This dissertation presents a detailed analysis of recorded seismic waves in terms of their source and their propagation through the Earth in multiple scenarios. First, I investigate the source mechanisms of some highly unusual seismic events associated with the formation of a large sinkhole at Napoleonville salt dome, Assumption Parish, Louisiana in August 2012. I implemented a grid-search approach for automatic detection, location and moment tensor inversion of these events. First, the effectiveness of this technique is demonstrated using low frequency (0.1-0.2 Hz) displacement waveforms and two simple 1D velocity models for the salt dome and the surrounding sedimentary strata for computation of Green’s functions in the preliminary analysis. In the revised, and more detailed analysis, I use Green’s functions computed using a finite-difference wave propagation method and a 3D velocity model that incorporates the currently known approximate geometry of the salt dome and the overlying anhydrite-gypsum cap rock, and features a large velocity contrast between the high velocity salt dome and low velocity sediments overlying and surrounding it. I developed a method for source-type-specific inversion of moment tensors utilizing long-period complete waveforms and first-motion polarities, which is useful for assessing confidence and uncertainties in the source-type characterization of seismic events. I also established an empirical method to rigorously assess uncertainties in the centroid location, MW and the source type of the events at the Napoleonville salt dome through changing network geometry, using the results of synthetic tests with real seismic noise. During 24-31 July 2012, the events with the best waveform fits are primarily located at the western edge of the salt dome at most probable depths of ~0.3-0.85 km, close to the horizontal positions of the cavern and the future sinkhole. The data are fit nearly equally well by opening crack moment tensors in the high velocity salt medium or by isotropic volume-increase moment tensors in the low velocity sediment layers. The addition of more stations further constrains the events to slightly shallower depths and to the lower velocity media just outside the salt dome with preferred isotropic volume-increase moment tensor solutions. I find that Green’s functions computed with the 3D velocity model generally result in better fit to the data than Green’s functions computed with the 1D velocity models, especially for the smaller amplitude tangential and vertical components, and result in better resolution of event locations and event source type. The dominant seismicity during 24- 31 July 2012 is characterized by the steady occurrence of seismic events with similar locations and moment tensor solutions at a near-characteristic inter-event time. The steady activity is sometimes interrupted by tremor-like sequences of multiple events in rapid succession, followed by quiet periods of little of no seismic activity, in turn followed by the resumption of seismicity with a reduced seismic moment-release rate. The dominant volume- increase moment tensor solutions and the steady features of the seismicity indicate a crack- valve-type source mechanism possibly driven by pressurized natural gas.

Accurate and properly calibrated velocity models are essential for the recovery of correct seismic source mechanisms. I retrieved empirical Green’s functions in the frequency range ~ 0.2–0.9 Hz for interstation distances ranging from ~1 to ~30 km (~0.22 to ~6.5 times the wavelength) at The Geysers geothermal field, northern California, from cross-correlation of ambient seismic noise recorded by a wide variety of sensors. I directly compared noise- derived Green’s functions with normalized displacement waveforms of complete single-force synthetic Green’s functions computed with various 1D and 3D velocity models using the frequency-wavenumber integration method, and a 3D finite-difference wave propagation method, respectively. These comparisons provide an effective means of evaluating the suitability of different velocity models to different regions of The Geysers, and assessing the quality of the sensors and the noise cross-correlations. In the T-Tangential, R-Radial, Z- Vertical reference frame, the TT, RR, RZ, ZR and ZZ components (first component: force direction, second component: response direction) of noise-derived Green’s functions show clear surface-waves and even body-wave phases for many station pairs. They are also broadly consistent in phase and relative inter-component amplitudes with the synthetic Green’s functions for the known local seismic velocity structure that was derived primarily from body wave travel-time tomography, even at interstation distances less than one wavelength. I also found anomalous large amplitudes in TR, TZ, RT and ZT components of noise-derived Green’s functions at small interstation distances (≲4 km) that can be attributed to ~10°-30° sensor misalignments at many stations inferred from analysis of longer period teleseismic waveforms. After correcting for sensor misalignments, significant residual amplitudes in these components for some longer interstation distance (≳ 8 km) paths are better reproduced by the 3D velocity model than by the 1D models incorporating known values and fast axis directions of crack-induced shear-wave anisotropy in the geothermal field. I also analyzed the decay of Fourier spectral amplitudes of the TT component of the noise-derived Green’s functions at 0.72 Hz with distance in terms of geometrical spreading and attenuation. While there is considerable scatter in the amplitudes of noise-derived Green’s functions, the average decay is consistent with the decay expected from the amplitudes of synthetic Green’s functions and with the decay of tangential component local-earthquake ground-motion amplitudes with distance at the same frequency.

