A Cluster-Based Modeling Framework for Rapid Acoustic Field Prediction in Deep Water

lightweight modeling framework is proposed for rapid prediction of underwater acoustic fields in deep-ocean environments characterized by anisotropic seabed bathymetry. The framework is motivated by the observation that bathymetries with similar geometric characteristics tend to induce acoustic fields with similar large-scale structures. Each bathymetric profile is first represented by a finite set of descriptive parameters. To identify an optimal parameter subset, two root-mean-square deviation (RMSD) matrices are constructed: one based on uniformly sampled bathymetric profiles and the other on the corresponding parameter vectors. The structural consistency between the two matrices is quantified using an eigenvector projection–based similarity measure. Based on the selected parameters, a robust K-means clustering algorithm groups similar bathymetries, and only one representative bathymetry per cluster is used for acoustic field computation. Numerical validation in the South China Sea shows that the proposed framework reduces the number of required acoustic field calculations from 1000 to 48 in an anisotropic region and to 29 in a relatively flat region, while preserving the dominant acoustic field structures, including direct-arrival regions and convergence zones.