the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 Jan 2024
Advancing Arctic sea ice remote sensing with AI and deep learning: now and future
Abstract. The revolutionary advances of Artificial Intelligence (AI) in the past decade have brought transformative innovation across science and engineering disciplines. Also in the field of Arctic science, we have witnessed an increasing trend in the adoption of AI, especially deep learning, to support the analysis of Arctic big data and facilitate new discoveries. In this paper, we provide a comprehensive review of the applications of deep learning in sea ice remote sensing domains, focusing on problems such as sea ice lead detection, thickness estimation, concentration, sea ice extent forecasting and motion detection as well as sea ice type classification. In addition to discussing these applications, we also summarize technological advances that provide customized deep learning solutions, including new loss functions and learning strategies to better understand sea ice dynamics. To promote the growth of this exciting interdisciplinary field, we further explore several research areas where the Arctic sea ice community can benefit from cutting-edge AI technology. These areas include improving multi-modal deep learning capabilities, enhancing model accuracy in measuring prediction uncertainty, better leveraging AI foundation models, and deepening the integration with physics-based models. We hope that this paper can serve as a cornerstone in the progress of Arctic sea ice research using AI and inspire further advances in this field.
This preprint has been withdrawn.
-
Withdrawal notice
This preprint has been withdrawn.
-
Preprint
(1022 KB)
Interactive discussion
Status: closed
-
CC1: 'Comment on egusphere-2023-2831', Andrew Shepherd, 24 Jan 2024
The sea ice thickness data shown in Fig. 1 and Table 2 are from the Centre for Polar Observation and Modelling as per the in-image credit, and not NSIDC as noted in the captions. They are reproduced by NSIDC. Please update your paper to give proper credit to the work; it can be cited as :
RL Tilling, A Ridout, A Shepherd, Estimating Arctic sea ice thickness and volume using CryoSat-2 radar altimeter data, Advances in Space Research 62 (6), 1203-1225I note that several others figures have been credited to NSIDC also; you should also check the provenance of those.Citation: https://doi.org/10.5194/egusphere-2023-2831-CC1 -
AC1: 'Reply on CC1', Wenwen Li, 31 Jan 2024
Thank you so much for pointing this out! We've updated the reference as you suggested for both Fig 1 and Table 2 to provide appropriate credit. These changes will be reflected in the revised manuscript. We look forward to receiving any other comments you may have.
Citation: https://doi.org/10.5194/egusphere-2023-2831-AC1
-
AC1: 'Reply on CC1', Wenwen Li, 31 Jan 2024
- RC1: 'Comment on egusphere-2023-2831', Anonymous Referee #1, 13 Mar 2024
-
RC2: 'Comment on egusphere-2023-2831', Anonymous Referee #2, 13 Mar 2024
Review of the manuscript “Advancing Arctic sea ice remote sensing with AI and deep learning: now and future” submitted by Li et al. (2024) to EGUsphere / The Cryosphere.
This summary is very timely and useful for scientists working on the topic. For this reason it deserves to be published. In fact, it has already been published as an EGUSphere preprint and can therefore already be used and cited. But I have serious concerns about accepting it for publication in The Cryosphere although it fits the scope of the Journal. Therefore, I recommend leaving the manuscript available as preprint but not to publish it in The Cryosphere.
First of all, I had a hard time coming to this decision. It is certainly well written and very interesting to read, in particular I enjoyed section 3. My major concerns are the following points which I explain in more detail below:
- References mostly include grey literature, preprints, conference proceedings, and articles in special issues.
- The language is more reminiscent of an advertisement for AI methods than of a critical, balanced review article.
- Missing critical examination of the methods and underlying data.
References
In fact, the topic is timely and evolving quickly. This is reflected in the references published with mostly short turnaround times. The number of references is appropriate but the quality is very often questionable. According to The Cryosphere guidelines, papers should make proper and sufficient reference to the relevant formal literature. Informal or so-called "grey" literature may only be referred to if there is no alternative from the formal literature.
