Project Description
Numerical wave models provide detailed information about wave conditions up to seven days in advance, and are critical to a number of industry applications – both for marine safety, and to inform the best time for complex offshore operations. However, virtually all operational wave models rely on phase-averaged wave models – meaning that rather than modelling the evolution of individual waves, the average properties of the sea state is modelled in frequency and directional space. As a result, and by virtue of being reliant on atmospheric models for wind input, spectral wave models often have difficulty predicting the magnitude and arrival time of wave events, particularly for swell events.
The project trained a machine learning algorithm, using historical and ongoing wave forecasts along with observations from wave buoys, to identify and correct consistent errors in wave forecast. Results from this prototype indicated that the machine learning methodology reduced forecast errors by as much as 20% and as a result, the Machine Learning system has been continually developed under sponsorship from the Bureau of Meteorology (BoM) and industry.
