This study characterizes the melt and refreeze evolution across a portion of the Wrangell and St. Elias mountains. The study offers an interesting method to characterize various zones of dry firn, wet firn, and melt by assessing the variability in the delay of wet snow/firn refreeze (as detected from Sentinel-1 SAR backscatter) compared to the 0˚C isotherm (derived using temperature lapse rates from weather stations across the region). The study is compelling and seamlessly combines remote sensing observations with in-situ observations and insights from models. The paper is also written at a very high quality, was generally easy to follow, and enjoyable to read. However, I have some major reservations about some of the results, particularly the classification of the ablation zone, and limitations associated with the melt detection algorithm.

Main comments

The level of writing of this paper is very high (definitely sufficient for publication). Still, readability would be improved with the removal or vast reduction of acronyms (S1, S1A/B, AOI, VV, VH, SEB, PDD, PR, K/H, ELA), especially for non-technical readers. I would highly recommend writing these out in full. This is ultimately not crucial and up to the discretion of the authors. Note some other acronyms not mentioned that are probably okay to use: SAR, AWS, HOBO, DOY.

The authors claim in the introduction and conclusion to be the first study to assess melt across the region, which is not true. Wells et al. (2026) also used Sentinel-1 SAR backscatter to produce melt onset time series for all glaciers >2 km2 in the region. It may be worth comparing these results in the supplement, which use the same -3 dB threshold from mean winter backscatter but for slightly different data (Wells et al. use cross-polarized SAR, this study uses co-polarized SAR) and slightly different onset elevation determination (Wells et al. use an ‘all melt’ threshold and hypsometric approach per-glacier, this study averages melt onset maps during elevation binning of all pixels in the study domain).

My main concern regards the firn melt zone classification results (as shown in Fig. 7). The accumulation area of the glaciers in this region is roughly 50-55% (Zeller et al., 2025), but this study only maps the ablation area as 12% of the area. In turn, this means that nearly 90% of the glacierized area is some classification of a firn zone, with over 70% being either wet firn, dry to wet firn, or dry firn. The authors need to motivate the physical meaning or interpretation of the transition zones, as the "wet firn to ablation" zone almost certainly contains no firn in reality. For many glaciers in the region, the "wet firn" zone itself is delineated as covering the entire glacier.

This may be the result of processing at the study domain scale, as opposed to per-glacier. As such, melt signals are mixed and aggregated into bins that span different parts of glaciers, especially across the large range of climates (maritime/coastal and continental/interior) and elevations (0 to 5000+ m a.s.l.) within the domain. While this is a major limitation of the study, it may still be okay as long as the authors make it explicitly clear that these results should be interpreted only as a domain-wide estimate and not necessarily indicative of any particular glacier (the authors don’t parse results for specific glaciers, but this could maybe still be explicitly stated). In this case, it also might make sense to remove the map in Fig. 7 and just show the zones in relation to the regional hypsometry to avoid misinterpretation of the results.

My other major concern arises from caution about the melt detection algorithm, particularly over heavy debris cover. Heavy debris can result in that portion of a glacier not exhibiting a clear melt signal. The current algorithm does not detect or identify these pixels as melting, and thus altogether excludes them from the analysis. This is clear from the maps D1-D7, where the terminus of Malaspina glacier, Logan glacier, Chitina glacier, and others (which are heavily debris covered) are never mapped in the melt onset and subsequent maps. The debris covered pixel behavior is also clear in cross-polarized SAR maps on, for example, Chitina glacier (https://alaskasnowlines.streamlit.app/plot_gif?name=Chitina&rgi_id=16307) and Logan glacier (https://alaskasnowlines.streamlit.app/plot_gif?name=Logan&rgi_id=16822), where the backscatter at the debris-covered terminus remains pretty much unchanged throughout time. In the study results, these parts of the glaciers are primarily the regions being mapped as “ablation area” or “wet firn to ablation zone”, even though no melt onset/refreeze interpretations where made over these areas of the glacier to begin with. One way to get around this issue would be to use some sort of “all melt” threshold to account for these lower pixels that are in fact melting but are either debris-covered or generally below the snowline, but this would likely require transitioning to a per-glacier analysis, at least to develop the spatially-distributed melt onset maps.

lastly, in Figure 7, there is an artifact resulting from plotting (hopefully it’s just a cosmetic error!), the DEM, or the mapping algorithm. The glacier areas near the edges of the figure are colored as ablation zones, cutting through other parts of the glaciers in a non-physical way, even in some accumulation areas.

