Data-driven testing of site amplification models and alternative proxies for large-scale seismic hazard and risk assessment
Year: 2025
Type: dissertation
Abstract: The rapid increase in available seismic data provides new opportunities for rigorous evaluation of ground motion models (GMMs), which are critical components of probabilistic seismic hazard assessment (PSHA). Despite scientific advancements and increase in ground motion recordings, the variability associated to GMMs is still high, with site variability being the largest contributor to the uncertainty. In this thesis, I therefore test the performance of site amplification models using new empirical data, while focusing on two aspects: the modeling of nonlinear soil behavior occurring at soft soils sites experiencing strong ground motions, and the usability of large-scale site condition models as site proxies for site amplification prediction. In my first study, I develop a framework to systematically test commonly used nonlinear site amplification models, using records from the KiK-net network in Japan. The models are tested against ground-motion residuals corrected for event and linear site effects, and compared against a linear site amplification model. 20 soft-soil stations that have recorded at least four strong motion records, here defined as peak ground acceleration (PGA) more than 0.05 g, was selected for the test. Results show that for most of the selected stations, the linear site amplification model performed better than the nonlinear models. This result indicates that classical nonlinear site amplification models are not able accurately capture the nonlinear soil behavior in the strong-motion range considered (PGA up to 0.2 g). The poor performance of the nonlinear site amplification models highlight a need for refinement in nonlinear site amplification modeling. Because the tested models are based on the time-averaged shear-wave velocity of the top 30 meters of the soil column (V_S30 and either PGA or spectral acceleration (SA), for the next step I therefore aim to test alternative site proxies and intensity measures against the nonlinear soil behavior. To asses the site-specific nonlinear behavior at stronger shaking, in my second study I compile a new harmonized ground-motion dataset with recent recordings, and compute the site response transfer function of KiK-NET stations using the surface-to-borehole ratio (SBr). In my third study, I then measure the nonlinearity as the change in amplitude and shift in frequency of each record SBr compared to the mean linear SBr. The station-specific nonlinearity is then derived by fitting a hyperbolic tangent function to the nonlinearity measurements of each station. The result shows that the PGA threshold for onset of nonlinearity obtained from the fitted functions has a high site-to-site variability, ranging between 0.2-1.15m/s² Next, the site-specific nonlinearity parameters are correlated against V_S30, as well as geology, depth to bedrock, sediment thickness and soil content. For most site parameters the correlation with the nonlinearity is low, with the highest correlation obtained between site-specific nonlinearity and depth to bedrock and parameters describing the shallow soil conditions. For sites with similar site condition, the mean nonlinear behavior shows that at moderate shaking PGA=1-3 m/s² nonlinear behavior is most prominent for soft soils and thin to intermediate sediment thickness. The mean nonlinear behavior across all sites regardless of grouping, however, is significant (defined as 10 \% change in amplitude or shift in frequency) only at PGA > 3 m/s². These results show that at PGA < 3 m/s², nonlinear site amplification modeling should strongly depend on the site conditions. In the final part of this thesis, I evaluate the ability of large-scale site proxies to predict linear site amplification. As an alternative to the widely used global model of V_S30 inferred from slope, a global geomorphological model for sediment thickness was evaluated as a site proxy for predicting site amplification. To evaluate the site proxies, empirical site amplification factors derived for Europe as site-to-site residuals are used to derive proxy-based site amplification models. The predictive ability of new and common proxies is then evaluated as the reduction in model-corrected site-to-site variability. The results show that the geomorphological sediment thickness performs better than or equally well as the traditional V_S30 derived from slope, while the best performance were found for the site amplification model based on geology and slope. Although the various proxy-based site amplification maps are largely similar, there are also distinct differences. This highlights the epistemic uncertainty associated with site amplification models used in large-scale seismic hazard or risk studies. In conclusion, this thesis demonstrates that while existing nonlinear site amplification models are not able to fully capture nonlinear soil behavior, the incorporation of additional site parameters, such as depth to bedrock or parameters describing shallow soil conditions, may improve the predictions. However, due to the high site-to-site variability in nonlinear behavior, the correlation between nonlinearity and the site parameters is generally low. For future studies, nonlinear soil behavior and the relation with site parameters should therefore also be systematically estimated for other networks and regions than KiK-net, for example, in Europe and the Mediterranean region, where the potential of large-scale site condition models from fields outside engineering seismology (e.g., hydrology and geomorphological) as alternative site proxies for site amplification predictions has been demonstrated.
Author: Løviknes, Karina
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6889641604
Cites:
Cited by:
Related to: