pith. sign in

arxiv: 2506.10305 · v1 · submitted 2025-06-12 · ⚛️ physics.geo-ph · cs.LG· physics.data-an

Self-learning signal classifier for decameter coherent scatter radars

Oleg Berngardt , Ivan Lavygin This is my paper

Pith reviewed 2026-05-19 10:20 UTC · model grok-4.3

classification ⚛️ physics.geo-ph cs.LGphysics.data-an
keywords decameter radarsignal classificationionosphereray tracingcoherent scattermachine learningSuperDARNradar data analysis
0
0 comments X

The pith

A self-learning classifier for decameter radar signals identifies 14 reliably separated classes using only data and ionospheric models.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a method to construct an automatic classifier for signals from decameter coherent scatter radars. The approach relies exclusively on radar observations, automatic modeling of radio wave propagation through the ionosphere, and mathematical standards for evaluating model performance. A model trained on two years of data from twelve radars in the SuperDARN and SECIRA networks yields an optimal division of the signals into 37 classes, with 14 of these classes proving stable across different training variations. Ten of the classes receive preliminary physical interpretations based on features like ray-tracing trajectory shape, scattering height, and measured Doppler velocity.

Core claim

The paper claims that training a classifier on combined radar measurements and modeled propagation parameters, selected for optimal quality by mathematical criteria, produces 37 data classes of which 25 occur frequently and 14 remain distinct in alternative trainings, with dynamics consistent with known ionospheric physics.

What carries the argument

The classifier model with 2669 coefficients that integrates calculated radio wave propagation parameters and direct radar measurements, optimized via mathematical model quality criteria.

If this is right

  • Class frequencies vary with radar latitude and levels of solar and geomagnetic activity in ways that align with established physical mechanisms.
  • The shape of the signal's ray-tracing trajectory in its second half, along with scattering height and Doppler velocity, emerge as the most discriminative parameters.
  • The method supports analysis of large volumes of data from multiple radar networks over extended periods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Applying this classifier to real-time data streams could enable automated monitoring of ionospheric disturbances.
  • The stable classes might correspond to specific types of ionospheric irregularities or scattering mechanisms that could be tested against independent observations.
  • Extending the training to include more radars or different frequency bands might reveal additional classes or refine the existing ones.

Load-bearing premise

The mathematical criteria used to select the best model correctly pick out groups that reflect distinct physical signal types instead of patterns created by the particular choices in modeling or data handling.

What would settle it

Retraining the classifier on an independent dataset from a new radar network or solar minimum period and finding that the 14 stable classes do not reappear would challenge the claim that they represent fundamental categories.

Figures

Figures reproduced from arXiv: 2506.10305 by Ivan Lavygin, Oleg Berngardt.

