Smith et al. present an important study using ICESat-2 altimetry data (ATL11 data) to evaluate three surface mass balance and firn models (MAR and two GSFC models) over the Greenland Ice Sheet. These two processes have been important to be considered when using altimetry data to estimate ice mass balance, in particular, to separate ice mass changes from firn compaction based on volumetric changes. Seasonally repeated surface elevation measurements from ICESat-2 provide an excellent opportunity to evaluate the firn models over an ice-sheet-wide scale which had been validated using sparsely distributed firn cores. The authors thoroughly compared ICESat-2-derived height changes with model-estimated height changes caused by surface mass accumulation+ablation and firn compaction. There are several points that need to be clarified and/or discussed. See comments below:

This study considers the surface height change anomalies and SMB/FAC anomalies over the areas with little variability of flow velocities. Although the ice-dynamic induced height changes (anomalies) can be neglected, how would the variations of local topography/roughness with (fast) ice flows affect the evaluation? This may have little impact for large-scale evaluation when the data are aggregated to a coarse resolution grid, but it would be good if the authors can comment/clarify on this point.

Correction of firn compaction has been a critical step when using altimetry data to estimate the ice mass changes. RACMO has been more widely used in literature to correct for this effect. Although it may fall out of the scope of this study, it would be very helpful for the community if the authors can comment/discuss the RACMO firn estimates as well.  

Line 19. Specify the names of the three FD/SMB models evaluated in this study.

Line 22. Specify the names of the two models mentioned here.

Line 25. Specify the name of the third model here.

Line 186. Why did the authors use MARv3.5.2 for this step? How would the difference between MARv3.11.5 and MARv3.5.2 affect the evaluation? The reasons and potential biases should be clarified.

Section 2.3.1. This part (especially the first two paragraphs) is difficult to follow. Could the authors use some equations to explain the regression analysis done here?

Section 3.2.3. This part is hard to follow too, with those scaling parameters and standard deviations. It would be helpful if the authors wrote some summary/topic sentences at the beginning of this section.  

Line 465. “..but the melt for GSFCv1.1 was based on a degree-day parametrization of the MARv3.11.5 melt..”. Here is confusing. Did the authors use MARv3.11.5 or MARv3.5.2 to calibrate the degree-day model? MARv3.5.2 was mentioned in the methods part.

This study evaluates output from 3 surface mass balance (SMB) and firn density (FD) models using ICESat-2 altimetry data over the Greenland Ice Sheet. Previously, SMB and FD models had been validated using dispersed firn cores or airborne radar surveys that were widely spaced. ICESat-2 surface height measurements provide seasonal temporal resolution of net surface height changes across the entire ice sheet, and provides a useful dataset for evaluating SMB and FD model performance. Smith et al. compare ICESat-2 height changes with simulated height changes from 3 combined SMB/FD models that can account for atmospherically-derived height changes and compaction-derived surface height changes. Because ice sheet surface height changes can be caused by thinning from ice dynamics, the authors focus on regions of the ice sheet with little variation in flow velocities. This is a very thorough and detailed study that will be very valuable to the firn community. The paper is well-written and the figures are well-presented. This paper is so detailed that it would benefit readers to provide more general points at the beginning of each section for them to latch onto, particularly if they are looking for main points of the analysis/results. I have a few suggestions that I think should be incorporated before publication. These comments are listed below:

ICESat-2 began measurements in October 2018. MacFerrin et al. recently published a firn compaction dataset, and I believe 2 of the sites have compaction measurements through 2019. It may be outside the scope of this study, but it may be interesting to compare ICESat-2 surface-height changes and modeled surface height changes at these two sites to examine the influence of firn compaction/atmospheric inputs to surface height changes and see how well the models capture these.

I would have liked to see the 3 models introduced earlier. It would be nice to list them in the abstract (e.g. Line 19) and in the introduction (e.g. Line 75).

It would be nice to clarify why you evaluate MARv3.11.5, but calibrate your degree-day parameterization in the GSFC model using MARv.3.5.2 (Sections 2.2.1 and 2.2.2).

The regression analysis sections are quite detailed, and a bit difficult to understand (which may be my own problem). In Section 2.3.1, I did not quite get the point until you gave the example of scaling SMB by 0.5 (Line 254). It may be useful in this section to give a summary how the regressions are used for readers to then understand the more detailed methodology. I believe this would also be useful for Section 3.2.3

Lines 282-284: Can you make this sentence clearer? You say “we can see that melt was considerably stronger in 2019 than it was in 2020”. Can you specify where in the table we look to come to that conclusion? It is difficult to find in the table by keeping track of the variables.

Lines 466 and 517: Here you say that  melt for GSFCv1.1 was based on degree-day parameterization of the MARv3.11.5 melt. Earlier in the methods section you mentioned that it used MARv3.2.5. Could you clarify this?

