Refine
Document Type
- Doctoral Thesis (5)
- Article (3)
- Master's Thesis (2)
- Bachelor Thesis (1)
- Habilitation Thesis (1)
Keywords
- seismology (12) (remove)
Institute
- Institut für Geowissenschaften (12) (remove)
We present an approach for rapidly estimating full moment tensors of earthquakes and their parameter uncertainties based on short time windows of recorded seismic waveform data by considering deep learning of Bayesian Neural Networks (BNNs). The individual neural networks are trained on synthetic seismic waveform data and corresponding known earthquake moment-tensor parameters. A monitoring volume has been predefined to form a three-dimensional grid of locations and to train a BNN for each grid point. Variational inference on several of these networks allows us to consider several sources of error and how they affect the estimated full moment-tensor parameters and their uncertainties. In particular, we demonstrate how estimated parameter distributions are affected by uncertainties in the earthquake centroid location in space and time as well as in the assumed Earth structure model. We apply our approach as a proof of concept on seismic waveform recordings of aftershocks of the Ridgecrest 2019 earthquake with moment magnitudes ranging from Mw 2.7 to Mw 5.5. Overall, good agreement has been achieved between inferred parameter ensembles and independently estimated parameters using classical methods. Our developed approach is fast and robust, and therefore, suitable for down-stream analyses that need rapid estimates of the source mechanism for a large number of earthquakes.
The mechanisms leading to large earthquakes are poorly understood and documented. Here we characterize the long-term precursory phase of the 1 April 2014 M(w)8.1 North Chile megathrust. We show that a group of coastal GPS stations accelerated westward 8months before the main shock, corresponding to a M(w)6.5 slow slip event on the subduction interface, 80% of which was aseismic. Concurrent interface foreshocks underwent a diminution of their radiation at high frequency, as shown by the temporal evolution of Fourier spectra and residuals with respect to ground motions predicted by recent subduction models. Such ground motions change suggests that in response to the slow sliding of the subduction interface, seismic ruptures are progressively becoming smoother and/or slower. The gradual propagation of seismic ruptures beyond seismic asperities into surrounding metastable areas could explain these observations and might be the precursory mechanism eventually leading to the main shock.
A comprehensive study on seismic hazard and earthquake triggering is crucial for effective mitigation of earthquake risks. The destructive nature of earthquakes motivates researchers to work on forecasting despite the apparent randomness of the earthquake occurrences. Understanding their underlying mechanisms and patterns is vital, given their potential for widespread devastation and loss of life. This thesis combines methodologies, including Coulomb stress calculations and aftershock analysis, to shed light on earthquake complexities, ultimately enhancing seismic hazard assessment.
The Coulomb failure stress (CFS) criterion is widely used to predict the spatial distributions of aftershocks following large earthquakes. However, uncertainties associated with CFS calculations arise from non-unique slip inversions and unknown fault networks, particularly due to the choice of the assumed aftershocks (receiver) mechanisms. Recent studies have proposed alternative stress quantities and deep neural network approaches as superior to CFS with predefined receiver mechanisms. To challenge these propositions, I utilized 289 slip inversions from the SRCMOD database to calculate more realistic CFS values for a layered-half space and variable receiver mechanisms. The analysis also investigates the impact of magnitude cutoff, grid size variation, and aftershock duration on the ranking of stress metrics using receiver operating characteristic (ROC) analysis. Results reveal the performance of stress metrics significantly improves after accounting for receiver variability and for larger aftershocks and shorter time periods, without altering the relative ranking of the different stress metrics.
To corroborate Coulomb stress calculations with the findings of earthquake source studies in more detail, I studied the source properties of the 2005 Kashmir earthquake and its aftershocks, aiming to unravel the seismotectonics of the NW Himalayan syntaxis. I simultaneously relocated the mainshock and its largest aftershocks using phase data, followed by a comprehensive analysis of Coulomb stress changes on the aftershock planes. By computing the Coulomb failure stress changes on the aftershock faults, I found that all large aftershocks lie in regions of positive stress change, indicating triggering by either co-seismic or post-seismic slip on the mainshock fault.
