A mesoscale eddy reconstruction method based on generative adversarial networks

Mesoscale eddies are phenomena that widely exist in the ocean and have a significant impact on the ocean's temperature and salt structure, as well as on acoustic propagation effects. Currently, utilizing the limited data on mesoscale eddy environments for refined acoustic field reconstruction in offshore conditions at mid-to-far-ocean distances is an urgent problem that needs to be addressed. In this paper, we propose a mesoscale eddy reconstruction method (EddyGAN) based on the generative adversarial network (GAN) model which is inspired by the concept of global localization. We adopt a hybrid algorithm for eddy identification using JCOPE2M high-resolution reanalysis data and Archiving, Validation, and Interpretation of Satellite Oceanographic (AVISO) satellite altimeter data to extract mesoscale eddy sound speed profile (SSP) sample data, and then apply the EddyGAN model to train this dataset and perform mesoscale eddy acoustic field reconstruction. We also propose an evaluation method for mesoscale eddy acoustic field reconstruction that uses RMSE, SSIM, and convergence zone (CZ) accuracy based on World Ocean Atlas (WOA) climate state data completion as indicators. The reconstruction result of this model achieves an RMSE of 1.7 m/s, an SSIM of 0.77, and an average CZ accuracy of over 70%. This method better characterizes the mesoscale eddy sound field than the native GAN and other reconstruction methods, improves the accuracy of mesoscale eddy acoustic field reconstruction, and provides superior performance, offering significant reference value for mesoscale eddy reconstruction technology and subsequent ocean acoustic research.