the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 30 Nov 2023
Auroral breakup detection in all-sky images by unsupervised learning
Abstract. Due to a large number of automatic auroral camera systems on the ground, the image data analysis requires more efficiency than what a human expert visual inspection can provide. Furthermore, there is no solid consensus on how many different types or shapes exist in auroral displays. We report the first attempt to classify auroral morphological forms by unsupervised learning method on an image set that contains both nightside and dayside aurora. We used six months of full-colour auroral all-sky images captured at a high-arctic observatory on Svalbard, Norway, in 2019–2020. The selection of images containing aurora was performed manually. These images were then input to a convolutional neural network called SimCLR for feature extraction. The clustered and fused features resulted in 37 auroral morphological classes. In the classification of auroral image data with two different time resolutions we found that the occurrence of eight morphological classes strongly increased when the image cadence was high (24 seconds), while the occurrence of 13 morphological classes experienced little or no change with changes in input image cadence. We therefore investigated the temporal evolution of the group of eight "active auroral classes". Time periods for which "active auroral classes" persisted for longer than two consecutive images with maximum cadence of six minutes coincided with ground-magnetic deflections and their occurrence was found clustered around the magnetic midnight. The active auroral onsets typically included vortical auroral structures and equivalent current patterns typical for substorms. Our findings therefore suggest that our unsupervised image classification method can be used to detect auroral breakups in ground-based image datasets with a temporal accuracy determined by the image cadence.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(3681 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3681 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2857', Anonymous Referee #1, 10 Jan 2024
General Comments
This article found that unsupervised learning resulted in 37 classes of aurora, with different time behaviour for some of them. Due to the high latitude of the observing site, both dayside and nightside aurora could be observed, although only in northern winter. High and low imaging rates were compared, and some of the classes showed dramatically different occurrence rates depending on image cadence.
Specific Comments
The classes found are not described in detail. Although this would take up space (likely quite a bit of it), if there is value in the classification it should be better described. The lack of description does not allow comparative study by others, nor comparison with previously established classes which are mentioned.
Technical corrections:
The article is well written but use of the word “the” should be reviewed, especially in the abstract. There is repetition of “https://doi.org” in some of the references. Arrowheads in Fig. 5 are difficult to see, and the caption refers to curl (which is likely correct as plotted) whereas the color bar label refers to Jz which is basically an assumption of the inversion method.
Citation: https://doi.org/10.5194/egusphere-2023-2857-RC1 -
AC1: 'Reply on RC1', Noora Partamies, 08 Mar 2024
We thank the referee for careful reading of the manuscript and valuable comments. We have done our best to take into account and implement all suggestions, and we feel that they greatly improve the manuscript. Below the comments are copied in bold and our answers in regular font.
Comments from Referee #1:
The classes found are not described in detail. Although this would take up space (likely quite a bit of it), if there is value in the classification it should be better described. The lack of description does not allow comparative study by others, nor comparison with previously established classes which are mentioned.
The individual clusters are not described in detail because they result from an unsupervised clustering based on automatically determined image features. They do not form any obvious groups of auroral structures with respect to what we know from before, but based on visual inspection properties like contrast, brightness, colour, alignment and the location of the aurora in the images may play a role. It is therefore not straightforward to compare the content of these individual clusters to earlier human classified or supervised learning results. To make the distinction clearer between a pre-determined class (ground truth) and a numerically calculated cluster we now call the unsupervised method clustering and the results 37 individual clusters.
The results of this study are not dependent on understanding of the individual clusters. We merely investigate changes in the occurrence of cluster groups as a function of the temporal resolution of the input image data.
Cluster groups called active and quiet aurora are determined by their occurrence rate change as a function of input data cadence. A better description of the category of active (8 clusters) and quiet aurora (14 clusters) will be included in the revised version of the manuscript to say that the active aurora mainly includes large vortex structures, very bright or rayed arc-like structures. The quiet aurora primarily includes arc-like and multiple arc-like structures, faint diffuse aurora, and daytime and afternoon overhead aurora (corona) with a notable red emission component. Images with moonlight are grouped into the same clusters in each category, although the auroral structures in those images are similar to those in other clusters within the same category. Instead of the old Figure 4 with a few examples of active and quiet aurora, we include figures with 10 random images of each individual cluster in the category of quiet and active aurora to support the description above.
Technical corrections:
The article is well written but use of the word "the" should be reviewed, especially in the abstract. There is repetition of "https://doi.org" in some of the references. Arrowheads in Fig. 5 are difficult to see, and the caption refers to curl (which is likely correct as plotted) whereas the color bar label refers to Jz which is basically an assumption of the inversion method.
