A comparison of three geoacoustic models using Bayesian inversion and selection techniques applied to wave speed and attenuation measurements.

Abstract

Many geoacoustic models have been developed to study sandy sediments. In this work, Bayesian inference techniques are used to compare three such models: the VGS(λ) model, the most recent of Buckingham's viscous grain-shearing models, the Biot-Stoll poroelastic model, and an extension to the Biot-Stoll model proposed by Chotiros called the corrected and reparametrized extended Biot (CREB) model. First, Bayesian inversion is applied to wave speed and attenuation measurements previously made in the laboratory to determine the degree to which each of the model input parameters can be resolved by wave speed and attenuation data. Then, Bayesian model selection techniques are utilized to assess the degree to which the predictions of these models match the measured data and to ascertain the Bayesian evidence in favor of each. Through these studies it is determined that the VGS(λ) and CREB models outperform the Biot-Stoll model, both in terms of parameter resolution and in their ability to produce predictions in agreement with measurements. The VGS(λ) model is seen to have the highest degree of Bayesian evidence in its favor.