A more symmetric inner core indicates higher likelihood of rapid intensification
Along the coastlines of the western Pacific, where typhoons have long humbled the most careful forecasters, a team in Shenzhen has placed a new kind of intelligence into the hands of those who watch the storms. China's National Meteorological Center has deployed a machine-learning ensemble model capable of predicting rapid intensification — the sudden, dangerous surge in a typhoon's power that has historically arrived without adequate warning. The system, born from the Chinese Academy of Sciences and tested against decades of storm data, now runs operationally, offering communities a few more hours to prepare for what the sea may send.
- Rapid intensification — a typhoon's wind speed surging more than 15 meters per second in a single day — has repeatedly blindsided forecasters, turning manageable storms into catastrophes that killed hundreds and erased billions in economic value.
- Traditional statistical and dynamical models struggle to capture the nonlinear chaos of a typhoon's inner life, producing missed warnings and false alarms that erode both public trust and emergency response effectiveness.
- Researchers identified two concrete physical signals — the symmetry of a typhoon's inner core and the ratio of ocean to land along its projected path — giving the model measurable anchors in an otherwise turbulent system.
- An ensemble of four machine-learning algorithms votes on whether rapid intensification will occur, with consensus required before a forecast is issued — a design that filters noise while preserving the signal that matters most.
- Tested against North Atlantic cyclone data and benchmarked against the U.S. National Hurricane Center's operational system, the model detected more true events and generated fewer false alarms, and is now processing live storms across the western Pacific.
A research team at the Shenzhen Institutes of Advanced Technology has built a machine-learning system designed to solve one of meteorology's most stubborn problems: predicting when a typhoon will suddenly and dramatically intensify. That system has now left the laboratory and entered operational use at China's National Meteorological Center, running alongside existing forecasting infrastructure as real storms develop across the western Pacific.
The danger it addresses is precise. When a typhoon's maximum sustained winds jump more than 15 meters per second in a single day — or 10 meters per second in 12 hours — a storm that seemed manageable can become catastrophic before coastal communities have time to respond. Typhoons Rammasun, Hato, and Yagi each underwent this kind of sudden strengthening before landfall, leaving casualties and economic wreckage in their wake. The difficulty of predicting these events led China's top scientific body to designate rapid intensification forecasting as one of the country's ten frontier scientific priorities in 2025.
The complexity of the problem lies in the overlapping, nonlinear forces that govern a typhoon's strength — inner-core structure, atmospheric moisture, ocean temperature, proximity to land. Traditional forecasting methods, built for more linear relationships, consistently struggle here. Lead researcher Li Qinglan approached the problem by identifying two physical signals that tend to precede rapid intensification: the symmetry of a typhoon's inner core, and a sea-land ratio measuring how much open ocean lies along the storm's projected path. These gave the team quantifiable markers to anchor their model.
Rather than relying on a single algorithm, the team built an ensemble of four machine-learning models that vote collectively. A forecast is only issued when a majority agrees — a mechanism that reduces false alarms while catching genuine events. When tested against historical North Atlantic cyclone data from 2016 to 2020 and compared directly with the U.S. National Hurricane Center's operational system, the new model outperformed on both detection rate and false alarm reduction.
The system is now operational, processing live storms. Whether its real-world performance matches its historical promise — and whether meteorological services beyond China's borders adopt similar approaches — remains the open question that only the coming typhoon seasons can answer.
A team of researchers in Shenzhen has built a machine-learning system that can predict when a typhoon will suddenly intensify—a shift that has historically caught forecasters off guard and left coastal communities vulnerable. The model, developed at the Shenzhen Institutes of Advanced Technology under the Chinese Academy of Sciences, has now moved from the laboratory into operational use at China's National Meteorological Center, where it runs alongside the country's existing forecasting infrastructure.
