Revolutionizing tsunami predictions: How an engineer's dose of AI could save lives

Now, two researchers from the University of California, Los Angeles, and Cardiff University in the U.K. have created a new warning system that utilizes advanced acoustic technology and artificial intelligence to quickly identify earthquakes and evaluate the possibility of an ensuing tsunami.

Interesting Engineering (IE) connected with the corresponding author of their paper on the work, Dr. Usama Kadri, to explore the potential implications of this advancement in predicting tsunamis globally.

AI predicts tsunamis by analyzing earthquake-produced sound signals 

"The published work on AI concerns one of the models we developed for early tsunami warnings," Kadri told IE.

"In particular, the model analyzes pressure recordings and provides the mode of strike (horizontal or vertical) and magnitude of the effective uplifting water layer."

To understand what the strike of an earthquake is, think of an earthquake as a crack that opens up in the ground. The strike of an earthquake is simply the direction in which that crack runs across the Earth's surface.

It's like drawing a line on a map to show where the earthquake started and which direction it moved in. Significantly, it's essential information that helps scientists study earthquakes.

"Tsunamis are associated with the vertical motion of the water layer. As such, the model can predict if a tsunami will occur after analyzing the sound signals that propagate from the source (i.e., earthquake)," he explained. 

Kadri discussed how he has been developing the technology for the early warning system since 2010. In addition to the AI model, the technology comprises independent analytical models.

He revealed that he had only recently hired a software engineer to assemble the different models into a single operational software. "We call [this] Global Real-time Assessment of Tsunami, or in short, GREAT," he said.

"I presented the software in the last steering committee meeting of UNESCO's ICG/NEAMTWS to help improve the current warning system."

The ICG/NEAMTWS is the Intergovernmental Coordination Group in the Northeastern Atlantic, the Mediterranean and Connected Seas Tsunami Early Warning and Mitigation System.

It is a cooperative effort among countries bordering the Atlantic Ocean and the Mediterranean Sea to establish a regional early warning system for tsunamis in response to the devastating Indian Ocean tsunami in 2004.

'At the speed of sound:' 3D signals race ahead of an ensuing tsunami 

"There is an AI model and an Analytical model," Kadri explained. "All models rely on analyzing pressure signals recorded on hydrophones. The signals travel at the speed of sound, which is much faster than the tsunami providing information on the source."

He said the AI model determines the earthquake's magnitude and analyzes whether the strike is horizontal (indicating no likely tsunami) or vertical (a possible tsunami). 

Independently, the analytical model calculates the properties of the vertical motion of the water layer using two independent models:

One is an inverse model that calculates the probability of the properties (geometry and dynamics). Whereas another model extracts the geometry and dynamics directly (non-probabilistic). 

"When the three models converge, the confidence in the analysis is high. Moreover, GREAT uses independent data sources once available, such as DART-buoys or tide gauges, which can be compared with the hydrophone analysis," Dr. Kadri stated. 

"Currently, I have access to recorded data on the Comprehensive Nuclear-Test-Ban Treaty Organisation's (CTBTO) hydrophones that are distributed in the Indian Ocean, the Atlantic, and the Pacific," he added. 

According to Kadri, the CTBTO's assistance and data have been crucial in enabling the testing of the AI models using actual earthquake and tsunami data. 

He also mentioned that only around two dozen more hydrophones need to be installed globally to create a comprehensive warning system. In other words, adding a relatively small number of underwater listening devices can significantly enhance the ability to detect and alert people to potential tsunamis worldwide.

"The specific locations depend on the region and the nature of tectonic activity there," he said. 

Back in 2017, a breakthrough solved real-time computations

"Personally, I have been studying low-frequency sound waves under the effects of gravity (we refer to these as acoustic-gravity waves) since 2010," Kadri revealed. "The early warning for tsunamis was only one of many applications."

Kadri noted that, in 2016, he met Dr. Bernardo Aliaga, a program specialist (at the time) at the Tsunami Unit of the Intergovernmental Oceanographic Commission of UNESCO, who seemed to see some potential in his team's approach and has been very supportive ever since. 

"In 2017, we had a breakthrough publication in terms of an early warning, where I, together with Prof. CC Mei (MIT), solved the three-dimensional propagation of sound signals from a slender fault analytically, which allowed real-time computations," he said. 

"In the meantime, other pieces of technology emerged, including the AI model that Dr. Bernabe Gomez- my Ph.D. student then (and the first author of this latest work)- worked tirelessly to develop and test."

Warning centers can integrate the new AI model, reducing false alarms 

He reasoned that tsunami warning centers that integrate this new AI approach into their assessment would benefit from a reduction in false alarms and thus increase the reliability of their regional alarm. This would also reduce financial costs associated with business shutdowns and evacuation of coasts. 

"It is sufficient that the software is deployed in a single warning center to provide a global assessment of tsunamis. Though, integrating efforts with regional warning centers is key to developing an efficient global warning for all," Kadri emphasized. 

Despite their progress, he also emphasized that the team is just scratching the surface of what needs to be accomplished, and there is much more work to be done. 

"For example, the current AI version of GREAT relies on 200 past earthquakes; only a couple were above magnitude 9. As such, we treat the larger magnitudes more conservatively," he explained. 

"In the meantime, we're working to include tens of thousands of available historical data to improve the models."

Kadri pointed out another example of the AI model's versatility, stating that it can identify and assess tsunamis from multiple sources, not just earthquakes.

However, he stressed that more testing is necessary to ensure the model's accuracy and effectiveness. While he didn't delve into the specifics of other potential tsunami triggers, we have an idea of what these could be.

For instance, when a volcanic eruption or landslide occurs underwater, it can cause a massive displacement of water that generates a tsunami. Similarly, a meteorite impact in the ocean can create a large wave. These events may not occur as frequently as earthquakes, but they still have the potential to cause devastating tsunamis.

"Once all available historical data is employed to test the software, we can then have more confidence in the accuracy and limitations," concluded Kadri.