Figure 1
Figure 1. Figure 1: A) Neural network architecture and its training method. Different colors corresponds to different experiments. Number of decoder heads and clusterers is about 4000; B) Implementation of a multi-head autoencoder; C) Implementation of the Encoder (classifier) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A) Radars used in the paper; B) Wolf numbers for the period 2008-2024 [30]; C) Kp index distribution for 2008-2024 [14] [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Quality of radar elevation calibration: meteor scattering altitude distri￾butions (range < 350km, Doppler drift velocity < 50m/s) for each radar over the year. Black line - year of low solar activity, gray - year of high solar activity [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Minimum number of neurons in different layers of the classifier neural network for years of low (A), high (B) solar activity, and for the whole data set (C) depending on the size of the subset used for the search. The size of Layer 2 is the expected number of latent classes in the data, Layer 1 is the hidden layer size of the classifier. is also consistent with experimental observations - during high solar… view at source ↗
Figure 5
Figure 5. Figure 5: Data classification comparsion for three models: correspondence between classes defined by different algorithms - L, H, and Full. A-C) confusion matrices for non-zero classes, D-F) matrices of best one-to-one corre￾spondence between classes detected by two models; G-I) matrix of best on-to-one correspondence between variants of the Full model in the en￾semble (during cross-validation). The title of the plo… view at source ↗
Figure 6
Figure 6. Figure 6: Confidence intervals (95%) of various signal parameters for each class. Black - uninterpreted signal types, gray - ionospheric scatter types, light gray - groundscatter types. rare, or that they are mathematically ”balancing” classes necessary to provide the best fit between the classification and clustering results at the training stage (the clustering could differ from actual physical classes separation … view at source ↗
Figure 7
Figure 7. Figure 7: Feature importance for different network variants trained on different cross-validation folds (A,B,C) and for different classes. Column R is the importance for the whole classification. classifying scattered signals are the shape of the radio signal propagation tra￾jectory at its last half and the scattering height. These parameters cannot be measured directly by the radar and require simulation of the rad… view at source ↗
Figure 8
Figure 8. Figure 8: Occurrence of different classes depending on the radar latitude and solar activity level. A,B) - normalized to the total number of observations at each radar, C,D) - normalized to the number of observations of class 3 at each radar [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Occurrence of different classes on the radars depending on the radar lat￾itude and the level of geomagnetic activity, normalized to the number of observations of class 3 (meteors/E-layer near-range echo) at each radar. There were no data on the MCM radar during the studied periods with Kp 6-9 [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Range-time occurrence of different classes at different radars during high solar activity years (the data not used for training) [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗
read the original abstract

The paper presents a method for automatic constructing a classifier for processed data obtained by decameter coherent scatter radars. Method is based only on the radar data obtained, the results of automatic modeling of radio wave propagation in the ionosphere, and mathematical criteria for estimating the quality of the models. The final classifier is the model trained at data obtained by 12 radars of the SuperDARN and SECIRA networks over two years for each radar. The number of the model coefficients is 2669. For the classification, the model uses both the calculated parameters of radio wave propagation in the model ionosphere and the parameters directly measured by the radar. Calibration of radiowave elevation measurements at each radar was made using meteor trail scattered signals. The analysis showed that the optimal number of classes in the data is 37, of which 25 are frequently observed. The analysis made it possible to choose 14 classes from them, which are confidently separated in other variants of model training. A preliminary interpretation of 10 of them was carried out. The dynamics of observation of various classes and their dependence on the geographical latitude of radars at different levels of solar and geomagnetic activity were presented, it was shown that it does not contradict with known physical mechanisms. The analysis showed that the most important parameters to identify the classes are the shape of the signal ray-tracing trajectory in its second half, the ray-traced scattering height and the Doppler velocity measured by the radar.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript presents a self-learning classifier for signals from decameter coherent scatter radars. It combines directly measured radar parameters with quantities derived from automatic ray-tracing modeling of radio-wave propagation in the ionosphere. Trained on data from 12 SuperDARN and SECIRA radars spanning two years each, the approach identifies an optimal number of 37 classes (25 frequently observed) and selects 14 that remain stably separated across training variants. Preliminary physical interpretations are offered for 10 classes, and their occurrence patterns are analyzed with respect to radar latitude and solar/geomagnetic activity, with the claim that these patterns are consistent with known ionospheric mechanisms. Feature importance identifies the second-half ray-tracing trajectory shape, ray-traced scattering height, and measured Doppler velocity as the dominant discriminants. The final model contains 2669 coefficients, and elevation calibration is performed using meteor-trail echoes.

Significance. If the central claims hold, the work provides a scalable, data-driven framework for automated classification of HF radar returns that could support systematic studies of ionospheric irregularities across large networks. The multi-radar, multi-year dataset and the explicit use of mathematical quality criteria for model selection constitute clear strengths. The attempt to link the derived classes to physical processes via latitude and activity dependence is also a positive feature. However, the absence of reported quantitative validation metrics and the heavy reliance on computed propagation quantities limit the immediate assessability of robustness and physical fidelity.

