Fujitsu to Use Supercomputer to Provide Tsunami Tracking
09-03-2021 | By Robin Mitchell
Recently, Fujitsu announced that it is working with researchers to develop a tsunami prediction warning system. Why is predicting the effects of tsunamis difficult, what system will Fujitsu be using, and how can it help save lives in the future?
Why Tsunamis are Near-Impossible to Predict
Tsunamis are well known for their incredibly destructive power, and when one is spotted, people usually have just minutes to respond. Out at sea, a tsunami looks like nothing more than a wave of a few feet high at the most, but as it approaches land, the shallow water results in a mighty and high wave that can easily scale coastal defence systems and flood major areas inland.
Weather events such as hurricanes, tornadoes, and storms have a degree of predictability as they can be seen forming (i.e. areas of high and low-pressure meeting with vortexes forming). However, a tsunami cannot be predicted as the cause of tsunamis is an underwater earthquake, and earthquakes themselves cannot be predicted.
To make matters more complex, even if an earthquake can be determined to happen within a given time frame (of years), the resulting earthquake's size cannot be determined, and the effects of a tsunami are always directly related to the size of the earthquake.
Fujitsu to Develop Tsunami Damage Predicting System
Even though tsunamis cannot be predicted, their damage can be. Using past data, researchers can look at how far tsunamis travel inland for given earthquake magnitudes and the distance to the epicentre.
But modelling even scenario for every earthquake energy is no small feat as there are millions of different parameters that can change the tsunami's outcome. Such parameters include the volume of water, how that water interacts with walls and dams, how buildings and low-lying areas affect the inrush of water, and if the earthquake has shifted the height of that land.
To help solve this, Fujitsu has teamed up with researchers worldwide to develop the worlds most advanced prediction system that will help emergency responses cope with tsunamis better. Instead of figuring out when a tsunami will happen, the system will predict which areas are most vulnerable. Thus when earthquakes are detected, these areas can be specifically evacuated.
The system being developed by Fujitsu is utilising Fugaku, the worlds fastest supercomputer. The Fugaku supercomputer is a $1 billion system with a speed of 442 PFLOPs, total data storage of 150 PB, over 150,000 nodes, and approximately 4 times as fast as the second-fastest supercomputer, the Summit.
Fujitsu will utilise AI to create a prediction model, and the system will be trained using over 20,000 tsunami scenarios developed by the supercomputer itself. The system's ingenious design is that the deep learning cycle is performed on the supercomputer, but once trained, can be executed on a desktop PC to provide a real-time prediction on the effects of a tsunami.
How AI Evacuation Will Save Lives
As stated previously, the new system is not trying to predict when a tsunami will happen, but instead, try to predict which areas are at risk. If proven to be accurate, the system can save thousands of lives from major tsunamis that can strike without warning.
However, the use of simulation generated models for learning raises a cause for concern. If the simulations themselves are not accurate, then the resulting AI will be fundamentally flawed and could result in wrong advice.
The 2011 Tōhoku tsunami is an example of how getting data wrong can be devastating. Instead of using AI models, the tsunami was detected by sea buoys which provided this data to early warning systems. However, these systems determined that the wave's maximum height would be 3 meters, but the resulting waves reached 50 meters in some areas; the result was the loss of over 20,000 lives.
Of course, researchers developing the new AI-based early warning system are most likely creating 3D models of Japan and considering elevation and obstacles such as sea walls. However, understanding how earthquakes, tsunamis, and wave height relate to each-other may be best left to real-life data as AI models can quickly amplify biases in datasets.
Read More