Finally, I investigated the relationship between mainshock-induced stress changes and associated seismicity parameters, in particular those of the frequency-magnitude (Gutenberg-Richter) distribution and the temporal aftershock decay (Omori-Utsu law). For that purpose, I used my global data set of 127 mainshock-aftershock sequences with the calculated Coulomb Stress (ΔCFS) and the alternative receiver-independent stress metrics in the vicinity of the mainshocks and analyzed the aftershocks properties depend on the stress values. Surprisingly, the results show a clear positive correlation between the Gutenberg-Richter b-value and induced stress, contrary to expectations from laboratory experiments. This observation highlights the significance of structural heterogeneity and strength variations in seismicity patterns. Furthermore, the study demonstrates that aftershock productivity increases nonlinearly with stress, while the Omori-Utsu parameters c and p systematically decrease with increasing stress changes. These partly unexpected findings have significant implications for future estimations of aftershock hazard.
The findings in this thesis provides valuable insights into earthquake triggering mechanisms by examining the relationship between stress changes and aftershock occurrence. The results contribute to improved understanding of earthquake behavior and can aid in the development of more accurate probabilistic-seismic hazard forecasts and risk reduction strategies.
The purpose of this thesis is to develop an automated inversion scheme to derive point and finite source parameters for weak earthquakes, here intended with the unusual meaning of earthquakes with magnitudes at the limit or below the bottom magnitude threshold of standard source inversion routines. The adopted inversion approaches entirely rely on existing inversion software, the methodological work mostly targeting the development and tuning of optimized inversion flows. The resulting inversion scheme is tested for very different datasets, and thus allows the discussion on the source inversion problem at different scales. In the first application, dealing with mining induced seismicity, the source parameters determination is addressed at a local scale, with source-sensor distance of less than 3 km. In this context, weak seismicity corresponds to event below magnitude MW 2.0, which are rarely target of automated source inversion routines. The second application considers a regional dataset, namely the aftershock sequence of the 2010 Maule earthquake (Chile), using broadband stations at regional distances, below 300 km. In this case, the magnitude range of the target aftershocks range down to MW 4.0. This dataset is here considered as a weak seismicity case, since the analysis of such moderate seismicity is generally investigated only by moment tensor inversion routines, with no attempt to resolve source duration or finite source parameters. In this work, automated multi-step inversion schemes are applied to both datasets with the aim of resolving point source parameters, both using double couple (DC) and full moment tensor (MT) models, source duration and finite source parameters. A major result of the analysis of weaker events is the increased size of resulting moment tensor catalogues, which interpretation may become not trivial. For this reason, a novel focal mechanism clustering approach is used to automatically classify focal mechanisms, allowing the investigation of the most relevant and repetitive rupture features. The inversion of the mining induced seismicity dataset reveals the repetitive occurrence of similar rupture processes, where the source geometry is controlled by the shape of the mined panel. Moreover, moment tensor solutions indicate a significant contribution of tensile processes. Also the second application highlights some characteristic geometrical features of the fault planes, which show a general consistency with the orientation of the slab. The additional inversion for source duration allowed to verify the empirical correlation for moment normalized earthquakes in subduction zones among a decreasing rupture duration with increasing source depth, which was so far only observed for larger events.
The Vogtland, located at the border region between the Czech Republic and Germany, is known for Holocene volcanism, gas and fluid emissions as well as for reoccurring earthquake swarms, pointing towards a high geodynamic activity. During the earthquake swarm in 2008/2009, a temporary array was installed close to Rohrbach (Germany), at an epicentral distance of about 10 km from the Nový Kostel focal zone (aperture ~0.75 km).
22 events of the recorded swarm were selected to set up a source array. Source arrays are spatially clustered earthquakes, which can be used in a similar manner as receiver array recordings of single events (Green’s functions reciprocity). The application of array seismology techniques like beam forming requires similar waveforms and precisely known origin times and locations. The resemblance of waveforms was assured by visual selection of events and quantified with the calculation of cross-correlation coefficients. We observed that the different events recorded at a single station generally show greater resemblances than the recordings of one event at all stations of the receiver array. This indicates a heterogeneous subsurface beneath the receiver array and a comparably homogeneous source array volume with respect to the frequency-dependent resolution of both arrays.
