Nicht-Scheranteile im seismischen Momententensor von Nachbeben und Labor- Erdbeben

Recent developments in waveform data-acquisition technology in combination with improved methods of modelling of seismic waves have enabled us to study the details of physical processes in earthquake sources with an unprecedented accuracy. This has revealed that the common assumption that earthquakes represent shear faulting on a planar rupture in an isotropic Earth’s crust is too simplistic and does not explain data adequately. The complexities of the earthquake process are, in particular, evidenced by the presence of non- double-couple (non-DC) components frequently observed in seismic moment tensors. Early studies showed, however, that non-DC components are difficult to determine accurately being often contaminated by numerical errors. The non- DC components are sensitive to noise in the input data, to errors in the modelling procedure, to a velocity model, to an earthquake location and to the amount of data in terms of the focal sphere coverage. In addition, their physical origin covers a broad range of possibilities, which further complicates the interpretation of the non-DC components. Therefore, high- quality data with good focal sphere coverage and detailed knowledge of the medium are essential for determining reliable non-DC components. This thesis deals with the determination and error analysis of the non-DC components of seismic moment tensors. In the thesis, the state-of-the-art moment tensor inversion algorithms are refined and extensive synthetic tests are performed in order to study the uncertainties and resolvability of the non-DC components. By analysing high-quality waveform data of earthquakes on various scales, the non-DC components are retrieved and suggestions for their physical origin are given. On the field scale, the aftershock sequence of the Mw=7.4 1999 Izmit earthquake in northwestern Turkey is analysed. The analysed earthquake catalogue contains waveforms of more than 4000 relative relocated events observed at 35 three-component short-period seismic stations surrounding the rupture zone. The Izmit earthquake is one of the best studied M>7 earthquakes worldwide and therefore detailed knowledge of the tectonic setting is available. To analyse the sensitivity of the given network to detecting source-related non-DC components, a synthetic case study is performed. In addition, two different moment tensor inversion approaches were tested – a linear full moment tensor inversion and a non-linear moment tensor inversion, which is constrained to the shear-tensile source model. A synthetic dataset mimicking seismic observations of aftershock recordings of the 1999 Izmit earthquake was modelled, adopting the shear-tensile source model. To analyse the resolution capability of the network, synthetic amplitudes were contaminated with artificial noise. In addition realistic errors in the location as well as in the velocity model were considered. The synthetic tests revealed that the error bounds for the ISO and the CLVD component remain considerable at approximately ±15%, while the orientations of P- and T-axes are well determined even when errors in the modelling procedure are high. In general, the tests showed that the aftershock recordings lack observations with near-epicentral distances (<15 km), which causes irregular focal sphere coverage. Thus, finding stable moment tensor solutions is a difficult task for the given network geometry. However, the non-linear inversion is constrained to a shear-tensile source model, which yields significantly smaller errors in the non-DC components compared to the full moment tensor inversion. In particular, the errors in the CLVD component are reduced. After completing the synthetic tests, the analysis of aftershock recordings was performed by inverting P- and S-wave amplitudes for the moment tensor. Prior to the inversion, strict selection criteria were applied to the event catalogue reducing the number of events from more than 4000 to only 33. This is mostly due to the poor focal sphere coverage. The 33 moment tensors display significant differences in the percentage of the non-DC components for the three distinct fault segments in the study area: the Izmit-Sapanca, Karadere- Düzce and the Akyazi segments. Events located in the Izmit-Sapanca and Karadere-Düzce segments show mainly strike-slip mechanisms with rather low percentages of the non-DC components, which are mainly positive if present. This correlates well with the predominant pure shear, strike-slip stress regime along this segment and also with the main shock rupture comprising a right lateral strike-slip. In contrast, we found substantial percentages of the non-DC components for events below the Akyazi Plain, which is a pull-apart structure. The observed non-DC components are entirely positive, indicating a tensional regime, and range from 20 to 48%, clearly exceeding the defined error bounds assessed from the synthetic case study. This observation is in accordance with the post-seismic setting following the Izmit main shock that left a remarkable slip deficit of 3.5 m below the Akyazi bend. The tension throughout the seismogenic layer might have allowed fluids to migrate upwards, thereby reducing effective normal traction (or friction) on coseismically stressed normal faults and accelerating the compensation of the slip deficit. On the laboratory scale, a