A domestic research team has developed a system that can precisely predict a range of weather conditions, including extreme rainfall, using artificial intelligence (AI) based solely on radar images.
The National Research Foundation of Korea (NRF) said on the 17th that a research team led by Hong Young-jun, a professor in the Department of Mathematical Sciences at Seoul National University, developed an AI model that can predict very short-term rainfall. The study was accepted by the international AI conference "ICLR 2026" and is scheduled to be presented in Apr.
Existing global precipitation prediction models have limits in real-time response due to massive computational demands, and they often struggle to precisely forecast Korea's localized rainfall patterns in complex terrain.
The researchers sought to overcome this limitation by designing the AI architecture with an applied mathematics approach. First, they introduced a "local attention mechanism" that allocates computational resources intensively to rapidly changing precipitation patterns in radar images. As a result, while accurately reconstructing the fine-grained information that triggers very short-term heavy rainfall over 1 to 6 hours, they minimized unnecessary computation to improve efficiency.
To verify the effectiveness of the developed model, the team evaluated its performance using real observation data from the United States, France, and the Korea Meteorological Administration. As a result, it showed higher accuracy than state-of-the-art global models, and it also improved accuracy in predicting Korea's extreme rainfall situations.
When applied to the 2023 torrential rain case, it was able to detect potential risk situations in advance, demonstrating that a Korea-tailored very short-term rainfall prediction model can be applied to practical early warnings and preemptive disaster response.
Hong Young-jun said, "This study is meaningful in that it overcame the AI 'black box' limitation based on mathematical rigor and implemented a model that operates reliably even under extreme weather conditions," adding, "I hope this technology will be integrated into practical disaster prevention systems and play an important role in protecting citizens' safety."