Goelzer et al. present the results from a multi-model ensemble of mass-loss/sea-level projections for the Greenland ice sheet, carried out under the PROTECT project. Compared to the Greenland projections for ISMIP6 (Goelzer et al. 2018, 2020), this study includes many improvements, notably (1) the projections begin from simulations that have run through the historical period, (2) global climate model output is downscaled by multiple RCMs, (3) the intermediate 4.5 scenario is included, and (4) some of the runs extends beyond the end of the century to 2300. Although there are fewer participating models than in ISMIP6, the other improvements make this study a valuable step forward. It is likely to be the most comprehensive set of Greenland sea-level projections available between now and when ISMIP7 results are published.

The paper is well organized and clearly written. It presents results in an accessible way without extraneous details. It’s possible to read the entire paper and come away with the key messages in an hour or two.

My main critique, which will be straightforward to address, is that some of the important results and conclusions do not appear in the Abstract or Section 5. Thus, busy readers (practitioners, for instance) won’t find the bottom-line results easily and might be left with questions. For example, the upper end of the 2300 range (3.1 m) is nearly half the total ice-sheet volume and is comparable to the most pessimistic projections for the West Antarctic Ice Sheet. However, this value assumes an extreme emissions scenario that arguably has a very low likelihood. Adding some caveats will reduce the chance that results will be misinterpreted.

Specific comments

l. 19: The Abstract does a good job of describing what was done, but it leaves out some key results. For example, one main result is that the projections are very sensitive to the climate scenario (as expected), moderately sensitive to the RCM choice, and relatively insensitive to the ice sheet model choice. The relatively high RCM sensitivity may be surprising for readers not familiar with Glaude et al. (2024). Also, I suggest stating the 2300 high-end value not only for the extended SSP5-8.5 scenario, but also for the projections with 2100 repeat forcing. Arguably, the repeat-forcing scenario is more realistic and thus more relevant for practitioners than the extended SSP5-8.5 scenario.

l. 55: This paragraph describes an important advance on ISMIP6, responding to the criticisms of Aschwanden et al. (2021). Are you able to say (not here, but later in the paper) how much difference this change makes, compared to initializations that do not include a historical simulation?

l. 95:  Please state the equation that describes this parameterization.

l. 99:  Can you say how the various percentile values of kappa were determined? In particular, how was the 5% value derived? The language suggests that there is a 5% probability that kappa is at least this high, but I suspect there are deep uncertainties here.

l. 229:  The paper doesn’t say how well the 2015 initial ice-sheet geometries compare to observations (e.g., BedMachine ice thickness). I suggest adding a map-view figure showing (1) the observed ice thickness from BedMachine and (2) the thickness error at 2015 for one ensemble member for each ISM, perhaps indicating on each panel the rms thickness error over the ice-covered area.

Optionally, you could also add a graph or table comparing (for each ISM) the simulated mass change over the historical period to the estimated observed mass loss.

l. 236: The text cites O’Neill et al. (2021) for the 2300 scenario. This should be O’Neill et al. (2016). I suggest stating that the maximum CO₂ concentration for this scenario is about 2200 ppm (see Fig. 5b in that paper), roughly double the value at 2100. This explains the large differences in 2300 between the repeat-forcing experiments and those based on ScenarioMIP.

l. 255:  “another problem on this timescale may be that the climate response to changing ice sheet geometry is not properly accounted for”. Isn’t this also true for the “natural extensions”? The IPSL and CESM2 extensions weren’t run with interactive ice sheets, so in neither case would the ESM simulate the climate response to changing ice sheet geometry. Or is there a subtlety I’m missing?

I want to suggest that the issues with schematic repeat forcing are not necessarily worse than those associated with CO₂ of 2200 ppm. Neither scenario is fully realistic, but the two approaches are complementary and are probably the best we can do with the CMIP6 output we have.

l. 323:  Can you say why the SMB in RACMO is less negative for future projections than those in MAR and HIRHAM? This is discussed by Glaude et al. (2024), but it would be helpful to give a summary here.

Figure 6:

• Please state in the caption that this is for 2100.
• In panel c, I suggest showing the ranges separately for 8.5 repeat forcing and 8.5 extended forcing, since the former is arguably more plausible than the latter.
• It would be helpful to use a different symbol for the CISM runs with extreme values of kappa – e.g., an open green circle instead of a closed green circle. This would confirm that the range of CISM results is close to that of the other ISMs when large kappa is excluded.

