the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Mar 2025
Multitemporal analysis of Sentinel-1 backscattering during snow melt using high-resolution field measurements and radiative transfer modeling
Abstract. The spatiotemporal evolution of snow melt is fundamental for water resources management and risk mitigation in mountain catchments. Synthetic Aperture Radar (SAR) images acquired by satellite systems such as Sentinel-1 (S1) are promising for monitoring wet snow due to their high sensitivity to liquid water content (LWC) and ability to provide spatially distributed data at a high temporal resolutions. While recent studies have successfully linked S1 backscattering to various phases of snowpack melting, a correlation with detailed snowpack properties is still missing. To address this, we collected the first dataset of detailed wet snow properties tailored for SAR applications over two consecutive snow seasons at the Weissfluhjoch field site in Switzerland. First, our dataset enabled the validation of previous methods relying on multitemporal SAR backscattering to characterize melting snowpacks and physically linked the increase in backscattering following the local minimum to the evolution of surface roughness. Then, the dataset was used as input to the Snow Microwave Radiative Transfer (SMRT) model to reproduce the S1 backscattering signal. Our simulations showed a general negative bias compared to the satellite data, with the most significant drivers being LWC early in the melt season and the surface roughness later on. The results also highlight several key challenges for reconciling S1 signals with radiative transfer simulations of wet snow: (i) the discrepancy in spatiotemporal variability of LWC as seen by the satellite and validation measurements, (ii) the lack of fully validated permittivity, microstructure and roughness models for wet snow in the C-band, (iii) the difficulty of capturing wet snow features potentially generating stronger scattering effects on a large scale, such as internal snowpack structures, soil features in case of low LWC, and surface roughness, which are not necessarily captured by point-wise measurements.
- Preprint
(36479 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2025-974', Anonymous Referee #1, 23 Apr 2025
General Comments
Overall, this is a well-presented manuscript and rigorously designed study. I appreciate the work the authors put in to ensure that the study design was thoroughly documented. The work is timely as it provides a detailed overview of the use of RT modelling for analysis of wet snow environments. Additionally, the use of SMRT is appreciated. The discussion is detailed and throughout the paper the authors appropriately acknowledge limitations of the study. The authors have presented information well, but I do have comments regarding the figures included in the manuscript. Minor technical corrections and science questions for an overall well-done study.
Specific Comments
I do not have major specific comments. The authors have done a good job clarifying the research methods.
L376 – It was unclear to me how many layers were used for the SMRT simulations. From the text it seems layers were varied to account for the C-band wavelength, which is fine. However, a short statement outlining the range of layers or some statistics on this parameter would be good to see.
Technical Comments
Most of my comments relate specifically to the figures included in the paper. Not the information but more related to formatting/design.
Fig 2. – The colours in panel a) are not clear. The white area is skiable domain, however, has a greener shade to it. Recommend using a hash or some sort of pattern to denote this area.
Panel b) and Panel c) – I realize that if I zoom in the legend is more visible, however, it is difficult to see at 100% zoom. The legend is also identical for both panels. Therefore, one larger legend with font size increased would be an improvement.
Figure 4. – This shows the setup for one SMRT simulation; however, nothing is marked as the SMRT layers used. Would this be possible to provide? Addresses my previous comment as well as providing some sort of example relate to the number of SMRT layers used.
Figure 5/6. – I think these are good figures illustrating the change in variables across the campaign. However, the text is not readable. Suggested changes: the different stages of snow melt could be included at the stop of only one pane as they are identical throughout the rest of the figure. Ideally the vertical lines could also be colour coded. Dates can be provided on only one pane as they are identical across. Also, font size is okay but a little small.
Figure 8 – Only one y-axis is needed between the figures. This would allow the size to be increased and increase readability.
Table 3 – Table layout is somewhat difficult to read. Horizontal outlines would aid in the interpretation of the data.
Figure 9 – This is a great figure. However, it is difficult to read. Two suggestions, 1) accompany the figure with a table which includes the specific explanations that way the table would be less busy, 2) use numbers or letters to point to specific events rather than having the arrows across the figure. This would again aid in improving the readability. Currently it takes too long to interpret it.
Table 4 – Similar comment to table 3.
Figure 11 – The bottom pane on this figure is unnecessary – it does not add anything to the interpretation. Simply providing the LWC for the top layer would be sufficient. Also, the LWC legend is unnecessary on the bottom. The RMSH legend is good.
