AI outperforming conventional models for typhoon track forecasts
TOKYO (Kyodo) -- Artificial intelligence-based typhoon track forecasts are outperforming conventional physics-based models, with AI accurately project
Tokyo (Kyodo) -- Machine learning-based typhoon track forecasts are proving more accurate than traditional physics-based models, as demonstrated by the recent path of Typhoon Champi. Private weather provider Weathernews Inc. employs AI forecasts, while Japan's meteorological agency is researching their potential use. However, forecasting typhoon intensity remains a challenge.
Conventional physics-based models rely on inputs like temperature, air pressure, and water vapor from supercomputers to predict outcomes based on physical laws. AI forecasts, on the other hand, analyze extensive historical weather data to detect patterns and tendencies, then apply these insights to estimate atmospheric movements.
Japan's Meteorological Agency primarily uses physics-based models but also consults external AI models. Typhoon Champi, the 30th typhoon of the year, formed over the Pacific on October 15 and initially headed west. Early physics-based forecasts from Japan and other countries predicted an approach to Taiwan before moving into China.
AI models, however, predicted a northward movement near Okinawa and a curve over the East China Sea. The typhoon indeed veered sharply after passing the Nansei Islands, a chain of islands southwest of Japan's main island Kyushu. AI is particularly adept at forecasting large-scale atmospheric patterns, including the Pacific high-pressure system and westerlies that influence typhoon paths, and accurately predicted the timing of the turn, according to Munehiko Yamaguchi of the Meteorological Research Institute's Typhoon and Severe Weather Research Department.
Weathernews has been issuing AI-driven typhoon forecasts to paid subscribers since last year, with track error for the first 26 typhoons last year being about 30% lower than physics-based models. Improving accuracy through optimal AI utilization is a priority, according to Kohei Sakamoto of Weathernews' forecasting center. Naoto Yamada, vice president of Tanemaki Joso, an agricultural company in Ibaraki Prefecture that grows cherry tomatoes, always considers AI forecasts, which he deems highly accurate.
When Typhoon Dujuan headed toward the Kanto region in September, Tanemaki Joso took early preventive measures, including installing flood barriers and adjusting staff schedules, based on Weathernews' AI predictions. The weather agency's initial forecast had indicated a path toward western Japan or Okinawa, while AI suggested a potential approach to Kanto, centered around Tokyo.
Yet, AI models also have limitations. Yamaguchi of the Meteorological Research Institute noted that AI struggles with predicting typhoon intensity, making precise rainfall and wind speed forecasts difficult. However, rapid technological advancements could soon address this issue. AI models cannot be disregarded; their application in disaster prevention information needs to be carefully considered, according to Yamaguchi.
Written by urgent.news from The Mainichi's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.