IBM and NASA Create a Virtual Model of the Sun to Forecast Upcoming Solar Storms

The Sun’s most intricate secrets may soon be unveiled through the power of artificial intelligence. On August 20, IBM and NASA introduced Surya, a foundational model focused on the sun. Trained on extensive datasets of solar activity, this AI tool strives to enhance human understanding of solar weather and to accurately foresee solar flares—outbursts of electromagnetic radiation emitted by our star that pose risks to both astronauts in space and communication systems on Earth.

Surya was developed using nine years of data from NASA’s Solar Dynamics Observatory (SDO), a satellite that has been orbiting the sun since 2010, capturing high-resolution images every 12 seconds. The SDO gathers observations at various electromagnetic wavelengths to estimate the temperature of different layers of the sun and takes precise measurements of its magnetic field—crucial information for understanding energy movement within the star and for predicting solar storms.

Interpreting this vast and complex data has historically been a significant challenge for heliophysicists. To tackle this issue, IBM reports that Surya’s creators utilized SDO data to build a digital twin of the sun—an evolving virtual model that updates with new data input and allows for more manageable study.

The development process started by standardizing the numerous data formats that fed into the model, enabling consistent processing. Following that, a long-range vision transformer was employed—an AI architecture that facilitates detailed analysis of high-resolution images and identifies relationships among their components, irrespective of distance.

The model’s effectiveness was enhanced using a technique called spectral gating, which cuts memory usage by up to 5 percent by eliminating noise in the data, thus improving the quality of processed information.

More Accurate Predictions in Less Time

Its developers claim that this design affords Surya a considerable edge: Unlike other algorithms needing extensive data labeling, Surya can learn directly from unprocessed data. This ability enables rapid adaptation to varying tasks and yields reliable outcomes in a shorter timeframe.

In testing, Surya showcased its ability to integrate data from additional instruments, including the Parker Solar Probe and the Solar and Heliospheric Observatory (SOHO), both of which monitor the sun. Surya also proved effective in various predictive applications, such as estimating flare activity and solar wind velocity.

According to IBM, conventional prediction models can forecast a flare only one hour in advance based on signals from specific regions of the sun. In contrast, “Surya provided a two-hour advantage by utilizing visual information. The model is considered the first to offer such advanced warnings. In initial tests, the team reported a 16 percent increase in solar flare classification accuracy, a notable improvement over current techniques,” the company stated.

NASA emphasizes that while the model is tailored for the study of heliophysics, its architecture is flexible enough to be applied in various fields, from planetary science to Earth observation. “By creating a foundational model trained on NASA’s heliophysics data, we’re streamlining the analysis of the sun’s behavior complexities with unprecedented speed and accuracy,” noted Kevin Murphy, NASA’s director of data science, in a statement. “This model fosters a greater understanding of how solar activity affects essential systems and technologies we rely on here on Earth.”

The threats posed by unusual solar activity are significant. A severe solar storm could disrupt global communications, damage electrical grids, and interfere with GPS navigation, satellite operations, internet connectivity, and radio communications.

Andrés Muñoz-Jaramillo, a solar physicist at the Southwest Research Institute in San Antonio, Texas, and the project’s lead scientist, stressed that Surya’s aim is to maximize the lead time for potential adverse scenarios. “We aspire to provide Earth with the longest possible advance notice. Our goal is for the model to have assimilated all critical processes underlying our star’s evolution over time so we can derive actionable insights.”

This story originally appeared on WIRED en Español and has been translated from Spanish.