the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 01 Sep 2026
Thermodynamic and dynamic process attribution for Arctic sea ice loss during RILEs in CESM2
Abstract. The decline in summer Arctic sea ice extent has been underway for several decades and is set to continue until summer Arctic sea ice disappears completely by the middle of the century, according to the latest climate projections. Based on observations and these climate projections, the rate of sea ice retreat is not linear: the decline in the Arctic sea ice cover is marked by periods of abrupt sea ice loss. Recent studies suggest that these rapid sea ice loss events (RILEs) will become increasingly likely in the coming decades. However, the processes driving RILEs remain poorly understood, in particular, the mechanisms controlling sea ice concentration evolution and the sources and sinks of sea ice mass during these climate events. Modeling tools that provide process-based analyses and budget decompositions across both historical and future climate simulations offer a robust framework for understanding the physical sources of these abrupt events. To investigate the processes behind these events, we characterize 31 September RILEs from the CESM2-lessmelt ensemble under the SSP3-7.0 forcing scenario across four periods spanning 1990–2065, and apply a two-stage process-based attribution framework to sea ice concentration and mass budget tendencies to identify the dominant mechanisms driving sea ice loss from a spatial perspective. RILEs are found to be a robust and recurrent feature of Arctic sea ice evolution, with a peak in occurrence around 2040, triggered across a wide range of initial sea ice states, from the thick and extensive cover of the early 1990s to the thin and reduced ice pack of the mid-21st century. Thermodynamic sea ice processes, and basal melt in particular, emerge as the dominant driver of the most intense and spatially significant sea ice losses during RILEs, while dynamic sea ice processes contribute a consistently minor fraction. Thermodynamic dominance strengthens as the century progresses and the ice pack thins, with dynamically dominated regions remaining confined to the vicinity of Greenland, the Canadian Archipelago, and the marginal ice zone. These results provide new process-level insight into the mechanisms of rapid summer Arctic sea ice loss and set the groundwork for the development of early warning indicators of the probability of RILE occurrence.
- Preprint
(3810 KB) - Metadata XML
-
Supplement
(2628 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-4173', Anonymous Referee #1, 01 Oct 2026
-
RC2: 'Comment on egusphere-2026-4173', Anonymous Referee #2, 31 Oct 2026
This paper analyzes periods of rapid sea ice loss (RILEs, defined as at least 4 years with trends more negative than -0.3 million km2 per year in the 5-year running SIE) in the CESM2-less melt ensemble of simulations. It characterizes RILES depending on their thermodynamic vs dynamic tendency terms, and then inspects thermodynamic melt terms (surface vs bottom melt). It exclusively focuses on the melt season. It compares RILEs across 4 different periods from the late 20th century to mid/late 21st century. It finds that thermodynamics dominate, with bottom melt being the main driver. Overall, the writing is adequate, though I note it could be shortened (especially the discussion) to make more clear and highlight the authors main results.
Major comments
My main issue is that the authors focus exclusively on the melt season, which ignores well known (mainly winter) dynamic processes that can drive sea ice age (and thickness) variability (via export out of the Arctic) which has been linked to summer sea ice extent variability (e..g, Rigor and Wallace 2004). This mechanism is also seen at regional scales, e.g., in the Beaufort (Babb et al., 2019). I also note that a study investigating sea ice thickness variability in CESM1, that accounted for year-round tendencies, found that dynamics dominated local scale variability, especially at intra-to-interannual timescales, which would be relevant for RILEs (Blanchard-Wrigglesworth and Bitz, 2014). These results from previous authors make me wonder how your results are dependent on focusing only on April-August tendencies. I recommend that the authors reconstruct continuous (monthly, year round) SIC and SIT into their dynamic and thermodynamic components, in the manner of Blanchard-Wrigglesworth and Bitz, 2014 (see their appendix c), which would allow for a more complete characterization of dynamic vs thermodynamic forcing. Note that melt season thermodynamic anomalies can be a result of winter season dynamic anomalies - e.g., an ice pack that has been anomalously thinned due to dynamics in winter is more likely to undergo more melt thermodynamically in the following melt season.
Minor comments
L62 Here I would add Rigor and Wallace 2004, Cullather and Tremblay 2008, Blanchard-Wrigglesworth and Bitz 2014.
L150 What is the ‘5-year running mean minimum SIE’ what is a ‘minimum SIE’?
L151 How does this compare with climatological rate of SIE loss?
L171 How are these correlations calculated? Across all grid cells?
