Insights from KiK-Net Data: What Input Parameters Should Be Addressed to Improve Site Response Predictions?

Site response models have been found to underpredict high-frequency ground motions in the aggregate, and we hypothesize these persistent biases are due to breakdowns in the one-dimensional (1D) site response assumptions and/or poorly characterized soil properties. Using data from Japan's Kiban-Kyoshin (KiK-net) network of vertical seismometer arrays, we test four physical explanations for this persistent bias by adjusting soil profiles and material parameters to address possible shortcomings in these model inputs. The physical adjustments include: (1) applying a depth-dependent shear-wave velocity (VS) gradient within layers, (2) decreasing the small-strain damping ratio, (3) increasing the small-strain shear modulus, and (4) randomizing the VS profile. Of these adjustments, we find that the usage of a depth-dependent VS gradient most greatly reduces the high-frequency bias, and produces more realistic strain profiles and surficial ground motions at the test site in this study (FKSH11). With regards to 1D site response model improvement, this study suggests that greater attention should be paid to soil profiles and material parameters, as some of these physical adjustments are more successful at reducing model bias than changing the constitutive model type.