Citation: https://doi.org/10.5194/egusphere-2025-974-RC1 - AC2: 'Reply on RC1', Francesca Carletti, 15 Jun 2025
-
RC2: 'Comment on egusphere-2025-974', Anonymous Referee #2, 20 May 2025
General comments/Main issues:
The manuscript addresses the temporal evolution of C-band backscatter intensity in high-Alpine terrain and its relation physical properties of the snow cover, in order to evaluate the feasibility for monitoring various phases of snowpack melting by means of satellite-borne SAR systems. This is a topic of great interest for hydrology and water resources management in regions for which the water supply relies on snowmelt runoff. On the sensor-side the investigations are based on data acquired by the C-band SAR sensor of the Copernicus-1 mission, a data base that has been used for operational monitoring of snowmelt area extent since several years.
The overall study approach is promising, comprising comprehensive, detailed field observations on physical snow properties, the analysis of Sentinel-1 backscatter time series during two snow cover seasons with particular focus on the snowmelt period, and the modelling of backscatter intensity. The presented material provides important new insights on the evolution of Alpine snow cover during different snowmelt phases and the related response of radar backscatter signals, offering essential information for advancing applications of satellite-based SAR observations for monitoring specific properties of the snow cover during the snowmelt period.
The completeness and traceability of the presentation differs, depending on the topic. The dense time sequence of very detailed snow property measurements during two snow seasons and the related analysis and discussion are a strong part of the study. On the other hand, the analysis and interpretation of the Sentinel-1 (S1) backscatter signals as well as the backscatter modelling part have weaknesses. Regarding S1 backscatter signatures, two main issues are: (i) the radiometric performance of the S1 backscatter intensity values of the rather confined cell used for signature analysis and validation tasks; (ii) the representativeness of the backscatter values of this reference cell in respect to backscatter in its immediate surroundings and to the typical backscatter signals of undisturbed snow areas in the same elevation zone.
- The radiometric accuracy for the backscatter data of the reference cell needs to be specified in detail. The reference to Marin et al. (2020) does not provide specific information on radiometric uncertainty for this particular cell of 20 m x 20 m extent. In S1 IW mode data it covers only about 4 independent samples (looks), resulting in high speckle-related uncertainty. Marin et al. apply for speckle reduction a multitemporal filter of 11 x 11 pixels and a gamma-MAP filter of 3 x 3 pixels. In both cases the radiometry is preserved if the window covers a homogeneous distributed target. This is not the case in the area surrounding of the refence cell. For the multitemporal filter different temporal response within the window may introduce additional radiometric biases for sigma-0 of individual pixels and dates.
- The representativeness of the S1 signatures of the reference cell in respect to the area in its surroundings should be assessed by analysing also data of other cells, preferably covering some larger area. Considering Fig. 2, it is unclear why cell 39 is used as single reference case. The data of cell 39 do not show a distinct incidence angle dependence and track 117 shows a different behaviour in the two years. The data of cell 18 of the two years, for example, are consistent. According to the elevation contour lines, this cell is located on an east-facing slope, and the two tracks of the ascending orbit show lower sigma-0 values, as expected for a back-slope. This may offers a possibility for studying impacts of the incidence angle.
For relating the backscatter modelling results to specific melt phases and for interpretation of the observed backscatter signatures, first of all it is important providing information on the backscatter contributions of the individual snow layers in dependence of the liquid water content. The relative contributions to total backscatter in dependence of depth below the snow surface should be specified for typical snow profiles shown in Figs. 5 and 6. Another concern is the limited information regarding properties of the scattering elements in the individual layers and the parametrization of the dense medium effect. In particular for the early melt phases, the properties of the top snow layer can be quite different, depending on size and shape of water inclusions (Arslan et al., 2003). Furthermore, information on the impact of incidence angle dependence of backscatter would be of interest, being of relevance for assessing roughness effects of wet snow surfaces.
Further issues:
Introduction and Discussion: During three winters Strozzi and Mätzler (1997; 1998) performed at the same test field above Davos C- and Ka-band backscatter measurements. Reference 1 (1997) is briefly cited in the manuscript, reference 2 (1998) is not cited. Results of these measurements are relevant within the context of the work presented by Carletti et al. and key points should be mentioned. Among issues addressed in the two papers are impacts of surface roughness and refrozen snow crusts based on measurements at different incidence angles, and the response to liquid water content.
P4, L121: The sensitivity to snow wetness depends on the local incidence angle, not directly on slope steepness. On steep fore-slopes the sensitivity is low.
P8, L212: A cutter of 55 cm length may not be suitable for resolving the density differences between individual layers that may be quite thin during the different melt phases. Please explain the limits in vertical resolution.
