Matched-field localization with sparse Bayesian learning wavenumber estimation

To address the challenge of shallow water waveguide passive localization being constrained by unknown seabed parameters, this paper proposes a passive localization method for unknown seabed parameters by integrating sparse Bayesian learning-based horizontal wavenumbers estimation. Under conditions where only the water column sound speed profile is available while seabed parameters are lacking, the method estimates the horizontal wavenumbers of each normal mode from vertical array data using sparse Bayesian learning and the finite difference method. Based on the estimated horizontal wavenumbers and the measured sound velocity profile, combined with the finite difference method, the replica field at various ranges and depths can be approximated for matched-field localization. However, the localizable range of this method is limited due to significant estimation errors in the horizontal wavenumbers of lower-order modes. To extend localization capability to distant sources, this paper inverts an equivalent geoacoustics model using the estimated horizontal wavenumbers of higher-order modes (characterized by smaller errors) from sparse Bayesian learning results, thereby correcting the horizontal wavenumber estimation of lower-order modes to enable long-range acoustic source localization. Simulation and experimental data processing results demonstrate that the inversion of the equivalent geoacoustics model effectively reduces estimation errors in lower-order mode horizontal wavenumbers (errors reduced by over an order of magnitude), thereby significantly extending the localizable range.