The problem the system addresses is specific and dangerous. Meteorologists define rapid intensification as a jump in maximum sustained wind speed of more than 15 meters per second in a single day, or more than 10 meters per second in 12 hours. When this happens, a storm that seemed manageable can become catastrophic in hours. Typhoon Rammasun in 2014, Typhoon Hato in 2017, and Typhoon Yagi in 2024 all underwent this kind of sudden strengthening before making landfall, each leaving behind a trail of deaths and economic damage. The unpredictability of these events has made them a persistent headache for meteorologists worldwide, and in 2025, the China Association for Science and Technology designated rapid intensification forecasting as one of the country's top ten frontier scientific problems.
The challenge lies in the sheer complexity of what drives a typhoon's intensity. A storm's strength depends on multiple overlapping factors—the structure of its inner core, the temperature and moisture of the surrounding air, the presence of land nearby, ocean currents below. These elements interact in ways that don't follow simple linear patterns. Traditional statistical and dynamical forecasting methods, which have served meteorology for decades, struggle to capture these nonlinear relationships. The result is that even the best existing systems miss rapid intensification events or issue false alarms that erode public trust.
Li Qinglan, who led the research team, approached the problem by identifying two measurable physical signals that appear to precede rapid intensification. The first is inner-core symmetry—the degree to which a typhoon's center develops a perfectly balanced, ring-like structure. A more symmetric core, the research showed, correlates strongly with the likelihood of rapid intensification. The second is what the team calls the sea-land ratio, a measure of how much ocean versus land lies along the path the storm will take. These indices gave the researchers concrete, quantifiable markers to work with.
Rather than building a single forecasting model, the team constructed an ensemble system that combines four different machine-learning algorithms. When more than half of these sub-models agree that rapid intensification will occur, the system issues a forecast. This voting mechanism acts as a filter, reducing the noise that any single algorithm might introduce while preserving the signal.
To test the system's reliability, the researchers ran it backward through historical data, simulating 24-hour rapid intensification events from tropical cyclones in the North Atlantic between 2016 and 2020. They then compared their results directly against the operational forecasting system used by the U.S. National Hurricane Center. The new model detected more genuine rapid intensification events and produced fewer false alarms—a combination that matters enormously in real-world forecasting, where a false alarm can trigger unnecessary evacuations and a missed event can cost lives.
Lyu Xinyan, a senior engineer at China's National Meteorological Center, described the 24-hour rapid intensification forecast capability as an important new tool for the country's typhoon forecasting operations. The system is now running operationally, processing real storms as they form and intensify across the western Pacific. What happens next depends partly on how well the model performs in the field, and partly on whether other meteorological services around the world adopt similar machine-learning approaches to a problem that respects no borders.
Notable Quotes
The evolution of typhoon intensity is controlled by multiple interacting factors that include inner-core structure, environmental background and land-sea surface interactions, making accurate forecast extremely difficult.— Li Qinglan, lead researcher at Shenzhen Institutes of Advanced Technology
24-hour rapid intensification forecast technology now provides an important reference for China's typhoon intensity forecasting— Lyu Xinyan, senior engineer at China's National Meteorological Center
The Hearth Conversation Another angle on the story
Why does a typhoon suddenly intensify? What changes?
Multiple things happen at once—the inner core becomes more symmetric, the surrounding air warms, the ocean beneath it stays warm. But these factors interact in ways we can't predict with old methods. That's the core problem.
So the researchers found a signal—symmetry—that appears before intensification happens?
Exactly. They noticed that before rapid intensification, the typhoon's center develops this perfectly balanced ring structure. More symmetry means higher risk. It's a physical marker you can measure.
How does a machine-learning ensemble actually improve on existing forecasts?
Instead of one model making a call, four different algorithms vote. If more than half agree intensification is coming, you issue a forecast. It's like having multiple experts in the room—you're less likely to miss something or cry wolf.
The comparison with the U.S. system showed better detection and fewer false alarms. Why does that matter operationally?
False alarms erode trust and waste resources. Missed events cost lives. You need both accuracy and restraint. This system appears to have found that balance.
Is this system now running on real typhoons?
Yes. It's operational at China's National Meteorological Center. It's processing actual storms as they form and intensify across the western Pacific.
What happens if it fails on a major storm?
That's the real test. Laboratory performance and field performance are different things. The system will prove itself or not based on what happens in the next typhoon season.