Hydraulic fracturing in tight gas and shale gas reservoirs is an essential stimulation technique for production enhancement. Often, hydraulic fracturing induces fracture patterns that are more complex than the planar geometry that has been assumed in the past models. These complex patterns arise as a result of the presence of planes of weakness, faults and/or natural fractures. In this thesis, two different 3D geomechanical models have been developed to simulate the interaction between the hydraulic fracture and the natural fractures, and to observe the impact of geomechanics on the potential microseismicity in these naturally fractured formations. Several cases were studied to observe the effects of natural fracture geometry, fracturing treatment, mechanical properties of the sealed fractures, etc. on the propagation path of the hydraulic fracture in these formations, and were found to be consistent with past experimental results. Moreover, the effects of several parameters including cohesiveness of the sealed natural fractures, mechanical properties of the formation, treatment parameters, etc. have been studied from the potential microseismicity standpoint. It is shown that the impact of geomechanics on potential microseismicity is significant and can influence the desired fracture spacing. In this thesis, the presented model quantifies the extent of potential microseismic volume (MSV) resulting from hydraulic fracturing in unconventional reservoirs. The model accounts for random geometries of the weak planes (with different dip and strike) observed in the field. The work presented here shows, for the first time, a fracture treatment can be designed to maximize the MSV, when the fractures form a complicated network of fractures, and in turn influence the desired fracture spacing in horizontal wells. Our work shows that by adjusting the fluid rheology and other treatment parameters, the spatial extent of MSV and the desired fracture spacing can be optimized for a given set of shale properties. / text

Understanding dykes is vital as they serve both as bodies that build the crust and as conduits that feed eruptions. The 2014-15 Bárðarbunga-Holuhraun rifting event comprised the best-monitored dyke intrusion to date and the largest eruption in Iceland in 230 years. Over a 13 day period magma propagated laterally from the subglacial Bárðarbunga volcano, Iceland, along a 48 km path before erupting in the Holuhraun lava field on 29 August 2014. A huge variety of seismicity was produced, including over 30,000 volcano-tectonic earthquakes (VTs) associated with the dyke propagation at ∼ 6 km depth below sea level, and long-period seismicity - both long-period earthquakes (LPs) and tremor - associated with the eruption processes. The Cambridge University seismic network in central Iceland recorded the dyke seismicity in unprecedented detail, allowing high resolution analyses to be carried out. This dissertation comprises two parts: study of 1) the volcano-tectonic dyke-induced seismicity and 2) the long-period seismicity associated with eruption processes. Volcano-tectonic earthquakes induced by the lateral dyke intrusion were relocated, using cross-correlated, sub-sample relative travel times. The ∼ 100 m spatial resolution achieved reveals the complexity of the dyke propagation pathway and dynamics (jerky, segmented), and allows us to address the precise relationship between the dyke and seismicity. The spatio-temporal characteristics of the induced seismicity can be directly linked in the first instance to propagation of the tip and opening of the dyke, and following this - after dyke opening - indicate a relationship with magma pressure changes (i.e. dyke inflation/deflation), followed by a general 'post-opening' decay. Seismicity occurs only at the base of the dyke, where dyke-imposed stresses - combined with the background tectonic stress (from regional extension over > 200 years since last rifting) - are sufficient to induce failure of pre-existing weaknesses in the crust, while the greatest opening is at shallower depths. Emplacement oblique to the spreading ridge resulted in left-lateral shear motion along the distal dyke section (studied here), and a prevalence of left-lateral shear failure. Fault plane strikes are predominately independent of the orientation of lineations delineated by the hypocenters, indicating that they are controlled by the underlying host rock fabric. Long-period earthquakes and tremor were systematically detected and located during the dyke propagation phase and the first week of the eruption. Clusters of highly similar, repetitive LPs were identified, with a peak frequency of ∼ 1 Hz and clear P and S phases followed by a long-duration coda. The source mechanisms were remarkably consistent between clusters and also fundamentally different to those of the VTs. The clusters were accurately located near each of three ice cauldrons (depressions formed by basal melting) that were observed on the surface of Dyngjujökull glacier above the path of the dyke. Most events were in the vicinity of the northernmost cauldron, at shallower depth than the VTs associated with lateral dyke propagation. At the two northerly cauldrons, periods of shallow seismic tremor following the clusters of LPs were also observed. Given that the LPs occurred at ∼ 4 km depth and in swarms during times of dyke-stalling, it is inferred that they result from excitation of magmatic fluid-filled cavities and indicate magma ascent. The tremor may then represent the climax of the vertical melt movement, arising from either rapid, repeated excitation of the same LP cavities, or sub-glacial eruption processes. This long-period seismicity therefore highlights magma pathways between the depth of the dyke-VT earthquakes and the surface. Notably, no tremor is detected associated with each cauldron, despite melt reaching the base of the overlying ice cap, a concern for hazard forecasting.

