New study proposes using AI to improve tsunami warnings in tourist areas

Tsunami warning sirens on Vancouver Island have never sounded for real, yet scientists know the lull is temporary. A new study says artificial intelligence could sharpen the split‑second calls that decide whether visitors and locals can reach higher ground.

The work zeroes in on Tofino, a surfing town whose beaches and hotels fill with travelers year‑round. Researchers tested different warning strategies and found that machine learning might save more people than the rules emergency crews use today.

Earth‑science professor Katsuichiro Goda at Western University, the Canada Research Chair in Multi‑Hazard Risk Assessment, led the project.

He and his team compared traditional statistical models with newer random forest algorithms and neural network tools, checking how long officials can wait before pressing the alert button.

High stakes for tofino

Tofino sits a short drive from the Cascadia Subduction Zone, the 600‑mile fault where the Juan de Fuca plate dives under North America.

A magnitude‑9 rupture on that boundary could send a 65‑foot wall of water ashore in roughly 20 minutes.

Japan’s S‑net network packs 150 ocean‑bottom sensors along deep‑sea cables, yet Vancouver Island relies on only four nodes for real‑time data.

That imbalance matters because Tofino’s hotels, marinas, and boardwalk shops represent more than 2 billion Canadian dollars in at‑risk assets during a severe inundation. 

Why minutes matter

Evacuation drills show that even fit hikers need up to 17 minutes to climb from the waterfront to designated hills. Alerts that come too late jam the single highway out of town, trapping beachgoers behind stalled traffic.

Canada lacks a local catalog of historic tsunamis, so the study fed the computer thousands of simulated waveforms instead of real past events.

The team discovered that performance swings wildly when the training set omits certain rupture styles, an uncertainty coastal planners rarely see in public briefings.

Adding synthetic events from Bayesian “digital twin” models could close that gap, a method now being tested for the Cascadia coast. 

Random forests vs human judgment

“When waiting times are too short, performances of the tsunami early warning models vary significantly in terms of success,” said Goda.

His favorite tool, the random forest, stacks dozens of mini decision trees, each weighing different seismic variables before voting on the safest moment to send a text alert.

In test runs the algorithm beat multiple‑linear regression by about 15 percent on successful evacuations, echoing results from international trials that link seafloor gauges with machine‑learning forecasts. 

Tsunami warning sirens

Japan’s 150‑station array beams pressure and seismic data along 3,600 miles of fiber‑optic cable, giving residents clearer information on wave height and arrival time.

Vancouver Island’s four‑sensor network, run by Ocean Networks Canada, limits early detection accuracy near tourist beaches like Cox Bay.

Goda argues that even a modest expansion – one extra sensor for each 25 miles of coastline, would feed the hungry algorithms and shorten the interval between quake and alert.

Tourist economies in the crosshairs

Peak season pushes Tofino’s population from 2,500 to nearly 20,000, magnifying the challenge of moving people uphill before a wave arrives.

Business owners note that a single false alarm can cost hundreds of thousands of dollars in canceled bookings, so accuracy and trust go hand in hand.

Locals say the success of any alert system depends on whether people trust it. That trust hinges on understanding how and why evacuation orders are made.

Workshops and school programs that explain tsunami dynamics and how warnings work have helped raise awareness in Tofino.

These efforts build a culture of readiness, making it more likely that people will act quickly when seconds count.

Toward smarter evacuations

“We need to start using more AI models to get better results, but they are more data‑hungry and only perform better with more data,” explained Goda.

The Western study suggests a counter‑intuitive tactic: wait a handful of extra seconds before issuing an alert so that messages include firmer wave‑height estimates.

Past disasters show why precision matters: surveys after the 2011 Tohoku tsunami found that repeated false alarms made some residents less likely to flee small waves later. 

Next fall Goda’s group will join residents during Tofino’s annual drill to test an AI‑assisted alert generator in real time. If the prototype works, the same code could guard resorts from Bali to Havana, shrinking the casualty gap between well‑monitored and overlooked coastlines.

The study is published in Coastal Engineering Journal.

—–

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. 

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

—–