Les données accumulées sur les mouvements du sol apportent des connaissances très importantes sur les processus de rupture des séismes, les caractéristiques du milieu de propagation, la relation entre le mouvement du sol et les dommages des structures... Cependant, les séismes de faible et moyenne amplitude étant plus fréquents que les grands événements sismiques, les bases de données de mouvements de sol utilisées dans le développement de modèles de prédiction du mouvement du sol ne contiennent pas beaucoup de données de forts séismes. Le point le plus critique concerne les stations proches de la rupture de la faille, pour lesquelles les bases de données restent mal échantillonnées. Les pays à sismicité modérée ou élevée pour lesquels des failles majeures peuvent se briser à proximité de ses grandes villes, sont donc confrontés à un risque sismique majeur, mais le manque d’enregistrements du mouvement ne permet pas une bonne prédiction des mouvements fort du sol. Il est donc nécessaire de simuler le mouvement fort en champ proche. Cette thèse est divisée en 2 parties. La partie 1 se concentre sur une meilleure compréhension de la rupture sismique et de son rapport avec le mouvement du sol proche de la faille. Les mécanismes de génération des valeurs de pics du mouvement du sol sont étudiés pour des ruptures homogènes et hétérogènes. Une analyse quantitative de sensibilité du mouvement du sol aux paramètres cinématiques de la rupture est présentée, pour des sites au voisinage de la rupture ainsi qu’en champ lointain. Un second chapitre est consacré à un effet de source majeur en champ proche: l’effet de directivité. Ce phénomène se produit lorsque la rupture se propage vers un site, avec une vitesse de rupture proche de la vitesse de l'onde de cisaillement Vs; les ondes se propageant vers le sites interfèrent de manière constructive et génèrent une onde de grande amplitude appelée pulse. Les caractéristiques de ce pulse, notamment sa durée, représentent des paramètres d’intérêt pour le génie parasismique. Une équation simple est présentée pour relier la durée du pulse à la configuration géométrique de la rupture et du site d'intérêt et aux paramètres de la source. La partie 2 est consacrée à une meilleure estimation de l’aléa sismique au Liban en simulant le mouvement fort pour des sites proches de la faille principale: la faille de Yammouneh. Le Liban est situé dans un environnement tectonique actif où le risque sismique est considéré comme modéré à élevé. Historiquement, des tremblements de terre destructifs se sont produits dans le passé, le dernier remontant à 1202. Cependant, en raison de la sismicité de grande ampleur actuellement peu fréquente, aucun mouvement fort n'a jamais été enregistré au Liban à ce jour. La faille de Yammouneh est une grande faille en décrochement traversant le Liban du Nord au Sud, situant toutes les villes et infrastructures à moins de 25km de la faille. Une tomographie de la structure de la croûte du Liban, en termes de vitesse des ondes de cisaillement Vs, est réalisée en utilisant le bruit ambiant. À notre connaissance, il s’agit de la première étude de la tomographie Vs 3D au Liban. Par la suite, une approche hybride est utilisée pour simuler le mouvement du sol en champ proche sur une large bande de fréquences (0.1-10Hz). Aux basses fréquences (≤1Hz), des ruptures potentielles de M7 sont simulées (comme définie dans les chapitres précédents), et les fonctions sources obtenues sont convoluées aux fonctions de Green calculées pour le modèle de propagation des ondes issu de la tomographie Vs afin d’estimer le mouvement du sol à proximité de la faille. Le mouvement du sol est complété par un contenu haute fréquence (jusqu’à 10 Hz), en utilisant un modèle stochastique calibré par des enregistrements en champ proche, et en tenant compte de la phase impulsive due à la directivité de la rupture. / Accumulated data of strong ground motions have been providing us very important knowledge about rupture processes of earthquakes, propagation-path, site-amplification effects on ground motion, the relation between ground motion and damage... However, most of the ground motion databases used in the development of ground motion prediction models are primarily comprised of accelerograms produced by small and moderate earthquakes. Hence, as magnitude increases, the sets of ground motions become sparse. Ground motion databases are poorly sampled for short source-to-site distance ranges (‘Near-fault’ ranges). However, the strongest ground shaking generally occurs close to earthquake fault rupture. Countries of moderate to high seismicity for which major faults can break in the vicinity of its major cities are facing a major seismic risk, but the lack of earthquake recordings makes it difficult to predict ground motion. Strong motion simulations may then be used instead. One of the current challenges for seismologists is the development of reliable methods for simulating near-fault ground motion taking into account the lack of knowledge about the characteristics of a potential rupture. This thesis is divided into 2 parts. Part 1 focuses on better understanding the seismic rupture process and its