Site-response high-frequency frontiers and the added value of site-specific earthquake record-based measurements of velocity and attenuation

Abstract

Listen

Translate article

One-dimensional (1D) ground-response analysis based on 1D velocity profiles and generic relationships between damping and velocity is frequently employed under the assumption that it can provide a reliable assessment of site effects. According to this hypothesis, epistemic uncertainty in site response primarily stems from inaccurately determined soil parameters. As shown by the recent development of ground-motion models in the Fourier domain, the underlying site-to-site variability of site response rises sharply at intermediate to high frequencies (>3 Hz), and this increase cannot be compensated for when relying on coarse velocity profiles and generic relationships between damping and velocity. In order to solve this high-frequency frontier in site response, we invert the spectral ratio of the horizontal-to-vertical components of strong motions based on the diffuse field concept for obtaining refined 1D velocity profiles and site-specific damping profiles. These updated equivalent structures of both velocity and damping significantly improve the performance of 1D ground-response analyses. This effect is particularly strong at intermediate frequencies (3–10 Hz), where both the deep sedimentary structure and the damping profile above the seismological bedrock have a strong influence on site response. While the use of refined 1D velocity and generic damping profiles for test sites of the Japanese KiK-net already provides a reduction of the intermediate-to-high-frequency residual and corrected site-to-site variability (and then an increase in the precision of predictions), the inclusion of site-specific damping allows this value almost to be halved. The results show that many sites are indeed too complex to be modeled by classical 1D ground-response analysis relying on generic damping relationships, while only site-specific record-based models allow a significant reduction in the bias.