Sunspots, flares, filaments, prominences and coronal transient events pop up from time to time, making the Sun look colorful. These phenomena are related to the evolution of magnetic fields. At present, the solar magnetic field is generally measured indirectly by means of the Zeeman effect. 

For the filter-based magnetograph, narrow-band linear or circular polarization maps are obtained firstly, and the transform process to convert these maps into the magnetogram is called magnetic field calibration. The traditional method is the linear calibration, but it has the disadvantage of magnetic saturation. 

A joint research team led by the National Astronomical Observatories of Chinese Academy of Sciences (NAOC) proposed machine learning (ML) method for nonlinear magnetic calibration. This method could effectively overcome the magnetic saturation and infer the strong magnetic regions much better than the linear calibration method. 

The study was published in Astronomy & Astrophysics on Feb. 4. Researchers from the Yunnan Observatories of the Chinese Academy of Sciences and Big Bear Solar Observatory (BBSO) of the U.S.A. were also involved in the study. 

"The success of this research depends on the powerful nonlinear approximation capability of ML," said Prof. JI Kaifan from Yunnan Observatories, the corresponding author of the study. "It is a new tool for the astronomical data processing." 

The produced magnetogram from ML method are much cleaner than the target maps, indicating the methods has the function of denoising. 

"More accurate magnetogram are expected to be provided for the scientific research with the new method," said Dr. BAI Xianyong from NAOC, one of the co-authors of the study. 

"This work has made contribution to the Full-disk Vector Magnetograph onboard ASO-S, the first solar observation satellite in China. And it also provides useful reference for data processing of other solar magnetic field telescopes," said Prof. DENG Yuanyong from NAOC.