Beam forming was applied on the Z, N and E component recordings of the source array events at 11 stations, and the results were analysed with respect to converted or reflected crustal phases. While the theoretical back azimuth of the direct phases match the beam forming results in case of the source array analysis, in case of receiver array beam forming derivations of 15°-25° are observed.
PS phases, closely following the direct P phase and presumably SP phases, arriving shortly before the direct S phase can be observed on several stations. Based on the time differences to the direct P and S phases we inferred a conversion depth of about 0.6-0.9 km. A second deeper source array was set up in order to interpret a structural phase arriving 0.85 s after the direct P phase on records of deeper events only.
Additionally to the source array beam forming method an analytical method with a fixed medium velocity and a grid search method, both for determining conversion/ reflection locations of phases traveling off the direct line between source and receiver array, were developed and applied to other observed phases.
In conclusion, we think that the distinct beam forming results along with the striking waveform resemblance reveal the opportunities of using source arrays consisting of small swarm events for the analysis of crustal structures.
Alfred Wegeners ideas on continental drift were doubted for several decades until the discovery of polarization changes at the Atlantic seafloor and the seismic catalogs imaging oceanic subduction underneath the continental crust (Wadati-Benioff Zone). It took another 20 years until plate motion could be directly observed and quantified by using space geodesy. Since then, it is unthinkable to do neotectonic research without the use of satellite-based methods.
Thanks to a tremendeous increase of instrumental observations in space and time over the last decades we significantly increased our knowledge on the complexity of the seismic cycle, that is, the interplay of tectonic stress build up and release. Our classical assumption, earthquakes were the only significant phenomena of strain release previously accumulated in a linear fashion, is outdated. We now know that this concept is actually decorated with a wide range of slow and fast processes such as triggered slip, afterslip, post-seismic and visco-elastic relaxation of the lower crust, dynamic pore-pressure changes in the elastic crust, aseismic creep, slow slip events and seismic swarms. On the basis of eleven peer-reviewed papers studies I here present the diversity of crustal deformation processes. Based on time-series analyses of radar imagery and satellited-based positioning data I quantify tectonic surface deformation and use numerical and analytical models and independent geologic and seismologic data to better understand the underlying crustal processes.
The main part of my work focuses on the deformation observed in the Pamir, the Hindu Kush and the Tian Shan that together build the highly active continental collision zone between Northwest-India and Eurasia. Centered around the Sarez earthquake that ruptured the center of the Pamir in 2015 I present diverse examples of crustal deformation phenomena. Driver of the deformation is the Indian indenter, bulldozing into the Pamir, compressing the orogen that then collapses westward into the Tajik depression. A second natural observatory of mine to study tectonic deformation is the oceanic subduction zone in Chile that repeatedly hosts large earthquakes of magnitude 8 and more. These are best to study post-seismic relaxation processes and coupling of large earthquake.
My findings nicely illustrate how complex fashion and how much the different deformation phenomena are coupled in space and time. My publications contribute to the awareness that the classical concept of the seismic cycle needs to be revised, which, in turn, has a large influence in the classical, probabilistic seismic hazard assessment that primarily relies on statistically solid recurrence times.
Earthquake modeling is the key to a profound understanding of a rupture. Its kinematics or dynamics are derived from advanced rupture models that allow, for example, to reconstruct the direction and velocity of the rupture front or the evolving slip distribution behind the rupture front. Such models are often parameterized by a lattice of interacting sub-faults with many degrees of freedom, where, for example, the time history of the slip and rake on each sub-fault are inverted. To avoid overfitting or other numerical instabilities during a finite-fault estimation, most models are stabilized by geometric rather than physical constraints such as smoothing.