The Centroid-Moment Tensor (CMT) method allows an optimal point-source description of an earthquake to be recovered from a set of seismic observations, and, for over 30 years, has been routinely applied to determine the location and source mechanism of teleseismically recorded earthquakes. The CMT approach is, however, entirely general: any measurements of seismic displacement fields could, in theory, be used within the CMT inversion formulation, so long as the treatment of the earthquake as a point source is valid for that data. We modify the CMT algorithm to enable a variety of near-field seismic observables to be inverted for the source parameters of an earthquake. The first two data types that we implement are provided by Global Positioning System receivers operating at sampling frequencies of 1,Hz and above. When deployed in the seismic near field, these instruments may be used as long-period-strong-motion seismometers, recording displacement time series that include the static offset. We show that both the displacement waveforms, and static displacements alone, can be used to obtain CMT solutions for moderate-magnitude earthquakes, and that performing analyses using these data may be useful for earthquake early warning. We also investigate using waveform recordings - made by conventional seismometers deployed at the surface, or by geophone arrays placed in boreholes - to determine CMT solutions, and their uncertainties, for microearthquakes induced by hydraulic fracturing. A similar waveform inversion approach could be applied in many other settings where induced seismicity and microseismicity occurs.

Automated location of seismic events is a very important task in microseismic monitoring operations as well for local and regional seismic monitoring. Since microseismic records are generally characterised by low signal-to-noise ratio, such methods are requested to be noise robust and sufficiently accurate. Most of the standard automated location routines are based on the automated picking, identification and association of the first arrivals of P and S waves and on the minimization of the residuals between theoretical and observed arrival times of the considered seismic phases. Although current methods can accurately pick P onsets, the automatic picking of the S onset is still problematic, especially when the P coda overlaps the S wave onset. In this thesis I developed a picking free automated method based on the Short-Term-Average/Long-Term-Average (STA/LTA) traces at different stations as observed data. I used the STA/LTA of several characteristic functions in order to increase the sensitiveness to the P wave and the S waves. For the P phases we use the STA/LTA traces of the vertical energy function, while for the S phases, we use the STA/LTA traces of the horizontal energy trace and then a more optimized characteristic function which is obtained using the principal component analysis technique. The orientation of the horizontal components can be retrieved by robust and linear approach of waveform comparison between stations within a network using seismic sources outside the network (chapter 2). To locate the seismic event, we scan the space of possible hypocentral locations and origin times, and stack the STA/LTA traces along the theoretical arrival time surface for both P and S phases. Iterating this procedure on a three-dimensional grid we retrieve a multidimensional matrix whose absolute maximum corresponds to the spatial and temporal coordinates of the seismic event. Location uncertainties are then estimated by perturbing the STA/LTA parameters (i.e the length of both long and short time windows) and relocating each event several times. In order to test the location method I firstly applied it to a set of 200 synthetic events. Then we applied it to two different real datasets. A first one related to mining induced microseismicity in a coal mine in the northern Germany (chapter 3). In this case we successfully located 391 microseismic event with magnitude range between 0.5 and 2.0 Ml. To further validate the location method I compared the retrieved locations with those obtained by manual picking procedure. The second dataset