AI Algorithm Takes Us Closer to Forecasting the Northern Lights

A group of researchers used artificial intelligence to sort nearly one billion images of the aurora borealis—the Northern Lights—which could help researchers understand and predict the remarkable natural phenomenon down the line.

The team developed a novel algorithm to sort through over 706 million images of the aurora borealis in the THEMIS all-sky images that were taken between 2008 and 2022. The algorithm sorted the images into six categories based on on their characteristics, showing the utility of the software for categorizing large-scale atmospheric datasets.

“The massive dataset is a valuable resource that can help researchers understand how the solar wind interacts with the Earth’s magnetosphere, the protective bubble that shields us from charged particles streaming from the sun,” said Jeremiah Johnson, a researcher at the University of New Hampshire and the study’s lead author, in a university release. “But until now, its huge size limited how effectively we can use that data.”

The team’s research—published last month in the Journal of Geophysical Research: Machine Learning and Computation—describes an algorithm trained to automatically label hundreds of millions of images of aurora, potentially helping scientists explore the ethereal phenomenon with speed at scale.

There have been plenty of auroras this year, in part because the Sun is at the peak of its solar cycle. The peak of the Sun’s 11-year solar cycle is defined by increased activity on the star’s surface, including eruptions of solar material (coronal mass ejections, or CMEs), and solar flares.

These events send charged particles out into space, and when those particles react with the particles in Earth’s atmosphere, they cause an ethereal glow in the sky: auroras. The particles can also disrupt electronics and power grids on Earth and in space, but we’re just talking about the pretty natural phenomena right now, not the merciless chaos that space weather can rain down on humankind.

“The labeled database could reveal further insight into auroral dynamics, but at a very basic level, we aimed to organize the THEMIS all-sky image database so that the vast amount of historical data it contains can be used more effectively by researchers and provide a large enough sample for future studies,” Johnson said.

The intensity of solar storms is difficult to predict because scientists can’t measure the solar outbursts they come from with precision until the particles are within an hour of arriving on Earth.

The team sorted the hundreds of millions of images into six categories: arc, diffuse, discrete, cloudy, moon, and clear/no aurora. Scientists may stand to gain from comparing the auroras with atmospheric data from the time the aurora occurred and linking the phenomena to the solar event that ultimately caused the light show.

Better understanding the chemical mix of solar particles and those in Earth’s atmosphere will help scientists determine which types of auroras arise from each scenario, and the ability to interrogate hundreds of millions of images with haste (compared to the rate of that work when done by humans) could be a boon to aurora research.