In April 2022, MIT engineers discovered a way to predict how waves break using machine learning and data from wave tank tests.

Wave breaking is the primary method for dissipating energy inputted into ocean waves by wind and transported across the spectrum by nonlinearity. It determines a sea state's attributes and is essential in ocean-atmosphere interactions, ocean pollution, and rogue waves. However, due to its turbulent character, wave breaking is too computationally demanding to solve via direct numerical simulations, especially in simple, short-duration situations. 

Image source: MIT News

To address this issue, the researchers propose a hybrid machine learning framework that combines a physics-based nonlinear evolution model for deep-water, non-breaking waves with a recurrent neural network to forecast the evolution of breaking waves.

Blended machine learning framework

MIT engineers have developed a novel method for modelling wave breakup. Using data from wave-tank tests and machine learning, the researchers modified previous equations to predict wave behaviour. Typically, engineers rely on such equations to help them construct offshore platforms and durable structures. However, until recently, equations were incapable of capturing the intricacy of breaking waves.

Researchers found that the updated model made more accurate predictions about how and when waves break. For example, the model predicted a wave's steepness before it broke and its energy and frequency after it broke than the usual wave equations.

Furthermore, offshore structures can be better if we know how these waves interact precisely. It can also help us figure out how the ocean and atmosphere work together. Scientists can figure out, for example, how much carbon dioxide and other gasses from the air the sea can absorb if they have a better idea of how waves break.

What is a learning tank?

Scientists usually do one of two things to try to figure out how a breaking wave will behave: 

• they either try to simulate the wave at the level of individual water and air molecules, or 
• They run experiments to figure out what waves are like by measuring them. 

The first method costs a lot to run on a computer and is hard to simulate even over a small area. The second method takes a long time to run enough experiments to get statistically significant results.

Instead, the MIT team took parts from both approaches and used machine learning to make a more accurate and efficient model. The researchers started with a standard set of equations that describe how waves behave. Then, they tried to improve the model by "training" it with data from real experiments that showed how waves broke.

Researchers learned how waves break by running tests in a 40-meter-long tank. At one end of the tank, there was a paddle that the team used to start each wave. First, the team put the paddle in place so that it would make a wave break in the middle of the tank. Then, as waves moved down the tank, the height of the water was measured by gauges.

How to build a safe harbour?

In total, the team did about 250 experiments. They used the data from these experiments to train a neural network, a machine-learning algorithm. In particular, the algorithm compares the real waves in experiments with the waves predicted by the simple model. If there are any differences, the algorithm adjusts the model to match reality.

After training the algorithm with their experiment data, the team took measurements from two separate experiments, each in a different wave tank with different dimensions. These tests found that the updated model made more accurate predictions than the simple model. For example, it was better at predicting how steep a breaking wave would be. The new model also accounted for an essential property of breaking waves called the "downshift."  

Conclusion

Low-frequency ocean waves move faster than high-frequency ones. Because of this, the wave will progress more quickly after the downshift. The new model can predict how the frequency of waves changes before and after they break. This approach could be beneficial for preparing for storms along the coast. Moreover, the team's updated wave model is in the form of an open-source code that other people could use, for example, in climate simulations of the ocean's ability to take in carbon dioxide and other gasses from the air. We can also use the code to test offshore platforms and coastal structures using simulations.