The processes associated with magma injection at shallow depths within the crust have been the topic of many geophysical studies, some investigating the seismicity that accompanies volcanic activity and others attempting to map the subsurface extent and geometry of the resulting magma bodies. The aim of this study is to obtain a better understanding of these processes by investigating the nature of seismic signals that accompany volcanic eruptions and by seismically imaging a magma body beneath a mid-ocean ridge, both located on, or adjacent to Iceland. The seismic phenomena associated with the 1996 Vatnajӧkull subglacial eruption in central Iceland, have been studied using data recorded by both temporary (HOTSPOT) and permanent (SIL) seismic networks. These networks comprise 60 broadband and short-period three-component seismographs and cover most parts of the country. Two very active volcanic systems, Bárdarbunga and Grimsvӧtn, are situated underneath the Vatnajokull ice cap. The volcanoseismic signals recorded there were categorised according to their waveform shape and frequency content, into three groups: (a) low-frequency events (1-2 Hz); (b) mixed-frequency events (1-4 Hz); and (c) volcanic tremor. The eruption was preceded by intense seismic activity which began with a = 5.6 earthquake located at the Bárdarbunga volcanic system. The epicentres of the earthquake swarm that followed the M(_w), = 5.6 event initially delineated the Bárdarbunga caldera rim and then migrated towards Grimsvӧtn, to a place where a fissure was later observed. Pre-eruptive tremor started at least two days before the eruption as a harmonic signal around five narrow frequency bands (0.5-0.7, 1.6, 2.2, 2.8 and 3.2 Hz). Co-eruptive tremor started as a broadband, continuous signal which evolved into low-amplitude background tremor interrupted by high-amplitude, cigar-shaped bursts. Further analysis revealed that continuous tremor and the cigar-shaped bursts had all the characteristics of low- dimensional chaotic signals. Geophysical and geochemical evidence suggest that a lateral migration of magma from Bárdarbunga facilitated the rupture of the roof of a magma chamber, situated at the fissure area, which subsequently erupted as tephra on the glacier. The second phase of the RAMESSES (Reykjanes Ridge Axial Melt Experiment: Structural Synthesis from Electromagnetic and Seismics) experiment involved the acquisition of multichannel seismic reflection data from 39 along- and across-axis lines shot over the magmatically active 57º 45'N axial volcanic ridge. The data from one along-axis line were processed using a variety of techniques that mainly aimed at reducing the large amount of coherent noise present, a result of scattered energy at the rough seabed. The final processed section revealed a number of reflection events that could be interpreted as intra-crustal reflections, originating from the interface between pillow lavas and sheeted dykes, and from the top part of a thin melt lens.

- / -

To obtain the rupture history of the Parkfield, California, earthquake, we perform 12 kinematic inversions using elliptical sub-faults. The preferred model has a seismic moment of 1.21 x 10^18 Nm, distributed on two distinct ellipses. The average rupture speed is ~2.7 km/s. The good spatial agreement with previous large earthquakes and aftershocks in the region, suggests the presence of permanent asperities that break during large earthquakes. We investigate our inversion method with several tests. We demonstrate its capability to retrieve the rupture process. We show that the convergence of the inversion is controlled by the space-time location of the rupture front. Additional inversions show that our procedure is not highly influenced by high-frequency signal, while we observe high sensitivity to the waveforms duration. After considering kinematic inversion, we present a full dynamic inversion for the Parkfield earthquake using elliptical sub-faults. The best fitting model has a seismic moment of 1.18 x 10^18 Nm, distributed on one ellipse. The rupture speed is ~2.8 km/s. Inside the parameter-space, the models are distributed according the rupture speed and final seismic moment, defining a optimal region where models fit correctly the data. Furthermore, to make the preferred kinematic model both dynamically correct while fitting the data, we show it is necessary to connect the two ellipses. This is done by adopting a new approach that uses b-spline curves. Finally, we relocate earthquakes in the vicinity of the Darfield, New-Zealand earthquake. 40 years prior to the earthquake, where there is the possibility of earthquake migration towards its epicentral region. Once it triggers the 2010-2011 earthquake sequence, we observe earthquakes migrating inside regions of stress increase. We also observe a stress increase on a large seismic gap of the Alpine Fault, as well as on some portions of the Canterbury Plains that remain today seismically quiet.