The overwhelming fraction of references in this manuscript has been published as preprints, in conference proceedings, and special issues with publishers such as MDPI. Some funding agencies, e.g. SNF, have stopped supporting publications in such special issues because of inadequate quality control. Of course, very good papers are also published in special issues but in general there needs to be a critical evaluation of references from lower quality sources. Otherwise, I see the danger that The Cryosphere's high quality standards will be tarnished. Furthermore, in addition to the references I was able to access, there are several references without DOI that I was unable to obtain. Other citations are clearly wrong and misleading, missing the original work. In summary, I find it difficult to see how this manuscript could be cured by a revision, as the references are essential to a review paper.
Advertisement
The authors do not hide the fact that they want to advance the use of AI. The paper concludes with the following sentences:
In conclusion, the fusion of AI and Arctic sea ice remote sensing is a testament to the incredible progress we have made in understanding one of the Earth’s most vulnerable and critical environments. It is our hope that this review paper serves as both a guide and an inspiration for researchers, engineers, and policymakers to further explore the limitless potential of AI in the service of the Arctic and our planet. Together, we can harness the power of AI to protect and preserve this unique and vital region for generations to come.
This paragraph exemplifies the opinion of the authors. I do not agree that AI has limitless potential for this purpose but there is a clear need to evaluate the potential of AI. I would just like to use this as an example that the language of the manuscript has a strong bias towards the use of AI methods and that it often lacks a critical evaluation. It sometimes reads more like an advertising brochure than a scientific review.
Missing critical examination of the methods and underlying data.
The use of AI could produce meaningless products that possibly seem reasonable at first glance but have no connection to real measurements of the quantity of interest. It starts with the primary data source for training the model. Unfortunately, the manuscript lacks a meaningful scientific weighting of methods and data. For example, consider the sea ice thickness estimation: There are instruments specifically designed to measure the thickness of sea ice and there are other sensors that provide only some proxy information of the thickness, e.g. the surface roughness. The first sensor listed in the section sea ice thickness is AMSR2. However, measurements show a clear limitation to a few centimeters by using these frequencies [Cho et al. 2024]. The authors mention CryoSat-2 only after highlighting the use of Sentinel 1 SAR for the estimation of sea ice thickness.
Cho, K. Naoki, M. Nakayama and T. Tanikawa, "The Relationship Between Microwave Brightness Temperature, Salinity, and Thickness of Sea Ice Acquired With a Tank Experiment," in IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1-9, 2024, Art no. 4300709, doi: 10.1109/TGRS.2024.3361904.
Citation: https://doi.org/10.5194/egusphere-2023-2831-RC2 -
AC2: 'Reply on RC2', Wenwen Li, 15 Mar 2024
Dear reviewer, thank you so much for your time reviewing our manuscript. While we are working on addressing your other comments, I would like to provide a brief response to the comments regarding references. This paper has reviewed 140 articles from a diverse range of outlets, including Science, Nature, PNAS, The Cryosphere, RSE, IEEE TGRS, and other major remote sensing journals. It also includes papers from top AI and machine learning conferences to connect the two domains - AI and sea ice - and to introduce new AI techniques that could be useful for sea ice research but have not yet been adopted by sea ice researchers. For example, you will frequently see outlets including ICML (International Conference of Machine Learning), NeurIPS (Conference on Neural Information Processing Systems), CVPR (IEEE/CVF Computer Vision and Pattern Recognition Conference), and ICLR (International Conference on Learning Representations), which are all top computer science conferences with very low acceptance rates (ICML: 21%; NeurIPS: 24%; CVPR: 26%; ICLR: 31%). Conference papers may not be considered as high quality in sea ice research, but these top AI conferences are where we read the high quality research papers. We acknowledge that the significant differences between fields present a challenge for conducting interdisciplinary research like ours.
In addition, in our selection of AI research papers, we did choose those which are groundbreaking work, although some may first appear on the archive, such as arXiv. In the revised version, we have carefully changed these to their official outlets. For the rest, we kept the arXiv citation because they are either very new work or really important foundational research released by big tech companies or organizations with authority, such as NSIDC. For example, the paper by Kingma and Welling (2013) on arXiv entitled "Auto-Encoding Variational Bayes," which we cited, has a citation count of 33,730. Touvron et al. (2023)'s work on "Llama 2: Open foundation and fine-tuned chat models," which we cited from arXiv, has already collected a citation count of 3,148 in less than one year of its "publishing."