P9, Table 1: The elevation contour lines of Fig. 2a indicate different slope steepness within the test field. Please explain to which points the cited incidence angles refer and show the overall range of angles for the test field.
P11, Fig: 2: Incidence angles of cells on sloping show lower sigma-0 for ascending orbits, as to be expected for back-slopes. Consequently, the difference in sigma-0 between individual tracks may offer the possibility for exploring incidence angle effects.
P12, L328: The characterization of snow microstructure is a critical issue for snow backscatter modelling. Exponential correlation functions have major deficiencies, in particular for multi-size and sticky cases (Chang et al., 2016). Furthermore, the phase functions of snow with liquid water inclusions are quite different from that of dry snow (Arslan et al., 2003).
P14, L363: In particular during a main part of winter 2023-2024 the base of the snowpack shows zero deg. temperature, implying unfrozen ground. This goes on throughout the snowmelt periods.
P16 Fig. 5: Between mid-April and early June 2023 sigma-0 of the two ascending tracks shows consistent differences of 3 to 5 dB. Please explain the reason. The high sigma-0 values are probably from track 117 which shows high variance in 2023-2024, differing from 2022-2023 (Fig. 2b).
P20, Table 2: Please specify the incidence angle. Besides, the validity of these numbers in respect to other incidence angles would be of interest.
P 27, Fig. 11: Please specify the incidence angle. Strozzi and Mätzler (1997) show the incidence angle dependence of backscatter of wet snow (for smooth and rough surfaces) based on backscatter measurements at the same test site.
P29, L625: The development of surface roughness after start of the snowmelt period depends also on the sequence and intensity of snowfall events, varying from year to year.
P29, L641ff: The parametrization of the scattering elements may as well be a reason for differences between recorded and modelled backscatter (see e.g. Arslan et al., 2003; Chang et al., 2016).
References:
Arslan, A.N. et al.: Scattering from wet snow applying strong fluctuation theory, J. Electromagnetic Waves and Applications, 17:7, 1009-1024, https://doi.org/10.1163/156939303322519081, 2003.
Chang, W., Ding, K.-H., Tsang, L., and Xu, X.: Microwave scattering and medium characterization for terrestrial snow with QCA-Mie and bicontinuous models: Comparison studies, IEEE Trans. Geosci. Remote Sens., 54 (6), 3637-3648, doi: 10.1109/TGRS.2016.2522438:, 2016.
Strozzi, T., Wiesmann, A., and Mätzler, C.: Active microwave signatures of snow covers at 5.3 and 35 GHz, Radio Science, 32(2), 479-495, 1997.
Strozzi, T. and Mätzler, C.: Backscattering measurements of Alpine snow covers at 5.3 and 35 GHz, IEEE Trans. Geosci. Remote Sens., 36(3), 838-848, 1998.
Citation: https://doi.org/10.5194/egusphere-2025-974-RC2 - AC1: 'Reply on RC2', Francesca Carletti, 15 Jun 2025
Status: closed
-
RC1: 'Comment on egusphere-2025-974', Anonymous Referee #1, 23 Apr 2025
General Comments
Overall, this is a well-presented manuscript and rigorously designed study. I appreciate the work the authors put in to ensure that the study design was thoroughly documented. The work is timely as it provides a detailed overview of the use of RT modelling for analysis of wet snow environments. Additionally, the use of SMRT is appreciated. The discussion is detailed and throughout the paper the authors appropriately acknowledge limitations of the study. The authors have presented information well, but I do have comments regarding the figures included in the manuscript. Minor technical corrections and science questions for an overall well-done study.
Specific Comments
I do not have major specific comments. The authors have done a good job clarifying the research methods.
L376 – It was unclear to me how many layers were used for the SMRT simulations. From the text it seems layers were varied to account for the C-band wavelength, which is fine. However, a short statement outlining the range of layers or some statistics on this parameter would be good to see.
Technical Comments
Most of my comments relate specifically to the figures included in the paper. Not the information but more related to formatting/design.
Fig 2. – The colours in panel a) are not clear. The white area is skiable domain, however, has a greener shade to it. Recommend using a hash or some sort of pattern to denote this area.
Panel b) and Panel c) – I realize that if I zoom in the legend is more visible, however, it is difficult to see at 100% zoom. The legend is also identical for both panels. Therefore, one larger legend with font size increased would be an improvement.
Figure 4. – This shows the setup for one SMRT simulation; however, nothing is marked as the SMRT layers used. Would this be possible to provide? Addresses my previous comment as well as providing some sort of example relate to the number of SMRT layers used.
