UCLA

Microtremor-based horizontal-to-vertical spectral ratios (mHVSR) are derived from Fourier Amplitude Spectra of ambient ground motions and can reveal site features such as resonances or lack thereof. Despite this promise, mHVSR curves have not previously found significant engineering application, likely because: (1) mHVSR data have not been available in sufficient quantities to support ground motion model development and (2) difficulties linking mHVSR parameters to site response. This research addresses both needs by enhancing mHVSR data availability and developing mHVSR-conditioned site response models for California sites. I extended and improved the available mHVSR data by temporarily deploying seismometers at sites of interest (often locations of permanent accelerometers) and downloading data from permanent broadband seismometers. At California vertical array sites, mHVSR curves were measured in expanding concentric circles and used to characterize spatial heterogeneity. In total, I added 1844 mHVSR curves from 1096 sites to a public database. Low-frequency microtremor signals are often influenced by oceanic and atmospheric effects unrelated to site stratigraphy. Such ordinates can be large and sensitive to processing details, while also being important for sites in deep basins. I documented issues affecting low-frequency ordinates for certain sensor types, investigated their causes, and provided recommendations to improve data quality. I propose a framework for developing parametric site response models conditioned on V_S30 and mHVSR-derived parameters. This framework predicts common site response features, including resonant peaks, ascending or descending ramps (reflecting energetic response at the upper plateau), or flat responses relative to ergodic models. mHVSR parameters help predict which features are most likely and the attributes of those features. Machine learning models are also provided conditional on mHVSR attributes. Both parametric and machine learning models reduce site-to-site standard deviations by 0.1-0.2 natural log units at long periods, which is a substantial benefit. Overall, results from this work highlight the value of mHVSR as a practical and informative site characterization tool. It also provides models that can be applied to improve the fidelity of site response modeling and reduce its associated uncertainties.