Relevance vector machines modeling of nonstationary ground motion coherency

Abstract

Conventional coherency models do not consider the nonstationary character of ground motions. Accurate modeling of nonstationary coherency can motivate accurate simulation of spatially incoherent ground motions. In this work, wavelet transform and Relevance Vector Machines are used to develop a regression model for nonstationary lagged coherency that characterizes the space-time variation of seismic ground motions. Earthquake data from four events recorded at dense seismograph SMART-1 array is used in the analysis. The nonstationary lagged coherency is computed using the wavelet transform of strong motion S-wave window, and then fed to the Relevance Vector Machine as training data. A homogeneous isotropic field is assumed. The proposed model does not require a fixed parametric functional form. It has been shown that the proposed model can estimate the lagged coherency for different separation distances at different time instants.