| Popis: |
This study presents an artificial intelligence (AI) model to forecast time-series wind waves in the US Atlantic Coast. The fundamental technique in the proposed model is the Long Short-term Memory model, which is able to learn patterns from data sequence. The model is trained using historical wind data, wave data, temperature data, and atmospheric pressure data. Then the model is used to forecast significant wave height, average wave period, and mean wave direction. A 2-year meteorological data measured at four NOAA buoy stations in the US Atlantic Coast are used to train and evaluate the model’s forecast skills. The results show the artificial neural network model converges fast and does not have over-fitting nor under-fitting issues. Furthermore, a short-term forecast (e.g., 1 to 6 h) achieves higher accuracy than a long-term forecast (e.g., 24 to 48 h). This study shows that sufficient forecast accuracy can be obtained by using the input hours equal to the forecast lead time. Comparison using rose plots confirms that the artificial neural network model reproduces wave height and wave period statistics very well. The artificial neural network model is used to forecast storm waves induced by Hurricanes Isaias and Eta in the US South Atlantic region, and two winter storms in the US North Atlantic region. The model forecast is compared with NOAA measurement. The results show that the artificial neural network model is able to accurately forecast both significant wave height and average wave period associated with storm events when the forecast lead time is relatively short (e.g., 1 to 6 h). The proposed artificial neural network model can serve as an alternative tool to traditional coastal models for wave prediction and storm forecast. |
Full Text from ScienceDirect