The blue economy includes activities such as tourism, marine energy, ship traffic that provide resources for economic growth and improved livelihoods. For the proper functioning and success of these activities, accurate predictions of wave height are of vital importance. Also, skillful forecasts are critical for releasing timely alerts to protect coastal communities and save millions of euros and lives. Several agencies forecast waves based on physics-based numerical models, artificial intelligence, and data assimilation. However, these systems are not faultless and they can lack accuracy, especially during extreme conditions. Thus, it is crucial to create methodologies that can contribute to developing more reliable forecast systems. In this regard, data assimilation from observations arises as a relevant tool. The so-called “Error prediction” method has proved to be a suitable approach to improve model predictions of different output variables at a specific location. This method requires 1) computing the errors between the modeled and observed values for past events, 2) developing techniques able to predict these errors and 3) incorporating the
predictions of the error into the forecast systems, increasing their accuracy. A pivotal aspect of this procedure is the prediction of the errors. In that sense, machine learning algorithms are very suitable tools for this purpose owing to their excellent ability to simulate nonlinear mechanisms governing physical processes. During the last years with the expansion of technologies and computational resources, we observe a rapid increase in marine big data, promoting the application of machine learning in oceanic and coastal studies. While the integration of data assimilation and machine learning has yielded promising results in other areas, this has not been deeply investigated in the field of wave forecast systems. Thus, the main goal of this project is to explore the suitability of data assimilation and machine learning techniques to improve the accuracy of the IberiaBiscay–Ireland (IBI) system to forecast wave conditions up to the next 24 hours. The Portuguese coast was selected as a study case where there is available wave information from observations and the IBI forecast system from 2018 to the present. In total, eight bouy stations monitor the wave
conditions along the coast and the data collected demonstrated that this coast is very exposed to energetic events. These extreme conditions represent a unique opportunity to explore the capabilities of data assimilation combined with machine learning models for improving forecast systems. Among the several machine learning techniques in this project, artificial neuronal networks and deep learning algorithms will be explored.