Strong motion data from sites having both an instrumented structure and free-field accelerograph are compiled to evaluate the conditions for which foundation recordings provide a reasonably unbiased estimate of free-field motion with minimal uncertainty. Variations between foundation and free-field spectral acceleration are found to correlate well with dimensionless parameters that strongly influence kinematic and inertial soil-structure interaction phenomena such as embedement ratio, dimensionless frequency (i.e., product of radial frequency and foundation radius normalized by soil shear wave velocity), and ratio of structure-to-soil stiffness. Low frequency components of spectral acceleration recorded on shallowly embedded foundations are found to provide good estimates of free-field motion. In contrast, foundation-level peak ground acceleration (both horizontal and vertical) and maximum horizontal velocity, are found to be de-amplified. Implications for ground motion selection procedures employed in attenuation relations are discussed, and specific recommendations are made as to how these procedures could be improved.

Ground motion models (GMMs) are used to predict ground motion intensity measures given parameters descriptive of source, path, and site conditions. These GMMs incorporate source, path, and site response models that represent approximately the average conditions in the database from which the GMMs were derived. In the case of NGA-type models, global data is used, potentially with path and site adjustments for large regions with ample data (e.g., California), so the predictions represent either global or regional averages. In contrast, when GMMs are applied for a specific engineering project, the source, path, and site response attributes of interest are those local to the site, which may depart from the global or regional averages represented by the GMM. In this context, I refer to the source, path, and site models in the GMM as ergodic. Alternative models that consider local, or site-specific features, are considered non-ergodic, and have the potential to significantly reduce the ground motion variability that is considered in probabilistic seismic hazard analysis.

My thesis work is concerned principally with the site response component of GMMs, and in particular, with evaluating the effectiveness of predictive models available for non-ergodic site response analysis. The ergodic site amplification within a GMM represents the global or regional average for the site’s value of time-averaged upper 30 meters shear wave velocity and basin depth. Many local effects may introduce departures in site response from the ergodic model, including strong impedance contrasts within the shear wave velocity profile, an unfavorable location relative to a basin edge, complexity of local terrain, and perhaps other factors. Therefore, the ergodic site response model has two drawbacks: (1) potential for a biased estimate of mean site response and (2) because the ergodic model averages over a diverse array of conditions having many different site responses, the model carries a relatively large standard deviation.

The alternative of non-ergodic site response takes into account the particular geologic conditions at a site that control site response. If applied properly, non-ergodic site response can produce unbiased estimates of site response and remove site-to-site variability from the total standard deviation, which is a significant contributor. One method of evaluating non- ergodic site response in practice is to utilize recordings at the site to evaluate misfits from a GMM, and then use this information to construct a median site response model. However, when on-site recordings are not available, site-specific analysis requires the application of various predictive models. The questions addressed in this research relate to the effectiveness of different predictive models for estimation of site response.

The general approach followed in this research was to develop a database of available recordings for sites in a study region, analyze the data to develop non-ergodic site responses, and then either 1) apply existing predictive models to the sites with “measured” (i.e., non- ergodic) site responses and then evaluate their effectiveness over the population of sites or; 2) develop a new predictive model where existing models cannot be reasonably applied. The first approach of evaluating existing tools is applied to a population of 159 sites in California. The second approach of developing a new model is applied to 7 sites in Obihiro (Japan), where soft soil conditions (VS30 = 102 to 211 m/s) require the development of a novel modeling framework.

For the California sites, the predictive models considered are ground response analysis (GRA; one-dimensional shear wave propagation through a soil column), square-root impedance method (SRI), and models conditioned on horizontal-to-vertical spectral ratio (HVSR) vs frequency plots. The GRA and SRI methods require a shear wave velocity (VS) profile for the site and models for material damping for each soil horizon in the profile. Among the 159 sites, the profile depth range is 30 to 255 m (profile period range is 0.06 to 1.02 sec). The HVSR model requires HVSR data, which can be derived from microtremors or earthquake recordings. A challenge that was encountered in the application of GRA and SRI methods was the lack of soil profiles to accompany VS profiles. I developed protocols for estimation of soil type parameters that allow geotechnical damping models to be applied. Additional damping models were also considered, including one that is informed by high-frequency spectral decay of site ground motions (κ0).

