Marine acoustics and seismic imaging
The research carried out in this theme is marked by a dual acoustic/geophysical culture. It aims to remove current and future scientific barriers in the environmental field (oceans and land), using high-performance computing with the SPECFEM code we are developing, as well as real data. The aim is to provide a detailed understanding and realistic modeling of wave propagation in natural environments, and to optimize the imaging of these environments. Applications mainly concern the impact of anthropogenic underwater noise pollution on marine ecosystems and, to a lesser extent, seismic exploration.
Permanent members : P. Cristini (CR), E. Debieu (IE), N. Favretto-Cristini (DR), R. Guillermin (IR), V. Monteiller (IR), S. Rakotonarivo (MCF AMU)
Seismo-acoustic wave modeling
The aim is to meet current and future challenges in predicting the impact of anthropogenic activities in coastal marine environments in terms of (i) underwater noise pollution, (ii) impact on the marine ecosystem, and (iii) induced geohazards (in particular, seismic risks at the coast), with a particular focus on the counter-mining neutralization of explosive devices (European MSCA Doctoral Network SEASOUNDS Project, 2024-27, https://seasounds-dn. cnrs.fr/; European EuroHPC ChEESE-2P project, 2023-26, https://cheese2.eu/) and offshore wind turbine installation work, such as pile driving (European MSCA Doctoral Network BETTER project, 2025-29).
To achieve this goal, we are developing a reliable prediction tool, based on 3D numerical modeling capable of accurately reproducing the propagation of low-frequency waves (typically a few Hertz to a few hundred Hertz) within large-scale shallow water environments, by integrating the underlying 1st-order physical phenomena. This modeling is driven by wave physics, which in turn is conditioned by the properties of the marine environment. Given the size of the marine zones generally considered, it also relies on very substantial computing resources. In contrast to other works published in the literature, our models take into account exchanges between acoustic propagation in the water layer and seismic propagation in the subsoil (notably through interface waves), as well as subsoil vibrations induced by anthropogenic activities. This is particularly important for assessing, for example, the impact of noise and vibrations on benthic marine species. The models developed also enable us to better understand and analyze the wave fields recorded on OBS or seabed seismometers.
This work is carried out in collaboration with various French (e.g., Shom, ENSTA Bretagne, Cerema, Marée SAS, Quiet Oceans) and international (e.g., UPC Spain, NTNU Norway, Aarhus Univ. Denmark, TU Delft Netherlands) laboratories, institutions and companies, as part of national (RAPID DGA MAESTRIA) and European (SEASOUNDS, ChEESE-2P, BETTER) projects.
3D HPC numerical modeling of the seismo-acoustic propagation generated by an underwater explosion in the Rade d'Hyères (France).
Development of methodologies adapted to 2D/3D Exascale HPC simulations
In parallel with studies aimed at optimizing the physics of wave propagation, we are working, as part of the European ChEESE-2P project, on the porting and design of specific workflows adapted to future European exascale machines. A reworking of the SPECFEM code is underway to facilitate these aspects. These developments have a major impact on the modeling and inversion capabilities of SPECFEM software, enabling, for example, the implementation of 3D numerical simulations of seismo-acoustic wave propagation that are closer to the reality of experiments at sea.
Inversion & Imaging
Acoustic imaging using waveform inversion aims to reconstruct certain physical parameters of a medium. The principle is to minimize the difference between data and numerically simulated recordings. To achieve this, numerical optimization algorithms are used, but these are expensive in terms of computing costs. This calls for intensive computing. The current trend towards exascale computing capacities means that we can envisage using more efficient optimization methods (Newton's method), but this requires the development of more complex algorithms and workflows. One of the aims of the European ChEESE-2P project is to implement such algorithms with the SPECFEM3D code, and to demonstrate the feasibility of such a workflow on European pre-exascale machines. The aim here is to take second derivatives into account in the optimization algorithm and then to realize a seismological application with data recorded by European seismological networks.
Alternatively, methods based on “experimental” models (data-based modeling) are developed to characterize the marine environment (water column, marine sediments) or underwater objects (mines, submarines) when a priori information on the environment is insufficient and does not allow the use of a numerical or analytical model for inversion. This type of approach requires the use of large sensor networks.
This research leads to instrumental and experimental developments for the implementation of large sensor networks in laboratory conditions. In addition, in order to predict the acoustic signature or parameters of the inspected environment in operational or in situ conditions, this work is based on the development of innovative antenna processing methods based on near-field measurements (e.g. acoustic holography) or far-field measurements (e.g. adapted filtering). The use of large sensor arrays is proving to be a difficult task in terms of experimental implementation, so there is a trend towards the use of parsimonious sensor arrays to characterize the environment. In this context, our work aims to develop characterization methods that optimize the ratio between partial a priori knowledge of the environment and parsimonious measurements of the environment. One of the approaches being studied is the implementation of physics-based deep learning approaches to tackle this problem.
This work is the subject of several national (UTC, Naval Group, ENSTA Bretagne) and international (Scripps Inst. Oceanograophy USA, NRL USA) collaborations.
This research also has applications in non-destructive testing and structural health monitoring. The outlook for these research activities focuses on the experimental implementation of such a methodology, as well as the coupling of the holography methods developed with physics-based artificial intelligence approaches.
*
Acoustic holography device comprising 200 mems microphone sensors for characterizing a cylinder in an airborne environment