Underwater 3D sound speed field reconstruction based on block term tensor decomposition

The three-dimensional sound speed field (SSF) is of great significance in underwater acoustic research; however, the high cost of maritime observation often leads to sparse and limited measurement data, making accurate SSF reconstruction a challenging yet valuable problem. Traditional inversion methods frequently suffer from high data requirements, an inability to process complex spatiotemporal features, and issues regarding accuracy and stability. To address these challenges, this paper proposes a 3D SSF reconstruction method that combines Block Term Tensor Decomposition (BTD) with sparse reconstruction. Specifically, BTD is utilized to extract latent information and structural features from high-dimensional historical ocean sound speed data, enabling precise 3D SSF reconstruction by integrating a small number of newly acquired regional observations through sparse reconstruction techniques. The proposed method was validated using the Argo dataset and experimental sea trial data from underwater gliders. Experimental results demonstrate that when reconstructing with limited observations, the BTD-based approach improves reconstruction accuracy by more than 49.25% compared to traditional Empirical Orthogonal Function and Tucker decomposition methods. Overall, utilizing BTD for 3D underwater sound speed field reconstruction represents a novel, high-precision, and cost-effective methodology that effectively overcomes data sparsity constraints.