Significant wave height retrieval from Sentinel-1 SAR imagery by convolutional neural network

Significant wave height (SWH) is an important wave parameter that is related to near-shore activities and research on the phenomenon near the air-sea interface. Here, we proposed a new method for retrieving SWH from C-band Sentinel-1 synthetic aperture radar (SAR) interferometric wide mode data based on convolutional neural network (CNN), which can directly establish the empirical relationship between normalized radar cross section and SWH. We collected 1597 Sentinel-1 SAR images matched with in situ buoys and conducted homogeneity tests for each of the matched sub-images, producing ~ 3330 matchups with 2028 of them being VV-polarization. After training, the VV-polarization data extracted by 0.5 threshold for homogeneity test perform better, and the comparison between these results and in situ buoy measurements in validation data indicates a SWH root mean square error of 0.32 m, a 23.58% scatter index and a 0.90 correlation coefficient. And the SWH from CNN-based method is also validated with radar altimeter data and Wavewatch3 data. These results demonstrate that the proposed CNN method is suitable for retrieving SWH from Sentinel-1 SAR imagery with some constraints on the matched dataset.