Authors: Ruizhu Chen (Stanford University), Junwei Zhao (Stanford University), Shea A. Hess Webber (Stanford)

The Sun’s far-side magnetic field is important for space weather forecasting and solar wind modeling, but currently it is not directly observed. Helioseismic far-side imaging can produce far-side acoustic maps from near-side Doppler observations continuously and in near-real-time, showing most of the medium- and large-size active regions. But these acoustic maps are given in acoustic-wave travel-time shifts, which cannot be directly used in space weather applications. Therefore, in this work we present a combined deep-learning approach to calibrate the far-side acoustic maps into far-side magnetic flux. For a limited time of about 4 years, the extreme ultraviolet (EUV) flux of the Sun’s far side is observed by STEREO, from which proxies of far-side magnetic flux can be generated using an EUV-to-magnetic-flux relation. We first train such an EUV-to-magnetic-flux relation using 8 years of near-side observations, including EUV images by SDO/AIA and magnetic-flux images by SDO/HMI. This machine-learned relation is then applied to STEREO EUV images, with necessary cross-instrument calibration, to generate far-side magnetic-flux proxies. The magnetic-flux results are further paired with far-side acoustic images to train an acoustic-to-magnetic-flux relation by a second deep-learning training. The final machine-learned relation can be used to calibrate future far-side acoustic maps to produce near-real-time far-side magnetic flux maps without the need of direct far-side observations.