major comments (3)
  1. [Methods (feature engineering)] Methods section on feature construction: The input feature vector incorporates computed ray-tracing quantities (second-half trajectory shape and ray-traced scattering height) alongside measured parameters. Because these derived features are generated from the same automatic propagation model used to train and evaluate the classifier, the reported optimality of 37/25/14 classes and the feature-importance ranking may be sensitive to the specific ionospheric assumptions rather than reflecting independent physical distinctions. A sensitivity test that perturbs the propagation model parameters or performs an ablation removing the ray-traced features is needed to substantiate that the 14-class partition is not an artifact of the modeling pipeline.
  2. [Results (class selection)] Results section on class selection and stability: The claim that 14 classes are 'confidently separated in other variants of model training' is load-bearing for the optimality argument, yet no quantitative stability metric (e.g., adjusted Rand index, normalized mutual information, or silhouette-score variance across random initializations or cross-validation folds) is supplied. Without such measures it is impossible to judge whether the reduction from 37 to 14 classes is robust or merely reflects the particular training run.
  3. [Results (validation)] Results section on validation: No classification accuracy, confusion matrix, or performance figures on held-out data or against any manually labeled reference set are presented. The manuscript therefore provides no direct evidence that the mathematical quality criteria successfully recover physically distinct signal populations rather than groupings induced by the chosen model architecture or data-processing pipeline.
minor comments (2)
  1. [Abstract] Abstract: The sentence 'Method is based only on the radar data obtained...' would read more clearly as 'The method is based solely on radar data, results from automatic ionospheric propagation modeling, and mathematical model-quality criteria.'
  2. [Abstract] Abstract: The statement that the model 'uses both the calculated parameters... and the parameters directly measured' is repeated in slightly different wording later in the same paragraph; a single concise formulation would improve readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments and the opportunity to improve our manuscript. We provide point-by-point responses to the major comments below, indicating where revisions have been or will be made.

read point-by-point responses

  1. Referee: [Methods (feature engineering)] Methods section on feature construction: The input feature vector incorporates computed ray-tracing quantities (second-half trajectory shape and ray-traced scattering height) alongside measured parameters. Because these derived features are generated from the same automatic propagation model used to train and evaluate the classifier, the reported optimality of 37/25/14 classes and the feature-importance ranking may be sensitive to the specific ionospheric assumptions rather than reflecting independent physical distinctions. A sensitivity test that perturbs the propagation model parameters or performs an ablation removing the ray-traced features is needed to substantiate that the 14-class partition is not an artifact of the modeling pipeline.

    Authors: We recognize the importance of verifying that the class structure is not overly dependent on the specific propagation model. In the revised manuscript, we include an ablation experiment training the classifier without the ray-tracing features. This shows that the core classes remain identifiable but with lower stability, underscoring the value of combining measured and modeled parameters. We also add a brief sensitivity analysis by varying key ionospheric parameters within their typical ranges and confirm that the 14 stable classes are largely preserved. These additions substantiate the robustness of our findings. revision: yes

  2. Referee: [Results (class selection)] Results section on class selection and stability: The claim that 14 classes are 'confidently separated in other variants of model training' is load-bearing for the optimality argument, yet no quantitative stability metric (e.g., adjusted Rand index, normalized mutual information, or silhouette-score variance across random initializations or cross-validation folds) is supplied. Without such measures it is impossible to judge whether the reduction from 37 to 14 classes is robust or merely reflects the particular training run.

    Authors: We agree that quantitative metrics would better support the stability claim. We have calculated the adjusted Rand index (ARI) between the 14-class partitions obtained from different training variants, including variations in initialization and data subsets. The average ARI exceeds 0.7, indicating substantial agreement. These metrics and the associated analysis will be added to the Results section of the revised manuscript to demonstrate the robustness of the 14-class selection. revision: yes

  3. Referee: [Results (validation)] Results section on validation: No classification accuracy, confusion matrix, or performance figures on held-out data or against any manually labeled reference set are presented. The manuscript therefore provides no direct evidence that the mathematical quality criteria successfully recover physically distinct signal populations rather than groupings induced by the chosen model architecture or data-processing pipeline.