Specific comments (these are mostly suggestions)

• L9: The way it’s written, it almost sounds like one needs freeze-up timing, air temperature, and elevation to classify the dry firn zone. The dry firn zone should basically be the elevation above the melt extent. I’m curious if you compare the dry firn zone delineated from your method to the maximum melt extent/melting pixels that you detect (maybe with some threshold to avoid a few individual pixel outliers).
• There’s also a very interesting study by Noël et al. (2018) that assesses past and ongoing changes in the firn structure of the Canadian arctic. This probably supports some of the statements you make in the introduction here.
• Fig. 1b: somewhere specify that you are only showing elevation of glacierized land. Maybe add a coastline to the map.
• I’m not sure how important the overall study domain area and percentage glacier cover is. Instead, it’s more interesting to highlight foremost the amount of glacierized area itself and possibly the number and size of glaciers.
• L68: Remove ‘dramatic’
• L77: I would similarly suggest writing ‘co-polarization (VV)’ and ‘cross-polarization (VH)’ to make this more readable to a non-technical audience.
• L83: possibly write: “...we only use path/frame from descending Sentinel-1 orbits, which are acquired before 10:00 local time, to eliminate biased … acquisitions (all scenes in our study area are acquired before 10:00 or after 18:00).”
• L95: Could you explain the rationale for using Nov-Mar scenes for the reference period? I think using Dec-Mar or Jan-Mar might make more sense given the slow refreeze of firn aquifers (especially, for example, on Donjek glacier). This would align with the statement on L227. To be clear, I don’t think this needs to be re-done as I don’t think this would make a difference in the results since the melt signal is so much larger than the refreeze signal, but I am curious about the rationale here. It could be worth showing a histogram of the difference between the mean winter backscatter from Nov-Mar compared to from Jan-Mar for one glacier or one path/frame.
• L96: It could be worth mentioning that you may classify areas below the snowline where the surface is ice as ‘dry’ scenes despite the fact that they are melting. This logic is sound in accumulation areas or anywhere above the firn line.
• L105: I appreciate the clarity of what the backscatter signal physically indicates.
• L110: I’m curious what the Sentinel-1 scene gaps look like (i.e., where you may have >12 days uncertainty in melt or refreeze dates). It would be interesting to plot a horizontal bar for each of the 6 path/frames from 2015-2024, with the bar shaded where scenes are available/used. It might be worth simply ignoring years without sufficient scenes (or, without sufficient scenes around the melt onset or refreeze time) for a given path/frame, instead of including this potentially biased result in the averaging of all years and prescribing a very high uncertainty.
• I’m also curious if you could use ascending scenes that pass in the evening to help get better resolution for the refreeze dates. While you mentioned this is biased and cannot be used for melt onset compared to the morning scenes, the slow, sub-surface refreezing process is not as susceptible to diurnal surface melt patterns. This could help give you better temporal resolution and extend the coverage of the middle of your area of interest to the earlier years without S1B data. This would at least be worth doing a quick investigation upon, to see if the refreeze dates from ascending scenes show systematic bias to the ascending scenes or not.
• L115: Any reason why the scenes were not downloaded directly in dB scale?
• L117: I would probably rephrase this as removing high-angle or very steep terrain.
• L144: I would be curious to see what these lapse rates look like
• L146: Very cool to see!
• L150: I was initially a bit confused reading these definitions. The ‘dry to wet firn’ and ‘wet firn to ablation zone’ are simply transition zones between the other zones, but this list doesn’t offer any insight into how these are classified or any additional physical intuition about these sites. I would perhaps rephrase the way this is brought up (e.g., something like, …we define 3 primary glacier zones and 2 transition zones which account for the complex nature of…), or define them more concretely like the other zones are.
• L161: “winter dry snow backscatter” is actually backscatter from ice, debris, or ice lenses in firn that is beneath the snow. I would just say “winter backscatter” since the SAR is not reflecting off of the dry snow itself. Can also probably remove the word “values”.
• I’m also curious how you identify pixels that never have a SAR melt signal due to either heavy debris very low on the glacier or the 3% of dry firn zone.
• L166: Nice–this adds a lot of useful context! Some mention of stable vs. transitional zones would have been useful before the listed zones because it was a bit confusing as someone unfamiliar with the study/method.
• L170: Super cool. I would reorganize this and the following paragraph a bit to explain the vertical gradient calculation before explaining what a high/low vertical gradient means in terms of firn zone stability.
• L171: Am I understanding the method correctly: you bin the ∆freeze into 100m elevations, and then do a 1D filtering on the elevation-dependent ∆freeze signal? Is the elevation binning and profiling done per glacier or for the entire study region at once? Does the elevation binning not already smooth the signal enough?
• L176: What does “bands <200 m thick” refer to? Is this referring to any detected firn melt zone that doesn’t span at least 200 m elevation range?
• Could temperate and cold firn play a role in some of these differences across the region? (not suggesting any changes, just curious about this)
• Fig. 5: I like this figure a lot. Why not include 2023 for (b) and (c)?
• L248: Outsized
• L254: This is very interesting to observe in the data–I’m glad you pointed it out.
• L262: I appreciate the discussion here, particularly tying in some of the nuances and work from previous studies.
• L283: Zeller et al. (2025) and Bevington and Menounos (2025) produce ELAs for basically all glaciers across the study area you assess. How do those compare with your mapped zones?
• L284: Feels strange to spell out and specify ‘ELA’ here after using it so many times already.
• L285 and Fig. 5: I’m a bit confused why the “freeze-up” is detected so late at low elevations. Many of the pixels at low elevations return to within 3 dB of the winter mean backscatter very early in the season, after the snowline has retreated above them. What prevents your algorithm from defining these as “dry” pixels after the snowline has retreated?

