Fast construction of 3-D solar coronal density distribution based on MAS simulation by deep learning

Magnetohydrodynamic (MHD) simulation provides a powerful approach to study the dynamics of the solar corona. However, this simulation process is quite complicated and requires a lot of computing resource and time. In this study, we consider a well-known deep learning model Pix2PixHD to generate 3D coronal electron number density distribution from photospheric solar magnetic fields. For this we consider 2D synoptic map in magnetic field as an input and coronal electron density for given solar radii as an output, which was simulated with the MHD Algorithm outside a Sphere (MAS) model. 4272 pairs of inputs and outputs are used for training and testing from 2010 June to 2020 May. For this work we train 21 deep learning models to cover from 2 to 30 solar radii. We find that the generated 3D electron densities are quite consistent with those of the simulated one at not only lower soar radii but also higher radii: very good mean correlation coefficient (0.95) and excellent mean Structure Similarity Index (SSIM) value (0.99). It is noted that the computing time of solar coronal density distribution from 2 to 30 solar radii by our deep learning models is about 35 secs under NVIDIA TITAN XP GPU, which is much less than a typical simulation time of MAS. The generated coronal density distribution can be used for space weather models on real-time basis. This work was supported by Institute for Information & communications Technology Promotion (IITP) grant funded by the Korea government (MSIP) (2018-0-01422, Study on analysis and prediction technique of solar flares).