|The authors are making progress on this study and have fixed some of the material that I objected to in the previous versions. I’m still not sure about what remains, and I think the authors need to spend some time thinking about the quality evidence that they have presented, and whether they, as referees of an article by a different set of authors, would be persuaded by the arguments and data they present.|
I remain skeptical of the authors’ differentiation between subglacial lakes and supraglacial lakes. The authors claim that the use of ATL03 and the Watta algorithm lets them detect elevation changes even when there is water in the lake, but they don’t make any distinction in the text or in the tables as to which elevation differences were calculated using the ATL03/Watta method. The one example they show of ATL11 and ATL03/Watta is in figure S1, which is not discussed in enough detail for me to be able to understand how they used the data, or what they thought was happening in that example. I had to cross-reference the coordinates on the figure with table S2 to figure out that this was HAYES_GLETSCHER_N_NN01. Then, looking at table S4, I could see that the authors listed elevation anomalies that almost certainly were measured on the floating ice on top of the lake alongside the elevation anomalies measured when the supraglacial lake may have been empty, and alongside the ATL03/Watta anomalies from August 2019.
The use of the ATL03/Watta algorithm is not possible for the ArcticDEM data. This means that the time series of elevations from mid-2018 and earlier are likely measuring changes in supraglacial water, or in lake ice atop supraglacial water. Unless the authors present good evidence to the contrary, this should be the assumption for what is going on in this earlier part of the record. As a result, examination of the ArcticDEM record does not confirm that the elevation anomalies are subglacial lakes- it just confirms that there is elevation variation in the past that continues into the ICESat-2 period.
In their rebuttal the authors show three examples where they claim that the irregular surfaces demonstrate that they are not measuring supraglacial lakes. I don’t understand why they make this claim—the first example (figure R1a) very clearly shows a supraglacial lake that has filled and drained seasonally. When water is present, the surface is not perfectly flat, but the irregularities could easily be caused by a rough lid of floating ice, or by spatially variable laser-light penetration into water and/or detector saturation effects from a bright water reflection. Figures R1b and R1c are much better examples of potential subglacial lake activity, but having seen the authors misinterpretation figure R1a, I am very worried about the quality of the interpretations of other data in the manuscript, and I would encourage the editor to request that the authors present the height profiles interpreted in the study one-by-one in the supplemental material to a revised manuscript. This would give the authors the opportunity to present the ATL11 tracks with the corresponding points sampled from ArcticDEM to provide a long-term record of change for each of the lakes. I suspect that in many cases, this would show that the ArcticDEM data sample flat surfaces (i.e. supraglacial water) during the high stands for many of the lakes.
The authors claim to have looked at Landsat imagery to confirm or deny the presence of liquid water in the lakes and, in their rebuttal, show an image for Academy05 that does not show much water on the surface. However, at the time of this image, Academy05 was at a low stand relative to the ICESat-2 measurements, so if it is a subglacial lake, we wouldn’t necessarily expect to see water at this time. Further, the landsat data can’t rule out floating ice for the lake.
The method for combining ArcticDEM and ICESat-2 time series seems to have the potential to generate nonsensical time series. The ICESat-2 profiles sample a small part of each lake basin, while the ArcticDEM data sample the whole basin, which means that a nonuniform pattern of filling and drainage (subglacial or supraglacial) will produce different values for the two datasets that are likely not comparable. I suggest sampling the ArcticDEM DEMs at the locations of the ATL11 measurements to construct a self-consistent time series, and investigating the extent to which these self-consistent time series agree with the full-basin records derived from ArcticDEM. If they don’t agree, the two should be presented separately, not combined as they currently are.
In my previous review, I pointed out that the spatial pattern of surface change was much more irregular than what we have seen in Antarctic subglacial lakes. In their rebuttal, the authors contend that under thin ice, the pattern of change associated with a subglacial lake can have sharp gradients, and cite two studies that looked at subglacial lakes that were under thin ice at the edges of glaciers. However, most of the lakes in this study are far from the edge of the ice sheet, and many are under ice of considerable thickness. The authors need to evaluate the ice thickness of their lakes and consider whether it makes sense that there would be large spatial variability for the locations they are considering.
In their response to my comment about uncertainties in the Watta algorithm, the authors again seem to interpret an elevation difference between a filled and a drained lake as evidence of subglacial lake activity (figure R3). Even using the Watta algorithm, most of the profile for 5 August 2019 is on floating lake ice, and the Watta algorithm only measures the bottom of the lake in a couple of small sections where the edge of the floating ice has melted. In these places, the bottom elevation is fairly close to the profiles from 2 Aug 2020 and 1 Nov 2020. The elevation change between 2 Aug 2020 and 1 Nov 2020 is fairly substantial (+5-10 m) but this is in an area where snowfall can be heavy, and it is not implausible that a local basin in the ice-sheet surface could trap a considerable amount of snow.
The authors quote the Fair et al study to say that the Watta algorithm can only measure ~7 m water depth, and quote Pope et al, 2016 to say that lakes are shallow (<10 m), but their own figure S1d shows a lake that is clearly more than 15 m deep (I’m assuming they too the refractive index of water into account in interpreting the apparent depth in S1c). These assumptions don’t seem to be valid and should not be relied upon.
One thing I don’t see in the manuscript is much critical assessment of the data. An example of this is the left panel of figure 1, where the authors plot a time series of elevations from Academy_01. This time series shows a gradual gain in elevation from 2012-2019, followed by a decline after 2019. The time series, however, includes several large upward and downward spikes in elevation, that are not explained in the text or in the caption. How do the authors interpret these spikes? I would suggest that they most likely represent errors in DEMs, but I don’t see that the authors recognize this, or that they acknowledge the possibility that other sharp features in time series for other lakes might be the result of DEM (or ICESat-2) errors. The authors need to acknowledge that the data that they are working from are fallible and need to explain how they differentiated between errors in the data and real signals.
Throughout the manuscript, the authors present elevation changes normalized to rates of change. This does not seem appropriate for changes that are episodic (i.e. seasonal drainage and filling of lakes), and especially in table S3 it makes the data difficult to compare against each other. Unless there is a good reason to the contrary, most of the changes should be presented as elevation differences, not elevation rates.
Figure 1 presents a really unusual lake as if it were a typical lake for the study. This lake is very large compared to the others in the study, and has a large, obvious, subglacially-driven change that is not typical of the other lakes, where the subglacial-vs-supraglacial difference is much less clear. It would be a much better use of space to present one or more ambiguous cases, and explain how each was interpreted, especially as the authors are claiming to make the very difficult (arguably impossible) distinction between change in supraglacial lakes and change in subglacial lakes paired with supraglacial lakes.
I don’t understand the time series presented in figure 2. There don’t seem to be enough points in the right-hand column relative to the number of ICESat-2 measurements in the left-hand column. Further, the profiles in 2a seem to show the lake filling, while the time series in 2b seems to show the lake draining during the ICESat-2 period.
The authors responded to some of my questions in their rebuttal without making corresponding changes in the manuscript (see the question of which lakes were sampled by RES).