The Universe

Predicting the Appearance of Sunspots on the Solar Surface Using Machine Learning

Image from animation showing forecast of the upcoming solar activity for active region AR11158
Animation of data flow that enables a forecast of the upcoming solar activity for active region AR11158, which was tracked throughout its evolution. On the left, the Solar Dynamics Observatory's Atmospheric Imaging Assembly (AIA) observations with an overplotted 9 by 9 grid shows area targeted for dopplergram analysis (insert at the bottom). On the right, data analysis of the target area from top to bottom: the continuum intensity, magnetic field, and acoustic power maps using dopplergrams for four ranges (2-3, 3-4, 4-5, and 5-6 millihertz). Each column shows the raw and tiled data and mean variations of the corresponding data within a tile. The pink curves correspond to the middle tile. Spiridon Kasapis, Nina McCurdy, Timothy Sandstrom, NASA/Ames

Predicting solar storms is becoming increasingly important to mitigate its impact on space exploration, near-Earth activities, and human health. To do so, we need to identify particular areas on the solar surface where large sunspots will form and can be a birthplace for strong, eruptive activity. Scientists at NASA’s Ames Research Center aim to provide a machine learning-driven technology to enable an early warning system for these space storms by analyzing the time evolution of turbulent convection on the Sun’s surface. The team uses observations of magnetic fields, continuum intensity, and dopplergrams obtained from the Helioseismic and Magnetic Imager (HMI) onboard the Solar Dynamics Observatory (SDO). 30-by-30-degree patches were tracked throughout the solar rotation to investigate how the dynamic and physical properties of these regions evolve before, during, and after sunspot formation. To identify these acoustic precursors to sunspots, the team is employing machine learning methods to analyze the dynamics of the turbulent motions on the Sun’s surface and automatically recognize changes in oscillations associated with an emerging flux from the solar interior, which drives the formation of sunspots. This research could pave the way for early forecasts of upcoming activity to identify locations for targeted analysis, helping provide a timely response to potential hazards.

Quick Facts

Pre-processing nine terabytes worth of solar observation data, while at the same time training different machine learning models on them, would never be possible without the NASA supercomputing resources.

Spiridon Kasapis,
NASA Ames Research Center
  • NASA’s Pleiades supercomputer at the NASA Advanced Supercomputing facility was used to process observations from SDO/HMI and run the machine learning models to predict the appearance of sunspots on the solar surface more efficiently.
  • Although this project is in a preliminary stage, it has already proven that machine learning can pick up patterns in solar acoustic power maps (processed observations from SDO/HMI) that humans cannot.
  • The resulting acoustic power maps dataset will be made publicly available, allowing for the investigation of a variety of solar phenomena, such as the evolution and dynamics of highly magnetized plasma and the self-organization processes that lead magnetic field configurations into pre-eruptive conditions.
  • This project will enable development of new robust machine learning-driven models to predict sunspot formation before it becomes visible on the solar surface. The team’s next step is to train a variety of machine learning methods with the entirety of the data they have produced.

Researchers

  • Spiridon Kasapis, NASA Ames Research Center
  • Irina Kitiashvili, NASA Ames Research Center

More Information