Artificial intelligence has been used to analyse more than a century of solar observations from the Kodaikanal Solar Observatory (KoSO) in India, converting historical hand-drawn records into structured scientific data. The study tracks magnetically active regions on the Sun from 1916 to 2007 and provides a long-term view of solar activity that influences space weather and satellite systems on Earth, according to the Department of Science and Technology (DST), Government of India.

Machine Learning Converts Hand-Drawn Solar Records

The research, led by scientists from the Aryabhatta Research Institute of Observational Sciences (ARIES) with collaboration from the Indian Institute of Space Science and Technology and the Indian Institute of Astrophysics, used a machine learning model based on a U-Net architecture. The model processed scanned “suncharts” recorded at KoSO, which contain daily drawings of sunspots, plages, filaments, and prominences.

These records span more than a century and were originally created through manual observation. Researchers first trained the system to identify the solar disc, including its position, size, and orientation in each image. It then extracted plages, which are bright regions linked to solar magnetic activity. The study covered nine solar cycles between 1916 and 2007, covering Solar Cycles 15 to 23.

Butterfly Diagram Maps Solar Magnetic Activity

The processed data was used to create time–latitude “butterfly diagrams”, which show how solar activity shifts over time. These diagrams illustrate how active regions on the Sun move toward the equator as each solar cycle progresses. Scientists compared the AI-generated results with Ca II K spectroheliograph observations from KoSO and found strong agreement, strengthening confidence in the reconstructed dataset.

The findings were published in The Astrophysical Journal. Researchers noted that combining multiple historical datasets helps fill gaps caused by inconsistent archival records, aging documents, and variations in early observational methods.

Long-Term Value For Space Weather And Climate Studies

Solar magnetic activity influences space weather, which can affect satellites, communication systems, navigation signals, and power grids. Long-term datasets are important for understanding solar cycle variability and improving predictive models.

According to the study team, digitising and standardising historical observations allows scientists to extend modern solar records further into the past. This helps improve reconstructions of solar energy output and supports better assessment of long-term space weather risks.

The research demonstrates how artificial intelligence can be applied to archival scientific data, transforming non-digital records into consistent datasets for modern analysis. The approach also highlights the scientific value of long-preserved observational archives such as those at KoSO.