the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Snow Water Equivalent Retrieval Over Idaho, Part B: Using L-band UAVSAR Repeat-Pass Interferometry
Zachary Marshall Hoppinen
Shadi Oveisgharan
Hans-Peter Marshall
Ross Mower
Kelly Elder
Carrie Vuyovich
Abstract. This study evaluates using interferometry on low frequency synthetic aperture radar (SAR) images to monitor snow water equivalent (SWE) over seasonal and synoptic scales. We retrieved SWE changes from nine pairs of SAR images, mean 8 days temporal baseline, captured by an L-band aerial platform, NASA's UAVSAR, over central Idaho as part of the NASA SnowEx 2020 and 2021 campaigns. The retrieved SWE changes were compared against coincident in situ measurements (SNOTEL and snow pits from the SnowEx field campaign) and to 100 m gridded SnowModel modeled SWE changes. The comparison of in situ to retrieved shows a strong Pearson correlation (R = 0.80) and low RMSE (0.1 m, n = 64) for snow depth change and similar results for SWE change (RMSE = 0.04 m, R = 0.52, n = 57). The comparison between retrieved SWE changes to SnowModel SWE change also showed good correlation (R = 0.60, RMSD = 0.023 m, n = 3.2e6) and especially high correlation for a subset of pixels with no modeled melt and low tree coverage (R = 0.72, RMSD = 0.013 m, n = 6.5e4). Finally, we bin the retrievals for a variety of factors and show decreasing correlation between the modeled and retrieved values for lower elevations, higher incidence angles, higher tree percentages and heights, and greater cumulative melt. This study builds on previous interferometry work by using a full winter season time series of L-band SAR images over a large spatial extent to evaluate the accuracy of SWE change retrievals against both in situ and modeled results and the controlling factors of the retrieval accuracy.
- Preprint
(9526 KB) - Metadata XML
- BibTeX
- EndNote
Zachary Marshall Hoppinen et al.
Status: open (until 06 Oct 2023)
-
RC1: 'Comment on tc-2023-127', Andrea Manconi, 11 Sep 2023
reply
The manuscript from Hoppinen et al. presents an analysis of Snow Water Equivalent (SWE) retrieved via remotes sensing (radar interferometry). The authors exploit L-Band SAR data acquired in Idaho with the UAVSAR platform and compare/validate the results against ground stations and model simulations. This work is of major importance for the remote sensing community, as well as for the development of cryospheric research. The manuscript is well written, the dataset is unique, the methods used analyses performed are of scientific sound, the figures are appropriate and the results and conclusions are of high relevance. The results are of major interest considering the future L-Band SAR missions (such as NISAR), thus I strongly support the publication of this manuscript, provided that the authors include some modifications and additional details to the current version. In particular:
(1) The authors state several times that they "utilized wrapped images when complete spatial or temporal coverage was necessary". However, this requires a clarification, especially to readers not aware of (or not used to) the differences between wrapped and unwrapped phase in radar interferometry. I suggest providing specific details what does it mean exactly and how you combined the results of wrapped phase and unwrapped phase
(2) The section 3.2. is unclear. I think I get the sense of what you mean when you need for a reference phase, but the process of how you get is not straightforward (at least not in your explanation) . I suggest to write down the formulas and also add a figure showing how the reference phase looks like.
(3) I have some doubts on the description of the theoretical 2pi limitation for phase wrapping, which I hope the authors can clarify:
(a) Spatial limitation: it is true that if the InSAR retrieved deformation field is smooth and continuous, implying also appropriate spatial sampling (pixel resolution), the wrapping limit is at 2pi. However, some discontinuities in the InSAR results might occur, i.e., the phase unwrapping (which is a gradient based approach, and needs thus continuity) would fail in providing accurate results. I don't have experience with L-Band interferograms related to snow height change, thus it is difficult for me to understand if the continuity condition is respected, especially in locations with high topographic relief. Including one or more interferograms (wrapped) either in the main text or in the supplementary would help in better understanding.
(b) Temporal limitation. The theoretical limit of phase aliasing between 2 acquisitions is = lambda/(4*dt). With lambda L-Band ca 24 cm this means that in case of changes larger than 1.5 cm/day on the same pixel, we would reach the ambiguity limit. If the spatial unwrapping works well (see point before) then it should be not a problem. However, what happens in the cases when the phase unwrapping does not work and you use the wrapped phase values?