consist in a pilot application performed in the Campania-Lucania region (southern Italy) using a 33 stations seismic network (Irpinia Seismic Network) with an aperture of about 150 km (chapter 4). We located 196 crustal earthquakes (depth < 20 km) with magnitude range 1.1 < Ml < 2.7. A subset of these locations were compared with accurate locations retrieved by a manual location procedure based on the use of a double difference technique. In both cases results indicate good agreement with manual locations. Moreover, the waveform stacking location method results noise robust and performs better than classical location methods based on the automatic picking of the P and S waves first arrivals. / Die automatische Lokalisierung seismischer Ereignisse ist eine wichtige Aufgabe, sowohl im Bereich des Mikroseismischen Monitorings im Bergbau und von Untegrund Aktivitäten,  wie auch für die lokale und regionale Überwachung von natürlichen Erdbeben. Da mikroseismische Datensätze häufig ein schlechtes Signal-Rausch-Verhältnis haben müssen die Lokalisierungsmethoden robust gegen Rauschsignale und trotzdem hinreichend genau sein. Aufgrund der in der Regel sehr hochfrequent aufgezeichneten Messreihen und der dadurch sehr umfangreichen Datensätze sind automatische Auswertungen erstrebenswert. Solche Methoden benutzen in der Regel automatisch gepickte und  den P und S Phasen zugeordnete Ersteinsätze und Minimieren die Summe der quadratischen Zeitdifferenz zwischen den beobachteten und theoretischen Einsatzzeiten. Obgleich das automatische Picken der P Phase in der Regel sehr genau möglich ist, hat man beim Picken der S Phasen häufig Probleme, z.B. wenn die Coda der P Phase sehr lang ist und in den Bereich der S Phase hineinreicht. In dieser Doktorarbeit wird eine Methode vollautomatische, Wellenform-basierte Lokalisierungsmethode entwickelt, die Funktionen des Verhältnisses "Short Term Average / Long Term Average"  (STA/LTA) verwendet und keine Pickzeiten invertiert. Die STA/LTA charakteristische Funktion wurde für unterschiedliche Wellenform Attribute getestet, um die Empfindlichkeit für P und S Phasen zu erhöhen. Für die P Phase wird die STA/LTA Funktion für die Energie der Vertikalkomponente der Bodenbewegung benutzt, wohingegen für die S Phase entweder die Energie der horizontalen Partikelbewegung oder eine optimierte Funktion auf Basis der Eigenwertzerlegung benutzt wird. Um die Ereignisse zu lokalisieren wird eine Gittersuche über alle möglichen Untergrundlokalisierungen durchgeführt. Für jeden räumlichen und zeitlichen Gitterpunkt werden die charakteristischen Funktionen entlang der theoretischen Einsatzkurve aufsummiert. Als Ergebnis erhält man eine 4-dimensionale Matrix über Ort und Zeit des Ereignisses, deren Maxima die wahrscheinlichsten Lokalisierungen darstellen. Um die Unsicherheiten der Lokalisierung abzuschätzen wurden die Parameter der STA/LTA Funktionen willkürlich verändert und das Ereignis relokalisiert. Die Punktwolke aller möglichen Lokalisierungen gibt ein Maß für die Unsicherheit des Ergebnisses. Die neu entwickelte Methode wurde an einem synthetischen Datensatz von 200 Ereignissen getestet und für zwei beobachtete Datensätze demonstriert. Der erste davon betrifft induzierte Seismizität in einem Kohlebergbau in Norddeutschland. Es wurden 391 Mikrobeben mit Magnituden zwischen Ml 0.5 und 2.0 erfolgreich lokalisiert und durch Vergleich mit manuell ausgewerteten Lokalisierungen verifziert.Der zweite Datensatz stammt von einem Anwednung auf des Regionale Überwachungsnetz in der Region Campania-Lucania (Süditalien)  mit 33 seismischen Stationen und einer Apertur von etwa 150 km. Wir konnten 196 Erdbeben mit Tiefen < 20 km und Magnituden zwischen Ml 1.1 und 2.7 lokalisieren. Eine Untergruppe der eigenen Lokalisierungen wurde mit den Lokalisierungen einer Standard Lokalisierung sowie einer  hochgenauen Relativlokalisierung verglichen. In beiden Fällen ist die Übereinstimmung mit den manuellen Lokalisierungen groß. Außerdem finden wir, dass die Wellenform Summations Lokalisierung ronbust gegen Rauschen ist und bessere Ergebnisse liefert als die Standard Lokalisierung, die auf dem automatischen Picken von Ersteinsatzzeiten alleine basiert.

