Interdisciplinary research led by a team of data scientists at University of Michigan could transform the way astronomers and physicists analyze solar activity and forecast space weather events, by creating a powerful composite data set through machine learning.
The team, led by David Fouhey (Computer Science and Engineering), started the project with seed funding from the MIDAS Propelling Original Data Science (PODS) Grant program. They have since gone on to secure a NASA Heliophysics Living with a Star Tools and Method grant to continue the research.
“We’ve created a system that will produce synthetic images of the magnetic field of the sun both really fast and really accurately thanks to machine learning.” said Fouhey, “This can produce virtual instruments that, as of right now, are not physically possible. These virtual instruments enhance the return value on existing missions and are only possible thanks to the wealth of data that’s available from each.”
By training a deep learning network off of previous observance data, the team is able to generate updated models in seconds which previously took upwards of 15 – 30 minutes of processing. Leveraging this new efficiency the team has begun aligning data from two different satellites, one covering a smaller observational surface area with higher resolution; one encompasses the entire view of the sun but with lower resolution. By pairing these two massive data sets Fouhey’s research team is now able to holistically produce more accurate magnetogram imaging for the entire sun which neither satellite could accomplish independently.
“Rather than just improving the time scale of producing an existing data product we synthesized a brand new data product which is more effective in many ways because it was informed by a greater spectral resolution than previously possible.” says Richard Higgins, a PhD. candidate in Computer Science and Engineering working with Fouhey.
These much improved magnetic imaging maps will enable a greater understanding of the magnetic field on the sun. For example, they can reveal potential solar flares or coronal mass ejections, which are two types of events that can lead to dangerous space weather that could seriously impact things on or surrounding the Earth such as telecommunications or the astronauts on the ISS.
Dr. Fouhey considers the success of this project to be in part thanks to the MIDAS pilot grant. “[Starting] new interdisciplinary work can be tricky because of the costs of moving into new fields, asking collaborators to explain the problems they are facing and trying to solve the problem takes a lot of time and effort.”
“The PODS program helped us start the research process and allowed us to get preliminary results to begin attracting more interest. What we have now has enough potential to kickstart at least 3 – 4 grants and is paving the way to show that working with computer vision and data science to astronomy can bring you these really cool results.”