Trans-dimensional inversion of seabed reflectivity data for poro-ealstic (dispersive) layered models

This paper considers trans-dimensional Bayesian inversion of seabed acoustic reflectivity data for frequency-dependent compressional- and shear-wave velocity and attenuation in arbitrarily layered poro-elastic media. Seabed reflectivity data are measured on a bottom-moored hydrophone and processed to give spherical-wave reflection coefficients as a function of frequency and angle. The seabed is modeled using Buckingham's viscous grain shearing model and layers are parametrized in terms of thickness, porosity, compressional and shear grain-to-grain moduli, material exponent, and the compressional visco-elastic time constant. These parameters are used to compute density, compressional- and shear-wave dispersion curves, and compressional and shear attenuation-frequency curves as a function of depth. The dispersion-depth curves are used to predict spherical-wave reflection coefficients over large frequency (300-3000 Hz) and grazing angle (12-75 degrees) bands, which include the effects of shear waves in arbitrarily layered media. The computationally expensive integration of the Sommerfeld integral is implemented massively parallel on a general purpose graphics processing unit. The seabed model is trans-dimensional and estimates the layering from the data for each parameter to ensure parsimonious parametrization. The ability to resolve dispersive shear-wave velocity and attenuation structure is studied using simulated data and data from the Tyrrhenian sea. Residual errors are modeled hierarchically using a trans-dimensional autoregressive model. [Work supported by the Office of Naval Research]