Near real-time significant wave height prediction along the coastline of Queensland using advanced hybrid machine learning models

Abstract

The accurate prediction of significant wave height is essential for coastal and offshore engineering, and is especially important for producing renewable ocean wave energy. However, H_s is traditionally predicted using empirical or numerical models, which lack accuracy, are computationally demanding, or require extensive datasets. Due to chaotic nature, it is very challenging for empirical or numerical models to precisely predict H_s. This study developed and tested several standalone machine learning models for H_s prediction and explored hybrid versions of these models based on additive regression to further improve model accuracy. Half-hourly H_s data along with common variables measured at ocean buoys were collected from four sations (i.e., Mooloolaba, Gladstone, Caloundra and Brisbane) along the coastline of Queensland, Australia and used to develop the ML models. The ML models were tested for their ability to accurately predict H_s at Mooloolaba station and were transferred to the three other stations to prove their spatial generalization capabilities. Overall, the results demonstrate that the ML models, and especially their hybrid versions, can accurately predict H_s at Mooloolaba as well as for other stations. Thus, the proposed models may serve as promising tools for improving ocean wave energy production.