It can be done in fractions of a second, buying time for people to take appropriate action. The technology uses machine learning to save lives.
About 18,500 people were killed in the disaster that hit northeast Japan in 2011. Predicting how high the water would reach at different points along the coastline would have saved many lives.
The coast is home to the largest network of ocean floor sensors. There are 150 offshore stations that give early warning of the waves. The data generated by the sensors needs to be converted into heights and extents for the coastline.
It takes about 30 minutes on a standard computer for this to be solved. A few minutes after the earthquake, the coast was hit by a wave.
Machine learning has been used to cut the calculation time in half.
The speed of predictions is important for early warning. Predicting the waves after 30 minutes is not possible. Predicting can be done within seconds.
The machine-learning system was trained using more than 3000 computer-generated events. They tested it with hundreds of other scenarios and three actual waves. The machine-learning model could only achieve 1% accuracy.
Disaster scenarios where time is of the essence could use the same deep- learning approach. "You can apply this method to any kind of disaster predictions where the time constraint is very limited," says Mulia, who is from Indonesia. I'm working on a storm surge prediction.
Nature Communications publishes the work.
The method only works for large waves that are over 1.5 meters, so Mulia and his team are trying to improve it for smaller waves.
Iyan E. Mulia and her team used machine learning to derive a prediction of the wave's path.
Journal information: Nature Communications
All Rights Reserved © 2024
