the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Jul 2026
Unravelling the source for diverging long-term ice loss trajectories within the Greenland Ice sheet Coupled Model Intercomparison Project (GrICMIP)
Abstract. The long-term evolution of the Greenland Ice Sheet (GrIS) remains a major source of uncertainty in projections of future sea-level rise. Recent advances in coupled climate–ice sheet models provide new opportunities to investigate the role of feedbacks between the ice sheet and the climate system, yet substantial divergence persists across climate-ice sheet model projections. Here, we present results from the Greenland Ice sheet Coupled Model Intercomparison Project (GrICMIP), using three coupled climate–ice sheet models to simulate GrIS evolution under multiple emission scenarios to the year 4000 CE. While projected sea-level contributions remain modest by 2100 (0.03–0.11 m), they diverge strongly on longer timescales, reaching up to 3.5 m by 2500. Under the high-emission scenario, full GrIS disintegration is projected as early as 3000 CE. By means of targeted sensitivity experiments, we identify the dominant sources of diverging GrIS trajectories. Across all models, changes in surface mass balance (SMB), and in particular net surface melt, control the long-term ice-sheet retreat. Differences in SMB formulation, especially in the underlying energy balance models, together with differences in the simulated climate, outweigh the influence of initial ice-sheet geometry. This establishes a hierarchy of uncertainties in which atmospheric processes and their representation within the model systems propagate non-linearly into ice-sheet evolution. Our results demonstrate that reliable long-term projections of the GrIS critically depend on improving the representation of climate and SMB, rather than on refining initial conditions alone.
Competing interests: At least one of the (co-)authors serves as editor for the special issue to which this paper belongs.
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 paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(10939 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-3384', Jeremy Fyke, 24 Aug 2026
-
RC2: 'Comment on egusphere-2026-3384', Anonymous Referee #2, 25 Aug 2026
Summary
Schanwell et al. have conducted simulations of the future evolution of the Greenland ice sheet (GrIS) and its contribution to global mean sea-level rise for the next 2000 years under three different climate change Scenarios using three different climate models that cover a (limited) spectrum of complexity and which include coupled ice sheet components. Their primary analysis is concerned with understanding the significant differences in this evolution as it is simulated by the different models, which they attribute primarily to differences in modelling the climate and its influence on how ice sheet surface mass balance is represented, downplaying the role in such projections of the initial state of the ice sheet model itself. The paper is quite clearly written - although sometimes a little too concise for my preferences - and the figures are generally appropriate to support the stated scope of the analysis - although the multi-panel figures are sometimes too small in hardcopy when printed out on A4 pages. I think the conclusions drawn could be caveated a little more with regard to the timescales over which they have identified these dominant factors and that the role of ice dynamics has not really been isolated independently from the climate and SMB representation in the model. With some more careful statements around the conclusions and some more contextual detail given in the model descriptions and description of results I think this would make a useful contribution to the emerging literature on coupled climate - ice sheet modelling.
General Comments
Although the paper does quote numbers for the sea level contributions from each model under each scenario, not a lot of emphasis is placed on these, preferring instead to try to understand the cause for the differences, which I think is a good use of simulations like these. Their main conclusion, and thus the focus of pretty much all that is written in the paper, is that the local climate evolution and SMB representation are the main factors. This point is made clearer by the fact that the experiment has been done in the same way across the three models, and in general the paper is a nice demonstration that model structural issues in representation of climate and downscaling (ie EBM) to allow ISM coupling need as much, if not more, attention than the coupling mechanism and the ice sheet model itself. That said, it would have been nice to see the MIP structure exploited further to understand the difference in retreat patterns seen in each of the models.
I find the paper quite brief in a few aspects where I think more contextual detail would help with a readers' understanding (eg model description). I also felt it could do with more context of the ice sheet simulation in each model including some evaluation of performance over recent historical period to help readers get a more complete view of what happens in the simulations
It would be good to note more clearly in the design of the factorial experiment, where the usable results come before the ice sheet has significantly evolved in each model, that they have not untangled the role of ice dynamics compared to the simulation of climate - this is an important limitation that I think could be made much clearer. The authors could also be more careful about stating the timescales over which unimportance of initial condition is true, since there is evidence that in the first century or so the ice sheet initial topography does play a significant role.
