The accuracy of earthquake source descriptions is a major limitation in high-frequency ($>1$ Hz) deterministic ground motion prediction, which is critical for performance-based design by building engineers. With the recent addition of realistic fault topography in 3D simulations of earthquake source models, ground motion can be deterministically calculated more realistically up to higher frequencies. We first introduce a technique to model frequency-dependent attenuation and compare its impact on strong ground motions recorded for the 2008 Chino Hills earthquake. Then, we model dynamic rupture propagation for both a generic strike-slip event and blind thrust scenario earthquakes matching the fault geometry of the 1994 Mw 6.7 Northridge earthquake along rough faults up to 8 Hz. We incorporate frequency-dependent attenuation via a power law above a reference frequency in the form $Q_0f^n$ ,with high accuracy down to Q values of 15, and include nonlinear effects via Drucker-Prager plasticity. We model the region surrounding the fault with and without small-scale medium complexity in both a 1D layered model characteristic of southern California rock and a 3D medium extracted from the SCEC CVMSi.426 including a near-surface geotechnical layer.

We find that the spectral acceleration from our models are within 1-2 interevent standard deviations from recent ground motion prediction equations (GMPEs) and compare well with that of recordings from strong ground motion stations at both short and long periods. At periods shorter than 1 second, Q(f) is needed to match the decay of spectral acceleration seen in the GMPEs as a function of distance from the fault. We find that the similarity between the intraevent variability of our simulations and observations increases when small-scale heterogeneity and plasticity are included, extremely important as uncertainty in ground motion estimates dominates the overall uncertainty in seismic risk. In addition to GMPEs, we compare with simple proxy metrics to evaluate the performance of our deterministic models and to determine the importance of different complexities within our model. We find that 3D heterogeneity, at both the long and short scale-lengths, is necessary to agree with data, and should be included in future simulations to best model the ground motion from earthquakes.

High-frequency (here defined for a maximum frequency (f_max) larger than about 1 Hz) ground motions are closely relevant to building response associated with small structures of engineering interests. Gaining a deeper understanding of the propagation of high-frequency seismic waves and characteristics of the resulting ground motions is therefore a principal goal for seismologists and earthquake engineers. Earthquake simulations, in particular those based on physics-based and deterministic models, have drawn significant attention from the seismic community in the most recent decades as a valuable supplement to recorded data. With the potential ability to accurately characterize broadband wavefields, numerical simulations have their own limitations, in particular difficulty in characterizing the underlying physical parameters to a sufficient resolution and computationally accommodating regional-scale models for seismic hazard analysis. The primary objective of this dissertations is to explore and rate the effects on high-frequency ground motions from various model features, which sheds light on their individual and mixed contributions and importance for future modelers. Chapter 1 is an introduction, providing background and motivation for each of the following chapters. Chapters 2 and 3 focus on model characteristics that govern the high-frequency ground motions. Chapter 2 proposes a calibration approach that enhances the near-surface velocity structure that is insufficiently resolved in community velocity models. In Chapter 3, we simulate a series of models to investigate the effects of topography, small-scale heterogeneities, frequency-dependent attenuation, and low near-surface velocities on the resulting wavefields and ground motions as the frequency extends up to 5 Hz. In Chapter 4, we propose a method to incorporate surface topography in constraining the 3D subsurface structure to predict site response. Chapter 5 studies nonlinear effects using dynamic simulations and proposes an equivalent kinematic source generator to emulate near-source plasticity in terms of the resulting peak ground velocities and spectral accelerations.

Ground motion prediction plays an important role in seismic hazard analysis. Deterministic wave propagation simulation has proven to be a powerful tool, widely used to predict ground motions in various applications. Recent advances in computing speed and sophistication of numerical methods have allowed ground motion prediction to achieve higher frequencies and broader applicability. However, realistic broadband simulations also require more accurate description of subsurface structure and more complicated physics. Thus, an important question is whether existing velocity models are sufficiently accurate for high frequency simulations. The main objectives of the research presented in this thesis are (1) to examine the efficacy of the current models for accurate ground motion prediction, and (2) if lacking sufficient accuracy, to establish model calibration procedures to deliver more accurate prediction of high-frequency ground motions. The introductory chapter first lays out the specific research questions addressed in this thesis. The following chapters demonstrate the procedure of using seismic data to constrain seismic frequency-dependent attenuation models and near-surface geotechnical layers. Finally, the calibrated models are used to demonstrate that deterministic ground motion simulation in both linear and nonlinear regimes indeed has the potential to be widely used in seismic hazard analysis.