the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Century-scale impacts of ice-sheet model initialization on Amundsen Sea Embayment, West Antarctica
Abstract. The glaciers of the Amundsen Sea Embayment (ASE) are some of the fastest-thinning in Antarctica. Future ice loss from this region depends on the trajectory of ocean warming. However, the response of glaciers to this warming depends to some extent on how they are initialised, or calibrated, to match observations. The relative importance of these dual factors of forcing and initialisation remains poorly understood. We carry out climate scenario-forced, synchronously coupled ice-ocean simulations of the ASE extending to the 23rd century. We conduct four experiments varying ice-sheet initialisation method (Snapshot versus Transient Calibration) and far-field ocean forcing (baseline climatology versus an RCP8.5 scenario). We find that the mode of ice-sheet initialisation dominates the trajectory of Thwaites Glacier volume loss throughout the 21st century, while climate forcing emerges as the primary control in subsequent centuries — a contrast explained by the low buttressing of Thwaites ice shelf prior to substantial grounding-line retreat. Under RCP8.5 forcing, Thwaites alone contributes up to 2.6 mm/a sea-level equivalent by 2200 following its retreat past Upper Thwaites Ridge, with the full ASE exceeding 3.2 mm/a. Under current ocean conditions, loss rates stabilise near 1.7 mm/a, partly due to growth of pinning points due to a thicker ice shelf compared to RCP8.5. Our results suggest that ASE ice loss remains sensitive to climate forcing even after retreat into the deep interior, implying that emissions mitigation could delay the trajectory of sea-level rise from this region – but that the effects of mitigation would not be felt for nearly a century.
- Preprint
(3340 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-3779', Anonymous Referee #1, 09 Sep 2026
- AC1: 'Reply on RC1', Dan Goldberg, 25 Sep 2026
-
RC2: 'Comment on egusphere-2026-3779', Rupert Gladstone, 18 Sep 2026
Reviewer comments on "Century-scale impacts of ice-sheet model initialization on Amundsen Sea Embayment, West Antarctica"
My limitations:My own research focusses on ice dynamic modelling and coupled modelling. I have a basic knowledge of ocean models, but I do not consider myself qualified to evaluate the ocean modelling methodology.
My review is late! Sorry! Several minor health issues filed in gaps between hard deadlines when I had intended to complete my review.
General comments.
This study takes the current state of the art of coupled ice sheet - ocean modelling and applies such a model to multi century projections of the Amundsen Sea Embayment in West Antarctica. While the coupling is justified, the impacts of coupling specifically are not assessed in this study. Instead, the impact of ice model initialisation is assessed in the context of a coupled model. Two approaches to ice model initialisation are compared, and their impacts on multi century projections are compared. The differences are significant, highlighting an important area of ice sheet model based projection that needs attention in the international community.
Sea level projections for PIG, Thwaites and the whole ASE are given for a base scenario and a very high climate warming scenario. These can be consiered as plausible point projections which contribute usefully to the sea level projection literature.
The impact of ice model initialisation is compared to the impact of climate forcing, and an index created to represent their relative importances. It is important to note that this index is very much specific to the design choices of the current experiment, and should not be generalised. The two approaches to model initialisation presented here do not necessarily span the range of possible ice model drifts imposed by initialisation methods in ice models, whereas the two climate forcing scenarios should be considered as end members of viable futures. For this reason, a mroe generic version of the metric presented here would probably give even higher weighting to the uncertainty arising from ice sheet model initalisation, giving a stronger result than that presented here. But however one views the metric, the results presented here highlight that ice sheet model initialisation is an important and poorly quantified problem in projections of ice sheet contributions to sea level.
RCP8.5 is a rather extreme forcing scenario. I'm not suggesting that you carry out additional runs at this point, but it would have been interesting to see a comparison between a control and a more intermediate scenario. Persumably this would increase the period over which the initialisation has high relative importance.
The fact that the simulations crashed with numeric instabilities is a bit concerning. I haven't seen evidence that the authors looked into whether these fatal instabilities were manifest in a non-catastrophic way earlier on in their simulations, potentially influencing the interpretation of results. I don't see a description of what the authors tried to do to mitigate these instabilities, perhaps with shortened timesteps or geometric smoothing for example. I don't see an anlysis of where thy occurred - were they all triggered from within the Thwaites catchment? That might increase confidence in the results for other catchments.
