Indian astronomers have employed Artificial Intelligence (AI) to trace the shift in magnetically active patches on the sun from 1916 to 2007 by scanning hand-drawn sun records of the past 100 year from the Kodaikanal Solar Observatory (KSO). This would lead to better understand long-term space weather risks that can affect technology on earth.
Long-term, consistent records of the sun’s magnetic activity are crucial because they let scientists compare how different solar cycles vary in strength and structure and improve reconstructions of how the sun’s energy output and magnetic influence have changed in the past, according to information shared by the Ministry of Science and Technology on Wednesday.
For more than a hundred years, scientists have been trying to understand how the sun’s magnetic activity rises and falls in rhythmic cycles. These cycles affect sunspots, flares, and eruptions, which can disrupt satellites, navigation, and power on earth. However, older observations are often incomplete and inconsistent, making long-term study difficult. That’s why historical records are very valuable, the Ministry said.
The new study, undertaken by researchers from the Aryabhatta Research Institute of Observational Sciences (ARIES), along with collaborators from the Indian Institute of Space Science and Technology, Thiruvananthapuram; Southwest Research Institute, USA; and Indian Institute of Astrophysics, Bangalore; has been published in The Astrophysical Journal.
The work, led by Dibya Kirti Mishra, shows that 100 years of hand-drawn sun records from the KSO can be turned into useful data using modern machine learning techniques. The observatory has a unique collection of observations, including daily ‘suncharts’ from 1904 to 2022, where features like sunspots, plages, filaments, and prominences were carefully drawn on a standard grid.
Before digital tools, scientists relied on careful drawings to record what they saw. KSO’s suncharts are valuable because they show solar activity over many cycles and include different features marked in specific ways. However, differences in drawing styles, paper aging and scan quality make it difficult to create a clean and consistent dataset using traditional methods.
To address the problem of messy, hand-drawn historical records, the research team used a supervised machine learning approach, U-Net, in two main steps. First, the model automatically found the sun’s disk in each scanned drawing, pinpointing the centre, size and tilt, so every feature could be placed in the correct location on the sun.
Next, it identified and traced magnetically active patches on the sun across drawings covering nine solar cycles from 1916 to 2007. This is important because such patches, called plages, are a reliable “fingerprint” of the sun’s magnetism and extracting them from old archives helps scientists connect today’s space-age measurements with what the sun was doing decades earlier.
By turning drawings into machine-readable data, the researchers were able to track how plage activity shifts over time, creating a ‘butterfly diagram’ that shows the solar cycle, the study said. They also found that the plage areas from these drawings match well with those derived from KSO’s full-disk observations, proving that the sun charts can help fill gaps and improve long-term solar data.