By analyzing seven centuries of data from more than a billion ocean waves, researchers from the University of Copenhagen and the University of Victoria used artificial intelligence to devise a predictive formula for these powerful ocean giants. New knowledge could make transportation safer.
Stories about monster waves, called "wild waves," have been legend among sailors for centuries. But in 1995, when a 26-meter wave hit the Draupner oil platform in Norway, digital instruments were there to capture and measure the North Sea monster. This is the first time Rogue has been measured and provides scientific evidence that unusual waves actually exist.
Since then, these extreme waves have been the subject of many studies. Now, researchers at the Niels Bohr Institute at the University of Copenhagen have used artificial intelligence methods to discover a mathematical model that sheds light on how, and especially when, unusual waves occur.
Using vast amounts of big data on ocean movements, researchers can predict the likelihood of being hit by a giant wave at any given time. "Often only very unfortunate things happen when these giant waves hit. They are caused by a combination of many factors that, until now, have not been combined into a single risk estimate. In this study, we mapped the causal variables that generate abnormal waves and used artificial intelligence to collect them into a model that can calculate the probability of abnormal wave formation," says Dion Häfner.
Häfner is a former Ph.D. student at the Niels Bohr Institute and first author of the scientific study, which has just been published in the prestigious journal Proceedings of the National Academy of Sciences (PNAS).
Wild waves happen every day
In their model, the researchers combined available data on ocean movement and sea conditions with water depth and bathymetric information. On top of that, wave data is collected from buoys at 158 different locations around the U.S. coast and overseas territories, which collect data 24 hours a day. Combined, the data from more than 1 billion waves contain 700 years of wave height and sea state information.
The researchers analyzed multiple types of data to find the cause of abnormal waves, which are defined as waves at least twice as high as surrounding waves, including extreme abnormal waves exceeding 20 meters in height. With machine learning, they turn it all into an algorithm and then apply it to their data set.
"Our analysis shows that unusual waves occur all the time. In fact, we registered 100,000 waves in the dataset that could be defined as wild waves. This corresponds to approximately 1 wild wave occurring per day at any random location in the ocean. However, they are not all wild waves," explains Johannes Gemmrich, second author of the study.
Artificial Intelligence Guest Scientist
In this study, the researchers were helped by artificial intelligence. They used a variety of AI methods, including symbolic regression, which gives an equation as output rather than just returning a single prediction like traditional AI methods.
By examining more than a billion waves, the researchers' algorithm worked its way through the causes of abnormal waves and condensed them into equations that describe why abnormal waves form. AI understands the cause and effect of a problem and communicates that cause and effect to humans in the form of an equation that researchers can analyze and incorporate into future studies.
"For decades, Tycho Brahe collected astronomical observations, and Kepler, through a lot of trial and error, finally arrived at Kepler's laws. Dion used machines to study waves, just as Kepler used machines to study planets. To me, the possibility of such a thing is still astounding," says Markus Jochum.
Phenomena known since the 1700s
The new research also shatters common beliefs about what causes the unusual waves. By far the most common cause of unusual waves is thought to be one wave briefly merging with another and stealing its energy, creating one large wave that keeps going.
However, the researchers found that the most dominant factor in the materialization of these weird waves is what's known as "linear superposition." This phenomenon, known since the 1700s, occurs when two wave systems cross each other over a short period of time and reinforce each other.
"If two wave systems meet at sea, thereby increasing the chance of high crests and deep troughs, there is a risk of huge waves. This knowledge has been around for 300 years, and we now back it up with data," says Dion Häfner.
Make transportation safer
The researchers' algorithm is good news for the shipping industry, which has about 50,000 cargo ships sailing around the world at any given time. In fact, with the help of algorithms, we can predict when this "perfect" combination of factors will occur, increasing the risk of large waves that could be dangerous to anyone at sea.
"Because shipping companies plan their routes in advance, they can use our algorithms to assess whether they are likely to encounter dangerous waves along the way. Based on this, they can choose alternative routes," said Dion Häfner.
The algorithms and research are publicly available, as is the weather and wave data deployed by the researchers. Therefore, interested parties such as public authorities and weather services can easily start calculating the probability of unusual waves, Dion Hafner said. Unlike many other models created using artificial intelligence, all intermediate calculations in the researchers' algorithm are transparent.
"Artificial intelligence and machine learning are often black boxes that do not add to human understanding. But in this study, Dionne used artificial intelligence methods to transform a huge database of wave observations into a new abnormal wave probability equation that is easy for humans to understand and related to the laws of physics." Thesis supervisor and co-author.
Reference: "Machine-guided discovery of real-world rogue wave models" by Dion Häfner, Johannes Gemmrich, and Markus Jochum, November 20, 2023, Proceedings of the National Academy of Sciences.
DOI: 10.1073/pnas.2306275120
Compiled source: ScitechDaily