When researchers analyzed satellite observations collected over the North Pole in 2014, they realized they had captured an entirely new type of space weather event—a giant cyclone-shaped aurora.

Unlike hurricanes on Earth, this storm was made of electrically charged particles (plasma) flowing through the upper atmosphere. The phenomenon, now known as a space hurricane, can interfere with satellite operations, radio communications, navigation systems, and radar. 

Despite its potential impact, finding these storms has remained surprisingly difficult. Researchers had to comb through vast collections of satellite images by hand, a slow and subjective process that made large-scale monitoring nearly impossible. 

Now, a team of Chinese researchers has devised a solution for this problem. “To overcome this, we developed an artificial intelligence system that can automatically spot and pinpoint space hurricanes in ultraviolet images from satellites,” the researchers said.

Uncovering a hidden storm above the poles

The challenge facing researchers was not a lack of data but a lack of efficient ways to analyze it. Space hurricanes occur in the Earth’s ionosphere and magnetosphere near the magnetic poles, where streams of energetic particles interact with the atmosphere. 

These events create giant rotating auroras that can span hundreds or even thousands of kilometers. Scientists only confirmed the first documented space hurricane in 2021, although the event itself occurred in 2014. 

At that time, researchers identified a plasma spiral roughly 1,000 kilometers wide that hovered above the North Pole for nearly eight hours. The discovery revealed a previously unknown type of space weather phenomenon.

What made the discovery particularly surprising was that the storm developed during exceptionally quiet geomagnetic conditions. Until then, scientists generally associated major space-weather disturbances with periods of intense solar and geomagnetic activity. 

The finding suggested that powerful energy-transfer processes can occur even when space weather appears relatively calm. Subsequent studies showed that these storms can inject large numbers of high-energy electrons into the polar ionosphere, potentially disrupting communication and navigation technologies.

However, identifying new events remained a major chokepoint. Researchers typically had to inspect ultraviolet auroral images captured by satellites manually. This process was time-consuming, inefficient, and vulnerable to human judgment.

Training AI to recognize space hurricanes

To overcome these limitations, the Chinese researchers developed a deep-learning system designed specifically to recognize the telltale signatures of space hurricanes. The researchers assembled an enormous dataset containing about 300,000 auroral images collected between 2005 and 2021 from both the Northern and Southern Hemispheres.

The images came from instruments aboard the US Air Force’s Defence Meteorological Satellite Program satellites, which monitor conditions in near-Earth space. From this archive, the team identified 570 confirmed space hurricane events. 

They also included large numbers of non-hurricane auroral images, including examples that closely resembled genuine space hurricanes, to teach the AI how to distinguish between similar-looking phenomena.

Using this dataset, the researchers trained multiple advanced computer-vision models. The systems were designed not only to recognize the distinctive spiral structures associated with space hurricanes but also to pinpoint their locations within satellite images, allowing researchers to identify and track events more efficiently.

According to the team, the best-performing model achieved nearly 98 percent detection accuracy on the global dataset, demonstrating a high level of reliability in identifying space hurricanes automatically.

“The model achieves high-precision automatic identification and pixel-level localization of space hurricanes, reaching an accuracy of 97.90% on a challenging global dataset,” the study authors note.

Moreover, the researchers built a complete software platform with a visual interface, allowing scientists to process and examine satellite imagery more efficiently. So instead of manually searching through thousands of images, researchers can now rely on AI to rapidly identify potential events and pinpoint where they occur.

From detection to forecasting

The new system arrives as space-weather missions begin generating unprecedented amounts of auroral data. 

One example is the Solar Wind Magnetosphere Ionosphere Link Explorer (SMILE), a joint China-Europe mission launched in May that will continuously capture high-resolution ultraviolet images of Earth’s auroras. 

For researchers, manually inspecting such vast datasets is becoming increasingly impractical. The AI tool offers a way to process this flood of observations automatically, helping scientists track space hurricanes and better understand how they form and evolve. 

However, while the system can identify these events with high accuracy, the next challenge is predicting them. This is why the researchers now plan to combine real-time satellite and ground-based observations to develop nowcasting and short-term forecasting capabilities.

The study is published in the journal Space Weather.

Recommended Articles