relation with the near-fault ground motion. The mechanisms of peak ground motion generating are investigated for homogeneous as well as for heterogeneous ruptures. A quantitative sensitivity analysis of the ground motion to the source kinematic parameters is presented, for sites located in the vicinity of the fault rupture, as well as far from the rupture. A second chapter is dedicated to a major near-fault source effect: the directivity effect. This phenomenon happens when the rupture propagates towards a site of interest, with a rupture speed close to the shear-wave speed (Vs); the waves propagating towards the site adds up constructively and generates a large amplitude wave called the pulse. The features of this pulse are of interest for the earthquake engineering community. In this chapter, a simple equation is presented that relates the period of the pulse to the geometric configuration of the rupture and the site of interest, and to the source parameters.Part 2 is dedicated to better estimate the seismic hazard in Lebanon by simulating the strong ground motion at sites near the main fault (the Yammouneh fault). Lebanon is located in an active tectonic environment where the seismic hazard is considered moderate to high. Historically, destructive earthquakes occurred in the past, the last one dates back to 1202. However, strong motion has never been recorded in Lebanon till now due to the presently infrequent large-magnitude seismicity, and therefore facing an alarming note of potential new ruptures. The Yammouneh fault is a large strike-slip fault crossing Lebanon, making all its regions located within 25km away from the fault. At first, the crustal structure tomography of Lebanon, in terms of Vs, is performed using the ambient noise, in order to characterise the wave propagation from the rupture to the ground surface. To our knowledge, this is the first study of the 3D Vs tomography in Lebanon. Afterwards, a hybrid approach is presented to simulate broadband near-fault ground motion . At low-frequencies (≤1Hz), potential ruptures of M7 are simulated (as defined in the previous chapters), and the generated slip rate functions are convolved with the Green’s functions computed for the propagation medium defined by the Vs tomography. The ground-motion is complemented by a high-frequency content (up to 10Hz), using a stochastic model calibrated by near-fault recordings and accounting for the presence of the directivity pulse. The computed peak ground acceleration is compared to the design acceleration in Lebanon.

Near surface measurements of seismic velocity and Q are useful in a number of situations, for example for when carrying out re-datuming and migration for depth images, or when analysing ground conditions for building. This thesis concentrates on the estimation of surface wave group and phase velocity as well as Q structure through the use of cross correlations of ambient noise recordings. Linearised tomography estimates are made for the British Isles, the Permain Basin of Texas and New Mexico, the Ekofisk Life of Field Seismic (LoFS) array and the Aquistore CO2 storage site. The results correspond well with the known geological structure and/or structure observed in velocity maps by other researchers. For the Ekofisk array a non-linearised non-linear method was also applied and the results estimated by these two methods for the Ekofisk LoFS array are compared. By non-linearised non-linear it is meant that the inversion method is fully non-linear and no linearisation has taken place in the method, this term will be used throughout this thesis for all methods which fall into this category. The tomography results from the two methods had similarities in their general structure but differences in the finer details, and so suggest that the substantial increase in time required for the non-linearised non-linear method is not justified. Linearised tomographic inversion of the Aquistore array was used to determine the potential of using ambient noise tomography for monitoring of CO2 storage sites. It was found that the repeatability of the tomography at the Aquistore site was not good enough to allow ambient noise tomography to be used for monitoring; however, it may be possible to apply the method at other sites. A Q and phase velocity inversion of the Ekofisk array is also presented, with results mostly showing excellent correlation with known geological features. It is shown that the higher frequencies are more sensitive to the effects of sea floor subsidence at the site, while lower frequencies are more sensitive to the effects of faulting. A final near surface method called ambient noise gradiometry was applied, this concentrates on estimating locations of sources of seismic energy within receiver arrays. Ambient noise gradiometry is applied to synthetic and real data for this purpose. It was found that using ambient noise gradiometry allows internal sources of energy to be identified but they produce a bias in the phase velocity tomography result. Two methods of reducing this bias are presented, both of which also provide an estimate of the source term for different sections of time of the recording.