Specific Comments
Title: I'm not sure you can 'unravel' something that doesn't explicitly have multiple parts. "Unravelling the sources of diverging ..." perhaps
line 56: In general there's a lot of this section that requires readers to go dig through the original model description papers to find detail beyond the submodel names and resolutions. In many places I feel like it would be useful for readers if this section (or at least the Appendix) had more detail of factors affecting relevant physics features and intepretation of how those details are likely to affect the model's simulation eg sliding laws, going beyond the simple naming of the sliding law in use in table A1.
line 111: the MPI-ESM section states that the AIS is prescribed in this study, but also makes specific reference to the resolution of the ice sheet model grid for Antarctica which makes the status of how the AIS is handled in this configurations a bit confusing. Could you clarify?
line 120: a similar comment to the one on general model descriptions above, but it's even more important here to describe all the relevant detail in how the snow and firn are modelled and the surface mass balance is calculated in each case. The three descriptions could be harmonised to cover the same aspects in the same way, rather than being three seemingly-independent descriptions. Eg, in each section a list of exactly which atmospheric variables are used, how refreezing is handled - which is touched on in the CLIMBER-X description but absent in others. As with the comment on the general model description, some interpretation of the relevant limitations and implications in the following simulations of the parameterisation choices wold help interpretation.
line 167: Again, at least little more detail on the construction of the extension of the SSP scenarios to yr 4000 and to what degree the idealised nature of what has been done may impact the relevance of the results to more realistic futures would be welcome, rather than diverting the reader to the Supplementary Material of a previous publication.The atmospheric CO2 concentrations that result from the CLIMBER-X emissions-driven simulations are shown, but no information is given here about what emissions have been used to create them.
line 179: I'm confused as to the point of doing standalone EBM calculations covering the full span to year 4000. Doing this is by no means equivalent to really running a full simulation of the climate system with eg, for experiment EA the climate and ice sheets model of MPI-ESM but the EBM calculation of AWI-ESM. The evolution of SMB from the standalone EBM will likely quickly become quite different from what would have happened in the full system because it is so dependent on the changing ice sheet topography on the timescales in question, and the change in topography it sees are those that were produced by the original MPI-ESM EBM, not the AWI-ESM EBM that is being tested, surely? It makes sense that the analysis to be done just focuses on a timeslice very near the beginning of the simulation (line 180), when the climate-sensitivity-dependent forcing signal has become different between models but the topography feedbacks have not yet become too significant - so perhaps you don't need to mention that the standalone EBM calculations were done for the later periods too?
line 186: "to the EBM used" would be better
line 217: Many factors control the simulated rate of AMOC decline in a model and I don't feel like ocean resolution alone is a strong or reliable enough factor that it's worth singling it out.
line 220 and on: are the numbers reported the T and P natively from the climate model or after they have been downscaled by the EBM / ice sheet coupling?
line 228: the CLIMBER-X SSP5-8.5 simulation invites a simplistic conclusion - there's an almost qualtitative diffrence in the sea-level contribution of the GrIS on a ~1000yr timescale linked to AMOC resilience (and melt-elevation feedback), where in the SSP5-85 simulations here by year 300 you have lost 1/3 of the GrIS if the AMOC shuts down, or see complete deglaciation if it does not. Is that going too far?
line 245: how do the spin-up icesheet extents compare to BedMachine for the 3 models? Could ice thickness differences (or surface height) compared to BedMachine also be plotted. Surface ice velocities compared to modern observations would also be useful to interpret whether the simulations of ice dynamics are likely to be in a reasonable ballpark. These features might help explain the different retreat patterns of the icesheet for each model, the differences between which are not discussed.
