Inferring depth-dependent plasma motions from surface observations using the DeepVel neural network

Coverage of plasma motions is limited to the line-of-sight component at Sun’s surface. Multiple tracking and inversion methods were developed infer transverse from observational data. Recently, DeepVel neural network was trained with computations performed by numerical simulations solar photosphere recover missing surface two additional optical depths simultaneously white light intensity in Quiet Sun. We argue that deep learning could provide spatial coverage existing observations form depth-dependent synthetic observations, i.e. estimates generated through emulation simulations. different versions using slices both Sun Active Region various geometrical atmosphere, upper convection zone establish lower limits which can generate reliable intensitygrams. Flow fields inferred low chromosphere ? ? [0.1, 1) are comparable inversions ( ? deemed be suitable for use as data assimilation processes data-driven This limit extends closer transition region 0.01) Sun, but not Regions. Subsurface flows intensitygrams fail capture small-scale features turbulent convective depth crosses a few hundred kilometers. suggest these reconstructions used first model’s velocity vector nowcast forecast short term activity space weather.