As a basis for the inversion approach of this study, we build on a new pseudo-dynamic rupture model (PDR) with only a few free parameters and a simple geometry as a physics-based solution of an earthquake rupture. The PDR derives the instantaneous slip from a given stress drop on the fault plane, with boundary conditions on the developing crack surface guaranteed at all times via a boundary element approach. As a side product, the source time function on each point on the rupture plane is not constraint and develops by itself without additional parametrization. The code was made publicly available as part of the Pyrocko and Grond Python packages. The approach was compared with conventional modeling for different earthquakes. For example, for the Mw 7.1 2016 Kumamoto, Japan, earthquake, the effects of geometric changes in the rupture surface on the slip and slip rate distributions could be reproduced by simply projecting stress vectors. For the Mw 7.5 2018 Palu, Indonesia, strike-slip earthquake, we also modelled rupture propagation using the 2D Eikonal equation and assuming a linear relationship between rupture and shear wave velocity. This allowed us to give a deeper and faster propagating rupture front and the resulting upward refraction as a new possible explanation for the apparent supershear observed at the Earth's surface.
The thesis investigates three aspects of earthquake inversion using PDR: (1) to test whether implementing a simplified rupture model with few parameters into a probabilistic Bayesian scheme without constraining geometric parameters is feasible, and whether this leads to fast and robust results that can be used for subsequent fast information systems (e.g., ground motion predictions). (2) To investigate whether combining broadband and strong-motion seismic records together with near-field ground deformation data improves the reliability of estimated rupture models in a Bayesian inversion. (3) To investigate whether a complex rupture can be represented by the inversion of multiple PDR sources and for what type of earthquakes this is recommended.
I developed the PDR inversion approach and applied the joint data inversions to two seismic sequences in different tectonic settings. Using multiple frequency bands and a multiple source inversion approach, I captured the multi-modal behaviour of the Mw 8.2 2021 South Sandwich subduction earthquake with a large, curved and slow rupturing shallow earthquake bounded by two faster and deeper smaller events. I could cross-validate the results with other methods, i.e., P-wave energy back-projection, a clustering analysis of aftershocks and a simple tsunami forward model.
The joint analysis of ground deformation and seismic data within a multiple source inversion also shed light on an earthquake triplet, which occurred in July 2022 in SE Iran. From the inversion and aftershock relocalization, I found indications for a vertical separation between the shallower mainshocks within the sedimentary cover and deeper aftershocks at the sediment-basement interface. The vertical offset could be caused by the ductile response of the evident salt layer to stress perturbations from the mainshocks.
The applications highlight the versatility of the simple PDR in probabilistic seismic source inversion capturing features of rather different, complex earthquakes. Limitations, as the evident focus on the major slip patches of the rupture are discussed as well as differences to other finite fault modeling methods.
Traveltime residuals for worldwide seismic stations are calculated. We use P and S waves from earthquakes in SE-Asia at teleseismic and regional distances. The obtained station residuals help to enhance earthquake localisation. Furthermore we calculated regional source dependent station residuals. They show a systematic dependence of the locality of the source. These source dependent residuals reflect heterogenities along the path and can be used for a refinement of earthquake localisation.
Rapidly growing seismic and macroseismic databases and simplified access to advanced machine learning methods have in recent years opened up vast opportunities to address challenges in engineering and strong motion seismology from novel, datacentric perspectives. In this thesis, I explore the opportunities of such perspectives for the tasks of ground motion modeling and rapid earthquake impact assessment, tasks with major implications for long-term earthquake disaster mitigation.
In my first study, I utilize the rich strong motion database from the Kanto basin, Japan, and apply the U-Net artificial neural network architecture to develop a deep learning based ground motion model. The operational prototype provides statistical estimates of expected ground shaking, given descriptions of a specific earthquake source, wave propagation paths, and geophysical site conditions. The U-Net interprets ground motion data in its spatial context, potentially taking into account, for example, the geological properties in the vicinity of observation sites. Predictions of ground motion intensity are thereby calibrated to individual observation sites and earthquake locations.