(4) related to the previous point, I find figure 7 of difficult reading. I know that it is convenient to put on a single graph several variables, but i think that for a better understanding you can put several graphs for different densities (using upper and lower boundaries) and/or different incidence angles. As mentioned in point (3a and 3b) spatial and temporal resolution play also an important role in the definition of the phase aliasing.
(5) Missing units on the Figure 9 (y-axis)
Citation: https://doi.org/10.5194/tc-2023-127-RC1 -
RC2: 'Comment on tc-2023-127', Mathieu Le Breton, 20 Sep 2023
reply
This study uses time-lapse data from an aerial radar and a SAR method, to estimate the properties of the snowpack. The development of this technology seems promising to study the snowpack over a large area, with better quality of measurements than with satellites. The article is written very clearly. Such data is also unique. From this point of view, this study is really worth publishing.
However, I am concerned by a potential methodological flaw. The article claims to retrieve the Snow Water Equivalent using aerial SAR data. Yet, on line 171-173, you say that for each image pair, you use the mean density (equivalent to dielectric permittivity) from insitu observation, in order to estimate snow depth and SWE from UAVSAR. The SAR measurement is therefore not independent, and formally the information it brings is snow depth. To explain: let us say you measure just snow depth from another method (e.g., LIDAR, photogrammetry), then using an estimate of average density from insitu measurement is sufficient to estimate SWE. As it is, I would rather say that the method estimates snow depth, by combining UAVSAR+insitu density measurements. SWE is then derived using again this average density. Following this point of view, comparing results that use insitu data, with the same insitu data, seems sloppy. In consequence, accuracy estimation (figure 10, figure 11, section 7.1), one of the article’s question, seems sloppy.
If I have misunderstood something, that is very good (because the study is otherwise very good). Then please clarify how the different data are used and compared (for example, adding a diagram describing the processing workflow)
If I am correct, you could either reduce the article’s claims, or, better, modify your processing workflow so that the UAVSAR estimation is fully independent from insitu data. It could be possible given some small approximations. Some examples using L band waves :
- Le Breton, M., Larose, É., Baillet, L., Lejeune, Y., van Herwijnen, A., 2023. Monitoring snow water equivalent using the phase of RFID signals. The Cryosphere 17, 3137–3156. https://doi.org/10.5194/tc-17-3137-2023
- Koch, F., Henkel, P., Appel, F., Schmid, L., Bach, H., Lamm, M., Prasch, M., Schweizer, J., Mauser, W., 2019. Retrieval of Snow Water Equivalent, Liquid Water Content, and Snow Height of Dry and Wet Snow by Combining GPS Signal Attenuation and Time Delay. Water Resources Research 55, 4465–4487. https://doi.org/10.1029/2018WR024431
Please find also some more minor comments for clarification :
- 175: the UAVSAR timeseries represents what ?
- 177: «retrieved mean snow depth or SWE» : is it snow depth or SWE ?
- 5 & 118 : What is SNOTEL, how does it measure SWE ? 97 : what are the telemetered stations ?
- 98: what are the models for SWE ?
- 112: Which nine pairs did you use ?
- Fig 2: Maybe indicate the buffer boundary here, to clarify relation with fig 3.
- Fig 3: What do you mean by clipping the data, and buffer zone ?
- Fig 4: Is it cumulative precipitations ? It looks there is no melting.
Fig 4: Can you add snow depth ? - 128: Did you use Liston’s model ? (you state it ‘can’ be used, not that you used it)
- 132: Not clear what you used for computing the snow model. (ok: explained later)
- What is the altitude of the plane ?
- How do you compute the phase ?
- What is the emitted frequency exactly ?
- How do you ensure in your method that the phase is dominated by the ground reflexion, and not by the reflexion on the top of the snow ?
- 166: What is a retrospective atmospheric to remove atmospheric phase ? Is it what is described in the previous paragraph ?
- 174: From the title of part 4, we expect results, yet this is still the method. Also, section 2 is called «method» (which it is) bu parts 3 and part 4, that are also methodological, are called differently. That is minor but a bit confusing.
- Fig. 7 is not straightforward to grasp and use. I may suggest a simpler graph more focused on the unwrapping limits. It could be for example a 2D graph with just a line representing «2pi» phase shift, depending on the angle of incidence and on SWE variation (several lines for several fixed densities). It would be more informative.
Citation: https://doi.org/10.5194/tc-2023-127-RC2
Zachary Marshall Hoppinen et al.
Zachary Marshall Hoppinen et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 148 | 48 | 7 | 203 | 3 | 1 |
- HTML: 148
- PDF: 48
- XML: 7
- Total: 203
- BibTeX: 3
- EndNote: 1
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1