Figure 5 - include climatological values, so we can compare
L304 ‘cat 1 expanding’ confusing language, as cat 1 loses area, suggest rewrite
Figure 6 I assume these are for September RILEs (not other months)? Caption should be clarified
Figure 7 - It would be great to see the actual SIC trends for these periods. I imagine that they vary spatially, and it would help to assign importance to the spatial variability in the attribution maps. For example, in panel A, I imagine during this period SIC trends north of CAA and Greenland are very small, especially compared to those along the Siberian/Alaskan sea ice edge?
Discussion - very lengthy at the moment, I would recommend shortening, and focusing on main findings
L439-454 Discussion on internal variability vs scenario uncertainty. I’m not sure what the point the authors are trying to make is. The paper does not analyze scenario uncertainty (you just analyze a single scenario).
Also, note how Cullather and Tremblay 2008 showed that RILEs are possible in long control runs (and with different driving mechanisms). One can just state that, given the authors definition of RILE, the forced response can make them more likely during times with fast forced sea ice loss.
L466 ‘suggesting that some of these events are not solely driven by areal changes but also involve changes in ice thickness’ - you could calculate the change in sea ice thickness between the end and beginning of a RILE and calculate its aera-weighted average; do you ever get a RILE without thinning? Probably not?
L473 - isn’t this inconsistent with previous sentence? Thin ice grows faster, so why would this stabilizing mechanism be weaker in a thinner ice pack?
L487 ‘we characterized sea ice loss during RILEs in terms of its driving processes for the first time’ I don’t think the authors are first to characterize RILEs in terms of driving processes. As the authors themselves state a few sentences later ‘Previous studies have identified thermodynamic processes as the primary driver of rapid Arctic sea ice loss, with dynamic processes playing a negligible direct role (Holland et al., 2006; Döscher and Koenigk, 2013)’ Note also other papers such as Cullather and Tremblay 2008 or Auclair and Tremblay 2015
L510 ‘For the early period, however, the ice-albedo feedback is less likely to explain the RILEs…the contribution of solar radiation absorbed through leads is likely minimal’ I’m not sure that this statement is well supported. CESM2-LESS MELT shows relatively low mean SICs in the marginal sea ice zone in the summer (see Figure 9A in Kay et al., 2022), so presumably the model’s ocean can absorb significant amounts of shortwave in the summer (downwelling shortwave at surface * (1-SIC)).
L522 Lateral melt and the hypothesis of its importance - Since this is output by the model (you show it in Fig S3), you could quantify if lateral melt is ever important for the RILEs? Here, the limitations of CESM2’s floe size distribution (I believe floes have fixed sizes of 300m in this model version) and their impact on lateral melt (which depends on floe size) could be commented.
L548 ‘The accelerating loss of Arctic summer sea ice’ But Arctic summer sea ice loss has not accelerated; in observations sea ice loss has decelerated over the past 20 years.
References:
Babb, D.G., Landy, J.C., Barber, D.G. and Galley, R.J., 2019. Winter sea ice export from the Beaufort Sea as a preconditioning mechanism for enhanced summer melt: A case study of 2016. Journal of Geophysical Research: Oceans, 124(9), pp.6575-6600.
Blanchard-Wrigglesworth, E. and Bitz, C.M., 2014. Characteristics of Arctic sea-ice thickness variability in GCMs. Journal of Climate, 27(21), pp.8244-8258.
Cullather, R.I. and Tremblay, L., 2008. Analysis of Arctic sea ice anomalies in a coupled model control simulation. Geophysical Monograph Series, 180, pp.187-211.
Rigor, I.G. and Wallace, J.M., 2004. Variations in the age of Arctic sea‐ice and summer sea‐ice extent. Geophysical Research Letters, 31(9).
Citation: https://doi.org/10.5194/egusphere-2026-4173-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 271 | 138 | 59 | 468 | 102 | 60 | 59 |
- HTML: 271
- PDF: 138
- XML: 59
- Total: 468
- Supplement: 102
- BibTeX: 60
- EndNote: 59
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This study uses a 9 member ensemble known as CESM2-lessmelt to try to understand and analyse the processes driving rapid ice loss events (RILEs) and trends in the occurrence of RILEs over a 75 year period. The authors analyse simulations using CMIP6 historical forcing and the medium-to-high-emission SSP3-7.0 scenario over 1990-2064. The 1990-2064 period is split into four distinct periods, and RILEs are identified with a focus on September. The September RILEs are analysed not only in terms of area loss but also mass loss, showing changes and trends across the four periods. The causes of the RILEs are identified as either dynamics dominant, thermodynamics dominant or discordant, where the process driving the accelerating trend differs from the process dominating ice loss in that grid cell. The thermodynamics processes are further broken down into surface and basal melt. The study finds that thermodynamics processes become increasingly dominant throughout the study period as the sea ice thins, with basal melt shown as the dominant factor.
