the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 04 Oct 2022
Sentinel-1 detection of seasonal and perennial firn aquifers in the Antarctic Peninsula
Abstract. In recent years, the existence of firn aquifers in the Antarctic Peninsula (AP) has been confirmed by in-situ observations. Given their importance for understanding the hydrology of the Antarctic ice sheet, a more spatially comprehensive assessment of AP firn aquifers is desirable. The purpose of this study is to map firn aquifers in the AP from space using C-band Synthetic Aperture Radar imagery from ESA’s Sentinel-1 mission. These observations enable the detection of firn aquifers at 1 × 1 km2 resolution. The method presented here is based on quantifying the characteristic, gradual backscatter increase during the (partial) refreezing of the liquid water in the firn layer after the peak melt season. When applied to the available time series, it detects perennial aquifers (existing year-round) for the period 2017 to 2020, as well as seasonal aquifers which do not persist through winter. We acknowledge that the backscatter signature in any given year is indistinguishable for seasonal and perennial aquifers. We detect seasonal firn aquifers in the north and northwest of the AP, as well as on the Wilkins and George VI ice shelves. Only in the north and northwest of the AP, aquifers are detected each year in the observation period, here taken as a proxy for perennial firn aquifers. Both distributions agree with model simulations. Further in situ and modelling studies and longer time series of satellite observations are needed to validate the results of this study.
- Preprint
(3347 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on tc-2022-127', Isis Brangers, 24 Oct 2022
The paper uses Sentinel-1 (S1) data to map perennial firn aquifers in the Antarctic Peninsula. I think it is the first study that uses Sentinel-1 for aquifers outside of Greenland. Other novelties of this work are the comparison of S1 backscatter with a firn model and SMRT radiative transfer modelling. Overall it is nice work. The paper is well written and the limitations are clearly stated. The proposed methodology is only slightly different from previous methods and has similar limitations.
Some comments:
What are the advantages of the presented method compared to the work of Brangers et al and Miller et al.? The latter could be more robust.
How was the threshold of 105 days chosen?
The dry snow zone was justly masked, however, in the bare ice zone there also can be no firn aquifers since there is no firn. It might improve the performance if this zone was also excluded. I believe slowly draining/freezing supraglacial lakes on bare ice could give similar backscatter responses as aquifers.
Part of the inter annual variability could potentially be explained by differences in weather. E.g. at an aquifer location the backscatter might increase faster in case of a colder fall than a relatively warm fall (especially at sites with deep water table). I agree part of the differences are seasonal aquifers, but this could also be part of the explanation.
The validation of the methodology is a bit limited. There is only a visual comparison with the model and a comparison of the total aquifer area. The authors could consider adding a more qualitative analysis.
Citation: https://doi.org/10.5194/tc-2022-127-RC1 -
RC2: 'Comment on tc-2022-127', Anonymous Referee #2, 07 Nov 2022
This manuscript details a new method for detecting firn aquifers over the Antarctic Peninsula (AP) from Sentinel-1 synthetic aperture radar data. The detection algorithm is based on the idea that in regions with firn aquifers, summer meltwater will refreeze more slowly within the firn due to high melt and warm subsurface temperatures. This refreezing process is reflected as a rebound of the SAR surface backscatter value from low values over wet firn in the melt season, to higher values over dry firn in the refreezing season. The authors set a detection threshold that firn aquifers exist where backscatter reaches 80% of its average September value on or after the 105th day of the year. Using this threshold, they map aquifers in the 2017-2020 seasons. They compare these mappings to predictions from the IMAU-FDM firn model and show that they are in broad agreement. Based on the large interannual variations in Sentinel-1 aquifer extent, they also conclude that seasonal firn aquifers may be common on the AP.
The motivation for satellite mapping of firn aquifers in Antarctic is clear and this paper advances that effort. However, the detection utility of the algorithm is limited by a lack of validation. This is not necessarily the authors’ fault – a robust validation data set for the AP that overlaps the Sentinel-1 era does not exist – but combined with the lack of uncertainty quantification in the current methods, the conclusions that can be drawn from the final mappings are, in my opinion, very limited.
Major Comments:
[1] It is not clear why the authors choose to develop a completely new Sentinel-1 firn aquifer detection algorithm for Antarctica when a reasonably robust algorithm was already developed in Brangers, et al (2020) for Greenland and validated against the OIB firn aquifer detections. What is the justification for not applying the Brangers, et al (2020) methodology to Antarctica? Of course, there will be some environmental differences between the two locations that can impact empirical thresholds, but given the lack of Antarctic validation data, the authors already have to choose essentially arbitrary thresholds in their method, and that does not seem to be a clearly better choice than adopting the thresholds tuned for Greenland. Presumably the accumulation and melt percolation processing the control refreezing rate will not be so different on ice shelves as to make the Brangers, et al (2020) results totally useless.
At a minimum, the paper would be significantly strengthened by a clear discussion of why a new detection algorithm is needed and how this new method improves up on the previous results of Brangers, et al (2020). Ideally, a quantitative comparison between aquifer extent calculated from the Brangers method and this newly proposed method would be presented so that the reader can assess to what extent the results are consistent.
[2] How does the algorithm deal with the possibility that backscatter values do not rebound to a stable September value that is consistent from year to year? (The Muller signal seems to be an example of this.) Assuming this indicates large quantities of near-surface liquid water that did not refreeze over the winter, I would expect that you would want to either include these areas as aquifers detections (maybe as a second filter in the algorithm), or completely discard them from the data set if they seem to be indicative of something like refreezing ponds.
[3] I do not think this paper can confidently make the conclusion that firn aquifers on the AP are largely seasonal from these data. Given that Sentinel-1 is only sensitive to the upper 7m, the data only suggest that there is a large seasonal variation in how quickly the upper 7m of the snow/firn column refreeze. I am sure that there is some cutoff where the refreezing rate is so rapid that it is no longer consistent with the presence of a temperate firn layer at depth, but without understanding physically where that threshold lies, one cannot rule out that the inter-annual variation in aquifer extent may be the result of noisy data, uncertainty in the validity of the DOY80 threshold, or controlled by variations in total accumulation and melt from year to year.
I would encourage the authors to at least try to quantify how much of the inter-annual variation in aquifer extent is robust to small changes in the DOY80 threshold (greater or less than 105) or to uncertainty in where the backscatter time series crosses the 80% threshold due to radiometric uncertainty or high frequency oscillations in the backscatter time series. (I recognize that the time series has been smoothed, but the smoothing window is pretty arbitrary, so there is some implicit uncertainty associated with the choice of that window.)
