90.00.00 GEOPHYSICS, ASTRONOMY, AND ASTROPHYSICS (for more detailed headings, see the Geophysics Appendix)
Refine
Has Fulltext
- yes (4)
Year of publication
- 2017 (4) (remove)
Document Type
- Doctoral Thesis (4)
Language
- English (4)
Is part of the Bibliography
- yes (4)
Keywords
Institute
Information on the contemporary in-situ stress state of the earth’s crust is essential for geotechnical applications and physics-based seismic hazard assessment. Yet, stress data records for a data point are incomplete and their availability is usually not dense enough to allow conclusive statements. This demands a thorough examination of the in-situ stress field which is achieved by 3D geomechanicalnumerical models. However, the models spatial resolution is limited and the resulting local stress state is subject to large uncertainties that confine the significance of the findings. In addition, temporal variations of the in-situ stress field are naturally or anthropogenically induced. In my thesis I address these challenges in three manuscripts that investigate (1) the current crustal stress field orientation, (2) the 3D geomechanical-numerical modelling of the in-situ stress state, and (3) the phenomenon of injection induced temporal stress tensor rotations. In the first manuscript I present the first comprehensive stress data compilation of Iceland with 495 data records. Therefore, I analysed image logs from 57 boreholes in Iceland for indicators of the orientation of the maximum horizontal stress component. The study is the first stress survey from different kinds of stress indicators in a geologically very young and tectonically active area of an onshore spreading ridge. It reveals a distinct stress field with a depth independent stress orientation even very close to the spreading centre. In the second manuscript I present a calibrated 3D geomechanical-numerical modelling approach of the in-situ stress state of the Bavarian Molasse Basin that investigates the regional (70x70x10km³) and local (10x10x10km³) stress state. To link these two models I develop a multi-stage modelling approach that provides a reliable and efficient method to derive from the larger scale model initial and boundary conditions for the smaller scale model. Furthermore, I quantify the uncertainties in the models results which are inherent to geomechanical-numerical modelling in general and the multi-stage approach in particular. I show that the significance of the models results is mainly reduced due to the uncertainties in the material properties and the low number of available stress magnitude data records for calibration. In the third manuscript I investigate the phenomenon of injection induced temporal stress tensor rotation and its controlling factors. I conduct a sensitivity study with a 3D generic thermo-hydro-mechanical model. I show that the key control factors for the stress tensor rotation are the permeability as the decisive factor, the injection rate, and the initial differential stress. In particular for enhanced geothermal systems with a low permeability large rotations of the stress tensor are indicated. According to these findings the estimation of the initial differential stress in a reservoir is possible provided the permeability is known and the angle of stress rotation is observed. I propose that the stress tensor rotations can be a key factor in terms of the potential for induced seismicity on pre-existing faults due to the reorientation of the stress field that changes the optimal orientation of faults.
Detection and Kirchhoff-type migration of seismic events by use of a new characteristic function
(2017)
The classical method of seismic event localization is based on the picking of body wave arrivals, ray tracing and inversion of travel time data. Travel time picks with small uncertainties are required to produce reliable and accurate results with this kind of source localization. Hence recordings, with a low Signal-to-Noise Ratio (SNR) cannot be used in a travel time based inversion. Low SNR can be related with weak signals from distant and/or low magnitude sources as well as with a high level of ambient noise. Diffraction stacking is considered as an alternative seismic event localization method that enables also the processing of low SNR recordings by mean of stacking the amplitudes of seismograms along a travel time function. The location of seismic event and its origin time are determined based on the highest stacked amplitudes (coherency) of the image function. The method promotes an automatic processing since it does not need travel time picks as input data.
However, applying diffraction stacking may require longer computation times if only limited computer resources are used. Furthermore, a simple diffraction stacking of recorded amplitudes could possibly fail to locate the seismic sources if the focal mechanism leads to complex radiation patterns which typically holds for both natural and induced seismicity.
In my PhD project, I have developed a new work flow for the localization of seismic events which is based on a diffraction stacking approach. A parallelized code was implemented for the calculation of travel time tables and for the determination of an image function to reduce computation time. In order to address the effects from complex source radiation patterns, I also suggest to compute diffraction stacking from a characteristic function (CF) instead of stacking the original wave form data. A new CF, which is called in the following mAIC (modified from Akaike Information Criterion) is proposed. I demonstrate that, the performance of the mAIC does not depend on the chosen length of the analyzed time window and that both P- and S-wave onsets can be detected accurately. To avoid cross-talk between P- and S-waves due to inaccurate velocity models, I separate the P- and S-waves from the mAIC function by making use of polarization attributes. Then, eventually the final image function is represented by the largest eigenvalue as a result of the covariance analysis between P- and S-image functions. Before applying diffraction stacking, I also apply seismogram denoising by using Otsu thresholding in the time-frequency domain.
