Increasing maritime safety with improved understanding of rogue waves

Abstract

Rogue waves are ocean surface waves larger than the surrounding sea that can pose a danger to ships and offshore structures. Fatal accidents are not just as a result of the wave size but the unexpected nature of the event. Despite rogue wave prediction being sought for decades, all current prediction methods are not operationally viable as they require complex measurement of the wave field and computationally intensive calculation, which is infeasible in most applications. Consequently there is a need for fast predictors.

Here we collate, quality control, and analyse the largest dataset of single-point field measurements from surface following wave buoys to search for predictors of rogue wave occurrence. We find that analysis of the sea state parameters in bulk yields no clear predictors, except spectral bandwidth parameters which display different probability distributions in rogue seas to normal seas, but these parameters are rarely provided in wave forecasts. When location is accounted for, trends can be identified in the occurrence of rogue waves as a function of the average sea state characteristics at that location. We find that frequency of occurrence of rogue waves and their generating mechanism is not spatially uniform, and each location is likely to have its own unique sensitivities which increase in the coastal seas. Further, we investigate the temporal variability rogue waves along the US western seaboard, to investigate regional trends in significant wave height and individual rogue waves. We find high spatial variability in trends in significant wave height and rogue waves across the region. Rogue wave occurrence displays a mostly decreasing trend, but the relative height – or severity – of the waves is increasing. We also identify seasonal intensification in rogue waves with increased rogue wave occurrence, of higher severity, in the winter than in the summer. Finally we investigate the feasibility of rogue wave prediction using existing technologies by applying our learnings to machine learning algorithms to build a predictive model based on the short-term sea state statistics that are forecast by wave models. We find that the rarity and complexity of the phenomenon leads to an imbalanced and overlapping dataset and consequently poor classification ability by machine learning models. The performance is deemed too low to be of practical use to the mariner at this time.

More information

Identifiers

ORCID for Dominic Hudson: orcid.org/0000-0002-2012-6255

Catalogue record

Export record