Figure 2: printed at A4, the 15-panel map figures such as this are really too small to make out some of the geographical detail in each panel
line 250: The lower end of the projection results here are all less than would be implied by a simple continuation of the currently observed 0.7mm/yr noted in the Introduction. Does this imply that the models in general do not reproduce currently observed mass loss rates? This deserves some comment. It might also be useful for some readers, since they generally look pretty linear, to quote the mass loss results as rates rather than just total amounts at different time horizons.
line 256: I think the consistent difference in retreat patterns across the models interesting. In theory this aspect is the sort of thing that a MIP is well-placed to understand and exploit in terms of perhaps saying which models behave more realistically than others, but this feature is mentioned and then not returned to?
line 284: whilst it conforms to expectation that surface melt would be found to be the dominant term in both SMB and the future GrIS mass budget generally, it is maybe worth noting that in all of the modelling systems here melt is the aspect to which most attention is paid in terms of process modelling and downscaling, and the aspect most amenable to lower resolution climate modelling of large-scale thermodynamics. That makes this conclusion a bit of circular argument, where other (more unlikely) factors such as fast ice dynamics, incursion of warm ocean water under the interior of the ice sheet where bedrock is below sea level and dynamics of atmospheric precipitation are barely represented, if present at all, in the class of climate models and ice sheet coupling used in this study.
Figure 3: Showing the information only as anomalies (and having defined calving mass loss as a positive-sense flux in equation 1) means it's really not obvious from the figure that the calving behaviour shown actually amounts to (line 77) reducing to zero across all models. Is the figure much less clear if absolute fluxes are shown?
Presenting the information as area-integrals when (presumably) both the average flux and the area of the ice sheet are both changing simulataneously can also hide some interesting behaviour - especially as here when the vertical scale chosen makes the accumulation changes appear small. Separate plots of extent and area-average flux might be interesting, if you haven't looked at those?
line 347: 3.5 felt like another section that would have benefited from more description of the results in question and walking the reader through from there to the author's conclusions. At the moment it feels a bit too condensed - the results, analysis and conclusion of all 9 of the factorial experiments trying to unravel the influence of each of the 3 main contributions selected take only one page of text. This section is where it might fit to attempt some explanation of why the main experiments with each model see the very different retreat patterns that they do. Might that lead to some insight into the model differences that could help suggest which of the projections might be more or less likely. The difference in rates of ice loss in the next few centuries in the SSP5-85 scenario between the MPI-ESM and CLIMBER-X simulations would imply very different strategies for sea-level adaption and any information to help reduce the uncertainty would be very valuable. The wording of the section title and second sentence could also be more precise - it is really just the influence of the initial surface topography that is beign isolated in this experiment design. Feedbacks between the evolving topography and the climate and melt calculation are clearly very important for the long-term evolution, and those are not isolated here.
line 358: I don't think I agree that that the topography is responsible for the least spread hypsometrically in figure 6(c,i), A6, A7. Although all three topographies have a similar ELA, below ~1000m where melting has the greatest effect the magnitude of the ablation seems to vary more significantly than that produced by varying the EBM
line 393, 396: is it true to conclude for all projection timescales in these results that initial geometry plays minor role in these results, I suspect on shorter timescales it is a much more important factor (as noted in line 390 that cites Goelzer2018), and that should not be forgotten in these conclusions - different readers will have different timescales in mind as a priority. This is not the first study that has noted this sort of conclusion on the limited influence of initial conditions, eg Coulon et al. Nat Comms (2025) for Antarctica
The sensitivity tests here can only be done before the evolving ice sheet topography starts to play a role. That being the case, I don't think the experiment design can rule out representation of ice model dynamics as a crucial factor in obtaining "reliable long-term projections of the GrIS" (Abstract). The conclusions written in this paper are only about the importance of the representation of climate and SMB compared to the initial condition which their analysis does support (and line 385 expresses well in terms on the initial value vs boundary-type problem), but I think this issue should be noted.