Figure 5/6. – I think these are good figures illustrating the change in variables across the campaign. However, the text is not readable. Suggested changes: the different stages of snow melt could be included at the stop of only one pane as they are identical throughout the rest of the figure. Ideally the vertical lines could also be colour coded. Dates can be provided on only one pane as they are identical across. Also, font size is okay but a little small.
Figure 8 – Only one y-axis is needed between the figures. This would allow the size to be increased and increase readability.
Table 3 – Table layout is somewhat difficult to read. Horizontal outlines would aid in the interpretation of the data.
Figure 9 – This is a great figure. However, it is difficult to read. Two suggestions, 1) accompany the figure with a table which includes the specific explanations that way the table would be less busy, 2) use numbers or letters to point to specific events rather than having the arrows across the figure. This would again aid in improving the readability. Currently it takes too long to interpret it.
Table 4 – Similar comment to table 3.
Figure 11 – The bottom pane on this figure is unnecessary – it does not add anything to the interpretation. Simply providing the LWC for the top layer would be sufficient. Also, the LWC legend is unnecessary on the bottom. The RMSH legend is good.
Citation: https://doi.org/10.5194/egusphere-2025-974-RC1 - AC2: 'Reply on RC1', Francesca Carletti, 15 Jun 2025
-
RC2: 'Comment on egusphere-2025-974', Anonymous Referee #2, 20 May 2025
General comments/Main issues:
The manuscript addresses the temporal evolution of C-band backscatter intensity in high-Alpine terrain and its relation physical properties of the snow cover, in order to evaluate the feasibility for monitoring various phases of snowpack melting by means of satellite-borne SAR systems. This is a topic of great interest for hydrology and water resources management in regions for which the water supply relies on snowmelt runoff. On the sensor-side the investigations are based on data acquired by the C-band SAR sensor of the Copernicus-1 mission, a data base that has been used for operational monitoring of snowmelt area extent since several years.
The overall study approach is promising, comprising comprehensive, detailed field observations on physical snow properties, the analysis of Sentinel-1 backscatter time series during two snow cover seasons with particular focus on the snowmelt period, and the modelling of backscatter intensity. The presented material provides important new insights on the evolution of Alpine snow cover during different snowmelt phases and the related response of radar backscatter signals, offering essential information for advancing applications of satellite-based SAR observations for monitoring specific properties of the snow cover during the snowmelt period.
The completeness and traceability of the presentation differs, depending on the topic. The dense time sequence of very detailed snow property measurements during two snow seasons and the related analysis and discussion are a strong part of the study. On the other hand, the analysis and interpretation of the Sentinel-1 (S1) backscatter signals as well as the backscatter modelling part have weaknesses. Regarding S1 backscatter signatures, two main issues are: (i) the radiometric performance of the S1 backscatter intensity values of the rather confined cell used for signature analysis and validation tasks; (ii) the representativeness of the backscatter values of this reference cell in respect to backscatter in its immediate surroundings and to the typical backscatter signals of undisturbed snow areas in the same elevation zone.
- The radiometric accuracy for the backscatter data of the reference cell needs to be specified in detail. The reference to Marin et al. (2020) does not provide specific information on radiometric uncertainty for this particular cell of 20 m x 20 m extent. In S1 IW mode data it covers only about 4 independent samples (looks), resulting in high speckle-related uncertainty. Marin et al. apply for speckle reduction a multitemporal filter of 11 x 11 pixels and a gamma-MAP filter of 3 x 3 pixels. In both cases the radiometry is preserved if the window covers a homogeneous distributed target. This is not the case in the area surrounding of the refence cell. For the multitemporal filter different temporal response within the window may introduce additional radiometric biases for sigma-0 of individual pixels and dates.
- The representativeness of the S1 signatures of the reference cell in respect to the area in its surroundings should be assessed by analysing also data of other cells, preferably covering some larger area. Considering Fig. 2, it is unclear why cell 39 is used as single reference case. The data of cell 39 do not show a distinct incidence angle dependence and track 117 shows a different behaviour in the two years. The data of cell 18 of the two years, for example, are consistent. According to the elevation contour lines, this cell is located on an east-facing slope, and the two tracks of the ascending orbit show lower sigma-0 values, as expected for a back-slope. This may offers a possibility for studying impacts of the incidence angle.
For relating the backscatter modelling results to specific melt phases and for interpretation of the observed backscatter signatures, first of all it is important providing information on the backscatter contributions of the individual snow layers in dependence of the liquid water content. The relative contributions to total backscatter in dependence of depth below the snow surface should be specified for typical snow profiles shown in Figs. 5 and 6. Another concern is the limited information regarding properties of the scattering elements in the individual layers and the parametrization of the dense medium effect. In particular for the early melt phases, the properties of the top snow layer can be quite different, depending on size and shape of water inclusions (Arslan et al., 2003). Furthermore, information on the impact of incidence angle dependence of backscatter would be of interest, being of relevance for assessing roughness effects of wet snow surfaces.