Line Comments:
Line 24 – perhaps specify “refreezing in the firn” here
Line 24 – consider more than one citation here, perhaps 1-2 supporting papers for each of the mechanisms discusssed (refreezing, runoff, supraglacial storage, etc…) to better support a general point about all ice shelves, rather than just the Roi Baudouin Ice Shelf discussed in the Lenaerts (2017) paper.
Line 26 – citation? I am not sure that we have a well-developed idea of what a “seasonal firn aquifer” really is, since it is not clear that such a thing has been observed in the field. Throughout the paper, the authors seem to use this as a catch-all term for a damp firn layer that refreezes over the winter, which I am not sure necessarily qualifies as an aquifer. It might be better just to refer to these seasonal signals as “transient liquid water storage” or something similar.
Line 57 – how is the data reprojected onto the EASE grid? What type of interpolation or binning is used to assign values to a grid cell? How are the native resolution data aggregated within a grid cell – taking the mean, for example? If data are interpolated or average, is this done in linear of dB space? Can the statistics of the native pixels assigned to each grid cell provide an estimate of the radiometric uncertainty within each EASE grid cell?
Line 59 – roughly how much do geometric parameters like look angle vary from RON to RON? At high look angles, is there any concern that the backscatter will hit the noise floor and bias the average values used for geometric bias correction?
Line 61 – does “the bias is determined pixel-wise” mean that a unique bias is calculated for each pixel? If so, how is the average calculated for the RON? Or does this mean that for any given image, the average is found, normalized by the average across all images used in the study, and then all pixels in the image are offset by that difference?
Line 67 – what set-up is used for the SMRT runs? IBA vs DMRT vs Mie/Rayleigh scattering? Autocorrelation function and correlation lengths?
Line 70 – how are these parameters chosen? How much does the reliable penetration depth vary with the density or grain size of the overlying snow or the liquid water content of the wet layer? This setup makes sense for understanding the impact of accumulation events over wet firn on detection, but may not be particularly representative of system sensitivity to wet firn layers deeper within an established firn pack.
Line 92 – how sensitive are the results to the smoothing threshold?
Line 130 – what electromagnetic model is used (IBA, etc) and how are the microstructural parameters chosen?
Line 133 – section 3.1 might be better placed with section 3.4, since the main purpose of 3.1 seems to be to convince the reader that changes in IMAU-FDM LWC should be comparable to the Sentinel-1 signal and to highlight some of the uncertainties in the results.
Line 149 – have you considered masking out regions of bare ice or ice slabs as simulated by IMAU-FDM when looking for firn aquifers? This might help avoid the potential for mixed signals from buried lakes.
Line 152/Figure 5 – is there a reason to show the spatial distribution of DOY80 since a strict cutoff threshold is used for aquifer detection? I actually do think it’s useful since the cutoff is uncertain, but it is worth discussing spatial patterns in that uncertainty in the text if you choose to show it.
Line 173 – I am not sure that melt underestimation in the model due to surface ponds should apply. Presumably firn aquifer areas should be largely mutually exclusive with areas of surface ponding?
Line 211 – where did 15m come from when your SMRT sensitivity study showed an effective penetration depth of only 7m?
Line 212 – why not try to incorporate this into the algorithm?
Line 200 - section 4.1 – this is a good and very important section that places the results of this work into the appropriate context of the uncertainty
Line 249 – again, I feel strongly that the uncertainty analysis and the system sensitivity are not sufficiently robust to make any real statements about seasonal variability.
Line 260 – any of the airborne OIB data available will be from November, so this requirement has been met by most if not all previous ice-penetrating radar data collection
Line 266 – much too strong of a statement given your level of evidence! Should you expect to see large floating ice shelves on the western side of the peninsula given the local ice dynamics and calving rates? Is there evidence of past ice shelf collapse along the western side of the peninsula in regions where you detect aquifers? You can speculate on these questions, but a statement that perennial firn aquifer development at the grounding line is a precursor for ice shelf collapse is not supported by the data in this manuscript.
Citation: https://doi.org/10.5194/tc-2022-127-RC2 -
RC3: 'Comment on tc-2022-127', Anonymous Referee #3, 20 Nov 2022
The manuscript decribes the development of new algorithm that uses an arbitrary set of thresholds and fixed dates to map ‘seasonal and perennial firn aquifers’ on the Antarctic Peninsula, and then comapres these results to IMAU-FDM. The manuscript further details the simulation of backscatter time series using IMAU-FDM and the SMRT radiative transfer model to ‘validate’ the comparison. ‘Seasonal firn aquifers’ are mapped if meltwater is detected in one year. ‘Perennial firn aquifers’ are detected ‘by proxy’ if meltwater is detected for one or more years. The authors acknowledge that they cannot tell the difference between the signals. Although the comparison between the S1 detection algorithm and IMAU-FDM appear to be in broad agreement when combined over the three-year time series, no year-by-year comparison is provided. Recent field measurements and OIB observations of an expansive perennial firn aquifer on the Wilkins Ice Shelf – which neither the S1 detection algorithm or IMAU-FDM are in agreement with – are not discussed. Alternatively, the authors suggest that coincident field observations are required for validation, which implies that perennial firn aquifer form and refreeze on a regular basis. The manuscript concludes with statements on perennial firn aquifers and ice shelf collapse that are not supported by the study.
Major Comments –
There are significant technical and theoretical issues throughout this manuscript detailed in the minor comments. Assumptions and conclusions are made that are simply not supported by the analysis. In particular, it is unclear how and why the thresholds were chosen other than to simply reproduce the IMAU-FDM simulations. Given that the study uses a radiative transfer model, it would be reasonable to expect that the threshold for subsurface meltwater would be chosen based on a series of simulations. However, that is not the case. The assumption that these thresholds remain stable temporally and spatially is not plausible. There is nearly a 10 C difference in the mean annual temperature between the northernmost islands and the southernmost George VI Ice Shelf. Many of the northern locations have a mean annual temperature near 0 C. This results in a significant difference in the melting seasons between locations. This is reflected in the final mappings which show the north and northwestern Antarctic Peninsula as primarily ‘perennial firn aquifers’, when what the algorithm is actually detecting is surface meltwater during long-duration melting seasons. This is further complicated by the shallow penetration depth of S1, which cannot detect meltwater beyond a few meters’ depth in the percolation zone. The lack of a detection can simply imply that meltwater is present, and has simply descended below the penetration depth prior to the fixed April 14th date. Overall, my conclusion is that the S1 detection algorithm is mapping both surface and subsurface meltwater within a fixed time interval and limited penetration depth. Unfortunately, I am unable to recommend publication of this manuscript.