The second study addresses the explicit incorporation of rupture forward directivity into ground motion modeling. Incorporation of this phenomenon, causing strong, pulse like ground shaking in the vicinity of earthquake sources, is usually associated with an intolerable increase in computational demand during probabilistic seismic hazard analysis (PSHA) calculations. I suggest an approach in which I utilize an artificial neural network to efficiently approximate the average, directivity-related adjustment to ground motion predictions for earthquake ruptures from the 2022 New Zealand National Seismic Hazard Model. The practical implementation in an actual PSHA calculation demonstrates the efficiency and operational readiness of my model. In a follow-up study, I present a proof of concept for an alternative strategy in which I target the generalizing applicability to ruptures other than those from the New Zealand National Seismic Hazard Model.
In the third study, I address the usability of pseudo-intensity reports obtained from macroseismic observations by non-expert citizens for rapid impact assessment. I demonstrate that the statistical properties of pseudo-intensity collections describing the intensity of shaking are correlated with the societal impact of earthquakes. In a second step, I develop a probabilistic model that, within minutes of an event, quantifies the probability of an earthquake to cause considerable societal impact. Under certain conditions, such a quick and preliminary method might be useful to support decision makers in their efforts to organize auxiliary measures for earthquake disaster response while results from more elaborate impact assessment frameworks are not yet available.
The application of machine learning methods to datasets that only partially reveal characteristics of Big Data, qualify the majority of results obtained in this thesis as explorative insights rather than ready-to-use solutions to real world problems. The practical usefulness of this work will be better assessed in the future by applying the approaches developed to growing and increasingly complex data sets.
Die genauen Einsatzzeiten seismischer P-Phasen von Erdbeben werden in SeisComP3 und anderen Auswerteprogrammen standardmäßig und in Echtzeit automatisch bestimmt. S-Phasen stellen dagegen eine weit größere Herausforderung dar. Nur mit genauen Picks der P- bzw. S-Phasen können die Erdbebenlokationen korrekt und stabil bestimmt werden. Darum besteht erhebliches Interesse, diese mit hoher Genauigkeit zu bestimmen. Das Ziel der vorliegenden Bachelorarbeit war es, vier verschiedene, bereits vorhandene S-Phasenpicker auf ausgewählte Parameter optimal zu konfigurieren, auf Testdaten anzuwenden und deren Leistungsfähigkeit objektiv zu bewerten. Dazu wurden ein S-Picker (S-L2) aus dem OpenSource SeisComp3-Programmpaket, zwei S-Picker (S-AIC, S-AIC-V) als kommerzielles Modul der Firma gempa GmbH für SeisComP3 und ein S-Picker (Frequenzband) aus dem OpenSource PhasePaPy-Paket ausgewählt. Die Bewertung erfolgte durch Vergleich automatischer Picks mit manuell bestimmten Einsatzzeiten. Alle vier Picker wurden separat konfiguriert und auf drei verschiedene Datensätze von Erdbeben in N-Chile und im Vogtland, Deutschland, angewandt. Dazu wurden regional bzw. lokal typische Erdbeben zufällig ausgewählt und die P- und S-Phasen manuell bestimmt. Mit den zu testenden S-Pickeralgorithmen wurden dieselben Daten durchsucht und die Picks automatisch bestimmt. Die Konfigurationen der Picker wurden gleichzeitig automatisch und objektiv durch iterative Anpassung optimiert. Ein neu erstelltes Bewertungssystem vergleicht die manuellen und die automatisch gefundenen S-Picks anhand von definierten Qualitätsfaktoren. Die Qualitätsfaktoren sind: der Mittelwert und die Standardabweichung der zeitlichen Differenzen zwischen den S-Picks, die Anzahl an übereinstimmenden S-Picks, die Prozentangaben über mögliche S-Picks und die benötigt Rechenzeit. Die objektive Bewertung erfolgte anhand eines Scores. Der Scorewert ergibt sich aus der gewichteten Summe folgender normierter Qualitätsfaktoren: Standardabweichung (20%), Mittelwert (20%) und Prozentangabe über mögliche S-Picks (60%). Konfigurationen mit hohem Score werden bevorzugt. Die bevorzugten Konfigurationen der verschiedenen Picker wurden miteinander verglichen, um den am besten geeigneten S-Pickeralgorithmus zu bestimmen. Allgemein zeigt sich, dass der S-AIC Picker für jeden der drei Datensätze die höchsten Scores und damit die besten Ergebnisse liefert. Dabei wurde für jeden Datensatz ein andere Konfiguration der Parameter des S-AIC Pickers als die am besten geeignete bezeichnet. Daher ist für jede Erdbebenregion eine andere Konfigurationen erforderlich, um optimale Ergebnisse mit diesem S-Picker zu bekommen.
