Deep Learning Fills the Gap Between Global Wave Data and the Water Near Shore

A deep-learning model can estimate wave conditions about a kilometer off the coast from coarse global offshore data, a team at Sun Yat-sen University reports in Communications Earth & Environment. In benchmark tests along the United Kingdom coast, the paper says, the model matched buoy observations with a coefficient of determination above 0.90.

Global wave reanalysis products, long-term reconstructions of past sea states, resolve the ocean at a spacing of about 30 kilometers, the authors write, a scale that they say "limits their direct applicability to the nearshore zone." The paper puts that zone at roughly one kilometer from shore and describes nearshore wave data as critical for coastal protection and environmental management.

The model combines a deep-learning surrogate, trained to reproduce high-resolution nearshore waves from offshore wave conditions, with what the authors call an "observation-driven residual correction sub-model," which adjusts the output using measurements. They describe the pairing as "a composite solution with both accuracy and efficiency."

Existing approaches, the authors write, "often face a trade-off between cross-regional applicability and high-resolution local accuracy." The 0.90 result is from the United Kingdom benchmark alone; for other coastlines, the authors report that evaluation at multi-regional buoys "further supports the transfer of the framework to global open coast transects."