AI innovations aid extreme weather management

Artificial intelligence has been increasingly applied to tackle the growing threat of extreme weather events such as heavy rainfall, hail and storms, according to groundbreaking innovations unveiled recently by the China Meteorological Administration.

The administration showcased 16 innovations in urban meteorological science at a conference in Beijing on Monday, which are part of 103 research breakthroughs in AI applications, radar networking and short-term heavy rainfall forecasting.

These achievements were delivered by the Urban Meteorological Science and Technology Alliance, initiated by the Beijing Meteorological Service last year and comprising meteorological departments from 38 major cities.

Beijing's Leadsee-Precip, a global deep-learning model designed to generate precipitation forecasts from meteorological circulation fields, is one of the domestically developed AI-powered forecasting systems that excels in predicting rainfall distribution and intensity.

Feng Jin, head of the Leadsee development team and a researcher at the Institute of Urban Meteorology, said deep-learning weather forecasting models have surpassed traditional numerical models in both accuracy and efficiency.

AI-driven global circulation models or GCMs, which integrate traditional numerical weather prediction techniques with machine learning, can deliver forecasts in no more than a minute - a dramatic improvement over the 30-minute processing time required by conventional models, Feng said.

However, current GCMs lack detailed atmospheric and precipitation data, which Leadsee compensates for by focusing on rainfall prediction, he added.

"Leadsee addresses critical gaps in AI global circulation models, particularly in forecasting extreme rainfall," Feng said, adding that the model ensures precision even with imbalanced rainfall data.

During Typhoon Gaemi, which made landfall in Fujian province and Taiwan in July, Leadsee successfully forecasted a shift in rainfall patterns over Beijing, enabling local authorities to adjust flood prevention strategies effectively.

The Beijing Meteorological Service conducted a comprehensive assessment based on references provided by Leadsee, concluding that Gaemi's impact on the Beijing area would significantly weaken, Feng said.

Additionally, evaluations of the model during this year's flood season demonstrated a 20 percent improvement in forecasting accuracy for heavy rainfall compared to mainstream models.

Beyond Beijing, the Shenzhen Meteorological Bureau in Guangdong province has developed an AI-based system for heavy rainfall nowcasting, and the effective lead time for nowcasting of heavy rainfall has been extended from one hour to two hours. Leveraging high-resolution datasets from radar, satellites and weather stations, the system has already surpassed traditional methods, according to the bureau.

The technological innovation has proved vital for disaster response and event planning during major events, such as the 40th anniversary celebrations of the Shenzhen Special Economic Zone in 2020 and the 25th anniversary of Hong Kong's return to China in 2022, the bureau added.

Chen Zhenlin, head of CMA, said Leadsee is an example of provincial meteorological departments actively exploring the field of artificial intelligence.

"These innovations provide robust technological support for further improving urban meteorological services," Chen said.