Sensitivity of 1-D ground motion predictions to analysis codes and material models using KiK-net vertical arrays

Abstract

One-dimensional (1-D) ground response analysis (GRA) predictions are highly sensitive to the analysis methods and dynamic soil models employed, contributing to significant scatter in analysis outputs. In this paper, a parametric study of 1-D GRA is conducted to evaluate the performance of various models using 60 ground motion records of varying intensity from five KiK-net downhole array stations in Japan. Four GRA codes are considered, including the equivalent-linear (EL) method, a modified EL method using frequency-dependent soil parameters (FDEL), and two nonlinear (NL) models computed in OpenSees and DEEPSOIL. The soil constitutive models are constrained using a set of five material curves representing a wide range of nonlinear soil behaviour. It was found that all models are sensitive to the material curves at high frequencies, and the NL models are highly sensitive to the parametrization of Rayleigh damping. As the levels of soil nonlinearity increase, all GRA methods tend to over-damp the higher frequency content of ground motions, with substantial discrepancies across the model predictions. Interestingly, the FDEL method displays a lower sensitivity to material curves and reduces the biases at high frequencies. The results of this study suggest that GRA codes and material curves contribute evenly to the overall dispersion when simulating strong ground motions. To quantify the variability of GRA models, along with the epistemic uncertainties, a linear regression relationship in bi-logarithmic space is established across all records between the total standard deviation and the shear strains developed within the soil profiles.