How Flat is Flat? Investigating the spatial variability of snow surface temperature and roughness on landfast sea ice using UAVs in McMurdo Sound, Antarctica

Martin, Julia; Dadic, Ruzica; Anderson, Brian; Pirazzini, Roberta; Wigmore, Oliver; Vargo, Lauren

doi:https://doi.org/10.5194/egusphere-2025-1601

Preprints

https://doi.org/10.5194/egusphere-2025-1601

Preprints

14 May 2025

| 14 May 2025

How Flat is Flat? Investigating the spatial variability of snow surface temperature and roughness on landfast sea ice using UAVs in McMurdo Sound, Antarctica

Julia Martin, Ruzica Dadic, Brian Anderson, Roberta Pirazzini, Oliver Wigmore, and Lauren Vargo

Abstract. How do snow distribution patterns influence the surface temperature of snow on sea ice? Despite its crucial role in the sea-ice energy balance, snow on Antarctic sea ice remains under-sampled and poorly understood. To address this knowledge gap, we used an Uncrewed Aerial Vehicle (UAV) and ground measurements to produce a Digital Elevation Model (DEM) of the snow topography and a map of snow surface temperature over relatively uniform landfast sea ice (2.4 ± 0.04 m thick) in McMurdo Sound, Ross Sea, Antarctica during our field season in November-December 2022. A key methodological innovation in this study is an algorithm that corrects thermal drift caused by Non-Uniformity Correction (NUC) events in the DJI Matrice 30T thermal camera. The new algorithm minimizes temperature jumps in the imagery, ensuring consistent and accurate high-resolution (9 cm/px) snow surface temperature maps. Our airborne maps reveal a mean snow depth of 0.16 ± 0.06 m and a mean surface temperature of -14.7 ± 0.4 °C. As expected, the largest surface temperature anomalies were associated with visible sediment depositions on the snow surface, which were manually identified. We found that the small-scale topography on a seemingly flat snow field significantly influences the incoming solar radiation (irradiance) at the point scale. Using a model that accounts for topographical effects on irradiance, we found that assuming uniform irradiance over our study (200x200 m) area underestimated irradiance variability due to relatively small-scale surface topography. The modeled mean irradiance, which accounts for surface topography, is 592 ± 45 Wm⁻²(1 Standard Deviation), whereas the mean measured irradiance at the point scale is 593 ± 20 Wm⁻². This shows that assuming a flat surface fails to represent the full irradiance range and may impact non-linear energy balance processes. While we initially hypothesized that snow depth was a key driver of snow surface temperature, our results indicate that sediment deposition and irradiance exert a far greater influence, overriding the effect of snow depth for this test site. Our results improve our understanding of snow’s spatial distribution, how it influences snow surface temperatures and how it may influence the sea-ice energy balance.

Received: 04 Apr 2025 – Discussion started: 14 May 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Julia Martin, Ruzica Dadic, Brian Anderson, Roberta Pirazzini, Oliver Wigmore, and Lauren Vargo

Status: final response (author comments only)

RC1:
'Comment on egusphere-2025-1601', Anonymous Referee #1, 20 Jun 2025
General
The authors investigate how snow distribution patterns affect the surface temperature of snow on Antarctic sea ice, a key but under-studied factor in the sea-ice energy balance. The introduction puts the topic of the paper well into context and provides a great overview of the current state-of-the-art. To contribute to this vast topic of polar sea ice, the authors applied UAV and ground measurements to create high-resolution maps of snow topography and surface temperature over uniform landfast sea ice in McMurdo Sound, Antarctica. The measurement site, the ground-based and airborne methods are introduced in detail. Regarding the latter, a novel algorithm was developed to correct thermal drift in UAV thermal imagery, ensuring consistent temperature data. Based on these maps and correlations, the authors investigated the reasons for observed surface temperature variations. As a result, the surface temperature anomalies were mainly linked to visible sediment on the snow, not snow depth, which has been the authors initial hypothesis. They further found that small-scale topography significantly affected local solar irradiance, and assuming uniform irradiance underestimated its variability. Overall, sediment and irradiance were found to have a stronger influence on snow surface temperature than snow depth, highlighting the importance of surface features in energy balance modelling on sea ice.