Results from synthetic experiments show that the proposed diffraction stacking provides reliable results even from seismograms with low SNR=1. Tests with different presentations of the synthetic seismograms (displacement, velocity, and acceleration) shown that, acceleration seismograms deliver better results in case of high SNR, whereas displacement seismograms provide more accurate results in case of low SNR recordings. In another test, different measures (maximum amplitude, other statistical parameters) were used to determine the source location in the final image function. I found that the statistical approach is the preferred method particularly for low SNR.
The work flow of my diffraction stacking method was finally applied to local earthquake data from Sumatra, Indonesia. Recordings from a temporary network of 42 stations deployed for 9 months around the Tarutung pull-apart Basin were analyzed. The seismic event locations resulting from the diffraction stacking method align along a segment of the Sumatran Fault. A more complex distribution of seismicity is imaged within and around the Tarutung Basin. Two lineaments striking N-S were found in the middle of the Tarutung Basin which support independent results from structural geology. These features are interpreted as opening fractures due to local extension. A cluster of seismic events repeatedly occurred in short time which might be related to fluid drainage since two hot springs are observed at the surface near to this cluster.
The timing and location of the two largest earthquakes of the 21st century (Sumatra, 2004 and Tohoku 2011, events) greatly surprised the scientific community, indicating that the deformation processes that precede and follow great megathrust earthquakes remain enigmatic. During these phases before and after the earthquake a combination of multi-scale complex processes are acting simultaneously: Stresses built up by long-term tectonic motions are modified by sudden jerky deformations during earthquakes, before being restored by multiple ensuing relaxation processes.
This thesis details a cross-scale thermomechanical model developed with the aim of simulating the entire subduction process from earthquake (1 minute) to million years’ time scale, excluding only rupture propagation. The model employs elasticity, non-linear transient viscous rheology, and rate-and-state friction. It generates spontaneous earthquake sequences, and, by using an adaptive time-step algorithm, recreates the deformation process as observed naturally over single and multiple seismic cycles. The model is thoroughly tested by comparing results to those from known high- resolution solutions of generic modeling setups widely used in modeling of rupture propagation. It is demonstrated, that while not modeling rupture propagation explicitly, the modeling procedure correctly recognizes the appearance of instability (earthquake) and correctly simulates the cumulative slip at a fault during great earthquake by means of a quasi-dynamic approximation.
A set of 2D models is used to study the effects of non-linear transient rheology on the postseismic processes following great earthquakes. Our models predict that the viscosity in the mantle wedge drops by 3 to 4 orders of magnitude during a great earthquake with magnitude above 9. This drop in viscosity results in spatial scales and timings of the relaxation processes following the earthquakes that are significantly different to previous estimates. These models replicate centuries long seismic cycles exhibited by the greatest earthquakes (like the Great Chile 1960 Earthquake) and are consistent with the major features of postseismic surface displacements recorded after the Great Tohoku Earthquake.
The 2D models are also applied to study key factors controlling maximum magnitudes of earthquakes in subduction zones. Even though methods of instrumentally observing earthquakes at subduction zones have rapidly improved in recent decades, the characteristic recurrence interval of giant earthquakes (Mw>8.5) is much larger than the currently available observational record and therefore the necessary conditions for giant earthquakes are not clear. Statistical studies have recognized the importance of the slab shape and its surface roughness, state of the strain of the upper plate and thickness of sediments filling the trenches. In this thesis we attempt to explain these observations and to identify key controlling parameters. We test a set of 2D models representing great earthquake seismic cycles at known subduction zones with various known geometries, megathrust friction coefficients, and convergence rates implemented. We found that low-angle subduction (large effect) and thick sediments in the subduction channel (smaller effect) are the fundamental necessary conditions for generating giant earthquakes, while the change of subduction velocity from 10 to 3.5 cm/yr has a lower effect. Modeling results also suggest that having thick sediments in the subduction channel causes low static friction, resulting in neutral or slightly compressive deformation in the overriding plate for low-angle subduction zones. These modeling results agree well with observations for the largest earthquakes. The model predicts the largest possible earthquakes for subduction zones of given dipping angles. The predicted maximum magnitudes exactly threshold magnitudes of all known giant earthquakes of 20th and 21st centuries.
The clear limitation of most of the models developed in the thesis is their 2D nature. Development of 3D models with comparable resolution and complexity will require significant advances in numerical techniques. Nevertheless, we conducted a series of low-resolution 3D models to study the interaction between two large asperities at a subduction interface separated by an aseismic gap of varying width. The novelty of the model is that it considers behavior of the asperities during multiple seismic cycles. As expected, models show that an aseismic gap with a narrow width could not prevent rupture propagation from one asperity to another, and that rupture always crosses the entire model. When the gap becomes too wide, asperities do not interact anymore and rupture independently. However, an interesting mode of interaction was observed in the model with an intermediate width of the aseismic gap: In this model the asperities began to stably rupture in anti-phase following multiple seismic cycles. These 3D modeling results, while insightful, must be considered preliminary because of the limitations in resolution.
