Geoacoustic inversion using Bayesian optimization with a Gaussian process surrogate model.

Area of Science:

• Oceanography
• Geophysics
• Computational Science

Background:

• Geoacoustic inversion is crucial for understanding the seafloor but is computationally intensive in high-dimensional spaces.
• Traditional methods often require thousands of forward model evaluations, limiting practical application.

Purpose of the Study:

• To demonstrate the efficacy of Bayesian optimization (BO) as an efficient global optimization method for geoacoustic inversion.
• To reduce the computational cost associated with estimating geoacoustic parameters in high-dimensional spaces.

Main Methods:

• Utilized Bayesian optimization (BO), an iterative global optimization technique.
• Employed Gaussian process surrogate models and heuristic acquisition functions (upper confidence bound, expected improvement) to guide the search.
• Defined the objective function as the Bartlett power for parameter estimation.

Main Results:

• Achieved accurate geoacoustic parameter estimation in a seven-dimensional space using fewer than 100 evaluations.
• Demonstrated comparable results to differential evolution optimization for both simulated and experimental shallow-water data.
• Showcased BO's ability to balance exploration and exploitation in parameter space.

Conclusions:

• Bayesian optimization offers a computationally efficient alternative for high-dimensional geoacoustic inversion.
• This method rapidly estimates optimal geoacoustic parameters, making it suitable for complex environments.
• BO provides a robust approach for analyzing both simulated and real-world acoustic data.