The manuscript provides an important contribution to the analysis of polar surface temperature variations. It presents interesting and valuable results, which help to identify gaps in common surface temperature retrievals assuming flat surfaces. It highlights the problems and mismatches they struggle with and introduces proper solutions. I highly recommend its publication after the authors have revised the manuscript regarding the comments listed below.

Major comment

Length: The paper is very long, which makes it difficult to keep the readers attention from the beginning to the end. However, I think there is some potential to shorten the paper significantly.
In my opinion, there is no reason to separate the results and discussion sections. On the contrary, the separation results in a lot of repetition, which unnecessarily lengthens the paper. By merging the two sections, this could be avoided, and the paper could be significantly shortened, which would also make it more focused. The individual sections already have similar topics, so this should be easy to do.

There are several graphs that represent more or less the same thing. For example, Fig. 1a,b and Fig. 5 a,b. One figure, either a or b, would be sufficient here. Some figures can be merged. For example Figure 8 and 9. Why not have one column with the Red Band Value (G and B are not really used), one column for the DEM and one for the temperature? Furthermore, there are graphs that in my

opinion are not needed at all. For example, Figures 10 and 13. I have also made

additional suggestions under Minor and Technical Corrections.

Some further sections can be skipped as own sections and merged into others. For example, the main message from 2.3.4 fits to the introduction of the camera, Section 4.5 belongs to the summary and conclusion part.

Figures: The figures are often not really introduced, but are only mentioned in brackets after certain statements, so that the reader has to find out for himself what is shown. This makes it difficult to read fluently and understand directly. It would be good to describe what is shown in the text with one or two sentences.

Minor comments:
Order: Figures are partly not numbered in the order they are used in the text.

Space signs: Space signs are multiple times missing between numbers and units, between figure abbreviation and figure number and in front of citations that are given in brackets.

Indices and units: Indices are sometimes written in italic letters and sometimes in non-italic letters. For reasons of consistency, you should write all indices in non-italic letters.

P2, L46: What means the original hypothesis was? It should stay the same, although it might have been rejected.

Sect. 2.3.2: You might have to revise this section to make it more clear to the reader what happens here. I had to read it several times and I am still not sure if I got it right. Do you match the in situ snow depth measurements by their GPS position into the RGB images? Or is it some kind of a stereographic method? Please revise it to make this easier understandable. Maybe also a sketch might help to better understand the procedure behind.

Sect. 2.3.4: I don’t think that this section is really needed. It should be enough, if you mention the conversion within one sentence, when you introduce the camera.

Sect. 2.3.5: This is a very well-thought-out method, which leads to convincing results. I just wonder how you derive the absolute calibration. Is it a predefined function for each pixel from the lab, which is then scaled for each pixel by the NUC? Or do you know the temperature of the shutter (which then acts like a black body) and in parallel is used as a homogeneous target to remove the non-uniformity of the single pixels? Furthermore, later in Line 305 you discuss a vignette effect, which originates from the lens properties. That brings me to the question where the shutter is installed. Is it installed in front of the lens or between the lens and the detector? For the latter you will imprint the structure of the lens-own temperature into the images, when you perform the NUC, which then leads to this vignette effect and might also changes the absolute calibration.

Sect. 3: The initial paragraph is a repetition and can be skipped.