Hydrothermal circulation at mid-ocean ridges plays an important role in the interaction between oceanic lithosphere and the overlying ocean. Changes in fluid flux within hydrothermal systems may directly impact ocean circulation, temperature, and chemistry, and hence the lives of biological organisms in hydrothermal vent environments. The permeability structure within a hydrothermal environment is an important control on fluid flow throughout the system. Common suggestions for mechanisms that might increase permeability within the system include thermal cracking due to contraction of the hot rock from interaction with cold seawater, fluid pressure of the water moving through the rock, tectonics, or tidal forces. Additional factors such as mineral precipitation can decrease permeability in the system, further complicating the permeability structure. Though there are many factors to consider within a hydrothermal system, few quantitative studies of these cracking mechanisms exist. This study examines the role of thermal cracking near hydrothermal vents via a numerical model created in Matlab. Flow was modeled using the Laplace equation, and the heat transfer equation was used to determine temperature differences in the rock, which lead to thermal cracking. The numerical results were compared with microearthquakes observed by Tolstoy et al. (2008) near a hydrothermal vent field along the East Pacific Rise. The model suggests that thermal cracking does occur, and this cracking occurs within the area of microearthquakes observed. Though thermal cracking is important for increasing permeability within the system, there are no obvious spatial or temporal trends within the earthquake data that support a direct relation between the modeled thermal cracking and observed earthquakes. It is likely that the observed earthquakes are due to a combination of cracking mechanisms, such as cracking due to fluid pressure, tectonics, or tides, in addition to the modeled thermal stresses.

En géophysique, les réseaux denses améliorent la caractérisation spatiale et fréquentielle des différents types d’ondes dans le milieu. Bien entendu, l’acquisition en surface est sujette aux ondes de surface qui sont très fortes. Les ondes de surface ont un fort impact sur les données géophysiques acquises au niveau du sol. Elles peuvent être considérées comme du bruit et être sujettes à la suppression puisqu’elles cachent l’information de sous-surface. Cependant, elles peuvent être utiles pour l’imagerie de proche surface si elles sont convenablement récupérées.Dans tous les cas, leur caractérisation est cruciale en géophysique d’exploration active et passive. Dans la surveillance microsismique passive, le bruit de surface ambiant est composé d’ondes de surface. L’objectif principal de la surveillance passive est de minimiser l’impact des ondes de surface sur les données microsismiques. Le fort bruit de surface diminue la sensibilité etl’efficacité des méthodes de détection et de localisation. De plus, les méthodes actuelles de localisation et de détection nécessitent généralement la connaissance d’informations telles qu’un un modèle de vitesse ou un modèle d’événement. Dans la sismique active, de fortes ondes de surface sont générés par des sources actives. Les stratégies actuelles de traitement sont généralement basées sur une sélection manuelle des ondes de surface afin de choisir lesquelles garder. Il s’agit là d’une tâche complexe, coûteuse et sujette à interprétation. Cependant, cette tâche est nécessaire pour l’imagerie de proche-surface et de sous-surface. Les ondes de surface peuvent être particulièrement difficiles à récupérer dans des acquisitions clairsemées.Nous proposons d’appliquer les techniques d’interférométrie et de formation de voies (telles que le Matched Field Processing) dans le contexte des réseaux denses. Une densité de traces importante ouvre de nouvelles possibilités dans les traitements géophysiques, qu’ils soient actifs ou passifs. Nous montrons que le bruit ambiant peut être utilisé dans le traitement microsismique pour extraire des informations importantes sur les propriétés du milieu. De plus, nous développons une approche de débruitage qui permet de supprimer les sources de bruit à la surface et détecter les événements microsismiques. Nous proposons également une méthode automatique de détection et de localisation qui se base sur une quantité minimale d’information préalable qui permet de récupérer la distribution des hétérogénéités du réservoir, dans le voisinage du puits. En ce qui concerne la sismique active, nous proposons une approche interférométrique et automatique de caractérisation des ondes de surface. Nous récupérons les noyaux de sensibilité de phase des ondes de surface entre deux points quelconques de l’acquisition. Ces noyaux de sensibilité sont par conséquent utilisés pour obtenir les courbes de dispersion multimodales. Ces courbes de dispersion permettent