Citation: https://doi.org/10.5194/egusphere-2026-3384-RC2 -
RC3: 'Comment on egusphere-2026-3384', Anonymous Referee #3, 31 Aug 2026
This study investigates the primary sources of divergence in multi-millennial (up to 4000 CE) Greenland Ice Sheet (GrIS) projections across three coupled climate-ice sheet models under various emission scenarios. While projected sea-level contributions remain relatively uniform and modest across models by 2100 (0.03-0.11 m), trajectories diverge dramatically on multi-centennial to millennial scales, reaching up to 3.5 m by 2500 and complete disintegration under high emissions by 3000 CE. Using targeted sensitivity experiments that isolate individual model components, the authors demonstrate that differences in the simulated atmospheric climate are the dominant drivers of long-term mass-loss spread, surpassing uncertainties in the representation of SMB processes, and far outweighing the influence of initial ice-sheet geometry.
The manuscript is well written, clearly structured, and features high-quality, informative figures. The experimental framework is well designed to address the primary research questions. However, several aspects regarding scenario setup, model resolution, temporal scales, and comparison with existing literature would benefit from further clarification and discussion. I believe that all of the comments below are relatively straightforward to address.
1. Contextualizing Findings with Previous Ensemble Studies:
The core finding, that differences in atmospheric forcing and SMB schemes dominate over other uncertainties, closely echoes the conclusions of earlier ensemble work (Holube et al., 2022, doi:10.5194/tc-16-315-2022). In particular, Figures 6a and 6b strongly support those earlier findings. It would be valuable to include an explicit discussion of how the findings from the fully coupled, advanced models used here complement the more comprehensive large ensemble and statistical variance partitioning of that earlier work.2. Ice-Ocean Interactions and Model Resolution:
While the conclusions regarding SMB dominance are well supported by the experimental setup, I wonder whether the relative importance of ice-ocean interactions might be underestimated due to the coarse resolution of the ice sheet models. Resolving narrow Greenlandic fjords and dynamic calving margins realistically requires high spatial resolution, which is necessarily limited in multi-millennial coupled runs. A discussion addressing the implications of this low resolution and how it might impact the balance between ocean-driven dynamic loss and atmospheric SMB loss would strengthen the paper.3. SSP Scenario Extension:
The simulation design is well suited to address the core research questions. However, the manuscript needs greater clarity on how the standard SSP scenarios were extended beyond 2100/2300 to the year 4000 CE. In particular, please clarify the exact role that CLIMBER-X played in determining the atmospheric greenhouse gas concentrations across the models for this extended period.4. Timescales of SMB Forcing and Atmospheric Variability:
The manuscript could provide a more explicit discussion of the relevant temporal scales governing SMB. This is highlighted by the use of dEBM, which relies on monthly mean input data while explicitly resolving the diurnal radiation cycle. This approach bridges two extremes but leaves out intermediate synoptic and weather-scale variability, which has been shown to strongly impact non-linear threshold processes and overall SMB (Zolles and Born, 2024, doi:10.5194/tc-18-4831-2024). In contrast, MPI-ESM utilizes hourly input forcing. Can the authors quantify or discuss how the choice of dEBM and the smoothing inherent to monthly averaging might affect the sensitivity experiments and calculated melt rates relative to higher-frequency forcing?5. Importance of Initialization:
The conclusion that initial ice-sheet topography (and initialization more broadly) plays a minor role in long-term divergence is only valid for the relatively narrow range of states tested here. Because all three models were initialized with very similar present-day configurations, the spread in initial geometry and thermal state is inherently limited. In a wider ensemble, or when comparing substantially different initialization techniques, initial conditions can introduce much larger uncertainties. The authors should explicitly caveat this finding and clarify that the conclusion reflects the specific, closely aligned initial states of the participating models rather than an exhaustive test of initialization uncertainty.6. Introduction and Scope of GrICMIP:
The acronym and initiative GrICMIP is introduced without sufficient background context. It remains unclear to the reader whether GrICMIP represents a broader, formally coordinated community initiative with an established protocol spanning multiple planned publications or if it is simply a project-specific acronym coined for this particular study and model trio.7. Structure and Readability:
Although the overall writing quality is very high, several sections contain dense blocks of text that make following the detailed model descriptions difficult. Breaking up longer sections into more and shorter paragraphs would significantly improve readability and flow.Citation: https://doi.org/10.5194/egusphere-2026-3384-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 116 | 88 | 14 | 218 | 11 | 9 |
- HTML: 116
- PDF: 88
- XML: 14
- Total: 218
- BibTeX: 11
- EndNote: 9
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General Comments
This study explores the reasons for substantially different long-term projections (i.e. past year 2100) of Greenland Ice Sheet evolution under anthropogenic forcing, using an ensemble of coupled ice sheet/climate models of various complexities. This paper provides a good example of intercomparison value: through a structured experimental design, among other results it highlights an important model process (SMB modelling) for the question at hand (long-term GrIS sea level rise contribution). This in turn should motivate the community to prioritize work on this arguably key process, rather than potentially less influential ones.
