China is putting serious effort into AI-powered weather forecasting, especially as extreme weather becomes more difficult to deal with. Several Chinese research teams and technology companies have built systems that can make forecasts very quickly.
Some have even matched or beaten traditional forecasting systems in certain tests. But this isn’t a simple story about computers becoming better than humans. AI has some impressive strengths, but it also has limits.
For now, the interesting part is how the old and new systems are being used together.
China’s AI Into the Weather Business
When Typhoon Dolphin was moving toward China, meteorologists weren’t sitting in front of one giant computer waiting for a single answer. Different forecasting systems were being used to understand what the storm might do next.
Among them were several AI models developed in China. Fengwu was developed by Shanghai AI Laboratory, while Huawei created Pangu and researchers at Fudan University developed Fuxi.
The idea behind them is quite different from the way traditional forecasting works. For decades, meteorologists have depended on numerical weather models running on powerful supercomputers.
These systems use the laws of physics to simulate what is happening in the atmosphere, taking information such as air pressure, temperature, wind, and humidity, then performing huge numbers of calculations to work out what the atmosphere might look like later.
AI doesn’t approach the problem the same way. Instead of calculating every part of the atmosphere from the ground up, AI models learn from enormous collections of historical weather observations.
They study what the atmosphere looked like in the past and what happened afterward, and over time the system learns patterns it can use to make new forecasts.
You could compare it to someone who has spent their entire life watching storms. After seeing thousands of them, that person might notice that certain combinations of wind, pressure, and temperature often lead to particular kinds of weather.
AI does something similar, except it can examine an amount of information that would be impossible for one person to handle.
And this is where the technology gets interesting: AI can produce some forecasts in a fraction of the time required by traditional systems, which could be especially useful in places where typhoons and other severe weather events are regular threats.
Minutes and Kilometers Start to Matter
A weather forecast can seem like a small part of everyday life. If it’s wrong, you might get wet. Annoying, but usually not a disaster.
Things are very different when a powerful storm is approaching a large population.
A forecast can influence whether people evacuate, whether boats stay in port, whether flights and trains need to change, and whether emergency workers are sent to certain areas.
This is why the speed of AI forecasting matters. Sun Zhi, chief technology officer of Techwind, the company working on Fengwu’s industrial applications, said that people need information to make decisions as extreme weather becomes more common.
He pointed to local and national governments, farmers, fishermen, and ordinary people as groups that can benefit from better forecasts.
According to Sun, Fengwu predicted the time and location of Typhoon Dolphin’s arrival on mainland China five days before the storm reached land, and the prediction was reportedly within about 30 minutes and 30 kilometers, or roughly 19 miles, of the actual event.
For something as unpredictable as a typhoon, that is a remarkable result.
Fengwu has also attracted attention because its developers reported that it performed better than Google’s GraphCast on around 80% of the weather variables they evaluated, and that it could produce useful global medium-range forecasts beyond 10 days.
That doesn’t mean you should throw away every weather app on your phone, but it does show that AI models are becoming serious competitors in a field long dominated by enormous computing systems.
China isn’t the only country interested in this. Google has developed GraphCast and GenCast. Nvidia has backed FourCastNet. The European Centre for Medium-Range Weather Forecasts has its own AI Forecasting System, known as AIFS.
The bigger story is global: technology companies, universities, and weather agencies are all trying to figure out how AI can make forecasting faster and more useful.
One Big Thing AI Struggles With
AI can be very good at some jobs and less reliable at others. One important weakness is storm intensity.
An AI model might do a good job predicting where a typhoon is going, but predicting exactly how powerful it will become is harder — and that difference can completely change how authorities prepare.
Traditional forecasting systems still have an advantage here. They’re based on physical models that describe how the atmosphere behaves, giving scientists a way to connect predictions with the actual processes happening in the air and ocean.
AI has learned from previous patterns, but learning patterns is not the same as fully understanding every physical process behind them.
There’s another problem when scientists look much farther into the future. Predicting a typhoon several days ahead is one challenge; predicting a major climate development many months ahead is a completely different game.
Sun gave the example of predicting an event such as El Niño 18 months in advance.
Even if an AI system produced such a prediction, people would still need a strong reason to believe it — they need evidence, testing, and proof the system has been right often enough to deserve their trust.
As Sun explained, researchers may need years of scientific work before people are comfortable relying on AI predictions about major climate events.
The Likely Future
The rise of AI weather models doesn’t necessarily mean traditional forecasting is about to disappear.
For quite some time, meteorologists may continue using both approaches at once, like having two tools in a toolbox, one old and reliable, one newer and faster, each used where it works best.
Traditional numerical models can continue providing detailed information based on atmospheric physics, while AI adds another layer by producing forecasts quickly and identifying patterns from huge amounts of historical data.
If the systems agree, meteorologists may have more confidence in the forecast. If they disagree, that difference becomes something worth studying.
For people living in areas regularly affected by typhoons, even a small improvement in forecasting can matter.
A more accurate track could give authorities extra time to organize evacuations. A faster forecast could help transport companies adjust schedules. Families could prepare before the storm becomes impossible to ignore.
AI cannot tell a typhoon to turn around. It can’t stop heavy rain from falling or prevent flooding by itself. What it can potentially do is give people a clearer picture of what may be coming.
That may sound less dramatic than a computer that can “predict the future,” but it is probably much more useful.
When a dangerous storm is heading your way, you don’t need a computer to know everything about the future. You just need it to tell you enough, early enough, to do something about it.
Sources:
Leave a Reply