la séparation des différents modes des ondes de surface, et fournissent l’information de proche surface suite à une simple inversion.Le réseau dense permet l’amélioration des méthodes présentées ci-dessus: elle permet des applications alternatives et innovantes dans le traitement du signal géophysique. / In geophysics, spatially dense arrays enhance the spatial and frequential characterization of the various waves propagating in the medium. Of course, surface array is subject to strong surface waves. Surface waves highly impact the processing of geophysical data acquired at ground level. They can be considered as noise and subject to suppression as they mask sub-surface information.However, they can be useful for near-surface imaging if they are well retrieved. In any case, their characterization is crucial in active and passive exploration geophysics. In passive microseismic monitoring, ambient surface noise consists of surface waves. The main goal of passive monitoring is to minimize the impact of surface waves on the actual microseismic data. The strong ambient surface noise lowers the sensitivity and the efficiency ofdetection and location methods. Moreover, current location and detection methods usually require strong a priori information (e.g., a velocity model or a template).Active sources generate strong surface waves. In active seismic, current processing strategies often consist in manually picking surface wave arrivals in order to use or remove them. This is often a complex, time consuming, and an ambiguous task. However, it is needed for near- and sub-surface imaging. Surface waves can be particularly difficult to retrieve in sparse arrays. We propose to apply the techniques of interferometry and beamforming (Matched Field Processing in particular) in the context of dense arrays. High trace density opens new possibilities in geophysical processing in both passive and active surveys. We show that the ambient noise can be explored in the case of microseismic monitoring to extract important information about the medium properties. Moreover, we develop a denoising approach to remove the noise sources at the surface and detect the microseismic event. Furthermore, we propose an automatic detection and location method with a minimum a priori information to retrieve the distribution of heterogeneities in the reservoir, in the well vicinity.In active survey, we propose an interferometric, automatic approach to characterize the surface waves. We retrieve phase-sensitivity kernels of surface waves between any two points of the acquisition. These kernels are consequently used to obtain multi-mode dispersion curves. These dispersion curves make it possible to separate different modes of surface waves and provide near-surface information if inverted.The above presented methodologies benefit from spatially dense arrays.Dense arrays of sources or receivers enable alternative, innovative applications in geophysical processing.

La multiplication du nombre de réseaux sismiques fait exploser le nombre de données sismologiques. Manuellement, leur traitement est long et fastidieux, d'où la nécessité d'automatiser la détection, la classification et la localisation des événements pour aider les observatoires qui surveillent continuellement la sismicité, mais aussi, dans un intérêt plus scientifique, rechercher et caractériser les phénomènes. La thèse se décompose en 2 axes majeurs : (1) la détection / localisation des séismes, avec le logiciel Waveloc. On a amélioré les outils pré-existants, ajouté de nouvelles fonctionnalités pour une analyse plus détaillée de la sismicité et validé le code avec les données du Piton de la Fournaise ; (2) la classification des séismes. Après calcul des attributs décrivant au mieux les signaux, on a démontré l'efficacité de 2 méthodes d'apprentissage supervisé (régression logistique et SVM) pour le Piton de la Fournaise et soulevé les difficultés pour un cas plus complexe (le Kawah Ijen). / For some time now the quantity of available seismological data has kept increasing. Manually, their processing is long and tedious. Then, the automation of the detection, location and classification of seismic events has become necessary and aims to help the local observatories and to search and characterize some rarer or not well-known phenomena. The work is divided into 2 main directions : (1) the detection and location of seismic events with the Waveloc software (we improved the pre-existing tools, added some new functions for a more detailed analysis of the seimicity and applied the code to data from the Piton de la Fournaise volcano) ; (2) their classification (after computing the seismic attributes, we proved the efficiency and reliability of 2 supervised learning methods - logistic regression and SVM - for the Piton de la Fournaise volcano, underlined the difficulties for a more complex case - the Kawah Ijen volcano - and tried to apply new strategies).