While I concur with conclusion that SMB/climate are very important to long-term GrIS trends, my primary general comment is that I’m not sure the present manuscript demonstrates this quite as conclusively as claimed. Particularly, because the sensitivity testing nominally meant to disentangle the climate/SMB/topography factors on GrIS response seem limited to 2100 and based on SMB model-only results, so results inferring their long term impacts on GrIS mass loss (a.k.a. sea level) remain somewhat just that, inferred. For example, over the 1900 remaining years of the multi-millennial simulations shown elsewhere in the paper, it could well be that topography differences emerge as a primary factor via the height/SMB feedback, or, that late-stage ice dynamical effects (e.g. shifting surface gradient patterns) become important. So in general I think greater justification is warranted during revisions, to try and highlight caveats related to using SMB-model-only, to-2100 behaviour to infer SMB/climate/topography impacts on multi-millennial timescales. That said, authors are welcome to push back on this general sentiment if they feel it is off-base.
Specific Comments
- Paper theme, and title: I think topic of this paper is, at its core, a targeted exploration of model uncertainty on long term GrIS change, using some sensitivity testing to do this. Maybe saying this clearly, in the title and elsewhere in the Abstract/Intro, would give readers a clearer cognitive expectation of the study contents. So for the title, instead of ‘Unravelling the source for diverging…’ what about ‘Partitioning model uncertainty sources behind diverging…’. Or even more directly: ‘Surface mass balance model formulation as a primary source for diverging…’?
- L2: ‘Since 1990…’ – suggest referring to an earlier, climatological period, outside of the transient historical period.
- L31-L~L38: ‘Almost exclusively…’, ‘…in its infancy…’, ‘…have been limited…’ I think the authors could possibly detune these claims a little bit. There has been a fair amount of model development/science over the years that falls into the categories described in these sentences. I don’t think these somewhat subjective claims on the history of the research field are necessary to frame the value of the current paper.
- L40: ‘By a factor of four…’ – would it be possible to convert this into a 0-order estimate of the difference in worldwide coastal inundation and/or displaced population (based on current population distribution)? This restatement would make this 4x factor immediately more relevant to decision making audiences, if formulated reasonably.
- Section 2: I think at the start of Methods, a paragraph or two is needed to very clearly restate, at a high level and in relatively plain language, the overall goal of the experiment, a statement of the experimental design, and the rational why the current experimental design achieves experimental goal. This is already touched on in the Introduction, but I think readers would benefit from having it again very directly and clearly restated at the beginning of Methods, so they have this context solidly in their mental RAM before they enter the ‘weeds’ of the different models and specific experimental protocols.
- Section 2.1.4: SMB is presumably quite dependent on the representation of snow/firn model in each of the models (e.g. multi-layer mixed-phase? Hydrologic/liquid water representation? Radiation-penetrating? Firn resolving?). More details on this, in this section, would seem a valuable addition. More generally, would it be possible for the authors to disentangle the effect of the EBM modelling differences and the snow/firn modelling differences, on the final result?
