When Typhoon Hagibis swept across Japan in 2019, bringing catastrophic flooding, the nation's early warning system missed crucial peak flows in certain regions—a gap that cost time and lives. Now, researchers at the Institute of Industrial Science at the University of Tokyo have developed a real-time data assimilation system that fixes exactly that problem, and the results are striking: it outperforms Japan's current early warning system across the vast majority of the country's major waterways.
The breakthrough hinges on a deceptively simple idea. Rather than relying solely on computer models to predict floods, the new system feeds hourly water-level observations from 1,800 in situ gauges across Japan's Ministry of Land, Infrastructure, Transport and Tourism monitoring network directly into the forecasting model. This constant stream of real-world data corrects the model's understanding of current conditions, allowing it to produce far more accurate predictions of what comes next.
The numbers tell the story. The data assimilation system delivers improved forecasts for 80% of Japan's major river reaches—a transformative leap over the existing system. Perhaps more importantly, it enables reliable flood prediction across more than 60% of the country's observation stations, dramatically expanding the geographic reach of trustworthy early warnings.
When Yingying Liu and her team at UTokyo-IIS tested the system against three major flood events—Typhoon Hagibis in 2019, the Northern Japan Flood in 2022, and the Akita flash flood in 2024—the results surprised even the researchers. In cases where the old early warning system had historically fallen short, the new approach captured peak flows it had previously missed. "We kept the model exactly the same and just used gauge observations to correct the model's initial state," Liu explained. "This action alone was sufficient to start capturing peak flows in flash flood cases that the previous early warning system had missed, such as in the Akita 2024 flood events."
What makes this breakthrough significant extends beyond Japan's borders. Data assimilation—the systematic integration of real-world measurements into computer models—is a powerful technique used in many fields, but its application to flood forecasting has historically been limited by data availability. Japan's dense network of monitoring stations provided the ideal testing ground. Now that the approach has proven effective, it opens a scalable pathway for flood-prone regions worldwide to improve their own early warning systems.
The practical impact is immediate and life-saving. By providing more accurate forecasts one day in advance, emergency managers and communities gain precious time to evacuate, secure infrastructure, and mobilize resources before floodwaters arrive. In a country where typhoons and heavy rainfall regularly threaten populated areas, those extra hours can mean the difference between disaster managed and disaster multiplied.
Professor Kei Yoshimura, who led the research team, envisions rapid adoption. "With more reliable one-day-ahead flood forecasts, emergency managers and communities have more time to act before the water arrives," he said. The team anticipates that the data assimilation framework will be integrated into Japan's operational early warning systems and applied across flood-prone regions globally, making communities worldwide more resilient in the face of extreme weather.