Improved detection of far-side solar active regions using deep learning

Context. The analysis of waves on the visible side of the Sun allows the detection of active regions on the far side through local helioseismology techniques. Knowing the magnetism in the whole Sun, including the non-visible hemisphere, is fundamental for several space weather forecasting applications. Aims. Seismic identification of far-side active regions is challenged by the reduced signal-to-noise ratio, and only large and strong active regions can be reliable detected. Here we develop a new method to improve the identification of active region signatures in far-side seismic maps. Methods. We constructed a deep neural network that associates the far-side seismic maps obtained from helioseismic holography with the probability that active regions lie on the far side. The network was trained with pairs of helioseismic phase-shift maps and Helioseismic and Magnetic Imager (HMI) magnetograms acquired half a solar rotation later, which were used as a proxy for the presence of active regions on the far side. The method was validated using a set of artificial data, and it was also applied to actual solar observations during the period of minimum activity of solar cycle 24. Results. Our approach shows a higher sensitivity to the presence of far-side active regions than standard methods that have been applied up to date. The neural network can significantly increase the number of detected far-side active regions, and will potentially improve the application of far-side seismology to space weather forecasting.