Minor Comments:
Line 36-38 – Both the Wilkins (Montgomery et al., 2020) and Muller ice shelves (MacDonell, 2021) have perennial firn aquifers confirmed via fieldwork. The perennial firn aquifer on the Wilkins Ice Shelf is confirmed throughout the winter of 2017-2018, again throughout the winter of 2018-2019, and 2019-2020. Firn cores confirmed ~20 meters of meltwater at depths of ~15 m below the surface. The perennial firn aquifer was also confirmed within a ~15 km radius by GPR surveys in December 2018.Line 56-57 – Why is horizontal transmit/receive polarization used? Were both channels analyzed?
Line 58 – The gridding is 1 km x 1km. The spatial resolution of the measurements is 20 m x 40 m.
Line 58 – What is the temporal resolution of the combined S1-A and S1-B? 12 days?
Line 60 – What is the incidence angle range of the measurements before averaging? How does averaging measurements with different orbital geometries (i.e., incidence and azimuthal angles) over the complex topography of the Antarctic Peninsula bias the measurements? Can the authors please cite or justify the averaging method?
Line 62-63 – What is the maximum and minimum number of data points ‘eliminated’ from each time series used in the algorithm? How does that influence the targeted signal spatially and temporally in terms of the algorithm performance, particularly in 2017-2018?
The authors use three (not four) years of data 2017-2018, 2018-2019, and 2019-2020. However, none of the provided plots or maps show 2017-2018, which is odd for such a short time series. Can the authors please provide a reason for this omission? If the data isn’t suitable to show, is it suitable to use in the S1 detection algorithm? Are the results viable?
Please cite the ice masked used in the detection algorithm.
Line 120-121 ‘IMAU-FDM does not simulate standing water or lateral water flow, so only a qualitative comparison can be made with S1 detected aquifers on the basis of the presence of irreducible liquid water content’.
If these processes were included in IMAU-FDM, can the authors please describe how a quantitative comparison could be made? The S1 detection algorithm is binary.
Line 124-128 – The S1 detections are 20 m x 40. The IMAU-FDM is 5.5 km x 5.5 km. Can the authors please describe how the time series and map comparisons were made?
Line 133-135 – An observational technique (S1 detection algorithm) can’t be ‘validated’ using a model simulation (SMRT) parametrized with a model simulation (IMAU-FDM). The simulations are simply not real.
Line 188-190 – ‘The South Shetland Islands show almost complete firn aquifer cover, confirmed by local observations (Jiahong et al., 1998; Travassos and Simoes, 2004; Macheret et al., 2009).’
These studies do not describe a ‘complete firn aquifer cover’. These results describe a ~1 m water table above the firn ice transition at ~ 30 - 50 m depth in a temperate glacier at field sites. These studies were conducted in 1985-1992, 1997-1998, and 2000-2006 – as much as 30 years prior to this study, and not coincident in time with S1. No comparisons between the S1 detections and these fields sites are provided.
Noting that the penetration depth of S1 is several meters – can the authors please justify using these field sites to ‘validate’ the detection algorithm?
Can the authors also please justify using these field sites to ‘validate’ the detection algorithm, when the field sites on the Wilkins Ice Shelf and the Muller Ice Shelf - which are coincident in time and do not appear to be consistent with the results of the S1 detection algorithm - are simply disregarded as ‘complex’, citing ‘dissimilar quantities in different time scopes’?
Line 202-203 – ‘The increase of backscatter over time after the peak melt season can be caused by the wet layer getting buried under fresh snow accumulation’
Backscatter from snow accumulation at C-band wavelengths is negligible.
Line 235-236 This issue will become less relevant in the coming years as the amount of data is increasing, although a complication is that Sentinel-1B has failed since December 2021 with no guarantee of operating again’.
Sentinel 1-B has failed. Only Sentinel 1-A is operational. The authors stated in the methods section that ‘sufficiently dense data coverage for our purpose was only reached after the launch of Sentinel-1B in April 2016’. If Sentinel-1B failed, and both satellites are required for ‘sufficiently dense data coverage’, can the authors please describe how is the amount of data increasing?
Line 253-254 - ‘Comparing the satellite detection results with in-situ measurements on the Wilkins and Müller ice shelves (Montgomery et al., 2020; MacDonell et al., 2021) is complex as dissimilar quantities in different time scopes are measured.’
This is an overly complicated sentence, when the comparison is simple. The authors are using S1 data with a resolution of 20 m x 40 m - comparable to a firn core. The authors claim to be ‘confidently’ mapping ‘seasonal and perennial firn aquifers’ over the Antarctic Peninsula. There is a field confirmed perennial firn aquifer on the Wilkins Ice Shelf in 2017-2018, in 2018-2019, and 2019-2020. Not simulated. Confirmed. At some point near the end of the austral meting season in 2017, 2018, and 2019, the seasonally recharged upper layers of the perennial firn aquifer were within the penetration depth of S1 and detectable. The authors appear to detect a ‘seasonal firn aquifer’ on the Wilkins Ice Shelf in 2018, however - as previously noted – the data is oddly not shown. The algorithm fails to detect even a ‘seasonal firn aquifer’ in 2019 or 2020.
Can the authors please justify the lack of a S1 detection?
Line 254-258 ‘On Wilkins Ice Shelf, the reported aquifer water tables were typically located at 6 to 22 meter the surface in 2014, and IMAU-FDM does not provide indications that the very deep aquifers have been recharged in recent years…'
The OIB-derived detections reported in Montgomery et al., (2020) extend over nearly the entire Wilkins Ice Shelf. Again, not simulated. These are observations. Furthermore, to attenuate the low-frequency MCoRDS signal, the detection implies that a relatively thick perennial firn aquifer with significant volumes of meltwater is present at depth. The assumption that the authors seem to make throughout the manuscript - that perennial firn aquifers form and refreeze regularly on the Wilkins Ice Shelf (and everywhere else) - is simply not plausible. Can the authors please justify this assumption?
IMAU-FDM does not appear to simulate the extent of the perennial firn aquifer confirmed by observations over the Wilkins Ice Sheet in 2014 (van Wessem et al., 2020). Nor does it simulate the perennial firn aquifer 2017, in 2018, and 2019. Can the authors explain the relevance of the fact that IMAU-FDM does ‘not provide indications that the very deep aquifers have been recharged in recent years’ to the S1 detection algorithm results? If IMAU-FDM is not capable of simulating the field confirmed firn aquifer on the Wilkins Ice Shelf, can the authors justify the comparison between the S1 detection algorithm and IMAU-FDM over the entire Antarctic Peninsula?
Line 259-261 – ‘further analysis of OIB flight measurements would be useful for comparison as they provide information on the firn aquifer water table on a larger scale. Especially collecting measurements before the peak melt season would be valuable as they allow detection of perennial firn aquifers which are more robustly defined than seasonal aquifers’
There are almost 20 years of OIB data, including OIB-derived perennial firn aquifer detections (Montgomery et al., 2020) publicly available and not used in this study.