We also counted the number of papers we cited from MDPI. Out of the 140 papers we cited, 16 are from MDPI Remote Sensing, which we consider a decent journal. Another MDPI paper (Li and Hsu 2022: "GeoAI for Large-Scale Image Analysis and Machine Vision: Recent Progress of Artificial Intelligence in Geography") is from MDPI IJGI, which was written by the authors of this paper and has been cited 43 times in less than 2 years since its publication. And altogether the MDPI journal citations are at 12% of our total references.
Citation: https://doi.org/10.5194/egusphere-2023-2831-AC2
Interactive discussion
Status: closed
-
CC1: 'Comment on egusphere-2023-2831', Andrew Shepherd, 24 Jan 2024
The sea ice thickness data shown in Fig. 1 and Table 2 are from the Centre for Polar Observation and Modelling as per the in-image credit, and not NSIDC as noted in the captions. They are reproduced by NSIDC. Please update your paper to give proper credit to the work; it can be cited as :
RL Tilling, A Ridout, A Shepherd, Estimating Arctic sea ice thickness and volume using CryoSat-2 radar altimeter data, Advances in Space Research 62 (6), 1203-1225I note that several others figures have been credited to NSIDC also; you should also check the provenance of those.Citation: https://doi.org/10.5194/egusphere-2023-2831-CC1 -
AC1: 'Reply on CC1', Wenwen Li, 31 Jan 2024
Thank you so much for pointing this out! We've updated the reference as you suggested for both Fig 1 and Table 2 to provide appropriate credit. These changes will be reflected in the revised manuscript. We look forward to receiving any other comments you may have.
Citation: https://doi.org/10.5194/egusphere-2023-2831-AC1
-
AC1: 'Reply on CC1', Wenwen Li, 31 Jan 2024
- RC1: 'Comment on egusphere-2023-2831', Anonymous Referee #1, 13 Mar 2024
-
RC2: 'Comment on egusphere-2023-2831', Anonymous Referee #2, 13 Mar 2024
Review of the manuscript “Advancing Arctic sea ice remote sensing with AI and deep learning: now and future” submitted by Li et al. (2024) to EGUsphere / The Cryosphere.
This summary is very timely and useful for scientists working on the topic. For this reason it deserves to be published. In fact, it has already been published as an EGUSphere preprint and can therefore already be used and cited. But I have serious concerns about accepting it for publication in The Cryosphere although it fits the scope of the Journal. Therefore, I recommend leaving the manuscript available as preprint but not to publish it in The Cryosphere.
First of all, I had a hard time coming to this decision. It is certainly well written and very interesting to read, in particular I enjoyed section 3. My major concerns are the following points which I explain in more detail below:
- References mostly include grey literature, preprints, conference proceedings, and articles in special issues.
- The language is more reminiscent of an advertisement for AI methods than of a critical, balanced review article.
- Missing critical examination of the methods and underlying data.
References
In fact, the topic is timely and evolving quickly. This is reflected in the references published with mostly short turnaround times. The number of references is appropriate but the quality is very often questionable. According to The Cryosphere guidelines, papers should make proper and sufficient reference to the relevant formal literature. Informal or so-called "grey" literature may only be referred to if there is no alternative from the formal literature.
The overwhelming fraction of references in this manuscript has been published as preprints, in conference proceedings, and special issues with publishers such as MDPI. Some funding agencies, e.g. SNF, have stopped supporting publications in such special issues because of inadequate quality control. Of course, very good papers are also published in special issues but in general there needs to be a critical evaluation of references from lower quality sources. Otherwise, I see the danger that The Cryosphere's high quality standards will be tarnished. Furthermore, in addition to the references I was able to access, there are several references without DOI that I was unable to obtain. Other citations are clearly wrong and misleading, missing the original work. In summary, I find it difficult to see how this manuscript could be cured by a revision, as the references are essential to a review paper.
Advertisement
The authors do not hide the fact that they want to advance the use of AI. The paper concludes with the following sentences:
In conclusion, the fusion of AI and Arctic sea ice remote sensing is a testament to the incredible progress we have made in understanding one of the Earth’s most vulnerable and critical environments. It is our hope that this review paper serves as both a guide and an inspiration for researchers, engineers, and policymakers to further explore the limitless potential of AI in the service of the Arctic and our planet. Together, we can harness the power of AI to protect and preserve this unique and vital region for generations to come.
