the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Estimating snow accumulation and ablation with L-band InSAR
Ryan W. Webb
Hans-Peter Marshall
Anne W. Nolin
Franz J. Meyer
Abstract. Snow is a critical water resource for the western US and many regions across the globe. However, our ability to accurately measure and monitor changes in snow mass from satellite remote sensing, specifically its water equivalent, remains a challenge in mountain regions. To confront these challenges, NASA initiated the SnowEx program, a multi-year effort to address knowledge gaps in snow remote sensing. During SnowEx 2020, the UAVSAR team acquired an L-band Interferometric Synthetic Aperture Radar (InSAR) data time series to evaluate the capabilities and limitations of repeat-pass L-band InSAR data for tracking changes in snow water equivalent (SWE). The goal was to develop a more comprehensive understanding of where and when L-band InSAR can provide snow mass change estimates, allowing the snow community to leverage the upcoming NASA-ISRO SAR (NISAR) mission. Our study analyzed three InSAR image pairs from the Jemez River Basin, NM, between 12–26 February 2020. We developed an end-to-end UAVSAR InSAR processing workflow for snow applications. This open-source approach employs a novel data fusion method that merges optical snow covered area (SCA) information with InSAR data. Combining these two remote sensing datasets allows for atmospheric correction and delineation of snow covered pixels. For all InSAR pairs, we converted phase change values to SWE change estimates between the three data acquisition dates. We then evaluated InSAR-derived retrievals using a combination of optical snow cover data, snow pits, meteorological station data, in situ snow depth sensors, and ground-penetrating radar (GPR). The results of this study show that repeat-pass L-band InSAR is effective for estimating both snow accumulation and ablation with the proper measurement timing, reference phase, and snowpack conditions.
Jack Tarricone et al.
Status: final response (author comments only)
-
CC1: 'Comment on tc-2022-224', Simon Gascoin, 18 Nov 2022
Congratulations to the authors for this study. My comment is about this statement "Optical fSCA data are needed to identify snow covered pixels (..) The Landsat 8 image data used in this study represented two of the very few cloud free days throughout the winter time series over VCNP. To account for the significant issue of cloud cover, future investigations should leverage optical sensor fusion and interpolation methods (...)"
Instead of fusion and interpolation, Sentinel-2 provides Landsat-like data (high quality multispectral imagery at 10-20 m resolution) with a 5 day revisit time. Just in February 2020, I find 4 clear sky images over the study area in Valles Caldera National Preserve.
Citation: https://doi.org/10.5194/tc-2022-224-CC1 -
AC2: 'Reply on CC1', Jack Tarricone, 21 Mar 2023
The comment was uploaded in the form of a supplement: https://tc.copernicus.org/preprints/tc-2022-224/tc-2022-224-AC2-supplement.pdf
-
AC2: 'Reply on CC1', Jack Tarricone, 21 Mar 2023
-
RC1: 'Comment on tc-2022-224', Cathleen Jones, 21 Dec 2022
The comment was uploaded in the form of a supplement: https://tc.copernicus.org/preprints/tc-2022-224/tc-2022-224-RC1-supplement.pdf
-
AC3: 'Reply on RC1', Jack Tarricone, 21 Mar 2023
The comment was uploaded in the form of a supplement: https://tc.copernicus.org/preprints/tc-2022-224/tc-2022-224-AC3-supplement.pdf
-
AC3: 'Reply on RC1', Jack Tarricone, 21 Mar 2023
-
RC2: 'Comment on tc-2022-224', Silvan Leinss, 22 Dec 2022
Review, manuscript "Estimating snow accumulation and ablation with L-band InSAR"
by J. Tarricone et al.General comments:
The authors describe the evaluation and interpretation of L-Band repeat-pass radar interferometry. Their aim is to observe changes of the snow water equivalent during partially wet snow conditions.
SWE estimation by radar interferometry is a very promising technique that has been developed within the past 20 years with increasing success. The main problems of this technique are a quick loss of coherence, correction for atmospheric phase delays, existence of phase reference points, and - especially for wet snow - the uncertainty and variability of the permittivity and signal penetration through the snow pack.
