Offshore islands make tsunamis stronger, not weaker

Enhanced computer models confirm what earlier research has suggested: offshore islands don't buffer mainlands from tsunami damage. In fact, they almost always make flooding much worse. (Credit: Wayan Vota/Flickr)

Instead of protecting the mainland from a tsunami, a new study shows that offshore islands do the exact opposite.

Rather than buffer, barrier islands focus the energy of the tsunami and can increase flooding on the mainland by up to 70 percent.


New computer modeling of tsunamis striking a wide variety of different offshore island geometries show no situation in which the mainland behind them fared better.

“This is where many fishing villages are located, behind offshore islands, in the belief that they will be protected from wind waves,” says Costas Synolakis of the University of Southern California Viterbi School of Engineering.

“Even Southern California residents believe that the Channel Islands and Catalina will protect them.”

Published in the Proceedings of the Royal Society A, the research was inspired by a field survey of the impact of the 2010 tsunami on the Mentawai Islands off of Sumatra.

The survey data showed that villages located in the shadow of small offshore islets suffered some of the strongest tsunami impacts, worse than villages located along open coasts.

Subsequent computer modeling by Jose Borrero, adjunct assistant research professor at the USC Viterbi Tsunami Research Center, showed that the offshore islands had actually contributed to—not diminished—the tsunami’s impact.

Machine learning

The researchers then set out to determine whether that was a one-of-a-kind situation, or the norm.

The team designed a computer model that took into consideration various island slopes, beach slopes, water depths, distance between the island and the beach, and wavelength of the incoming tsunami.

“Even a casual analysis of these factors would have required hundreds of thousands of computations, each of which could take up to half a day,” Synolakis says. “So instead, we used machine learning.”

Machine learning is a mathematical process that makes it easier to identify the maximum values of interdependent processes with multiple parameters by allowing the computer to “learn” from previous results.

200 models, not thousands

The computer starts to understand how various tweaks to the parameters affect the overall outcome and finds the best answer quicker.
As such, results that traditionally could have taken hundreds of thousands of models to uncover were found with 200 models.

“This work is applicable to some of our tsunami study sites in New Zealand,” says Borrero, who is producing tsunami hazard maps for regions of the New Zealand coast. “The northeast coast of New Zealand has many small islands offshore, similar to those in Indonesia, and our modeling suggests that this results in areas of enhanced tsunami heights.”

“Substantial public education efforts are needed to help better explain to coastal residents tsunami hazards, and whenever they need to be extra cautious and responsive with evacuations during actual emergencies,” Synolakis says.

Researchers from Ecoles Normales de Cachan in Paris and University College Dublin contributed to the research. EDSP of ENS-Cachan, the French Embassy in Dublin; the ERC, SFI, and University College Dublin provided funding.

Source: USC