Abstract
Short-term ocean waves forecasting requires a high degree of skill and knowledge, as one should observe the available model forecast and real-time measurement and reach a combined estimation. This paper presents a deep learning model providing a short-term wave height prediction derived from recent in-situ measurements and an available mid-range forecast. The model is based of a gated recurrent unit, which is common in time-series forecasting. The model is able to improve significant wave height RMSE by as much as 76% for 1 h forecasts and converge to ∼12% improvement for forecasts over 7 h. The model is also shown to be easily transferable to another station and achieves good performance without further training in a ”zero-shot” learning process. This model can prove valuable to various off-shore operations, allowing for data-driven decision making instead of skilled human operator and experience-based evaluation.
Original language | English |
---|---|
Article number | 113389 |
Journal | Ocean Engineering |
Volume | 268 |
DOIs | |
State | Published - 15 Jan 2023 |
Keywords
- Data-based wave forecasting
- Machine learning
- Numerical wave forecasting models
- Short-term wave forecast
- Wave buoy measurements