Inferring Plasma Flows at Granular and Supergranular Scales With a New Architecture for the DeepVel Neural Network

Tremblay, Benoit; Attié, Raphaël

doi:10.3389/fspas.2020.00025

Cited by 10 publications

(9 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…where v and v 0 represent the estimated and simulated horizontal velocity vectors, respectively. This definition is consistent withTremblay & Attie (2020) andTremblay et al (2021b). This dot product indicates the difference of the orientation of the vectors.…”

supporting

confidence: 90%

“…The results of DeepVel and LCT are similar when they are averaged, however, DeepVel still has the advantage of reproducing the kinetic power spectra on sub-supergranule scales. Tremblay & Attie (2020) developed a new architecture for DeepVel using the U-NET architecture and found that it is more effective than other tracking methods. However, their accuracies were not verified at various spatial scales.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Multi-Scale Deep Learning for Estimating Horizontal Velocity Fields on the Solar Surface

Ishikawa,

Nakata,

Katsukawa

et al. 2021

Preprint

View full text Add to dashboard Cite

Context. The dynamics in the photosphere is governed by the multi-scale turbulent convection termed as granulation and supergranulation. It is important to derive three-dimensional velocity vectors to understand the nature of the turbulent convection and to evaluate the vertical Poynting flux toward the upper atmosphere. The line-of-sight component of the velocity can be obtained by observing the Doppler shifts. However, it is difficult to obtain the velocity component perpendicular to the line-of-sight, which corresponds to the horizontal velocity in disk center observations. Aims. To develop a new method based on a deep neural network that can estimate the horizontal velocity from the spatial and temporal variations of the intensity and vertical velocity and to suggest a new measure for examining the performance of the method. Methods. We developed a convolutional neural network model with a multi-scale deep learning architecture. The method consists of multiple convolutional kernels with various sizes of the receptive fields, and it performs convolution for spatial and temporal axes. The network is trained with data from three different numerical simulations of turbulent convection. Furthermore, we introduced a novel coherence spectrum to assess the horizontal velocity fields that were derived at each spatial scale.Results. The multi-scale deep learning method successfully predicts the horizontal velocities for each convection simulation in terms of the global-correlation-coefficient, which is often used for evaluating the prediction accuracy of the methods. The coherence spectrum reveals the strong dependence of the correlation coefficients on the spatial scales. Although coherence spectra are higher than 0.9 for large-scale structures, they drastically decrease to less than 0.3 for small-scale structures wherein the global-correlation-coefficient indicates a high value of approximately 0.95. By comparing the results of the three convection simulations, we determined that this decrease in the coherence spectrum occurs around the energy injection scales, which are characterized by the peak of the power spectra of the vertical velocities. Conclusions. The accuracy for the small-scale structures is not guaranteed solely by the global-correlation-coefficient. To improve the accuracy in small-scales, it is important to improve the loss function for enhancing the small-scale structures and to utilize other physical quantities related to the non-linear cascade of convective eddies as input data.

show abstract

supporting

confidence: 90%

Section: Introductionmentioning

confidence: 99%

Multi-Scale Deep Learning for Estimating Horizontal Velocity Fields on the Solar Surface

Ishikawa,

Nakata,

Katsukawa

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…but also scales up the size of the sub-images presented during training (DeepVelU: Tremblay and Attie, 2020;Tremblay et al, 2019). Our recent efforts have also focused on modifying the inputs of the neural network in order to accept a combination of consecutive intensitygrams (i.e., the default inputs), magnetograms and Dopplergrams (Tremblay and Attie, 2020). This is motivated by the magnetic induction equation which relates transverse plasma motions to the magnetic field and line-of-sight plasma motions.…”

Section: Discussionmentioning

confidence: 99%

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

Tremblay

Cossette

Kazachenko

et al. 2021

J. Space Weather Space Clim.

Self Cite

View full text Add to dashboard Cite

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

show abstract

“…The results of DeepVel and LCT are similar when they are averaged; however, DeepVel still has the advantage of reproducing the kinetic power spectra on sub-supergranule scales. Tremblay & Attie (2020) developed a new architecture for DeepVel using the U-NET architecture and found that it is more effective than other tracking methods. However, their accuracies have not been verified at various spatial scales.…”

Section: Introductionmentioning

confidence: 99%

Multi-scale deep learning for estimating horizontal velocity fields on the solar surface

Ishikawa

Nakata

Katsukawa

et al. 2022

A&A

View full text Add to dashboard Cite

Context. The dynamics in the photosphere is governed by the multi-scale turbulent convection termed as granulation and supergranulation. It is important to derive three-dimensional velocity vectors to understand the nature of the turbulent convection and to evaluate the vertical Poynting flux toward the upper atmosphere. The line-of-sight component of the velocity can be obtained by observing the Doppler shifts. However, it is difficult to obtain the velocity component perpendicular to the line of sight, which corresponds to the horizontal velocity in disk center observations. Aims. We present a new method based on a deep neural network that can estimate the horizontal velocity from the spatial and temporal variations of the intensity and vertical velocity. We suggest a new measure for examining the performance of the method. Methods. We developed a convolutional neural network model with a multi-scale deep learning architecture. The method consists of multiple convolutional kernels with various sizes of receptive fields, and performs convolution for spatial and temporal axes. The network is trained with data from three different numerical simulations of turbulent convection. Furthermore, we introduced a novel coherence spectrum to assess the horizontal velocity fields that were derived for each spatial scale. Results. The multi-scale deep learning method successfully predicts the horizontal velocities for each convection simulation in terms of the global correlation coefficient, which is often used to evaluate the prediction accuracy of the methods. The coherence spectrum reveals the strong dependence of the correlation coefficients on the spatial scales. Although the coherence spectra are higher than 0.9 for large-scale structures, they drastically decrease to less than 0.3 for small-scale structures, wherein the global correlation coefficient indicates a high value of approximately 0.95. By comparing the results of the three convection simulations, we determined that this decrease in the coherence spectrum occurs around the energy injection scales, which are characterized by the peak of the power spectra of the vertical velocities. Conclusions. The accuracy for the small-scale structures is not guaranteed solely by the global correlation coefficient. To improve the accuracy on small scales, it is important to improve the loss function for enhancing the small-scale structures and to utilize other physical quantities related to the nonlinear cascade of convective eddies as input data.

show abstract

Inferring Plasma Flows at Granular and Supergranular Scales With a New Architecture for the DeepVel Neural Network

Cited by 10 publications

References 34 publications

Multi-Scale Deep Learning for Estimating Horizontal Velocity Fields on the Solar Surface

Multi-Scale Deep Learning for Estimating Horizontal Velocity Fields on the Solar Surface

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

Multi-scale deep learning for estimating horizontal velocity fields on the solar surface

Contact Info

Product

Resources

About