P24, Fig 12: It would be great to have such a “correlation” plot for “degree of sedimentation” vs. temperature. I guess here you would find a significant high correlation. Of course, I see the point with the difficulties related to the varying camera settings as you write in Line 337 and the shadows you mention in Line 408. But shouldn’t be the first issue solved by your NUC calibration? Furthermore, you could use your DEM to extract regiones, which might be affected by shadows. Adapting both, you would be able to select scenes, which are unaffected by shadows and after a normalization you can extract an arbitrary value for the degree of sedimentation, which might range from 0 to 1.

Sect. 4.5: Belongs to summary and conclusions. There, it fits well at the very end.

Sect. 5: I would expect some words regarding the correlations and the main findings resulting from them.

Technical comments
Sun: “Sun” is a proper name and should be written with capital letter. It appears several times throughout the manuscript.

P1, L13: … our study (200x200 m) area → …our study area (200 x 200 m²)

P2, L26: Wendisch et al. 2023 might be cited as well as it summarizes a huge project with its focus on Arctic amplification.

P2, L53: → …(), colder

P3, L78: Acronym

P3, L85: Which infrared? Near-, thermal-, far-?

P4, Fig. 1: Only Fig. 1a is needed. Figure 1b does not offer to many more details. If you prefer to keep both, add E(ast) and S(outh) in Fig. 1b and draw (a) and (b) into the Figures. I have overseen it for a long time. Furthermore, there is a missing space sign in the figure caption before (yellow rectangle).

P4, L92: Add an outline at the end of the introduction?

P5, L115: Size and time period are already mentioned in the sentence before. Skip the repetition.

P5, L129: Italic index, which might be a typo. Also, two lines later.

P6, Table 1, caption: Skip “used in this paper”. Should be clear.

P6, Table 1: Maybe include some empty lines in between the different parameters. Since some parameters use two lines it is hard to see, which entries belong to the same parameter.

P6, L136: This last sentence belongs two the first sentence of this section. You should move it there. It will fit better.

P8, L190: back → black

P9, Fig. 2: The transparency is almost not visible on a printed version.

P9, L214: Here you jump over six figures to show something in a figure, which is not yet introduced. Please avoid this.

P10, L247: Differences in Kelvin, not in °C

P14, Fig. 5: … and boxplots (b) → …and (b) boxplots; actually I think that only one of the figures is needed. They more or less show the same.

P14, L300: to or by 0.53°C?

P15, L320: Fig.6a → Fig. 6a and the same for Fig. 6b a few lines later

P15, L326: … higher-resolution (10 s) sea… → … temporally higher resolved (10 s) sea…

P15, L338: better to use radiation instead of light.

P18, Fig. 8: Here, I have several things. (i) Uncommon Lat/Long values. I don’t know how to interpret them. (ii) The colour bar is unintuitive. To me blue colours give the impression of lower snow depth, but it is the other way round. (iii) The scale ranges from 0 m to 0.6 m. Does it mean that there are parts with bare ice in the images?

P19, L 398: Missing brackets for the citation.

P19, 403: skip “and a small fraction of clean snow.” It will be said in Line 405.

P19, L405: QGIS?

P19, L 408: RGB and in the next line R/G/B

P21, Fig. 10: Not really introduced and not really needed. What is the aspect?

P21, L432: flight legs instead of flight lines?

P23, L453: can’t → cannot

P23, L469: Don’t start sentence with an abbreviation.

P24, Fig. 12: Would be helpful to add a name list to the single figures. Maybe as a fourth column left or right of the graph, were you just write (vertically) All, Sediment, and No-Sediment.

P25, L486: … area of 200 m. → is no area

P26, L500: e.g. → e.g.,

P27, L557: … traps the light is a not so well expression. Try to avoid it.

P28, Fig. 14: Add All, Sediment, No-Sediment to the figures.

P28, L568: … the the …

P28, L575: Missing space sign in front of citation. Appears three times more within this and the next paragraph.

P30, L627: viable → sustainable?

P32, Fig. A3: x axis of graph b is incomplete.

P33, Fig. A4: (a) and (b) are missing inside the graphs. However, graph (b) is not really needed. It is already well visible in (a).

