the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Evaluating Snow Depth Retrievals from Sentinel-1 Volume Scattering over NASA SnowEx Sites
Abstract. Snow depth retrievals from spaceborne C-band synthetic aperture radar (SAR) backscatter have the potential to fill an important gap in the remote monitoring of seasonal snow. Sentinel-1 SAR data have been used previously in an empirical algorithm to generate snow depth products with near-global coverage, sub-weekly temporal resolution, and spatial resolutions on the order of hundreds of meters to 1 km. However, there has been no published independent validation of this algorithm. In this work we develop the first open-source software package that implements this Sentinel-1 snow depth retrieval algorithm as described in the original papers, and evaluate the snow depth retrievals against nine high-resolution lidar snow depth acquisitions collected during the winters of 2019–2020 and 2020–21 at six study sites across the western United States as part of the NASA SnowEx Mission. Across all sites, we find poor agreement between the Sentinel-1 snow depth retrievals and the lidar snow depth measurements, with a mean RMSE of 0.92 m and a mean Pearson correlation coefficient R of 0.46. Algorithm performance improves slightly in deeper snowpacks and at higher elevations. We further investigate the underlying Sentinel-1 data for a snow signal through an exploratory analysis of the cross-polarization backscatter ratio relative to lidar snow depths. We find a significant correlation between this cross ratio and snow depth over ~1.5 m but no relationship to a slight negative correlation for snow depths less than ~1.5 m. We attribute poor algorithm performance to a) the variable amount of apparent snow depth signal in the S1 cross ratio and b) an algorithm structure that does not adequately convert S1 backscatter signal to snow depth. Our findings provide an open-source frame work for future investigations, along with insight into the applicability of C-band SAR for snow depth retrievals and directions for future C-band snow depth retrieval algorithm development. C-band SAR has the potential to address gaps in radar monitoring of deep snowpacks; however, more research into retrieval algorithms is necessary to better understand the physical mechanisms and uncertainties of C-band volume scattering-based retrievals.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(19119 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(19119 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-1018', Anonymous Referee #1, 10 May 2024
review of paper egusphere-2024-1018, “Evaluating Snow Depth Retrievals from Sentinel-1 Volume Scattering over NASA SnowEx Sites”
Hoppinen et al.
General
A novel method to retrieve snow depth from C-band Sentinel-1 observations was previously introduced by Lievens et al. (2019). While potentially very valuable as a novel source of snow information, the study has also caused a degree of debate in the community due to the conventionally assumed insensitivity of radar backscatter at this frequency to snow accumulation. Although new theoretical analysis suggests that such sensitivity may be possible in certain cases indirectly through the sensitivity of cross-polarized backscatter to anisotropic snow structures, an independent validation of the method beyond some case studies has been largely lacking. The study by Hoppinen et al. attemps just this, providing a potentially valuable contribution to the community, also by giving access to an open-source software replicating the snow depth retrieval method.
The study itself is well written and clear. I find the presentation of results convincing and thorough. I recommend publication of the paper, after considering the following few minor suggestions.
Minor comments
- Abstract line 8: 2020-2021
- Abstract line 9 and throughout the paper; “poor agreement”. While I agree with the authors that the agreement is indeed poor, I would still suggest another way of pointing this out. Also, I’m not sure anyone has quantified what is “poor”… e.g. just stating the R value and nRMSE should be sufficient for readers to make their own conclusions. You can always cite requirements placed on e.g. nRMSE, as you have done in several places. Also you could cite values obtained with PMW (e.g. Mortimer et al., 2022). In the abstract you could just say the achieved accuracy “is considerably lower than requirements placed for remotely sensed observations of SD” or something similar.
- Introduction, line 36 “passive microwave measurements saturate” only applies to the 37 GHz frequency typically used in retrievals. Please reword.
- Introduction, around line 60. Here, it would be appropriate to briefly acknowledge the diverse methods SAR could be used to retrieve SD/SWE: 1) repeat-pass InSAR at low frequencies to obtain deltaSWE 2) single-pass InSAR (DEM differencing) to obtain SD 3) volume scattering approach to obtain SWE directly. Please add also a few appropriate references. You can then tie Lievens et al. more or less to the volume scattering approach, giving a motivation for section 1.1. Please also recap some of the difficulties associated with the volume scattering approach, namely the separation of ground and snow backscattering contributions, as well as the influence of snow microstructure.
- Section 1.1, line 75. Surface scattering contributions for dry snow should be very small compared to volume scatter and ground backscatter. This could be good to point out, perhaps with a reference?
- section 1.1 line 77 “Some SAR-based methods…” brings up the question of what other methods there are. See previous comment #4, please reword.
- section 1.2, lines 111 and 115, maybe elsewhere: suggest to change “in Lievens et al.” to “by Lievens et al.”
- Section 2.2 lines 163-164. Sentence seems out of place/complementary of what comes on lines 165-166. Remove?
- section 2.2 line 174 & elsewhere. I guess the parameters A, B and C are not actually dimensionless? delta SD should come out as meters?
- Section 2.2 line 187. I find the rather limited correlation surprising, it is a pity that apparently this could not be pursued further. I’m also a bit lost why the intercomparison between the products produces a R value of 0.64, when comparisons to lidar data are apparently very similar between the two… can you elaborate still on the comparisons to lidar data? Did this represent e.g. a subset of the product intercomparison? Maybe even some map comparisons between C-SNOW and your retrievals could be in order e.g. in the Appendix?
- Figure 3 panel b: I guess x-axis label could be just “Snow depth” since both lidar and S1 SDs are presented.
Citation: https://doi.org/10.5194/egusphere-2024-1018-RC1 - AC4: 'Reply on RC1', Zachary Hoppinen, 03 Jul 2024
-
RC2: 'Comment on egusphere-2024-1018', Anonymous Referee #2, 21 May 2024
Comment on “Evaluating Snow Depth Retrievals from Sentinel-1 VolumeScattering over NASA SnowEx Sites” by Z. Hoppinen et al.
General Comments:
This article reports on the evaluation of an algorithm for retrieval of snow depth based on C-band backscatter intensity images of the Sentinel-1 mission. The first version of the retrieval algorithm, presented by Lievens et al. (2019), applied an empirical change detection method using temporal changes of the cross- to co- polarization backscatter ratio (VH/VV) to compute a snow index that is rescaled by means of empirical parameters in order to obtain the snow depth. A modified version of the snow depth algorithm was applied for snow depth retrievals over the European Alps (Lievens et al., 2022). This version of the algorithm employs the VH/VV ratio and the VV backscatter intensity, accounts for the fractional forest cover, and uses several empirical scaling and weighting parameters. A critical constituent of the algorithm is the determination of empirical scaling factors that relate the backscatter intensity to snow depth.
A comprehensive independent validation of the algorithm has been lacking by now. The work by Hoppinen at al. addresses this open issue, evaluating the performance of Sentinel-1 snow depth maps derived by means of the 2022 version of the algorithm. Reference data for performance assessment are available from nine high-resolution airborne lidar acquisitions over extended study sites in the Western US, an excellent data set for algorithm validation. The scaling factors are derived from a subset of these data by optimizing the correlation coefficient and minimizing the mean absolute error.
The paper is a valuable contribution to the topic of SAR-based approaches for mapping snow depth and snow mass. It addresses a very relevant question in this context. The data analysis, results and conclusions are well described and conclusive. However, there are still some issues to be checked and clarified, addressed below. In particular Section 1.1 on the theoretical background needs major revision. Besides, information on the available Sentinel-1 coverage (orbits, repeat coverage) would be of interest.
