Research Profile

Understanding wave propagation is the central issue for the interpretation of seismic records.  I studied extensively the regional propagation of short period waves using numerical simulation and data analysis.  I showed the frequency dependence of the classical quality factor of rocks and the importance of scattering.  I proposed maps of quality factors and discovered zones of extinction of guided waves across the mountain ranges of Alps and Pyrenees.  A method of simulation of wave propagation based on boundary integral equations was developed to investigate such complex geological structures.  This method was also applied to problems of seismic hazard assessment (effect of basin or topography) and of discrimination of earthquakes and explosions. 

The quantitative analysis of seismograms convinced me to turn my interests to the problem of scattering of seismic waves in the Earth.  This represents a significant change of perspective in my career since I had to investigate some problems that are traditionally more relevant to physics than to seismology.  We develop a model of coda decay based on the radiative transfer theory in which the traditional seismological parameter Qcoda is interpreted as a time of residence of diffuse waves in the Earth crust.  In modern physics, the concept of ‘mesoscopic’ regime opened a new field of investigation of a world of phenomena between microscopic and macroscopic behaviours.  We found that multiple scattering of seismic waves is relevant of this approach.  We demonstrated the existence of the phenomena of equipartion of energies in the seismic coda, and we showed that energy ratios are in agreement with theory.  We showed also the first demonstration of weak localisation of seismic waves, a typical mesoscopic effect in which diffuse average energy is exhibiting a behaviour associated with reciprocal properties of the wave field and interferences. 

An interesting consequence of this preservation of phase effect in supposedly random fields is the possibility of passive reconstruction of the response of the Earth between two distant points from the correlation of the field recorded at these points.   We demonstrated the pertinence of this concept with actual coda waves.  Then we applied the same approach to records obtained in absence of earthquake, that is with ambient noise records.  In this case, and in the period band we considered, the so-called seismic noise consists of the permanent agitation produced by the interaction between ocean waves and the solid Earth.  

Using long continuous noise records, we have reconstructed the Earth response between stations and use these virtual seismograms to extract surface waves and to produce 3D tomographic images of the Earth interior with an unprecedented resolution.

The development of this technique was a breakthrough, which rapidly led to numerous applications ranging from near-surface exploration to deep Earth structures such as the mantle-core boundary.  

We studied carefully the precision and stability of this novel type of data.  We found that the measurements can be remarkably stable.  We therefore proposed to use it for monitoring temporal changes at depth with repeated noise-based measurements.  We showed a new type of signal preceding the eruptions of a volcano.  The long-term work on random fields and mesoscopy led to new approaches to investigate the changing Earth. My present priority is to develop new seismological tools to monitor slight changes in fault zones that could be associated with the evolution of the tectonic stress.  This includes new imaging techniques based on the radiative transfer theory with, for the first time, the coupling of body and surface waves. 

In the last years, we studied the large slow slip events recently discovered in Mexico by using jointly GPS and seismic data. We developed techniques that allow for imaging the slip evolution during the events and for detecting low magnitude events. We have seen the triggering of slow slip by a distant mega-earthquake and the triggering of a large local event by a slow slip event. 

One of our findings is the intermittent nature of the large slow slip events that can be decomposed in a cascade of short transient slips by using simultaneously geodetic and seismological data.  This discovery challenges the present-day widely accepted models of friction. 

Recently we started a project for applying deep learning to data analysis. A chair in the newly installed Interdisciplinary Institute for Artificial Intelligence of UGA (MIAI) supports this program.   The goal of this project, based on multiscale representations of the signals, is to integrate geodetic and seismological data in the same classification.  Scatnet is an analysis tool developed as part of the ERC F-IMAGE and used for a number of applications.

We recently obtained positive results: predicting GPS time series from seismic data.  We are now reaching another of our goals: extract from raw seismograms a feature space in which it is possible to blindly separate the different sources of waves, but also to disentangle source and medium change effects. 

This opens up the possibility of extracting data directly from temporal variations in rock properties (due to meteorological forcing or deep deformation) and from changes in weak signal sources at depth.