Reference:
Wendisch, M., et al.: Atmospheric and Surface Processes, and Feedback Mechanisms Determining Arctic Amplification: A Review of First Results and Prospects of the (AC)3 Project, Bull. Amer. Meteorol., 104 (1), E208–E242, doi:10.1175/BAMS-D-21-0218.1, 2023.
Citation: https://doi.org/10.5194/egusphere-2025-1601-RC1
- AC1: 'Reply on RC1', Julia Martin, 01 Aug 2025
  
  Author response in PDF (01_AC_01_review.pdf).
  
  Citation: https://doi.org/10.5194/egusphere-2025-1601-AC1
RC2:
'Comment on egusphere-2025-1601', Anonymous Referee #2, 01 Jul 2025

The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-1601/egusphere-2025-1601-RC2-supplement.pdf

Citation: https://doi.org/10.5194/egusphere-2025-1601-RC2
- AC2: 'Reply on RC2', Julia Martin, 01 Aug 2025
  
  Author response in PDF (02_AC_02_review.pdf).
  
  Citation: https://doi.org/10.5194/egusphere-2025-1601-AC2
RC3:
'Comment on egusphere-2025-1601', Ghislain Picard, 02 Jul 2025

Review of “How flat is flat? Investigating the spatial variability of snow surface temperature and roughness on landfast sea ice using UAVs in McMurdo Sound, Antarctica”
This article examines the spatial variations in snow surface temperature at the plot scale (hundreds of meters) on sea ice, aiming to identify the primary drivers. Snow temperature is highly sensitive to even minor changes in incoming fluxes (such as shortwave and longwave radiation) and serves as a means to explore the surface energy budget (SEB) and its unique characteristics on snow. To address this question effectively, the authors collected a unique and exceptional dataset in the Antarctic. The paper holds significant interest for the TC community but has some shortcomings, one of which is notable yet addressable.
The main issue lies in how the paper's question and objective are approached. The hypothesis proposed by the authors to justify and structure the study is unconvincing from my perspective, and they ultimately demonstrate that this hypothesis is indeed not validated. Instead, they identify one/two more adequate hypotheses, which, to my knowledge, are more obvious. While it is appreciable to build a paper around a clearly stated hypothesis, the fact that this hypothesis is not commonly supported in the literature and can be dismissed with magnitude order calculations undermines the paper's construction. What if the hypothesis had been, "Does local topography explain most of the surface temperature variations at the plot scale?" Research on the link between topography and SEB is prevalent in the literature, and the current title aligns more closely with this hypothesis. With the same dataset and results, the introduction and discussion would flow better, aligning more effectively with the results and conclusion.
Apart from this shortcoming, the paper is excellent and has many strengths, including significant work on TIR camera correction and an exceptional dataset. The paper is clear and well-written, with an easy-to-follow logic (except the order of the figures) and a generally sufficient level of detail (see the exception on the model below). However, the abstract is difficult to understand without having read the paper, as the logical progression is unclear (unlike in the paper itself). To attract more readers, I suggest rewriting the abstract from scratch, incorporating more results and fewer hypotheses.
The paper is quite long and my recommendation for the review is to shorten where possible, but avoid lengthening (except for the model description!).
Detailed comments:
L 6 “Our airborne maps reveal a mean snow depth of 0.16 ± 0.06 m”. The mention of snow depth measurements is new in this sentence. The previous sentence is about surface temperature and the end of this sentence is about surface temperature. A reorganisation is necessary.
L11 “seemingly flat snow field” can you give a number, this statement is relative to reader’s expectation of what flat is.
L13-14. It is not clear what the variability accounted in the uniform irradiance model.
What is meant exactly by “the incoming solar radiation (irradiance) at the point scale”. In principle incoming solar radiation is measured w/r to a horizontal surface. Do you mean the solar radiation received / perceived by the surface ?
L17 “While we initially hypothesized that snow depth was a key driver of snow surface temperature,” this hypothesis should be stated in the first part of the abstract (L5) to position the problem addressed in the paper, and it should be justified, backed by literature because this hypothesis is not intuitive, at least to me. From the SEB equation, I’d expect surface temperature to depend on irradiance first, the snow depth does not appear in this equation unless the conductive term is written with the Fourier law. However even in this case, the snow is far too insulating to allow this term to become significant with respect to the others in summer, especially the incoming irradiance.
L34-35: “through the satellite period” “since record-keeping began”. Should indicate the starting year to avoid ambiguity.
L47. While the authors are free to make the hypothesis they want as long as it is clearly stated – and I acknowledge it is very well done here w/r to the literature in general -- still I found this hypothesis strange. The physical reasoning behind this hypothesis should be developed a bit and examples from the literature could help.
