A massive tsunami devastated northeast Japan in 2011, killing around 18,500 people. Since then, the country has been focused on avoiding a repeat of the situation.
According to a press release issued by the RIKEN Prediction Science Laboratory on Monday, new study has employed machine learning to accurately anticipate tsunami impacts in less than one second.
“The key benefit of our method is the speed of predictions, which is critical for early warning,” said Iyan Mulia, the project’s lead and a RIKEN scientist.
“Traditional tsunami modelling predicts after 30 minutes, which is too late. However, our model can generate predictions in seconds.”
150 offshore stations
To do this, the coast currently has the world’s greatest network of sensors for monitoring ocean floor movement. This network consists of approximately 150 offshore stations that collaborate to provide tsunami warnings.
However, in order to function properly, the data provided by the sensors must be transformed into tsunami heights and extents along the coastline.
This often involves solving challenging nonlinear equations, which can take a regular computer roughly 30 minutes to complete. This obviously does not give people enough time to evacute.
This is why the RIKEN AI model is so important in saving lives. It gives them at least a half-hour head start from where the tsunami will strike.
The RIKEN team used over 3,000 computer-generated tsunami events to train their machine-learning system, which was then tested with 480 different tsunami scenarios and three genuine tsunamis.
Accurate for any time-sensitive disaster
They discovered that their machine-learning-based model could achieve equal accuracy while requiring only 1% of the computational work of traditional approaches. They now claim that their model is applicable to any time-sensitive natural disaster.
“The sky is the limit: you can apply this method to any type of disaster forecasting where time constraint are very limited,” added Mulia.
“I’m now focusing on predicting storm surges with machine learning.”
In February 2021, RIKEN produced, in partnership with Fujitsu, a sophisticated predictive AI platform that enabled real-time predictions of tsunami flooding. Fugaku, the world’s fastest supercomputer, was employed in the development of the new tsunami prediction programme.
Although the model had to be trained using the enormous computing power of Fugaku, it was built to be loaded onto normal PCs and can make predictions in a matter of seconds.
Researchers developed a novel technique to detect tsunamis in December 2021 that uses the magnetic fields that tsunamis produce as they move across the conductive water of the ocean. These magnetic fields can be noticed a few minutes before the sea level rises, providing some extra time for action that could save lives.
Both innovations are impressive, but they cannot match with RIKEN’s latest innovation. However, the system is currently only accurate for huge tsunamis that are greater than 1.5 metres. Mulia and his colleagues are now striving to improve the accuracy of the model for smaller tsunamis.