I didn't find a description of calving in the ice model. I presume that calving is not included in this model [edit: this is first mentioned in 4.4, but I would like to see it mentioned in the methodology], which raises the questions: What happens when a lot of thinning occurs? Is the shelf allowed to thin to zero or is there a minimum thickness imposed? Does this impact on buttressing? Separately, on the ocean side, I presume you force the ocean from CESM2 outputs at the surface (I wasn't sure whether your description of ocean forcing applied only to the vertical boundaries or also the upper surface?), in which case you are affecting future ocean behaviour by keeping the ice fronts fixed and not allowing upper surface forcing to evolve?
Compass directions are discussed in the text, but not shown in the figures. Roughly speaking, north is left, but this is not precise and a reader not familiar with the domain might assume north is upward. I suggest you add a compass graphic in Figure 1 to orient the reader.
Line by line comments.
Line 45. While coupled ice - ocean modelling is likely the best way forward, more advanced parameterisations for sub shelf melt, such as machine learning based parameterisations (e.g. Rosier's work) or 2D ocean models (e.g. LADDIE) may offer viable compromises, especialy given that high fidelity ocean modelling can be prohibitively expensive for large scale coupled projections, leading to resolution compromises. I agree that fully coupled modelling is the current gold standard, but not necessarily the "only" option going forward.
Line 59. Here you talk about SSP but in the abstract you talk about RCP. RCP is mentioned later on too. I suggest you search and ensure consistency.
Figure 1. Surely the 85 simulations have more retreat than the base simulations? In which case, why are the base ocean domains expanded at 2200 but not the 85 domains?
Section 2.1.3. This seems like an important section. I donÄt see any discussion of errors in the observed product. I havent read the Gourmelen paper. But it says it is based on remotely sensed data, and seems to cover the whole PIG shelf, so I guess it is mainly satellite data. If this is based on surface elevation change and flux divergence calculations, it's accuracy may be dependent on observational accuracy, displacement from floatation, and density variations. The region near the grounding zone is where the flotation assumption is least accurate. Are you sure that the observed melt rates are correct here? Is there a chance you might be putting effort in tuning your melt to an inaccurate product? It looks like you end up with something that underestimates melt near the GL but overestimates it elsewhere. It is hard to estimate the impact of thie error on the uncertainty in your final results.
Equation 3. I am missing an explanation of alpha squared? Maybe alpha squared is the Coulomb limit and beta squared is the Weertman coefficient? What value is used for alpha squared?
Line 198 "timd" typo
Figure 4. The elevation rate comparison (a to c) is very clear. But the velocities are dominated by the base pattern rather than the difference to observations. How about showing difference to observed in e and f? Do you think that would convey the differences more clearly? Can you add the reference for 4a to the caption? Seems unfair to reference the velocity obs but not the elevation rate obs at this point...
Section 2.3. This seems like a good choice of coupling strategy, but it is slightly less tight than I was expecting given the description in the abstract that the ice and ocean were "synchronously coupled". Maybe it is ok to leave it like this, and certainly the strategy is justified. Perhaps it would help the community to define terminology for different levels of synchronicity. This feels to me to be "almost synchronous"...
Line 240. OK, the ocean domain was expanded when needed. But the ice mesh stayed the same as I understand. Did the GL ever retreat into the coarser resolution region? It looks like it got close for Thwaites... could this impact on retreat rates after this point?
Line 250. typo "oecean"
Line 253. Can you be slightly clearer about this? I think you present your outputs up until the point of simulation failure. Can you state this specifically in this paragraph? Before I looked ahead at the results figures I didnt know how to interpret this paragraph.
Figure 5 caption typo. Forgot to include (g)-(i).
Figure 5. I guess the melt rate smoothing was due to the actual rates being very noisy. Is that noise purely internal variability in the ocean, or is some of it coupling artefact, e.g. jumping as GL retreats, perhaps the instability you mentioned above? The reader naturally wonders whether non-physical artefacts occurred long before catastrophic failure of the simulation.
Figure 6b is very noisy. It isn't clear whether the 6 month tiem smoothing was applied to the melt rates here or not. If it wasn't maybe apply it? If it was, maybe try a longer time to smooth it more? And maybe also increase the upper limit of the y axis?
Figure 6. I don't know why you have a solid line at y=0.75. I don't see the significance of this number. I suggest to remove it.
Figure 8 the y axis label is cut in half.
Figure 8. Has the PIG/Thwaites catchment boundary migrated? Or was it always slightly out compared to the model?