- Section 2.1.4: If different SMB models A and B were ‘perfect’ representations of real SMB processes (C), they would all also theoretically be functionally identical to each other (if A=C and B=C, A=B). So, following this line of reasoning, differences between SMB models stem from the different, necessary choices that each modelling group took to simplify infinitely complex reality down a tractable computer-based SMB model code. For this reason I wonder if this section could explicitly highlight, to the author’s views, the primary simplifications/caveats of each SMB model. Definitely not to criticize any of them per se. But rather just to help objectively provide the reader with the potentially primary reasons behind the different SMB model (and resulting long term sea level) results. And perhaps point to community-wide priority improvement areas.
- Section 2.1.4: AWI-ESM dEBM description: ‘the melt period is defined as the faction of the diurnal cycle during which incoming shortwave radiation exceeds outgoing longwave radiation’: but I am confused since I can easily imagine cases where no melting occurs, even if the former exceeds the latter (specifically, at very cold baseline temperatures, well below 0C). Should ‘melt period’ be called ‘potential melt period’ instead perhaps? Additionally, it’s not clear how PDDs are used, in this EBM architecture. This all presumably points to confusion on my (i.e. the readers’) part instead of dEBM modelling issues. In which case, perhaps a clearer description of the dEBM approach is needed, instead of just references to a secondary paper.
- L179: ‘…are performed with the different EBMs in standalone mode.’ So, to be clear, the core of the study is using static EBMs, uncoupled to either climate or ice sheet. In this case think it’d be worthwhile to make this very clear – and potentially, worth reweighting the methodology description towards a much deeper description of the EBMs, in lieu of the coupled climate models more broadly.
- L181. Somewhat by definition, the energy balance at the terrestrial surface (ice-covered or otherwise) will impact overlying atmospheric conditions in reality (and coupled climate models). This self-consistent nature is, arguably, a big benefit of coupled modelling. Imposing fixed atmospheric conditions developed with one coupled models setup (in other words, where the atmospheric conditions have a ‘signature’ resulting from existing above the particular EBM formulation) onto another EBM as prescribed forcing breaks or ‘decouples’ this link. As a conceptual example of an impact of this: the transfer of atmospheric heat content into latent heat absorption to melt ice cools the near-surface air, in reality or a 2-way coupled model. But this atmospheric cooling effect - reflecting conservation of energy - is presumably lost in an uncoupled setup where the air temperature is prescribed. In other words in this case, in this situation, some heat is effectively created, presumably resulting in some level of over overestimated localized melting..? Could the authors potentially describe or at least speculate what impact or signature this ‘decoupling’ could have on their results?
- Use of different scenarios: different SSP scenarios add an extra dimension to the overall analysis (including nonlinearities due to model and scenario-specific AMOC collapse), that I wonder might provide more confusion than value? Is including the complication of multiple scenarios actually necessary at all to meet the primary objective of the study? It seems like the approach of just focusing 100% on SSP245 (e.g. as in L266 and onwards) and completely excluding results from the other scenarios entirely, could be expanded study-wide, to achieve essentially all of the study’s goals. I.e. ‘Here we focus exclusively on results obtained using SSP245, noting that extensive sensitivity testing resulted in qualitatively similar results…’ Would this be feasible, in the author’s view?
- General: ensure that all acronyms are well defined. And in more general, I suggest doing an overall edit while taking the naïve perspective of a new reader. The authors may find, in emulating this mindset, that many short-hand terms that are intrinsically familiar to them might benefit from a bit more plain-language description (‘spin-up procedures’, ‘dome structure’, for example).
