The EN regression models are implemented for estimating significant wave height (SWH) from quad-polarization Gaofen-3 SAR wave mode data based on the collocated data set of ~11200 Gaofen-3 imagettes matched with SWH from ERA5 reanalysis. The importance of SAR features for SWH estimation from EN is analyzed. The model performance is evaluated through a comparison with observations from buoys and altimeters. The results show that the 20 EOF spectral parameters, NRCS, cvar, and θ are significant for EN to estimate SWH from Gaofen-3 SAR. The EN models achieve good performance with RMSEs smaller than 0.5 m. The co-polarization models show better performance at low sea states but worse performance at high sea states compared to the cross-polarization models.

The Surface Wave Investigation and Monitoring (SWIM) aboard the China-France Oceanic Satellite (CFOSAT) provides the ocean wave spectrum (70–50 m wavelength range). However, the accuracy of this data is affected by speckle noise, low-frequency parasitic peaks, and missing information in the short wavelength range. To improve the accuracy of the SWIM wave spectrum, this letter introduces a vision transformer (ViT) deep learning (DL) model combined with a deconvolution block, which leverages buoy wave spectrum and full wavenumber wind wave spectrum to improve the SWIM wave spectrum with high precision and wide wavelength range. The results show that the linear correlation coefficient of the improved wave spectrum has increased from 0.510 to 0.833. Furthermore, the accuracy of spectrum parameters is enhanced. Particularly, compared with the original SWIM spectrum, the root mean square error (RMSE) for the mean wave period (MWP) and peak wave period (PWP) decreased by 70.19% and 71.68%, respectively.

The Surface Waves Investigation and Monitoring (SWIM) omnidirectional wave spectra and its shape parameters are compared by matching the SWIM wave spectrum data with the buoy wave spectra of the National Data Buoy Center (NDBC). The results show that the omnidirectional wave spectrum of SWIM has significant overestimation at low frequencies and significant underestimation at high frequencies. And there are spurious peaks at low frequencies. The overestimation and underestimation of the SWIM 6° beam wave spectrum are the most obvious. We believe that the presence of spurious peaks is due to the amplification of the noise floor in the SWIM omnidirectional spectrum at low frequencies. This phenomenon is greatly relieved with the increase of Significant Wave Height (SWH) . The difference between the SWIM spectrum shape parameters frequency spread (σf) and the "peakedness" of the omnidirectional spectrum (Qp) and the buoy is large. The difference between them decreases significantly with increasing SWH.

The 200 dual-polarized Sentinel-1 SAR images covering typhoons from 2018 to 2022 are collocated with ERA5 reanalysis data. The SAR-ERA5 collocations are randomly divided into two subsections: one for model training (80%), and the other for independent testing (20%). Based on the selected training samples, three machine learning models, including the Back Propagation Neural Networks (BPNN), the Random Forest (RF), and the Gaussian Process Regression (GPR), are built for the estimation of typhoon sea surface wind speed (SSWS) from VV data. The results show that the three machine learning models achieve significantly better performance than the traditional method. BPNN has the best performance with the bias, root mean square error (RMSE) and scattering index (SI) respectively being -1.33 m/s, 3.25 m/s and 1.5%. The GPR model performs slightly better than BPNN at higher wind speeds. However, the performance of RF is relatively poor. The additional introduction of VH information improves the model performance.

The dependence of azimuth cutoff wavelength (λc) on significant wave height (SWH) and wind speed (U) at C-band VV, HH, VH, and HV polarizations is analyzed based on the collocations between the quad-polarization Gaofen-3 SAR wave mode images and the ERA5 wind and wave reanalysis. Then the influence of pixel spacing on the dependence is discussed. The results show the co-polarized (VV and HH) λchas an evident positive dependence with SWH (with U), while for the cross-polarization (VH and HV), it is relatively weaker. In addition, the cross-polarized λcis more related to$U$than to SWH. Moreover, the size of pixel spacing affects the estimated value of λc. The dependence of λcon SWH and$U$shows a decreasing trend with pixel spacing increasing in all four polarizations. But this trend is more significant for co-polarization, especially for VV.

The upwind-downwind asymmetry in radar return from the sea surface is well known. This paper develops a modified two-scale model to better describe the difference between upwind and downwind of the radar backscatter at moderate incidence angles caused by skewness of the non-Gaussian sea surface. The unknown parameter in the modified model is estimated by fitting the model to CMOD5.n at different incidence angles under various wind conditions. Then the modified model is compared with CMOD5.n and the advanced scatterometer (ASCAT) backscatter measurements. The results show that the model predictions are in rather good agreement with the reference data. The modified model can accurately describe the difference between the upwind and downwind normalized radar cross section (NRCS) with the root mean square difference being about 0.02 dB compared with that of CMOD5.n.