Evaluation of Biases in Predictions from Total-stress and Effective-stress Ground Response Modeling Methods by Using Vertical Arrays in Taiwan

When earthquake waves propagate through geologic deposits, frequency content and amplitude of the waves would be changed. Ground response analysis can be performed to model the wave propagation process by incorporating detailed site conditions. These analyses can be in either frequency domain or time domain. In engineering practice, equivalent linear frequency analysis is widely used because of its simplicity in parameter selection and code usage. However, equivalent linear analyses may not be able to accurately model the nonlinear behavior of soil, especially when shear strain level is high. Nonlinear time domain analyses can accurately represent the soil nonlinear behavior and thus have the potential to better predict the ground response (compared to equivalent-linear analysis). Also, nonlinear time domain analyses can allow the soil stress to be evaluated in terms of total stress or effective stress. In addition, pore water pressure generation and dissipation can be evaluated through effective stress ground response analyses. The construction of surface-downhole monitoring network is one of the important tasks for strong motion monitoring by the Central Weather Bureau in recent years. Almost fifty vertical arrays have been installed in Taiwan until Year 2015. The objective of this research is to evaluate the performance of different ground response analysis methods by utilizing the ground motion data recorded at selected vertical arrays. To facilitate this evaluation, geotechnical models were first developed and the downhole records were used as the input motions for equivalent-linear frequency-domain and nonlinear time domain analyses. It is found that some of the shear wave profiles provided by the Central Weather Bureau may contain errors. Improved predictions can be obtained when the profiles are modified according to nearby site data. In addition, predicted surface response from different analysis methods are generally similar as most input motions are weak motions. By comparing the predicted and recorded surface response spectra, there is under-prediction at short periods while over-prediction can be observed at long periods.