Line 264 – 268 - ‘Finally, it is noteworthy that the existence of perennial firn aquifers at the grounding line and the presence of extensive floating ice shelves appear to be mutually exclusive on both sides of the AP (Fig. 7). This suggests that perennial firn aquifers play an important role in ice shelf viability and demise: if a perennial firn aquifer develops around the grounding line or on the ice shelf, it is a precursor for collapse.
Big conclusion that is not supported by the results of this study.
Line 264 – 268 – ‘An exception is Wilkins ice shelf, but it is well known that this ice shelf has partly disintegrated in recent years (Braun et al., 2009).’
How exactly did the formation of perennial firn aquifers at the grounding line of the Wilkins Ice Shelf contribute to the collapse and disintegration of the ice front?
Figure 2.
SMRT is a snow model. It is not appropriate for modeling deeper subsurface meltwater within the complex stratigraphy of the percolation facies. During refreezing, the dominant backscattering response is via volume scattering from ice pipes and lenses – which significantly reduces penetration depth. This is well-documented in the literature. Additionally, in perennial firn aquifer areas, SMRT does not include latent heat release from refreezing meltwater.The SMRT simulated ‘dry’ and ‘wet’ layers are inconsistent with the actual backscatter data shown in Figure 4.
The ‘dry layer’ is simulated with backscatter values of between ~ -25 dB to -15 dB.
These values are consistent with volume scattering from dry snow in the interior of the ice sheet – not the percolation zone where volume scattering from ice pipes and lenses dominates the response. A realistic ‘dry layer’ in the percolation zone that included ice pipes and lenses would plausibly have backscatter values between ~ -5 dB and 0 dB, which is consistent with the plots shown in Figure 4, except for the Muller.
The ‘wet layer’ is simulated with backscatter values between ~ -15 dB and almost -40 dB.
However, in the percolation zone, it is uncommon for the backscatter value to drop below ~ -25 dB at C-band. This value is consistent with significant surface melting, not subsurface meltwater. So, in terms of observations, the minimum value (not simulated or the ‘radiometric accuracy’) should reasonably be set at ~ -25 dB. This is also consistent with peak melt on any of the plots shown in Figure 4.
Setting the lower threshold at -25 dB would decrease the ‘detection limit’ to ~1 m.
Can the authors justify a reasonable detection of ‘seasonal and perennial firn aquifers’ at these shallow depths?
Alternatively, can the authors justify their modeling approach? Specifically, why are uniform density, grain size, and LWC values used to depths of 7 m, rather than depth-dependent parameters? Is the 5% LWC value consistent with field measurements? How does the detection limit change with increasing and decreasing LWC values?
Figure 3
The authors state that this plot represents an ‘idea case’ – why was this plot chosen? What are the dates? What are the lat/lon coordinates? Is it a ‘seasonal’ or perennial’ firn aquifer? If you can’t tell the difference – why is it ideal?
Can the authors explain the significant differences between the ‘ideal case’ and the Muller Ice Shelf site shown in Figure 4, which obviously shows extended, possibly year-round surface melting? How does that influence the detection algorithm?
Can the authors show that the ‘ideal case’ characterizes the majority of the ‘perennial firn aquifers’ in the northern Antarctic Peninsula + islands? Or do those sites look more like the Muller Ice Shelf site?
Figure 4
The authors need to provide lat/lons for each of these sites. The Wilkins Ice Shelf and the Muller Ice Shelf sites appear to be field locations from other studies (Montgomery et al., 2020; MacDonell et al., 2021) The authors need to acknowledge this work, and describe what was found at these sites.
Note: There are issues with using a 15 -day smoothing filter on the S1 backscatter and distinguishing surface melting from subsurface meltwater, particularly at the end of the melting season. The smoothed signal with ‘mimic’ the subsurface meltwater signal if there is late season surface melting. Similarly, the IMAU-FDM is simulated on 10-day intervals, which makes time series comparisons difficult to interpret.
Note: The authors state that the peak melting season is between 1 and 15 January – yet many of the time series, suggest this data is much later, with significant variability between years. As an example, in 2020, the peak melting season appear to be in March, with similar shifts on the George VI in both 2019, and 2020.
The authors state that ‘Line 143-145 - there is a clear correspondence between the presence of liquid water in IMAU-FDM and the corresponding forward modelled SMRT S1 backscatter time series on the one hand, and the observed S1 backscatter time series on the other hand’.
Given that the IMAU-FDM is the primary input into the SMRT – it is expected that there would be a ‘clear correspondence’ between the two.
However, there is no obvious correspondence between the SMRT and S1 backscatter on the Larsen C, George VI, or Wilkins. The SMRT does not simulate the identified subsurface meltwater signal at the end of the melting season – which suggests that the dominant backscattering responses are not simulated in SMRT, or that the IMAU-FDM is not adequate simulating the actual firn conditions at the surface. In 2019 – there are several months where the there is clearly significant subsurface meltwater detected by SI.
The Wilkins Ice Shelf, in particular, is a field validated perennial firn aquifer site that the IMAU-FDM is not simulating and the SI algorithm is not detecting in either year. This is a strong indication that the IMAU-FDM will not simulate accurate results. Furthermore, the 2019 Wilkins Ice Sheff site and 2020 Larsen C Ice Shelf site (Line 135-136 - ‘a location with rapid meltwater refreezing and therefore small amounts of retained liquid water ‘) appear to be near identical in terms of the S1 detection – shifting a threshold would identify aquifers on both sites in different years. This is a strong indication that the S1 detection algorithm is not viable.
On the Muller Ice Shelf in 2019 – the backscatter values appear to be near -15 dB at the start of the time series, and experience a limited drop in backscatter values during the melting season. This suggests that the firn is initially wet at or very near the surface and gets wetter throughout the melting season. It would be difficult to justify these backscatter values from subsurface meltwater underlying ice pipes and lenses. Once surface meltwater began the refreezing process, strong scattering would increase backscatter values very quickly. The authors potentially attribute this to supraglacial lakes (Line 149-150 - The S1 backscatter signal could represent persistent or slowly buried supraglacial lakes’), however, this appears to be the field confirmed perennial firn aquifer site described in MacDonell et al., 2021, where supraglacial lakes are not observed. In 2020, there is obviously surface melting throughout the entire year. Even smoothed, the backscatter time series show significant variability. Backscatter values appear to be below -15 dB prior to the start of the 2020-2021 melting season. The detection thresholds are chosen during active surface melting at this site. The algorithm is detecting active surface melting, not subsurface meltwater. Even know this is a field confirmed perennial firn aquifer site, subsurface meltwater is masked. It is simply not possible to detect ‘seasonal and perennial firn aquifers’ during active surface melting. This is a second strong indication that the S1 detection algorithm is not viable.
