Fujitsu Limited and Yokohama National University have achieved a groundbreaking milestone by developing the world’s first real-time prediction system for multiple tornadoes associated with typhoons. Utilizing Fujitsu’s Fugaku supercomputer and an enhanced version of the Cloud Resolving Storm Simulator (CReSS), this innovation significantly improves disaster preparedness by delivering more accurate and faster tornado forecasts.
This advanced system reduces prediction times from over 11 hours to just 80 minutes, allowing forecasts up to four hours ahead of tornado occurrences. Notably, the prediction model only used 5% of Fugaku’s computational capacity, demonstrating the potential for larger-scale simulations and even quicker predictions.
Addressing Tornado Prediction Challenges
About 20% of tornadoes in Japan are linked to typhoons, making accurate predictions crucial. Tornado warnings were first introduced in Japan in 2008 due to increasing tornado damage. However, current warnings are limited to about one hour, with demand growing for longer lead times. Tornadoes are difficult to predict due to their small size and short duration compared to other weather phenomena.
In November 2022, Fujitsu and Yokohama National University launched a collaborative project to enhance typhoon and tornado prediction. This research focuses on understanding typhoon formation and improving typhoon-related tornado forecasting accuracy, conducted under the Fujitsu Small Research Lab initiative.
Technology Behind the Breakthrough
The CReSS simulator was originally designed for high-resolution weather predictions spanning cloud scales (50m to 2,000m) to mesoscales (2km to 2,000km). While effective at simulating thunderstorms and tornado formation, it was not suitable for large-scale parallel processing on thousands of nodes, leading to impractical computation times.
To overcome this, Fujitsu developed a lightweight version of CReSS that maintained high prediction accuracy while significantly reducing computational demands. Optimized parallel processing on Fugaku allowed for faster simulations, achieved by streamlining computation and data output.
Successful Typhoon No. 10 Prediction
The enhanced CReSS simulator was tested on 8,192 Fugaku nodes during Typhoon No. 10, which struck Japan’s Kyushu region in August 2024. The simulation used 3D spatial data temperature, pressure, humidity, wind speed, and direction to reproduce multiple tornadoes that hit Kyushu’s eastern coast. This simulation accurately visualized tornado formations across a 20km by 20km area, demonstrating the simulator’s predictive power.
Seishi Okamoto, EVP and Head of Fujitsu Research, expressed pride in the collaboration’s achievements. “This success will further propel the integration of our expertise and technologies. By advancing research in typhoon prediction, we aim to mitigate weather-related disasters and reduce their impact, contributing to global environmental solutions,” he said.
Professor Hironori Fudeyasu of Yokohama National University highlighted the importance of the breakthrough in improving early warning systems. “The high-resolution modeling and Fugaku’s supercomputing capabilities have enabled real-time predictions crucial for mitigating the impact of these devastating events,” Fudeyasu stated.
Fujitsu and Yokohama National University plan to release the enhanced CReSS simulator to the public research community by the end of the fiscal year. The organizations will also explore integrating AI technology to further improve prediction speed and accuracy. These efforts align with Fujitsu’s broader mission to address global environmental challenges and reduce disaster risks.