by Until now, giant waves, also known as rogue waves, have been difficult to predict. Now researchers have used artificial intelligence to solve the mystery of giant waves that appear out of nowhere. Based on data from more than a billion ocean waves, the AI system determined potentially triggering factors. From this, a prediction formula for these rogue waves was finally created in comparison with existing physical models. This can now help shipping better predict risk.
They come out of nowhere: Giant waves, also known as freak waves or rogue waves, are individual waves that suddenly reach more than twice the height of all the waves before and after them on the high seas. In 1995, for example, a giant wave 25 meters high hit the Draupner oil platform in the North Sea, and many ships have already fallen victim to these “killer waves”. But how do they happen? And how can you predict them?
So far, these questions can only be partially answered. Although scientists can reconstruct giant waves in the laboratory and in computer models, some contributing factors are also known. However, it is not yet possible to predict rogue waves.
Data from a billion waves
Now artificial intelligence can provide a solution: researchers led by Dion Häfner of the Niels Bohr Institute in Copenhagen have investigated whether coupled neural networks can identify the causal factors of giant waves and derive a prediction formula from them. To do this, they fed the AI systems with data from measuring buoys from different oceans that recorded parameters such as wave speed and frequency, wave height, wave steepness and direction. In addition, the depth of the water, the topography of the seabed and other factors were also taken into account.
In total, the dataset included data from more than a billion waves – both normal and giant waves. “Our analysis demonstrates that abnormal waves occur all the time. “We detected 100,000 waves in our dataset that can be defined as rogue waves,” reports co-author Johannes Gemmrich from the University of Victoria in Canada. “This means that every day there is about one monster wave somewhere in the ocean – although not all of them are of extreme height.”
What factors are crucial for a tidal wave?
For their study, the researchers first had several AI systems analyze the entire dataset or just individual sections of it – for example, just summer data or just winter data or just shallow or deep water waves – for possible causal factors. They then checked which factors the neural networks agreed on, regardless of the dataset used. After further testing and training sessions, Häfner and his team selected an AI model that they continued working with.
Then another important step followed: the researchers used the symbolic regression method common in computer science and mathematics to generate an equation from the results of the AI system. Artificial intelligence evaluates the extent to which the various formulas and models provided to it reflect causal relationships and uses them to create a suitable equation. “The result is a new formula for the probability of rogue waves that is based on the laws of physics and that is understandable to people,” explains Häfner’s colleague Markus Jochum.
Overhead, slope and summit height
The result confirms some of the previous assumptions and findings, but also provides new insights. Giant waves can arise when normal waves meet at a certain angle and overlap. Whether it becomes a monster wave also depends on the steepness and height difference between the wave’s trough and crest, as the team explains. “If two wave systems meet in a way that creates high wave crests followed by deep troughs, then the risk of giant waves increases,” explains Häfner.
It also became evident that giant waves in shallow water and deep water in the open ocean follow slightly different laws. On the one hand, they are influenced to varying degrees by the topography of the seabed and, on the other hand, the shape of the wave that favors them also differs: “In deeper waters, the risk of turbulent waves increases with the steepness of the waves. the wave”, the researchers report. “In shallow water it’s exactly the opposite: here we find a clearly negative link with wave steepness.”
Better-than-usual forecasting models
But how good is the AI-generated formula at predicting giant waves? Häfner and his colleagues tested this based on measurement data that had not previously been included in the training datasets. The results of several previously common models were compared with those of your neural network, but also with the result of the prediction formula generated with its help.
The result: “Our models predict tidal wave frequency better across all different test cases than current methods,” write Häfner and his colleagues. The prediction using the neural network was slightly more accurate than the formula derived from symbolic regression. However, both agreed better with real giant wave events than with all other formulas and models, the team reports.
Shipping help
According to the researchers, their model opens up new opportunities to better predict the risk of giant waves in the future than before. For example, it could be used to identify potentially risky weather and sea conditions in certain ocean areas. “Shipping companies often plan their routes in advance. “You could therefore use our algorithm to determine whether there is a risk of rogue waves on this route,” says Häfner. “If that is the case, you can modify your route accordingly.”
The team has made its data and algorithm publicly available for use and testing by anyone interested. At the same time, they are working on refining the model, which should make particularly powerful giant waves more predictable. (Annals of the National Academy of Sciences, 2023; doi: 10.1073/pnas.2306275120)
Source: University of Copenhagen – Faculty of Science
November 21, 2023 – Nadja Podbregar