L90. The objective would benefit to be rewritten without this simple hypothesis. A more neutral approach would be to list all the potential factors influencing the small scale variability of snow surface temperature with a short literature review for each, and reframe the goal into investigating/quantifying which term is the key driver in the specific context of this study (summer, sea-ice with small snow depth).
L116: “The other four sites” I’d remove this sentence, it diverts from the objective of the paper.
L136. Please check the correspondence between he height of installation vs the footprint areas (1.6m2 versus 0.35m2). Is the angle of installation at the sediment site different ?
L214: Check that the figures are referred in order (it seems Fig 8 is referred before others)
L215 L223: Check figure A1 reference
Figure 2. Can you show a scatterplot (+ r2 and RMSE) between the inferred snow depth proxy and the magnaprobe measurements as a validation of the approach ?
L272: It is not obvious how 0.5 °C was found based on Fig 4a. I guess it is empirical but how sensitive to conditions is it ?
L318: by curiosity, how large is this correction ?
L321: I don’t understand what this RMSE is, between what and what it is calculated (+ typo: with an RMSE => with a RMSE)
L329: “ The RMSE of the residuals of this linear fit”. RMSE → RMS or remove residuals
L330. “the square of the thermal” and “the square of the RMSE associated “ check if square is correct in both cases. I’d recommend to completely rephrase or write as an equation.
Section 2.3.8. It is not clear how the impurities are detected. Using a threshold or just visually ? If just visually, this section could be removed, and a line or two in the results section is sufficient.
Section 2.3.9. Given the critical role of the model, the level of detail should match that given for the drone. It is necessary to provide the main aspect of the model (e.g. workflow) and present the equations or refer to the equations in the cited papers for each main calculation step.
Main questions are: the diffuse component, the resolution of the calculation (in relation with the positioning uncertainties), cast shadows, multiple scattering (esp for the shadows).
L360. How does 2.4 ± 0.04 m translate into 1% variation and where 0.04m is coming from ? The histogram seems to indicate larger deviation. How relative variation is defined and calculated ?
L395. Fig 12 is referred, check the order.
L413. Isn’t it due to the correlation between snow depth and impurities ?
L469. Why is this relevant in this section ? It is well known that the irradiance depends on the local incidence angle, and not on the slope.
L476: The aspect distribution is uniform, not gaussian. The slope distribution is not Gaussian.
Figure 12. For convenience, adding titles on the rows directly in the graph would help quickly read the figure, without having to read the caption.
The x-axis scale is very large, for just a few outliers. I suggest to reduce the range to -17°C - -10°C or so. It would make the graphs H and I more convincing for instance.
L491: I’m not sure but I think that the algorithm not only correct for NUC jumps but also for other trends in the camera which are usually very large.
Note that other cameras do a “better” job in NUC smoothing which makes jumps more difficult to detect… while still be necessary to applied the necessary corrections. See for instance: Arioli, S., Picard, G., Arnaud, L., Gascoin, S., Alonso-González, E., Poizat, M., and Irvine, M.: Time series of alpine snow surface radiative temperature maps from high precision thermal infrared imaging, Earth Syst. Sci. Data, 16, 3913–3934, doi: 10.5194/essd-16-3913-2024, 2024
Ideally, one would access the raw data… but it seems that camera manufacturer prefer to overprotect their (insufficient) algorithms.
L500. I’d advocate for more accurate sensors than Apogee sensors when absolute value is important (e.g. close to 0°C, see Arioli et al. 2024)
L539. “While the red band values do not directly affect the surface energy balance, we use them as a proxy for impurities.” I don’t understand this statement. Maybe the verb “affect” is incorrect.
L550. I would suggest to coarsen the resolution a bit to account for the positioning uncertainty and to see how this correlation increases. Mathematically, the correlation always increases with smoothing, but here the idea is to see how quick it increases.
L567: the the => the
L570: while this result is sound, the statistical demonstration would require first to demonstrate that topography and impurities are independent in your case. For instance if the impurities areas had more north looking slopes, the relationship is biased. It is frequent (in the mountains) that the sun facing slopes are more likely to have dust emerging at the surface than the colder faces.
L580 I suggest to also mention multiple scattering which is likely important in the cast shadows areas and cavity effects in the LW which is probably negligible with slopes <10°. Ref: A. Robledano, G. Picard, L. Arnaud, F. Larue, I. Ollivier, Modelling surface temperature and radiation budget of snow-covered complex terrains, The Cryosphere, 16, 559–579, doi:10.5194/tc-16-559-2022, 2022
L608: This is not necessarily a drawback. If only the irradiance is changing (not Tair, not wind), observing two different Ts give a lot of information on the balance between SW and the other terms of the SEB.
L630: “offering valuable tools for many users”. Same comment as before. To my experience, using more expensive cameras with better NUC correction make the proposed solution not applicable and the problem more severe… A recommendation could be to buy or develop open-source cameras or at least cameras that have been evaluated by others and for which correction algorithms exist.

Citation: https://doi.org/10.5194/egusphere-2025-1601-RC3
- AC3: 'Reply on RC3', Julia Martin, 01 Aug 2025
  
  Author response in PDF (03_AC_GP_review.pdf).
  
  Citation: https://doi.org/10.5194/egusphere-2025-1601-AC3

Julia Martin, Ruzica Dadic, Brian Anderson, Roberta Pirazzini, Oliver Wigmore, and Lauren Vargo

Viewed

Total article views: 887 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
769	92	26	887	11	26

HTML: 769
PDF: 92
XML: 26
Total: 887
BibTeX: 11
EndNote: 26

Views and downloads (calculated since 14 May 2025)

Month	HTML	PDF	XML	Total
May 2025	132	35	6	173
Jun 2025	62	16	7	85
Jul 2025	66	13	6	85
Aug 2025	150	14	6	170
Sep 2025	348	12	1	361
Oct 2025	11	2	0	13

Cumulative views and downloads (calculated since 14 May 2025)

Month	HTML	PDF	XML	Total
May 2025	132	35	6	173
Jun 2025	62	16	7	85
Jul 2025	66	13	6	85
Aug 2025	150	14	6	170
Sep 2025	348	12	1	361
Oct 2025	11	2	0	13

Viewed (geographical distribution)

Total article views: 868 (including HTML, PDF, and XML) Thereof 868 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 07 Oct 2025

Short summary

This study examines how snow distribution affects Antarctic sea ice surface temperature, a key factor in its energy balance. Using drone and ground-based data, we mapped snow depth and surface temperature on 2.4 m thick sea ice in McMurdo Sound. We corrected thermal camera inconsistencies and found that surface temperature is more influenced by topography-driven solar radiation than snow depth. Our findings highlight the need to account for small-scale processes in sea ice energy balance models.


Total:	0
HTML:	0
PDF:	0
XML:	0