Unfortunately, the backscatter time series on the Muller Ice Shelf are characteristic of the entire north and northwestern Antarctic Peninsula.
Citation: https://doi.org/10.5194/tc-2022-127-RC3
Status: closed
-
RC1: 'Comment on tc-2022-127', Isis Brangers, 24 Oct 2022
The paper uses Sentinel-1 (S1) data to map perennial firn aquifers in the Antarctic Peninsula. I think it is the first study that uses Sentinel-1 for aquifers outside of Greenland. Other novelties of this work are the comparison of S1 backscatter with a firn model and SMRT radiative transfer modelling. Overall it is nice work. The paper is well written and the limitations are clearly stated. The proposed methodology is only slightly different from previous methods and has similar limitations.
Some comments:
What are the advantages of the presented method compared to the work of Brangers et al and Miller et al.? The latter could be more robust.
How was the threshold of 105 days chosen?
The dry snow zone was justly masked, however, in the bare ice zone there also can be no firn aquifers since there is no firn. It might improve the performance if this zone was also excluded. I believe slowly draining/freezing supraglacial lakes on bare ice could give similar backscatter responses as aquifers.
Part of the inter annual variability could potentially be explained by differences in weather. E.g. at an aquifer location the backscatter might increase faster in case of a colder fall than a relatively warm fall (especially at sites with deep water table). I agree part of the differences are seasonal aquifers, but this could also be part of the explanation.
The validation of the methodology is a bit limited. There is only a visual comparison with the model and a comparison of the total aquifer area. The authors could consider adding a more qualitative analysis.
Citation: https://doi.org/10.5194/tc-2022-127-RC1 -
RC2: 'Comment on tc-2022-127', Anonymous Referee #2, 07 Nov 2022
This manuscript details a new method for detecting firn aquifers over the Antarctic Peninsula (AP) from Sentinel-1 synthetic aperture radar data. The detection algorithm is based on the idea that in regions with firn aquifers, summer meltwater will refreeze more slowly within the firn due to high melt and warm subsurface temperatures. This refreezing process is reflected as a rebound of the SAR surface backscatter value from low values over wet firn in the melt season, to higher values over dry firn in the refreezing season. The authors set a detection threshold that firn aquifers exist where backscatter reaches 80% of its average September value on or after the 105th day of the year. Using this threshold, they map aquifers in the 2017-2020 seasons. They compare these mappings to predictions from the IMAU-FDM firn model and show that they are in broad agreement. Based on the large interannual variations in Sentinel-1 aquifer extent, they also conclude that seasonal firn aquifers may be common on the AP.
The motivation for satellite mapping of firn aquifers in Antarctic is clear and this paper advances that effort. However, the detection utility of the algorithm is limited by a lack of validation. This is not necessarily the authors’ fault – a robust validation data set for the AP that overlaps the Sentinel-1 era does not exist – but combined with the lack of uncertainty quantification in the current methods, the conclusions that can be drawn from the final mappings are, in my opinion, very limited.
Major Comments:
[1] It is not clear why the authors choose to develop a completely new Sentinel-1 firn aquifer detection algorithm for Antarctica when a reasonably robust algorithm was already developed in Brangers, et al (2020) for Greenland and validated against the OIB firn aquifer detections. What is the justification for not applying the Brangers, et al (2020) methodology to Antarctica? Of course, there will be some environmental differences between the two locations that can impact empirical thresholds, but given the lack of Antarctic validation data, the authors already have to choose essentially arbitrary thresholds in their method, and that does not seem to be a clearly better choice than adopting the thresholds tuned for Greenland. Presumably the accumulation and melt percolation processing the control refreezing rate will not be so different on ice shelves as to make the Brangers, et al (2020) results totally useless.
At a minimum, the paper would be significantly strengthened by a clear discussion of why a new detection algorithm is needed and how this new method improves up on the previous results of Brangers, et al (2020). Ideally, a quantitative comparison between aquifer extent calculated from the Brangers method and this newly proposed method would be presented so that the reader can assess to what extent the results are consistent.
[2] How does the algorithm deal with the possibility that backscatter values do not rebound to a stable September value that is consistent from year to year? (The Muller signal seems to be an example of this.) Assuming this indicates large quantities of near-surface liquid water that did not refreeze over the winter, I would expect that you would want to either include these areas as aquifers detections (maybe as a second filter in the algorithm), or completely discard them from the data set if they seem to be indicative of something like refreezing ponds.
[3] I do not think this paper can confidently make the conclusion that firn aquifers on the AP are largely seasonal from these data. Given that Sentinel-1 is only sensitive to the upper 7m, the data only suggest that there is a large seasonal variation in how quickly the upper 7m of the snow/firn column refreeze. I am sure that there is some cutoff where the refreezing rate is so rapid that it is no longer consistent with the presence of a temperate firn layer at depth, but without understanding physically where that threshold lies, one cannot rule out that the inter-annual variation in aquifer extent may be the result of noisy data, uncertainty in the validity of the DOY80 threshold, or controlled by variations in total accumulation and melt from year to year.
I would encourage the authors to at least try to quantify how much of the inter-annual variation in aquifer extent is robust to small changes in the DOY80 threshold (greater or less than 105) or to uncertainty in where the backscatter time series crosses the 80% threshold due to radiometric uncertainty or high frequency oscillations in the backscatter time series. (I recognize that the time series has been smoothed, but the smoothing window is pretty arbitrary, so there is some implicit uncertainty associated with the choice of that window.)
Line Comments:
Line 24 – perhaps specify “refreezing in the firn” here
Line 24 – consider more than one citation here, perhaps 1-2 supporting papers for each of the mechanisms discusssed (refreezing, runoff, supraglacial storage, etc…) to better support a general point about all ice shelves, rather than just the Roi Baudouin Ice Shelf discussed in the Lenaerts (2017) paper.
Line 26 – citation? I am not sure that we have a well-developed idea of what a “seasonal firn aquifer” really is, since it is not clear that such a thing has been observed in the field. Throughout the paper, the authors seem to use this as a catch-all term for a damp firn layer that refreezes over the winter, which I am not sure necessarily qualifies as an aquifer. It might be better just to refer to these seasonal signals as “transient liquid water storage” or something similar.
Line 57 – how is the data reprojected onto the EASE grid? What type of interpolation or binning is used to assign values to a grid cell? How are the native resolution data aggregated within a grid cell – taking the mean, for example? If data are interpolated or average, is this done in linear of dB space? Can the statistics of the native pixels assigned to each grid cell provide an estimate of the radiometric uncertainty within each EASE grid cell?
