Serge Sambolian

Tomographie des pentes à cohérence cinématique fondée sur des solveurs eikonals et la méthode de l’état adjoint : Théorie et applications à la construction de modèles de vitesses et localisation d'événements: La construction du modèle de vitesse est crucial en imagerie sismique puisqu'elle contrôle la précision avec laquelle des méthodes d'imagerie haute résolution telles que la migration ou l'inversion des formes d'ondes complètes (FWI) peuvent imager le sous-sol. La stéréotomographie, une méthode de tomographie des pentes qui tire efficacement profit de la densité des données sismiques modernes, a été proposée comme une alternative aux approches classiques de tomographie en réflexion fondées sur le pointé d'horizons continus dans le volume sismique. La stéréotomographie est en revanche fondée sur le pointé semi-automatique d’événements localement cohérents, paramétrés par le temps double et les pentes aux sources et récepteurs et liés à des diffractants dans le sous-sol. Plus récemment, une variante de la stéréotomographie a été proposée en remplaçant le tracé de rai par un solveur eikonal dans le problème direct et l'inversion de la matrice des dérivés de Fréchet par la méthode de l'état adjoint dans le problème inverse. Cette nouvelle approche est massivement parallèle et de ce fait adaptée à des applications de grande dimension. Néanmoins, et de manière comparable à l'approche initiale, la position des diffractants et les paramètres du milieu sont conjointement mis à jour.Durant cette thèse, j'ai proposé une nouvelle formulation de la stéréotomographie qui gère plus efficacement le couplage vitesse-profondeur, inhérent aux approches en réflexion. Via une migration cinématique, je résous le problème de localisation et le projette dans le sous-problème principal de l'estimation des vitesses. Cette projection de variable garantit la consistance cinématique entre les deux classes de variables, consistance qui n'est pas garantie quand les deux classes de variables sont mis à jour conjointement. Par ailleurs, la projection de variable induit une paramétrisation compacte du problème inverse où une classe d'observables, en l’occurrence une pente, est utilisée pour mettre à jour une classe de paramètres, les vitesses. Je développe cette approche avec un solveur eikonal et la méthode de l'état adjoint pour des milieux TTI. Son évaluation sur deux cas d'étude synthétiques et réel confirme sa meilleur résilience au modèle initial et une vitesse de convergence plus rapide que l'approche conjointe.La stéréotomographie est principalement utilisée pour des dispositifs de sismique réflexion (flûte sismique) pour lesquels les sources et les récepteurs sont finement échantillonnés. Pour exploiter des dispositifs modernes à forts déports, j'ai introduit dans l'inversion les premières arrivées issues indifféremment de dispositifs de sismique réflexion multitrace ou de sismique grand-angle (OBN, OBC, terrestre). Dans un premier temps, j'ai illustré l'apport des pentes dans la tomographie des temps des premières arrivées (FATT) pour réduire l’ambiguïté temps-profondeur avec un cas synthétique et un cas réel sur la zone de subduction de Nankai. J'ai aussi évalué la tomographie des pentes en première arrivée pour construire un modèle initial pour la FWI avec un modèle complexe représentatif du Golfe du Mexique où la présence de sel génère de forts contrastes. J'ai pu illustrer la capacité de ma méthode à reconstruire les corps de sel tout en notant les difficultés héritées de l'éclairage incomplet de la zone située sous le sel. Cela m'a incité à combiner des données de sismique réflexion et des données grand-angle pour effectuer l'inversion conjointe des pentes et des temps de trajet des premières arrivées et des arrivées réfléchies pour bénéficier d'un éclairage angulaire optimal du milieu illustrée par des applications sur la zone de Nankai.Finalement, j'ai étendu l'utilisation de ma méthode par projection de variable pour localiser l'hypocentre des séismes en utilisant l'estimation des vitesses et des temps origine comme proxys. Cette approche originale a été validée avec deux exemples synthétiques.

SUMMARY: This study focuses on the hydrophone component of a dense ocean bottom cable data set from the North Sea. This data had already been used in the past to illustrate the high resolution power of full waveform inversion based strategies. We have developed a highly scalable implementation of a visco-acoustic full waveform inversion engine making it possible to double the frequency content of the inverted data compared to previous studies, using simultaneously up to almost 50 000 CPU. This results in a multiparameter reconstruction of the subsurface, where the P-wave velocity, the density and the quality factor are reliably reconstructed down to 2 km depth, with a resolution of about 10 m.

