Uncertainty analysis in matched-field geoacoustic inversions

Area of Science:

• Geophysics
• Ocean Acoustics
• Statistical Modeling

Background:

• Accurate parameter estimation in geoacoustic inversions relies on quantifying data uncertainties.
• Ambient noise and modeling errors contribute to uncertainty, often modeled by error variance.
• The error variance parameter is typically unknown a priori, complicating uncertainty quantification.

Purpose of the Study:

• To develop and compare Bayesian methods for quantifying uncertainty in matched-field geoacoustic inversion parameter estimates.
• To address the challenge of unknown error variance in Bayesian geoacoustic inversion.
• To evaluate the performance of full and empirical Bayesian approaches in handling error variance uncertainty.

Main Methods:

• Implemented full Bayesian approach to propagate error variance uncertainty through parameter estimation.
• Developed empirical Bayesian approach conditioning posterior distributions on a point estimate of error variance.
• Utilized both synthetic and experimental data for validation and comparison.

Main Results:

• Demonstrated that both full and empirical Bayesian approaches can quantify uncertainty in parameter estimates.
• Showcased the impact of error variance uncertainty on the reliability of inversion results.
• Provided comparative analysis of the two Bayesian methods using diverse datasets.

Conclusions:

• The proposed Bayesian methods offer robust frameworks for uncertainty quantification in geoacoustic inversions.
• The empirical Bayesian approach provides a computationally feasible alternative for handling error variance uncertainty.
• Accurate uncertainty quantification is crucial for reliable interpretation of geoacoustic model parameters.