Further issues:
Introduction and Discussion: During three winters Strozzi and Mätzler (1997; 1998) performed at the same test field above Davos C- and Ka-band backscatter measurements. Reference 1 (1997) is briefly cited in the manuscript, reference 2 (1998) is not cited. Results of these measurements are relevant within the context of the work presented by Carletti et al. and key points should be mentioned. Among issues addressed in the two papers are impacts of surface roughness and refrozen snow crusts based on measurements at different incidence angles, and the response to liquid water content.
P4, L121: The sensitivity to snow wetness depends on the local incidence angle, not directly on slope steepness. On steep fore-slopes the sensitivity is low.
P8, L212: A cutter of 55 cm length may not be suitable for resolving the density differences between individual layers that may be quite thin during the different melt phases. Please explain the limits in vertical resolution.
P9, Table 1: The elevation contour lines of Fig. 2a indicate different slope steepness within the test field. Please explain to which points the cited incidence angles refer and show the overall range of angles for the test field.
P11, Fig: 2: Incidence angles of cells on sloping show lower sigma-0 for ascending orbits, as to be expected for back-slopes. Consequently, the difference in sigma-0 between individual tracks may offer the possibility for exploring incidence angle effects.
P12, L328: The characterization of snow microstructure is a critical issue for snow backscatter modelling. Exponential correlation functions have major deficiencies, in particular for multi-size and sticky cases (Chang et al., 2016). Furthermore, the phase functions of snow with liquid water inclusions are quite different from that of dry snow (Arslan et al., 2003).
P14, L363: In particular during a main part of winter 2023-2024 the base of the snowpack shows zero deg. temperature, implying unfrozen ground. This goes on throughout the snowmelt periods.
P16 Fig. 5: Between mid-April and early June 2023 sigma-0 of the two ascending tracks shows consistent differences of 3 to 5 dB. Please explain the reason. The high sigma-0 values are probably from track 117 which shows high variance in 2023-2024, differing from 2022-2023 (Fig. 2b).
P20, Table 2: Please specify the incidence angle. Besides, the validity of these numbers in respect to other incidence angles would be of interest.
P 27, Fig. 11: Please specify the incidence angle. Strozzi and Mätzler (1997) show the incidence angle dependence of backscatter of wet snow (for smooth and rough surfaces) based on backscatter measurements at the same test site.
P29, L625: The development of surface roughness after start of the snowmelt period depends also on the sequence and intensity of snowfall events, varying from year to year.
P29, L641ff: The parametrization of the scattering elements may as well be a reason for differences between recorded and modelled backscatter (see e.g. Arslan et al., 2003; Chang et al., 2016).
References:
Arslan, A.N. et al.: Scattering from wet snow applying strong fluctuation theory, J. Electromagnetic Waves and Applications, 17:7, 1009-1024, https://doi.org/10.1163/156939303322519081, 2003.
Chang, W., Ding, K.-H., Tsang, L., and Xu, X.: Microwave scattering and medium characterization for terrestrial snow with QCA-Mie and bicontinuous models: Comparison studies, IEEE Trans. Geosci. Remote Sens., 54 (6), 3637-3648, doi: 10.1109/TGRS.2016.2522438:, 2016.
Strozzi, T., Wiesmann, A., and Mätzler, C.: Active microwave signatures of snow covers at 5.3 and 35 GHz, Radio Science, 32(2), 479-495, 1997.
Strozzi, T. and Mätzler, C.: Backscattering measurements of Alpine snow covers at 5.3 and 35 GHz, IEEE Trans. Geosci. Remote Sens., 36(3), 838-848, 1998.
Citation: https://doi.org/10.5194/egusphere-2025-974-RC2 - AC1: 'Reply on RC2', Francesca Carletti, 15 Jun 2025
Data sets
SnowTinel high-temporal-resolution ground truth dataset for SAR remote sensing of snow Francesca Carletti et al. https://doi.org/10.16904/envidat.574
Interactive computing environment
multitemporal-s1-backscattering Francesca Carletti https://github.com/carlettif/multitemporal-s1-backscattering
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 435 | 131 | 23 | 589 | 16 | 25 |
- HTML: 435
- PDF: 131
- XML: 23
- Total: 589
- BibTeX: 16
- EndNote: 25
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1