The shallow Earth’s layers are at the interplay of many physical processes: some being driven by atmospheric forcing (precipitation, temperature...) whereas others take their origins at depth, for instance ground shaking due to seismic activity. These forcings cause the subsurface to continuously change its mechanical properties, therefore modulating the strength of the surface geomaterials and hydrological fluxes. Because our societies settle and rely on the layers hosting these time-dependent properties, constraining the hydro-mechanical dynamics of the shallow subsurface is crucial for our future geographical development. One way to investigate the ever-changing physical changes occurring under our feet is through the inference of seismic velocity changes from ambient noise, a technique called seismic interferometry. In this dissertation, I use this method to monitor the evolution of groundwater storage and damage induced by earthquakes. Two research lines are investigated that comprise the key controls of groundwater recharge in steep landscapes and the predictability and duration of the transient physical properties due to earthquake ground shaking. These two types of dynamics modulate each other and influence the velocity changes in ways that are challenging to disentangle. A part of my doctoral research also addresses this interaction. Seismic data from a range of field settings spanning several climatic conditions (wet to arid climate) in various seismic-prone areas are considered. I constrain the obtained seismic velocity time-series using simple physical models, independent dataset, geophysical tools and nonlinear analysis. Additionally, a methodological development is proposed to improve the time-resolution of passive seismic monitoring.
Assuming that liquid iron alloy from the outer core interacts with the solid silicate-rich lower mantle the influence on the core-mantle reflected phase PcP is studied. If the core-mantle boundary is not a sharp discontinuity, this becomes apparent in the waveform and amplitude of PcP. Iron-silicate mixing would lead to regions of partial melting with higher density which in turn reduces the velocity of seismic waves. On the basis of the calculation and interpretation of short-period synthetic seismograms, using the reflectivity and Gauss Beam method, a model space is evaluated for these ultra-low velocity zones (ULVZs). The aim of this thesis is to analyse the behaviour of PcP between 10° and 40° source distance for such models using different velocity and density configurations. Furthermore, the resolution limits of seismic data are discussed. The influence of the assumed layer thickness, dominant source frequency and ULVZ topography are analysed. The Gräfenberg and NORSAR arrays are then used to investigate PcP from deep earthquakes and nuclear explosions. The seismic resolution of an ULVZ is limited both for velocity and density contrasts and layer thicknesses. Even a very thin global core-mantle transition zone (CMTZ), rather than a discrete boundary and also with strong impedance contrasts, seems possible: If no precursor is observable but the PcP_model /PcP_smooth amplitude reduction amounts to more than 10%, a very thin ULVZ of 5 km with a first-order discontinuity may exist. Otherwise, if amplitude reductions of less than 10% are obtained, this could indicate either a moderate, thin ULVZ or a gradient mantle-side CMTZ. Synthetic computations reveal notable amplitude variations as function of the distance and the impedance contrasts. Thereby a primary density effect in the very steep-angle range and a pronounced velocity dependency in the wide-angle region can be predicted. In view of the modelled findings, there is evidence for a 10 to 13.5 km thick ULVZ 600 km south-eastern of Moscow with a NW-SE extension of about 450 km. Here a single specific assumption about the velocity and density anomaly is not possible. This is in agreement with the synthetic results in which several models create similar amplitude-waveform characteristics. For example, a ULVZ model with contrasts of -5% VP , -15% VS and +5% density explain the measured PcP amplitudes. Moreover, below SW Finland and NNW of the Caspian Sea a CMB topography can be assumed. The amplitude measurements indicate a wavelength of 200 km and a height of 1 km topography, previously also shown in the study by Kampfmann and Müller (1989). Better constraints might be provided by a joined analysis of seismological data, mineralogical experiments and geodynamic modelling.