Research article 06 Jan 2022
Research article | 06 Jan 2022
Mapping liquid water content in snow at the millimeter scale: an intercomparison of mixed-phase optical property models using hyperspectral imaging and in situ measurements
Christopher Donahue et al.
Download
- Final revised paper (published on 06 Jan 2022)
- Preprint (discussion started on 12 Aug 2021)
Interactive discussion
Status: closed
-
RC1: 'Comment on tc-2021-247', Ryan Webb, 26 Aug 2021
This study uses NIR spectral reflectance measurements to estimate the LWC of snow, and compare those estimates to an independant dielectric measurement. This investigation conducts laboratory scale tests to determine which reflectance model is most appropriate. This model is then applied in the field to a single snow pit face to demonstrate the field applicability. This study was able to show a final uncertainty of ~1% for larger grain sizes as methods were not very promising for small grain sizes (not surprising, but excellent effort that was worth a shot).
Overall, I really liked reading this paper. It is well written and easy to follow. These methods will advance capabilities in observing LWC in the snow, a long-standing challenge in the field. With that being said, there are a few improvements that I think could really help expand the impact of the final paper. Most of these are relatively minor (at least I think so) and should not be too difficult to address. I put the comments in order of appearance in the paper, generally listed by line number. One general comment is that there are a number of recent studies investigating water flow through snow that could/should be cited (some are listed in the comments below).
Title: given the recent advances in mapping liquid water content at various scales (from pore scale to hillslope and watershed scales), I think specifying what scale this paper addresses in the title is justified.
L 23: "unprecedented detail" may be a bit of an overstatement. I think the method has the potential to do so in the future, but not in the current study. The previous paper that comes to mind is Williams et al. (2010) who were able to provide a 3-D model of a 1 m x 1 m cube of snow at 1 cm^3 resolution showing the meltwater flow patterns throughout. While the current study has higher resolution, it is limited to a single pit face. However, the Williams study should be referenced and potentially added as a comparison in the discussion.
Williams, M.W., Erickson, T.A. and Petrzelka, J.L. (2010), Visualizing meltwater flow through snow at the centimetre-to-metre scale using a snow guillotine. Hydrol. Process., 24: 2098-2110. https://doi.org/10.1002/hyp.7630
L66-70: Please clarify a little more as these processes often happen at the same time in a natural snowpack. Especially early in the melt season (Hirashima et al., 2019; Eiriksson et al., 2013). These studies could also improve the discussion as to potential future applications of the presented methods.
Hirashima, H., Avanzi, F., & Wever, N. (2019). Wet-snow metamorphism drives the transition from preferential to matrix flow in snow. Geophysical Research Letters, 46, 14548– 14557. https://doi.org/10.1029/2019GL084152
Eiriksson, D., Whitson, M., Luce, C.H., Marshall, H.P., Bradford, J., Benner, S.G., Black, T., Hetrick, H. and McNamara, J.P. (2013), An evaluation of the hydrologic relevance of lateral flow in snow at hillslope and catchment scales. Hydrol. Process., 27: 640-654. https://doi.org/10.1002/hyp.9666
L80: The Williams et al. (2010) study quantified spatial distribution using dye tracers.
L207-209: So an entire snowpit profile was taken prior to imaging. What effect do you think this had on the liquid water content of the pit face. For example, Shea et al. (2012) found that 90% of temperatures changed in a statistically significant manner in the first 90 seconds of exposure to the air and continued to change over time. I think you still demonstrate the applicability of this method in the field, but this should be a consideration for future studies that focus on the field applications.
Shea, C., Jamieson, B., and Birkeland, K. W.: Use of a thermal imager for snow pit temperatures, The Cryosphere, 6, 287–299, https://doi.org/10.5194/tc-6-287-2012, 2012.
L225-226: Maybe an iterative approach could overcome this? Essentially, the sum of LWC mass and dry density should equal the density cutter bulk density. For example, if the SLF sensor gives you 6% LWC, that is 60 g of water for a 1 L density cutter, and if the dry density estimate for that is 350 kg/m^3, then you can compare 410 g (350 + 60) to the actual bulk measurement from the cutter and adjust accordingly. The SLF sensor records and logs the measured permittivity with the calculated LWC so this should be relatively straight forward using the equations in the manual. Feel free to reach out if this does not make sense.
L242: I do not recall if you specify that you are referring to volumetric LWC. Please double-check that this is done somewhere.
Figure 7: This is a really cool figure. It did take me a minute, though, to get used to blue being dry and yellow being wet. Please consider inverting the color scheme. Also, a scale bar and dashed lines for the SLF sensor footprint oculd be quite helpful to see how much variability occurs in the footprint of the instrument. Possibly even histograms of just the SLF sensor footprint, if they differ from the whole ROI.
L338: did you take grain size measurements at the end of the experiments also to look at grain growth? This might also have an impact (as you mention in the discussion).
L3339: "this data" please correct to "these data"
L345: the 2% RMSE is also the stated accuracy for the SLF sensor I think (maybe worth mentioning), so these are great results!
L474: Please see previous comment for iterative approach.
L511: I was really hoping for a lot more discussion towards this in the 'discussion' section. Questions kept popping up like: why is this useful? How might this method be helpful in the field for future studies? What improvements can you suggest to reduce the impact of the rapid changes after exposing the face (as previously mentioned) and to capture the entire pit face down to the ground? etc. A paragraph or two to this affect could really help the impact of the paper.
L515-523: This final paragraph reads much more like discussion than conclusions, in my opinion. I recommend moving to the discussion section.
I would like to re-iterate how much I like this paper and my comments are meant to be constructive. Please feel free to reach out by email or in the interactive discussion if something that I wrote is unclear or you wish to discuss further.
-Ryan Webb
- AC1: 'Reply on RC1', Christopher Donahue, 11 Oct 2021
-
RC2: 'Comment on tc-2021-247', Chander Shekhar, 15 Sep 2021
The article is written well and convey properly the work carried out by the authors. The objectives were mentioned clearly by authors in lines 59-63 of paper. This study focused on quantitative two dimensional retrievals of snow surficial liquid water content (LWC) using near infrared hyperspectral imaging (NIR-HSI) measurements in two scenarios. Firstly, LWC was mapped in controlled laboratory scenario by comparing reflectance spectra (measured using NIR-HSI) of snow samples (07Nos.) at different time steps with simulated reflectance spectras of three theoretical models by tuning LWC and grain size parameters. The retrieved LWC values were compared with dielectric measurement (SLF Sensor) based LWC of same snow samples. It led to selection of best matching hyperspectral imaging based retrieval model among the three. RMSE was found to be around 1% in LWC for large retreived snow grain size values (>176µm and <1000 µm). The LWC retrieval models did not performed well for small grain sizes with complex shapes. Secondly, in field scenario the applicability potential of selected NIR-HIS based model was demonstrated for quantitative two dimensional surficial LWC retrieval on the wall of snow pit with an artificial source of illumination. The efforts by authors are appreciated as limited work is available on quantitative two dimensional LWC using hyperspectral imaging sensors.
There are a few observations and suggestions that i think will help to improve the overall impact of the paper. Specific comments include concerns based on curiosity for actual applications and other minor comments/suggestions have been put in order of appearance in the paper, listed by the line numbers.
Specific omments
- For field retrievals of LWC using NIR-HSI, dry snow densities could not be measured that lead to certain uncertainties in LWC. Also, few of the potential sources of uncertainty were discussed in section 5.3 without any quantitative values for discussed ones (like spherical grain size, dry snow density, illumination source etc.). For lab scenario, what LWC would have been obtained if one proceeds without accounting for dry snow density, as happened in field scenario. The authors had already worked out for lab samples by inclusion of dry snow density values. Based on this, an uncertainty figure may emerge (for 07 samples) and it will provide confidence for actual field retrievals. An observation on intital and final snow parameters in lab experiment may also help. Authors can include discussion on this in uncertainty section (Section 5.3).
- The envisaged potential applications areas (in field/ air/ space borne modes) using proposed high resolution NIR-HSI method for quantitative two dimensional LWC may kindly be mentioned in introduction or background section (Line 95). It will provide clarity to readers why LWC details at high resolutions are required with advantages/ limitations over low resolution LWC. Suitable references may also be included.
- Power source of halogen lamps may be mentioned which were used in field conditions for illumination purpose. A discussion on advantages/ disadvantages of dielectric based (e.g. SLF sensor) and NIR-HSI based sensors (in terms of outputs obtained by both instruments, performance time, cost effectiveness, uncertainties involved, portability in field, preferable method/instrument for specific application etc.) may help to provide better insights to readers.
- The parameters of actual snowpack and those measured on snow pit wall are expected to be different as exposure of snow pit wall while digging leads to changes in snow pack parameters on the wall. It would be intersting to know whether illumination using halogen lamps and opaque tarp is imperative for HSI based field measurements. In a natural sceraio, the signatures from snow pit wall under shadow will be of diffused sunlight and may affect the reflectance measurements and hence retrievals based on simulated reflectances. I am cusrious to know about applicability potential of NIR-HSI retreival method in actual field scenarios.
- Refer Figure 10: To obtain the snow pack stratigraphy, one has to dig a snow pit. Ice layers can be easily recognised visually in layers of snow. It is known fact that LWC will have high values above the impermeable ice layer for a wet snow pack and can be measured easily using portable SLF sensor in any illumination condition. Definitely, resolution of NIR-HSI is better that SLF sensor. From application point of view, it is curiosity to know where and how this high resolution details will actually help, knowing the fact that (i) LWC has high spatial and temporal variability in snow pack and (ii) hyperspectral information retrieval is constrained to surface measurements only. Kindly mention.
The specific comments express concern towards application potentail of NIR-HSI based LWC retrieval methods with due recognition to actual field constraints, variability in field spectral reflectance signatures caused by number of dynamic factors (e.g. dynamic snow parameters, viewing/ illumination conditions and geometry etc.) and limited penetration of NIR radiations into the snow.
Other minor comments/suggestions:
L 1-3: Inclusion of appropriate scale in the title of paper at which work had been performed will provide clarity. As this work had not been tested for air/space borne sensors to cover large scales.
L 15: ‘determine’ to be changed to ‘determined’.
L23: ‘unprecedented detail’ appears to be an exaggeration. Kindly use appropriate word/details.
L149: Inclusion of workflow diagram in Section 3 will provide quick easy understanding to the readers about the work that was carried out.
L 175: In laboratory setup, the distance (m) of halogen lamps and NIR-HSI imager from snow samples may kindly be mentioned.
Table 1: A column may be added for retrieved grain size values using SBA method for dry snow samples, that corresponds to grain size legend of Figure 8.
L 205-206: The distance (m) of halogen lamps and NIR-HSI imager from snow pit wall may be mentioned.
L 216: What were the implications of using 36% reflectance calibration panel during field experiments in place of 99% (that was used in lab scenario)?
Figure 4: The figure may be modified to depict clearly that upper 110cm of snowpit wall was imaged using NIR-HSI. It appear as if image was taken upto 150cm depth.
L 225: For words ‘a small error’, a qunatified value may be mentioned.
L 250–254: It appeared that single scattering properties were tuneable using only two parameters LWC and re. Are there any other parameters also which have not been considered, under some assumptions/ approximations ? These can be mentioned.
Figure 7: SLF sensor measured LWC values corresponding to time steps in (A to D) may also be mentioned.
L 301-304: Criteria or reference statistical parameter (RMSE, Bias etc.) can be mentioned for the quoted relative best or poor performance of models.
Figure 8: Legends are same for Figure 8 (A – C) and can be placed outside the three figures to have better representation. Author can take a decision on this based on editor’s suggestions.
Table 2: (1) The numder of sample points used to arrive at RMSE and Bias corresponding to each sample need to be mentioned. The confidence level may also be mentioned. (2) The word ‘RSME’ need to be spelled correctly.
Caption of Figure 10: The words ‘depth average of the SLF sensor area only’ may be modified to provide clarity that it is HSI measurement corresponding to area covered by SLF sensor only. Description of ‘NaN’ can also be mentioned.
Table 3: Spelling corrections: ‘Senor’ to be replaced by ‘Sensor’.
L 454-457: Kindly check whether these lines convey the correct message. It seems like inclusion of certain additional points has led to smaller SBA. Kindly verify.
L 469: A quantitative figure or reference for uncertainty of SLF sensor may be mentioned.
L 515: ‘un-precedented detail’ appears to be an exaggeration. Kindly use appropriate word/ details.
L 519-524: While using HSI based retreival methods, kindly suggest the ways to account for the expected variability in refelctance signatures caused by illumination changes for air/space borne sensors. In this air/space borne sensor scenario, the level of uncertainty expected in LWC mapping using proposed simulated spectra aproach may also be commented.
The article presented analysis on LWC retrived from spectral reflectance signatures (in image form) and LWC measured using dielectric based, point form data acquired in synchronization for fair comparisons. It is nice set of information that will help research community to understand the potentail of hyperspectral data for retrieval of LWC parameter of snow. The comments/ observations have been written with a constructive and curious spirit to improve the impact of the paper.
Best wishes.
-
Chander Shekhar
- AC2: 'Reply on RC2', Christopher Donahue, 11 Oct 2021