Artificial Intelligence Helps Find Magnetic Eruptions in Space

A warning will appear in your email: The latest spacecraft observations are ready. You now have 24 hours to sift through 84 hours of data and select, in a split second, the most promising moments you can find. The data points you select will be downloaded from the spaceship in the highest possible resolution, depending on their ranking. Researchers can spend months analyzing them. Everything else will be overwritten as if it was never collected.

These are the missions that the Scientist in the Loop faces, one of the most important roles in the Magnetospheric Multiscale (MMS) mission team. 73 volunteers share responsibility and work for a week each to make sure the best data gets down to earth. It takes a keen and meticulous eye, which is why it has always been left to a carefully trained person – at least until now.

A recently published paper in Limits in astronomy and space science describes the first algorithm for artificial intelligence that gives the scientist a (virtual) hand in the loop.

“MMS is the first big one NASA Mission that implements machine learning into their mission operations, ”said Matthew Argall, space physicist at the University of New Hampshire and lead author of the paper.

The algorithm performs a single task: detecting when the spaceship passed from the earth’s magnetic field to the sun or vice versa. But it is only the first of many special algorithms that could change the way MMS science works.

Burst the bubble of the earth

An invisible force field surrounds our planet, a giant bubble that flies more than 40,000 miles into space. This is our magnetic field and it serves us in several ways. It keeps things away and deflects harmful cosmic rays that would otherwise hit the surface of the earth and endanger life. But it also keeps things under control and sets traffic patterns for the particles buzzing through near-earth space. Tiny and light electrons spin tight pirouettes around the earth’s magnetic field lines. heavier ions move in slower, wider loops.

But the earth’s magnetic field is nothing compared to that of the sun. Particles blown away by the sun, known as solar wind, carry our star’s magnetic field far beyond the orbit of Neptune. The particles contained therein follow the sun’s magnetic field lines and collide with the Earth’s magnetic bubble. The collision points form an invisible boundary that scientists refer to as the magnetopause.

By and large, magnetopause keeps you strong – but not always. If the conditions are right and the magnetic fields align, the solar wind can pierce our magnetic bubble. The location of the injury is known as the electron diffusion region or EDR, and locating these locations is the main goal of the MMS mission.

Within an EDR, the magnetic field lines of the sun and earth merge, cancel each other out and disappear. Excited and unbound electrons move back and forth in a chaotic pandemonium.

“It’s like they lost their lane while someone stepped on their accelerators,” said Barbara Giles, senior project scientist at MMS.

These particle bursts set off a chain reaction that ignites the northern and southern lights – they can even endanger astronauts and spacecraft on their way. EDRs erupt across the universe, from the center of the solar flares to the edges of the black holes. MMS looks for them closer to home on the edge of the earth’s magnetic field.

But it’s extremely difficult to catch you in the act. EDRs are displayed without notice, are only two miles in diameter (within a 14 billion mile search space), and last only a tenth of a second. In five years of continuous searching, MMS measured just over 50. But every time it crosses the magnetopause where our magnetic field meets that of the sun, it has another chance to see one.

Hunt for magnetopause transitions

So the scientist in the loop searches the data of each orbit and looks for magnetopause transitions. They don’t necessarily stand out in the data, however. By identifying them, you are more likely to be able to tell when a drizzle turns into rain. A single orbit data can contain only two or up to 100 magnetopause transitions with false alarm look-a-likes in between. To find it, the scientist simply has to enter the time in the loop.

“In the earliest days, it was basically a full-time job,” said Rick Wilder, a space physicist at the Atmospheric and Space Physics Laboratory in Boulder, Colorado. Since then, Wilder has helped to optimize the scientist’s workflow in the loop and trained new employees to become seasoned experts. Today an experienced scientist only needs a few hours a week up to date. But it’s still a strain on researchers who volunteer for busy schedules. “Fatigue is always in the back of your mind,” said Wilder.

They had always planned to automate parts of the scientist in the role of the loop, but finding an algorithm that would match human performance was a challenge. Scientists can spot larger trends in data, which most algorithms are difficult to do. “One scientist is examining, among other things, the time course of the data,” said Argall. “For example, to be able to recognize that you are at a point in the magnetosphere and use that to influence how [you see] the data evolves. ”

Argall and his team have developed an algorithm that tries to emulate how people read data. It takes the form of a neural network, a brain-inspired computing technique. In contrast to conventional algorithms, neural networks program themselves through trial and error. Argall showed the network examples for magnetopause transitions and then tested them on new cases. If it was answered incorrectly – a non-intersection was selected or a real intersection was overlooked – it would send an error signal and trigger a cascade of adjustments before the next test. Like human scientists in the loop, the network learned to identify magnetopause transitions from experience.

Most neural networks, however, process data in isolated snapshots as scientists watch measurements unfold over time. The team approached the scientist’s ability by using Gates to save the data the network had just seen as well as the data that came up next. When the network decides whether or not it is a magnetopause crossing, it can access surrounding data points to help. “The algorithm adds past and future input data to provide the context for the decision it is currently making,” said Argall.

It is the first of many algorithms. The team envisions building several specialty detectors to work together in a hierarchy. (A collection of specialists, others have found outperforms an all-rounder algorithm.) At the lowest level, “region classifiers” examine the data to find out where the spaceship is in space. They pass their output on to region-specific “event classifiers” that look for the phenomena researchers want to find. With success over the next few years, MMS was able to automatically detect far more than magnetopause transitions.

“We could take requests such as a certain signature in the data and store them in real time,” said Giles. “In that sense, it becomes a systems observatory – a community resource.”

That is still a long way off. The new algorithm currently corresponds to human judgments about 70% of the time. (Even scientists don’t agree 100%.) Since October 2019, Scientist in the Loop has treated it every week like an assistant checking their work and catching errors.

“But I’m sure that with the techniques he’s developing, he’ll get rid of the scientist on the loop in a couple of years,” said Giles. “We’ll know when that day comes because all they’re going to do is go in, check a box, and move on.”

With a trusted algorithmic assistant by their side, scientists could focus on the problems in the data that they can’t yet flag. We may see a future in which algorithms are less tools than employees and work with scientists as both learn from new data together.

Reference: “MMS SITL Ground Loop: Automating the Burst Data Selection Process” by Matthew R. Argall, Colin R. Small, Samantha Piatt, Liam Breen, Marek Petrik, Kim Kokkonen, Julie Barnum, Kristopher Larsen, Frederick D. Wilder, and Mitsuo Oka, William R. Paterson, Roy B. Torbert, Robert E. Ergun, Tai Phan, Barbara L. Giles and James L. Burch, September 1, 2020, Limits in astronomy and space science.
DOI: 10.3389 / fspas.2020.00054