the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Scale invariance in kilometer-scale sea ice deformation
Abstract. Large-scale modeling of sea ice dynamics assumes scale-invariance that is used to calibrate and validate current models. Validity of this assumption, particularly its lower spatial limit, remains poorly understood. Identifying when, where, and why scale-invariance does not apply is essential for linking meter-scale sea ice mechanics with large-scale sea ice dynamics and climate models. Here we address this challenge by employing unique high-resolution ship radar imagery from MOSAiC expedition in an analysis based on novel deep learning-based optical flow technique. Together these allow capturing sea ice kinematics consistently at unprecedented 20-meter spatial and 10-minute temporal resolutions over an entire winter season and into summer over a 10-kilometer spatial domain. We show that the sea ice within this domain remains largely quiescent for extended periods, with distinct events revealing a 102-meter lower limit for scale-invariance that endures as the ice cover undergoes seasonal evolution. This threshold remains stable throughout the winter, even as deformation features become more localized and distinct, suggesting an intrinsic mechanical constraint that is invariant under varying external conditions. Once the ice transitions to a floe-dominated configuration in summer, no comparable scaling signature emerges. Our results give a limit under which continuum models fail to capture critical fine-scale processes, highlighting the need for approaches accounting for detailed description of discontinuous spatial and temporal behavior of sea ice.
Competing interests: At least one of the (co-)authors is a member of the editorial board of The Cryosphere. The authors have no other competing interests to declare.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.- Preprint
(4273 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 27 Mar 2025)
-
RC1: 'Comment on egusphere-2025-311', Anonymous Referee #1, 06 Mar 2025
reply
This paper is pushing the boundaries of observing scale invariance in sea ice deformation. The authors state that there is a lower limit on scale invariance, which they find to be present across a variety of sea ice dynamical regimes in the MOSAiC drift from November until April, and in the MIZ in June. If the findings are robust, it is an important contribution.
I do, however, have one major concern. In order to create 150m resolution ice drift from RADAR, the authors use a deep learning method to track RADAR targets. The authors have developed a method that is effective at providing such high resolution maps, which is a vast improvement on previous methods for RADAR tracking. For example, the tracking method of Thomas et al. (2011) has been applied to RADAR to provide detailed motion maps, however this method imparts a large scale filter on the motion field that yields a product inappropriate for scale analysis. I have personal experience of trying to use these older methods for this goal of decreasing the resolution cut off for scale analysis of deformation, and realized that new motion tracking methods needed to be developed to achieve this goal. So I am excited to see progress towards this. My one concern is that we have not examined the accuracy of the Uunsinoka et al. (2025) tracking method at all scales. Uusinoka (2025) figure 1b3 shows how noise in the RADAR imagery is manifest. This noise, which we assume is from waves and wind "wobble", appears to have a course grain structure that is of order 100m wide. I am guessing the dimensions of the course graining comparing to the scale in figure 1. How does this impact the RAFT solution at 10^2 m (100m) length scales? Is the "surprising" result that scaling behavior disappears below 100m simply that you are observing noise between vectors at this scale. If you are sampling noise you would expect beta and alpha to go to zero. If this noise has a particular course grain structure that is imparted by the methodology, you would expect L_c to be constant across the analysis.
While this concern means I do not believe this particular finding, I still find the paper highly valuable. This is the first time we have observed ice motion across scales of 1 to 10 km with such fidelity. The results could be combined with buoy analysis from the MOSAiC DN to extend to 1000km scales (which is outside of the scope of this paper, but I would like to see done). I am looking forward to a response and learning more about the accuracy of the new motion tracking method at the smaller scales. Should noise be an issue, the main finding of the paper needs to be refocussed - because you are not finding a lower limit on scaling that is applicable to the modeling community.
I have some smaller suggestions.
1. The paper in many places has sentences with grammar that I found hard to follow. A copy editor should be able to help here. I honestly stopped trying to correct them in my read. I will happily do this on a second draft of the paper.
2. The gray lines in figure 3 were hard to see in my print out.
3. The units of each term in figure A2 do not match, or is it that I am rusty in Einstein notation (I admit I prefer reading vector notation).Citation: https://doi.org/10.5194/egusphere-2025-311-RC1
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
115 | 28 | 4 | 147 | 5 | 6 |
- HTML: 115
- PDF: 28
- XML: 4
- Total: 147
- BibTeX: 5
- EndNote: 6
Viewed (geographical distribution)
Country | # | Views | % |
---|---|---|---|
Finland | 1 | 69 | 43 |
United States of America | 2 | 49 | 30 |
China | 3 | 9 | 5 |
Canada | 4 | 5 | 3 |
France | 5 | 5 | 3 |
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1
- 69