Machine learning is helping to uncover some of the mysteries behind Jupiter’s aurorae. By relying on computers to classify the types of aurorae on the gas giant, astronomers have found that the solar wind plays an even smaller part in forming the beautiful events than previously suspected.
Like aurorae on Earth, Jupiter’s aurorae are charged particles interacting with the planet’s magnetic field. But whereas Earth’s aurorae are charged predominantly by the solar wind, charged particles flowing from the surface of the Sun, the bulk of Jupiter’s dazzling displays seems unconnected to solar inputs.
“Most of these things don’t really seem to care about what the solar wind is doing, which is a little bit surprising,” said Jonathan Nichols, a researcher at the United Kingdom’s University of Leicester. Nichols and his colleagues developed a computer program to analyze images of Jupiter’s atmosphere captured by the Hubble Space Telescope in 2016, as NASA’s Juno spacecraft approached the planet. They found that most of Jupiter’s aurorae are driven by something other than the solar wind. “[That] tells us that Jupiter’s magnetosphere is fundamentally different from that of the Earth,” Nichols said.
According to Nichols, this is the first time any quantitative method has been applied to studying Jupiter’s aurorae.
Jupiter hosts almost 80 known moons. Most of them are small, but the four Galilean moons are substantial. One of these, Io, is the most volcanically active body in the solar system and plays a strong role in creating Jupiter’s aurorae.
When the volcanoes covering Io erupt, the small moon doesn’t have the gravity to hold on to the material. The vented material winds up orbiting Jupiter, creating a nebulous ring around the planet. At the mercy of the gas giant, the thin ring finds itself rotating at the same speed as Jupiter, pulled by the electric currents of the planet’s magnetic field. These currents create aurorae that dot the planet.
The solar wind also plays its own role. On Earth, the pressure of the solar wind is fairly constant, and the aurorae continually light up the sky around our planet’s poles. But Jupiter orbits more than 5 times farther out than Earth, and at those distances the solar wind behaves differently. The fast-moving components of the solar wind leave behind the slower components, creating a snowplow effect as fast particles pile up behind slower particles that left the solar surface earlier. The result is long, slow periods of almost no solar particles followed by short, high-intensity bursts of strong solar wind. The intense bursts squish the planet’s atmosphere, and the energy lights up as aurorae.
“On Jupiter, the auroras are much more complicated [than on Earth],” Nichols said. “We just don’t understand what drives” certain regions of these aurorae.
By using machine learning to classify the different features, Nichols and his colleagues split the aurorae into six different types. They found that one type, identified as “class 5,” was the only one that occurred during intervals of solar wind. The other five happened whether or not solar particles were present, driven instead by Io’s interactions.
Nichols said that the idea that the solar wind plays very little role in Jupiter’s aurorae is not unexpected. Scientists already knew that Jupiter’s magnetic field was driven primarily by rotation and by its interaction with Io.
“Even if the solar wind doesn’t drive most of the auroras, like on Earth, we would expect it to at least modulate what’s going on,” Nichols said. That’s “quite surprising.”
The research was published in the Journal of Geophysical Research: Space Physics.
“Small Variations” Revealed by Machine Learning
Nichols and his colleagues weren’t the first to take a look at the 2016 Hubble data. Juno scientist Denis Grodent had previously combed through the images to better classify the aurorae. But unlike Nichols, the Juno team didn’t use machines.
“We did it the old way, by using our eyes, our common sense, and, unfortunately, our subjectivity,” said Grodent, a researcher at the University of Liège in Belgium.
By relying on machine learning, Nichols and his colleagues removed the subjectivity from the process, a step that Grodent applauded.
“When you’ve got a lot of data, at some point it’s no longer possible to find these small variations which contain a lot of information,” Grodent said. “At some point, your eyes, your common sense, [are] no longer sufficient to retrieve information from [these] data.”
When astronomers sort the aurorae, their classifications aren’t strongly codified. “You subjectively get the idea that this kind of image or pattern in auroras occurs when the magnetosphere is hit strongly by the solar wind,” Nichols said. “But ‘having a feeling that something happens all the time’ is not scientific.”
Since its arrival at Jupiter, Juno has already collected millions of images to sort through. “It’s becoming more important to have automatic procedures,” Grodent said. “You can’t do that by hand.”
As astronomy and planetary science collect an ever-increasing amount of data, Grodent thinks machine learning will become mandatory for researchers. Nichols’s program is one of several possible techniques, and Grodent hopes to see more develop in the near future, allowing him to directly apply them to his own research.
Nichols plans to expand his study to include more data coming in from both Hubble and Juno.
“It’s really a very exciting time,” Nichols said. “It could be the case that machine learning could bring out behaviors that we have otherwise missed.”
—Nola Taylor Redd (@NolaTRedd), Freelance Science Journalist
Redd, N. T. (2019), Computers tease out secrets of Jupiter’s aurorae, Eos, 100, https://doi.org/10.1029/2019EO136742. Published on 21 November 2019.
Text © 2019. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.