Shear-wave splitting and azimuthal variations of seismic attributes are two major anisotropic effects induced by vertically-aligned fractures. They both have influences on seismic data processing and interpretation, and provide information on fracture properties. Azimuthal variations in P-wave data have been intensively studied to improve imaging and obtain fracture parameters. However, azimuthal variations in PS-converted wave seismic data, particularly the velocity variation in PS-converted wave data, have not been well studied. Shear-wave splitting has been frequently used to estimate fracture directions and densities. However, its influence on the azimuthal variations of PS-converted wave data has also lacked a proper analysis. In this thesis, I analyse the anisotropic behaviour of PS-converted wave seismic data in the presence of azimuthal anisotropy, which includes the azimuthal variation of the PSconverted wave and PS-converted wave splitting. First, I demonstrate the robustness of PS-converted wave splitting for fracture characterisation. PS-converted wave seismic data is also influenced by the splitting effect due to its upgoing shear-wave leg. This important feature enables the application of shear-wave splitting analysis to PS-converted wave seismic data. I use synthetic data to show the necessity for separation of the split PS-converted waves. Then I apply the PS-converted wave splitting analysis to Sanhu 3D3C land seismic data. By separation of the fast and slow PS-converted waves and compensation for the time delays, the imaging quality has been improved. Dominant fracture properties obtained from the splitting analysis show a good correlation with the stress-field data. However, this work is accomplished by assuming only one set of vertical fractures in processing a given time window. In future work a specific layer-stripping algorithm could be constructed and applied. . Second, I study azimuthal variations of velocities in PS-converted wave seismic data. It involves two major parts: analysing azimuthal variations of NMO velocities to improve imaging, and examining the sensitivity of azimuthal variations to different fluid saturations. For a layer with HTI anisotropy induced by a set of vertical fractures, seismologists usually analyse the azimuthal behaviour exhibited on the radial and transverse components, on which PS-converted wave data are recorded. However, PS-converted waves also undergo shear-wave splitting, which complicates the azimuthal variations of PS-converted wave data. I demonstrate that it is essential to separate the fast P-SV1 wave from the slow P-SV2 wave, before applying any azimuthal analysis. I derive an equation describing the azimuthal variation in PSconverted wave NMO velocities, which shows the variation can be approximated into an ellipse. Based on this theory, I build a workflow to analyse the azimuthal variations of velocities in PS-converted wave data and apply this workflow to synthetic data. The imaging quality can be improved by using this workflow. Different fluid saturations in fractures have different influences on the azimuthal variations of both P-wave and PS-converted wave data. I perform a numerical study to understand how dry or water-saturated fractures control the azimuthal variations. Through theoretical and synthetic studies, I find that the azimuthal variation of velocities in PS-converted wave data is sensitive to different fluid saturations. By analysing the azimuthal variation, the fracture properties can also be estimated, but results are not as robust as those from PS-converted wave splitting analysis. I find that azimuthal variations of fast P-SV1 and slow P-SV2 waves show in-phase characteristics in dry fractures, but exhibit out-of-phase characteristics in water-saturated fractures. This important feature could open a new application for using PS-converted wave seismic data to distinguish oil-filled fractures from gas-filled fractures. In cases where multiple HTI layers are involved, I have developed a specific layer-stripping method to analyse both azimuthal variations and splitting effects of PS-converted waves. By applying this method to synthetic data, the fracture properties of each HTI layer can be estimated. The analysis of azimuthal variations in PS-converted wave velocities is applied to Daqing 3D3C land data. By using azimuthal velocity models in the PS-converted wave seismic data processing, the imaging quality is improved, especially in the anticline area where intensive fractures are likely to be developed. Furthermore, all fracture information obtained from analysis of azimuthal variations and splitting effects is compared with the stress-field data. The results from splitting analysis show a better correlation with the stress-field study. Finally, it is important to conclude that the analysis of PS-converted wave splitting is a robust method to estimate fracture directions and densities. However, it is not sensitive to different fluid saturations, which limits its application to fractured reservoir characterisation. Azimuthal variations of PS-converted wave seismic data can be analysed to improve imaging quality. Moreover their sensitivity to fluid saturations may provide a new way to discriminate between oil-filled and gas-filled fractures. However, the analysis of azimuthal variations is not as robust as the analysis of splitting effects, and it may require appropriate calibration with other fracture characterisation methods.

