Abstract
Accurate significant wave height prediction is essential for fuel-efficient ship operation and weather routing, as wave-induced resistance directly affects propulsion demand and fuel consumption. This study proposes a Residual U-Net-based deep-learning correction model to improve long-range SWH forecasts from WAVEWATCH III (WW3). WW3 global forecast fields were corrected using the proposed model, with CMEMS reanalysis data used as the ground-truth reference. The corrected outputs, denoted as WW3_UNET, were evaluated against 10 min resolution main engine fuel oil consumption (ME1_FOC) records and onboard wave observations from a commercial vessel traversing the South Atlantic in 2025. WW3_UNET showed markedly improved agreement with ship observations compared with the raw WW3 forecast across all lead times from 0 to 288 h. When a 24 h moving average was applied, WW3_UNET achieved a correlation of 0.720 with ME1_FOC at the 168–180 h lead time, closely approaching the 0.736 obtained from onboard wave measurements. These results indicate that AI-corrected forecasts can provide observation-consistent wave information up to 7–8 days in advance. The proposed approach can support fuel-aware weather routing and voyage planning, thereby contributing to improved maritime energy efficiency and decarbonization.
Showing the abstract — retrieve the full paper via the Exa API.