Reconstructing the ocean sound speed field via regularized tensor network decomposition

Area of Science:

• Oceanography
• Applied Mathematics
• Geophysics

Background:

• Accurate reconstruction of 3D oceanic sound speed fields (3D SSFs) is crucial for underwater acoustics and oceanographic research.
• Current methods often struggle with accuracy due to insufficient global correlation analysis.

Purpose of the Study:

• To develop a novel, high-accuracy method for 3D SSF reconstruction.
• To address the limitations of existing techniques by leveraging global correlations and local smoothness.

Main Methods:

• Formulated the 3D SSF as a third-order tensor.
• Introduced a low-rank tensor modeling framework: the fully connected tensor network (FCTN) model.
• Incorporated factor-based Tikhonov regularization into the FCTN framework, creating the FCTN-T model.
• Developed an efficient proximal alternating minimization (PAM) algorithm with an acceleration strategy for model optimization.

Main Results:

• The FCTN-T model demonstrated superior performance compared to state-of-the-art methods.
• Achieved higher accuracy in reconstructing the ground-truth 3D SSF.
• Showcased significant improvements in computational speed.

Conclusions:

• The FCTN-T model represents a significant advancement in oceanic 3D SSF reconstruction.
• The proposed method provides more faithful approximations of oceanic sound speed fields.
• Offers enhanced accuracy and computational efficiency for researchers.