AI helps power tsunami early warning system

The team combined state-of-the-art acoustic technology with AI tools to monitor tectonic activity in real time and improve the accuracy of existing tsunami alerts. 

To train the AI model, the researchers relied on sound recordings captured by underwater microphones, called hydrophones. These sounds were used to measure the acoustic radiation produced by 200 earthquakes that happened in the Pacific and Indian Oceans.

Then, the researchers leveraged an AI computational model to triangulate the source of the tectonic event and classify earthquakes' properties, including length and width, uplift speed, and duration, which are used to reveal the size of the tsunami.

"Tsunamis can be highly destructive events causing huge loss of life and devastating coastal areas, resulting in significant social and economic impacts as whole infrastructures are wiped out," said Dr Usama Kadri, a co-author of the research.  

"Our study demonstrates how to obtain fast and reliable information about the size and scale of tsunamis by monitoring acoustic-gravity waves, which travel through the water much faster than tsunami waves, enabling more time for evacuation of locations before landfall."

At the moment, tsunami warning systems rely on waves reaching sea buoys before tsunami warnings are triggered. This method is often inaccurate and leaves little time for evacuations.

Instead, the Cardiff researchers chose to focus on acoustic-gravity waves: naturally occurring sound waves that move through the deep ocean at the speed of sound and can travel thousands of kilometres in the water.

This acoustic radiation also carries information about the originating source of the tectonic event and its pressure field can be recorded at distant locations, even thousands of kilometres away from the source.

"Tectonic events with a strong vertical slip element are more likely to raise or lower the water column compared to horizontal slip elements," said co-author Dr Bernabe Gomez Perez.  "So, knowing the slip type at the early stages of the assessment can reduce false alarms and complement and enhance the reliability of the warning systems through independent cross-validation."

The team’s work predicting tsunami risk is part of a long-running project to enhance natural hazard warning systems across the globe. Their latest development features in user-friendly software which is set to be hosted in national warning centres later this year.

In May 2022, a team of scientists at Los Alamos National Laboratory revealed it was also working on a deep-learning model that would be able to pick up on the gravity waves generated by an earthquake and thus predict the risk of an ensuing tsunami. 

The researchers' findings were published today in the journal Physics of Fluids.