Magnetopause Detection under Low Solar Wind Density Based on Deep Learning

Round 1

Reviewer 1 Report

============================================================

Summary

============================================================

In this work, the authors show a machine learning application in which by using deep learning architecture they segment soft X-ray imaging (SXI) simulated data and use it to obtain the location and shape of the magnetopause. They use as ground truth an MHD simulated model of variable solar wind density. They conclude that their approach is significantly better than the traditional methods used.

============================================================

Recommendation

============================================================

Overall, the work is particularly interesting both in terms of result and on the future application regarding the upcoming joint mission of ESA and Chinese Academy of Sciences mission, SMILE. The application is straightforward, and the methodology and results are generally clearly stated. However, there are a few issues regarding the representation and the readability of the article. With a few adaptations, primarily to the figure, to the data/code availability and to some discussion points, this article will be ready for publication.

Recommendation: Moderate Revision

============================================================

 Moderate Concerns

============================================================

Open research & code/data availability:

You wrote on the Data Availability Statement that it is “Not applicable”. I think you have both data and potentially a code that would be very beneficial to the community to share. This will not only help to the reproducibility and validation of the result, but also to future work of other research groups that could potentially use your methodology and adopt it to their work. Please consider adding the dataset to some standard .csv file and the code/training as well. This can be done either through some repository, dataset site (e.g., zenodo) or sites like Kaggle.

Figure adaptations: In general, the figures require heavy optimization to make them readable and impactful.

Figure 1: Please add labels, x and y-axis, units. Furthermore, make the whole figure readable, now it’s not clear what is shown there. Crucial is also to use the same color bar, now the color scale changes on each figure, making a comparison impossible. Make the figure also larger, and the numbers of the color bar bigger. Finally, (d) says 20 cm^(-3) while it should be 30.

Figure 2 – 4: Same as figure 1.

Figure 5: Since all these are the same as written in line 244, consider adding one and mention that they are the same and the reasoning in the text. Furthermore, some similar issues as Figures 1-4.

Figure 6-15: Similar issues are shown here. Make things larger, clearer and most importantly, save all figures as either .pdf or .eps to be in vector format so that the quality doesn’t decrease when they are in the manuscript.

Please also include that the distances and the MP are given in Earth radius (Re) in most figures. Overall, please take care of the figures, they are really decreasing the readability and impact of your article at this point, which is a shame since the application itself is particularly interesting.  

============================================================

Minor Concerns

============================================================

Changes of model and other existing libraries:

The changes you made to [1] are particularly interesting, and it is surprising that you obtain much better results. Nevertheless, it has come to my attention that there are a few recent libraries out that could potentially be useful for this or future research on a similar note to the one you are working on. In particular, https://github.com/UX-Decoder/Segment-Everything-Everywhere-All-At-Once [2] and https://github.com/facebookresearch/segment-anything [3].

Do you think that these models could be useful for your work?

Lines 140 – 144: The explanation here is not clear. Later, you explain in more detail (Lines 284-293), which makes things clearer. Consider rephrasing the part in the introduction to a clearer text with more details or omitting it, as currently it’s difficult to follow.

Line 161: Southward IMF of Bz = -5 nT seems a bit extreme. Is there a specific reason such upstream value was used? It would be interesting to check the accuracy of variable IMF conditions that may also affect the MP shape/distance. Furthermore, you provide the thermal pressure for the solar wind, does this mean that the temperature/thermal speed of the SW used changes depending on the density to have a constant thermal pressure? I was unable to find a clear answer in the text on that.

Train/test split and section 3: Consider adding more information on how the train/validation/test split was done. Is it a random split? Also, make sure that it is clear whether the result shown in the figures are obtained from the test set.

Line 34-36 & section 4 (discussion):  This is more a question/discussion point to the authors.

It would be very interesting to consider in the future using a similar technique for another task. Upstream of the magnetopause, there is another important boundary layer surface, the bow shock. At the bow shock and downstream of it, there are known large-scale structures, named high-speed plasma jets (see e.g., [4,5]). Both the bow shock of the Earth and the large-scale jets can accelerate particles and potentially produce a signal observable by X-rays. Probably significantly weaker than the magnetopause, but still present.

Do you think it would be possible to:

(a)    Track the bow shock movement and location change using similar segment methods? This will be particularly useful as the bow shock is dynamically evolving and the determination of its location even with integration times of ~30 seconds would be extremely useful for future space weather applications.

(b)    Would it be possible to capture large-density high-speed flows (e.g., [5]) that travel through the magnetosheath by using SMILE and your technique? It is a well-known problem in the magnetospheric community that we are unable to have a clear overview of how frequently these and similar structures occur throughout the magnetosheath. An estimation of their presence would also be particularly useful to evaluate their effect in the inner magnetosphere environment.

Section 4 (discussion):

Another interesting adaptation of your work would be to have simulations that produce a more accurate shape/position of the magnetopause. These simulations could include global Hybrid or Particle-in-cell codes that consist of a more accurate description of the magnetospheric environment (see e.g., article [6] and future talk on [7]). This would be an even better comparison as it will establish a more accurate ground truth to compare with.

============================================================

Edits/Typos

============================================================

Line 92, 258 ++: For some reason there are some rendering issues around the manuscript with the text appearing different. Check if this is something on your end from the template used.  

Line 262: Verification Validation

Lines 411-416: I think the syntax here is off, making it hard to read. Consider rephrasing to a more natural tone.

============================================================References

============================================================

[1] Chen, L. C., Papandreou, G., Kokkinos, I., Murphy, K., & Yuille, A. L. (2017). Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs. IEEE transactions on pattern analysis and machine intelligence, 40(4), 834-848.