The authors tackle some of these problems and show that, even for melting conditions, coherence is maintained at L-band for the 7 and 14 day repeat time of their three acquisitions. They correct large-scale atmospheric phase delays by referencing the InSAR phase to snow free areas in the scene which is useful if no topography-dependent atmospheric phase exists. The authors reference the local InSAR phase to 8-9 pixels around a local snow pit and assume the remaining phase originates from changes in SWE (accumulation or melt) rather than local km-scale atmospheric delays. It is not clear where the km-scale phase patterns originate from (SWE change, permittivity changes or atmosphere).
Unfortunately, during the 14 days of the InSAR observation period, the snow pit data showed hardly any change in SWE which makes it very difficult for the authors to find a statistically convincing relation between the InSAR phase and SWE changes. Therefore, the authors rely on interpretation of local patterns which they assume to be caused by snow melt. Sentinel-2 imagery, as suggested by Simon Gascoin https://doi.org/10.5194/tc-2022-224-CC1, supports the authors assumption and should be considered. See provided images in the supplements (Channel 8 provides some information about snow wetness).
In addition to the unfortunate meteorological conditions I have some concern because the authors use the dry-snow-SWE-to-InSAR equation (Guneriussen 2001, Leinss 2015) even though wet snow (melting condidions) are considered. This equation depends on the permittivity of snow which changes significantly (from 1.5 to at least 2.2) when the snow becomes wet at constant SWE. Therefore, (some of) the observed phase change might be due to increasing wetness rather than a change in SWE.
The paper is very well structured and good to read, but need to be improved by better focusing on the main results; in addition to addressing the above mentioned concerns which could require a major revision of the paper.
Specific comments:line 50: "where shorter wavelengths (..) have been used to estimate SWE (references)": The authors detail some technical challenges faced when estimating SWE from backscatter. It would be interesting, to provide a rough estimate or precision based on the conclusions from the cited authors, that apply these techniques, to indicate how well they were able to actually estimate SWE from backscatter. For example, accoording to the scattering physics that happen in snow, radar backscatter does not necessarily show a monotonically increasing relation to backscatter. I would claim, that determining the required in-situ parameters to derrive SWE might be even more difficult that determining SWE directly.
58: If the authors think it would be beneficial, they could cite Stefko/Leinss(2022) which provides a completely new approach to analyze the radar backscatter from snow (even though it does not claim to derrive SWE or snow height).
Stefko and Leinss et al. "Coherent backscatter enhancement in bistatic Ku- and X-band radar observations of dry snow", TC 2022, https://tc.copernicus.org/articles/16/2859/2022/91: "accumulation and ablation": It is not totally clear what the difference is of this study compared to the study discussed in the paragraph before (Marshall 2021). In the previous paragraph "a wide range of (..) snow conditions" is specified. This study/paragraphs seems to adress "both snow accumulation and ablation". Do you mean ablation of dry snow by wind drift or evaporation, or do you mean ablation my melt an runoff? In line 92 you mention that the "UAVSAR-based approach" has been only applied to dry snow "but not melt". Does that indicate that you adress melt or, at least, wet snow periods? Try to better describe the differences betwen the two studies (you could also add the study side here, instead of in line 96.).
125 - 130: The equation from Guneriussen requires knowledge about snow density rho_s and the snow permittivity epsilon_s. Due to a lack of spatially (and vertically) distributed information, these two variables seem to be determined from measurements in two snow pit data as written in line 255-257 which could potentially introduce a significant bias on the derrived SWE values. However, as shown in Leinss et al. (1015) [Figure 8left, Figure 9], for dry snow (and only for dry snow), there is an almost linear relationship between SWE and the InSAR phase which does hardly depend on snow density rho_s or epsilon_s. I think it is worth considering or mentioning this as it simplifies SWE determination.