The detection and the analysis of unusual seismic events such as slow/low-stress-drop earthquakes in the Mendocino Triple Junction (MTJ), and nonvolcanic tremors and repeating earthquakes near Parkfied-Cholame can be used to provide a clearer picture and a better understanding of fault rheology, mechanics, and active tectonics in California. Modeling of cataloged slow earthquakes, which show large high-frequency/low-frequency magnitude differentials, indicates that the differences come from the source rather than from strongly attenuated propagation paths. Because the restricted number of these events limits their study and a precise comprehension of the complex processes occurring in the MTJ, a new procedure is proposed using continuous scanning of long-period seismic waveforms and moment tensor inversions to automatically detect and characterize regular and slow events, with magnitude larger than 3.5 occurring in the region. A similar approach using quasi-finite-source Green's functions in a point-source inversion method is proposed for the fast detection and characterization of large and potentially tsunamigenic earthquakes along the Cascadia Subduction Zone as well as in other subduction zone regions such as offshore Japan. Furthermore, this approach has the potential to be included in future earthquake and tsunami early warning systems.

Nonvolcanic tremors on the other hand, near Parkfield, inform on the variations in the state of stress in the deep fault zone and as a consequence in possible loading of the San Andreas Fault (SAF) at the proximity of the 1857 Fort Tejon earthquake rupture. Tremors are indeed strongly modulated by small stress changes (i.e. in the order of kPa) transmitted by local and regional earthquakes into their source region, below the seismogenic zone. The effects on tremors can last from a few seconds to several years after local, regional, and/or teleseismic earthquakes. Understanding and quantifying the tremor activity is essential for understanding the physics underlying the earthquake and fault zone processes such as earthquake nucleation, stress dependences, and state of fault stress. Episodes of tremors are found for example to exhibit deep M~5 slow slip events along the SAF, where no other datasets have yet revealed them. Hence, the occurrence of large events might be better understood through modeling of atypical, and numerous micro-earthquakes, whose recurrence times are shorter.

Understanding the characteristics of the unusual seismicity in northern California gives insights on the processes responsible for the occurrence of regular small and large earthquakes in the region that might help the seismological community to develop better earthquake forecasting models in the future.

This dissertation demonstrates the utility of the complete waveform regional moment tensor inversion (Dreger and Woods, 2002; Dreger, 2003, Minson and Dreger, 2008) for nuclear event discrimination. I explore the source processes and associated uncertainties for explosions and earthquakes under the effects of limited station coverage, compound seismic sources, assumptions in velocity models and the corresponding Green’s functions, and the effects of shallow source depth and free-surface conditions. The motivation to develop better techniques to obtain reliable source mechanism and assess uncertainties is not limited to nuclear monitoring, but they also provide quantitative information about the characteristics of seismic hazards (e.g. Petersen et al., 2014), local and regional tectonics and in-situ stress fields of the region (Hardebeck and Hauksson, 2001; Hardebeck and Michael, 2006).

This dissertation begins with the analysis of three sparsely recorded events: the 14 September 1988 US-Soviet Joint Verification Experiment (JVE) nuclear test at the Semipalatinsk test site in Eastern Kazakhstan, and two nuclear explosions at the Chinese Lop Nor test site. We utilize a regional distance seismic waveform method fitting long-period, complete, three-component waveforms jointly with first-motion observations from regional stations and teleseismic arrays. The combination of long period waveforms and first motion observations provides unique discrimination of these sparsely recorded events in the context of the Hudson et al. (1989) source-type diagram. We demonstrate through a series of Jackknife tests and sensitivity analyses that the source-type of the explosions is well constrained. One event, a 1996 Lop Nor shaft explosion, displays large Love waves and possibly reversed Rayleigh waves at one station, indicative of a large tectonic release. We demonstrate the behavior of Network Sensitivity Solutions [NSS] (Ford et al., 2010) for models of tectonic release (Toksöz et al., 1965) and spall-based tensile damage (Patton and Taylor, 2008) over a range of F-factors and K-factors.

