New AI technology can predict tsunami impacts in less than a second

The method can be used on any time-sensitive natural disaster.
Loukia Papadopoulos
Illustration of a tsunami hitting a city.jpg
Illustration of a tsunami hitting a city.

kurosuke/iStock 

In 2011, northeast Japan was struck by a devastating tsunami that claimed the lives of about 18,500 people. Since then, the nation has been focused on preventing a similar outcome in the future.

Now, new research out of the RIKEN Prediction Science Laboratory has used machine learning to accurately predict tsunami impacts in less than one second, according to a press release by the institution published on Monday.

"The main advantage of our method is the speed of predictions, which is crucial for early warning," explained Iyan Mulia, the work's lead and a scientist at RIKEN.

 "Conventional tsunami modeling provides predictions after 30 minutes, which is too late. But our model can make predictions within seconds."

150 offshore stations

To achieve this, the coast now boasts the world's largest network of sensors for monitoring the movement of the ocean floor. About 150 offshore stations make up this network and work together in order to provide early warnings of tsunamis. 

To function effectively, however, the data generated by the sensors needs to be converted into tsunami heights and extents along the coastline.

This normally requires solving difficult nonlinear equations, which can take about 30 minutes on a standard computer. Needless to say, this does not give people enough time to evacuate.

New AI technology can predict tsunami impacts in less than a second
Tsunamis are devastating natural disasters.

That's why the RIKEN AI model is so crucial to saving lives. It allows people to get at least half an hour head start from where the tsunami will strike.

The RIKEN team trained their machine-learning system using more than 3,000 computer-generated tsunami events and tested it with 480 other tsunami scenarios and three actual tsunamis. 

Accurate for any time-sensitive disaster

They found that their machine-learning-based model could achieve comparable accuracy at only one percent of the computational effort of conventional approaches. Now, they claim their model could work for any time-sensitive natural disaster. 

"The sky's the limit—you can apply this method to any kind of disaster prediction where the time constraint is very limited," added Mulia. 

Most Popular

"I'm now working on a storm surge prediction, also using machine learning."

Back in February of 2021, RIKEN, in collaboration with Fujitsu, developed a powerful predictive AI tool that enabled real-time predictions of the flooding caused by a tsunami. The hardware used for the development of the new tsunami prediction tool was Fugaku, the world's fastest supercomputer.

Though the model required the immense computational power of Fugaku for training, it was designed to be loaded onto regular PCs where it can carry out predictions in seconds.

In December 2021, researchers conceived of a new method that detects tsunamis by the magnetic fields they generate when moving through the ocean's conductive water. These magnetic fields can be detected a few minutes before rising sea levels, giving a few potentially life-saving extra moments of response time.

Both inventions are impressive, but they simply can't compete with RIKEN's latest development. Currently, however, the method is only accurate for large tsunamis that are higher than about 1.5 meters. Mulia and his team are now working to improve its accuracy for smaller tsunamis.