Geoacoustic inversion with generalized additive models.

Area of Science:

• Geophysics
• Ocean Acoustics
• Machine Learning

Background:

• Geoacoustic parameter estimation is crucial for understanding underwater environments.
• Acoustic signal propagation is influenced by environmental factors like sediment properties.
• Traditional methods for geoacoustic estimation can be complex and time-consuming.

Purpose of the Study:

• To develop a machine learning framework for geoacoustic parameter estimation.
• To establish a relationship between acoustic signal features and geoacoustic properties (sound speed, attenuation).
• To enable efficient prediction of geoacoustic parameters using generalized additive models.

Main Methods:

• Feature extraction from broadband acoustic time-series.
• Application of generalized additive models (GAMs) with smoothing splines for non-linear regression.
• Training GAMs on noise-free data from known environments.
• Prediction of geoacoustic properties using noisy data from diverse environments.

Main Results:

• Distinct acoustic signal structures correlate with different propagation environments.
• Extracted features (peak amplitudes, kurtosis, signal strength, decay, time differences) effectively characterize signals.
• The proposed multivariate GAM accurately predicts sediment sound speed and attenuation.

Conclusions:

• Machine learning, specifically GAMs, provides an efficient and effective approach for geoacoustic parameter estimation.
• Feature engineering from acoustic time-series is key to successful geoacoustic prediction.
• This method offers a robust way to analyze underwater acoustic data for environmental characterization.