A potential issue for moment tensor inversion of explosions is that Green’s functions have vanishing amplitudes at the free surface. Because explosions are detonated at very shallow depths, this can result in bias in the moment tensor solution (Stevens and Murphy, 2001). It is important to understand these free surface effects on discriminating shallow explosive sources for nuclear monitoring purposes. It may also be important in natural systems that have shallow seismicity such as volcanoes and geothermal systems. To tackle this problem, we examine the effects of the free surface on the moment tensor via synthetic testing, and apply the moment tensor based discrimination method to well-recorded chemical explosions. These shallow chemical explosions represent rather severe source-station geometry in terms of the vanishing traction issues. We show that the combined waveform and first motion method enables the unique discrimination of these events, even though the data include unmodeled single force components resulting from the collapse and blowout of the quarry face immediately following the initial explosion. In contrast, recovering the announced explosive yield using seismic moment estimates from moment tensor inversion remains challenging but we can begin to put error bounds on our moment estimates using the NSS technique.

The estimation of seismic source parameters is dependent upon having a well-calibrated velocity model to compute the Green’s functions for the inverse problem. Ideally, seismic velocity models are calibrated through broadband waveform modeling (e.g. Dreger and Helmberger, 1990; Bhattacharyya et al., 1999), however in regions of low seismicity velocity models derived from body or surface wave tomography may be employed (e.g. Tape et al., 2010; Shen et al., 2013; Porritt et al. 2014). Whether a velocity model is 1D or 3D, or based on broadband seismic waveform modeling or the various tomographic techniques, the uncertainty in the velocity model can be the greatest source of error in moment tensor inversion. These errors have not been fully investigated for the nuclear discrimination problem. To study the effects of unmodeled structures on the moment tensor inversion, we set up a synthetic experiment where we produce synthetic seismograms for a 3D model (Moschetti et al., 2010) and invert these data using Green's functions computed with a 1D velocity mode (Song et al., 1996) to evaluate the recoverability of input solutions, paying particular attention to biases in the isotropic component. We then evaluate source inversions for real data using Green's functions for 1D and 3D velocity models in which the Green's functions were computed by utilizing the principle of source-receiver reciprocity (Aki and Richards, 2002; Dahlen and Tromp, 1998), and the finite-difference method (Appelo and Petersson, 2008; Eisner and Clayton, 2001; Graves and Wald, 2001). Using the full waveform moment tensor inversion method we analyze earthquakes and explosions at NTS using 1D and 3D Earth models and compare the solutions and associated uncertainties at different frequency bands.

The synthetic experiment results indicate that the 1D model assumption is valid for moment tensor inversions at periods as short as 10 seconds for the 1D western U.S. model (Song et al., 1996). The correct earthquake mechanisms and source depth are recovered with statistically insignificant isotropic components as determined by the F-test. Shallow explosions are biased by the theoretical ISO-CLVD tradeoff but the tectonic release component remains low, and the tradeoff can be eliminated with constraints from P wave first motion. Path-calibration to the 1D model can reduce non-double-couple components in earthquakes, non-isotropic components in explosions and composite sources and improve the fit to the data. When we apply the 3D model to real data, at long periods (20-50 seconds), we see good agreement in the solutions between the 1D and 3D models and slight improvement in waveform fits when using the 3D velocity model Green’s functions. At high frequencies the advantage of the 3D model is limited except for paths from NTS to the San Francisco Bay, where we see a marked improvement in waveform fit. However, we do not see a clear reduction in source uncertainties when using a 3D model. A larger sample size is required to make useful interpretations about the use of 3D models in estimating source uncertainties. Our results indicate that the 3D model for the western U.S. (Moschetti et al., 2010) still needs further refinement to adequately model wave propagation at high frequencies and that path-averaged 1D models derived from the 3D model may be a more attractive approach than the more costly 3D simulation for short period inversions.

The objective of this research was to develop methodologies of interest to the geothermal industry to estimate critical parameters needed to characterize the fracture network generated during Enhanced Geothermal System (EGS) development. This work was supported by the U.S. Department of Energy Geothermal Technologies Office (DOE GTO). The research focused on seismicity at The Geysers geothermal field located 120 km north of San Francisco, California. The Geysers geothermal field is the world’s largest geothermal reservoir with approximately 1.6 GW installed electric capacity, 22 geothermal power plants with current average production of over 900 MW. Geothermal energy has been produced at The Geysers geothermal field since the early 1960s. Most of the earthquakes occur at shallow depths and are related to stress and hydrological perturbations due to geothermal energy operations including, on average, more than 57 million liters of treated wastewater injected daily. While the locations of earthquakes, as well as the timing and rates of their occurrence correlate with geothermal energy activities, greater understanding about the physical mechanisms is needed.

