A fast and accurate hybrid method for short-term forecasting significant wave height

Abstract

This paper proposes a hybrid method for forecasting significant wave height (SWH). The wavelet decomposition algorithm is applied to decompose the original signal into different components, and suitable models are selected to predict wave heights for each component. To validate the performance of the proposed hybrid model in forecasting SWH with varying characteristics, wave data measured from three stations are utilized for both model training and accuracy validation. For each station, a total of 2000 h of continuously observed wave data is selected. The initial 80% of the data is used as training samples, while the remaining 20% is utilized as test samples. Additionally, a data normalization procedure is conducted. In addition to the hybrid model, a set of benchmark models is employed to evaluate the performance of the hybrid model using a series of statistical scoring metrics. The hybrid model, along with benchmark models, is utilized to perform 1-h and multiple-hour lead SWH predictions. The results validate that the proposed hybrid model provides accurate and efficient short-term SWH predictions. The correlation coefficients between the one- and three-step predictions and their corresponding observations are greater than 0.99, while they are greater than 0.96 for the five and eight-step predictions.