The capability to observe ocean swell using spaceborne Synthetic Aperture Radar (SAR) has been demonstrated starting with ERS-1 mission in 1992. This dissertation shows how ocean swell properties can be used to combine swell observations of heterogeneous quality and acquired at various times and locations for the observation and forecast of ocean swell fieldsusing ASAR instrument on-board ENVISAT. The first section is a review of how ocean swell spectra can be derived from the SAR complex images of the ocean surface using a quasi-linear transformation. Then, significant swell heights, peak periods and peak directions from in situ measurements are used to assess the accuracy of the SAR observed swell spectra. Using linear propagation in deep ocean, a new swell field reconstruction methodologyis developed in order to gather SAR swell observations related to the same swell field. Propagated from their generation region, these observations render the spatio-temporal properties of the emanating ocean swell fields. Afterwards, a methodology is developed for the exclusion of outliers taking advantage of the swell field consistency. Also, using the irregularly sampled SAR observations, quality controlled estimations of swell field integral parameters are produced on a regular space-time grid. Validation against in situ measurements reveals the dramatic impact of the density of propagated observations on the integral parameters estimated accuracy. Specifically, this parameter is shown to be very dependent on the satellite orbit. Finally, comparisons with the numerical wave model WAVEWATCH-III prove it could potentially benefit from the SAR swell field estimates for assimilation purposes.

Les signaux enregistrés à la surface du globe sont composés de trois types d'ondes : des ondes directes, des ondes réfléchies plus ou moins facilement interprétables, et des ondes multiplement diffusées beaucoup plus complexes à interpréter. Certaines propriétés physiques de ces ondes multiplement diffusées permettent de les assimiler au bruit sismique ambiant. Nous appliquons des techniques de corrélation de bruit sismique développées ces dernières années à la coda sismique, afin de tirer des informations sur le caractère multiplement diffusant du milieu. En particulier, nous montrons que le théorème reliant fonction de Green et champ ambiant peut être utilisé dans la coda. La dynamique temporelle de la reconstruction de la fonction de Green est alors un indicateur de la répartition de l'énergie dans le milieu. En reconstruisant la fonction de Green à l'aide de corrélations pour plusieurs fenêtres de temps dans la coda, nous montrons qu'il est possible de relier la symétrie des parties causales et acausales de la fonction de Green au libre parcours moyen du milieu. Nous développons ensuite des simulations numériques de propagation d'onde acoustique en 2D. Nous observons que l'évolution de la reconstruction des fonctions de Green observée dans ce milieu et celle prédite par la théorie est proche. La même approche est ensuite appliquée à des données sismologiques de terrain, sans permettre de retrouver la même dynamique. Nous développons alors une méthode de type MCMC permettant de reconstruire les fonctions de Green du milieu de manière optimale. / Most of the waves recorded by seismometers are hard to interpret because of the complexity of the propagation medium, especially the late part of the seismic coda. These multiply scattered coda waves are close in nature to the ambient noise. We are applying recent noise correlation techniques to coda waves in order to retrieve information about the scattering medium. We show how the relationship between ambient noise and Green's function can be used in the case of the seismic coda. The quality of the Greens function retrieved by cross correlation of time windows in the coda is a proxy indicating the energy partition in the propagation medium. In particular, we establish a link between the symmetry of the causal and acausal parts of the reconstructed Green's function and the mean free path. We validate this theoretical approach with acoustical 2D numerical simulations. The same approach seems inefficient on a seismological dataset, due to the high S/N ratio and the non optimal repartition of receivers. Hence, we develop an MCMC based algorithm in order to optimally reconstruct the green's function in the seismic coda.

A comprehensive study of the soil shaking under the seismic wave-field ex- citation is presented. It includes theoretical, geological, geotechnical, data analysis and numerical simulations aspects. The aim is to quantify the main parameters allowing the estimate of the soil shaking in urban areas for better mitigating seismic risk due to future earthquakes. The city of Rome has been chosen as a case study because of its high density of popula- tion and large concentration of historical monuments with high earthquake vulnerability. This study improves significantly the knowledge concerning the detailed near-surface geology of the chosen study area of Rome, ful- fills the absence both of knowledge concerning its geotechnical properties and earthquake data recordings in the city. Among others, it allows for a better explanation of the spatial damage pattern observed in the city due to earthquakes in the past. The main innovations include the construction and long-term operation a seismic array in the city, analysis of the natural seismic noise, and instrumental recordings of the 2009 L'Aquila earthquake sequence. The 3D array (including a borehole sensor at 70-m depth) is the first one in Italy planned, realized and operated within an urban area, and the first one that recorded a significant earthquake in...