I infer Figure 9 is also at the point of failure (though the GL speed represents the history). This doesn't seem to be explicitly stated. Add to the Fig 9 caption for clarity?
Line 324. Well, it is hardly going to occur after model failure...
Figure 11b and c. Leaves me wondering how much this is changing relative to the total grounded area, which is presumably pretty small since we're just looking at a couple of pinning points. From this perspective I think it would be more informative to show the absolute (rather than relative to 2267) grounded area. You don't have to include y=zero on the axis, so the plot would look the same, but the axis would be slightly more informative. Especially c. You can't tell in Fig 10 how much of the pinning point has gone. Has it halved? More Less?
Section 4.1. Dynamic regrounding (downstream advection of thicker ice in response to accelerated flow; see also https://tc.copernicus.org/articles/12/2425/2018/) may be more likely in the base than RCP85 experiments due to the differences in melt rates. Might be relevant here.
Figure 12. Since the buttressing number is calculated in the ice flow direction, a gentle smattering of flowlines might aid the reader in interpreting this plot.
Figure 12. Idle thought: I wonder if you can combine your melt rates with the buttressing number somehow to come up with a "melt impact" number...
Lines 385-386. While I like the overall context and use of the buttressing number as a proxy for sensitivity to ocean induced melting, the wording here underplays the potential of the grounding zone to control ice dynamics. Technically "upstream of the GL" could include the elements/grid cells immdeiately connected to the GL, but that's probably not the implication most readers will take from this. I suspect that the parameterisation of drag in the grounding zone, along with the way in which tuning determines the spatial pattern of these parameters, can have a very strong impact on the model drift once transient simulations start. See also this paper, in which different ways of implementing effective pressure in regularised coulomb type sliding laws has a massive impact on projections of Thwaites catchment (and we found the differences in basal drag resulting from these choices were highest in the grounding zones): https://www.nature.com/articles/s41467-025-58375-4
Line 420. If I were you I would probably be tempted to go one step further here, and suggest that any theshold for instability must be defined in terms of both climate forcing and GL position/geometry (the "traditional" thrshold perspective for MISI).
Line 423. This confuses me. Both "algorithmic uncertainty" and "model uncertainty" sound like "structural uncertainty" to me and neither sounds like "parametric uncertainty". Structural and paramteric uncertainties are the two modes of uncertainty I am familiar with in the context of model based experiments designed to quantify uncertainty.
Line 429. Typo biase
Line 449. Missing "ice"? i.e. "advection of less damaged ice"?
Line 467. Typo? full stop in the brackets should be comma or semi colon?
Lines 468-469. This is a very Thwaites-centric comment, and to my mind not quite strong enough based on what you've shown. Surely the more complete statement is that the benefits of climate mitigation are indicated by this study to be very strong on a multi century timescale, but that these benefits become apparent at different rates for different catchments. The implication being that if you want to assess the benfits over many generations you need to be careful in your choice of which catchments to look at to see the benefits.
As for your final line, yes, sure, but the more specific point here is that ice model initialisation practices tend to tune a model drift, and that this drift can impact sea level projections. This study highlights how important the ice model intialisation is, whether you have a coupled system or not, and while coupled ice ocean models surely do need to be improved, that general comment isn't really about what you've specifically shown here.
Citation: https://doi.org/10.5194/egusphere-2026-3779-RC2 -
AC2: 'Reply on RC2', Dan Goldberg, 25 Sep 2026
Please see the response as PDF in the supplement.
Citation: https://doi.org/10.5194/egusphere-2026-3779-AC2 - AC3: 'Reply on RC2', Dan Goldberg, 25 Sep 2026
-
AC2: 'Reply on RC2', Dan Goldberg, 25 Sep 2026
Data sets
Code, Input, and Outputs for Century-scale impacts of ice-sheet model initialization on Thwaites Glacier, West Antarctica, Daniel Goldberg https://doi.org/10.5281/zenodo.20796236
Model code and software
Code, Input, and Outputs for Century-scale impacts of ice-sheet model initialization on Thwaites Glacier, West Antarctica, Daniel Goldberg https://doi.org/10.5281/zenodo.20796236
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 180 | 111 | 33 | 324 | 25 | 29 |
- HTML: 180
- PDF: 111
- XML: 33
- Total: 324
- BibTeX: 25
- EndNote: 29
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
see attached pdf file