In this thesis, I use multiple techniques to investigate the active faulting and deformation of the Altay Mountains, Western Mongolia. The Altay are an intracontinental transpressional mountain range, which are deforming in the far-field of the India-Asia collision. An anastomosing network of dextral faults strikes NNW-SSE, and accommodates NE-SW oriented shortening by rotating anticlockwise about vertical axes. I begin by characterising the Altay faults, and add to what is already known about their surface expression with new observations of active faulting and three previously undescribed ancient earthquake ruptures. I use ¹⁰Be cosmogenic dating and uranium-series dating on pedogenic carbonate to estimate the average Quaternary rate of slip for two of the major fault zones in the Altay. The slip rate on the Ölgiy fault is constrained to 0.3-2.1 mm/yr^-1. Results from the Hovd fault are ambiguous, demonstrating the complications encountered with application of Quaternary dating techniques. I measure palaeomagnetic directions from Cretaceous to Pliocene-aged sediments in the eastern Altay to constrain the degree of anticlockwise rotation. Results from thermal demagnetisation of specimens indicate that the eastern Altay has not undergone significant rotation, in contrast with previous studies from the Siberian Altay that reveal almost 40 degrees of anticlockwise rotation. This suggests that the eastern-most Altay fault is too young to have experienced significant rotation, or is kinematically different from the Siberian Altay. I apply apatite fission track (AFT) dating and track length modeling to the central Altay. Results from AFT dating show rapid cooling in the late Cretaceous due to the distal assembly of Central Asia, suggesting that there was pre-existing topography at the start of the Late Cenozoic phase of deformation, the timing of which is constrained to have initiated at least 20 Myr ago. My work demonstrates that combining results from techniques that cover a variety of time scales quantifies the evolution of active faulting and deformation in the region.

In this thesis I use Interferometric Synthetic Aperture Radar (InSAR) and GPS geodetic observations, along with numerical models, to examine the distribution of strain, assess seismic hazard, and study the dynamics of deformation across Turkey and Iran. I measure interseismic strain accumulation across the Ashkabad fault using InSAR, and find that atmospheric corrections using MERIS (Medium Resolution Imaging Spectrometer) data are necessary in order to retrieve the tectonic signal in the presence of large atmospheric delays. I estimate a slip rate of 5-12 mm/yr for the Ashkabad fault which is faster than previous geodetic estimates. I also attempt to validate atmospheric corrections derived from the ERA-Interim numerical weather model and find that they do not work satisfactorily for this region. I produce InSAR-derived velocity maps for five overlapping tracks in Eastern Turkey, covering both the North Anatolian Fault (NAF) and East Anatolian Fault (EAF), and measure slip rates for the NAF and EAF of 20+/-3 mm/yr and 10+/-2 mm/yr respectively. I calculate a velocity field for Eastern Turkey from these InSAR data and a compilation of GPS data, and find that strain is mainly localised across the NAF and EAF and that there is negligible differential vertical motion across the Anatolian Plateau. I construct a thin viscous sheet model for Iran and find that the GPS velocity field is well described by deformation of a ductile lithosphere. Contrary to previous suggestions, a rigid central Iran is not required to match the kinematics of Iranian deformation, but buoyancy forces acting in the lithosphere are found to play an important role. I develop a new method to assess slip rates and therefore seismic hazard on major faults in Iran from this continuum model. In this thesis I have measured slip rates across three major strike-slip faults using InSAR; the first time this has been achieved for the Ashkabad fault and the EAF. I have demonstrated the importance of atmospheric correction for these results, and have shown that Iran deforms as a continuous medium.

L’explosion du nombre et de la variété des données géodésiques, sismologiques et tsunami disponibles est une opportunité exceptionnelle pour produire de nouveaux modèles de la source sismique. Mais ces données n’apportent pas toutes la même information et sont soumises à différentes sources d’incertitudes, rendant la solution au problème inverse non-unique. Dans cette thèse, nous utilisons une méthode d’échantillonnage bayésien pour produire de nouveaux modèles de glissement moins assujettis au sur-ajustement des données et permettant une estimation réaliste de l’incertitude associée aux paramètres estimés. Nous l’appliquons à l’étude du glissement dans trois contextes tectoniques différents : le séisme de Landers (1992, Mw=7.3), la zone de subduction équato-colombienne où s’est produit le séisme de Pedernales (2016, Mw=7.8), et le séisme intra-plaque de Tehuantepec (2017, Mw=8.2). À travers ce travail, nous démontrons l’importance de la considération rigoureuse des incertitudes et les atouts de l’approche bayésienne pour l’étude des différentes phases du cycle sismique. / The explosion in the amount and variety of available geodetic, tsunami, and seismological observations offers an outstanding opportunity to develop new seismic source models. But these data are sensitive to different sources of uncertainty and provide heterogeneous information, which makes the solution of the inverse problem non-unique.In this thesis, we use a Bayesian sampling method to propose new slip models, which benefit from an objective weighting of the various datasets by combining observational and modelling errors. These models are less affected by data overfit and allow a realistic assessment of posterior uncertainties. We apply this method to the study of slip processes occurring in three different tectonic contexts: the Landers earthquake (1992, Mw=7.3), the Ecuador-Colombia subduction zone which hosted the Pedernales earthquake (2016, Mw=7.8), and the intraslab Tehuantepec earthquake (2017, Mw=8.2). Through these analyses, we demonstrate how the study of the seismic cycle can benefit from rigorous uncertainty estimates and Bayesian sampling.

