“For predicting precipitation 1 or 6 hours ahead, traditional numerical forecast models are more accurate. However, for predicting precipitation 2 to 5 hours ahead, AI prediction models may be more accurate.”

At the 'AI Nowcasting Pilot Project (AINPP)' workshop held on the 24th of last month at the National Institute of Meteorological Sciences in Seogwipo City, Jeju, Lee Hae-sook, head of the AI Meteorological Research Division, stated this. With the advancement of AI weather prediction technology, some forecasts are surpassing the accuracy of traditional numerical forecast models.

Lee stated, “When a disaster occurs due to the climate crisis, at least 6 hours are needed to evacuate people or block roads,” and added, “The goal of the National Institute of Meteorological Sciences is to support domestic AI models overseas so that the whole world can secure this time.”

● Domestic AI Weather Forecasting System ‘Now Alpha’

This workshop was the second AINPP meeting organized by the World Meteorological Organization (WMO). Over 70 participants, including major national meteorological agencies, the U.S. National Center for Atmospheric Research (NCAR), the European Centre for Medium-Range Weather Forecasts (ECMWF), and big tech company representatives attended. WMO aimed to verify AI weather forecast models from each country at this meeting and find prediction models applicable to developing countries by 2027.

South Korea introduced ‘Now Alpha,’ an AI nowcasting precipitation prediction model that has been used in actual forecasting since May. The National Institute of Meteorological Sciences established the AI Forecasting Research Team in 2019 to research AI forecast models tailored to domestic climate characteristics. As a result, Now Alpha was unveiled in July last year. Just as generative AI learns vast text data to understand language, Now Alpha learns weather data to predict precipitation up to 6 hours ahead in 10-minute intervals.

Traditional numerical forecast models, used since the past, predict weather by inputting factors affecting weather, such as temperature and sunlight, into equations explaining atmospheric conditions and movements. Since the Earth's atmosphere is divided into numerous grids and calculated separately, it requires immense time and resources for computation. Woo Jin-kyu, a forecast analyst at the Korea Meteorological Administration, stated, “AI forecast models consume a lot of computational resources to train and create, but once trained, they produce results at a very fast speed.”

Professor Choi Jae-sik's team from KAIST AI Graduate School first applied an ‘explainable AI’ forecast model to Now Alpha. Explainable AI is an AI that provides reasons and evidence along with judgment results, contrasting with ‘black box AI,’ which does not explain the conclusions obtained through algorithms. Now Alpha presents typhoons, monsoon fronts, etc., as evidence for meteorological predictions to forecasters. The National Institute of Meteorological Sciences plans to develop a medium-range weather forecast model optimized for the East Asian region by next year and a more versatile high-performance model by 2029.

● Google and NVIDIA Also Enter AI Weather Forecasting

AI forecast models are evaluated to have almost caught up with traditional numerical forecast models in other advanced meteorological countries as well as domestically. ECMWF has been operating an AI program ‘AIFS’ since February, which learns the results of traditional numerical forecast models. ECMWF explains that AIFS performs up to 20% better than traditional models in predicting paths of tropical cyclones like typhoons. ECMWF's traditional numerical forecast models are renowned for their global accuracy.

Big tech companies have also entered the development of AI weather forecasting systems. NVIDIA is developing ‘ForecastNet,’ and Google is working on ‘GraphCast.’ Jeff Adie, a senior engineer at NVIDIA who attended AINPP, stated, “Predicting climate change with AI is one of NVIDIA's key research projects,” and added, “There is a dedicated team directly overseen by CEO Jensen Huang.”

Experts emphasized the need to narrow the gap in weather forecasting capabilities between developed and developing countries. Yuki Honda, head of WMO's Integrated Processing and Prediction Systems Division, stated, “WMO's role is global cooperation and coordination, establishing AI technology guidelines, and supporting member countries,” and evaluated, “South Korea is an important partner in international cooperation due to its outstanding AI weather research capabilities.” David Gagne, leader of NCAR's Earth System Machine Learning Group, stated, “Models applicable to both developed and developing countries must be developed,” and added, “Accurately reflecting the geographical characteristics of weather data in AI models is key to long-term climate prediction.”

Popular News