The technique developed in this thesis for cross-scale modeling of seismic cycles can be used to study the effects of multiple seismic cycles on the long-term deformation of the upper plate. The technique can be also extended to the case of continental transform faults and for the advanced 3D modeling of specific subduction zones. This will require further development of numerical techniques and adaptation of the existing advanced highly scalable parallel codes like LAMEM and ASPECT.
Observational and computational extragalactic astrophysics are two fields of research that study a similar subject from different perspectives. Observational extragalactic astrophysics aims, by recovering the spectral energy distribution of galaxies at different wavelengths, to reliably measure their properties at different cosmic times and in a large variety of environments. Analyzing the light collected by the instruments, observers try to disentangle the different processes occurring in galaxies at the scales of galactic physics, as well as the effect of larger scale processes such as mergers and accretion, in order to obtain a consistent picture of galaxy formation and evolution. On the other hand, hydrodynamical simulations of galaxy formation in cosmological context are able to follow the evolution of a galaxy along cosmic time, taking into account both external processes such as mergers, interactions and accretion, and internal mechanisms such as feedback from Supernovae and Active Galactic Nuclei. Due to the great advances in both fields of research, we have nowadays available spectral and photometric information for a large number of galaxies in the Universe at different cosmic times, which has in turn provided important knowledge about the evolution of the Universe; at the same time, we are able to realistically simulate galaxy formation and evolution in large volumes of the Universe, taking into account the most relevant physical processes occurring in galaxies.
As these two approaches are intrinsically different in their methodology and in the information they provide, the connection between simulations and observations is still not fully established, although simulations are often used in galaxies' studies to interpret observations and assess the effect of the different processes acting on galaxies on the observable properties, and simulators usually test the physical recipes implemented in their hydrodynamical codes through the comparison with observations. In this dissertation we aim to better connect the observational and computational approaches in the study of galaxy formation and evolution, using the methods and results of one field to test and validate the methods and results of the other.
In a first work we study the biases and systematics in the derivation of the galaxy properties in observations. We post-process hydrodynamical cosmological simulations of galaxy formation to calculate the galaxies' Spectral Energy Distributions (SEDs) using different approaches, including radiative transfer techniques. Comparing the direct results of the simulations with the quantities obtained applying observational techniques to these synthetic SEDs, we are able to make an analysis of the biases intrinsic in the observational algorithms, and quantify their accuracy in recovering the galaxies' properties, as well as estimating the uncertainties affecting a comparison between simulations and observations when different approaches to obtain the observables are followed. Our results show that for some quantities such as the stellar ages, metallicities and gas oxygen abundances large differences can appear, depending on the technique applied in the derivation.
In a second work we compare a set of fifteen galaxies similar in mass to the Milky Way and with a quiet merger history in the recent past (hence expected to have properties close to spiral galaxies), simulated in a cosmological context, with data from the Sloan Digital Sky Survey (SDSS). We use techniques to obtain the observables as similar as possible to the ones applied in SDSS, with the aim of making an unbiased comparison between our set of hydrodynamical simulations and SDSS observations. We quantify the differences in the physical properties when these are obtained directly from the simulations without post-processing, or mimicking the SDSS observational techniques. We fit linear relations between the values derived directly from the simulations and following SDSS observational procedures, which in most of the cases have relatively high correlation, that can be easily used to more reliably compare simulations with SDSS data. When mimicking SDSS techniques, these simulated galaxies are photometrically similar to galaxies in the SDSS blue sequence/green valley, but have in general older ages, lower SFRs and metallicities compared to the majority of the spirals in the observational dataset.
In a third work, we post-process hydrodynamical simulations of galaxies with radiative transfer techniques, to generate synthetic data that mimic the properties of the CALIFA Integral Field Spectroscopy (IFS) survey. We reproduce the main characteristics of the CALIFA observations in terms of field of view and spaxel physical size, data format, point spread functions and detector noise. This 3-dimensional dataset is suited to be analyzed by the same algorithms applied to the CALIFA dataset, and can be used as a tool to test the ability of the observational algorithms in recovering the properties of the CALIFA galaxies. To this purpose, we also generate the resolved maps of the simulations' properties, calculated directly from the hydrodynamical snapshots, or from the simulated spectra prior to the addition of the noise.
Our work shows that a reliable connection between the models and the data is of crucial importance both to judge the output of galaxy formation codes and to accurately test the observational algorithms used in the analysis of galaxy surveys' data. A correct interpretation of observations will be particularly important in the future, in light of the several ongoing and planned large galaxy surveys that will provide the community with large datasets of properties of galaxies (often spatially-resolved) at different cosmic times, allowing to study galaxy formation physics at a higher level of detail than ever before. We have shown that neglecting the observational biases in the comparison between simulations and an observational dataset may move the simulations to different regions in the planes of the observables, strongly affecting the assessment of the correctness of the sub-resolution physical models implemented in galaxy formation codes, as well as the interpretation of given observational results using simulations.