A system capable of precisely forecasting a wide range of weather conditions, including extreme precipitation, has been developed in South Korea. This technology is expected to contribute to advancing real-time disaster response and weather warning systems.


On March 17, the National Research Foundation of Korea announced that Professor Youngjun Hong and his team from the Department of Mathematical Sciences at Seoul National University have developed an artificial intelligence (AI) model for “nowcasting precipitation” by combining a local spatiotemporal attention mechanism with a novel upsampling structure.


Precipitation forecast flowchart using the 'ultra-short-term precipitation prediction' artificial intelligence (AI) model. Korea Research Foundation

Precipitation forecast flowchart using the 'ultra-short-term precipitation prediction' artificial intelligence (AI) model. Korea Research Foundation

View original image

Typically, global precipitation forecasting models face limitations in real-time response due to the immense computational requirements. Moreover, South Korea's unique topographical characteristics have posed challenges in accurately predicting localized precipitation patterns.


To address these issues, the research team designed the AI architecture through an applied mathematical approach. By introducing a “local attention mechanism” that intensively allocates computational resources to rapidly changing precipitation patterns in radar imagery, the model accurately reconstructs fine-grained information that triggers short-term heavy rainfall within 1 to 6 hours, while simultaneously minimizing unnecessary calculations. This maximized efficiency proved to be a key success factor.


The effectiveness of the developed model was validated using actual observational data from meteorological agencies in the United States, France, and Korea, with clear results. Compared to recent global models, it demonstrated higher accuracy. In particular, for extreme precipitation events in Korea, the computational efficiency improved more than twentyfold, resulting in a marked enhancement in prediction accuracy.


Professor Youngjun Hong, National Research Foundation of Korea

Professor Youngjun Hong, National Research Foundation of Korea

View original image

When the model was applied to the heavy rainfall events of 2023, it successfully detected high-risk situations in advance. This outcome is seen as evidence that the Korean-style nowcasting precipitation model can predict extreme precipitation events ahead of time, enabling early warnings and proactive disaster response.


Professor Hong stated, “This research is significant in that it overcame the ‘black box’ limitations of AI based on mathematical rigor and realized a model that operates stably even under extreme weather conditions. I hope the technology developed by our team will be integrated into actual disaster prevention systems and play an important role in safeguarding the public.”



Meanwhile, the results of this research (paper) have been accepted for presentation at ‘ICLR 2026,’ an international conference in the field of AI, and will be announced next month. ICLR is one of the world’s top three academic conferences in AI, boasting overwhelming citation counts and influence, ranking at the top of all science and technology fields when applying Google Scholar’s journal impact metric (h5-index).


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing