Impact of the sampling procedure on the specific surface area of snow measurements with the IceCube

Martin, Julia; Schneebeli, Martin

doi:https://doi.org/10.5194/egusphere-2022-501

Preprints

https://doi.org/10.5194/egusphere-2022-501

Preprints

27 Jun 2022

| 27 Jun 2022

Impact of the sampling procedure on the specific surface area of snow measurements with the IceCube

Julia Martin and Martin Schneebeli

Abstract. The specific surface area (SSA) of snow is directly measured by X-ray computed tomography or indirectly using the reflectance of near-infrared light. The IceCube is a well-established spectroscopic instrument using a near-infrared wavelength of 1310 nm. We compared the SSA of six snow types measured with both instruments. The IceCube measured significantly higher values with a relative percentage difference between 20 to 52 % for snow types with an SSA between 5 to 25 m² kg⁻¹. There is no significant difference for snow with an SSA between 30 to 80 m² kg⁻¹. The difference is statistically significant between snow types but not uniquely related to the SSA. We suspected that artificially created particles were the source of the difference. These were sampled, measured and counted. Numerical simulations with radiation transfer solver TARTES confirm the observation.

Received: 16 Jun 2022 – Discussion started: 27 Jun 2022

Julia Martin and Martin Schneebeli

Status: closed

RC1:
'Comment on egusphere-2022-501', Anonymous Referee #1, 18 Jul 2022
Paper # https://doi.org/10.5194/egusphere-2022-501

Impact of the sampling procedure on the specific surface area of snow measurements with the IceCube

This manuscript is potentially valuable as it contains an important headline result – that the manner in which snow is sampled impacts up to approximately 50% the SSA values less than 30 m² kg^-1 from the IceCube (IC) instrument in comparison to micro-CT (CT). Potentially, this has big implications for 1310 nm snow reflectance measurements as the IceCube (n.b. also potentially DUFISSS in pre-production, or IRIS in non-commercial form), is an increasingly common and robust in-field instrument for objective measurements of snow microstructure. Micro-CT, a ‘gold standard’, is as good a direct measure of snow microstructure as we currently have. Behind this headline result, there are a number of issues that need to be addressed for the community to have confidence in the currently proposed message. It may mean the scope of the message needs to be refined with greater detail, and the implications limited to particular snow types.

The 5-25 m² kg^-1 range of snow SSA that shows significant different between IC and CT are often associated with important snow types, e.g. depth hoar or wind slab in Arctic and sub-Arctic snowpacks, that are not part of the experiment. At best this paper needs to be explicitly limited to Alpine snow, otherwise unintended mis-interpretation could occur. Could more detail be provided to describe the Alpine snow types measured, e.g. densities from volumetric sampling? More details of the snowpack from which the samples were extracted would be highly beneficial so the reader can get a feel for snow types which the interpretation is both relevant for and limited to.

The sample preparation process is a key conclusion explaining the difference between IC and CT. However, there is ambiguity in the description of this method. It seems the sample was reduced to size through cutting of unsuitable material and then brushed gently to remove loose particles and measured by IC. Secondly, the ‘default method following Gallet et al. (2009)’ was followed, then any remaining loose particles were knocked off, then the sample was remeasured using IC. The default method in section 2 of Gallet et al. (2009) refers to the sample measurement face being shaved off with a spatula, in which they state it was difficult to obtain a ‘perfect surface’. Hence more needs to be included about the shaving process and how it was applied in this experiment. I got the impression that from Gallet et al. (2009) the shaving/smearing of the surface grains by the spatula (especially when close to freezing) could have had an impact on surface optical reflectance. I expect this not to be the case in cold labs at -15 degrees Celsius, but it requires a more detailed discussion about how preparation of IC sample surfaces effect SSA. Discussions at the Davos Grain Size Measurement workshop in 2014 and my own experience of making IC measurements suggest that the SSA from IC is (thankfully) not very sensitive to sample preparation. The pressure required to cause sintering as part of the sampling process is highly unlikely to be achieved. Rather, making sure the sample container is completely full by addition of snow to fill any gaps in the extracted sample, and light compaction of snow to be flush with the container surface is preferable so that reflectance is less likely to come from the edges of the sample container. This negligible impact of sample preparation appears to be shown in the comparison of distributions of IC + particles and IC – particles in Figure 2, where distributions overlap. As both field experience and results in Figure 2 contradict the message that sample surface preparation is crucial, this message needs to be revisited.

Serious consideration needs to be made as to whether relative percentage difference is a fair way to present the results, particularly when the mean or median values range from <10 to >50 m² kg^-1. I suggest presenting the measurement uncertainty in m² kg^-1 is more appropriate, e.g. a bias or RMSE. This is illustrated by Figure 2, where the actual difference between extents of upper and lower quartiles between CT out of IC / CT reference and IC respectively, either overlap for type C and E, or are approximately 2-3 m² kg^-1 apart for type A and D. And when the four distributions of CT are considered against the two IC distributions, overlap of distributions is more common than not. Some discussion about what level of natural SSA variability might be expected within a sample (CT or IC) needs to be added here. Depending on the orientation of the sample in IC measurement I would expect variability in spectral reflectance, particularly in snow types that are not highly homogenous in structure, size and orientation. Hence SSA variability of the order 2-3 m² kg^-1 may well be within measurement noise. As an exemplar, Fig 2 shows that differences between distributions of CT surface and CT mid are on the same order of similarity to the difference between IC + particles and CT out of IC. While I expect CT mid to be the best measurement to compare other measurements to, the fact that there is such spatial variability within a CT sample, suggests that the comparison between CT and IC is not drastically worse than the within CT measurements. Can this be discussed in further detail as it appears to add sensible uncertainty caveats to one of the headline conclusions, which is there is a SSA difference of 20-52% in the 5-25 m² kg^-1 range when measured by IC and CT.

The visual and statistical comparison of distributions (Fig 1 and 2) is good, but this raises a concern at the low number of sample values (Table 1 shows n=1-8) which make up these distributions. I appreciate the time required to make CT measurements, so this not being a high n-value is understandable. However, how was the n value calculated for IC? Was it a single sample measured in different orientations? Or were there a number of samples in the same snow layer? Considering that the IC is designed for field use and implications of results increasingly tend to be considered in recent literature when using larger distributions (n >10) of measurements in similar layers, these are very low sample numbers to be making robust conclusions. However, there is a balance to be struck here. These initial results are useful for the community to see, but I think that it points the way for future work, rather than being definitive about the applicability of IC measurements at SSA <30 m² kg^-1. It might be a suggestion for a revised version of this manuscript to be a brief communication rather than a full article in TC?

The application of TARTES to provide an explanation of potential impact of the surface grain size on optical reflectance is good. However, as a non-expert, does the choice of a 1m thick snow substrate matter, rather than a sample thickness of 300 mm which is a common depth of the IC sample container?

What is nature-identical new snow (largely >30 m² kg^-1) in the context it is presented? Figure 1 looks like the distribution of IC + particles (more realistically how samples would be measured in the field) are very similar to the CT distribution, so we need more detail on what is ‘nature-identical’. It suggests the rest of the samples are dissimilar to nature, which is a worry when drawing implications from this experiment. This may just be a terminology issue, but it needs to be addressed.

Rather than a list of minor comments, at this stage I would encourage the authors to:

Expand on the introduction to contrast IC and CT to a broader range of microstructural measurements and implications for their use.

Increase the clarity of the ‘hypothesis statement’ at the end of the introduction (i.e. this paper does, this, and this, and this…)

Add clarity on what is being presented in the box plots (are they median and IQR, or mean and standard deviation - as in Table 1).

Add more details throughout (as per issues raised above), and in the discussion section add more on the implications of these results for a) field measurement using IC, and b) what using measurements from IC may mean for applications where SSA is crucial.

Check the cross references are correct (e.g. ‘Tab 3’ is cross-referenced, but is not in the main manuscript).
Citation: https://doi.org/10.5194/egusphere-2022-501-RC1
- AC1: 'Reply on RC1', Julia Martin, 08 Aug 2022
  
  We very much appreciate the time and effort of the first reviewer and would like to address the significant points mentioned by the reviewer.
  We agree about the limitation of the snow type. Our study used only alpine snow samples and nature-identical snow samples. We will include measurements (IC and micro-CT) from a field campaign in Greenland to improve the data set. Furthermore, we also agree with the benefits of a more detailed description of the snowpack we extracted the samples and provide additional snow physical properties (i.e., density measured with micro-CT).
  To clarify the sampling procedure, we will provide a step-by-step illustration and elaborate on the number of samples used for the different sampling steps in more detail.
  We will also add the bias and RMSE for the result presentation as suggested by the reviewer and expand the discussion towards the mentioned points.
  Indeed we missed the reference to the production and properties of nature-identical snow. We will add the reference and the conditions of production.
  We agree with the list of changes at the end of the review and will adapt the manuscript accordingly.
  Nevertheless, we would like to keep the format as a full article, as a Brief communication does not provide the scope essential to incorporate the required changes.
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-AC1
RC2:
'Comment on egusphere-2022-501', Anonymous Referee #2, 27 Jul 2022

The paper is concise and relatively clear though the format is a bit disorganized and the terminology is not always consistent. The figures positively contributed to the representation of the data and methodology though more specific figures related to sample preparation would improve the explanation of this process.

The experimental design was thoughtful and could be impactful to the field. Refining our understanding of the caveats associated with field instruments will help improve future measurements and use. The paper would benefit if the authors add more information about sample preparation (with additional visual aids) and context for the snow samples used in the experiment. It would also be beneficial to describe the micro-CT methods and analysis, including any differences associated with two nominal scan resolutions (15 &18um). Could the authors also better describe the "manufactured micro-CT sampling kit" and how it is used?

Generally speaking, it would be helpful to have the results presented in the same order as the data, or it would help to introduce and explain the format in which the data will be presented. Figures would be more accessible if they were adjacent to where they are references in the text and the figure captions could be improved if more information is added to them and if they are written in complete sentences.

Line 6: Please describe what these "artificially created particles" are.

Line 20: I recommend re-writing the sentence "We focus on.." to improve clarity.

Line 22: Is it necessary to mention black carbon at all? It is the only time it is mentioned in the entire paper.

Line 39: Recommended change "Next, a second SSA is measured with ..."

Line 46-48: Can you comment on any variability associated with different micro-CT scan resolutions?

Line 54: I recommend rewording the sentence "Eight pictures..." for improved clarity.

Line 68: I recommend rewording the sentence "As the desired output variable.." for improved clarity.

Line 71: In the results section, Table 3 is both spelled out and abbreviated as "Tab.3".

Line 79: Recommend changing sentence to " However, the SSA measured by the micro-CT is 24% smaller than that measured by the IC."

Figure 1 caption: The figure is described as having a "left, middle, and right". Do you mean top, middle, and bottom? It might be better to refer to "panels (a), (b), and (c)" instead.

Figure 3: Please describe the manufactured micro-CT sampling kit. There is no reference to the (b) and (c) labels on this kit.

Citation: https://doi.org/10.5194/egusphere-2022-501-RC2
- AC2: 'Reply on RC2', Julia Martin, 08 Aug 2022
  
  We are thankful for the time and effort of the second anonymous reviewer and would like to address the points mentioned by the reviewer.
  We will add a detailed step-by-step illustration of the sampling process to provide sufficient information for the reader to reproduce the procedure.
  Furthermore, we realized that the provided information about the micro-CT sampling kit is insufficient for someone who does not use the micro-CT. We will provide more details on the sampling procedure with the micro-CT sampling kit and how we used it. We will adapt the figure captions accordingly.
  Also, we will restructure the manuscript to improve the flow of reading.
  We will adapt the manuscript according to the list of changes suggested at the end of the review.
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-AC2
CC1:
'Comment on egusphere-2022-501', Florent Dominé, 10 Aug 2022

First of all, I would like to thank Julia Martin and Martin Schneebeli for this interesting contribution. Measuring SSA can be very delicate. Errors, mostly due to an inadequate sampling procedure, are common. A detailed study on one such error type is therefore very welcome. I also thank both anonymous Reviewers for contributing to the discussion and therefore help improve sampling protocols.
Thousands of SSA measurements were done with DUFISSS and Ice Cube since (Gallet et al., 2009) was written. I have taken many new Ice Cube users to the field for training to realize that the sampling procedure described in that paper and in the Ice Cube sampling manual does not describe in sufficient detail many common potential sources of errors. I often thought of writing a detailed paper but this has unfortunately remained wishful thinking, my apologies.
One error is indeed the production of small particles when cutting the sample surface. As (Martin and Schneebeli, 2022) nicely describe, the magnitude of the error depends essentially on the hardness of the snow, not on the SSA.
The usual sampling procedure is to place the sampler vertically on the snow surface of interest and to sample downward. The 35 mm-thick sample is then placed in the 25 mm-thick sample-holder so that there is 10 mm of snow protruding above the sample holder. This is shaved off with a spatula. If the snow is sufficiently hard not to fall apart, I recommend shaving off the extra 10 mm of snow by placing the snow surface vertically. The particles formed then mostly fall out rather than inside the sample. Then, a soft brush is used to remove most remaining particles. I have done many comparisons of both protocols (1) shaving off horizontally and no brushing, and (2) shaving off vertically with brushing. For hard snows, I found a difference similar to the data of (Martin and Schneebeli, 2022). The largest difference was for about 12 m deep firn in Greenland. Values with vertical shaving off and brushing were around 5 m² kg^-1 while without these precautions, SSA was almost doubled because of the effort required to cut the sample. I do not want to interfere with the writing options of the Authors, but I suggest (Martin and Schneebeli, 2022), if possible, stress this protocol for hard snows.
Another frequent errors I have seen is measuring soft snow of too low density. Then the optical depth is insufficient as detailed in (Gallet et al., 2009). The 1310 radiation reaches the bottom of the sample holder where it is absorbed. This reduces reflection and inferred SSA. To overcome this problem, DUFISSS uses the 1550 nm radiation, where ice is more absorbent and the e-folding depth therefore shallower. Ice Cube however does not have the 1550 nm option. The alternative is to compact the snow to about 200 kg m^-3 to decrease the e-folding depth. Tests showed that for soft snows, compaction does not affect structure and SSA in a detectable manner.
Lastly, it may be useful to mention difficulties in measuring large soft loose depth hoar crystals, as frequently encountered in the Arctic. Such snow samples cannot be shaved off and can be scooped directly onto the sample holder. The critical part is getting a surface at the correct level. If crystals stick out of the surface of the sample holder, reflection is enhanced, while the reverse is observed if the snow crystal level is too low. This can lead to SSA variations exceeding 20%. Great care is required to ensure that the average crystal level is flush with the top of the sample holder.
Coming back to the paper of (Martin and Schneebeli, 2022), it may be worthwhile specifying that reflectance measurements at 950 nm are less sensitive to a layer of fine particles than at 1310 nm because the e-folding depth is about twice as much at 950 than at 1310 nm. At 950 nm, the thin surface layer then contributes less to the signal. However, precision is lower at 950 nm, as detailed in Figure 1 of (Gallet et al., 2009). This was one reason for choosing 1310 nm.
References
Gallet, J.-C., Domine, F., Zender, C. S., and Picard, G.: Measurement of the specific surface area of snow using infrared reflectance in an integrating sphere at 1310 and 1550 nm, The Cryosphere, 3, 167-182, https://doi.org/10.5194/tc-3-167-2009, 2009.
Martin, J., and Schneebeli, M.: Impact of the sampling procedure on the specific surface area of snow measurements with the IceCube, EGUsphere, 2022, 1-13, 10.5194/egusphere-2022-501, 2022.

Citation: https://doi.org/10.5194/egusphere-2022-501-CC1
- AC3: 'Reply on CC1', Julia Martin, 23 Aug 2022
  
  Dear Florent,
  we are thankful for the time and effort you took to review our paper.
  We appreciate the share of your broad experience with the IceCube and will try to incorporate your valuable points of discussion (i.e., brushing the sample surface, and sample preparation).
  Furthermore, as mentioned in an earlier reply we will include a data set with snow from Greenland to present a higher variation of snow types to make our study more robust.
  Also, we will extend the discussion about the NIR wavelength.
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-AC3
CC2:
'Comment on egusphere-2022-501', Julien Meloche, 22 Aug 2022

I would to thank both authors for their work.
I'm affraid my comment is mostly anecdotal (but with some data!) and relate well to this work.
At the University of Sherbrooke, with Prof. Alain Royer and Prof. Alexandre Langlois, we work with our own version (IRIS) of IceCube as you probably know. The instrument is really similar to IceCube with some differences in the design and sampling procedure.
In a recent publication (Meloche et al., 2022) with arctic snow type, a dataset of snow pit mesurements from two different regions in the canadian arctic was presented; Trail Valley Creek (TVC) with SSA derived from IceCube and Cambridge Bay (CB) with SSA from IRIS. Figure 5 a) presented the SSA of each site and year. It can be noted that SSA from TVC is larger for all years and layer types (wind slab and depth hoar). Based on this, it seems that IRIS underestimate the SSA compare to IceCube since IRIS was used in CB and IceCube in TVC. I originally thought this was due to a difference in the location of the observations as TVC can be caracterized with sub arctic vegetation. However, I'm beginning to think that based on this paper, the difference is probably due to the instruments and the sampling procedures.
Since both instruments are used extensively in the field, I think the snow community would benefit from analysis like this on error estimate of SSA cause by the sampling procedure.

Citation: https://doi.org/10.5194/egusphere-2022-501-CC2
- CC3: 'Reply on CC2', Julien Meloche, 22 Aug 2022
  
  Reference for CC2
  
  Meloche, J., Langlois, A., Rutter, N., Royer, A., King, J., Walker, B., Marsh, P., and Wilcox, E. J.: Characterizing tundra snow sub-pixel variability to improve brightness temperature estimation in satellite SWE retrievals, The Cryosphere, 16, 87–101, https://doi.org/10.5194/tc-16-87-2022, 2022.
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-CC3
- AC4: 'Reply on CC2', Julia Martin, 23 Aug 2022
  
  Dear Julien,
  thank you for your insights and descriptions. We will consider including an outlook on how to best handle the source of error we found in our study. In any way, we think it would be highly advisable to make comprehensive tests based on micro-CT imaging before deploying an SSA instrument (if resources in that matter are available).
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-AC4

Status: closed

RC1:
'Comment on egusphere-2022-501', Anonymous Referee #1, 18 Jul 2022
Paper # https://doi.org/10.5194/egusphere-2022-501

Impact of the sampling procedure on the specific surface area of snow measurements with the IceCube

This manuscript is potentially valuable as it contains an important headline result – that the manner in which snow is sampled impacts up to approximately 50% the SSA values less than 30 m² kg^-1 from the IceCube (IC) instrument in comparison to micro-CT (CT). Potentially, this has big implications for 1310 nm snow reflectance measurements as the IceCube (n.b. also potentially DUFISSS in pre-production, or IRIS in non-commercial form), is an increasingly common and robust in-field instrument for objective measurements of snow microstructure. Micro-CT, a ‘gold standard’, is as good a direct measure of snow microstructure as we currently have. Behind this headline result, there are a number of issues that need to be addressed for the community to have confidence in the currently proposed message. It may mean the scope of the message needs to be refined with greater detail, and the implications limited to particular snow types.

The 5-25 m² kg^-1 range of snow SSA that shows significant different between IC and CT are often associated with important snow types, e.g. depth hoar or wind slab in Arctic and sub-Arctic snowpacks, that are not part of the experiment. At best this paper needs to be explicitly limited to Alpine snow, otherwise unintended mis-interpretation could occur. Could more detail be provided to describe the Alpine snow types measured, e.g. densities from volumetric sampling? More details of the snowpack from which the samples were extracted would be highly beneficial so the reader can get a feel for snow types which the interpretation is both relevant for and limited to.

The sample preparation process is a key conclusion explaining the difference between IC and CT. However, there is ambiguity in the description of this method. It seems the sample was reduced to size through cutting of unsuitable material and then brushed gently to remove loose particles and measured by IC. Secondly, the ‘default method following Gallet et al. (2009)’ was followed, then any remaining loose particles were knocked off, then the sample was remeasured using IC. The default method in section 2 of Gallet et al. (2009) refers to the sample measurement face being shaved off with a spatula, in which they state it was difficult to obtain a ‘perfect surface’. Hence more needs to be included about the shaving process and how it was applied in this experiment. I got the impression that from Gallet et al. (2009) the shaving/smearing of the surface grains by the spatula (especially when close to freezing) could have had an impact on surface optical reflectance. I expect this not to be the case in cold labs at -15 degrees Celsius, but it requires a more detailed discussion about how preparation of IC sample surfaces effect SSA. Discussions at the Davos Grain Size Measurement workshop in 2014 and my own experience of making IC measurements suggest that the SSA from IC is (thankfully) not very sensitive to sample preparation. The pressure required to cause sintering as part of the sampling process is highly unlikely to be achieved. Rather, making sure the sample container is completely full by addition of snow to fill any gaps in the extracted sample, and light compaction of snow to be flush with the container surface is preferable so that reflectance is less likely to come from the edges of the sample container. This negligible impact of sample preparation appears to be shown in the comparison of distributions of IC + particles and IC – particles in Figure 2, where distributions overlap. As both field experience and results in Figure 2 contradict the message that sample surface preparation is crucial, this message needs to be revisited.

Serious consideration needs to be made as to whether relative percentage difference is a fair way to present the results, particularly when the mean or median values range from <10 to >50 m² kg^-1. I suggest presenting the measurement uncertainty in m² kg^-1 is more appropriate, e.g. a bias or RMSE. This is illustrated by Figure 2, where the actual difference between extents of upper and lower quartiles between CT out of IC / CT reference and IC respectively, either overlap for type C and E, or are approximately 2-3 m² kg^-1 apart for type A and D. And when the four distributions of CT are considered against the two IC distributions, overlap of distributions is more common than not. Some discussion about what level of natural SSA variability might be expected within a sample (CT or IC) needs to be added here. Depending on the orientation of the sample in IC measurement I would expect variability in spectral reflectance, particularly in snow types that are not highly homogenous in structure, size and orientation. Hence SSA variability of the order 2-3 m² kg^-1 may well be within measurement noise. As an exemplar, Fig 2 shows that differences between distributions of CT surface and CT mid are on the same order of similarity to the difference between IC + particles and CT out of IC. While I expect CT mid to be the best measurement to compare other measurements to, the fact that there is such spatial variability within a CT sample, suggests that the comparison between CT and IC is not drastically worse than the within CT measurements. Can this be discussed in further detail as it appears to add sensible uncertainty caveats to one of the headline conclusions, which is there is a SSA difference of 20-52% in the 5-25 m² kg^-1 range when measured by IC and CT.

The visual and statistical comparison of distributions (Fig 1 and 2) is good, but this raises a concern at the low number of sample values (Table 1 shows n=1-8) which make up these distributions. I appreciate the time required to make CT measurements, so this not being a high n-value is understandable. However, how was the n value calculated for IC? Was it a single sample measured in different orientations? Or were there a number of samples in the same snow layer? Considering that the IC is designed for field use and implications of results increasingly tend to be considered in recent literature when using larger distributions (n >10) of measurements in similar layers, these are very low sample numbers to be making robust conclusions. However, there is a balance to be struck here. These initial results are useful for the community to see, but I think that it points the way for future work, rather than being definitive about the applicability of IC measurements at SSA <30 m² kg^-1. It might be a suggestion for a revised version of this manuscript to be a brief communication rather than a full article in TC?

The application of TARTES to provide an explanation of potential impact of the surface grain size on optical reflectance is good. However, as a non-expert, does the choice of a 1m thick snow substrate matter, rather than a sample thickness of 300 mm which is a common depth of the IC sample container?

What is nature-identical new snow (largely >30 m² kg^-1) in the context it is presented? Figure 1 looks like the distribution of IC + particles (more realistically how samples would be measured in the field) are very similar to the CT distribution, so we need more detail on what is ‘nature-identical’. It suggests the rest of the samples are dissimilar to nature, which is a worry when drawing implications from this experiment. This may just be a terminology issue, but it needs to be addressed.

Rather than a list of minor comments, at this stage I would encourage the authors to:

Expand on the introduction to contrast IC and CT to a broader range of microstructural measurements and implications for their use.

Increase the clarity of the ‘hypothesis statement’ at the end of the introduction (i.e. this paper does, this, and this, and this…)

Add clarity on what is being presented in the box plots (are they median and IQR, or mean and standard deviation - as in Table 1).

Add more details throughout (as per issues raised above), and in the discussion section add more on the implications of these results for a) field measurement using IC, and b) what using measurements from IC may mean for applications where SSA is crucial.

Check the cross references are correct (e.g. ‘Tab 3’ is cross-referenced, but is not in the main manuscript).
Citation: https://doi.org/10.5194/egusphere-2022-501-RC1
- AC1: 'Reply on RC1', Julia Martin, 08 Aug 2022
  
  We very much appreciate the time and effort of the first reviewer and would like to address the significant points mentioned by the reviewer.
  We agree about the limitation of the snow type. Our study used only alpine snow samples and nature-identical snow samples. We will include measurements (IC and micro-CT) from a field campaign in Greenland to improve the data set. Furthermore, we also agree with the benefits of a more detailed description of the snowpack we extracted the samples and provide additional snow physical properties (i.e., density measured with micro-CT).
  To clarify the sampling procedure, we will provide a step-by-step illustration and elaborate on the number of samples used for the different sampling steps in more detail.
  We will also add the bias and RMSE for the result presentation as suggested by the reviewer and expand the discussion towards the mentioned points.
  Indeed we missed the reference to the production and properties of nature-identical snow. We will add the reference and the conditions of production.
  We agree with the list of changes at the end of the review and will adapt the manuscript accordingly.
  Nevertheless, we would like to keep the format as a full article, as a Brief communication does not provide the scope essential to incorporate the required changes.
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-AC1
RC2:
'Comment on egusphere-2022-501', Anonymous Referee #2, 27 Jul 2022

The paper is concise and relatively clear though the format is a bit disorganized and the terminology is not always consistent. The figures positively contributed to the representation of the data and methodology though more specific figures related to sample preparation would improve the explanation of this process.

The experimental design was thoughtful and could be impactful to the field. Refining our understanding of the caveats associated with field instruments will help improve future measurements and use. The paper would benefit if the authors add more information about sample preparation (with additional visual aids) and context for the snow samples used in the experiment. It would also be beneficial to describe the micro-CT methods and analysis, including any differences associated with two nominal scan resolutions (15 &18um). Could the authors also better describe the "manufactured micro-CT sampling kit" and how it is used?

Generally speaking, it would be helpful to have the results presented in the same order as the data, or it would help to introduce and explain the format in which the data will be presented. Figures would be more accessible if they were adjacent to where they are references in the text and the figure captions could be improved if more information is added to them and if they are written in complete sentences.

Line 6: Please describe what these "artificially created particles" are.

Line 20: I recommend re-writing the sentence "We focus on.." to improve clarity.

Line 22: Is it necessary to mention black carbon at all? It is the only time it is mentioned in the entire paper.

Line 39: Recommended change "Next, a second SSA is measured with ..."

Line 46-48: Can you comment on any variability associated with different micro-CT scan resolutions?

Line 54: I recommend rewording the sentence "Eight pictures..." for improved clarity.

Line 68: I recommend rewording the sentence "As the desired output variable.." for improved clarity.

Line 71: In the results section, Table 3 is both spelled out and abbreviated as "Tab.3".

Line 79: Recommend changing sentence to " However, the SSA measured by the micro-CT is 24% smaller than that measured by the IC."

Figure 1 caption: The figure is described as having a "left, middle, and right". Do you mean top, middle, and bottom? It might be better to refer to "panels (a), (b), and (c)" instead.

Figure 3: Please describe the manufactured micro-CT sampling kit. There is no reference to the (b) and (c) labels on this kit.

Citation: https://doi.org/10.5194/egusphere-2022-501-RC2
- AC2: 'Reply on RC2', Julia Martin, 08 Aug 2022
  
  We are thankful for the time and effort of the second anonymous reviewer and would like to address the points mentioned by the reviewer.
  We will add a detailed step-by-step illustration of the sampling process to provide sufficient information for the reader to reproduce the procedure.
  Furthermore, we realized that the provided information about the micro-CT sampling kit is insufficient for someone who does not use the micro-CT. We will provide more details on the sampling procedure with the micro-CT sampling kit and how we used it. We will adapt the figure captions accordingly.
  Also, we will restructure the manuscript to improve the flow of reading.
  We will adapt the manuscript according to the list of changes suggested at the end of the review.
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-AC2
CC1:
'Comment on egusphere-2022-501', Florent Dominé, 10 Aug 2022

First of all, I would like to thank Julia Martin and Martin Schneebeli for this interesting contribution. Measuring SSA can be very delicate. Errors, mostly due to an inadequate sampling procedure, are common. A detailed study on one such error type is therefore very welcome. I also thank both anonymous Reviewers for contributing to the discussion and therefore help improve sampling protocols.
Thousands of SSA measurements were done with DUFISSS and Ice Cube since (Gallet et al., 2009) was written. I have taken many new Ice Cube users to the field for training to realize that the sampling procedure described in that paper and in the Ice Cube sampling manual does not describe in sufficient detail many common potential sources of errors. I often thought of writing a detailed paper but this has unfortunately remained wishful thinking, my apologies.
One error is indeed the production of small particles when cutting the sample surface. As (Martin and Schneebeli, 2022) nicely describe, the magnitude of the error depends essentially on the hardness of the snow, not on the SSA.
The usual sampling procedure is to place the sampler vertically on the snow surface of interest and to sample downward. The 35 mm-thick sample is then placed in the 25 mm-thick sample-holder so that there is 10 mm of snow protruding above the sample holder. This is shaved off with a spatula. If the snow is sufficiently hard not to fall apart, I recommend shaving off the extra 10 mm of snow by placing the snow surface vertically. The particles formed then mostly fall out rather than inside the sample. Then, a soft brush is used to remove most remaining particles. I have done many comparisons of both protocols (1) shaving off horizontally and no brushing, and (2) shaving off vertically with brushing. For hard snows, I found a difference similar to the data of (Martin and Schneebeli, 2022). The largest difference was for about 12 m deep firn in Greenland. Values with vertical shaving off and brushing were around 5 m² kg^-1 while without these precautions, SSA was almost doubled because of the effort required to cut the sample. I do not want to interfere with the writing options of the Authors, but I suggest (Martin and Schneebeli, 2022), if possible, stress this protocol for hard snows.
Another frequent errors I have seen is measuring soft snow of too low density. Then the optical depth is insufficient as detailed in (Gallet et al., 2009). The 1310 radiation reaches the bottom of the sample holder where it is absorbed. This reduces reflection and inferred SSA. To overcome this problem, DUFISSS uses the 1550 nm radiation, where ice is more absorbent and the e-folding depth therefore shallower. Ice Cube however does not have the 1550 nm option. The alternative is to compact the snow to about 200 kg m^-3 to decrease the e-folding depth. Tests showed that for soft snows, compaction does not affect structure and SSA in a detectable manner.
Lastly, it may be useful to mention difficulties in measuring large soft loose depth hoar crystals, as frequently encountered in the Arctic. Such snow samples cannot be shaved off and can be scooped directly onto the sample holder. The critical part is getting a surface at the correct level. If crystals stick out of the surface of the sample holder, reflection is enhanced, while the reverse is observed if the snow crystal level is too low. This can lead to SSA variations exceeding 20%. Great care is required to ensure that the average crystal level is flush with the top of the sample holder.
Coming back to the paper of (Martin and Schneebeli, 2022), it may be worthwhile specifying that reflectance measurements at 950 nm are less sensitive to a layer of fine particles than at 1310 nm because the e-folding depth is about twice as much at 950 than at 1310 nm. At 950 nm, the thin surface layer then contributes less to the signal. However, precision is lower at 950 nm, as detailed in Figure 1 of (Gallet et al., 2009). This was one reason for choosing 1310 nm.
References
Gallet, J.-C., Domine, F., Zender, C. S., and Picard, G.: Measurement of the specific surface area of snow using infrared reflectance in an integrating sphere at 1310 and 1550 nm, The Cryosphere, 3, 167-182, https://doi.org/10.5194/tc-3-167-2009, 2009.
Martin, J., and Schneebeli, M.: Impact of the sampling procedure on the specific surface area of snow measurements with the IceCube, EGUsphere, 2022, 1-13, 10.5194/egusphere-2022-501, 2022.

Citation: https://doi.org/10.5194/egusphere-2022-501-CC1
- AC3: 'Reply on CC1', Julia Martin, 23 Aug 2022
  
  Dear Florent,
  we are thankful for the time and effort you took to review our paper.
  We appreciate the share of your broad experience with the IceCube and will try to incorporate your valuable points of discussion (i.e., brushing the sample surface, and sample preparation).
  Furthermore, as mentioned in an earlier reply we will include a data set with snow from Greenland to present a higher variation of snow types to make our study more robust.
  Also, we will extend the discussion about the NIR wavelength.
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-AC3
CC2:
'Comment on egusphere-2022-501', Julien Meloche, 22 Aug 2022

I would to thank both authors for their work.
I'm affraid my comment is mostly anecdotal (but with some data!) and relate well to this work.
At the University of Sherbrooke, with Prof. Alain Royer and Prof. Alexandre Langlois, we work with our own version (IRIS) of IceCube as you probably know. The instrument is really similar to IceCube with some differences in the design and sampling procedure.
In a recent publication (Meloche et al., 2022) with arctic snow type, a dataset of snow pit mesurements from two different regions in the canadian arctic was presented; Trail Valley Creek (TVC) with SSA derived from IceCube and Cambridge Bay (CB) with SSA from IRIS. Figure 5 a) presented the SSA of each site and year. It can be noted that SSA from TVC is larger for all years and layer types (wind slab and depth hoar). Based on this, it seems that IRIS underestimate the SSA compare to IceCube since IRIS was used in CB and IceCube in TVC. I originally thought this was due to a difference in the location of the observations as TVC can be caracterized with sub arctic vegetation. However, I'm beginning to think that based on this paper, the difference is probably due to the instruments and the sampling procedures.
Since both instruments are used extensively in the field, I think the snow community would benefit from analysis like this on error estimate of SSA cause by the sampling procedure.

Citation: https://doi.org/10.5194/egusphere-2022-501-CC2
- CC3: 'Reply on CC2', Julien Meloche, 22 Aug 2022
  
  Reference for CC2
  
  Meloche, J., Langlois, A., Rutter, N., Royer, A., King, J., Walker, B., Marsh, P., and Wilcox, E. J.: Characterizing tundra snow sub-pixel variability to improve brightness temperature estimation in satellite SWE retrievals, The Cryosphere, 16, 87–101, https://doi.org/10.5194/tc-16-87-2022, 2022.
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-CC3
- AC4: 'Reply on CC2', Julia Martin, 23 Aug 2022
  
  Dear Julien,
  thank you for your insights and descriptions. We will consider including an outlook on how to best handle the source of error we found in our study. In any way, we think it would be highly advisable to make comprehensive tests based on micro-CT imaging before deploying an SSA instrument (if resources in that matter are available).
  
  Citation: https://doi.org/10.5194/egusphere-2022-501-AC4

Julia Martin and Martin Schneebeli

Data sets

IceCube_microCT_Snow_grainsize Martin, Julia; Schneebeli, Martin https://doi.org/10.16904/envidat.333

Model code and software

IceCube_microCT_Snow_grainsize Martin, Julia; Schneebeli, Martin https://doi.org/10.16904/envidat.333

Julia Martin and Martin Schneebeli

Viewed

Total article views: 630 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
435	168	27	630	8	9

HTML: 435
PDF: 168
XML: 27
Total: 630
BibTeX: 8
EndNote: 9

Views and downloads (calculated since 27 Jun 2022)

Month	HTML	PDF	XML	Total
Jun 2022	89	29	5	123
Jul 2022	86	31	5	122
Aug 2022	95	39	14	148
Sep 2022	26	17	0	43
Oct 2022	20	8	0	28
Nov 2022	24	4	1	29
Dec 2022	21	6	0	27
Jan 2023	27	7	0	34
Feb 2023	32	23	0	55
Mar 2023	15	4	2	21

Cumulative views and downloads (calculated since 27 Jun 2022)

Month	HTML	PDF	XML	Total
Jun 2022	89	29	5	123
Jul 2022	86	31	5	122
Aug 2022	95	39	14	148
Sep 2022	26	17	0	43
Oct 2022	20	8	0	28
Nov 2022	24	4	1	29
Dec 2022	21	6	0	27
Jan 2023	27	7	0	34
Feb 2023	32	23	0	55
Mar 2023	15	4	2	21

Viewed (geographical distribution)

Total article views: 537 (including HTML, PDF, and XML) Thereof 537 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 21 Mar 2023

Short summary

The grain size of snow determines how light is reflected and other physical properties. The IceCube measures snow grain size at the specific near-infrared wavelength of 1320 nm. In our study the preparation of snow samples for the IceCube creates a thin layer of small particles. Comparisons of the grain size with computed tomography, particle counting and numerical simulation confirm the observation. We conclude that snow grain size measurements at this wavelength underestimate the grain size.


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