I believe the study provides important insight into RILEs. The paper is well written and the figures are presented nicely, though there are some questions which need addressing and clearer justification of the methods used before the paper is ready for publication.
General Comments
The paper should be more descriptive about the sea ice model setup and some of the parameterisations in the sea ice model particularly. Do the authors for example use any of the non-default parameterisations? Which melt pond parameterisation is used? These choices may be particularly relevant to the study because the attribution of sea ice loss to thermodynamics is a central part of the analysis.
Some justification for why the medium-to-high-emission SSP3-7.0 scenario was used for the study is necessary. Is this because RILEs are more present in this scenario, or is there another reason for selecting it?
In CESM2-lessmelt, only four members are smoothed to remove the spurious signal. Could the authors clarify if this partial treatment affects the identification or characteristics of the RILEs? In particular, does the different treatment of different ensemble members affect the number and timing of RILEs identified? There should be more discussion of this within the method and/or discussion section.
The suitability of the CESM2-lessmelt for studying the physical mechanisms of RILEs also requires further discussion. CESM2-lessmelt includes modifications to snow on sea ice, the snow melt onset temperature and the snow albedo. These modifications were introduced to improve the simulated mean sea ice state however, the improved agreement with observations may not necessarily mean the parameter values are correct. This is especially important because the modified parameter values directly affect the thermodynamic process attribution in this study (surface sea ice melt). There should be a discussion about how the modified parameterisation may affect the attribution of the RILEs and how robust the results may be across different sea ice model configurations.
I am wondering why only surface and basal melt are considered in the thermodynamic process attribution, rather than also including lateral melt? Is this not available in the authors sea ice model setup? The contribution from lateral melt is relatively small but the authors themselves mention the possibility that lateral melt may become increasingly important as the ice thins, so it would be useful to understand why it is excluded from the analysis. Would it be possible to add this to the study?
Specific comments:
L86: "The outline of the paper is organised as follows" should be "The outline of the paper is as follows" or "The paper is organised as follows"
Figure 1: The caption should state that the dashed grey horizontal line represents the definition of a sea ice free Arctic
L184-197: In the paragraph describing the conditions for dynamics-dominant, thermodynamics-dominant, and discordant RILEs, I think the conditions would be much clearer if presented as a bulleted list rather than a series of long sentences and difficult to follow.
L185: Should say "(c) discordant"
L186: "Simulateneously" should be "simultaneously"
L213: Should say "thermodynamic and dynamic processes"; in the next line the word "processes" is then missing
L244: "occurrence" should be "occurrence"
L260-270: The wording is confusing here as the authors sometimes describe percentages explicitly, but sometimes more vaguely (majority/half of the RILEs) making it somewhat difficult to follow The first sentence beginning "While the majority of RILEs..." is also somewhat redundant given that the authors say "Majority" and "nearly a quarter" and then give the actual percentages in the next sentence. Please clarify the meaning of this paragraph.
L273-282: I believe "paradoxically" is not the most appropriate choice of words here. The ice should be thicker in the early period and therefore more resilient to area loss, while in the later periods the ice is much thinner and hence melt leads to more area becoming ice free. Consequently the SIV loss should be higher in the earlier period and smaller in the later period and vice-versa for the SIE . This appears consistent with the analysis later in the paper, particularly the shift in the thickness distribution shown.
Figure 6: It is difficult to read the percentages within the bars. The figure should be enlarged or the labels made more legible or the data displayed in some other way.
Figure 8: This figure could be replotted with the discordant categories separated into thermodynamics-dominant/dynamics-driven and dynamics-dominant/thermodynamics driven cases. I believe this would add useful information and additional understanding and further show whether a type of discordant event becomes more prevalent through time. The figure is also a little small.
L492: "across all four RILEs period" should say "across all four RILE periods"
L513: "enhanced simulated adecline" contains a typo
L526: "are important role"
L653: "driving these rapid sea ice decline" should be "driving these rapid sea ice declines" or "driving this rapid sea ice decline" depending on the intended meaning