The Cambrian Rogersville Shale is a part of a hydrocarbon system in the Rome Trough of eastern Kentucky and West Virginia that can only be produced unconventionally. In Kentucky, the Rogersville Shale ranges in depth from ~1,800 to ~3,700 m below the surface with the crystalline basement ~1,000 m lower than the formation’s base. Baseline Rome Trough microseismicity data were collected, focusing on wastewater injection wells and recently completed and planned unconventional hydrocarbon test wells in the Rogersville Shale, using thirteen broadband seismic stations installed between June, 2015 and June, 2016 and existing University of Kentucky and central and eastern United States network stations. In addition, the network’s minimum detection threshold, the magnitude at which the theoretical signal exceeds the noise by a factor of 3 between 1 and 20 Hz for at least 4 stations, was estimated for the project area. Thirty-eight local and regional events were located and magnitudes were calculated for each event. No events were proximal to operating disposal or hydrocarbon test wells, nor did any occur in the eastern Kentucky’s Rome Trough. The minimum detection threshold varies between 0.4 and 0.7 Mw from 0000-1100 UTC and 0.6 to 0.9 Mw from 1100-2300 UTC.

Rapid stress release at the source of an earthquake produces seismic waves. Observations of the particle motions from such waves are used in source inversion to characterise the dynamic behaviour of the source and to help in understanding the driving processes. Earthquakes either occur naturally, such as in volcanic eruptions and natural geothermal fields, or are linked to anthropogenic activities including hydrofracture of gas and oil reservoirs, mining events and extraction of geothermal fluids. Source inversion is very sensitive to uncertainties in both the model and the data, especially for low magnitude, namely microseismic, events. Many of the uncertainties can be poorly quantified, and are often not included in source inversion. This thesis proposes a Bayesian framework enabling a complete inclusion of uncertainties in the resultant probability distribution using Bayesian marginalisation. This approach is developed for polarity and amplitude ratio data, although it is possible to use any data type, provided the noise model can be estimated. The resultant posterior probability distributions are easily visualised on different plots for orientation and source-type. Several different algorithms can be used to search the source space, including Monte Carlo random sampling and Markov chain Monte Carlo sampling. Relative information between co-located events may be used as an extension to the framework, improving the constraint on the source. The double-couple source is the commonly assumed source model for many earthquakes, corresponding to slip on a fault plane. Two methods for estimating the posterior model probability of the double-couple source type are explored, one using the Bayesian evidence, the other using trans-dimensional Markov chain Monte Carlo sampling. Results from both methods are consistent with each other, producing good estimates of the probability given sufficient samples. These provide estimates of the probability of the source being a double-couple source or not, which is very useful when trying to understand the processes causing the earthquake. Uncertainty on the polarity estimation is often hard to characterise, so an alternative approach for determining the polarity and its associated uncertainty is proposed. This uses a Bayesian estimate of the polarity probability and includes both the background noise and the arrival time pick uncertainty, resulting in a more quantitative estimate of the polarity uncertainty. Moreover, this automated approach can easily be included in automatic event detection and location workflows. The inversion approach is discussed in detail and then applied to both synthetic events generated using a finite-difference code, and to real events acquired from a temporary seismometer network deployed around the Askja and Krafla Volcanoes, Iceland.

L'étude des signes précurseurs d'un effondrement brutal au-dessus de cavités souterraines, dont le recouvrement est caractérisé par la présence d'un banc massif et raide, est un problème majeur pour la sécurité publique. Aussi, pour progresser dans la compréhension et l'évolution des mécanismes mis en jeu, une cavité saline, située dans le NE de la France, a été suivie en temps réel, entre 2004 et 2009, jusqu'à son effondrement. Celle-ci a été exploitée par dissolution, jusqu'à atteindre une dimension critique (de l'ordre de 180 m) sous un recouvrement de 180 m, armé d'un banc raide de Dolomie situé à 120 m de profondeur.Un système de mesures multi-paramètres haute résolution visait à caractériser les signes précurseurs ainsi qu'à suivre l'effondrement lui-même. Il comprenait des dispositifs à la fois géotechniques et géophysiques, dont des mesures de nivellement de surface et un réseau permanent d'écoute microsismique. Ce dernier, dont les données font l'objet principal de cette thèse, était constitué de neuf sondes équipées de géophones 40 Hz (5 unidirectionnels et 4 tridirectionnels), réparties autour et à l'aplomb de la cavité, dont une dans le banc