The article has been reviewed and some words “the” removed. The reference list has been reworked into a more coherent list with no repeating doi.org’s. Figure 5 has been replotted with cropped area for the maps, which makes the arrowheads better visible. The new colour bar label is Z component of the curl of the equivalent current. We use the same terminology in the caption, but also mention that this vertical component of the curl of the equivalent current can be used as an estimate of the field-aligned current, which is what is done in the text later on.
Citation: https://doi.org/10.5194/egusphere-2023-2857-AC1
-
AC1: 'Reply on RC1', Noora Partamies, 08 Mar 2024
-
RC2: 'Comment on egusphere-2023-2857', Anonymous Referee #2, 20 Feb 2024
Review of: Auroral breakup detection in all-sky images by unsupervised learning
By
Partamies, et al.---
General Comments: This work represents one of the first attempts to make a robust method for automatically detecting substorms in auroral image data. The methodology is clearly explained as are their results. As such methods become more powerful and accurate, they will be of great scientific value for the study of the aurora, because the aurora contains so many varied morphologies often with our clearly definable boundaries between them. This research is timely and well-presented and I therefore recommend that it be returned to the authors for minor/technical corrections.
Specific Comments:Line 93: It is possible that normalizing the brightness of each image could be introducing biases. For example, faint arcs could then get lumped together with much brighter ones, which may not be a good thing for categorizing the aurora. Perhaps this could be mentioned in the discussion.
Lines 75 to 95: It might be useful to add an example image, starting with the original All sky image, then how it looks at the different stages of this processing.
Citation: https://doi.org/10.5194/egusphere-2023-2857-RC2 -
AC2: 'Reply on RC2', Noora Partamies, 08 Mar 2024
We thank the referee for careful reading of the manuscript and valuable comments. We have done our best to take into account and implement all suggestions, and we feel that they greatly improve the manuscript. Below the comments are copied in bold and our answers in regular font.
Specific comments from Referee #2:
Line 93: It is possible that normalizing the brightness of each image could be introducing biases. For example, faint arcs could then get lumped together with much brighter ones, which may not be a good thing for categorizing the aurora. Perhaps this could be mentioned in the discussion.
Sounds reasonable, the new version will mention in the discussion that it should be investigated further if the brightness normalisation leads to unnecessary high weight on faint auroral structures as compared to bright auroral structures.
Lines 75 and 95: It might be useful to add an example image, starting with the original All sky image, then how it looks at the different stages of this processing.
Good point. Rather than showing each stage of the pre-processing of the images, some of which are visually minor, we include a figure with a couple of examples of the “raw” quicklook images prior to the pre-processing and the corresponding images after the pre-processing in the new version of the manuscript. They help illustrating the effect of cropping, colour enhancement and brightness normalisation.
Citation: https://doi.org/10.5194/egusphere-2023-2857-AC2
-
AC2: 'Reply on RC2', Noora Partamies, 08 Mar 2024
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2857', Anonymous Referee #1, 10 Jan 2024
General Comments
This article found that unsupervised learning resulted in 37 classes of aurora, with different time behaviour for some of them. Due to the high latitude of the observing site, both dayside and nightside aurora could be observed, although only in northern winter. High and low imaging rates were compared, and some of the classes showed dramatically different occurrence rates depending on image cadence.
Specific Comments
The classes found are not described in detail. Although this would take up space (likely quite a bit of it), if there is value in the classification it should be better described. The lack of description does not allow comparative study by others, nor comparison with previously established classes which are mentioned.
Technical corrections:
The article is well written but use of the word “the” should be reviewed, especially in the abstract. There is repetition of “https://doi.org” in some of the references. Arrowheads in Fig. 5 are difficult to see, and the caption refers to curl (which is likely correct as plotted) whereas the color bar label refers to Jz which is basically an assumption of the inversion method.
Citation: https://doi.org/10.5194/egusphere-2023-2857-RC1 -
AC1: 'Reply on RC1', Noora Partamies, 08 Mar 2024
We thank the referee for careful reading of the manuscript and valuable comments. We have done our best to take into account and implement all suggestions, and we feel that they greatly improve the manuscript. Below the comments are copied in bold and our answers in regular font.
Comments from Referee #1:
The classes found are not described in detail. Although this would take up space (likely quite a bit of it), if there is value in the classification it should be better described. The lack of description does not allow comparative study by others, nor comparison with previously established classes which are mentioned.
The individual clusters are not described in detail because they result from an unsupervised clustering based on automatically determined image features. They do not form any obvious groups of auroral structures with respect to what we know from before, but based on visual inspection properties like contrast, brightness, colour, alignment and the location of the aurora in the images may play a role. It is therefore not straightforward to compare the content of these individual clusters to earlier human classified or supervised learning results. To make the distinction clearer between a pre-determined class (ground truth) and a numerically calculated cluster we now call the unsupervised method clustering and the results 37 individual clusters.