Reference: Leinss et al. "Snow water equivalent of dry snow measured by differential interferometry", IEEE JSTARS (2015), https://doi.org/10.1109/JSTARS.2015.2432031125 - 130, continued:
In contrast to dry snow, for melting conditions, the linear relation between the InSAR phase change and SWE does not hold anymore. Please provide some information about the permittivity of wet snow compared to dry snow. From the references [Sihvola 1986, Webb 2021, Hallikainen 1986, fig. 9] I obtain that the real part of the permittivity increases from 1.5 to about 2.2...2.6 for the same parameters of density (300 kg/m^3), LWC (5%) and frequency (1 GHz). This should cause a significant change in the observed phase change even for a constant SWE.
For references, see comment about Table 2 further below.Furthermore, the penetration depth into wet snow could vary considerably from the penetration into dry snow, at least for higher frequencies. For L-band and wet snow, the expected penetration should be checked to be larger than the snow depth.
129/130: Please mention how strongly the liquid water content can affect the real and imaginary part of the permittivity of wet snow. The real part has a stronger effect and changes are considerable. See further references and hints in the comment below for Table 2.
132-142: To better understand the study area, I highly suggest showing Sentinel-2 images as suggested by Simon Gascoin to illustrate the land-cover of the studied area. There are S2 images from 2020-02-06, 2020-02-16, 2020-02-21, 2020-02-26.
Figure 1: The length Delta R_a does not correspond to the illustrations in (Guneriussen 2001, Leins 2015) and to the underlying physics of equation (2).
Note that it might be better to reference the peer reviewed paper of Guneriussen rather than the IGARSS proceeding (here and other places, possibly cite both).
T. Guneriussen, K. A. Høgda, H. Johnsen, and I. Lauknes, “InSAR for estimation of changes in snow water equivalent of dry snow,” IEEE Trans. Geosci. Remote Sens., vol. 39, Art. no. 10, 2001, doi: http://dx.doi.org/10.1109/36.957273.118-121 (145-155): How was the interferometric data processed? Was there a perpendicular (across-track) baseline between the different flight tracks that caused a topographic phase that had to be removed? Why was the SRTM used for processing and not a better DEM with lower height noise (see e.g. Fig. 6)? Could the phase differences along the south-west exposed slopes be an artifact resulting from a DEM not perfectly coregistered with the radar data? Nevertheless, Sentinel-2 data suggest, that indeed, snow melt is observed here.
Figure 2, caption: Are the six CZO snow depth sensors all at the same location? I see only one black diamond.
160: Section 2.3.2 (Snow Pit) contains no information about SWE estimation even though table 2 lists SWE values. How were the listed SWE values determined?
164/165 and table 2: Does the change of epsilon_s result from an increase in density or a change in liquid water content?
Table 2: The listed mean epsilon_s values appear to be only the real-value of the permittivity. However, as the snow condition is melting, I would at least mention the order of magnitude of the imaginary part of epsilon (permittivity of wet snow). For an imaginary part of epsilon on the same order (or larger) as the real part, equation (3) is not valid any more because the refractive index depends on both, the real and imaginary part. Fortunately, from references about the permittivity of wet snow around 1 GHz [Sihvola 1986, Webb 2021, Hallikainen 1986, fig. 9] I obtain eps''_wetsnow = 0.05..0.1 at 1 GHz, density 300 kg/m^3, liquid water volume content LWC=5%, which is by a factor of 5-10 smaller than the change of the real part of the permittivity: epsilon'_wetsnow increases from 1.5 to 2.2...2.6 for the same parameters of density, LWC and frequency.
References:
- M. Hallikainen, F. Ulaby, and M. Abdelrazik, “Dielectric properties of snow in the 3 to 37 GHz range,” IEEE Trans. Antennas Propag., vol. 34, Art. no. 11, 1986, doi: 10.1109/TAP.1986.1143757.
- A. Sihvola and M. Tiuri, “Snow fork for field determination of the density and wetness profiles of a snow pack,” IEEE Trans. Geosci. Remote Sens., Art. no. 5, 1986.
