| 11 Jun 2026
A Shifting Balance: Dynamic and Thermodynamic Controls on Arctic Sea Ice Thickness
Abstract. Arctic sea ice thickness has declined rapidly over recent decades, yet the relative roles of thermodynamic growth and dynamic redistribution in driving this change remain poorly constrained at basin scale. We quantify thermodynamic and dynamic contributions to sea-ice thickness change together with their uncertainties across the Arctic from 2002 to 2020 by combining satellite-derived thickness with sea-ice model simulations (Icepack) along trajectories. Separating dynamical thickening (30%), dynamical thinning (−24%), and lead-ice growth (10.2%) shows that dynamic processes contribute nearly as much to the average winter ice growth of 0.21 m per month as thermodynamic processes (35.8%). Regional, seasonal, and thickness-dependent variability is consistent with large-scale dynamic patterns and the ice-growth feedback. We quantify the effects of the overly smooth deformation forcing, which leads to an underestimation of large dynamic events and a substantial noise floor during dynamically quiet periods, and relate their magnitude to other sources of uncertainty. Analyzing the long-term trend from 2002–2020, we resolve a weak increase in median sea ice deformation (1.5% per year) and net dynamic thickness change (12% per year) within the limits of our study setup. Overall, our results suggest that increasing deformation in the Arctic enhances net dynamic thickness change and acts as a negative feedback in the pan-Arctic winter thickness budget.
Review for "A Shifting Balance: Dynamic and Thermodynamic Controls on Arctic Sea Ice Thickness." von Albedyll et al, 2026
The authors combine satellite observations of sea ice drift & thickness with 1D sea ice model simulations to decompose the dynamic & thermodynamic mechanisms controlling Arctic winter ice thickness evolution. They use a Lagrangian method to track satellite-altimetry ice thickness changes along drift trajectories, then use sea-ice deformation derived from differential ice drift and reanalysis forcing to drive model simulations of sea-ice thermodynamic growth and dynamics (ridging, open water formation, lead ice growth) along the same trajectories. Significant trends in sea ice deformation and dynamic thickness gain are reported across the study period 2022-2020 and then used as evidence for a negative sea-ice volume feedback mechanism related to increasing sea-ice deformation.
We decided to share the review to focus on our separate expertise in sea ice observation and modelling. First off, credit to the authors for completing such a challenging and comprehensive study on an important topic. An improved understanding of the processes driving winter sea-ice growth and redistribution will benefit sea-ice model development for more accurate projections of future Arctic climate, among other applications. The main findings are generally convincing and contextualized with a detailed uncertainty analysis. Special credit to the authors for spending so much time evaluating sources of uncertainty, which pre-emptively answers many of the obvious questions for readers.
The paper deserves to be published but would benefit from some clarifications and for the authors to consider how they present their findings, especially regarding trends. More carefully defining the proposed dynamic & thermodynamic mechanisms would allow the reader to better compare the results to past studies and understand the implications of trends. Although the study focused on winter and early spring, there should be some discussion on the role of summer processes in mediating the proposed new feedback mechanism.
Please feel free to get in touch with specific questions on our review, Marek Muchow & Jack Landy
Major comments
Particular clarification is required for the dynamical thinning term. How do you define ice thickness – is it the thickness of the ice-covered area, or the effective thickness scaled by the ice concentration? If the former, which is typically the processing assumption for CryoSat-2 observations, it is unclear how the dynamical thinning fits in, since it should be zero in the case of ice divergence. Are there other mechanisms involved in this term related to the advection of ice into or out of the tracked area, or other? If the latter definition is true, then there are questions around the conservation of ice volume for the other terms. How does the definition of dynamic thinning fit to processing assumptions of the observations and model setup?
Other questions re. Definitions. The definition of lead ice growth in Section 3.3. is quite clear, but it gets confusing later on what the authors include as “thermodynamics” in their discussion of the budget and trends. Consider renaming thermodynamics to more obviously differentiate between new ice growth and existing ice thickening thermodynamically.
Minor comments/edits
Line 4-6. Needs to be more specific. Net contribution to growth from dynamics would be ~16% right, so not similar. You mean gross contribution? Also presented in the abstract this way, it is questionable why lead ice growth doesn’t constitute thermodynamics.
L28. Explain what you mean by “start regions of the TPD”
L29. “dynamics contribute negative...” what does this mean? What are the mechanisms?
