The Shenzhen Institutes of Advanced Technology (SIAT) at the Chinese Academy of Sciences has deployed an ML-based ensemble model for predicting typhoon rapid intensification at China's National Meteorological Center. The model -- "Machine Learning Ensemble Model for Tropical Cyclone Rapid Intensification Forecast" -- uses four ML algorithms and two novel indices: the sea-land ratio (land-sea distribution along a typhoon's track) and the symmetric ratio (inner-core convective symmetry). A majority-vote signal from at least half the sub-models triggers a forecast. Rapid intensification is defined as maximum sustained wind speed increasing by more than 15 m/s in 24 hours. Testing against the US National Hurricane Center's operational system on North Atlantic data (2016-2020) showed higher probability of detection and lower false alarm rates. A companion 12-hour rapid intensification forecast service is also now live.
What happened
The Shenzhen Institutes of Advanced Technology (SIAT), Chinese Academy of Sciences, has completed operational deployment of a new machine-learning ensemble model for typhoon rapid intensification forecasting at China's National Meteorological Center, Xinhua reported on June 15, 2026. The deployment covers a 24-hour rapid intensification forecast model and a 12-hour rapid intensification forecast service.
The model and methodology
The model, formally named the "Machine Learning Ensemble Model for Tropical Cyclone Rapid Intensification Forecast," was developed by a team led by Li Qinglan at SIAT. Four machine-learning algorithms are integrated into an ensemble: when more than half of the sub-models predict rapid intensification, the system issues a forecast signal. This majority-vote design is intended to improve accuracy while reducing false alarms.
Two new quantitative indices drive the ensemble. The sea-land ratio captures variations in land-sea distribution along a typhoon's track, reflecting how coastal geometry modulates intensity. The symmetric ratio describes inner-core convective symmetry -- "prior to rapid intensification, a typhoon's inner core typically develops a highly symmetric ring-like structure," Xinhua quoted Li as explaining. "A more symmetric inner core indicates a higher likelihood of rapid intensification occurrence."
Technical benchmark
In meteorology, rapid intensification is defined as maximum sustained wind speed increasing by more than 15 m/s within 24 hours, or more than 10 m/s within 12 hours. The team tested the model against the US National Hurricane Center's operational system using North Atlantic tropical cyclone data from 2016 to 2020. The new model showed a higher probability of detection and a lower false alarm rate, according to Xinhua.
Operational context
Conventional statistical-dynamical methods fail to capture the nonlinear dynamics of typhoon intensity change, Li noted. The problem is high-stakes: rapid intensification events such as Rammasun (2014), Hato (2017), and Yagi (2024) all struck land after sharp intensification, causing significant casualties and economic losses. In 2025, the China Association for Science and Technology designated typhoon rapid intensification forecasting as one of the top ten frontier scientific problems in the country.
National Meteorological Center senior engineer Lyu Xinyan said the 24-hour rapid intensification technology "now provides an important reference for China's typhoon intensity forecasting," per Xinhua.
Significance for practitioners
This deployment illustrates combining an ensemble majority-vote scheme over multiple ML algorithms with domain-driven engineered features -- inner-core symmetry and coastal geometry indices -- to outperform a major operational numerical forecasting baseline on a safety-critical task. The use of physically motivated feature engineering (rather than raw image/field inputs alone) shows how domain knowledge can improve ML model reliability in earth-system science applications.
Scoring Rationale #
Operational deployment of an ML ensemble model for typhoon rapid intensification at China's National Meteorological Center is a concrete applied-ML advance relevant to earth-system practitioners. The model introduces domain-driven feature engineering (sea-land and symmetric ratios) benchmarked favorably against the US NHC, placing it in the solid-to-notable band without reaching landmark research status.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.