A significant wave height data-driven modeling for digital twins of marine environment

Abstract

This study explores the development and imple-mentation of a predictive modeling framework specifically for forecasting Significant Wave Height (VHMO) in marine environments, leveraging the capabilities of digital twins (DTs). The primary variable, VHMO, represents the average height of the highest one-third of waves observed and is a crucial indicator for maritime operations and safety. Utilizing a Gated Recurrent Unit (GRU) neural network, the model was trained and evaluated using three distinct Wave Height datasets from in situ sensors on Tarragona, Barcelona, and the EMSO-OBSEA observatory, at the western Mediterranean. The methodology involves rigorous data preprocessing, including normalization and sequence cre-ation, followed by the training and validation of the model to forecast Significant Wave Height. Outliers are detected through residual analysis, establishing a threshold based on the statistical distribution of residuals. The model demonstrates high accuracy and robustness to data gaps, with Pearson correlation coefficients of 0.93, 0.95, and 0.88 for the Tarragona, Barcelona, and OBSEA datasets, respectively. The findings highlight the model's efficacy in predictive accuracy, contributing to enhanced monitoring and decision-making in maritime operations.

Postprint (published version)