This study explores how the Greenland Ice Sheet has changed over the past 24 kyr by running 100 high-resolution simulations and comparing them with real-world data on past ice margins. The results reveal that a major retreat happened between 16 kyr and 14 kyr years ago, mainly driven by ocean warming, even though the air was still too cold to cause much surface melting. No simulation from the ensemble is able to well represent both the Last Glacial Maximum (LGM) extent, deglaciation history and pre industrial extent of the ice sheet. Still, by identifying regional mismatches in model-data agreement, especially in northern and eastern Greenland, and by investigating the causes behind this mismatch, the study helps to improve how we simulate past ice sheet changes.

This is a well-written and well-structured study that clearly reflects the extensive work behind it. As a modeler I know how hard might be this kind of exercises and I congratulate the authors for their work. To my understanding, this is the first comprehensive paleo-modelling effort that simulates the full deglaciation of the Greenland Ice Sheet since the LGM. Importantly, it compares these simulations against a spatially distributed set of paleo extent records, which adds significant value and originality to the work. Undoubtedly, this work stands out as a valuable contribution to the field.

Still, I recommend some modification before being considered for publication in TC. 

General comments: 

• The use of the terms “ice-sheet extent” or “margin” could benefit from clarification: do you refer to the grounding line or the ice shelf front? Where and when are ice shelves present in your simulations? This distinction is important, particularly given that sub-shelf melting appears to be a key driver of mass loss in the early Holocene. While most of the discussion and figures clearly focus on the grounded ice sheet, knowing the extent and presence of ice shelves is essential. For example, plotting the ice shelf extent in Figure 17 would be helpful. The presence of ice shelves could delay the onset of deglaciation due to their buttressing effect, and this deserves more attention.

• The manuscript is quite long, and the detailed descriptions of individual simulations (e.g., the best LGM and best deglaciation simulations) could be streamlined. You might consider focusing on the top five simulations after the full selection process and moving some other figures and discussions to the Supplementary Material. This would help improve the readability and flow for the reader.

• I’m surprised that sub-shelf melting is not explored within the ensemble, despite the large uncertainties in SST forcing. Unlike for surface air temperature, no parameters in the sub-shelf melt parameterization are varied, nor is any uncertainty in SST explicitly addressed. This is particularly striking given that the retreat at the onset of the last deglaciation is attributed primarily to oceanic forcing. Statements such as: “Between 16 and 14 thousand years ago, the ice sheet lost most of its ice grounded on the continental shelf. This marine-sector demise, associated with up to seven times greater mass loss rates than observed today, was predominantly caused by ocean warming while air temperatures possibly remained too cold to generate surface melt” seem premature in the absence of a more thorough investigation of the sub-shelf melt scheme and its sensitivity to oceanic conditions. For example, how robust is the timing of SST increases around 16 ka? This uncertainty deserves more discussion. See also my specific comments.

• Many of the model-data misfits likely stem from uncertainties in the climatic forcing. I won’t elaborate further on this point, as it has already been discussed in detail by another reviewer. However, I do agree that it would be helpful to compare your climate forcing, particularly air temperature and precipitation, with that used in Badgeley et al. (2020). This could provide useful context and help assess the robustness of your forcing choices (see also specific comments).

• Your simulations do not show an expansion of the GrIS in the northeastern sector during the LGM, despite several studies suggesting that the margin likely reached the continental shelf break. It’s true that many models struggle to reproduce this feature, and you do acknowledge this limitation in the text. However, I believe the discussion could be strengthened by incorporating some of the additional points I’ve outlined in the specific comments.

Specific comments:

Line 233: please, write the equation with \cdot.

Lin 255: Could you describe the simple hydrologic model from Tulaczyk et al., 2000 with more detail? how is the N_till calculated?

Line 375: Please add in this paragraph that temperature and precipitation used to force the SMB model will be described in section 2.1.3. 

Line 391: in section 2.1.3 you describe the SST and salinity signal used to force the sub-shelf method, which is only briefly described here. Please, add a section describing the sub-shelf melting scheme in more detail (the basics of the three equation formulation of the ocean fluxes, if I understand correctly) and how are the SST and salinity taken into account in this formulation. I guess that what’s especially interesting here is the fact that this method takes the SST as a forcing to compute temperature and melting at the base of the ice shelf. 

Line 423: Figures are generally far away in the text from where they are cited. Please, try to put them closer. Also, how do you “combine” the two GIA (local + non-local) signals? Simply by adding them up?

Line 483: what do you mean by “reference height air temperature and precipitation”? Do you mean you apply a lapse rate to correct the surface air temperature with respect to the topography? How is this lapse rate applied to precipitation if there is one? 

Figures 4-8: I would suggest to restructure the figures of climatic forcing so that they are separated from those from sea surface temperature. In that case you’d have figure 6 with precipitation only or you could add a third panel to figure 4 with precipitation fluxes and fig. 6 would only have SST. 

Line 488: which is the temporal resolution of the transient iTRACE experiment?