[2] Zou, X., Yang, J., Zhang, H., Li, F., Li, L., Gao, J., & Lee, Y. J. (2023). Segment Everything Everywhere All at Once. arXiv preprint arXiv:2304.06718.

[3] Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., ... & Girshick, R. (2023). Segment anything. arXiv preprint arXiv:2304.02643.

[4] Raptis, S., Aminalragia-Giamini, S., Karlsson, T., & Lindberg, M. (2020). Classification of magnetosheath jets using neural networks and High Resolution OMNI (HRO) Data. Frontiers in Astronomy and Space Sciences, 7, 24.

[5] Raptis, S., Karlsson, T., Vaivads, A., Pollock, C., Plaschke, F., Johlander, A., ... & Lindqvist, P. A. (2022). Downstream high-speed plasma jet generation as a direct consequence of shock reformation. Nature communications, 13(1), 598.

[6] Ala‐Lahti, M., Pulkkinen, T. I., Pfau‐Kempf, Y., Grandin, M., & Palmroth, M. (2022). Energy Flux Through the Magnetopause During Flux Transfer Events in Hybrid‐Vlasov 2D Simulations. Geophysical Research Letters, 49(19), e2022GL100079.

[7] Guo, J., Sun, T., Lu, S., Lu, Q., Lin, Y., Wang, X., Huang, K., and Wang, R.: Soft X-ray Imaging of Earth’s Magnetopause under Different Solar Wind Conditions: Three-Dimensional Global Hybrid Simulations, EGU General Assembly 2023, Vienna, Austria, 24–28 Apr 2023, EGU23-6104, https://doi.org/10.5194/egusphere-egu23-6104, 2023.

 

 

See the overall comments/suggestions above regarding potential edits. Furthermore, consider revisiting once your manuscript to remove some repetition that takes place. 

Author Response

Dear reviewer, we would like to express our deepest appreciation to you for your time and comments. After getting the comments, our research group discussed the solutions to each comment and tried our best to make improvement on the paper. We have responded to concerns one by one in the following.

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Review of the manuscript titled “Magnetopause Detection under Low Solar Wind Density Based  on Deep Learning” by  Yujie Zhang, Tianran Sun, Wenlong Niu, Yihong Guo, Song Yang, Xiaodong Peng and Zhen Yang

 

  The manuscript investigated the interesting problem connected with X-rays emission which result from the interaction between heavy solar wind ions and exospheric neutrals. It is a relative new point in the magnetospheric physics. Using some a soft X-ray detector, we can produce two-dimensional images of the X-rays around the magnetosphere and understand a shape of the magnetopause.

 

 But my conclusion is that manuscript can not be published in any similar presented form and must be resubmitted after major revision.

 

For unclear for me reasons authors did not included in the list of references two papers by Samsonov et al., JGR, 2022 which studied  a similar problem connected with incoming mission SMILE:

1.     Samsonov, A., Sembay, S., Read, A., Carter, J. A., Branduardi-Raymont, G., Sibeck, D., & Escoubet, P. (2022). Finding magnetopause standoff distance using a Soft X-ray Imager: 2. Methods to analyze 2-D X-ray images. Journal of Geophysical Research: Space Physics, 127, e2022JA030850. https://doi.org/10.1029/2022JA030850; 

2.     Samsonov, A., Carter, J. A., Read, A., Sembay, S., Branduardi-Raymont, G., Sibeck, D., & Escoubet, P. (2022). Finding magnetopause standoff distance using a soft X-ray imager: 1. Magnetospheric masking. Journal of Geophysical Research: Space Physics, 127, e2022JA030848. https://doi.org/10.1029/2022JA030848

 

My main doubt concern to possibility to publish manuscript in any scientific journal is that it did not included any information about some real existing object. It started not from real Earth’s magnetosphere magnetopause but from magnetopause MHD model (it is not real object but some simulation results) and finished by some reconstruction of virtual X ray flux which can be detected by virtual detector. Maybe it is interesting problem, but how is it connected with the magnetopause of some solar system planet magnetosphere?

 

Connected with this point but other reason to be a pessimist to recommend this manuscript to publishing in the journal is that manuscript not included any discussion current status of the SMILE mission. I propose that it is important for potential readers it will be launched or not. But anyway, I delegate these questions to editor who understand the journal approach better. 

 

I propose that in a revised version of the manuscript in particular Samsonov’s papers must be discussed and authors must clear demonstrated what is new in their study and why it will be interested for future readers. 

 

Let me go to more partial comments.

 

1.     I propose that structure and thiсkness of the magnetopause is important for X ray emissions by charge exchange solar wind ions and geocorona neutral. Observational signal can be strong different for smooth MHD surface and real magnetopause current which scale is electron Larmour radius as demonstrated by fine time resolution MMS data. Why authors did not used any information about real magnetopause I did not known.

 

2.     We well know that we have a two cusp shape deformations at north and south magnetopause. Authors did not discuss this point.

 

3.      It is not clear for me what kind of conclusions can be made after comparison of modelled subsolar magnetopause distance with observational coordinates some spacecrart magnetopause crossing points.

 

4.     I saw very little differences between 4 figure with different solar wind density. I propose that it is enough to be limited by one of them.

Comments for author File: Comments.pdf

Author Response

Dear reviewer, we would like to express our deepest appreciation to you for your time and comments. After getting the comments, our research group discussed the solutions to each comment and tried our best to make improvement on the paper. We have responded to concerns one by one in the following.

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

I am fine with your replies to my comments. Most of them have been accepted in the revised version&

For me your English is enough.