Despite the depth of the profiles considered in this work being relatively modest, ground response analyses (or square-root-impedance analyses) are able to improve site response predictions relative to ergodic models for approximately 36% of sites (for periods less than or equal to the site period). The inability of site-specific methods to improve prediction accuracy for the 64% sites could stem from three potential sources: (1) simulations of one-dimension wave propagation do not accurately characterize the physics of site response; (2) the measured VS profile from the site does not accurately represent site conditions, either because of strong site heterogeneity or inaccurate measurements; (3) portions of the site profile beneath the profile depth significantly impact the site response in the frequency range of the measured profile. These problems are common to some extent in virtually all site response simulations, so understanding their collective impact is of practical importance. The unknown influence of these factors introduces epistemic uncertainties, which we quantify. Lacking any knowledge of whether a given site is well represented with one-dimensional simulations, this epistemic uncertainty is only slightly reduced from that of the site-to-site variability in ergodic models within soil column period range. For the subset of sites where this modeling is effective, the epistemic uncertainty is more substantially reduced by amounts ranging from 0.05-0.10 in natural log units.

The HVSR model considered in this work (adapted from a model in literature) uses the frequency and amplitude of peaks in HVSR spectra. I identify three populations of sites based on microtremor data – those for which a clear HVSR peak is evident (40%), those for which no peak occurs (40%), and intermediate/ambiguous cases (20%). When the ergodic model is used, sites with a peak are observed to have higher bias and site-to-site variability than sites without peaks; as a result, commonly used models for site-to-site variability represent a blending of these condition because the occurrence of peaks is not accounted for. Use of the HVSR model for sites with peaks does not appreciably change the bias but reduces dispersion at long periods (> 1 sec) relative to what is obtained with an ergodic model. The lack of improvement at short period could be caused by false positives (peaks in HVSR that do not appear in site response) and not well-aligned peak positions between HVSR and site response, and may also be influenced by the model used in our analyses having been derived for conditions in Japan. I recommend a California-specific bias correction for sites without a peak.

For the Obihiro (Japan) sites, I developed a region-specific site amplification model applicable to the peaty organic soils in this region. The analysis of site response from regional data required removal of source-specific biases and careful consideration of source-to-site path effects. These considerations were essential to avoid mapping source- or path-related model misfits into estimates of site response. I considered two subduction ground motion models as reference models. By paying special attention to the conditions for which the path models are effective, and making adjustments for between-island path misfits (Hokkaido to Honshu and vice-versa), I found the proposed approach effectively identifies site effects, and that the results are insensitive to the selected ground motion model. Observed site responses are characterized by strong resonances at first-mode site frequencies as derived from HVSR measurements.

The Next-Generation Attenuation (NGA) projects provide uniformly-processed ground motion data from earthquakes recorded in different tectonic settings and regions around the world. Since 2008, three separate NGA projects have been developed: NGA-West and NGA-West2 for shallow crustal earthquakes in active tectonic regions like California; and NGA-East for stable continental regions like central and eastern North America. Currently, the NGA-Subduction project is under development, focusing on gathering data for regions affected by subduction-zone earthquakes like the Pacific Northwest (PNW) and Alaska regions of North America, Japan, Taiwan, Mexico, and South America, among other areas.

As part of the effort of assembling and characterizing the ground motion dataset for South America, this thesis describes the development of a Chilean ground motion database for the NGA-Subduction project. First, the tectonics and seismicity of South America are discussed to highlight the importance of these subduction-zone earthquakes, with a particular emphasis on Chilean events. Subsequently, an overview of the seismic data providers in Chile is presented, describing the evolution of the different institutions and seismic networks that have recorded ground motions in the country. In addition, prior works on ground motion datasets and ground motion modeling for South America and Chile are introduced and briefly described. The importance of regional considerations in the development of global models, particularly with respect to path terms, is emphasized by comparing attenuation features observed in past work on the 2010 M 8.8 Maule Chile and M 9.0 Tohoku Japan events.