Line 495: Buizert et al., 2018 provides transient spatially variable fields for temperature since 21 kyr BP. How do you build the 1D glacial index from these 2D fields? Do you compute a different climatic index for every 5x5 km cell across the transient 2D fields, otherwise I don’t get how would you end up with a different signal for each location (Figure 5). Please specify. Also, could you show the precipitation time series for different ice core locations using the same index approach? How does this compare with Badgeley et al., 2020? 

Line 512: surface salinity: do you mean here you take the 2D equilibrium iCESM simulations for 21, 11, 9, 6, 3 kyr BP and PI and interpolate them linearly?

Line 559: Fixing the parameters to their mid values for the spin-up and then vary them for the ensemble transient runs might bring to some inconsistencies in the first years of simulation after 24 kyr BP. I believe this is not a crucial point, but I would like to see a sentence that discusses this. 

Figure 11: could you highlight the best 5 simulations in both panels? Could you also add a mark for the present grounded area (1.7 10^6 km^2)?

Line 849: please add the reference O Cofaigh et al., 2025 “Shelf-edge glaciation offshore of northeast Greenland during the last glacial maximum and timing of initial ice-sheet retreat” which further supports the maximum extent of the NE Greenland during the LGM. 

Figure 12: I think “basal mass fluxes” should be better replaced by “sub-shelf mass fluxes” to avoid confusion between that and grounded basal mass balance. 

Line 1077-1081: I would be careful about this paragraph. You don’t model a central ice divide migration for NGRIP, but your simulated LGM maximum extent in the NE is significantly underestimated. In fact, as you noted earlier and as several recent publications suggest, the expected extent should likely reach close to the continental shelf break. Although previous modelling work seems not to suggest that the NGRIP summit was migrated during the LGM (e.g Tabone et al., 2024), this discrepancy might imply a more stable and less elevated ice divide in the central-north central region of the ice sheet, even if that may not have been the case. Also, a thickening of the NE stream at the LGM is found from geomorphological records (Lane et al., 2023). This is somehow seen in figure 15, panel b, but it might well be underestimated due to the limited margin extent. Please, add some comments on this. 

Figure 15: could you make the maps within the panels bigger?

Lines 1120-1134: yes. I would add some comments on the ice discharge in the northeast. Although recent radar measurements suggest that the upper part of the present-day NEGIS was fully developed only during the last 2000 years (Jansen et al., 2024), there is the evidence at least of a paleo ice stream that was flowing before and likely into the Holocene in the northeast (Franke et al., 2022) and this is not captured in your LGM simulations. This suggests that the northeast region of the ice sheet could have been more dynamic as your simulations show.

Line 1457: as you pointed out earlier, this is due to the onset of sub-shelf melting around 16 kyr BP in some regions? Could you add some comments on the reliability of the temperature forcing in these regions from Osman et al., 2021? Is there any comparison you could make between these simulations and available paleo records, or was it done by Osman et al.?

Line 1538: I am not surprised that thinning can be fairly well reproduced for the last 8 kyr BP as generally what is hard to replicate/explain is the thinning at the early Holocene due to the demise of the IIS, LIS, as well as big uncertainties in the climate forcing and ice-sheet dynamics. Please add a sentence commenting on this and citing Lecavalier et al., 2017 (for Camp Century) and Tabone et al., 2024 (for NGRIP). 

Line 1638: I would add here a discussion on how your transient precipitation forcing compares with Badgeley et al., 2020, as they suggest different low- to high- precipitation scenarios for the early Holocene. 

Lines 2045-2054: there is a modelling work able to reproduce the retreat of the NE Greenland since the LGM fairly well, from a fully expanded NE Greenland to its present day margin (Tabone et al., 2024). In their work the onset of the paleo NEGIS at the early Holocene is key for the NE retreat. Some comparison between your work and theirs could be useful to investigate the causes of your run-data misfits during the deglaciation.  

Robert et al., 2024a and Roberts et al., 2024 b are the same publication. Same for O Cofaigh et al., 2023 a and b. Please correct. 

References:

Badgeley, J. A., Steig, E. J., Hakim, G. J. & Fudge, T. J. Greenland temperature and precipitation over the last 20 000 years using data

assimilation. Climate 16, 1325–1346 (2020).

Cofaigh, Colm Ó., et al. "Shelf-edge glaciation offshore of northeast Greenland during the last glacial maximum and timing of initial ice-sheet retreat." Quaternary Science Reviews 359 (2025): 109326.

Franke, S. et al. Holocene ice-stream shutdown and drainage basin reconfiguration in northeast Greenland. Nat. Geosci. 15, 995–1001

(2022).

Jansen, D., Franke, S., Bauer, C.C. et al. Shear margins in upper half of Northeast Greenland Ice Stream were established two millennia ago. Nat Commun 15, 1193 (2024). https://doi.org/10.1038/s41467-024-45021-8

Lane, Timothy P., et al. "The geomorphological record of an ice stream to ice shelf transition in Northeast Greenland." Earth Surface Processes and Landforms 48.7 (2023): 1321-1341.

Tabone, I., Robinson, A., Montoya, M., Alvarez-Solas, J., "Holocene Thinning in Central Greenland Controlled by the Northeast Greenland Ice Stream", Nature Communications, 15, 6434 (2024). https://doi.org/10.1038/s41467-024-50772-5