Wave energy converter (WEC) siting is hindered by discrepancies between coarse-resolution reanalysis, such as the Copernicus Marine Service (CMEMS), and local wave dynamics. This study propagates wave-regime labels from fourteen Acoustic Doppler Current Profiler (ADCP) stations, spanning 2015–2025, onto the full CMEMS grid along the Omani coast at three-hourly resolution. A deep autoencoder that combines wave parameters with local bathymetry, capturing depth-induced transformations, identifies four physically interpretable regimes: Golden Swell, Storm, Confused Sea, and Ambient/Calm. A cost-sensitive stacking ensemble propagates these regimes grid-wide, trained only on the twelve station segments that pass a Fidelity Gate, which admits an anchor where CMEMS reproduces its wave climate to within a correlation of 0.80 and a wave-height bias of 0.50 m. The ensemble attains a Balanced Accuracy of 70.7% (95% CI: 66.3–74.6), and the gate improves it by 5.5 points across ten seeds (p < 0.001). Storm detection reaches an AUC of 0.93, four cyclones are flagged as anomalies, and off-cyclone storm activity follows the Southwest Monsoon cycle. Bi-objective screening of cells shallower than 100 m yields a fifteen-cell Pareto front trading swell potential against storm exposure, while leave-one-station-out validation (48.7 ± 12.4% Balanced Accuracy) bounds the atlas to monitored regions.
No takes yet. Share an insight, caveat, or question.
Shahmiri et al. (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: