Skip to main content
SearchLogin or Signup

Inferring Depth-dependent Plasma Motions from Surface Observations using Deep Learning

Published onAug 18, 2020
Inferring Depth-dependent Plasma Motions from Surface Observations using Deep Learning

Coverage of plasma motions is limited to the line-of-sight component at the Sun’s surface. Multiple tracking and inversion methods were developed to infer the transverse motions from observational data. Recently, the fully convolutional DeepVel (Asensio Ramos et al., 2017) and DeepVelU (Tremblay & Attié, 2020) neural networks were used to recover the transverse velocity vector at the surface, in the chromosphere and the upper convection zone from surface observations. Using computations from a detailed magnetohydrodynamics simulation as a reference, DeepVel and DeepVelU are trained to map a combination of intensitygrams, magnetograms and/or Dopplergrams to the depth-dependent horizontal component of the plasma motions. Both neural networks differ in the mapping function they learn. We use observations as input in DeepVel and DeepVelU to generate instantaneous depth-dependent synthetic velocity vectors inside the Quiet Sun and sunspots, i.e. plasma motions that emulate the physics of a numerical model but are made to look as though they were observed by a specific instrument. These could be combined with other observations or reconstructions to infer quantities such as the electric field or the Poynting flux, or be used as boundary conditions from which to drive MHD simulations of the solar atmosphere.


Comments
0
comment

No comments here