Currently, the source mechanisms of global large and moderate earthquakes are routinely constrained and reported by multiple agencies using long-period seismic signals. These conventional analyses simplify the rupture process of an earthquake as single point source. On one hand, it becomes not appropriate if this earthquake has a fault dimension comparably larger than the signal's dominant wavelength or includes multiple subevents with different focal mechanisms. On the other hand, these source complexities can yield distinct seismic waves, and therefore might be deciphered through more careful waveform modeling. Such additional information is crucial for the study of earthquake physics as well as post-earthquake hazard evaluation and emergency assistance. The first part of this dissertation introduces a nonlinear inverse algorithm, which approximate an earthquake as multiple double-couple sources (MDC) and constrain them simultaneously using seismic data. The number of double-couples is determined by statistic F-test. This method has been applied to local and global large earthquakes using long-period signals, teleseismic body waves and local broadband strong-motion records. The results demonstrated that more source information could be extracted using this MDC algorithm than the conventional moment tensor inversion algorithms, though the spatiotemporal resolution relies on the station coverage and frequency contents of seismic signals. The implements of high-quality broadband seismograph networks globally and advances in parallel computational capability enable us to conduct the MDC analysis automatically for global large earthquakes and provide more precise priori information for the subsequent finite fault inversions. It then shall be an important contribution to the routine real-time earthquake hazard analysis.

The second part of this thesis focuses on detecting the aftershocks occurring temporally close to the corresponding mainshocks with the MDC analysis and subsequently analyzing their dynamic interactions with the mainshocks using complete Coulomb stress failure criteria. The two examples presented in Chapter 3 and 4 are the 2012 Mw 7.3 Honshu earthquake and 2000 Mw 8.1 New Ireland earthquake. Our results indicated that both events are doublets. The first subevent of the 2012 Honshu earthquake is a Mw 7.3 oblique thrust earthquake beneath the seaside of the Japan trench axis at a depth of ~50 km, and followed ~13 s later by a Mw 7.3 pure normal fault rupture ~50 km to the N260oE, at a depth of 25-30 km beneath the island side of the trench axis, a classic example of the plate bending. The first subevent of the 2000 Mw 8.0 New Ireland earthquake is Mw 8.0 right lateral strike-slip earthquake, which ruptured a fault plane of 140 km and following 90 s later by a Mw 7.4 normal fault earthquake beneath the outer rise region of subducted Solomon sea plate, 263 km to the south of the first subevent. Our subsequent stress calculations indicated that the rupture of the 2011 Mw 9.1 Tohoku earthquake increases the stress levels at the hypocenters of two 2012 Honshu subevents. While the Coulomb stress increment is significantly smaller than the co-seismic stress drop of the first subevent, which at the second subevent is compatible to co-seismic stress drop. At the hypocenter of the second subevent, the rupture of the first subevent produced negative static Coulomb stress. The dynamic Coulomb stress carried by the direct P wave is negative as well, but that associating with the direct S is positive with a peak amplitude of 3.8 MPa.. Although the inverted rupture initiation time of the second subevent is 1.7 s earlier than when the positive stress reaches the maximum, this is within the uncertainty of our kinematic modeling. For the 2000 New Ireland doublet, our calculations indicate that the rupture of first subevent produces either a negative or positive but with negligible amplitude (6x10^-4MPa) static Coulomb failure stress at the hypocenter of second subevent. In contrast, a 0.91 MPa positive dynamic Coulomb stress pulse excited by the stopping phase associating with the rupture of the dominant asperity of the first subevent reaches the hypocenter of the second subevent at 86 s and keeps positive until 90 s. November 16 2000 Mw 8.0 New Ireland earthquake is then a unique example that passing of seismic waves caused a nearly instantaneous triggering of an Mw >7 normal fault earthquake 150 km away. In Chapter 5, we have performed MDC analysis and finite fault joint inversion for 2013 Mw 7.7 Balochistan earthquake using teleseismic body waves and surface waves. Volumetric and Coulomb stress perturbation are then carefully investigated to shed light on revealing the possible dynamic interaction between mud volcano eruption and seismic process of the 2013 Pakistan earthquake. In short, for the all three cases, we find that these subsequent events are correlated with the dynamic stress perturbation, rather than static stress perturbation, excited by the earlier events.