A significant effort was made to collect and characterize required metadata to accompany the ground motion database, including source, path, and site information. This thesis is primarily focused on presenting the development of parameters describing seismic sources. To date, the NGA-Subduction event catalog for the South American region consists of 826 earthquakes that extend from 1985 to 2016, 689 of which have been recorded in Chilean territory. The moment magnitude ranges from M 2.5 to M 8.8. In terms of number of recordings, the data obtained in Chile (4,213 time histories) represents approximately 68% of the total dataset in South America. The event database is controlled by interface and instraslab earthquakes, having 404 and 223 events respectively, with considerably fewer shallow crustal and outer-rise events. When possible, seismic sources are defined using suitable finite fault models (FFM) from the literature. Processes are introduced here to interpret published FFMs in a way that the most salient portion of the fault plane is used for site-to-source distance calculations. These processes improve upon those that had been used previously and have been adopted elsewhere in the NGA-Subduction project (e.g., Japan). In the case of the events without available FFMs, source characterization is similar to that used previously in NGA-projects.

Path and site parameters are currently being characterized and preliminary results are briefly explained in the final chapter. The results presented in this thesis regarding source information, along with the results from path and site characterization, will be part of a data PEER report describing the Chilean dataset.

For engineering design and seismic risk assessments, earthquake ground motions are characterized using intensity measures such as peak acceleration or spectral accelerations. Ground motion models (GMMs) estimate statistical distribution of intensity measures given information about the source, source-site path (distance), and site condition. This research pertains to data resources used to develop GMMs for subduction zone regions and adapts a global GMMs to a target region. My initial research developed portions of the NGA-Subduction (NGA-Sub) project database. I worked with other experts to characterize seismic sources and source-to-site paths. The database contains data from 1,880 earthquakes in Alaska, Cascadia, Central America and Mexico, Japan, New Zealand, South America, and Taiwan. Source parameters are related to the earthquake focus and finite fault representations. An important issue was the assignment of event types (intraslab and interface). I developed event type classification procedures that consider hypocenter locations relative to the interface, focal mechanisms, and other factors. The second phase of research involved collaborating with researchers in Mexico to enhance data resources for that region (relative to NGA-Sub) and to use the information to develop a regionally customized GMM. I selected Mexico because it has high seismic hazard, its available GMMs have deficiencies, and it has substantial data from recent events that were not considered in NGA-Sub. The database for Mexico is extended by considering small magnitude (M < 6) events and three large events (M 7.2-8.3) in 2017 and 2018. The latter are particularly important, because they are well recorded over a broad distance range and apply for hazard-critical conditions. These changes increase the size of the database from 593 recordings from 66 events (NGA-Sub) to 1882 recordings from 121 events. These data are examined relative to a global NGA-Sub model, which reveals a number of interesting features including faster anelastic attenuation in backarc than forearc regions and generally similar levels of site response as in the global GMM. A novel model for site response in Mexico City is proposed that is properly centered with respect to GMMs and that accounts for the unique Lake Texcoco geology.

We develop empirical relationships to predict nonlinear (i.e., amplitude-dependent) amplification factors for 5% damped response spectral acceleration as a continuous function of average shear wave velocity in the upper 30 m, Vs-30. We evaluate amplification factors as residuals between spectral accelerations from recordings and modified rock attenuation relationships for active regions. Amplification at low- and mid-periods is shown to increase with decreasing Vs-30 and to exhibit nonlinearity that is dependent on Vs-30. The degree of nonlinearity is large for NEHRP Category E (Vs-30~300 m/s. The results can be used as Vs-30-based site factors with attenuation relationships. The results also provide an independent check of site factors published in the NEHRP Provisions, and apparent bias in some of the existing NEHRP factors is identified. Moreover, the results provide evidence that data dispersion is dependent on Vs-30.