- Webb, R.W.; Marziliano, A.; McGrath, D.; Bonnell, R.; Meehan, T.G.; Vuyovich, C.; Marshall, H.-P. In Situ Determination of Dry and Wet Snow Permittivity: Improving Equations for Low Frequency Radar Applications. Remote Sens. 2021, 13, 4617. https://doi.org/10.3390/rs13224617Section 2.4:
First, you mention a high-pass and low-pass filtering sequence, then you mention that the wet delay is caused by spatial variations in water vapor. But finally, if I understood right, there is simply a linear phase ramp removed. Reading line 202-211 I would expect a more advanced atmospheric filtering method. So maybe, you could better justify that removing a linear phase ramp is sufficient.215-225: Is there any difference between the PLC and slant-range? If both is equivalent, then I guess, you are only removing a linear (or possibly quadratic) ramp from the interferogram in the slant-range radar geometry.
215-222: Partially redundant information (PLV). Please check and remove redundancies.
228: raw data: do you mean the unfiltered interferogram? In the context of radar, raw data usually refers to unfocused radar raw data.
237: "significant errors within the original SRTM DEM": Are these "errors" larger than the specified accuracy of the SRTM? I don't think so. I think these undulations are due to phase noise in the SRTM interferograms. I would expect, that the same noise level is observed for snow free areas at lower altitude.
262: "the Delta SWE values for the eight surrounding pixels" - do you mean phase values?
275/276: Would averaging the HH and VV polarization (or the complex interferograms) improve the phase noise?280: "likely caused by snowpack LWC attenuating the radar signal": Is there any independent indication that LWC in this area is larger than in the surrounding area of same altitude? When looking at the backscatter image, I interpret the very low coherence and the significantly lower backscatter (compared to the surrounding separated by a sharp transition from high to low backscatter) rather as liquid surface water or wetland, possibly below ice covered by a thin layer of snow. Could you check? You mention "riparian area" indicating wetland.
292-298: You write about SWE changes and SWE mapping, even though Figure 8 shows results of the InSAR-derrived Delta SWE results. Try to make clearer in this Section (3.2) that you describe InSAR-derrived results, and not field-measured SWE changes.
331: Looking at Figure 11, I think VG shows changes in snow height, not in SWE.
334: "we used new rho_s": Do you mean, "we used a density of rho_s = 240 kg/m^3" for new snow?
338, Section 3.5: Comparing Figure 13 a and b (Delta SWE vs. fSCA), some patterns agree. However, in the south-eastern corner, there is a 1km large area that shows -100% change of snow while the surrounding area shows no change. In contrast, the SWE change map shows no spatial variation at all of SWE for this area. Can you explain that?
Furthermore, in the south-western corner, south of the GPRI measurements, there is a few hundreds meter high mountain where Delta SWE is negative while fSCA is positive. Could you explain also this?381: "Our GPR data provide further justification in our analysis beyond single point comparisons." Looking at Figure 12, I consider the GPR data as not convincing enough to provide reasonable justification.
technical comments:
Typo in Figure 2: Meteorlogical Staitons
Figure 3: UAVSAR swath extend: add "(black outline)" to clarify that the fSCA map is clipped to the outline from UAVSAR.
161-163: Could you put a label for HQ and BA into Figure 2c?
164: for each of the two snow pits?
Table 2: UAVSAR start, Pit start: I guess this a local times. Could you add [hhmm] to the table header line or "start time of snow pit creation" to the table caption?
235: "line of site" -> line of sight
Figure 11: Could you add to the caption of the figure the elevation of Redondo peak and HQ?-
AC1: 'Reply on RC2', Jack Tarricone, 21 Mar 2023
The comment was uploaded in the form of a supplement: https://tc.copernicus.org/preprints/tc-2022-224/tc-2022-224-AC1-supplement.pdf
-
AC1: 'Reply on RC2', Jack Tarricone, 21 Mar 2023
Jack Tarricone et al.
Model code and software
Estimating snow accumulation and ablation with L-band InSAR: R and Python code for analysis and figure creation Jack Tarricone https://doi.org/10.5281/zenodo.7199791
Jack Tarricone et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
530 | 202 | 19 | 751 | 9 | 8 |
- HTML: 530
- PDF: 202
- XML: 19
- Total: 751
- BibTeX: 9
- EndNote: 8
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1