References

• Wells et al. (2026) https://doi.org/10.1038/s41612-026-01321-y
• Zeller et al. (2025) https://doi.org/10.1017/jog.2024.65
• Noël et al. (2018) https://doi.org/10.1029/2017JF004304
• Bevington and Menounos (2025) https://doi.org/10.1088/1748-9326/adc9ca

July 13, 2026

https://doi.org/10.5194/egusphere-2026-2431

Review of: Seasonal surface melt onset and firn freeze-up across the central Wrangell and St. Elias Mountains

Kindstedt et al.

This study reports on firn melt dynamics using C-band SAR and modelled climate data.

General comments:

The manuscript is well written.  More detail is required relating to some methods and results.    

There are issues pertaining to the climate modelling and instrument data that needs to be addressed before the manuscript is acceptable for publication. 

The interpretation of the SAR data is done with a climate model that requires work.  There needs to be a rigorous definition of melt that is independent of the remote sensing used to detect melt.  Snow and ice are biphasic and contain liquid water as a function of temperature. Is there other data that can independently verify the interpretation of the SAR data?  The inclusion of optical albedo and surface infrared thermal satellite data would help with the interpretation. 

Most of the validation data comes from the east side of the precipitation divide, but the parameters from these data are applied to the west side of the icefield that has much different temperature and precipitation characteristics.  Your own statistical model indicates distance from the ocean is very important.  Much more detail is required in understanding how the model choices and spatial heterogeneity of data influence this studies results. 

Radiosonde and re-analysis should be consulted for lapse rate evaluation. 

Detailed comments:

Many instances of undefined words need revision.  For example, in the Abstract, “all elevations”, “mid-elevation”, “highest elevations”, “area of interest” are all undefined.  Please review the manuscript the remainder of the manuscript for this issue.  Examples include, “some regions” on line 21, “high annual snow accumulation” on line 45. 

The last sentence of the Abstract needs to better convey the importance of this work.

Line 28 – by climate records do you mean ice cores?  There are many types of climate records. 

Line 30-32 – this sentence is obvious.  It is unclear to me why this citation is necessary. 