Ces dernières années ont vu le développement d'une nouvelle méthode d'imagerie des structures géologiques basée sur l'utilisation du bruit sismique continu. Dans ce travail nous avons utilisé cette approche dans deux problématiques différentes. La première consiste à réaliser le suivi temporel des vitesses des ondes sismiques dans la croûte lors de séismes lents qui ont eu lieu dans la région de Guerrero au Mexique. Les séismes lents sont des glissements asismiques et transitoires qui ont été découverts récemment dans la lacune sismique de Guerrero. Ils sont considérés comme une part importante de la relaxation des déformations dans le cycle sismique. Les séismes lents affectent le processus de chargement et déchargement de l'interface, il est donc important de comprendre le comportement mécanique de la subduction dans cette région pour mieux évaluer le risque sismique. Dans notre étude, nous avons mesuré les perturbations de vitesse des ondes associées à deux séismes lents en 2006 et 2009-2010 à partir des enregistrements continus du bruit sismique. Pour chacun des deux séismes lents on observe une chute de vitesse : elle s'élève à 0.2% pour celui de 2006 et à 0.8 % pour celui de 2009-2010. Au cours de ces séismes lents, les ondes de longues périodes (>10 s) sont perturbées. A courtes périodes, aucune variation de vitesse n'est observée ce qui suggère qu'un endommagement des couches superficielles de la croûte ne peut pas être à l'origine du changement de vitesse. Par ailleurs, la perturbation de vitesse est reliée au taux de dé- formation plutôt qu'à la déformation elle-même. Cette observation suggère que pendant de forts séismes lents, la croûte chevauchante présente un comportement mécanique non linéaire. Nous pouvons donc utiliser les variations de vitesse comme des marqueurs du taux de déformation du milieu. Enfin, une corrélation entre les trémors non volcaniques et les variations de vitesse suggère qu'une part importante de la déformation résultant des séismes lents est accommodée par la croute chevauchante. La deuxième problématique abordée dans cette thèse est l'imagerie de structures à fort contraste de vitesse et dans lesquelles la propagation des ondes de surface est complexe. Imager et comprendre la propagation des ondes dans la vallée de Mexico est crucial pour l'estimation du risque sismique à la capitale. Nous mesurons la dispersion des ondes de Rayleigh reconstruites à partir d'intercorrélations de bruit de fond sismique. Pour identifier les modes nous utilisons une mesure du rapport spectral des composantes horizontales sur la composante vertical (H/V) sur la coda des séismes que l'on compare avec le rapport H/V théorique. Grâce à cette identification des modes, nous pouvons retrouver le modèle de vitesse de la structure. / Recent years have seen the development of a new method for imaging geological structures based on continuous seismic noise. In this work we used this approach in two different problems. The first is to monitor seismic waves velocity in the crust during slow slip events that occurred in the region of Guerrero in Mexico. These slow slip events are aseismic transients that were observed recently in the seismic gap of Guerrero. They are considered an important part of the strain relaxation in the seismic cycle. Since slow slip events affect the process of loading and unloading of the interface, it is important to understand the mechanical behavior of the subduction in this region to better assess the seismic risk. In our study, we measured wave velocity perturbations associated with two slow slip events in 2006 and 2009-2010 from continuous recordings of seismic noise. For both events we observed a drop in wave speed : it reaches 0.2 % in 2006 and 0.8 % in 2009-2010. During these slow slip events waves of long periods (> 10 s) are disturbed. At short period, no velocity variation is observed suggesting that damage of the superficial layers of the crust cannot produce such velocity perturbation. Moreover, the wave speed change is related to the strain rate rather than the deformation itself. This observation suggests that during strong slow slip events the overriding crust presents a nonlinear mechanical behavior. We can therefore use the velocity variations as a proxy of the strain rate of the medium. Finally, a correlation between non-volcanic tremor and changes in waves speed suggests that part of the deformation resulting from the slow slip events is accommodated by the overriding crust. The second issue addressed in this thesis is imaging geological structures with high velocity contrast in which the propagation of surface waves is complex. Charac- terizing and understanding wave propagation in the Valley of Mexico is crucial for the estimation of seismic risk in Mexico City. We measure the dispersion of Rayleigh waves reconstructed from cross-correlations of seismic noise. To identify the modes of Rayleigh waves we use a measure of the spectral ratio of the horizontal components to the vertical component (H / V) in the coda of earthquakes which are compared with the theoretical H / V. With this identification method, we can find the velocity model of the structure.