Line 59 – roughly how much do geometric parameters like look angle vary from RON to RON? At high look angles, is there any concern that the backscatter will hit the noise floor and bias the average values used for geometric bias correction?
Line 61 – does “the bias is determined pixel-wise” mean that a unique bias is calculated for each pixel? If so, how is the average calculated for the RON? Or does this mean that for any given image, the average is found, normalized by the average across all images used in the study, and then all pixels in the image are offset by that difference?
Line 67 – what set-up is used for the SMRT runs? IBA vs DMRT vs Mie/Rayleigh scattering? Autocorrelation function and correlation lengths?
Line 70 – how are these parameters chosen? How much does the reliable penetration depth vary with the density or grain size of the overlying snow or the liquid water content of the wet layer? This setup makes sense for understanding the impact of accumulation events over wet firn on detection, but may not be particularly representative of system sensitivity to wet firn layers deeper within an established firn pack.
Line 92 – how sensitive are the results to the smoothing threshold?
Line 130 – what electromagnetic model is used (IBA, etc) and how are the microstructural parameters chosen?
Line 133 – section 3.1 might be better placed with section 3.4, since the main purpose of 3.1 seems to be to convince the reader that changes in IMAU-FDM LWC should be comparable to the Sentinel-1 signal and to highlight some of the uncertainties in the results.
Line 149 – have you considered masking out regions of bare ice or ice slabs as simulated by IMAU-FDM when looking for firn aquifers? This might help avoid the potential for mixed signals from buried lakes.
Line 152/Figure 5 – is there a reason to show the spatial distribution of DOY80 since a strict cutoff threshold is used for aquifer detection? I actually do think it’s useful since the cutoff is uncertain, but it is worth discussing spatial patterns in that uncertainty in the text if you choose to show it.
Line 173 – I am not sure that melt underestimation in the model due to surface ponds should apply. Presumably firn aquifer areas should be largely mutually exclusive with areas of surface ponding?
Line 211 – where did 15m come from when your SMRT sensitivity study showed an effective penetration depth of only 7m?
Line 212 – why not try to incorporate this into the algorithm?
Line 200 - section 4.1 – this is a good and very important section that places the results of this work into the appropriate context of the uncertainty
Line 249 – again, I feel strongly that the uncertainty analysis and the system sensitivity are not sufficiently robust to make any real statements about seasonal variability.
Line 260 – any of the airborne OIB data available will be from November, so this requirement has been met by most if not all previous ice-penetrating radar data collection
Line 266 – much too strong of a statement given your level of evidence! Should you expect to see large floating ice shelves on the western side of the peninsula given the local ice dynamics and calving rates? Is there evidence of past ice shelf collapse along the western side of the peninsula in regions where you detect aquifers? You can speculate on these questions, but a statement that perennial firn aquifer development at the grounding line is a precursor for ice shelf collapse is not supported by the data in this manuscript.
Citation: https://doi.org/10.5194/tc-2022-127-RC2 -
RC3: 'Comment on tc-2022-127', Anonymous Referee #3, 20 Nov 2022
The manuscript decribes the development of new algorithm that uses an arbitrary set of thresholds and fixed dates to map ‘seasonal and perennial firn aquifers’ on the Antarctic Peninsula, and then comapres these results to IMAU-FDM. The manuscript further details the simulation of backscatter time series using IMAU-FDM and the SMRT radiative transfer model to ‘validate’ the comparison. ‘Seasonal firn aquifers’ are mapped if meltwater is detected in one year. ‘Perennial firn aquifers’ are detected ‘by proxy’ if meltwater is detected for one or more years. The authors acknowledge that they cannot tell the difference between the signals. Although the comparison between the S1 detection algorithm and IMAU-FDM appear to be in broad agreement when combined over the three-year time series, no year-by-year comparison is provided. Recent field measurements and OIB observations of an expansive perennial firn aquifer on the Wilkins Ice Shelf – which neither the S1 detection algorithm or IMAU-FDM are in agreement with – are not discussed. Alternatively, the authors suggest that coincident field observations are required for validation, which implies that perennial firn aquifer form and refreeze on a regular basis. The manuscript concludes with statements on perennial firn aquifers and ice shelf collapse that are not supported by the study.
Major Comments –
There are significant technical and theoretical issues throughout this manuscript detailed in the minor comments. Assumptions and conclusions are made that are simply not supported by the analysis. In particular, it is unclear how and why the thresholds were chosen other than to simply reproduce the IMAU-FDM simulations. Given that the study uses a radiative transfer model, it would be reasonable to expect that the threshold for subsurface meltwater would be chosen based on a series of simulations. However, that is not the case. The assumption that these thresholds remain stable temporally and spatially is not plausible. There is nearly a 10 C difference in the mean annual temperature between the northernmost islands and the southernmost George VI Ice Shelf. Many of the northern locations have a mean annual temperature near 0 C. This results in a significant difference in the melting seasons between locations. This is reflected in the final mappings which show the north and northwestern Antarctic Peninsula as primarily ‘perennial firn aquifers’, when what the algorithm is actually detecting is surface meltwater during long-duration melting seasons. This is further complicated by the shallow penetration depth of S1, which cannot detect meltwater beyond a few meters’ depth in the percolation zone. The lack of a detection can simply imply that meltwater is present, and has simply descended below the penetration depth prior to the fixed April 14th date. Overall, my conclusion is that the S1 detection algorithm is mapping both surface and subsurface meltwater within a fixed time interval and limited penetration depth. Unfortunately, I am unable to recommend publication of this manuscript.
Minor Comments:
Line 36-38 – Both the Wilkins (Montgomery et al., 2020) and Muller ice shelves (MacDonell, 2021) have perennial firn aquifers confirmed via fieldwork. The perennial firn aquifer on the Wilkins Ice Shelf is confirmed throughout the winter of 2017-2018, again throughout the winter of 2018-2019, and 2019-2020. Firn cores confirmed ~20 meters of meltwater at depths of ~15 m below the surface. The perennial firn aquifer was also confirmed within a ~15 km radius by GPR surveys in December 2018.Line 56-57 – Why is horizontal transmit/receive polarization used? Were both channels analyzed?
Line 58 – The gridding is 1 km x 1km. The spatial resolution of the measurements is 20 m x 40 m.
Line 58 – What is the temporal resolution of the combined S1-A and S1-B? 12 days?
Line 60 – What is the incidence angle range of the measurements before averaging? How does averaging measurements with different orbital geometries (i.e., incidence and azimuthal angles) over the complex topography of the Antarctic Peninsula bias the measurements? Can the authors please cite or justify the averaging method?
Line 62-63 – What is the maximum and minimum number of data points ‘eliminated’ from each time series used in the algorithm? How does that influence the targeted signal spatially and temporally in terms of the algorithm performance, particularly in 2017-2018?
