Mesoscale eddies (MEs), a prevalent natural phenomenon in the oceans, are pivotal to the oceans temperature and salt structure, as well as the acoustic propagation mechanism. In the offshore environment of the mid- and far-ocean, the effective use of scarce ME environmental data for accurate acoustic field reconstruction is a pressing issue in marine scientific research. To address this, our study introduces a hybrid eddy identification algorithm, merging JCOPE2M high-resolution reanalysis data and AVISO satellite altimeter data. This algorithm aims to capture sound velocity profile sample data of MEs, laying a robust foundation for subsequent analysis and modeling. We then delve into the training and reconstruction of the generative adversarial network model based on these valuable sample data. The primary goal of this step is to characterize the sound field of MEs to more accurately reflect their actual physical properties. To ensure the reliability and validity of the reconstruction method, we design a comprehensive evaluation system with RMSE and SSIM, and convergence zone (CZ) accuracy as the main evaluation indexes. After a series of experimental validations, the MAT model reconstruction method we adopted demonstrated a significant improvement in accuracy. Specifically, the RMSE value of the 2D reconstruction results is as low as 1.9 m/s, the SSIM value is as high as 0.72, and the average accuracy of the CZ is more than 60%. Compared with other previous reconstruction methods, ours is more accurate in characterizing the ME sound field and achieves a significant improvement in the accuracy of field prediction. This result provides a new technical path for ME fine-grained sound field reconstruction and a valuable reference for more underwater acoustic research and application.