We developed and implemented a suite of methodologies to support the geothermal industry during EGS development. Our research focused on 1.0 < MW < 5.0 seismicity at The Geysers geothermal field, California to remotely estimate critical source parameters needed to characterize the fracture network generated during the EGS development. Initially 53 larger magnitude M > 3 earthquakes occurring throughout The Geysers geothermal field from 1992 to 2014 were investigated using an approach designed to assess resolution and uncertainty of seismic moment tensors. Deviatoric and full moment tensor solutions were computed, and statistical tests were implemented to assess solution stability, resolution and significance, particularly with respect to possible volumetric or tensile components. Several source models were examined including double-couple (DC), pure isotropic (ISO; volumetric change), and volume-compensated linear vector dipole (CLVD) sources, as well as mixtures of these such as DC + CLVD, DC + ISO and shear-tensile sources. In general, it was found that The Geysers earthquakes as a population deviate significantly from northern California seismicity in terms of apparent volumetric source terms and complexity. The mean volumetric component of earthquake mechanisms at The Geysers is approximately 30% compared to essentially zero for earthquake mechanisms outside The Geysers geothermal field. The volumetric moment tensor components of The Geysers earthquakes could arise from the flashing of injected water to steam, or more likely from the tensile stress induced from rapid cooling of the hot rock through the introduction of water.

In another study, seismicity in the vicinity of an EGS demonstration project in the Northwest Geysers at the Prati 32 (P-32) injection well and the Prati State 31 (PS-31) production well is investigated to determine earthquake focal mechanisms, moment magnitudes, and in-situ stress of events that occur before and during reservoir stimulation starting October 6, 2011. We compiled a catalog of 167 waveform-based seismic moment tensors ranging in moment magnitude from 0.6 to 3.9. Owing to the large number of events a semi-automated approach was developed to align the data with the Green’s functions used in the inversion. The automatic results were then reviewed to find the optimal shift for the inversion. The moment tensor catalog is subsequently used to invert for the stress tensor and to investigate possible temporal stress changes resulting from fluid injection. The relatively small uncertainties of the recovered stress tensors demonstrate the quality of the input mechanisms from the seismic moment tensor catalog. An approximate 15-degree counter-clockwise rotation of the least compressive stress σ3 was found to occur during injection. More remarkable was a change in the orientation of the maximum compressive stress σ1 from subhorizontal to vertical when injection operations temporarily halted. The orientation of σ1 returned to subhorizontal when injection operations resumed. It is found that there was a systematic reduction in the stress shape factor, R, as the injected water volume increases, indicating an evolution toward a more transtensional stress state. This work shows that effective tracking of the evolution of the state of stress in a system due to the introduction of water may be accomplished utilizing a semi-automatic full waveform moment tensor method. In addition, the full waveform approach was found to provide more robust estimates of the magnitudes of the micro-earthquakes at The Geysers.

Finally, a finite-source inverse approach was used to estimate the rupture area for ten earthquakes ranging in magnitude from MW 1.0 to 5.0. The slip models for these small earthquakes are complex with non-uniform distribution and in some cases with multiple asperities of slip release. They have complexity comparable to what is commonly found for larger earthquakes. We found the rupture area scaling with moment magnitude to be consistent with published relationships developed for globally distributed earthquakes with larger magnitudes (MW ≥ 5.5). This scaling relationship in conjunction with results from the stress inversion was subsequently applied to a set of earthquakes with high location accuracy to generate a statistical representation of the 3-D fracture network that was activated during the EGS demonstration project at Prati 32 injection well. We found the highest concentration of fractures located approximately 300 m to the north and 400 m below the bottom of the injection and production wells. Although this approach is approximate with respect to the actual orientation and dimension of the micro-earthquakes, it is a means to estimate a representation of the stimulated fracture network and to evaluate fracture density which could be used to help industry target where to place steam recovery wells. Additionally, the detailed source analysis and scaling information together with characterizations of rates of earthquakes can help to inform on possible seismic hazard related to such operations. Finally, an objective of the research was the development of portable tools that may be applied to study seismicity in other geothermal systems which could help assess their energy and economic potential.