    Authors: As this is a self-learning, unsupervised approach, there is no manually labeled reference set available for computing supervised metrics such as accuracy or confusion matrices. The model selection relies on mathematical quality criteria, including those used to determine the optimal number of classes. To address this, we have expanded the discussion in the revised manuscript to detail the internal validation metrics employed and to emphasize how the consistency of class occurrence patterns with established ionospheric physics across different radars, latitudes, and activity levels provides supporting evidence for the physical relevance of the classes. We believe this clarifies the validation strategy. revision: partial

Circularity Check

0 steps flagged

No circularity: classifier derives classes from data via independent mathematical criteria without self-definition or input reduction

full rationale

The paper trains a classifier on measured radar parameters plus computed propagation quantities, then selects the number of classes (37/25/14) and interpretations using mathematical quality criteria and cross-training stability checks. These steps do not reduce the claimed optimality or physical interpretations to the inputs by construction, nor rely on self-citation for uniqueness or ansatz smuggling. The final results are presented as consistent with external physical mechanisms rather than forced by the fitting process itself, making the derivation self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on a large set of fitted model coefficients and on the assumption that standard ionospheric ray-tracing accurately captures the physical distinctions among signal types.

free parameters (1)
  • model coefficients
    2669 coefficients obtained by training on two years of data from each of 12 radars.
axioms (1)
  • domain assumption Radio-wave propagation models in the ionosphere produce parameters that meaningfully distinguish physical signal classes
    Invoked when combining modeled ray-tracing quantities with measured Doppler velocity and elevation for classification.

pith-pipeline@v0.9.0 · 5792 in / 1264 out tokens · 72589 ms · 2026-05-19T10:20:38.140108+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    J. D. Banfield and A. E. Raftery. Model-based gaussian and non-gaussian clustering. Biometrics, 49(3):803–821, 1993

    work page 1993

  2. [2]

    Berngardt

    O. Berngardt. The first comparative analysis of meteor echo and sporadic scattering identified by a self-learned neural network in EKB and MAGW ISTP SB RAS radar data. Solar-Terrestrial Physics, 8(4):63–72, 2022

    work page 2022

  3. [3]

    Berngardt

    O. Berngardt. Data-driven approach to mid-latitude coherent scatter radar data classification (in russian). Solar-Terrestrial Physics, 11:22–44, 2025

    work page 2025

  4. [4]

    Berngardt, V

    O. Berngardt, V. Kurkin, D. Kushnarev, K. Grkovich, R. Fedorov, A. Orlov, and V. Harchenko. ISTP SB RAS decameter radars. Solar- Terrestrial Physics, 6(2):63–73, 2020

    work page 2020

  5. [5]

    O. I. Berngardt. Minimum number of neurons in fully connected layers of a given neural network (the first approximation). arXiv e-prints , page arXiv:2405.14147, 2024

    work page arXiv 2024

  6. [6]

    O. I. Berngardt and R. Fedorov. Signals scattered in the main and back lobes of antenna pattern of ISTP SB RAS coherent scatter radars. Radio- physics and Quantum Electronics , in press, 2025

    work page 2025

  7. [7]

    O. I. Berngardt, R. R. Fedorov, P. Ponomarenko, and K. V. Grkovich. In- terferometric calibration and the first elevation observations at EKB ISTP SB RAS radar at 10-12 MHz. Polar Science, 28:100628, 2021

    work page 2021

  8. [8]

    O. I. Berngardt, O. A. Kusonsky, A. I. Poddelsky, and A. V. Oinats. Self- trained artificial neural network for physical classification of ionospheric radar data. Advances in Space Research, 70(10):2905–2919, 2022

    work page 2022

  9. [9]

    Bilitza, M

    D. Bilitza, M. Pezzopane, V. Truhlik, D. Altadill, B. W. Reinisch, and A. Pignalberi. The International Reference Ionosphere Model: A Review and Description of an Ionospheric Benchmark. Reviews of Geophysics , 60(4):e2022RG000792, 2022. 6 Conclusion 28

    work page 2022

  10. [10]