In engineering practice, it is common for the ground motion intensity measures used in design to be estimated using a combination of probabilistic and deterministic procedures. Formal probabilistic seismic hazard analyses are performed to estimate intensity measures (IMs) for rock site conditions. This is followed by a deterministic modification of the rock IMs to account for site effects, which is typically done using prescribed site factors available in the literature or in seismic code provisions. In this article we investigate the extent to which ground motions estimated using this semi-probabilistic approach approximate ground motions evaluated in a fully probabilistic context in which the nonlinear site response is integrated into the hazard calculations. Using two existing California sites as examples, we demonstrate that the deterministic application of nonlinear site factors underestimates the ground motions evaluated using a formal probabilistic approach. This misfit arises from multiple sources including different standard deviation terms for rock and soil sites and different controlling earthquakes. In particular, sites having a significant nonlinear site response tend to attract larger contributions from distant earthquakes than do rock sites. Fortunately, a new set of ground motion prediction equations developed through the Next Generation Attenuation project directly incorporate nonlinear site response effects, and hence as those models are integrated into hazard codes, there will be no need to continue in practice the semi-probabilistic approach for ground motion estimation.

In this dissertation, we study the effects of site response on earthquake ground motions, the uncertainty in site response, and incorporating site response in probabilistic seismic hazard analysis. We introduced a guideline for evaluation of non-ergodic (site-specific) site response using (a) observations from available recorded data at the site, (b) simulations from one-dimensional ground response analysis, or (c) a combination of both. Using non-ergodic site response is expected to be an improvement in comparison to using an ergodic model which is based on the average of a global dataset conditional on site parameters used in ground motion models. The improvement in prediction when using non-ergodic analysis results in the removal of site-to-site variability which is a part of the uncertainty in ground motion prediction. The site-to-site variability is evaluated by partitioning the residuals to different sources of variability. We illustrate application of these procedures for evaluating non-ergodic site response, and use examples to show how the reduction in site response uncertainty results in less hazard for long return periods.

We utilize a dataset of recordings from vertical array sites in California in order to study the effectiveness of one-dimensional ground response analysis in predicting site response. We use the California dataset for comparing the performance of linear ground response analysis to similar studies on a dataset from vertical arrays in Japan. We use surface/downhole transfer functions and amplification of pseudo-spectral acceleration to study the site response in vertical arrays. For performing linear site response analysis for the sites, we use three alternatives for small-strain soil damping namely (a) empirical models for laboratory-based soil damping; (b) an empirical model based on shear wave velocity for estimating rock quality factor; and (c) estimating damping using the difference between the spectral decay () at the surface and downhole. The site response transfer functions show a better fit for California sites in comparison to the similar results on Japan. The better fit is due to different geological conditions at California and Japan vertical array sites, as well as the difference in the quality of data for the two regions. We use pseudo-spectral acceleration residuals to study the bias and dispersion of ground response analysis predictions. The results of our study shows geotechnical models for lab-based damping provide unbiased estimates of site response for most spectral periods. In addition, the between- and within-site variability of the residuals do not show a considerable regional between California and Japan vertical arrays.

In another part of this dissertation, we develop ground motion models for median and standard deviation of the significant duration of earthquake ground motions from shallow crustal earthquakes in active tectonic regions. The model predicts significant durations for 5-75%, 5-95%, and 20-80% of the normalized Arias intensity, and is developed using NGA-West2 database with M3.0-7.9 events. We select recordings based on the criteria used for developing ground motion models for amplitude parameters as well as a new methodology for excluding recordings affected by noise. The model includes an M-dependent source duration term that also depends on focal mechanism. At small M, the data suggest approximately M-independent source durations that are close to 1 sec. The increase of source durations with M is slower over the range M5 to 7.2-7.4 than for larger magnitudes. We adopt an additive path term with breaks in distance scaling at 10 and 50 km. We include site terms that increase duration for decreasing VS30 and increasing basin depth. Our aleatory variability model captures decreasing between- and within-event standard deviation terms with increasing M. We use the model for validating the duration of ground motion time series produced by simulation routines implemented on the SCEC Broadband Platform. This validation is based on comparisons of median and standard deviation of simulated durations for five California events, and their trends with magnitude and distance, with our model for duration. Some misfits are observed in the median and dispersion of durations from simulated motions and their trend with magnitude and distance. Understanding the source of these misfits can help guide future improvements in the simulation routines.