Extension strategies for full-waveform inversion (FWI) rely on introducing additional degrees of freedom to the FWI problem, which expands the search space. This search space extension helps by relaxing the nonconvexity of the problem and thereby alleviating the cycle-skipping issue. The receiver-based extension strategy introduces the receiver position as the additional degree of freedom to FWI to improve the fit between the observed and calculated data at early iterations. This helps circumvent the cycle-skipping phenomenon. In this study, we make this receiver position time dependent, meaning that the receiver positions vary as a function of the acquisition time. The resulting mathematical problem is a two-nested-loop minimization, where the outer loop is the conventional FWI loop to update the subsurface mechanical parameters and the inner loop aims at finding the optimal time-dependent virtual receiver positions. This inner-loop problem is heavily nonlinear and nonconvex. Finding the global minimum is therefore a challenging task. To do so, we use a computational intelligence technique, particle swarm optimization (PSO). PSO makes it possible to thoroughly explore the search space with few iterations. Numerical experiments using a North Sea exploration 2D synthetic model, starting from crude initial models, illustrate that the method is robust and very easy to tune.

SUMMARY: Modelling seismic wavefields in complex 3-D elastic media is the key in many fields of Earth Science: seismology, seismic imaging, seismic hazard assessment and earthquake source mechanism reconstruction. This modelling operation can incur significant computational cost, and its accuracy depends on the ability to take into account the scales of the subsurface heterogeneities varying. The theory of homogenization describes how the small-scale heterogeneities interact with the seismic waves and allows to upscale elastic media consistently with the wave equation. In this study, an efficient and scalable numerical homogenization tool is developed, relying on the similarity between the equations describing the propagation of elastic waves and the homogenization process. By exploiting the optimized implementation of an elastic modelling kernel based on a spectral-element discretization and domain decomposition, a fully scalable homogenization process, working directly on the spectral-element mesh, is presented. Numerical experiments on the entire SEAM II foothill model and a 3-D version of the Marmousi II model illustrate the efficiency and flexibility of this approach. A reduction of two orders of magnitude in terms of absolute computational cost is observed on the elastic wave modelling of the entire SEAM II model at a controlled accuracy.

SUMMARY: First-arrival traveltime tomography is one of the most used velocity model building techniques especially in sparse wide-angle acquisitions for deep crustal seismic imaging cases. Relying on the inversion of a picked attribute, the absolute traveltimes, the approach is ill-posed in terms of non-uniqueness of the solution. The latter is remedied by proper regularization or the introduction of prior information. Indeed, since traveltime kernels are vulnerable to the velocity–depth ambiguity, the inversion is stabilized by the introduction of complementary data like reflections and explicit reflectors in the velocity models. Here, we propose to supplement first-arrival traveltimes by their slopes, in other words the horizontal component of the slowness vectors at the sources and/or receivers. Slopes are a crucial attribute in state of the art scattering-based or reflection-based tomographic methods like slope tomography or wavefront tomography where the differential information is needed in order to locate the scattering events position or to parametrize the wavefront. The optional but valuable injection of slopes as an objective measure in first-arrival traveltime tomography stabilizes the problem by constraining the emergence angle or in turn implicitly the turning point depth of the rays. We explain why slopes have a tremendous added value in such a tomographic problem and highlight its remedial effect in cases where the medium is unevenly illuminated. We also show that the contribution of slopes become even more significant when the acquisition is sparse as it is generally the case with ocean-bottom seismometer surveys. The inferred models from such an extended time-attributes tomography will be used as initial guesses in a full-waveform inversion workflow context. The proposed strategy is benchmarked in 2-D media against a dip section of the SEG/EAGE overthrust model and then followed by a revisit of ocean bottom seismometers data from the eastern-Nankai subduction margin as a real deep crustal case study.