L’imagerie satellite radar propose un point de vue intéressant pour l’étude et la compréhension des océans. Là où l’altimétrie, reconnue et utilisée mondialement, a su s’imposer comme une source de données majeure, les observations de houle issues du SAR (de l’anglais « Synthetic Aperture Radar ») restent encore largement sous exploitées. L’objet de cette thèse est de promouvoir l’utilisation de ces données en proposant un modèle pour l’analyse et la prévision de la houle à l’échelle du globe qui soit indépendant des modèles numériques classiques comme Wavewatch-III. Ce travail s’inscrit dans une logique de pérennisation de la mesure de houle depuis l’espace avec le lancement dans les trois années à venir des trois missions satellites Sentinel-1 A et B et CFOSAT. Un des principaux résultats de ce travail est la capacité de la méthode développée à fournir une information plus précise que celle des modèles existants. Cette méthode permet également une meilleure caractérisation des mesures utilisées en entrée et des pistes d’amélioration de ces dernières sont dégagées pour les futures activités de calibration/validation. Ces travaux ouvrent également des perspectives sur les possibilités d’assimilation des sorties de ce nouveau modèle dans les modèles numériques classiques. / The capability to observe ocean swell using spaceborne Synthetic Aperture Radar (SAR) has been demonstrated starting with ERS-1 mission in 1992. This dissertation shows how ocean swell properties can be used to combine swell observations of heterogeneous quality and acquired at various times and locations for the observation and forecast of ocean swell fieldsusing ASAR instrument on-board ENVISAT. The first section is a review of how ocean swell spectra can be derived from the SAR complex images of the ocean surface using a quasi-linear transformation. Then, significant swell heights, peak periods and peak directions from in situ measurements are used to assess the accuracy of the SAR observed swell spectra. Using linear propagation in deep ocean, a new swell field reconstruction methodologyis developed in order to gather SAR swell observations related to the same swell field. Propagated from their generation region, these observations render the spatio-temporal properties of the emanating ocean swell fields. Afterwards, a methodology is developed for the exclusion of outliers taking advantage of the swell field consistency. Also, using the irregularly sampled SAR observations, quality controlled estimations of swell field integral parameters are produced on a regular space-time grid. Validation against in situ measurements reveals the dramatic impact of the density of propagated observations on the integral parameters estimated accuracy. Specifically, this parameter is shown to be very dependent on the satellite orbit. Finally, comparisons with the numerical wave model WAVEWATCH-III prove it could potentially benefit from the SAR swell field estimates for assimilation purposes.

[Truncated abstract] Interferometric gravitational wave detectors are being built around the world with continually improving measurement sensitivities. Noise levels from sources that are intrinsic to these detectors must be reduced to a level below the gravita- tional wave signal. Seismic noise in the low frequency range, which is within the gravitational wave detection bandwidth, is a concern for earth-based detectors. This thesis presents research and development of a high performance vibration isolation system that is designed to attenuate seismic noise. The final design will be used as part of a fully working interferometer at the Australian International Gravitational Observatory (AIGO). Pendulums and springs are conventionally used for the horizontal and vertical vibration isolation components respectively. A complete system comprises of a cascade of these components, each stage dramatically improving the level of isola- tion. The residual motion at the test mass level is thus reduced but is dominated by the normal mode resonances of the chain. A simple and effective method to reduce residual motion further is to add ultra-low frequency pre-isolation stages which suspend the chain. The Roberts Linkage is a relatively new and simple geometrical structure that is implemented in the pre-isolation stages. Here we present experimental results of improving isolation based on mathematical mod- elling. The attenuation of seismic noise in the vertical direction is almost as important as that in the horizontal direction, due to cross-coupling between the two planes. To help improve the vertical performance a lightweight Euler spring that stores no static energy was implemented into the AIGO suspension system. ... Theoretical and experimental results are presented and discussed. Currently the AIGO laboratory consists of two 80 m length arms. They are aligned along the east and south directions. One of AIGO's top priorities is the installation of two complete vibration isolators in the east arm to form a Fabry-Perot cavity. Assembling two suspension systems will enable more accurate performance measurements of the tuned isolators. This would significantly reduce the measurement noise floor as well as eliminate the seismic noise spectrum due to referencing with the ground motion. The processes involved in preparing such a task is presented, including clean room preparation, tuning of each isolator stage, and local control schematics and methods. The status of the AIGO site is also presented.

