A New Algorithm for Significant Wave Height Estimation From X-Band Radar Images

Accurate estimation of significant wave height (SWH) is critical for understanding ocean surface dynamics and supporting marine operations. In this study, a novel approach combining wavelet scattering transform (WST) and convolutional neural network (CNN), termed WST-CNN, is proposed for SWH estimation using X -band marine radar images. The hierarchical and multi-scale features extracted by WST effectively capture the intricate patterns of ocean surface motion while being robust to noise and translation invariance. Then, these features are processed by a CNN to map the radar data to corresponding SWH values. The radar data in this work were collected in 2008 from a ship-mounted Decca radar operating in a sea area approximately 300 kilometers off the coast of Halifax, Canada. Ground truth SWH values were measured by buoys deployed around the data collection vessel. The CNN method is also applied to the radar data and used for comparison with the proposed method. Results show that the WST-CNN method achieves a better accuracy, with a root mean square error (RMSE) reduced by 0.03 m and bias decreased to 0.01 m compared to CNN before averaging. Additionally, the proposed WST-CNN method achieves an RMSE reduced to 0.15 m after averaging, along with an improved correlation coefficient (CC) of 0.98.