The wave hindcast NORA3, with a 3 km resolution in a Pan-Arctic domain, has been widely used for weather window marine operation, climate trends, extreme analysis and downscaling. The hindcast has produced very good results. In this project we want to use NORA3 as a training set for a data-driven (ML) wave forecast. Since the training is done with analysis and not with forecast data it is expected that the data-driven forecast outperforms the regular forecast at some lead times.
The Norwegian Meteorological Institute (MET) provides several wave forecasts used mainly for navigation, extreme warnings, ship routing, safety at sea and maritime operations. In a typical year the forecast fails to be generated between 4 to 8 times. As a secondary goal we propose to establish this data-driven forecast as an alternative forecast system for such situations. The new forecast system will be based on the Anemoi framework which has been used successfully at MET for other applications.