Relative seismic velocity variations in the Earth’s crust can be estimated by using ambient seismic noise records from a pair of stations. Velocity variations can be caused by stress perturbations in the subsurface. Therefore, information on stress changes in the crust can possibly be retrieved from measured velocity variations in the medium. The measurement is done by comparing the coda part of two cross-correlation functions (CCFs) obtained from ambient noise recordings at two seismic stations; a current CCF that is considered to represent the actual state in the study medium at a specific time and a reference CCF that is considered to represent its average state. Here, the method is applied to the area around Katla volcano in southern Iceland. Katla is an active subglacial volcano and therefore frequent stress changes can be expected to take place there. Long-term changes (of the order of 1-2 months) in relative seismic velocity were estimated over a period of 7 months in 2011. These changes were of the order of about 0.1% for a frequency range of 0.2-1.0 Hz. For this frequency range, surface waves around Katla have been estimated to be most sensitive to velocity changes taking place at depths of about 1-5 km but the sensitivity kernels also have a peak at the surface. The scattering volume (in this case area since we are working with surface waves) depends on both the inter-station distance and how far into the coda the measurements are made. The inter-station distances vary between 5.8 and 23.4 km. Measurements are made 30 s into the coda. This results in scattering areas on the order of 100 km2. The velocity variations have a negative trend over July and over a two month period from the end of August until early November, and a positive trend in August and from early November until the end of the study period in late December. These variations are possibly the results of a combination of changes in the ground water level beneath the glacier, surface load changes and possibly hydrothermal and magmatic pressurization changes. No significant velocity change was estimated in the area associated with the tremor event that took place in early July in 2011. / Seismiska vågor är vibrationer i jordytan som genereras av jordbävningar, explosioner eller andra processer som skakar jorden. Seismiska vågor färdas genom jordens lager och innehåller därför information om jordens inre struktur. Dessa vibrationer kan hämtas med ett känsligt instrument som kallas seismometer. Seismiska vågor färdas med en viss hastighet som beror på hur hård och tung berggrunden är. Förändringar av dessa egenskaper kan därför resultera i förändringar av hastigheten. Dessa förändringar kan orsakas av spänningsförändringar under marken, till exempel trycket i porer eller variationer i vikten ovanför marken, exempelvis från en glaciär. I denna uppsats studeras förändringar av seismiska vågors hastighet kring vulkanen Katla på Island under 7 månader, 2011. Katla är en av Islands mest aktiva vulkaner och är belägen under en glaciär, Mýrdalsjökull. Detta görs genom att använda omgivande seismiskt brus, som består av seismiska vågor. Bruset genereras av tryckvariationer i samband med havsvågor. Bruset analyseras med en korrelationsanalys som bland annat isolerar spridda vågor från detaljer i strukturen och variationer av dessa med tid kan användas til mätningar av hastighets förändringar. Resultaten tyder på förändringar i relativ seismik hastighet avstorleken 0.1% som varar i en till två månader. Hastigheten minskar i juli och över en tvåmånadersperiod från slutet av augusti till början av november, men ökar i augusti och från början av november till slutet av december. Dessa variationer kan ha orsakats av en kombination av förändringar i grundvattennivån under glaciären, förändringar i glaciärens vikt och magmatiska processer. Inga tydliga förändringar i samband med sekvenser av små jordbävningar som ägde rum i början av juli 2011 kunde observeras frånförändringar i relativ seismisk hastighet runt Katla.