- Figure 2: A really interesting graphical depiction of model uncertainty. I think it’s worth noting that there are other, non-ice-sheet examples in climate modelling that display similar levels of specific intermodal heterogeneity (‘lack of consensus’) as part of coupled future projections. Like, changes in overall wind strength patterns, Amazon response, El Nino variability… I almost wonder if making a ‘metapoint’ in this paper, that the differences in GrIS response across models under the same scenario is in many ways consistent with sometimes quite significant differences in non-ice-sheet metrics evaluated in other multimodel analyses (including some citations to relevant, analagous studies). In other words, it’s not GrIS alone that does quite different things in different coupled climate models, for sometimes hard-to-understand reasons…
- Figure 3: the increase in calving anomaly here could be confusing to readers..? Essentially, the positive anomaly values here are indicating a less negative flux as calving reduces due to more land-terminating ice (right?). And this is being compared to less positive SMB budgets over time (the solid lines). I wonder if a clarification in the caption, or, a splitting into two plots (where calving flux is considered a positive number) would be useful.
- Figure 3: it appears that accumulation is periodically net-negative past year ~3000 for AWI-ESM and MPI-ESM. This seems like at least an integrated SMB accounting error – i.e. how can there be negative accumulation (that sort of implies upwards snowfall)? It perhaps points to an issue in the determination of different SMB terms for this plot..?
- L356-L357 – ‘…it becomes evident that different surface topographies cause the smallest differences in SMB.’ Do the authors mean, essentially the original spun-up topographies, have little effect on SMB spatial distribution at year 2100? This could be a bit confusing to interpret for readers, particularly as they ponder how increasingly divergent topographies (in response to different SMB fields) contribute in turn to even more divergent SMB mass trends, via the positive (amplifying) height/SMB feedback. Perhaps the authors could add some nuance here, to reflect this time dependency..?
- Section 3.5: I suggest separating into three different subsections (or at least, 3 different paragraphs). Especially since this seems like a central aspect of the study. It is really hard to parse this single ~40-line paragraph, that covers all three sensitivity assessments.
- Surface air temperature is actually not, itself, the direct driver of melting. Instead, it is a key regulator the energy fluxes between the ice sheet surface and the atmosphere; and, it is these fluxes that cause melting. Really, the advantage of EBM modelling, is that it captures this temperature/melt relationship in a physically consistent way, rather than relying on empirical relationships like PDDs. To that end, given the advertised importance of using EBM modelling, I think the paper could use a bit more nuance when tying GrIS changes to changes in air temperatures specifically. This nuance might uncover why, exactly, the SMB models are behaving differently under similar climate/topography conditions, as they ‘transform’ near surface air temperature into the melt response, via the EBM modelling structure. In general, I feel like this investigation would be a valuable contribution if it were possible, since it would point to priority areas for EBM modelling improvement for better sea level rise projections.
- ECS is a measure of global average sensitivity. But the additional, equally important ‘bridge factor’ that relates global ECS value to high latitude conditions that matter for GrIS, is polar amplification. For example, one can imagine a similar GrIS response, for reasonable high ECS/low PA versus low ECS/high PA models. In fact, this has been demonstrated in a controlled setting (https://link.springer.com/article/10.1007/s00382-014-2050-7, full disclosure, I am lead author of this somewhat dated paper). For this reason (amongst others), I’m not entirely surprised that, for example, models with similar ECS display quite different GrIS responses, and leads me to agree with the statement ‘underlining the critical role of the characteristics of the climate model for the future evolution of the GrIS’.
- Climate feedbacks (including those related to ice sheet evolution) ‘turn on’ as a function of the applied perturbation. So in this sense, scenarios with strong climate perturbations (e.g. SSP5-8.5) should display strong feedback response, and thus be reasonable places to investigate climate-ice sheet interactions (contra the hypothesis on L436, if I understand it correctly). More generally: if the goal is to study these processes, I would argue that simple step-function perturbation forcing scenarios (e.g. instantaneously doubled CO2, or 3x CO2 as in the above-referenced paper) are the most appropriate, because there is less conflation between gradual forcing changes, and ice sheet response. Recognizing that perhaps these exact experiments are not available for the current study, it might be nonetheless worth mentioning as a way to further understand relative roles of topography, SMB formulation, and climate, in long-term response.