New machine learning models improve early tsunami warnings

History has a way of repeating itself. Unlike science, however, based on general principles and testable theories about the natural world, history uses evidence and interpretations to examine past events and human behavior. This portrayal is important when predicting earthquakes and tsunami waves on Canada's west coast. Researchers do not yet have the scientific data needed to make the community safe, so current calculations are informed by historical natural disasters in remote locations like Japan and Indonesia.

To combat this lack of data, new western studies supported traditional statistics with new machine learning techniques using artificial intelligence and tested early warning models for Tofino, a popular surfing and tourist destination on the west coast of Vancouver Island.

Overall, the collected data indicates that we are waiting long to issue an early warning message for the tsunami. This shows better performance in terms of evacuating the most people safely and neatly than shorter response times. However, this approach is problematic as warning messages become less effective with longer wait times and can ultimately be more fatal for evacuees and emergency responders.

“Our model shows that if latency is too short, the performance of early tsunami warning models differs significantly in terms of success,” he says, as a professor of geoscience, Canada's research committee chair and multi-hazard risk assessment research committee chair. “We need to continue collecting data and develop robust tsunami early warning models for Tofino and other coastal communities on Vancouver Island, using multiple data sets.”

Low terrain, high risk

Tofino is located near the Cascadia subduction zone, which increases the risk of tsunamis. Located 100-200km from the Pacific coast of North America, the Cascadia subduction zone is a convergent (structural) plate boundary that can generate earthquakes over 9.0 and tsunamis that can reach heights of 20-30 meters.

Local tsunamis generated by earthquakes in the Cascadia subduction zone can reach chlorine within 15-20 minutes, with little warning time. The community is actively working to prepare for the tsunami, including the potential construction of the tsunami tower.

“We don't have any data to make sure the model and predictions are correct. Since a record-based tsunami hits Tofino, we can inform the model to study historical cases like Tohoku. “But we know that it will happen. That's not the problem, it's when.”

Mitigate fatal disasters

On March 11, 2011, a series of fatal tsunami waves collided with Japan's Tohoku. An earthquake of 9.0 size off the northeast coast of Japan's largest island caused a tsunami, causing nearly 20,000 deaths. The tsunami also caused a major nuclear accident at a coastal power plant.

“Early warnings of tsunamis are always a hot topic as their impact can be dire and deadly at the highest levels,” Goda said. “Thoku is a classic example that was very devastating. And of course, Sumatra had the 2004 tsunami. These are two major disasters that motivated me.”

Born in Japan, Goda received his master's degree in agriculture from Kyoto University before completing his Ph.D. Western civil engineering.

Research published from Coastal Engineering Journalgoda showed random forest models (machine learning algorithms that use decision trees to make predictions), and was the most accurate system when compared to neural networks, human brain-inspired AI, and traditional statistical methods, multiple linear regression models.

However, GODA argues that all tested models need to provide valuable data, utilize and cross-reference.

“Multiclass regression is a long-standing baseline model that we've used for decades to predict tsunamis,” Goda said. “To get better results, we need to start using more AI models, but we are hungry for more data and only improve performance with more data.”

And there is another problem with predicting the Tofino tsunami. Canada's west coast relies on data from four submarine sensors in part on Canada's Ocean Networks (owned and operated by the University of Victoria) from Vancouver Island, where more than 150 Japanese coastlines are monitored in the Tohoku region alone.

“Japan has deployed over 150 sensors. It's a very expensive system and other countries can't afford that many people,” Goda said. “Ocean Networks Canada monitors four sensors 24 hours a day, deployed in a highly strategic location, including one off the coast of Vancouver Island.

“With a nearby station, you should expect to get much better data. That was another motivation for the study to show how good it would be to warn you of an imminent tsunami with more data.”

The perfect storm

Tofino and its unique geographical location are renowned for its mild climate, including year-round surfing opportunities. With over 35km of sandy beaches, it attracts surfers of all levels and makes it a paradise for beginners and experienced surfers.

Unfortunately, its exposed waterfront and low terrain make Tofino an ideal location for potential tsunami waves. In fact, during the tsunami, the majority of the $2 billion worth of economic assets in tourist hotspots are at risk.

As a leading community researcher, Goda studies risky coastal communities, including Tofino, Havana, Cuba, Bali and Bali, as well as the leading researcher of infrastructure resilience for the Climate Geographical Long-Term Effects (CIRCLE) project. The interdisciplinary and international research initiatives are conducting multi-hazard impact assessments of physically interconnected infrastructures to better identify and protect vulnerable people and communities along the global coastline.

“The people of Tofino face many challenges when prompt evacuation is needed due to the possibility of a single road and road flood linking the town to Victoria,” Goda said.

Tofino is actively involved in preparing for tsunamis, including annual evacuation drills and early tsunami warning tests. Plans are also underway to build a vertical shelter in the town.

In another study, Goda and his collaborators outlined how earthquake mapping and tsunami risks for shaking are needed for coastal communities like Tofino, facing the imminent danger of offshore earthquakes. Large-scale seismic hazard assessments like Goda and one Goda or Partners created by partners identify relative differences in regional and regional shaking and require local and local seismic hazard mapping to which they are mapped. This process is known as microzoning.

Goda and post-doctor scholar Nova Ruthmawati will visit Tofino for the town's upcoming tsunami warning test in October, leading the workshop and sharing the latest findings with community members. The team also has a workshop scheduled for September 11th in Bali.