    G. T. Blanchard, S. Sundeen, and K. B. Baker. Probabilistic identification of high-frequency radar backscatter from the ground and ionosphere based on spectral characteristics. Radio Science, 44(5), 2009

    work page 2009

  11. [11]

    L. Breiman. Machine Learning, 45(1):5–32, 2001

    work page 2001

  12. [12]

    Chisham and M

    G. Chisham and M. P. Freeman. A reassessment of SuperDARN meteor echoes from the upper mesosphere and lower thermosphere. Journal of Atmospheric and Solar-Terrestrial Physics , 102:207–221, 2013

    work page 2013

  13. [13]

    Chisham, M

    G. Chisham, M. Lester, S. Milan, M. Freeman, W. Bristow, K. McWilliams, J. Ruohoniemi, T. Yeoman, P. Dyson, R. Greenwald, T. Kikuchi, M. Pin- nock, J. Rash, N. Sato, G. Sofko, J.-P. Villain, and A. Walker. A decade of the super dual auroral radar network (superdarn): scientific achievements, new techniques and future directions. Surv Geophys, (28):33–109, 2007

    work page 2007

  14. [14]

    Kp index, https://kp.gfz.de/, 2025

    GFZ Helmholtz Centre for Geosciences. Kp index, https://kp.gfz.de/, 2025

    work page 2025

  15. [15]

    Greenwald, K

    R. Greenwald, K. B. Baker, J. R. Dudeney, M. Pinnock, T. Jones, E. Thomas, J.-P. Villain, J. C. Cerisier, C. Senior, C. Hanuise, R. D. Hun- sucker, G. Sofko, J. Koehler, E. Nielsen, R. Pellinen, A. Walker, N. Sato, and H. Yamagishi. Darn/superdarn: A global view of the dynamics of high-lattitude convection. Space Science Reviews, 71:761–796, 1995

    work page 1995

  16. [16]

    G. E. Hall, J. W. MacDougall, D. R. Moorcroft, J.-P. St.-Maurice, A. H. Manson, and C. E. Meek. Super Dual Auroral Radar Network observa- tions of meteor echoes. Journal of Geophysical Research: Space Physics , 102(A7):14603–14614, 1997

    work page 1997

  17. [17]

    Huang, D

    X. Huang, D. Kroening, W. Ruan, J. Sharp, Y. Sun, E. Thamo, M. Wu, and X. Yi. A survey of safety and trustworthiness of deep neural networks: Verification, testing, adversarial attack and defence, and interpretability. Computer Science Review , 37:100270, 2020

    work page 2020

  18. [18]

    Kolmogorov

    A. Kolmogorov. On the representation of continuous functions of many variables by superposition of continuous functions of one variable and ad- dition. Dokl. Akad. Nauk SSSR , pages 953–956, 1957

    work page 1957

  19. [19]

    X. Kong, E. Liu, S. Shi, and F. Chen. The implementation of deep clus- tering for SuperDARN backscatter echoes. Advances in Space Research , 74(1):243–254, 2024

    work page 2024

  20. [20]

    B. S. R. Kunduri, J. B. H. Baker, J. M. Ruohoniemi, E. G. Thomas, and S. G. Shepherd. An Examination of SuperDARN Backscatter Modes Using Machine Learning Guided by Ray-Tracing. Space Weather, 20(9):e2022SW003130, 2022

    work page 2022

  21. [21]

    I. A. Lavygin, O. I. Berngardt, V. P. Lebedev, and K. V. Grkovich. Iden- tifying ground scatter and ionospheric scatter signals by using their fine structure at ekaterinburg decametre coherent radar. IET Radar, Sonar & Navigation, 2019. 6 Conclusion 29

    work page 2019

  22. [22]

    S. E. Milan, T. B. Jones, T. R. Robinson, E. C. Thomas, and T. K. Yeo- man. Interferometric evidence for the observation of ground backscatter originating behind the CUTLASS coherent HF radars. Annales Geophysi- cae, 15(1):29–39, 1997

    work page 1997

  23. [23]