L’atténuation sismique est un paramètre physique très utile pour décrire et imager les propriétés du sous-sol, et tout particulièrement les roches saturées et les nuages de gaz. Les approches classiques analysent l’amplitude du spectre des données ou bien la distorsion de ce spectre, avec des méthodes asymptotiques. L’inversion des formes d’onde (Full Waveform Inversion en anglais, FWI) est une approche alternative qui prend en compte les aspects de fréquences finies. En pratique, à la fois les vitesses et l’atténuation doivent être déterminées. Il est connu que l’inversion multi-paramètre ne conduit pas à un résultat unique.Ce travail se focalise sur la détermination des vitesses et de l’atténuation. La dispersion liée à l’atténuation produit des modèles de vitesse équivalents en termes de cinématique. Je propose une inversion hybride : la « relation cinématique » est un moyen de guider l’inversion des formes d’onde non-linéaire. Elle se décompose en deux étapes. Dans un premier temps, l’information cinématique est remise à jour, et ensuite les vitesses et l’atténuation sont modifiées, pour une cinématique donnée. Différentes approches sont proposées et discutées au travers d’applications sur des données synthétiques 2D, en particulier sur les modèles Midlle-East et Marmousi. / Seismic attenuation is a useful physical parameter to describe and to image the properties of specific geological bodies, e.g., saturated rocks and gas clouds. Classical approaches consist of analyzing seismic spectrum amplitudes or spectrum distortions based on ray methods. Full waveform inversion is an alternative approach that takes into account the finite frequency aspect of seismic waves. In practice, both seismic velocities and attenuation have to be determined. It is known that the multi-parameter inversion suffers from cross-talks.This thesis focuses on retrieving velocity and attenuation. Attenuation dispersion leads to equivalent kinematic velocity models, as different combinations of velocity and attenuation have the same kinematic effects. I propose a hybrid inversion strategy: the kinematic relationship is a way to guide the non-linear full waveform inversion. The hybrid inversion strategy includes two steps. It first updates the kinematic velocity, and then retrieves the velocity and attenuation models for a fixed kinematic velocity. The different approaches are discussed through applications on 2D synthetic data sets, including the Midlle-East and Marmousi models.

Although much progress has been made in determining the three dimensional distribution of seismic wave velocities in the Earth, substantially less is known about the three dimensional distribution of intrinsic attenuation. In this study variations in attenuation and shear velocity of the Earth's mantle are constrained using measurements of differential travel time and attenuation. The data are broadband displacement SH seismograms filtered to have energy in the period range 8 to 20 s. Broadband data are used as they should allow a more accurate estimation of body wave attenuation to be made. The seismograms are obtained from over 600 globally distributed earthquakes of magnitude, Mw, 5.5 or greater. Two new methods for determining differential travel times and differential t* values from multiple S phases are presented. The first of these, referred to as the "waveform fitting method" is used to analyse approximately 4300 SS and S waveforms and around 1000 SSS and SS waveforms resulting in differential SS-S and SSS-SS travel times, and corresponding values of differential attenuation represented by t*. The second method, referred to as the "spectral ratio method" is used to analyse approximately 3200 SS and S and around 900 SSS and SS waveforms. The differential travel times and t* values are inverted to obtain models of the lateral variation of shear velocity and lateral variation of q(mu) where q(mu) =1/Q(mu). The models explain the data well but have limited depth resolution. The velocity models show good correlation with previous studies, in particular, low velocities are observed underlying spreading ridges and convergent margins and high velocities are observed under continental regions. The q(mu) model shows shield regions to be less attenuating than PREM, with ridges appearing as highly attenuating features. Models of shear velocity and attenuation are also obtained by combining the body wave dataset of this study with the surface wave datasets of Van Heijst (1997) and Selby (1998).

- / -