Line 500: What about using these models to predict ice sheet mass changes in the future using these SMB/FD models? It seems important that these models overpredict FAC changes associated with high-elevation melt events, which will likely be more frequent in the future.

Figures 4, 6, 7: Could you write out what each colored histogram represents in the figure caption? That would have benefited me while reading.

MacFerrin et al. citation:

MacFerrin, M. J., Stevens, C. M., Vandecrux, B., Waddington, E. D., and Abdalati, W.: The Greenland Firn Compaction Verification and Reconnaissance (FirnCover) dataset, 2013–2019, Earth Syst. Sci. Data, 14, 955–971, https://doi.org/10.5194/essd-14-955-2022, 2022.

Smith et al. take up the challenge to assess the performance of combined SMB-FD models against laser-derived observations of elevation change over parts of the Greenland ice sheet where they assume ice-dynamical effects to be negligible. This is important, as it allows us to understand how to improve altimetry-based estimated of GrIS mass balance using firn and SMB models.

The paper is clearly written (in most places), and the scientific analyses are sound (in most places).

My only major critique of this paper is that the analysis of the dh correction, as well as the scaling experiments in section 3.2, are only presented in terms of histograms over the entire ice sheet, or over aggregated sections of the ice sheet.

Much more insight would be provided to the reader if time series were presented from a selection of locations across the ice sheet. What do time series of dh, dhm, dhc, dhFAC and dhSMB look like for an individual location in, for example, the lower western accumulation area, the southern interior, the northern interior, the northeast and the southeast? Rather than having to guess the physical reasons for improved agreement (reduced residuals), it would become clear at a process level from the time series.

As this is my only major concern, I encourage the authors to expand the paper to accommodate for it. It would strengthen further the discussion about the scaling experiments, because the authors will have figures with time series of dhFAC and dhSMB that immediately make obvious why scaling of dhFAC works and that of dhSMB won't.

Line by line comments:

L 29: heights vary -> elevation varies

L 38: perhaps good to clarify that you are referring to a climatologically mean surface mass balance here

L 51: This is a confusing statement. FD models are forced by meteorological parameters as well as by mass fluxes, both of which are computed by an SMB model.

L 68: Here you focus mostly on the densification part of an FD. However, the thermodynamical part of FD models is usually also evaluated against observations of deep temperature.

L 71: In Munneke et al. (2015), laser-observed dh/dt was tested against an FD model at selected locations in order to evaluate their model.

L 93: I assume that the separation of 3.3 km refers to the ground projection of the lasers, not of the lasers themselves

L 105: please reformulate: a strategy does not measure anything.

L 105: “At each of a set of reference points…” this sentence does not flow well

L 129: Why is it safe to assume that the errors derived from release-003 products are not too optimistic?

Table 1: Listing the internal model variables feels redundant since they are never referred to in the manuscript. This table can either be moved to the supplementary materials or removed entirely.

L 220: In 2008, Helsen et al. showed that systematic surface elevation change can be the delayed result of multi-decadal or even centennial variability in SMB. In the present setup of your study, this effect is not accounted for. Rather, like in other studies, changes are defined with respect to a reference period (in your case, 1980-1995) over which no change is assumed. However, in the interior over which you evaluate the SMB/FD models, any residual between observations and models could be caused by these very long-term effects originating from quite deep in the firn.

L 305 (figure 2): see major issue above. The 32 tiled maps are a very comprehensive way of presenting the data, but it lacks in detail, making it hard to judge the models against observations at key locations. My suggestion would be to add a figure with time series of dh, dhm and dhc for a few selected locations (e.g. west coast, southern interior, northern interior, NE coast, SE coast). In that way, it becomes much easier to appreciate the temporal simulation of elevation change by the models compared to the observations.

L 330 (figure 4): perhaps clarify here that the scaling factors X were defined such that dh - X dhm == 0

L 345: what does the scaling imply? Is surface density not sufficiently captured? Is there a structural overestimation of snowfall and/or melt? Is there a structural error in the ICESat observations?

L 425 (figure 7): the effect of only scaling dhFAC (light green line) is invisible in the graph.

L 425 (figure): why does rescaling the dhSMB make almost no difference, as opposed to rescaling dhFAC? Please elaborate on this.

L 456: why does it help to isolate errors in high-elevation melt when the agreement at lower elevations is good? Can we simply assume that an SMB model will perform well at higher elevations (snow albedo dominated) when it does so at lower elevations (ice albedo dominated)?

L 471: The elevation change in GFSCv1.2 is much less sensitive to melt events than the other two models. At the same time, its surface density has increased to 327-387 kg/m3, which is higher than the mean 315 kg/m3 reported by Fausto et al. in 2018. The surface elevation change associated with a melt event is approximately the amount of melt per unit area divided by the density of the melted snow:  . Why do you think the higher surface density cannot explain the lower sensitivity of GFSCv1.2 to melt events?

L 514: GFSCv1.1 and GFSCv1.1 -> GFSCv1.1 and GFSCv1.2