Solar flares that can release enough energy to interfere with satellite communications or disrupt power grids on the surface of Earth might be predictable using machine learning and artificial intelligence, solar physicists say.
Researchers Sebastien Couvidat and Monica Bobra of Stanford University have analyzed the most extensive set of solar observations ever gathered by the Solar Dynamics Observatory, or SDO, which generates more amounts of data than any other satellite NASA has ever sent into space.
The goal of their study is to identify which features in the data might be the most useful in predicting solar flares.
The Stanford Solar Observatories Group stores and processes the 1.5 terabytes of data the SDO data gathers every day, and the group has been looking at ways to efficiently examine the massive amounts of data.
Creating prediction from so much data based on pure theory can be difficult, so they turned for inspiration to an online class in machine learning offered by Stanford computer science professor Andrew Ng.
Machine learning software is very good at assigning information into sets of established or known categories, then scanning for patterns which can identify which data is relevant for predicting the occurrence of certain categories.
"Machine learning is a sophisticated way to analyze a ton of data and classify it into different groups," Bobra says.
The more data that is available the better machine learning becomes in its predicting, the researchers say.
So Bobra and Couvidat wanted to find out how successful machine learning could be at predicting the strength of solar flares by using massive amounts of SDO data on sunspots, features on the sun's surface often associated with flared.
Couvidat applied the algorithms to the data and Bobra characterized the features of the two strongest classes of solar flares, M and X.
While M-class flares can result in minor solar storms that could endanger astronauts and cause brief radio blackouts at Earth's poles, flares of X-class are much more powerful and potentially more dangerous.
The researchers catalogued flaring and non-flaring areas of the sun's surface using a database of around 2,000 active regions, then categorized those regions by 25 features including energy, current and gradients of their magnetic fields.
They then put the data through the learning machine to train it to identify relevant features.
Using just a few of the 25 features, machine learning was able to discriminate between active regions that would flare and those that would not flare, providing a prediction result, the researchers report in The Astrophysical Journal.
The next step in solar flare prediction would be to incorporate data from the sun's atmosphere, says Bobra, who admits to a passion for solar physics.
"It's exciting because we not only have a ton of data, but the images are just so beautiful," she says.
