Session: 02-05-01 Data-driven and AI-based Models

Submission Number: 180924

A Data-Driven Machine Learning Approach for the Reconstruction of Ocean Wave Storm Events 

In the field of ocean engineering, the availability of complete and high-resolution wave data is essential. In particular, large databases containing comprehensive storm event histories are critical for improving the reliability of design estimations for offshore and coastal structures, as well as marine energy devices. In this context, this work proposes a data-driven machine learning approach for the reconstruction and generation of ocean wave storm events. The model is trained on storm datasets extracted from historical wave time series provided by NOAA. Storm events are identified and characterized in terms of significant wave height (Hs), peak period (Tp), and mean wave direction (Dir), from which a set of synthetic parameters is derived. These include, for example: maximum significant wave height, storm duration, peak period and wave direction at the storm peak, total storm energy, duration of the growth and decay phases, and an asymmetry index describing storm shape. A significance analysis is performed to identify the most relevant parameters influencing storm dynamics and structure. The trained machine learning model is then used for two main tasks: (1) reconstruction of storm histories with missing data or low temporal resolution (e.g., upsampling Hs sequences), and (2) generation of new, plausible storm sequences conditioned on key input parameters such as intensity, duration, and energy content. A case study is presented to evaluate the model’s performance for both reconstruction and generation tasks.

Presenting Author: Valentina Laface Mediterranea University of Reggio Calabria

Presenting Author Biography: xxxxx

Authors:

Valentina Laface Mediterranea University of Reggio Calabria
Felice Arena Mediterranea university of Reggio Calabria