Consistency assesment of wave directional spectrum predictions from machine learning based ship-as-a-wave-buoy methods

Real-time knowledge of the ocean wave directional spectrum is valuable for maritime safety and navigation. When end-to-end machine learning is used to learn ocean wave directional spectra from ship motions, the spectrum predictions may be inconsistent with measured ship motions from a physical point of view. In order to assess this consistency, a method is proposed that makes use of a numerical seakeeping code to recalculate ship motions from predicted wave spectra. The recalculated ship motions are subsequently compared to the measured ship motions. The method is applied using data from three different vessels, each having a different level of data completeness. For these
cases it is concluded that the consistency can be assessed when the sea state is not too extreme. The method may be improved by using higher fidelity seakeeping codes.