L31-34. We would suggest defining dynamics in the intro here, since it is unclear what negative dynamics actually means? Are you referring to advection too, or just divergence/convergence? Is new ice growth in divergent events characterized as a dynamic or thermodynamic (TD) process?
L33. What is the reasoning to hypothesise a shift around the 2010s?
L37. “dynamical thinning due to open water formation” requires clarification. Is this characterizing the effective thickness scaled by SIC? Or do you mean new TD ice formation from open water acts to reduce the mean thickness within an area? but that is captured in your next term.. so what does this negative term actually represent?
L50. We had questions here on the possible representation errors for a parcel of 75 km width, but we mostly answered later on. Could be worth pre-emptively referring to the representation error results here.
L54-55. We guess the snow depth evolves relatively smoothly if using warren/PMW, but what about the ice type and its impacts on the SIT processing chain? e.g. ice density. do trajectories cross ice type boundaries and how do the ice type fraction time series for parcels look? do you filter on the ice type? Are there any jumps when a time series “changes” ice type..?
Fig 1 caption. What do simulations mean in this context?
L66. What does it mean that an IMB buoy was “not suspected to have experienced dynamic events”? What metric did you follow to make this selection?
L68. bit ambiguous. you mean model simulations performed along DA-SIT trajectories or the trajectories are simulated?
L79. Is the initial area of the Icepack simulations always based on the area of a circle with a radius of 250 km? If this is the case, we would suggest approximating the spatial scale then based on the area of a circle as currently the spatial scale has a difference of around 50 000 km2. Additionally, this can help to understand the initial Icepack setup.
L84. What are the “starting points”? The trajectory starting points or the starting points of your analysis?
L92. Averaging or summing histograms? Does this account for the different number of samples between histograms?
L93-94. Getting a bit lost here between corresponding date and particular date.. these are monthly distributions compiled from all trajectories grouped together that start on a similar date?
L101. consideration of missed deformation with a 1-day interval? opening and closing of a larger magnitude than represented by a 1-day polygon change?
Section 3.3. It could be helpful to make the introduction of Icepack as well as the description of the setup used more concise with more focus on the here used setup. For example, the list of possible tracers can cause additional confusion on what variables you focused on. It might make sense to explain what variables you use from the Icepack simulations (e.g. Section 3.3.3) before you explain what the specific settings are.
L147. “5146 monthly differences” explain this, why so similar to the number of daily observations? what are monthly differences?
L156. Both abbreviations RMSE and MAD are only introduced in the appendix. It could help the reader to include them here as well.
L163. Is there a limitation of not accounting for advection of thinner or thicker ice into or out of trajectory clusters. how much does this matter?
L166-167. Explain how these parameters are obtained from the polygon deformation estimates.
L173. this is calculated monthly right? so are the simulations re-initialized monthly? how much impact does the initial SIT distribution have, and does it fit well to the ITD updates arriving with new altimetry passes?
L191-195. It would be valuable to know the fraction of data removed in each of these curation steps, if possible.
L201-204. It is hard to follow why this data curation (“step 2”) was done. Especially as later in the manuscript often results from both the curated and non-curated data sets are compared with each other with only minimal differences (e.g., L249-252). Is this curation necessary or does it introduce additional uncertainties (e.g., L603-606) as well as makes the methods (and presentation of the results) more complex and complicated to follow?
L205. What filtering? Does this reference the curation mentioned above? Maybe use “curation” instead of filtering, so that it is easier to draw connections.
L209-210. Is there a danger with this approach that deficiencies in the Icepack TDs are up-weighted by selecting only SIT-DA that approx. match, thereby biasing all dynamics terms pulled out of the simulations? Have you checked whether there are any dependencies (initial mean SIT, snow depth change, total SIT change, ice type, season etc) for the pairs removed?
L210-216. CS2 DA-SIT is the thickness of the ice-covered area, right? so will it be expected to account for the dynamic thinning term (thin ice TD term also..), without some correction to effective SIT?
L210-215. How many values actually use the 90% as a boundary or how many are using the other “special boundaries”? Is it possible to only use one or the other? What is the effect on the data selection?
L221-222. Can you interpret what this means? Poorer agreement in low SIC, MIZ, marginal seas? Could this have any impact on the generalization of your results?
Fig 2. Could the opposite approach have yielded a different result? Tuning or optimizing the Icepack simulations to match the full distribution of DA-SIT change, rather than discarding DA-SIT to match Icepack? Additionally, Fig. 2a is missing a description on the y-axis.
