high accuracy and resolution in time. Our method isolates the effect of the ground frost and describes how it affects the horizontal wavefield. Our findings show how AI-based strategies can help to identify and understand hidden patterns within seismic data caused either by medium or source changes.

Hierarchical Exploration of Continuous Seismograms With Unsupervised Learning.

Continuous seismograms contain a wealth of information with a large variety of signals with different origin. Identifying these signals is a crucial step in understanding physical geological objects. We propose a strategy to identify classes of signals in continuous single-station seismograms in an unsupervised fashion. Our strategy relies on extracting meaningful waveform features based on a deep scattering network combined with an independent component analysis. Based on the extracted features, agglomerative clustering then groups these waveforms in a hierarchical fashion and reveals the process of clustering in a dendrogram. We use the dendrogram to explore the seismic data and identify different classes of signals. To test our strategy, we investigate a two-day-long seismogram collected in the vicinity of the North Anatolian Fault, Turkey. We analyze the automatically inferred clusters' occurrence rate, spectral characteristics, cluster size, and waveform and envelope characteristics. At a low level in the cluster hierarchy, we obtain three clusters related to anthropogenic and ambient seismic noise and one cluster related to earthquake activity. At a high level in the cluster hierarchy, we identify a seismic burst that includes around 200 events with similar waveforms and high-frequent signals with correlating envelopes and an anthropogenic origin. The application shows that the cluster hierarchy helps to identify particular families of signals and to extract subclusters for further analysis. This is valuable when certain types of signals, such as earthquakes, are under-represented in the data. The proposed method may also successfully discover new types of signals since it is entirely data-driven.

Seismic tremor reveals active trans-crustal magmatic system beneath Kamchatka volcanoes.

The occurrence and the style of volcanic eruptions are largely controlled by the ways in which magma is stored and transported from the mantle to the surface through the crust. Nevertheless, our understanding of the deep roots of volcano-magmatic systems remains very limited. Here, we use the sources of seismovolcanic tremor to delineate the active part of the magmatic system beneath the Klyuchevskoy Volcanic Group in Kamchatka, Russia. The tremor sources are distributed in a wide spatial region over the whole range of crustal depths connecting different volcanoes of the group. The tremor activity is characterized by rapid vertical and lateral migrations explained by fast pressure transients and dynamic permeability. Our results support the conceptual model of extended and highly dynamic trans-crustal magmatic systems.

Clustering earthquake signals and background noises in continuous seismic data with unsupervised deep learning.

The continuously growing amount of seismic data collected worldwide is outpacing our abilities for analysis, since to date, such datasets have been analyzed in a human-expert-intensive, supervised fashion. Moreover, analyses that are conducted can be strongly biased by the standard models employed by seismologists. In response to both of these challenges, we develop a new unsupervised machine learning framework for detecting and clustering seismic signals in continuous seismic records. Our approach combines a deep scattering network and a Gaussian mixture model to cluster seismic signal segments and detect novel structures. To illustrate the power of the framework, we analyze seismic data acquired during the June 2017 Nuugaatsiaq, Greenland landslide. We demonstrate the blind detection and recovery of the repeating precursory seismicity that was recorded before the main landslide rupture, which suggests that our approach could lead to more informative forecasting of the seismic activity in seismogenic areas.