Line 44 – extensively is better than heavily. 

Line 54 – Define what is “novel’. 

Figure 1 – is there a representation of the precipitation divide that can be included on this figure?  It appears the stations are all on the west side of the divide. 

Line 94 – a pixel is a picture element on a computer screen.  The resolution of the pixel is determined by the properties of the device and thus are variable across platforms.  There are better descriptions of gridded remote sensing data, such as array element, grid cell, etc. 

Line 96 – is there a citation for the 3dB value.  If not, a justification for its use is required.  Reconcile the RS definition of melt to physical definition of melt.

Line 112 – there needs to be meta data included that documents the time of capture of the RS data.  Any trend in time of capture or variation in time of capture can have unevaluated effect on the interpretations. 

Line 116 – median filters will smooth data – what is the effect on detecting melt across spatial distributions?

Line 117 – How are slopes determined?  If with a DEM then analysis is required to show how robust the method is.  DEMs are notoriously inaccurate over snow and ice. 

Line 122 – Define physically meaningful - the sentence is difficult to interpret and confusing.  How does the SD compare to a "drop" of 3dB?  SD is a measure of variation around a central tendency.     

Figure 2 – increase font size.

Section 2.6. 

Provide background information on HOBO temperature loggers.  Make, model, precision, integration, shielded, movement (x,y,z) over the study period, etc.

The highest station is well below 6000 m, so not 0 – 6000 m.  And the data records are of various lengths, with PR-Col being the shortest.  The lapse rates will be of various heights depending on the lengths of the data records.  There are going to be calibration issues related to the PR-Col station that need to be discussed.  Typically radiometric instruments need to be calibrated every two years.  Explain/discuss. 

Table 1: Add months to the Time period covered field. 

Lines 150-154 – What are the criteria for these classifications?

Line 167 – “transitional” should not be a quote. 

Line 171 – citation for filter and usage?

Section 2.8:

This section requires a substantial amount of work.  To gain a better understanding of the factors that influence glacier melt models I suggest you read some papers found here: https://scholar.google.com/citations?user=86dWcwkAAAAJ&hl=en

The energy balance requires a more mature treatment of albedo.  Albedo has been measured in the St. Elias, see: https://www.sciencedirect.com/science/article/pii/S1873965216300500

Also, see section 4d of this paper which you have cited elsewhere: https://journals.ametsoc.org/view/journals/clim/33/8/jcli-d-19-0405.1.xml

Albedo contribution to glacier melt can be found here: https://www.nature.com/articles/s43247-025-02503-x

Much more detail regarding the Brutsaert (1975) model needs to be included.  For example, longwave radiation if largely the result of the bulk atmosphere in the bottom 1 km.  If surface temperature is used without modification for lw radiation, then the values will be much larger than actual.  Clear sky e will be much lower than cloudy sky e.  Thus the amount of lw will be greater in all sky conditions compared to clear sky.  There is a recursive shortwave radiation relationship between high albedo snow and clouds – thus somewhat counterintuitively the downwelling solar radiation can be larger under partial cloud cover than during clear sky.  Has this effect been accounted for?

Line 214 – define generally

Line 218 - nice

Line 221 – dramatic

Line 217-224 – unclear.  Revise for clarity.  Eg., positive relationship with elevation – this appears to be backward. 

Line 226-227 – explain why freeze up is so late.  How is freeze up determined?  With which data?

Figure 4 is difficult to read and needs to be revised to better convey the meaning.  As currently presented I can’t evaluate this figure. 

Figure 8: to my knowledge net radiation has not been measured in the St. Elias.  I assume that net radiation here is modelled.  If modelled from the parameters provided earlier in the manuscript, then these values will be inaccurate.  A calculation of error is required.

Line 255 - indeed

Line 265 – percolation and refreeze models exist – see work by Shawn Marshall and others. 

Line 268 – drilling of what?

Appendix A

Table A1 – Include dates.  If difficult to include individual dates then include average date each year and standard deviation.  Separate by number of scenes for melt and refreeze.  The reader needs to be able to evaluate if the number and timing of observations influences the results.