L231. there is quite a difference in the gradients of the pattern though, with Icepack showing much lower variability between locations. how much might this impact the derived terms?
L233. “weaker in Icepack than in DA-SIT" why is this though, since icepack is initialised with da-sit?
L234. Are the CAA results valid with the uncertainties in CS2 data and PMW ice drift trajectories close to the coast or within the islands?
L238. Might the lead ice growth be weaker overall because the curated dataset removes lower and negative SIT changes with respect to the full dataset? Are the situations where Icepack would tend to produce lead ice growth removed from the DA-SIT initialization data?
L244. “thermodynamic thickening” this is a situation where it is not totally clear where lead ice growth fits
L243-245. “First, dynamical thickening is of comparable magnitude to thermodynamic thickening and is therefore a major contributor to total thickness change. Second, the positive and negative dynamic terms largely offset one another, yielding a much smaller net dynamic contribution of 0.06 m per month.” The way these two sentences are currently presented feels contradictory. Isn’t the “real” dynamic contribution the 0.06 m per month? These sentences are an example where it would be good to address that, in this case, dynamic thickening seems to mainly refer to the influence of macroporosity on the ice thickness.
L250. Wouldn't it make more sense to offset ridging with dynamic thinning instead of with lead-ice growth within the framework of this study? After all, the ice which got removed (dynamic thinning) got redistributed in the ridges, because that is where the ice goes (regarding the ice mass balance) and then we gain net positive on the thickness from ridging via the macroporsity (while the ice volume stays the same). Then the other positive effect is from the new ice grown in leads, for which ridging is a necessity.
L249-254. It is a bit surprising that the numbers are so similar between full and curated datasets given that the distributions in Fig 2a are quite different. Re. To our comment on L209-210, does this imply there are not dependencies in the data removed in the curated version? It would be great to confirm this. Additionally, it would also be possible to just choose one of these datasets for the results if the results are so similar.
L265-267. We are unsure about the argument about the “growing buckling strength”. On the one hand, Fig. 6 in Hopkins (1998) shows ice load/force measurements from one simulation; thus, this might be a wrong citation. On the other hand, thicker ice has a higher buckling strength, which makes it harder to ridge thicker ice. It is unclear to us how this observation relates to the ice volume?
Fig. 5. Is this figure necessary? It seems like a potentially more confusing summary of Fig. 3 and is not referenced in the manuscript separately.
Fig. 6. Wouldn’t it make more sense to have the orange part (dynamical thinning) with a negative sign like in Fig. 4? Additionally, a “zero” line would be helpful. How was the initial ice thickness determined? The participation function within sea-ice redistribution schemes do not account for deformed or undeformed ice while selecting the ice to be deformed. Or does this information come from observational data? Or the initial ice thickness distribution? This question is true for all further figures referring to an initial thickness.
Fig 7. Typo on the lowest panel (should be dynamic thinning). The contrast between the positive and negative mean thickness change could be a bit bigger. The seasonal cycle of the DA-SIT also appears to be different to the Icepack simulations, with the simulations peaking towards Jan whereas as the observations are quite stable before a jump to higher SIT change in Feb and March. What is your interpretation of this?
L284. “increasing ice thickness towards spring” but you say above that dynamical thickening is largest for thicker ice, so why does it drop in spring? Does the reduced frequency of ridging compensate for the increased ridging magnitude per event for thicker ice?
L285 until the end of the paragraph. The logical cohesion of this paragraph is a bit off. It would be great to first give an overall impression on the whole period before talking about the differences in satellite products as the focus of Fig. 8 is on the overall period first and foremost. Or maybe even detangle both topics into different paragraphs.
L289-290. What are the means for these two periods? Is there a significant trend for the first period? This transition in 2012-2014 seems important in your budget analysis, so can you suggest some mechanistic change in sea ice before and after this transition? elicited by the 2012 summer min and 2013-14 thick winter ice covers? Additionally, it is hard to compare the dynamical thickening trend over the whole period with the trend after 2014/2014 as they are presented in different ways (percentage and m per month).
Fig. 8. This figure could be improved by highlighting the line at 0 with a black line to indicate the sign of the different contributions.
