Robust regional moment tensor determination in complex geological settings: Quantifying uncertainties and exploring Bayesian approaches

Earthquakes are transient failures of the natural state of the Earth, generated by a sudden release of accumulated stress. They manifest themself on the Earth's surface as ground motions that may result in significant damage to the socio-economic landscape. We use seismometers to record these ground motions and analyse them with physical and mathematical methods in modern high-preformance-computations. Derived models give us an valuable insight into the Earth's structure and the physical processes that lead to earthquakes. Understanding its source mechanism is crucial to quantify the potential geological hazard of a region and the local stress regime.

The main goal of this study is the quantative description of the earthquake source mechanism in a moment tensor (MT) representation. The underlying point source approximation of the earthquake is a robust standard approach in seismology. The equivalent forces fully describe the generated wavefield that interacts with the Earth's structure and causes motions at its surface. State-of-the-art MT modeling now often incorporates a Bayesian formulation of associated errors in the data and model assumptions. ... mehrFor this thesis, I created a new python-based MT inversion routine \enquote{Am$\Phi$B - uniXtree} that uses a Bayesian error formulation to mitigate the effects of uncertainties due to different sources of errors.

My area of interest is the Lesser Antilles Arc (LAA) subduction zone in the eastern Caribbean Sea, where I use data recorded on ocean bottom seismometers (OBS). OBS data are affected by strong seismic noise and issues related to the sensor installation, notably mostly affecting the horizontal recordings. Waveform data have so far been often discarded or limited for source modeling. In this thesis, the complete three components of the recorded waveforms are used for the first time in an application for MT inversion. With the newly derived source mechanisms, the regional MT database could be expanded significantly and a detailed analysis of the stress regime could be given.

Additionally, based on this data, this study shed light on the 2007 intermediate-depth $M_\text{w}$7.4 Martinique earthquake in the central parts of the LAA, which is hinted to have had a complex source process in all existing MT solutions. Furthermore, the location and size of this earthquake are not only unusual within the LAA but also in comparison to the global seismic intermediate-depth range (70-300~km).

I conducted several additional studies in collaboration. Two of them are featured in this thesis and used to benchmark my inversion code and examine further prominent error sources of MT solutions. For the 2020 Monte Christo Range earthquake in Nevada, USA, I extended my inversion routine to incorporate two individual faults. For an induced earthquake sequence near Blackpool, UK, I added a restricted cross-correlation time update to counter arrival time uncertainties between modeled and observed data.

My study demonstrates that we can use valuable OBS data for MT source modeling if we quantify prominent error sources in a Bayesian approach to mitigate potential effects on the source solution. The newly collected source mechanism data in the LAA offered a detailed insight into the tectonic setting and highlighted the source complexity of the Martinique earthquake, which I interpret as a source doublet consisting of two different faulting styles on an ancient subducted mid-oceanic ridge transform structure.