This paragraph exemplifies the opinion of the authors. I do not agree that AI has limitless potential for this purpose but there is a clear need to evaluate the potential of AI. I would just like to use this as an example that the language of the manuscript has a strong bias towards the use of AI methods and that it often lacks a critical evaluation. It sometimes reads more like an advertising brochure than a scientific review.
Missing critical examination of the methods and underlying data.
The use of AI could produce meaningless products that possibly seem reasonable at first glance but have no connection to real measurements of the quantity of interest. It starts with the primary data source for training the model. Unfortunately, the manuscript lacks a meaningful scientific weighting of methods and data. For example, consider the sea ice thickness estimation: There are instruments specifically designed to measure the thickness of sea ice and there are other sensors that provide only some proxy information of the thickness, e.g. the surface roughness. The first sensor listed in the section sea ice thickness is AMSR2. However, measurements show a clear limitation to a few centimeters by using these frequencies [Cho et al. 2024]. The authors mention CryoSat-2 only after highlighting the use of Sentinel 1 SAR for the estimation of sea ice thickness.
Cho, K. Naoki, M. Nakayama and T. Tanikawa, "The Relationship Between Microwave Brightness Temperature, Salinity, and Thickness of Sea Ice Acquired With a Tank Experiment," in IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1-9, 2024, Art no. 4300709, doi: 10.1109/TGRS.2024.3361904.
Citation: https://doi.org/10.5194/egusphere-2023-2831-RC2 -
AC2: 'Reply on RC2', Wenwen Li, 15 Mar 2024
Dear reviewer, thank you so much for your time reviewing our manuscript. While we are working on addressing your other comments, I would like to provide a brief response to the comments regarding references. This paper has reviewed 140 articles from a diverse range of outlets, including Science, Nature, PNAS, The Cryosphere, RSE, IEEE TGRS, and other major remote sensing journals. It also includes papers from top AI and machine learning conferences to connect the two domains - AI and sea ice - and to introduce new AI techniques that could be useful for sea ice research but have not yet been adopted by sea ice researchers. For example, you will frequently see outlets including ICML (International Conference of Machine Learning), NeurIPS (Conference on Neural Information Processing Systems), CVPR (IEEE/CVF Computer Vision and Pattern Recognition Conference), and ICLR (International Conference on Learning Representations), which are all top computer science conferences with very low acceptance rates (ICML: 21%; NeurIPS: 24%; CVPR: 26%; ICLR: 31%). Conference papers may not be considered as high quality in sea ice research, but these top AI conferences are where we read the high quality research papers. We acknowledge that the significant differences between fields present a challenge for conducting interdisciplinary research like ours.
In addition, in our selection of AI research papers, we did choose those which are groundbreaking work, although some may first appear on the archive, such as arXiv. In the revised version, we have carefully changed these to their official outlets. For the rest, we kept the arXiv citation because they are either very new work or really important foundational research released by big tech companies or organizations with authority, such as NSIDC. For example, the paper by Kingma and Welling (2013) on arXiv entitled "Auto-Encoding Variational Bayes," which we cited, has a citation count of 33,730. Touvron et al. (2023)'s work on "Llama 2: Open foundation and fine-tuned chat models," which we cited from arXiv, has already collected a citation count of 3,148 in less than one year of its "publishing."
We also counted the number of papers we cited from MDPI. Out of the 140 papers we cited, 16 are from MDPI Remote Sensing, which we consider a decent journal. Another MDPI paper (Li and Hsu 2022: "GeoAI for Large-Scale Image Analysis and Machine Vision: Recent Progress of Artificial Intelligence in Geography") is from MDPI IJGI, which was written by the authors of this paper and has been cited 43 times in less than 2 years since its publication. And altogether the MDPI journal citations are at 12% of our total references.
Citation: https://doi.org/10.5194/egusphere-2023-2831-AC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 547 | 388 | 29 | 964 | 19 | 13 |
- HTML: 547
- PDF: 388
- XML: 29
- Total: 964
- BibTeX: 19
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1