Answers are grounded in this paper and cite their sources.
Conventional wave measurements from moored buoys are limited to coastal waters, leaving a critical gap in observing complex sea states with high spatial resolution in the open ocean. To address this deficiency, the present study proposes a novel methodological framework that synergistically integrates synthetic aperture radar (SAR) imagery with deep learning techniques. The principal innovation of this approach resides in the derivation of a first-guess wave spectrum from a SAR-retrieved wind field, which is subsequently employed as a Joint North Sea Wave Project (JONSWAP) prior within a transformer-based 1-D wave spectrum inversion algorithm. Model training and validation are conducted using collocated observational records sourced from the National Data Buoy Center buoy network. A transformer-based model was developed for wind speed retrieval from SAR images. Validation against approximately 70 000 collocated Haiyang-2 scatterometer measurements demonstrated high accuracy (root-mean-squared error of 1.04 m/s, r= 0.95), with a case study confirming numerical stability and the absence of inversion failures. A transformer-based wave spectral inversion algorithm was developed by incorporating three physically complementary input features: the SAR intensity spectrum, a first-guess spectrum constructed via the JONSWAP formulation using SAR-derived wind speed, and vertical–vertical polarization modulation transfer functions. Rigorous validation against the surface waves investigation and monitoring spectrometer, altimeter, and buoy data demonstrated that the transformer-based retrievals of the 1-D wave spectrum and significant wave height (SWH) exhibit strong agreement with reference measurements. The transformer-based model outperformed extreme gradient boosting across all datasets, effectively capturing spatial variations of SWH and demonstrating consistency with altimeter measurements, excelling in complex wave systems beyond the capabilities of traditional methods.
No takes yet. Share an insight, caveat, or question.
Hu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: