“ Brief communication: Impact of common ice mask in surface mass balance estimates over the Antarctic ice sheet

Abstract. Regional climate models compute ice sheet surface mass balance (SMB) over a mask that defines the area covered by glacier ice, but ice masks have not been harmonised between models. Intercomparison studies of modelled SMB therefore use a common ice mask. The SMB in areas outside the common ice mask, which are typically coastal and high-precipitation regions, is discarded. Ice mask differences change integrated SMB by between 40.5 and 140.6 Gt yr−1 (1.8 % to 6.0 % of ensemble mean SMB), equivalent to the entire Antarctic mass imbalance. We conclude there is a pressing need for a common ice mask protocol.


1. Understanding why there are differences between the ice masks seems relevant to investigate in this study. The authors spend a few lines in the beginning of the Methods on the background for each model, but we need to more information on the native ice masks, what are they made from? Do they have a time stamp? How to they deal with ice shelves, grounded ice? etc.? etc.? Asking (and answering …) these questions and more provides a better understanding of "problems" with the native ice masks -being too old, old coarse, etc. Perhaps this can be used for more easily defining a concrete recommendation in addition to the 3 steps already defined? We

USGS) Earth Resources Observation and Science (EROS) Center and consist of Advanced Very High Resolution
Radiometer (AVHRR) data in 1 km resolution collected from 1992 to 1993 (EIDENSHINK and FAUNDEEN, 1994); MetUM at 0.44•resolution, has an ice mask created from the International Geosphere-Biosphere Programme (IGBP) data in combination with 1 km AVHRR from the period of 1992 to 1993 (Loveland et al., 2000); MARv3.10 at 0.32•resolution, with ice mask created from Bedmap2, which consist of a combination of different data sources such as satellite images, radar and laser altimetry gathered between 2000and 2010(Fretwell et al., 2013; and RACMO2.3p2 at 0.25• resolution, with ice mask made from a 1 km DEM that was created from the combination of ERS-1 data from 1994and ICESat data from 2003to 2008(Bamber et al., 2009." 2. There is a tendency when reading the manuscript that the "truth"/true AIS mask is the common mask. See for instance Table 1 and L 71 -"…smallest deltaSMB integrated over the 27 basins…". Why the comparison to the common ice mask? The answer is of course easy, it's what is common. https://doi.org/10.5194/tc-2021-317 However, throughout the study there is an implied desire for deltaArea and deltaSMB to be as small as possible, but that does not make much sense as the extent of the common is likely some distance from real world observations. However, I understand that without a new ice mask/product ready to be used, the authors need something to compare against -and in reality, it is highlighting the differences between models that needs to be conveyed. A way of addressing this, and conveying the need for a common usable product, would be to investigate/digitize the coastline of a few selected basins where differences are both small and large. As the comparison is based on regridded data to 0.11 degr., the resolution should just need to be better that this for now. These areas of interest could be displayed as a new figure 2 or be added as panels to figure 1. The main text should include a paragraph outlining the problem between real world observations, ice masks of each model, and comparison to a common ice mask (due to the lack of a new…). This could be used to better convey why a comparison to the common mask is used for now. Following this, the remained of the text (wording/phasing) should be reassessed in this perspective.
To clarify a small delta is not necessarily more true, We have this sentence above Furthermore, we have added a comparison between the mask from the DEM REMA and the common mask over the Peninsula in figure 1, and added this text above figure 1 (Line 86-88 in the revised version): "Moreover, it is also shown how well the common mask agreeswith the Reference Elevation Model of Antarctica (REMA, Howat et al. (2019)) mask over the AP, it is clear that the common mask is smaller than the REMA area, Fig. 1

panel B"
We have rewritten L71 to this (Line 94-95 in the revised version): "COSMO-CLM2 also has the smallest ∆SMB Gtyr -1 integrated over the 27 basins, which show that COSMO-CLM2 is least affected by the change in ice mask, Tab 1" Introduction there is a range in the SMB values, but also in the uncertainties (listed 80 -> 122 gt/yr) equal to 3.7 % -4.7%, which to me seems very, very low, in particular given the results of this study. This is a very interesting comment and we agree with the reviewer that the uncertainties as quoted seem low. Uncertainty quantification in the true sense of the meaning of uncertainty on a modelled quantity compared to a measured quantity is extremely difficult for SMB calculated from models. The convention in the field is to use the standard deviation on SMB equating in effect to interannual variability to estimate uncertainty in SMB estimates. As this subject is worth a paper in its own right we use the standard deviation as calculated in Mottram et al and added these as uncertainties on the SMB in 13. L57-58 incl rightmost column in table 1: I understand from a sea level contribution perspective it is interesting to include the numbers/highlight in impact, but in the introduction, you clearly stated that your ice mask comparison includes shelves etc. I agree that is should be included, it warrants a follow-up, but there is a need to mention this in the Methodswhat is defined as grounded ice and are there differences in grounded ice extent in the native ice masks? Clearly defining the ground ice also becomes an issue in the discussion L80-87(see also below) 14. L83-85: To some extent it makes sense to compare against IMBIE2 but is it a full-on apples-to-apples comparison? Is there "another" common extent for the ground ice only? In which case the comparison makes sense (grounded v grounded), but from reading the manuscript I am not sure where the numbers for the comparison originate from. Please clarify in the methods, results, and here in the discussion We compare grounded ice sheet SMB with the total AIS mass imbalance as calculated by IMBIE2 (109±56 Gt yr -1) , which is also only applied over the grounded part of the ice sheet so the comparison is valid. We have rewritten this in the discussion section to make it clearer (Line 111-115 in the revised version  -2021-317 in 63 combinations of model coverage, in addition to the common…" 18. L94-98: this brings back the point about the origin of each native ice mask. Following this discussion/mentioning of post-creation modification can be discussed/added. We have rewritten it to (Line 125-132 in the revised version): "These different area coverage combinations around the ice sheet are partly driven by differences in ice masks and partly by differences in resolution. However, we cannot identify any systematic or model-specific biases on a regional scale. The native ice masks vary around the coastline arbitrarily, which is partly due to the time when the ice mask was created and what data are used to create the native ice mask. For example, the HIRHAM5 and MetUM ice masks are created from data collected three decades ago, so there have been multiple calving events since the data collection. The native ice mask from COSMO-CLM2 and MARv3.10 are created from data collected lore recently and in higher resolution. The higher resolution is also a benefit over AP and in coastal areas where there is complex orography, where a higher resolution also can change the orographic precipitation." 19. L105-124: I really like this part. As part of Step 2, I would suggest/recommend proper surface delineation with a specific year associated, e.g. Antarctic summer 2018/19, or what compares well with the remainder datasets, e.g REMA, etc., to ensure a common data platform that can "easily" be updated every X years. I realize that some datasets such as RGI are getting "old", but I will encourage pursuing more recent data/time stamps for the grids, DEMs, etc., and hope that/encourage other data produces will update their data too.
That is a good idea to add a time step for when the data is from, within the files, we have added this at the end of this paragraph: "Finally, we suggest that it is noted in the mask file, which data sets are used to create it, and when the data sets are last updated" 20. L117: incl. "the" before tool Done 21. Both the grounding line and ice shelf extent change through time. Should the ice mask not dynamically evolve both over the reanalysis era and in the future? The authors only discuss creating one, static, ice mask. This in itself will create uncertainty as it will only be "correct" at one point in time. And what point in time should that be?
That are some good points, below I have answered them point by point: a. What recommendations can be made in terms of using a common ice mask?
That the common mask is easily accessible for all and with clear guidelines on how to use it.
b. Should the ice mask not dynamically evolve both over the reanalysis era and in the future?
Ideally, we would of course evolve ice sheet masks through time, but this requires a dynamical ice sheet model to be fully two-way coupled within a regional climate model. The development of regional coupled climate and ice sheet models is still in its infancy and not yet available within the constraints of computing resources. Over the recent past we do not have sufficient observational data to reconstruct multiple ice sheet extents for most of Antarctica.
Even the highest resolution models are still relatively coarse compared to observed changes in ice sheet extent, e.g. the calving of large icebergs from ice shelves. Updates to the ice mask are therefore usually only necessary for long timescale simulations. We have added the following sentence, in the end of the discussion: "Ideally the ice mask would evolve over time as the continent changes, however, we feel it is important to first standardize the approach to creating and using ice masks. Thus, we suggest that it is noted in the mask file, which data sets are used to create it, and when the data sets are last updated." c. This in itself will create uncertainty as it will only be "correct" at one point in time. And what point in time should that be?
There is no single point in time when we have a fully complete and accurate ice mask for Antarctica, all datasets are made up of a mosaic of different types of data assembled into a single land surface cover dataset. More recent datasets are generally higher resolution with https://doi.org/10.5194/tc-2021-317 better quality data, which is why we suggest using the REMA data to construct a common mask.
27. L16: Remove comma after "Although" Done 28. L26: Can you provide an approximate grid cell length in km as well? Yes, it is approximately 12.5 km, this is added in the sentence "...onto the same grid at 0.11•(≈12.5 km) resolution" 29. L35-: Can you please provide more information for these various datasets?
Over what time period(s) were each developed? How were the masks developed? What does IGBP mean? On line 95, you go on to speculate why there differences may exist in the ice masks, but this is only very briefly discussed, and in my opinion, insufficiently discussed.
This is added at the beginning of the Method section, see the answer to comment #1 Regarding the line 95, we have added more to this, seethe answer to comment #18 30. L45: change first instance of defined -> define Corrected 31. L58: adn -> and Corrected