A Korea Meteorological Administration (KMA) official explains the AI-powered short-term precipitation forecasting model. (Photo courtesy of KMA)
SEOUL, Nov. 25 (Korea Bizwire) – South Korea will introduce AI-driven weather forecasting to the public next year, marking a significant advancement in meteorological technology.
The National Institute of Meteorological Sciences (NIMS) announced on November 20 that it plans to provide real-time AI-based short-term precipitation forecasts during the disaster prevention period from May 15 to October 15, 2025. This will be the first public application of AI in weather forecasting in the country.
The AI model, developed by NIMS, leverages advanced transformer technology—the same used in generative AI tools like OpenAI’s ChatGPT—to predict precipitation patterns.
Trained on seven years of radar imagery and ground observation data, the system generates short-term forecasts up to six hours ahead, in 10-minute intervals. For public use, predictions will cover up to two hours.
Starting next summer, citizens will be able to access these forecasts on the Korea Meteorological Administration’s (KMA) website and the Weather Alarm app, enabling users to track rain cloud movements in real-time.
The AI model demonstrates a high accuracy rate, with over 90% precision in predicting whether it will rain. Performance metrics such as the Critical Success Index (CSI), which measures the accuracy of forecasts during rainfall events, have shown promising results.
For instance, CSI scores for one-hour lead times averaged 0.5888 during the May-September 2024 test period.
While traditional numerical weather prediction models like the European Centre for Medium-Range Weather Forecasts (ECMWF) remain the backbone of global forecasting, AI models are quickly becoming viable complements.
Unlike traditional models that require extensive computational resources, AI models offer faster predictions, with the NIMS system generating six hours of forecasts in just 38–42 seconds.
AI-driven forecasting has advanced rapidly in recent years, with notable models such as Google DeepMind’s GraphCast and Huawei’s Pangu-Weather outperforming some traditional numerical systems in accuracy. South Korea’s NIMS asserts that its AI model rivals global counterparts and plans to expand its capabilities.
Future goals include addressing the “black box” challenge in AI systems by developing explainable AI and creating a global-scale weather forecasting model akin to GraphCast.
As extreme weather events such as heavy rains and heat waves become more frequent due to climate change, some consumers are considering reducing or discontinuing their use of dawn delivery services. (Image courtesy of Yonhap)
However, resource constraints remain a hurdle. AI model training requires extensive GPU resources, and NIMS currently operates with only eight NVIDIA A100 GPUs. These are significantly outpaced by newer H100 GPUs, which are up to 30 times faster in inference tasks.
Earlier this month, NIMS was designated as a National Strategic Technology Research Center, securing ₩15 billion in funding through 2028. The institute aims to develop a world-class AI-powered meteorological foundation model within the next five years.
Lamenting the secrecy surrounding weather AI technologies among major tech firms, NIMS official Lee Hye-sook emphasized the importance of investing in independent development. “We have the expertise to build models on par with GraphCast. With the right resources, South Korea can lead in AI meteorology,” she stated.
As AI continues to revolutionize weather forecasting, South Korea’s efforts highlight the transformative potential of combining advanced technology with public services.
Ashley Song (ashley@koreabizwire.com)
Kevin Lee (kevinlee@koreabizwire.com)
