A generated solar image becomes a search query

A generated image can become a request to a solar archive: find an observed magnetic region with a similar configuration. A 2026 study demonstrates that sequence by linking an image generator to physically meaningful controls and an image-retrieval model.[1]

The real observations come from Space-weather HMI Active Region Patches, or SHARPs, derived from NASA’s Solar Dynamics Observatory. Each patch tracks a concentration of magnetic activity through its visible lifetime, with measurements every 12 minutes. Alongside magnetic-field maps, the original SHARP pipeline calculates quantities including total magnetic flux and spatial field gradients. It therefore provides both a pattern to examine and numerical descriptions of the magnetic region.[2]

Turning a property into a query

The authors trained on 49,552 maps from 1,839 distinct patches observed between May 2010 and July 2023. A generative adversarial network creates images. Support-vector machines identify directions in its internal coordinates that change summed field strength or separation between magnetic polarities. A third model retrieves similar archive images.[1]

For 1,000 generated queries, the unsigned-field sums of queries and retrieved observations had a Pearson correlation of 0.78; polarity separation correlated at 0.28. Here, 1 would mean a perfect positive linear relationship. Some physical properties therefore matched more closely than others. Exact calibration of the controls to physical units remains future work.[1]

The result gives researchers a route from a hypothetical magnetic configuration to a real observation they can investigate.[1]

Report an error or suggest a correction