Near-real-time two-dimensional wave spectrum retrieval from Gaofen-3 wave mode imagettes

Near-real-time ocean monitoring using synthetic aperture radar (SAR) satellites aims to rapidly characterize the sea states from SAR data. Although recent studies have demonstrated near-real-time retrieval of significant wave height and mean wave period, these integrated parameters often fail to capture the full complexity of sea states, such as peak direction and peak wavelength. The two-dimensional (2-D) wave spectrum, by contrast, provides a more complete and physically informative representation. However, the inversion of 2-D wave spectra often needs prior information, which limits the development of SAR-based near-real-time ocean monitoring. To eliminate the requirement of priors, we combine SAR ocean wave imaging theory with deep learning to perform near-real-time 2-D wave spectrum retrieval from Chinese Gaofen-3 (GF-3) wave mode imagettes. Specifically, a trained deep learning model generates a normalized wave spectrum from the SAR spectrum. The ultimate wave spectrum is then recovered by refining the absolute intensity values through the iterative Levenberg–Marquardt optimization. Retrieved spectra from GF-3 SAR are validated against allocated 2-D buoy spectra. The averaged retrieval time is under 20 seconds, making the proposed method suitable for near-real-time SAR ocean wave monitoring.