Les éboulements rocheux de volume intermédiaire (103-105 m3) posent un problème sérieux dans les régions montagneuses en raison de leur fréquence d'occurrence relativement élevée et de leur fort pouvoir destructeur. En conséquence, il est difficile de mener des travaux de protection pour réduire l'aléa, ce qui souligne le besoin de techniques de suivi et d'alerte précoce avant la rupture. Après l'éboulement, peu d'informations quantitatives sont disponibles sur la phase de propagation en raison de la soudaineté du phénomène et de sa localisation dans des pentes raides et difficiles d'accès. Dans ce travail de thèse, une approche expérimentale est proposée pour extraire des informations à partir d'enregistrements de signaux sismiques durant les phases pré- et post-rupture d'éboulements.La première partie de ce travail vise à tester la pertinence des vibrations ambiantes pour le suivi temporel de la réponse dynamique de compartiments rocheux lors de la phase pré-rupture. Cette technique (couramment employée en génie civil pour le suivi de l'intégrité des structures) permet d'extraire les fréquences de résonance d'une structure, dont la décroissance peut traduire l'endommagement. Une étude antérieure menée sur une colonne calcaire instable a montré une décroissance d'environ 30% de la fréquence fondamentale de résonance (f1) environ deux semaines avant la rupture, interprétée comme une diminution de la rigidité du contact avec le massif stable adjacent. / Mid-size rockfalls (103-105 m3) represent a substantial hazard in mountainous areas, because of relative high rate of occurrence and destructive power. Consequently, few protection means can be applied, emphasizing the need for monitoring techniques and early warning prior to the collapse. After the rupture, quantitative information on the rockfall propagation phase is scarce, owing to their suddenness and location in steep and rugged slopes. In this thesis work, an experimental approach is proposed to derive valuable information from seismic records during rockfall pre-rupture and post-rupture phases. The first part of this work aims at testing the applicability of the ambient vibration technique to monitor unstable rock compartments dynamic response in the pre-rupture phase. This technique (commonly employed in civil engineering for structural health monitoring) reveals the resonant frequencies of a structure, a decrease in frequency revealing potential damage. A previous case study of an unstable limestone compartment brought to light a #30% decrease in fundamental resonant frequency (f1) two weeks before the collapse. Following this innovative work, we selected and instrumented four prone-to-fall medium-size rock compartments located in the Occidental Alps, showing various geological contexts (limestones, argillite and shale-sandstone series), deformation patterns and failure mechnisms. Ambient vibrations recorded on-site revealed caracteristic seismic noise features. Spectral peaks were observed and attributed to resonant frequencies of the rock compartments, the fundamental resonant frequency (f1) showing clear polarization parallel to the line of maximum slope gradient, and perpendicular to the main bounding fracture observed at most of the sites. Similar findings were made for an unstable rock compartment located in a volcanic caldera, characterized by rapid morphological changes and intense rockfall activity. The dynamic response of the rear fracture network was explored, showing that spectral content of seismic noise is controlled by the caldera structure in the 0.5-5 Hz range. The direction of vibration is polarized perpendicularly to the fractures, while vibration amplitudes are linked to compartment uncoupling from the rock massif. In this case, the physical origin of seismic noise amplification may be due to complex 2D or 3D resonance effects. For the four alpine sites, the fundamental frequency f1 was monitored over more than one year, showing fluctuations clearly correlated with temperature oscillations. The thermal control over f1 is highly complex, showing both positive and negative correlations, depending on site morphology and destructuration, as well as on the studied oscillations periods (daily or seasonal). No change in fundamental frequency resulting from damage was observed over this time span. One site, characterized by intense rock fracturing and a deep-open rear fracture, showed high f1 sensitivity to temperature changes. Thermo-mechanical numerical simulations revealed that both material contraction-dilation and thermal dependancy of the elastic modulus control f1 fluctuations. In addition, high amplitude seasonal f1 oscillations were explained by ice formation in the rear fracture. A criterion was developped to separate thermal-induced f1 fluctuations from damage effects, under the hypothesis that thermal sensitivity of a rock compartment increases towards failure. The second part of this work relates to the post-rupture phase of rockfalls. The seismic records generated by two mid-size rockfalls {one natural, one provoked{ that occured in the same place were analyzed, showing complex enveloppe and spectrogram features. Both events showed close magnitude, duration and spectral content. The seismic signals of the provoked event were calibrated using video shots, allowing estimation of fallen material velocity during the successive propagation phases.