Participants: Natalia Poiata, Mariano Supino, Cyril Journeau, Natalia Galina, Gaspard Farge, Jean-Pierre Vilotte

The main goal here is to analyze observations from active geological regions where STER-related seismic signals have been continuously recorded by dense seismic networks and during long periods of time. Below, the main datasets pre-selected for the analysis in our proposal are described.

– Japan represents an exceptional natural laboratory for the study of STER seismicity as a complex subduction with its in abundant seismicity. Systematic analysis of the Japanese observations is one of the cornerstone tasks of this project. The analysis of Japanese data is done in collaboration with Dr. Obara and his team, at the Earthquake Research Institute (ERI) in Tokyo.

– Kamchatka: in this region, the targeted STER seismicity is associated to various very active volcanoes. We are using the unique seismological dataset that was obtained through the collaboration with Russian colleagues and institutions that include records from permanent monitoring stations, and the results of the large-scale field seismological experiments KISS that have been organized in collaboration with Russian and German Scientists in 2015-2016.
KISS seismic experiment in Kamchatka

– Mexico: In particular, the Guerrero region is also an exceptional natural laboratory for studying tectonic tremorsx, because of its flat subduction geometry that results in a large and well-exposed frictional transition zone. The analysis of Mexican data is done in collaboration with the Dr. Vladimir Kostoglodov from UNAM, Mexico and with Dr. William Frank from the MIT, USA.

– Piton de la Fournaise on La Réunion Island is one of the most frequently active volcanoes in the world. We collaborate with the Piton de la Fournaise volcanic observatory that operates seismic and geodetic networks on this volcano.

Participants: Natalia Poiata, Mariano Supino, Cyril Journeau, Natalia Galina, Gaspard Farge, Francis Tong, Julien de Rosny, Gael Varoquaux, Jean-Pierre Vilotte

According to different STER seismicity regimes, the methods that we develop can be grouped as having two main goals. The first goal is to study the impulsive LFEs and the second goal is to analyze “fully mixed” tremors that may be considered as stochastic and non-stationary signals.

– Goal 1: exploring impulsive component of tremors, LFEs. To constrain STER processes at the ELFS scale, a first goal is to build large high-resolution LFE catalogs from the studied data sets including locations, waveform information, and source properties (magnitides, durations, mechanisms, etc). Distributions and scaling laws of these different parameters must be characterized in a probabilistic inference framework, e.g. with a Bayesian approach that might include machine learning and stochastic optimization. The main challenge is that classical earthquake methods fail with low signal-to-noise ratios LFEs. Therefore, we need to develop and to apply novel methods specifically adjusted for their analysis.

– Goal 2: statistical analysis of seismic tremor signals. A part their impulsive component (LFEs), tremors are difficult to detect, characterize, and locate because they appear as weak signals with no definite beginning, and they are almost buried in the noise. Therefore, we plan to develop and to apply advanced statistical methods for the analysis of tremor signals.

Main families of methods:

• Advanced data-driven time–frequency-scale approaches for STER signal analysis.
• Exploring waveform similarity for signal detection and classification.
• Advanced network-based approaches for signal analysis.
• Inferences from continuous seismic data based on pattern recognition and machine learning.

Participants: Véronique Dansereau, Gaspard Farge, Claude Jaupart, Jean-Pierre Vilotte, Edouadr Kaminski, Vladimir Lyakhovsky, Oleg Melnik

– Elementary low frequency sources (ELFS). A central problem for the understanding of STER related seismicity is establishing the basic model of the seismogenesis. So far, in many previous studies of tremors and LFEs, this problem was bypassed when tremors and/or LFE rates were simply and arbitrary considered as proxies to deformation/slip rates (for tectonic STER seismicity) or to fluid pressure (for volcanic STER seismicity). Such simplification is obviously not satisfactory and, therefore, we develop a physically consistent model of STER-related seismic sources compatible with observations of seismic signals.

– The interaction between ELFS that results in the time and space clustering is a fundamental phenomenon that controls the collective behavior of LFEs during structural reorganization of finite regions and the dynamics of tremors. The level of these interactions depends on forcing from the driving STER processes resulting in complex and nonlinear collective ELFS behavior during strongest tremor episodes. Mechanisms controlling the ELFS interactions are closely related with the physical processes generating LFEs and might include the elastic stress transfer, the stress transfer by slow deformation, and/or migration of fluid pressure.

– Statistical physical modeling of the STER activity. On the basis of observations accumulated during the past decades, a number of striking features of STER activity in slowly driven dissipative volcanic and fault systems are now emerging. Non-stationary, long-duration tremor signals might be a distinct signature of the cooperative reorganization of finite regions, associated often to spatial ELFS clustering at variable recurrence interval. From the high-resolution catalogues of tremors and ELFS, we intend to statistically analyze and model the STER activity during the slow evolution of volcanic and tectonic systems in different regions.

Participants: Geneviève Moguilny, Francis Tong

The main goal is to systematically analyze several large archives of geophysical data that have been collected over recent decades and to develop mechanical and statistical physics models for the STER sources and activity.

The main characteristics of the planned data analyses are: (1) large volumes of raw data (up to hundreds of terabytes); (2) possibly even larger volumes of intermediate data files, because of the cross-multiplication required in many algorithms, which results in power law growing; (3) very large amounts of basic and repetitive operations (e.g., Fourier transforms, dot products, and others), (4) complex high-end statistical data analysis (HDA) together with ensemble of high-performance simulations (HPC). This implies running application on data-intensive computational platforms (GRICAD at UGA and S-CAPAD at IPGP).