The objective of this thesis is to improve the physical understanding of earthquake ground motion characteristics related to source, path and nonlinear site effects and our ability to model those effects with engineering models. This was achieved through four research studies consisting of: (1) calibrating broadband simulation procedures to remove previously recognized sources of bias in distance attenuation and standard deviation; (2) enhancing a site database used for assigning site parameters to ground motion recordings, particularly with regard to the level of rigor and transparency with which the database is populated; (3) leveraging a state–of–the–art ground motion database and recent simulation–based studies to develop a nonlinear site amplification model suitable for use in ground motion predictions equations (GMPEs) and relatively simplified building code applications; and (4) developing GMPEs that provides mean and standard deviation of ground motion intensity measures in active crustal regions.

The high–frequency component of the simulation procedure considered in this study combines deterministic Fourier amplitude spectra (dependent on source, path, and site models) with random phase. Significantly too–fast distance attenuation bias identified in prior work has been removed by increasing the quality factor (Q). We introduced random site–to–site variations to Fourier amplitudes using a log–normal standard deviation ranging from 0.45 for M < 7 to zero for (M8 to achieve dispersion terms that are more compatible with those from empirical models but remain lower at large distances (e.g., > 100 km).

Site database work was performed within the context of the NGA–West 2 project. Starting with the site database from original (2008) NGA project (last edited in 2006), we provided site classifications for 2538 new sites and re–classifications of previous sites. The principal site parameter is the time–averaged shear wave velocity in the upper 30 m (V_s30), which is characterized using measurements where available, and proxy–based relationships otherwise. We improved the documentation and consistency of site descriptors used as proxies for the estimation of V_s30, developed evidence–based protocols for Vs30 estimation from available proxies, and augmented estimates of various basin depth parameters.

Site factors typically have a small–strain site amplification that captures impedance and resonance effects coupled with nonlinear components. Site factors in current NEHRP Provisions are empirically–derived at relatively small ground motion levels and feature simulation–based nonlinearity. We show that current NEHRP site factors have discrepancies with respect to the site terms in the original NGA GMPEs both in the linear site amplification (especially for Classes B, C, D, and E) and the degree of nonlinearity (Classes C and D). We analyzed the NGA–West 2 dataset and simulation–based models for site amplification to develop a new model. The model has linear and nonlinear additive components. The linear component is fully empirical, being derived from worldwide ground motion data (regional effects were examined but found to not be sufficiently important to be included in the model). The model features linear V_s30–scaling in a log–log sense below a corner velocity (V_c), and no V_s30–scaling for velocities faster than V_c. The nonlinear component is developed from consideration of empirical data analysis and simulation results within a consistent context. The resulting nonlinearity operates principally at short periods and soft soils. This model is suitable for use as a site term in GMPEs and was applied to develop a proposal for updating the NEHRP site factors. The recommended factors remove a discrepancy between the reference condition used in the site factors and the national seismic hazard maps published by USGS.

We have developed empirical equations for predicting the average horizontal component of earthquake ground motions from active crustal region earthquakes worldwide. The equations build upon a previous ground-motion model by Boore and Atkinson in 2008. Significant new features of the proposed GMPEs include: modified site terms; a modified magnitude scaling function that produces a higher degree of saturation at large magnitude for high–frequency ground motions; region–specific apparent anelastic attenuation term; basin depth correction factors that are centered on the average level of basin amplification conditional on V_s30; standard deviation terms that depend on M for between–event standard deviations and M–, R_jb– and V_s30–dependent within–event standard deviations. The resulting equations are applicable for events over a magnitude range of 3 to 8.5 for strike–slip or reverse–slip events (M3 to 8 for normal–slip events), distance range up to 400 km, and site conditions ranging from V_s30 = 150 to 1500 m/s. The equations are useful for prediction of the ground motion intensity measures (IMs) PGA, PGV, and PSA at periods T = 0 to 10 sec.

We review site parameters used in ground motion prediction equations (GMPEs) for various tectonic regimes and describe procedures for estimation of site parameters in the absence of site-specific data. Most modern GMPEs take as the principal site parameter the average shear wave velocity in the upper 30 m of the site (V_s30) either directly or as the basis for site classification into categories. Three GMPEs developed for active regions also use basin depth parameters. We review estimation procedures for V_s30 that utilize surface geology, terrain-based site categories, ground slope, or combinations of these. We analyze the relative efficacy of those procedures using a profile data set from California assembled in a recent NGA project. The results indicate that no single procedure is most effective and that prediction dispersion is lower for young sediments than for stiff soils or rock.