The use of distributed acoustic sensing (DAS) presents unique advantages for earthquake monitoring compared with standard seismic networks: spatially dense measurements adapted for harsh environments and designed for remote operation. However, the ability to determine earthquake source parameters using DAS is yet to be fully established. In particular, resolving the magnitude and stress drop is a fundamental objective for seismic monitoring and earthquake early warning. To apply existing methods for source parameter estimation to DAS signals, they must first be converted from strain to ground motions. This conversion can be achieved using the waves' apparent phase velocity, which varies for different seismic phases ranging from fast body waves to slow surface and scattered waves. To facilitate this conversion and improve its reliability, an algorithm for slowness determination is presented, based on the local slant-stack transform. This approach yields a unique slowness value at each time instance of a DAS time series. The ability to convert strain-rate signals to ground accelerations is validated using simulated data and applied to several earthquakes recorded by dark fibers of three ocean-bottom telecommunication cables in the Mediterranean Sea. The conversion emphasizes fast body waves compared to slow scattered waves and ambient noise and is robust even in the presence of correlated noise and varying wave propagation directions. Good agreement is found between source parameters determined using converted DAS waveforms and on-land seismometers for both P and S wave records. The demonstrated ability to resolve source parameters using P waves on horizontal ocean-bottom fibers is key for the implementation of DAS-based earthquake early warning, which will significantly improve hazard mitigation capabilities for offshore earthquakes, including those capable of generating tsunami.

The use of Distributed Acoustic Sensing (DAS) presents unique advantages for earthquake monitoring compared with standard seismic networks: spatially dense measurements adapted for harsh environments and designed for remote operation. However, the ability to determine earthquake source parameters using DAS is yet to be fully established. In particular, resolving the magnitude and stress drop, is a fundamental objective for seismic monitoring and earthquake early warning. To apply existing methods for source parameter estimation to DAS signals, they must first be converted from strain to ground motions. This conversion can be achieved using the waves' apparent phase velocity, which varies for different seismic phases ranging from fast body-waves to slow surface- and scattered-waves. To facilitate this conversion and improve its reliability, an algorithm for slowness determination is presented, based on the local slant-stack transform. This approach yields a unique slowness value at each time instance of a DAS time-series. The ability to convert strain-rate signals to ground accelerations is validated using simulated data and applied to several earthquakes recorded by dark fibers of three ocean-bottom telecommunication cables in the Mediterranean Sea. The conversion emphasizes fast body-waves compared to slow scattered-waves and ambient noise, and is robust even in the presence of correlated noise and varying wave propagation directions. Good agreement is found between source parameters determined using converted DAS waveforms and on-land seismometers for both P- and S-wave records. The demonstrated ability to resolve source parameters using P-waves on horizontal ocean-bottom fibers is key for the implementation of DAS based earthquake early warning, which will significantly improve hazard mitigation capabilities for offshore and tsunami earthquakes.

SUMMARY: We revisit the hypocentre–velocity problem, which is of interest in different fields as for example microseismics and seismology. We develop a formulation based on kinematic migration of two picked kinematic attributes in the 2-D case, the traveltime and the slope (horizontal component of the slowness vector), from which we are able to retrieve the location and subsequently the origin time correction and the subsurface parameters mainly velocity. We show how, through a variable projection, the optimization problem boils down to a physically consistent and parsimonious form where the location estimation is projected into the subsurface parameter problem. We present in this study a proof of concept validated by a toy test in two dimensions and a synthetic case study on the Marmousi model. The method presented in this study is extendible to three dimensions by incorporating the crossline slope or the backazimuth as a supplementary attribute.

The first leg of the SEFASILS cruise took place in November 2018 onboard the RV Pourquoi-Pas ? Up-to-date multichannel and wide-angle seismic data were acquired offshore Monaco, from margin to basin, aiming at providing a renewed vision of the complex North Ligurian backarc system. The compressive and extensive tectonic phases that have closely alternated in time and space over the last 45 My have yielded fairly contrasting structures, whose understanding is rendered even more challenging by the strong overprint of the Messinian salt tectonics. There is ample evidence of a compressive reactivation of the North Ligurian margin since 5 Ma at least, especially to the East, along the Gulf of Genoa. Such deformation is associated with some notable seismicity originating from faults and mechanisms that remain poorly apprehended. Yet, this seismicity peaked at one historical Mw ~6.6-6.9 destructive event (1887 Ligurian earthquake). The main objective of the SEAFASILS effort is a better characterization of the crustal structures, and chiefly, of the active crustal faults and their potential interplay with salt tectonics beneath the margin and the northernmost part of the basin, both featuring seismicity. Linking these aspects with broader-scale lithospheric processes within the Southern Alps/Northern Apennines, addressed in the AlpArray initiative, is also of great importance. Here we present preliminary results of these seismic investigations, with time and prestack depth migrated MCS data. Emphasis was put on the construction of some suitable velocity models to get optimal focusing of structures from surface to depth. Some active crustal tomographic velocities derived from the dense OBS deployment providing complementary insight will also be presented.