The authors use three (not four) years of data 2017-2018, 2018-2019, and 2019-2020. However, none of the provided plots or maps show 2017-2018, which is odd for such a short time series. Can the authors please provide a reason for this omission? If the data isn’t suitable to show, is it suitable to use in the S1 detection algorithm? Are the results viable?
Please cite the ice masked used in the detection algorithm.
Line 120-121 ‘IMAU-FDM does not simulate standing water or lateral water flow, so only a qualitative comparison can be made with S1 detected aquifers on the basis of the presence of irreducible liquid water content’.
If these processes were included in IMAU-FDM, can the authors please describe how a quantitative comparison could be made? The S1 detection algorithm is binary.
Line 124-128 – The S1 detections are 20 m x 40. The IMAU-FDM is 5.5 km x 5.5 km. Can the authors please describe how the time series and map comparisons were made?
Line 133-135 – An observational technique (S1 detection algorithm) can’t be ‘validated’ using a model simulation (SMRT) parametrized with a model simulation (IMAU-FDM). The simulations are simply not real.
Line 188-190 – ‘The South Shetland Islands show almost complete firn aquifer cover, confirmed by local observations (Jiahong et al., 1998; Travassos and Simoes, 2004; Macheret et al., 2009).’
These studies do not describe a ‘complete firn aquifer cover’. These results describe a ~1 m water table above the firn ice transition at ~ 30 - 50 m depth in a temperate glacier at field sites. These studies were conducted in 1985-1992, 1997-1998, and 2000-2006 – as much as 30 years prior to this study, and not coincident in time with S1. No comparisons between the S1 detections and these fields sites are provided.
Noting that the penetration depth of S1 is several meters – can the authors please justify using these field sites to ‘validate’ the detection algorithm?
Can the authors also please justify using these field sites to ‘validate’ the detection algorithm, when the field sites on the Wilkins Ice Shelf and the Muller Ice Shelf - which are coincident in time and do not appear to be consistent with the results of the S1 detection algorithm - are simply disregarded as ‘complex’, citing ‘dissimilar quantities in different time scopes’?
Line 202-203 – ‘The increase of backscatter over time after the peak melt season can be caused by the wet layer getting buried under fresh snow accumulation’
Backscatter from snow accumulation at C-band wavelengths is negligible.
Line 235-236 This issue will become less relevant in the coming years as the amount of data is increasing, although a complication is that Sentinel-1B has failed since December 2021 with no guarantee of operating again’.
Sentinel 1-B has failed. Only Sentinel 1-A is operational. The authors stated in the methods section that ‘sufficiently dense data coverage for our purpose was only reached after the launch of Sentinel-1B in April 2016’. If Sentinel-1B failed, and both satellites are required for ‘sufficiently dense data coverage’, can the authors please describe how is the amount of data increasing?
Line 253-254 - ‘Comparing the satellite detection results with in-situ measurements on the Wilkins and Müller ice shelves (Montgomery et al., 2020; MacDonell et al., 2021) is complex as dissimilar quantities in different time scopes are measured.’
This is an overly complicated sentence, when the comparison is simple. The authors are using S1 data with a resolution of 20 m x 40 m - comparable to a firn core. The authors claim to be ‘confidently’ mapping ‘seasonal and perennial firn aquifers’ over the Antarctic Peninsula. There is a field confirmed perennial firn aquifer on the Wilkins Ice Shelf in 2017-2018, in 2018-2019, and 2019-2020. Not simulated. Confirmed. At some point near the end of the austral meting season in 2017, 2018, and 2019, the seasonally recharged upper layers of the perennial firn aquifer were within the penetration depth of S1 and detectable. The authors appear to detect a ‘seasonal firn aquifer’ on the Wilkins Ice Shelf in 2018, however - as previously noted – the data is oddly not shown. The algorithm fails to detect even a ‘seasonal firn aquifer’ in 2019 or 2020.
Can the authors please justify the lack of a S1 detection?
Line 254-258 ‘On Wilkins Ice Shelf, the reported aquifer water tables were typically located at 6 to 22 meter the surface in 2014, and IMAU-FDM does not provide indications that the very deep aquifers have been recharged in recent years…'
The OIB-derived detections reported in Montgomery et al., (2020) extend over nearly the entire Wilkins Ice Shelf. Again, not simulated. These are observations. Furthermore, to attenuate the low-frequency MCoRDS signal, the detection implies that a relatively thick perennial firn aquifer with significant volumes of meltwater is present at depth. The assumption that the authors seem to make throughout the manuscript - that perennial firn aquifers form and refreeze regularly on the Wilkins Ice Shelf (and everywhere else) - is simply not plausible. Can the authors please justify this assumption?
IMAU-FDM does not appear to simulate the extent of the perennial firn aquifer confirmed by observations over the Wilkins Ice Sheet in 2014 (van Wessem et al., 2020). Nor does it simulate the perennial firn aquifer 2017, in 2018, and 2019. Can the authors explain the relevance of the fact that IMAU-FDM does ‘not provide indications that the very deep aquifers have been recharged in recent years’ to the S1 detection algorithm results? If IMAU-FDM is not capable of simulating the field confirmed firn aquifer on the Wilkins Ice Shelf, can the authors justify the comparison between the S1 detection algorithm and IMAU-FDM over the entire Antarctic Peninsula?
Line 259-261 – ‘further analysis of OIB flight measurements would be useful for comparison as they provide information on the firn aquifer water table on a larger scale. Especially collecting measurements before the peak melt season would be valuable as they allow detection of perennial firn aquifers which are more robustly defined than seasonal aquifers’
There are almost 20 years of OIB data, including OIB-derived perennial firn aquifer detections (Montgomery et al., 2020) publicly available and not used in this study.
Line 264 – 268 - ‘Finally, it is noteworthy that the existence of perennial firn aquifers at the grounding line and the presence of extensive floating ice shelves appear to be mutually exclusive on both sides of the AP (Fig. 7). This suggests that perennial firn aquifers play an important role in ice shelf viability and demise: if a perennial firn aquifer develops around the grounding line or on the ice shelf, it is a precursor for collapse.
Big conclusion that is not supported by the results of this study.
Line 264 – 268 – ‘An exception is Wilkins ice shelf, but it is well known that this ice shelf has partly disintegrated in recent years (Braun et al., 2009).’
How exactly did the formation of perennial firn aquifers at the grounding line of the Wilkins Ice Shelf contribute to the collapse and disintegration of the ice front?
Figure 2.
