By David Cyranoski of Nature magazine

On 11 March, the Japan Meteorological Agency (JMA) warned that a 3-metre-plus tsunami would hit northeastern Japan. In fact, the wave that came ashore stood more than 10 metres high -- reaching 50 metres in some places -- and is thought to have claimed the lives of more than 20,000 people.

At a press conference on Monday, a somber Akira Nagai, head of the agency's earthquake and tsunami observations division, acknowledged that the mistaken forecast "led to a slow evacuation" -- a painful admission in a country that prides itself on its earthquake and tsunami preparedness. He then presented a revised early-warning system that recognizes the uncertainty in predicting natural phenomena, as well as the difficulty in translating this knowledge into public information.

The main change is that in response to earthquakes of magnitude eight or above, the agency will no longer forecast a tsunami wave's height, warning instead of "the possibility of a huge tsunami".

Imperfect prediction

The tsunami early-warning system measures seismic signals during the first minute after an earthquake and feeds these into computer simulations. The models estimate the size of tsunami wave that will hit each prefecture, and send predictions -- using eight categories of wave height, ranging from 0.5 metres to more than 10 metres -- two minutes later.

For large earthquakes lasting more than a minute, however, there is a tendency to underestimate the quake, and thus the tsunami. The March earthquake continued for more than two minutes after the JMA computers took readings and began their simulations, during which time the tsunami's energy increased.

The JMA didn't issue a corrected figure, based on Global Positioning System wave gauge readings, until 20 minutes later (see Japan faces up to failure of its earthquake preparations).

The revised system is still quick, issuing a warning within 3 minutes. But it sacrifices precision in an attempt to avoid misunderstanding, cutting the number of wave categories from eight to five. The range of each category is still under discussion.

"The new system will avoid underestimated values, which should help," says Fumihiko Imamura, a tsunami specialist at the Disaster Control Research Center of Tohoku University, and a member of the working group that put out the draft. "Not giving numbers [for the biggest quakes] is a good idea."

The broad-brush approach is "like going back 20 years", to the days before Japan started quantifying tsunami forecasts, says Imamura.

Koji Fujima, who researches tsunami-wave propagation at the National Defence Academy in Yokosuka, favours a different course. "How important is it to get a warning within 3 minutes that a 'huge wave' is coming?" he asks. Without more detail, he says, it is hard to tell whether a wave will breach tsunami defences, and where to go for safety.

Fujima would prefer to take more time, and issue an alert that a tsunami warning will come 3 minutes after seismic measurements are complete. Regions that can't afford to wait that long would be told that the warning will not come in time. "That seems a better for the people," he says, because, unlike this year's warnings, it will not provide false reassurance.

Interim measures

The JMA's proposal will be open for comment for a month, and will probably become policy by the end of the year.

But it is a short-term fix. The agency will start discussions of a more thorough overhaul in September. Imamura hopes that, instead of the present single figure for the height of an incoming wave, the next revision will include plans for detailed inundation maps estimating what levels a tsunami would reach at different locations on shore. The United States has been working on such a system since the Indian Ocean tsunami in 2004.

The ¥1 billion (US$13 million) that such a system would cost is worth paying, says Imamura, pointing out that in March even people who made it safely to some evacuation sites were washed away. Warnings could be issued with precise maps on computers, televisions or mobile phones, and would help to overcome the difficulty people have in interpreting warnings.

"What we do is natural science. But how people read these numbers is a difficult problem that requires human science. We need more research on that," says Imamura.

This article is reproduced with permission from the magazine Nature. The article was first published on August 11, 2011.