Extended nucleation phases of earthquakes have been regularly observed, yet the underlying mechanisms governing the initiation phase of rupture are yet to be understood in detail. Currently two end member models exist to explain earthquake nucleation: one model claiming that the nucleation phase of a small earthquake is indistinguishable from that of a large one, while the other proposes fundamental differences in the underlying process. Previous studies have been using the same seismological observations to argue for either model, leaving the need of further investigations into the nucleation behavior of earthquakes across scales and different settings. The thesis at hand contributes to the current discussion on earthquake nucleation by providing additional observational evidence for extended nucleation phases, complex rupture interaction and growth across a number of different scales and settings. Here, earthquake nucleation is investigated for three different scenarios, each with varying degrees of complexity: 1) the controlled case of induced seismicity in hydraulic stimulations of geothermal reservoirs, where rupture growth is assumed to be primarily governed by anthropogenic activity, 2) the partly-controlled setting of a geothermal field with a long history of fluid injection and production, and 3) the uncontrolled case of natural seismicity in the central Sea of Marmara, where earthquake nucleation is purely governed by the regional tectonics.
First, the temporal evolution of seismicity and the growth of observed moment magnitudes for a range of past and present hydraulic stimulation projects associated with the creation of enhanced geothermal systems are analyzed. They reveal a clear linear relation between injected fluid volume/hydraulic energy and cumulative seismic moments. For most projects studied, the observations are in good agreement with existing physical models that predict a relation between injected fluid volume and maximum seismic moment of induced events. This suggests that seismicity results from a stable, pressure controlled rupture process at least for an extended injection period. Overall evolution of seismicity is independent of tectonic stress regime and is most likely governed by reservoir specific parameters, such as the preexisting structural inventory. In contrast, a few stimulations reveal unbound increase in seismic moment suggesting that for these cases evolution of seismicity is mainly controlled by stress field, the size of tectonic faults and fault connectivity. The uncertainty over whether or not a transition between behavior is likely to occur at any point during the injection is what motivates the need for a next generation monitoring and traffic-light system accounting for the possibility of unstable rupture propagation from the very beginning of injection by observing the entire seismicity evolution at high resolution for an immediate reaction in injection strategy. Furthermore, the majority of pressure-controlled stimulations shows the potential of actively controlling the size of induced earthquakes, if an injection protocol is chosen based on continuous feedback from a near-real-time seismic monitoring system. Second, moderate sized earthquakes at The Geysers geothermal field (California), where years of injection and production across hundreds of wells have led to a unique physical environment, are studied. While overall seismicity at The Geysers is generally governed by anthropogenic activities, contributions of individual wells or injection activities are hard to distinguish, thus making detailed managing of occurring magnitudes challenging. New high-resolution seismicity catalogs framing the occurrence of 20 ML > 2.5 earthquakes were created. The seismicity catalogs were developed using a matched filter algorithm, including automatic determination of P and S phase onsets and their inversion for absolute hypocenter locations with corresponding uncertainties. The selected 20 sequences sample different hypocentral depths and hydraulic conditions within the field. Seismic activity and magnitude frequency distributions displayed by the different earthquake sequences are correlated with their location within the reservoir. Sequences located in the northwestern part of the reservoir show overall increased seismic activity and low b values, while the southeastern part is dominated by decreased seismic activity and higher b values. Periods of high injection coincide with high b values and vice versa. These observations potentially reflect varying differential and mean stresses and damage of the reservoir rocks across the field. Additionally, a systematic search for seismicity localization using a multi-step cross-correlation analysis was performed. No evidence for increased correlation between the occurring seismicity and the mainshock for any of the 20 sequences could be seen, indicating that each main nucleation spot was seismically silent prior to the main rupture. However, a number of highly inter-correlated earthquakes for sequences below the reservoir and during high injection activity is observed. Under these conditions, the seismicity surrounding the future mainshock source region is more concentrated and might be evidence for a cascading nucleation process. About 50% of analyzed sequences exhibit no change in seismicity rate in response to the large main event. However, we find complex waveforms at the onset of the main earthquake, suggesting that small ruptures spontaneously grow into or trigger larger events, consistent with a cascading type nucleation.
Third, the spatiotemporal evolution of seismicity during a sequence of moderate (MW4.7 and MW5.8) earthquakes occurring in September 2019 at the transition between a creeping and a locked segment of the North Anatolian Fault in the central Sea of Marmara (Turkey) was analyzed. A matched filter technique was applied to continuous waveforms from the regional network, substantially reducing the magnitude threshold for detection. Sequences of foreshocks preceding the two mainshocks are clearly seen, exhibiting different behaviors: a migration of the seismicity along the entire fault segment on the long-term and a concentration around the epicenters of the large events on the short-term. Suggesting that both seismic and aseismic slip during the foreshock sequences change the stress state on the fault, bringing it closer to failure. Furthermore, the observations also suggest that the MW4.7 event contributed to weaken the fault as part of the preparation process of the MW5.8 earthquake.
Combining the results obtained from different settings, it becomes apparent that, regardless of the tectonic setting and degree of anthropogenic control over the seismicity, there is a wide range of complex nucleation behaviours not yet explained by any of the current models of earthquake nucleation. A simplistic view of earthquake nucleation as either a deterministic or a stochastic process seems inconsistent with the obtained results and fails to account for a more complex nucleation behaviour. Observations from The Geysers and the western Sea of Marmara earthquake sequence, suggest that both cascade triggering and aseismic slip can play major roles in the nucleation of moderate sized earthquakes. Both mechanisms seem to jointly contribute to fault initiation, even within the same rock volume. A separation of the two mechanisms can potentially be thought of at The Geysers, where cascade triggering seems to dominate in highly damaged parts of the reservoir, suggesting that the anthropogenic activity can at least partially influence the nucleation behavior of the occurring seismicity. This would be in agreement with the results obtained from analysis of hydraulic stimulations, where during the pressure-controlled phase of injection rupture growth is controlled by the injected fluid.