Author(s): Dosso Stan, Gosselin Jeremy, Cassidy John
This paper presents Bayesian environmental inversion of ambient noise data in both marine and terrestrial settings. The marine setting involves inverting seabed Scholte-wave dispersion curves estimated from ambient noise data recorded on a seafloor hydrophone array, while the terrestrial setting involves inverting Rayleigh-wave dispersion from ambient noise data recorded at a seismometer array. In all cases nonlinear Bayesian (probabilistic) inversion methods are applied to quantify the information content of the ambient-noise data to resolve geophysical profile structure and provide rigorous uncertainty estimates. An important component of this work involves model selection methods to determine appropriate profile parameterizations that are consistent with the resolving power of the data and with available prior information (e.g., the expectation of geophysical profiles characterized either by discrete layering or by smooth gradients). Layered profiles are effectively parameterized in trans-dimensional (trans-D) inversion, which samples probabilistically over an unknown number of layers and includes the uncertainty of the parameterization within the parameter uncertainty estimates. A relatively new approach to smooth-gradient parameterizations is based on a linear combination of Bernstein-polynomial basis functions, with the polynomial order determined using the Bayesian information criterion. This approach can represent general smooth gradients with a small number of parameters (basis function coefficients), and allows the data, rather than subjective choices, to define the form of the profile.
Name: Prof Stan Dosso