SMRT is a snow model. It is not appropriate for modeling deeper subsurface meltwater within the complex stratigraphy of the percolation facies. During refreezing, the dominant backscattering response is via volume scattering from ice pipes and lenses – which significantly reduces penetration depth. This is well-documented in the literature. Additionally, in perennial firn aquifer areas, SMRT does not include latent heat release from refreezing meltwater.The SMRT simulated ‘dry’ and ‘wet’ layers are inconsistent with the actual backscatter data shown in Figure 4.
The ‘dry layer’ is simulated with backscatter values of between ~ -25 dB to -15 dB.
These values are consistent with volume scattering from dry snow in the interior of the ice sheet – not the percolation zone where volume scattering from ice pipes and lenses dominates the response. A realistic ‘dry layer’ in the percolation zone that included ice pipes and lenses would plausibly have backscatter values between ~ -5 dB and 0 dB, which is consistent with the plots shown in Figure 4, except for the Muller.
The ‘wet layer’ is simulated with backscatter values between ~ -15 dB and almost -40 dB.
However, in the percolation zone, it is uncommon for the backscatter value to drop below ~ -25 dB at C-band. This value is consistent with significant surface melting, not subsurface meltwater. So, in terms of observations, the minimum value (not simulated or the ‘radiometric accuracy’) should reasonably be set at ~ -25 dB. This is also consistent with peak melt on any of the plots shown in Figure 4.
Setting the lower threshold at -25 dB would decrease the ‘detection limit’ to ~1 m.
Can the authors justify a reasonable detection of ‘seasonal and perennial firn aquifers’ at these shallow depths?
Alternatively, can the authors justify their modeling approach? Specifically, why are uniform density, grain size, and LWC values used to depths of 7 m, rather than depth-dependent parameters? Is the 5% LWC value consistent with field measurements? How does the detection limit change with increasing and decreasing LWC values?
Figure 3
The authors state that this plot represents an ‘idea case’ – why was this plot chosen? What are the dates? What are the lat/lon coordinates? Is it a ‘seasonal’ or perennial’ firn aquifer? If you can’t tell the difference – why is it ideal?
Can the authors explain the significant differences between the ‘ideal case’ and the Muller Ice Shelf site shown in Figure 4, which obviously shows extended, possibly year-round surface melting? How does that influence the detection algorithm?
Can the authors show that the ‘ideal case’ characterizes the majority of the ‘perennial firn aquifers’ in the northern Antarctic Peninsula + islands? Or do those sites look more like the Muller Ice Shelf site?
Figure 4
The authors need to provide lat/lons for each of these sites. The Wilkins Ice Shelf and the Muller Ice Shelf sites appear to be field locations from other studies (Montgomery et al., 2020; MacDonell et al., 2021) The authors need to acknowledge this work, and describe what was found at these sites.
Note: There are issues with using a 15 -day smoothing filter on the S1 backscatter and distinguishing surface melting from subsurface meltwater, particularly at the end of the melting season. The smoothed signal with ‘mimic’ the subsurface meltwater signal if there is late season surface melting. Similarly, the IMAU-FDM is simulated on 10-day intervals, which makes time series comparisons difficult to interpret.
Note: The authors state that the peak melting season is between 1 and 15 January – yet many of the time series, suggest this data is much later, with significant variability between years. As an example, in 2020, the peak melting season appear to be in March, with similar shifts on the George VI in both 2019, and 2020.
The authors state that ‘Line 143-145 - there is a clear correspondence between the presence of liquid water in IMAU-FDM and the corresponding forward modelled SMRT S1 backscatter time series on the one hand, and the observed S1 backscatter time series on the other hand’.
Given that the IMAU-FDM is the primary input into the SMRT – it is expected that there would be a ‘clear correspondence’ between the two.
However, there is no obvious correspondence between the SMRT and S1 backscatter on the Larsen C, George VI, or Wilkins. The SMRT does not simulate the identified subsurface meltwater signal at the end of the melting season – which suggests that the dominant backscattering responses are not simulated in SMRT, or that the IMAU-FDM is not adequate simulating the actual firn conditions at the surface. In 2019 – there are several months where the there is clearly significant subsurface meltwater detected by SI.
The Wilkins Ice Shelf, in particular, is a field validated perennial firn aquifer site that the IMAU-FDM is not simulating and the SI algorithm is not detecting in either year. This is a strong indication that the IMAU-FDM will not simulate accurate results. Furthermore, the 2019 Wilkins Ice Sheff site and 2020 Larsen C Ice Shelf site (Line 135-136 - ‘a location with rapid meltwater refreezing and therefore small amounts of retained liquid water ‘) appear to be near identical in terms of the S1 detection – shifting a threshold would identify aquifers on both sites in different years. This is a strong indication that the S1 detection algorithm is not viable.
On the Muller Ice Shelf in 2019 – the backscatter values appear to be near -15 dB at the start of the time series, and experience a limited drop in backscatter values during the melting season. This suggests that the firn is initially wet at or very near the surface and gets wetter throughout the melting season. It would be difficult to justify these backscatter values from subsurface meltwater underlying ice pipes and lenses. Once surface meltwater began the refreezing process, strong scattering would increase backscatter values very quickly. The authors potentially attribute this to supraglacial lakes (Line 149-150 - The S1 backscatter signal could represent persistent or slowly buried supraglacial lakes’), however, this appears to be the field confirmed perennial firn aquifer site described in MacDonell et al., 2021, where supraglacial lakes are not observed. In 2020, there is obviously surface melting throughout the entire year. Even smoothed, the backscatter time series show significant variability. Backscatter values appear to be below -15 dB prior to the start of the 2020-2021 melting season. The detection thresholds are chosen during active surface melting at this site. The algorithm is detecting active surface melting, not subsurface meltwater. Even know this is a field confirmed perennial firn aquifer site, subsurface meltwater is masked. It is simply not possible to detect ‘seasonal and perennial firn aquifers’ during active surface melting. This is a second strong indication that the S1 detection algorithm is not viable.
Unfortunately, the backscatter time series on the Muller Ice Shelf are characteristic of the entire north and northwestern Antarctic Peninsula.
Citation: https://doi.org/10.5194/tc-2022-127-RC3
Data sets
Modelled and Sentinel-1 detected firn aquifers areas in the Antarctic Peninsula Lena G. Buth, Sanne B. M. Veldhuijsen, Bert Wouters, Stef Lhermitte, Michiel R. van den Broeke https://doi.org/10.5281/zenodo.7113603
Model code and software
Sentinel-1 detection of seasonal and perennial firn aquifers in the Antarctic Peninsula Lena G. Buth https://gitlab.awi.de/lenbuth/tc-aquifers
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 906 | 399 | 47 | 1,352 | 45 | 38 |
- HTML: 906
- PDF: 399
- XML: 47
- Total: 1,352
- BibTeX: 45
- EndNote: 38
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1