- As a preventive measure against damage from tsunamis, you developed tsunami flooding prediction technology using K computer in 2015. Would you tell us about this technology?

Yusuke: It's a technology for real-time simulation of how far a tsunami will inundate the land before it reaches the coast. We tested with K computer to produce a real-time tsunami flooding simulation and succeeded in formulating a high-resolution simulation in a short period of time. By entering an estimate of an offshore tsunami's status, we were able to predict the flooded areas of the 2011 Great East Japan Earthquake in a matter of two minutes.

- In February 2021, you developed a method of producing high-resolution flooding predictions on a PC using Fugaku and AI, correct?

Yusuke: We had to use the supercomputer to produce detailed tsunami flooding predictions on a real-time basis. Of course, supercomputers aren't ubiquitous, so we used AI technology, which has evolved rapidly in the decade following the earthquake. Based on the simulations made by Fugaku, we generated 20,000 sets of "training data" to teach the AI the relationship between the tsunami waveform observed offshore and the inundated status of the coastal areas.
Production of training data requires a large volume of high-resolution simulations, which is not possible without a supercomputer. After inputting training data into the AI, the AI was educated along a set of parameters. This "educated AI," which is small in size and requires a smaller calculations to make predictions, can work on a computer. This has brought us closer to putting the system to practical use.
Specifically, when we feed the AI with observed data, including sea-level changes detected offshore at the occurrence of a tsunami, it can analyze the inundation extent, the height and the arrival time of the tsunami on land. We are now considering how to put it to practical use in the disaster prevention administrations of municipalities, for example.

Attachments

  • Original Link
  • Original Document
  • Permalink

Disclaimer

Fujitsu Ltd. published this content on 10 December 2021 and is solely responsible for the information contained therein. Distributed by Public, unedited and unaltered, on 10 December 2021 01:31:02 UTC.