Bayesian calibration of underwater propagation models under uncertainty

Abstract

Propagation models are notorious for the uncertainty of important parameters such as source strength, speed of sound profiles, and reflecting surface profiles. In many cases, one calibrates a model to measured data (e.g., sound levels or transmission loss) for the purposes of estimating these model parameters, i.e., for inverse modeling. Bayesian calibration methods have been developed that are extremely useful for calibration of models where parameters have high levels of uncertainty and problems may be under or over determined. The Kennedy and O’Hagen framework which uses a Gaussian process surrogate model to replace the model under calibration is especially useful when the underlying model is computationally expensive, and so, it may be difficult to apply many optimization based calibration methods. In this talk, we describe the application of the Kennedy and O’Hagen Bayesian Calibration framework to the calibration of an underwater ray tracing propagation model. The source strength and parameters for the sound speed profile are considered as highly uncertain. The Bayesian calibration technique is shown to improve model prediction and reduce the uncertainty of the unknown propagation parameters.