Growing AI role in weather forecasts

Artificial intelligence-powered weather forecasting models have played a critical role in responding to extreme rainfall, snowstorms and other hazardous weather events this year, though major challenges remain in predicting sudden and highly localized events in northern China, meteorologists said.

AI-based forecasting systems provided key technical support during this year's flood season in North China, including a series of extreme rainstorms, as well as during recent snowfall events, experts from the China Meteorological Administration said on Friday while introducing upgrades to several models in Xiong'an New Area, Hebei province.

Used in conjunction with traditional numerical forecasting, the models have improved the timeliness and accuracy of warnings, helping strengthen disaster prevention and mitigation efforts, they said.

Cao Yong, head of the weather forecasting technology research division at the National Meteorological Center, said the medium- and short-range AI forecasting model known as Fengqing has been rolled out nationwide, with pilot versions deployed in regions including Hebei province.

During the flood season this year, prolonged rainstorms scattered across different locations in North China posed major forecasting challenges. "In 96-hour forecasts, the Fengqing model successfully captured the overall trend of the event," Cao said.

The model also performed well during North China's first snowfall of the season earlier this month, accurately predicting the timing, duration and intensity of the snowfall, he added.

China has also upgraded its AI-based "now-casting" system, known as Fenglei, which focuses on short-term and imminent weather forecasts. Zhang Xiaowen, head of the center's Fenglei research team, said the model allows forecasts to better adhere to atmospheric science principles while improving stability and precision.

The Fenglei system has also made breakthroughs in forecasting short-duration heavy rainfall and extreme downpours. In June, it accurately predicted a sudden severe convective storm in Beijing and an extreme rainstorm in Henan province.

"The model has shown a marked improvement over existing operational products in forecasting extreme short-duration rainfall exceeding 50 millimeters per hour, providing effective support for forecasters issuing early warnings for such extreme precipitation events," Zhang said.

However, predicting extreme weather events remains particularly challenging in northern regions, said Lu Bo, vice-president of the Xiong'an Institute of Meteorological Artificial Intelligence.

Compared with southern China, where flood-season rainfall is often long-lasting and relatively stable, North China is more prone to sudden severe convective storms or short-duration downpours triggered by distant typhoons. "These events are typically intense, highly uneven in space and time, and more difficult to predict," Lu said.

For example, the Fengshun model, an AI-based seasonal forecasting system, successfully predicted the July rain belt in North China by the end of June, but its performance lagged behind that of traditional numerical prediction models for the August rain belt, Lu said.

According to the Earth System Forecasting Development Strategy (2025-2035), a road map recently released by the CMA, a new generation of forecasting models will be operationally deployed and a unified foundational framework for meteorological AI will be built within the next five years.

Upgraded AI weather models will provide broader and more precise support for short-term warnings and extreme climate alerts, while also facilitating energy dispatching and agricultural planning, officials said.

Unlike traditional forecasting models that can be slow and strictly follow physical laws, AI forecasting models are extremely fast and accurate, though they can struggle with "unseen" record weather events.