Articles | Volume 25, issue 18
https://doi.org/10.5194/acp-25-11517-2025
© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.Global ionospheric sporadic E intensity prediction from GNSS RO using a novel stacking machine learning method incorporated with physical observations
Download
- Final revised paper (published on 29 Sep 2025)
- Preprint (discussion started on 25 Apr 2025)
Interactive discussion
Status: closed
Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor
| : Report abuse
-
RC1: 'Comment on egusphere-2025-1549', Yosuke Yamazaki, 12 May 2025
- AC1: 'Reply on RC1', Xiaohua Xu, 04 Jul 2025
- AC2: 'Reply on RC1', Xiaohua Xu, 04 Jul 2025
-
RC2: 'Comment on egusphere-2025-1549', Anonymous Referee #2, 27 Jun 2025
- AC3: 'Reply on RC2', Xiaohua Xu, 04 Jul 2025
Peer review completion
AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload
AR by Xiaohua Xu on behalf of the Authors (04 Jul 2025)
Author's response
Author's tracked changes
Manuscript
ED: Publish as is (20 Jul 2025) by John Plane

AR by Xiaohua Xu on behalf of the Authors (23 Jul 2025)
Review on "Global ionospheric sporadic E intensity prediction from GNSS RO using a novel stacking machine learning method incorporated with physical observations"
Reviewed by Yosuke Yamazaki, Leibniz Institut of Atmospheric Physics, University of Rostock
This is a well-written and insightful paper on machine learning modeling of sporadic E (Es) layer intensity. The study is particularly novel in two key aspects: (1) the application of a stacking machine learning (SML) approach, and (2) the integration of atmospheric parameters as model inputs. The authors convincingly demonstrate that their model not only captures the salient features of Es climatology but also outperforms existing models. The topic and overall quality of the paper are well aligned with the journal's scope. Below are my comments on the manuscript. Most are minor, though a few may be considered major.
1. Abstract
The abstract does not clearly highlight the novelty of the study, as it omits key details that are essential for understanding the significance of the work. The authors should briefly state what the SML method is. I found this information later in the main text.
"SML combines the advantages from different ML models to obtain better performance than a single ML model."
Also, the use of vertical ion convergence (VIC) and gravity wave (GW) potential energy as model inputs should be explicitly stated. The present abstract only mentions "physical observations" without explaining what they are.
2. (l. 33) "vertical motion of gravity waves"
This sounds strange. Do the authors mean "vertical air motion due to gravity waves"?
3. (l. 35) "Therefore, the high precision modeling and ..."
This "Therefore" sounds out of place. The authors did not explain why "the high-precision modeling and prediction of Es layers" are crucial for space weather forecasting.
4. (l. 48) "physical observations"
It is unclear what they are. Please be specific.
5. (l. 57) "there were little data on Es-related physical mechanisms"
What are considered as "data on Es-related physical mechanisms"? Please be specific.
6. (l. 57) "On the other hand"
This sounds out of place.
7. (l. 65) "better accuracy and generalization"
I understand "better accuracy" but not "better generalization". What does the latter mean?
8. (l. 66) "the stacking strategy"
Please elaborate on what the stacking strategy means.
9. (l. 72) "the physical observations"
Again, it is unclear what they are.
10. (l. 74) "vertical ion drift (VIC)"
This should be "vertical ion convergence (VIC)".
11. (l. 78) "the meta model"
The concept of meta model should be explained.
12. (l. 79) "from different aspects"
Please clarify what "different aspects" refer to.
13. Figure 1
What are the blue lines?
14. (l. 97) "remove the profiles with negative TEC values and the bottom heights higher than 90 km"
Roughly how many percent of the data are rejected in this quality control process? Are they already removed in Figure 1?
15. Figure 2
It appears that the numbers following the station names indicate the years of measurements. If so, please clarify this explicitly.
16. (l. 278) "ground truth"
What are "Ground Truth" shown in Figure 8? Are they grid averages over many years including different seasons and local times? Please clarify.
17. (l. 310) "The clearly reconstruction of these two ..."
I suggest the authors rewrite this sentence; the wording sounds a bit awkward (e.g. "two larger EsI with different reasons" and "SML model can simultaneously consider different physical mechanisms").
18. (l. 353) "Figure 12"
State that Figure 12 represents the results for quiet periods.
19. (l. 367) "Compared to quiet days, the EsI distributions during storm days are more complex"
It is not clear if this is associated with geomagnetic storms or other factors (season, F10.7, VIC, etc.). If the authors want to demonstrate the impact of geomagnetic storms, they could run the SML model with a fixed Dst value of 0 and compare the results to those obtained using variable Dst values.
20. Figure 14
Does this include all the ionosonde data from different latitudes? The results seem to show that Smax values are sometimes much greater than those anticipated from foEs. Can the authors comment on whether data from certain ionosondes are responsible for those discrepancies?
21. Figure 15
It should be stated that in contrast to Figure 14, the model-data comparisons presented in Figure 15 do not incorporate the spatiotemporal window of (0.5°, 0.5°, 1 h).
22. Figure 15
Why are these comparisons limited to the three mid-latitude stations in the Northern Hemisphere? Since the authors claim that the model can capture Es variability caused by different mechanisms, the comparisons should be extended to include high-latitude stations.
23. Table 1
This is an interesting evaluation. However, the current approach to assessing the importance of input variables (VIC, GW, F10.7, and Dst) limits the insights that can be drawn. I suggest that the authors evaluate the contribution of each input variable individually by systematically removing them one at a time. As it stands, I am not fully convinced that the GW and Dst parameters meaningfully enhance the model's performance.
24. (l. 431) "Tian et al. (2023) conducted ..."
Could this be more specific? If Tian et al. (2023) did not include VIC, what lower atmospheric parameters did they use?
25. (l. 433) "the physical observations"
Please specify exactly what they are.
26. (l. 456) "with the improvements of 20.1%-40.5%"
State that these numbers represent the improvements in RMSE.