Data-Driven Phase-Resolved Sea Surface Reconstruction From Synthetic X-Band Radar Data

Accurate phase-resolved information on the sea surface can be crucial for well-founded decision making in the maritime sector. This information can be obtained by analyzing X-band radar measurements, which contain sparse information on the surrounding sea. This analysis, referred to as sea surface reconstruction, usually relies on simplified physical models or computationally expensive optimization procedures, thus creating a trade-off between accuracy and computational cost. This work proposes a purely data-driven approach, which aims at providing accurate sea surface reconstruction from X-band radar data in real-time. For this, state-of-the-art methods from computer vision and deep learning were combined to a model, which maps successive historic radar images to the 2D phase-resolved sea surface. The training data were generated using the high-order spectral method to model nonlinear hydrodynamic effects, and features a wide range of governing sea-state conditions. The synthetic radar images were derived from these simulations using a numeric radar model. The results show that the proposed data-driven approach is capable of faithfully reconstructing the 2D sea surface from sparse radar information over a wide range of governing sea state parameters. Moreover, the approach is able to extrapolate over the radar blind zone, yielding a complete reconstruction of the sea surface within the radar radius.