Figure 8: I found this figure confusing because there are experiments of the same type on both sides of the dashed lines, but it’s hard to compare them due to the different vertical scales. I suggest using a different color for each type of forcing: o2300, r2300, x2300, and e2200. Another possibility might be having a separate panel for each type (maybe combining e2200 with x2300), with a different vertical scale for the natural extensions.

l. 371:  Please state the ranges for Goelzer et al. (2020) and Payne et al. (2021). It might be worth noting that the Payne et al. ranges are much larger than those in Goelzer et al., and similar to the ranges in this study, because of the high warming in some CMIP6 models compared to CMIP5.

l. 374:  “A solid number” is vague; can you make this more quantitative?

Somewhere in the Discussion, I suggest commenting on the value of a “mini-MIP” like this one.  Does it save time, without diminishing the results, to run just four models instead of a larger number? Other MIP groups might see this study as a template for efficiently updating projections between IPCC reports when resources are limited.

I also suggest stating that the lack of climate feedbacks (as would be present in an ESM with interactive ice sheets) is an important limitation of the study.

l. 411:  Since some readers will skim the Abstract and then go straight to the Conclusions, I suggest adding a paragraph summarizing the main results. I think some redundancy with the Discussion is okay. For example:

• Give the total GMSL ranges for the different scenarios and time frames.
• Restate the result that under high forcing, the ESM choice is the largest contributor to the uncertainty, followed by RCM choice and ISM choice. The rather large sensitivity to RCM choice points to the need for better understanding of the downscaling.
• You might add some thoughts on how ISMIP7 will build on this study.

Minor fixes

Please make sure there is a carriage return between each paragraph, or some other indication of a paragraph break.

For figures, put just the label (e.g., Figure 1) in boldface, instead of the entire caption.

l. 87 I think it’s more common to denote surface temperature by ‘TS’ than ‘ST’.

Table 1: Change “RCP5.8” to “RCP8.5” in the header.

l. 126:  No hyphen in “grid cells”

l. 162:  CO2 -> CO₂ (also in l. 164)

l. 174: The units here are not formatted correctly.

l. 350:  6% (no space)

l. 356:  Add a comma after “scenario SSP1-2.6” for clarity.

l. 423:  I think a word is missing after “text”.

l. 427:  “heigh” -> “height”

l. 454:  “in function” -> “as a function”?

l. 459:  The “/ 4” means that a quarter of the anomaly was applied? Maybe change the wording for clarity.

l. 460:  “compared” -> “compared to”

l. 491: This is a sentence fragment; it needs a verb.

l. 495: Extraneous decimal point in “SSP5-.8.5”

The authors present results from a multi-model ensemble of future Greenland ice sheet projections carried out within the PROTECT project. This study includes several improvements compared to the ISMIP6 simulations, i.e., it includes different regional climate models for the downscaling (compared to just one in ISMIP6), the protocol has been extended over the historical period, and intermediate CO2 scenario runs have been included. The authors present ranges of potential sea level contributions with respect to the scenarios, regional climate models, and ice sheet models used. This study clearly adds nicely to the ISMIP6 results and comprises valuable results for assessing the potential behavior of the Greenland ice sheet under future warming and the uncertainties associated with it.

The paper is well structured and well written. The figures are of high quality and informative.

General comments:

• I think it would be helpful for the reader to include additional figures regarding the initial ice sheet state and the climate forcing used. Time series of the global mean temperature would allow for assessing the effect of the ESM’s climate sensitivity on the ice sheet response. As one huge advantage of this study compared to ISMIP is the historical simulations, I think it would be helpful to include time series of the SMB for different models and/or a map of the initial ice sheet showing how well current observations of the ice sheet extent are captured by the ensemble.
• The authors state the importance of emulators and that their model results are geared towards emulator use. However, I miss some more explanation of this. As I understand, the ensemble has been updated with additional experiments to meet the needs of emulators. Except for this, were there additional requirements for the model output to be met in order to facilitate the use of emulators?

Specific comments:

L28f: Given that a large share of the spread is due to different scenarios, it might be more informative to give the GMSL spread here scenario-specific.

L95: Please consider showing the function for the parametrization here.

L99: After reading Slater et al. (2019), it became clear to me what you mean by “can be expressed probabilistically and is sampled …”. However, I think it would be helpful to include the PDF here and indicate which values you take for your simulations.

L125: “GIMP DEM”: not mentioned earlier. Please explain the acronym.

Table 2: Please avoid the page break in the table.

L134: Should be Appendix A, I assume.

L247: Should be Appendix C, I assume.

L264: Should be Appendix C, I assume. Here, I hoped to see some time series of the forcing data, e.g., global mean air temperature.

Figure 5: Given the large spread of the CISM results, would it be possible to use different markers for the different retreat parametrizations?

L366: Should be “rightmost” instead of “leftmost”