A Bayesian Inversion Approach for Site Characterization Using Surface Wave Measurements

This paper presents a Bayesian inference method for the characterization of soil properties and stratigraphy for site response analysis using surface wave measurements. The method is evaluated using numerically simulated data for surface wave methods (e.g. SASW) and ambient noise methods (e.g. H/V) using horizontal, homogeneous, layered soil models. Wave propagation-both vertically propagating shear waves and horizontally propagating surface waves-through one-dimensional horizontally layered media and a linear soil constitutive model are assumed for the simulation of measured data. The inversion process is performed using error functions defined as the difference between selected features of model-predictions and measurements. The considered features include dispersion curves obtained from surface wave propagation and the estimated transfer functions from H/V spectral ratios of ambient noise assuming vertically propagating shear waves. The numerical study was performed to evaluate the proposed approach for horizontally layered sites where the number of layers, layer thicknesses, and layer properties are varied. Different levels of measurement noise, modeling errors, number of updating parameters, number of data sets, and error functions are considered. The numerically simulated sensor data are polluted with independent Gaussian white noise vectors at three different noise levels (0.5%, 1%, and 2% in terms of response root-mean-square) and their effects on bias and covariance of updating parameters are studied. Furthermore, the sensitivity of the results to the updating parameters (e.g., layer height, density, damping, and stiffness), available measurements (type, quantity, quality) and the used error functions (e.g., selected points on dispersion curves or H/V transfer function) are investigated. The value of information will be assessed for the inclusion of the ambient noise data in the Bayesian inversion in terms of reduction in site parameter uncertainty.