The dynamic motion of an earthquake begins abruptly, but is often initiated by a short interval of weak motion called the seismic nucleation phase (SNP). Ellsworth and Beroza [1995, 1996] concluded that the SNP was detectable in near-source records of 48 earthquakes with moment magnitude (Mw), ranging from 1.1 to 8.1. They found that the SNP accounted for approximately 0.5% of the total moment and 1/6 of the duration of the earthquake. Ji et al [2010] investigated the SNP of 19 earthquakes with Mw greater than 8.0 using teleseismic broadband data. This study concluded that roughly half of the earthquakes had detectable SNPs, inconsistent with the findings of Ellsworth and Beroza [1995]. Here 69 earthquakes of Mw 7.5-8.0 from 1994 to 2011 are further examined. The SNP is clearly detectable using teleseismic data in 32 events, with 35 events showing no nucleation phase, and 2 events had insufficient data to perform stacking, consistent with the previous analysis. Our study also reveals that the percentage of the SNP events is correlated with the focal mechanism and hypocenter depths. Strike-slip earthquakes are more likely to exhibit a clear SNP than normal or thrust earthquakes. Eleven of 14 strike-slip earthquakes (78.6%) have detectable NSPs. In contrast, only 16 of 40 (40%) thrust earthquakes have detectable SNPs. This percentage also became smaller for deep events (33% for events with hypocenter depth>250 km). To understand why certain thrust earthquakes have a visible SNP, we examined the sediment thickness, age, and angle of the subducting plate of all thrust earthquakes in the study. We found that thrust events with shallow (<50 km) hypocenters with thick seafloor sediments (>600 m) on the subducting plate tend to have clear SNPs. If the SNP can be better understood in the future, it may help seismologists better understand the rupture dynamics of large earthquakes. Potential applications of this work could attempt to predict the magnitude of an earthquake seconds before it begins by measuring the SNP, vastly improving earthquake Early Warning Systems for populated areas.

Our understanding of earthquakes that occurred prior to the establishment of the World-Wide Standardized Seismographic Network (WWSSN) is generally limited by the availability of high-quality geophysical observations. As a result, significant variability exists among source studies for important seismic events such as the historic 1952 Kern County, California earthquake. Here, combining reported geodetic observations with a collection of previously unused, local seismic recordings, we conduct a series of inversions to constrain a slip model for the main rupture. Our results suggest that it initiates on a low-angle fault plane with dominant strike-slip motion (strike=49±3°; dip=35±1°; and rake=11±5°) then triggers an abnormally energetic rupture on a high-angle fault plane (strike = 51°, dip = 75°), 2 s later. This energetic rupture, contained within a 9×6 km patch near the hypocenter, accumulates 6–7 m of slip and has a high average static stress-drop (larger than 50 MPa). P-waves excited by this powerful sub-event saturates seismic recordings as far as Berkeley (approximately 430 km away). The total rupture has a duration 23–26 s and a cumulative moment of 7.61×10^{19} Nm, or Mw 7.18. The majority of moment release occurs within a 36 km section in the southwest portion of White Wolf fault (assumed to be 60 km long). The weighted, average rake-angle over the southwest segment of the fault is 47–57°, falling between previous results based on individual seismic or geodetic data sets. Investigation of the regional velocity structure reveals high P-wave velocities near the southwest portion of the White Wolf fault, adjacent to the presumed hypocenter.

Accurate images of the growth of slip and faulting are critical in understanding the physics of earthquakes, and subsequently aid in the prediction of the ground motion that could be expected from future rupture events. Many methods, namely finite fault inversions, have been developed that use observed seismic waveforms and static deformations to constrain the spatiotemporal rupture evolutions of great earthquakes. This work explores the development of novel inversion methods to further the investigation into properties of earthquake physics.

