A tree pushed over by strong wind during typhoon Matmo is pictured by a street in Wenchang, south China's Hainan Province, Oct. 5, 2025. Typhoon Matmo, the 21st named storm of the 2025 Pacific typhoon season, made landfall along the eastern coast of Xuwen County, Zhanjiang City in south China's Guangdong Province around 2:50 p.m. on Sunday, according to Guangdong's meteorological service. Heavily affected by Matmo, many parts of Hainan Island, which sits to the south of Guangdong, still suffer from strong wind and rainfall late on Sunday. Photo: Xinhua

Chinese scientists have developed a machine learning-based typhoon rapid intensification forecasting model which has been deployed and tested in operational use at the National Meteorological Center and the Hong Kong Observatory, marking a significant breakthrough for China in the field of forecasting sudden changes in tropical cyclone rapid intensity which has long remained a major challenge worldwide, the Global Times learned from the developer of the model on Wednesday. 

Meteorologists define typhoon rapid intensification as a process in which a typhoon's intensity increases by more than 15 meters per second within 24 hours, or by more than 10 meters per second within 12 hours. Such changes are highly destructive. 

Super typhoons including Rammasun in 2014 and Yagi in 2024 all rapidly intensified before making landfall, resulting in heavy casualties and significant economic losses, China Science Daily reported. 

The machine learning-based typhoon rapid intensification forecasting model developed by the research team led by Li Qinglan, a researcher at Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, is the first in China to be put into operational use for 24-hour typhoon rapid intensification forecasting, with a 12-hour forecasting model also launched, marking the integration of independently developed domestic AI typhoon rapid intensification forecasting technology into the national meteorological system.

According to Li, typhoon rapid intensification is hard to anticipate, making timely preparedness and response difficult. 

Typhoon intensity evolution is governed by inner-core structure and the underlying sea-land surface conditions, making intensity forecast even more challenging. 

Traditional numerical models cannot accurately represent the evolution of typhoon intensity due to resolution and parameterization limits. As a result, typhoon intensity forecast, particularly rapid intensification, remains a long-standing challenge, Li told the Global Times. 

Li noted that typhoons often develop a symmetric ring-shaped inner core before rapid intensification. A more uniform and symmetrical typhoon inner-core structure signals a higher chance of rapid intensification.

In recent years, artificial intelligence and machine learning, which excel at processing large datasets and uncovering complex nonlinear relationships, have emerged as key technologies for improving tropical cyclone forecasting, Shenzhen Evening News reported. 

The team combined four machine learning algorithms including Decision Trees, Random Forests, AdaBoost, and LightGBM into an ensemble system that forecasts rapid intensification to improve prediction accuracy, according to Li. 

The model was tested against all North Atlantic 24-hour tropical cyclone rapid intensification cases from 2016 to 2020 and it has achieved higher detection rates and lower false alarms, showing strong operational performance compared with the US National Hurricane Center's best system, Li said.

China Meteorological Administration released on April 1 its climate trend forecast for the main flood season from June to August this year. The administration expects 24 to 26 typhoons will form over the Northwest Pacific and the South China Sea in 2026, with 7 to 9 making landfall in China - higher than the long-term average. Typhoon tracks are expected to be mainly westward and northwestward, primarily affecting the coastal regions of East China and South China, with overall stronger-than-normal intensity, CCTV News reported on April 1.