The results of this study are not dependent on understanding of the individual clusters. We merely investigate changes in the occurrence of cluster groups as a function of the temporal resolution of the input image data.
Cluster groups called active and quiet aurora are determined by their occurrence rate change as a function of input data cadence. A better description of the category of active (8 clusters) and quiet aurora (14 clusters) will be included in the revised version of the manuscript to say that the active aurora mainly includes large vortex structures, very bright or rayed arc-like structures. The quiet aurora primarily includes arc-like and multiple arc-like structures, faint diffuse aurora, and daytime and afternoon overhead aurora (corona) with a notable red emission component. Images with moonlight are grouped into the same clusters in each category, although the auroral structures in those images are similar to those in other clusters within the same category. Instead of the old Figure 4 with a few examples of active and quiet aurora, we include figures with 10 random images of each individual cluster in the category of quiet and active aurora to support the description above.
Technical corrections:
The article is well written but use of the word "the" should be reviewed, especially in the abstract. There is repetition of "https://doi.org" in some of the references. Arrowheads in Fig. 5 are difficult to see, and the caption refers to curl (which is likely correct as plotted) whereas the color bar label refers to Jz which is basically an assumption of the inversion method.
The article has been reviewed and some words “the” removed. The reference list has been reworked into a more coherent list with no repeating doi.org’s. Figure 5 has been replotted with cropped area for the maps, which makes the arrowheads better visible. The new colour bar label is Z component of the curl of the equivalent current. We use the same terminology in the caption, but also mention that this vertical component of the curl of the equivalent current can be used as an estimate of the field-aligned current, which is what is done in the text later on.
Citation: https://doi.org/10.5194/egusphere-2023-2857-AC1
-
AC1: 'Reply on RC1', Noora Partamies, 08 Mar 2024
-
RC2: 'Comment on egusphere-2023-2857', Anonymous Referee #2, 20 Feb 2024
Review of: Auroral breakup detection in all-sky images by unsupervised learning
By
Partamies, et al.---
General Comments: This work represents one of the first attempts to make a robust method for automatically detecting substorms in auroral image data. The methodology is clearly explained as are their results. As such methods become more powerful and accurate, they will be of great scientific value for the study of the aurora, because the aurora contains so many varied morphologies often with our clearly definable boundaries between them. This research is timely and well-presented and I therefore recommend that it be returned to the authors for minor/technical corrections.
Specific Comments:Line 93: It is possible that normalizing the brightness of each image could be introducing biases. For example, faint arcs could then get lumped together with much brighter ones, which may not be a good thing for categorizing the aurora. Perhaps this could be mentioned in the discussion.
Lines 75 to 95: It might be useful to add an example image, starting with the original All sky image, then how it looks at the different stages of this processing.
Citation: https://doi.org/10.5194/egusphere-2023-2857-RC2 -
AC2: 'Reply on RC2', Noora Partamies, 08 Mar 2024
We thank the referee for careful reading of the manuscript and valuable comments. We have done our best to take into account and implement all suggestions, and we feel that they greatly improve the manuscript. Below the comments are copied in bold and our answers in regular font.
Specific comments from Referee #2:
Line 93: It is possible that normalizing the brightness of each image could be introducing biases. For example, faint arcs could then get lumped together with much brighter ones, which may not be a good thing for categorizing the aurora. Perhaps this could be mentioned in the discussion.
Sounds reasonable, the new version will mention in the discussion that it should be investigated further if the brightness normalisation leads to unnecessary high weight on faint auroral structures as compared to bright auroral structures.
Lines 75 and 95: It might be useful to add an example image, starting with the original All sky image, then how it looks at the different stages of this processing.
Good point. Rather than showing each stage of the pre-processing of the images, some of which are visually minor, we include a figure with a couple of examples of the “raw” quicklook images prior to the pre-processing and the corresponding images after the pre-processing in the new version of the manuscript. They help illustrating the effect of cropping, colour enhancement and brightness normalisation.
Citation: https://doi.org/10.5194/egusphere-2023-2857-AC2
-
AC2: 'Reply on RC2', Noora Partamies, 08 Mar 2024
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 170 | 68 | 22 | 260 | 12 | 13 |
- HTML: 170
- PDF: 68
- XML: 22
- Total: 260
- BibTeX: 12
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Noora Partamies
Bas Dol
Vincent Teissier
Liisa Juusola
Mikko Syrjäsuo
Hjalmar Mulders
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3681 KB) - Metadata XML