Earthquakes lead to the overall reduction of stress across the ruptured fault plane, and stress drop is a key parameter in accurately estimating the strong ground motion. Thus, we have developed a new procedure to determine if it is possible to robustly constrain an uncertainty range for the co-seismic stress drop of large earthquakes. For a given earthquake, we loop through a series of target stress drops, and for each case we conduct a modified finite fault inversion to find the solution that matches the seismic and/or geodetic data, and also matches a prescribed stress drop value. From this, we can determine a relationship between the resulting misfit between our synthetic model and the observations, and the pre-assigned target stress drop value.

This new inversion technique is applied to several rupture events with varying datasets in order to test its robustness. First, we examine the 2014 Mw 7.9 Rat Islands earthquake using far-field data and discovered that only the lower bound of the average stress drop could be well constrained. To investigate whether such a conclusion also holds for near field data, the slip distribution and the stress drop of the 2015 Mw 7.8 Gorkha, Nepal earthquake was studied using GPS and InSAR data. Our results revealed similar patterns: that even a comprehensive geodetic dataset could also only constrain the lower bound of stress drop. Furthermore, one of the important uses of stress drop for earthquake physics is in the study of energy partitioning. The average stress drop is equivalent to twice the ratio of apparent available energy to the total seismic potency, and so our result has a direct impact on the study of the earthquake energy budget. As stress drop is proportional to the available seismic energy, our results imply that only the lower bound of the available energy can be constrained.

The November 14th 2016 MW 7.8 Kaikoura, New Zealand earthquake occurred along the northern part of the South Island. Available information indicates that this earthquake involved multiple fault segments of the Marlborough fault system (MFS). Additionally, slip might also have occured on the subduction interface of the Pacific Plate under the Australian Plate, beneath the MFS. The exact number of involved fault segments as well as the temporal co-seismic rupture sequence has not been fully determined. We propose two inversion strategies to determine the fault geometry and spatiotemporal rupture history: first we use teleseismic and strong motion waveforms to determine point-source focal mechanisms for all of the faults that participated in the rupture; second, we use seismic and geodetic data to invert for the kinematic rupture parameters on a limited number of fault segments. The first approach allows us to determine a rupture timing sequence for different fault segments. Once we have the timing among fault segments, we invert the seismic and geodetic data for the spatial and temporal kinematic parameters. Both of these methods allow us to evaluate the potential of slip on the Hikurangi subduction interface.

Instrumented geotechnical field sites are designed to capture the infrequent but critically important in situ case histories of ground response, deformation, and liquefaction during significant earthquakes that generate high intensity ground shaking and large strains. The University of California at Santa Barbara has been monitoring densely instrumented geotechnical array field sites for almost three decades, with continuous recording now for more than a decade. When seismic waves travel into soil with sufficiently large ground motions, the soil behaves nonlinearly meaning the shear modulus of the material decreases from the linear value observed during weak ground motions. The degraded shear modulus can continue to affect a site for a period of time by changing the soil response during smaller ground motions after the large event. Decreased shear modulus is inferred when a decrease of shear wave velocity between two sensors in a vertical downhole array is observed. This velocity is calculated by measuring the difference in shear wave arrival times between the sensors using normalized cross correlation. The trend of decreasing shear wave velocity with increasing peak ground acceleration is observed at multiple geotechnical array field sites. The length of time the decreased velocity remains following stronger shaking is analyzed using more than 450 events over more than a decade at the Wildlife Liquefaction Array (WLA). Using both monthly and yearly velocity averages between sensors, there is evidence that suggests the shear wave velocity remains low over a period of months following larger significant shaking events at the site. In addition, at WLA there is evidence that the decrease in shear wave velocity can be detected at ground motion levels as low as 20 cm/s2.

Additionally at the Garner Valley Downhole Array, a permanent cross-hole experiment is used to measure velocity changes in the soil with changing water table height. An underground hammer source swings once a week and is recorded on two geophones at the same depth in a line adjacent to the source. Data collected from December 2010 to June 2012 and again from August 2015 to June 2017 is analyzed. That results shows a strong correlation between water table height and the shear wave velocity in the sediment, with changes of almost 5% over the course of seasonal water table variation.