raide.L'évolution de la cavité a été marquée par deux épisodes majeurs d'activité microsismique :- au printemps 2008, la reprise de la dissolution dans la cavité a engendré l'apparition de crises répétées avec plusieurs milliers d'événements en quelques jours, traduisant un changement de régime microsismique, marqueur de l'instabilité de la cavité ;- en février 2009, suite à ces observations, l'exploitant a décidé de provoquer l'effondrement, par le rabattement intensif de saumure dans la cavité. Pendant les trois jours d'opération, plus de 30000 événements ont été enregistrés (sur 60000 depuis 2004).L'étude de la signature des événements apporte des renseignements essentiels pour la surveillance opérationnelle et la discrimination de ces deux périodes. En particulier, alors que les valeurs maximales atteintes en amplitude, énergie au capteur et fréquence fondamentale apparente, sont assez stables au cours des crises, les sauts marqués durant l'effondrement, permettent de présumer de son imminence. L'évolution de la distribution des microséismes en termes d'énergie libérée et d'occurrence, calculée de manière similaire à la loi de Gutenberg-Richter, bien que souvent difficile à interpréter, a pu être associée à des hausses du niveau piézométrique, ainsi qu'à de petites accélérations de l'affaissement mesuré en surface. Pendant la période d'effondrement, le nombre d'événements microsismiques augmente en suivant une loi en puissance.La localisation des microséismes a nécessité la mise en place d'une stratégie adaptée pour garantir la qualité et l'homogénéité des résultats (sélection des enregistrements, calibrage, étude paramétrique).Cependant, l'utilisation d'un modèle de vitesse constant sur toute la période s'est révélé impossible, compte tenu de l'évolution rapide et permanente du milieu. Il a donc été entrepris d'établir des modèles de vitesse différents en fonction des périodes d'évolution de la cavité.Les distributions spatio-temporelles des foyers ainsi localisés montrent l'existence de structures préférentielles de rupture et souligne le rôle majeur du banc raide.Croisées avec les autres mesures acquises sur le site, ces résultats ont permis d'établir un scénario probable d'évolution de la cavité et de proposer quelques recommandations pour la surveillance opérationnelle / The study of the precursory signs of a brutal collapse above underground caverns, with an overburden characterized by the presence of a massive and stiff bench, is a major problem for public safety. Thus, to progress in the comprehension and the evolution of the concerned mechanisms, a salt cavern, located in the NE France, was monitored in real-time, since 2004 to 2009, until its collapse. This cavern was mined by solution, until reaching its critical dimension (about 180 m) under a covering of 180 m thick, armed with a stiff Dolomite bench located at 120 m of depth.A multi-parameter high resolution monitoring system aimed at characterizing the precursory signs and following collapse itself. It included both geotechnical and geophysical devices as surface leveling measurements and a permanent microseismic network. This one, which data are the principal subject of this thesis, consisted in nine probes equipped with 40 Hz geophones (5 1D and 4 3D), distributed around and directly below the cavern, including one located in the stiff bench.The evolution of the cavern was marked by two major episodes of microseismic activity:- at the beginning of spring 2008, the dissolution restart in the cavern which caused repeated crisis with several thousand events in a few days, this represent a change in the microseismic regime and marked the cavern instability;- in February 2009, following these observations, the owner decided to trigger the collapse by intensive brine pumping in the cavern. During the three days of the operation, more than 30,000 events were recorded (against 60,000 since 2004).The study of the event signature provides essential information for operational monitoring and the discrimination of these two periods. Particularly, while maximal values reached in amplitude, energy and apparent fundamental frequency are quite stable during the 2008 episodes, the rises of this values are important during the collapse period (prior to the peak of activity), allowed us to suppose its imminence. The evolution of the microseism distribution in terms of energy released and occurrence, calculated similarly to the Gutenberg-Richter law, although often difficult to interpret, has been associated with piezometric level rises, and with small accelerations of surface subsidence. During the collapse, the microseismic activity acceleration follows a power law.Microseisms location required the establishment of an appropriate strategy to ensure the quality and the consistency of the results (record selection, calibration, parametric analysis). However, the use of a constant velocity model over all the period was impossible due to the fast and permanent evolution of the environment. Thus, several models were used, according to the a priori known cavern evolution.The event spatiotemporal distributions, thus located, revealed the existence of preferential failure structures and highlight the role of the stiff bench, located at 120 m depth