newly developed moment tensor inversion approach was implemented and applied to acoustic emission waveform recordings. During rock deformation experiments in the laboratory with differential loading, the seismic velocities become highly anisotropic and waves were attenuated due to opening or closure of microcracks. In this part of the thesis, the sensitivity of the moment tensor inversion to anisotropy of P-wave velocities and attenuation of the rock sample was studied. Neglecting attenuation and anisotropy of P-wave velocities in the moment tensor inversion can produce significant uncertainties in the retrieved moment tensors and may lead to misinterpretations of the source mechanisms. Interestingly, the sensitivity depends on the fracturing mode of AEs. The tensile events are more sensitive to anisotropy of the P-wave velocity and attenuation of the sample than the shear events. Furthermore, the geometry of faulting in anisotropic rock samples should be studied using the source tensors, since the P and T axes of the moment tensors are affected by anisotropy of the P-wave velocity and deviate from the true orientation of faulting. The stronger the anisotropy the higher the deviations are. Finally, the moment tensor inversion was applied to a large dataset of AEs to find optimum values of the attenuation parameters of the rock sample in a grid-search scheme. The resolution of the method is sufficiently high for quantitatively distinguishing anisotropic behaviour of attenuation of P-waves. The method is capable of measuring differences between the P-wave attenuation in the horizontal and vertical directions in the sample and allows detection of the effects of dilatant cracking according to the stress regime. The results obtained here for datasets on the laboratory as well as on the field scale demonstrate the challenges in reliably determining and interpreting the non-DC components. The analysis revealed the importance of evaluating the non-DC components for studying fracture modes during earthquake processes and for understanding their driving mechanisms. In addition, the analysis showed that a joint inversion for moment tensors of a family of earthquakes is capable of providing not only information on seismic sources but also on seismic anisotropy and attenuation in the focal zone.Neuste Entwicklungen in dem Bereich der Akquisition von seismischen Wellenfelddaten in Kombination mit verbesserten Konzepten der Momententensor- Inversion bilden die Basis für die vorliegende Dissertation. In den letzten Jahrzehnten hat sich in der seismologischen Fachwelt die Überzeugung durchgesetzt, dass die Annahme eines reinen Scherbruchs nicht ausreicht, um z.B. signifikante Nicht-Scheranteile im Herdprozess zu beschreiben. Erste Studien dazu haben ergeben, dass die Nicht-Scheranteile sich sensitiv gegenüber Rauschen in den Daten, Fehlern in der Modellierung und der Herdflächenüberdeckung verhalten. Daher sind hochqualitative Daten mit einer optimalen Herdflächenüberdeckung und detaillierte Kenntnisse über das durchlaufene Medium essentiell für eine zuverlässige Bestimmung von Nicht- Scheranteilen. Die vorliegende Dissertation beschäftigt sich mit der Bestimmung und Analyse von Nicht-Scheranteilen auf zwei verschiedenen Skalen, um deren Auflösung und physikalischen Ursprung unter Verwendung von hochqualitativen Daten und der Weiterentwicklung von Momententensor- Inversionsalgorithmen zu untersuchen. Auf der Feld-Skala wurde die Nachbebenserie des Mw=7.4 Izmit Erdbebens von 1999 im Nordwesten der Türkei analysiert. Der Erdbebenkatalog umfasst mehr als 4000 relokalisierte Erdbeben, welche von einem 35 Stationen-Netzwerk, ausgestattet mit kurzperiodischen, 3-Komponenten Seismometern, aufgezeichnet wurden. Vor der Analyse der Nachbeben in Bezug auf das Auftreten von möglichen Nicht-Scheranteilen wurde eine umfangreiche synthetische Studie durchgeführt, um das Auflösungsvermögen des Netzwerkes zur Erfassung von quellgebundenen Nicht-Scheranteile zu analysieren. Weiterhin wurden zwei verschiedene Inversionsalgorithmen getestet – eine lineare Momententensor-Inversion, in welcher der volle Momententensor bestimmt wird, und eine nichtlineare Momententensor-Inversion, welche auf das Model des kombinierten Scher-Zugbruches beschränkt ist. Dazu wurde ein synthetischer Datensatz generiert, welcher seismische Beobachtungen des Izmit- Erdbebens mit dem Modell des Scher-Zugbruches simuliert. Um das Auflösungsvermögen des Netzwerkes zu untersuchen, wurden die synthetischen Amplituden mit zufällig verteiltem Rauschen überlagert und es wurden weiterhin Fehler in den Lokalisierungen und Fehler im Geschwindigkeitsmodell eingebaut. Die synthetische Studie hat ergeben, dass die Fehlergrenzen der Nicht- Scheranteile beträchtlich sind und annähernd ±15% betragen. Die Ausrichtung der P- und T-Achsen dagegen ist gut bestimmt auch für den Fall verhältnismäßig großer Fehler in der Modellierung. Die nichtlineare Inversion, welche auf den Scher-Zugbruch beschränkt ist, ergibt kleinere Fehler in den Nicht- Scheranteilen als die lineare Momententensor-Inversion, welche den vollen Momententensor bestimmt. Insbesondere Fehler in der CLVD (engl. compensated linear vector dipole) Komponente können deutlich reduziert werden. Insgesamt hat die Studie gezeigt, dass viele der Nachbebenaufzeichungen eine schlechte Überdeckung der Herdfläche aufweisen, da Beobachtungen mit kurzen epizentralen Abständen (