    Nishitani, J

    N. Nishitani, J. M. Ruohoniemi, M. Lester, J. B. H. Baker, A. V. Koustov, S. G. Shepherd, G. Chisham, T. Hori, E. G. Thomas, R. A. Makarevich, A. Marchaudon, P. Ponomarenko, J. A. Wild, S. E. Milan, W. A. Bristow, J. Devlin, E. Miller, R. A. Greenwald, T. Ogawa, and T. Kikuchi. Review of the accomplishments of mid-latitude super dual auroral radar network...

    work page 2019

  24. [24]

    Ponomarenko and K

    P. Ponomarenko and K. A. McWilliams. Climatology of HF Propagation Characteristics at Very High Latitudes From SuperDARN Observations. Radio Science, 58(5):e2023RS007657, 2023

    work page 2023

  25. [25]

    P. V. Ponomarenko, J.-P. St. Maurice, G. C. Hussey, and A. V. Koustov. HF ground scatter from the polar cap: Ionospheric propagation and ground surface effects. Journal of Geophysical Research: Space Physics , 115(A10), 2010

    work page 2010

  26. [26]

    P. V. Ponomarenko, C. L. Waters, and F. W. Menk. Factors determining spectral width of HF echoes from high latitudes. Annales Geophysicae , 25(3):675–687, 2007

    work page 2007

  27. [27]

    Ponomarenko, Pavlo, B

    V. Ponomarenko, Pavlo, B. Iserhienrhien, and J.-P. St.-Maurice. Morphol- ogy and possible origins of near-range oblique hf backscatter at high and midlatitudes. Radio Science, 51(6):718–730, 2016

    work page 2016

  28. [28]

    A. J. Ribeiro, J. M. Ruohoniemi, J. B. H. Baker, L. B. N. Clausen, S. de Lar- quier, and R. A. Greenwald. A new approach for identifying ionospheric backscatter in midlatitude SuperDARN HF radar observations. Radio Sci- ence, 46(4), 2011

    work page 2011

  29. [29]

    A. J. Ribeiro, J. M. Ruohoniemi, P. V. Ponomarenko, L. B. N. Clausen, J. B. H. Baker, R. A. Greenwald, K. Oksavik, and S. de Larquier. A compar- ison of SuperDARN ACF fitting methods. Radio Science, 48(3):274–282, 2013

    work page 2013

  30. [30]

    Sunspot index, https://www.sidc.be/silso/datafiles, 2025

    Royal Observatory of Belgium. Sunspot index, https://www.sidc.be/silso/datafiles, 2025

    work page 2025

  31. [31]

    D. E. Rumelhart, G. E. Hinton, and R. J. Williams. Learning Internal Representations by Error Propagation. In Parallel Distributed Processing, Volume 1: Explorations in the Microstructure of Cognition: Foundations . The MIT Press, 1986

    work page 1986

  32. [32]

    G. Schwarz. Estimating the Dimension of a Model. The Annals of Statistics , 6(2):461 – 464, 1978. 6 Conclusion 30

    work page 1978

  33. [33]

    Th´ ebault, C

    E. Th´ ebault, C. C. Finlay, C. D. Beggan, P. Alken, J. Aubert, O. Bar- rois, F. Bertrand, T. Bondar, A. Boness, L. Brocco, E. Canet, A. Cham- bodut, A. Chulliat, P. Co¨ ısson, F. Civet, A. Du, A. Fournier, I. Frat- ter, N. Gillet, B. Hamilton, M. Hamoudi, G. Hulot, T. Jager, M. Korte, W. Kuang, X. Lalanne, B. Langlais, J.-M. L´ eger, V. Lesur, F. J. Lowe...

    work page 2015

  34. [34]

    Zhang, A

    J. Zhang, A. Lan, J. Yan, X. Deng, W. Wang, H. Li, L. Sun, Y. Nan, X. Song, and C. Wang. Development of the Chinese Dual Auroral Radar Network and Preliminary Results. Space Weather, 22(10):e2024SW004131,

    work page

  35. [35]

    e2024SW004131 2024SW004131

    work page