Section 4.2. This whole section feels very hard to follow, which is likely coming from different levels of details in different sections of the paragraph. For example, the general topic and general method are introduced until L306. Then, the more detailed uncertainty (here the comparison of Envisat and CryoSat-2) is introduced before the bigger picture of the trends are even addressed. Wouldn’t the other way around make more sense? At least L313 to 317 before L307. Further, it would be good if trends could use a consistent unit (e.g. L321 cm per month per season and L327 s-1 per season).
L300. owing to a negative trend in SIT? what trend does your data suggest?
L314-315. bit vague. add some more specific details for readers unfamiliar with these uncertainties
L321. 12% more dynamic growth per month sounds extremely high and might be misinterpreted given that it is relative to a mean value that is very small. the data should potentially be normalised before calculating relative monthly change, or the % change shouldn't be highlighted so much and only the absolute should be reported.
Fig. 10. We would suggest identifying the significant trends with hashed grid cells compared to the not significant trends.
L370. Is it reasonable to assume all DA-SIT measurements within a parcel are uncorrelated, based on covariance lengths of same-track SSHA interpolation errors or auxiliary snow depth/ice type data that have correlated errors on 10s km scales? The median DA-SIT uncertainty of 8 cm sounds low, but how much would it affect your later uncertainty budgeting if the spatial measurements were assumed to be correlated and parcel SIT errors were more like 30-40 cm?
Sections 4.3.1 and 4.3.2. Thanks for providing such a complete error analysis.
L399. What ice thickness does Icepack overestimate? The mean ice thickness calculated from the ITD?
L465 onwards. Point (3) feels less thorough compared to the other points. As detailed above, the choice of ridging parametrization could be assessed as Icepack offers two different ridging functions. Further, the observations regarding the “uniform ice” configurations are really connected to the ridging scheme or are more a consequence of how to implement a Lagrangian framework.
L487-489: Where does the information from the first sentence come from? If this information is also presented in Fig. 12 it would be good to reference this and remove the word “further” in the next sentence.
L494-497. It would be useful to add this as another panel to Fig. 12. Additionally, the variable is not explained.
Fig. 12a. Are the sharp contours parallel to the 1:1 line coming from the data curation filters? How much do the numbers at L483 change when using the full dataset?
Fig. 12b. The negative signs on the y-axis are hard to see at –0.3. It could be helpful to plot the zero lines differently than the other lines.
Fig. 13. This figure would be improved if the thermodynamics from Icepack in b) would not have the same colour as the DA-SIT changes in a).
L500 onwards. The discussion around Fig. 12b and Fig. 13 is presented as Icepack model physics deficiencies with respect to the DA-SIT thickness changes as a reference “truth”. Is this the author’s intention? We could imagine some of the stronger DA-SIT_dyn monthly changes, at the tails of the distribution, are representing periods with high systematic error in the DA-SIT data (even if for most of the DA-SIT the systematic error is relatively low). For example, an artificially thick or thin DA-SIT at the beginning or end of a 30-day interval, caused by e.g. a SSHA or freeboard bias, or snow depth bias, or jump between ice types, would produce a strong DA_SIT_dyn response. So, how convinced are you that Icepack physics are underestimating the true dynamic response, versus the stronger DA-SIT_dyn response for extremes is overestimating the truth? The section could be revised if the authors did not intend the Icepack analysis to be framed this way.
L500-503. We have troubles following the interpretation of Fig. 12b here. DA-SIT_dyn describes the net dynamic thickness change. Thus, negative DA-SIT_dyn describes dynamic thinning and positive DA-SIT_dyn describes dynamic thickening, right? Then, both dynamic thinning and dynamic thickening are then less strongly represented in Icepack or underestimated. But where does Fig. 12b show that small dynamic thickness changes around zero are overestimated?
L542. Formatting of the reference Anheuser et al. (2023) needs correction.
L556. Missing space between “month” and start of the reference “(Koo”.
L563-570. Since the ice topography was not a direct focus of the study, this text feels more appropriate as motivation for the intro than discussion of results.
Section 5.3. Throughout the whole manuscript the mechanical redistribution (both participation function and ridging function) is seen as a given, while their assumptions are not mentioned. How do they contribute to the uncertainty between simulations and observations?
L702-710. Although the analysis is performed in winter, can you comment whether summer processes may have any impact on the reported trends in dynamic thickening or deformation? For instance, by changing the initial conditions of the ice at trajectory start points over time.
Fig. 17. It looks like several paths are plotted overlaying each other, which complicates understanding the regional changes.
L869. Fix reference for Fig. XX.