Specific Comments:
Section 1.1 SAR volume scattering snow depth retrieval theory:
This section refers to the radar signal interaction with snow. The description of processes having an impact on the C-band backscatter signal for snow over ground is incomplete and lacks quantitative information. Volume scattering cannot be considered as a stand-alone process for deriving physical snow parameters from backscatter intensity. In support of interpretation and discussion of the results of the study, I recommend including a concise description of the main contributions to the observed backscatter signal (including not only snow, but also ground and vegetation) and their impact on retrievals of snow depth. Because large parts of the study sites are covered by forest, the impact of forests on C-band signal propagation should be addressed. Figure 1 needs to be revised. The information to be conveyed by this figure is unclear. It implies that the incoming radar signal is reflected within the snow volume as a main source and the signal increases directly with increasing snow depth. Besides, the figure shows incoming and reflected beams in bistatic configuration.
Line 12 (Abstract): the term “cross-polarization backscatter ratio” should be defined at its first usage.
Line 13 (Abstract): Please provide a number for “significant correlation”
Line 99, 100: “as new snow increases the cross-polarized energy that is backscattered toward the sensor” This is not in accordance with experimental data and theory. Due to the low C-band scattering albedo of fresh snow the backscatter signal of a medium below with higher scattering albedo (e.g. coarse-grained metamorphic snow, refrozen snow) is attenuated when propagating through fresh snow.
Line 125: Please check these two references: “Frerebeau et al., 2023; Lebrun et al., 2020”; both refer to “Dose Rate Estimation from in-Situ Gamma-Ray Spectrometry” and not to SAR.
Line 126: “European Space Agency, 2021” missing in the reference list
Line 130: The speckle-related uncertainty of the selected grid size would be of interest.
Line 133 to 135: The backscatter intensity at different incidence angles is not an “artifact” but contains relevant information related to physical properties of a medium which is suppressed of data with different incidence angles are merged.
Line 175 to 177: According to this information a single S1 image per lidar acquisition data set (a subset of the total S1 data set) is used for deriving optimized scaling parameters. However, the snow depth retrieval algorithm is not based on single images but on changes of backscatter intensity in time over an extended period (Appendix A).
Line 208, 209: From the histograms in Fig. 3b, the agreement between the medians of these three sites is not obvious (due to the log-scaling). Besides, Banner 2020 and Fraser 2020 show a rather high negative bias for average snow depth (Table 2).
Line 209: The acronyms “ICC”, and in Fig 3b “LCC”, refer probably to Little Cottonwood?
Fig. 3a: The linear trendline (on which the correlation coefficient is based) should be included, as its slope is an indication for the S1 sensitivity in respect to snow depth.
Line 240 to 244 and Fig.7: The SNOTEL snow depth times series should be compared to the Sentinel-1 CR data of areas in the vicinity of the SNOTEL stations rather than to the site-wide mean cross ratio. Surface elevation has a major impact on the state of the snowpack and its backscatter properties.
Line 245 to 248 and Fig. 8: The two classes with deep snow (2.5-3, 3+) showing a distinct rise in delta-CR comprise only 1.2 % of the total sample. Hardly a suitable basis for a statistically significant conclusions regarding the retrieval performance for deep snow.
Page 14, Fig. 4 caption: the labels for FC and elevation in the figures and caption are mixed up.
Page 15, Fig. 5: These histograms are probably also log-scaled.
Line 252: The correlation coefficient is not a suitable parameter for assessing the performance of the spatial distribution.
Line 255: “Frasier” typo
Line 261-262: Fig. 6a shows a decrease of the relative error with lidar snow depth, but the absolute error increases with snow depth for snow depth > 1 m. Also, in Fig. 5a the S1 and lidar histogram for snow depth > 2 m show the largest disparity. This is not a clear evidence for improved performance for deep snow.
Line 267: “SAR signals primarily interact with layers within the snowpack rather than individual snow grains”. Please check this statement. The scattering elements within layers are grains and grain clusters.
Line 267-271: Experimental data on propagation losses show for C-band power penetration length in dry seasonal snow typical numbers in excess of 10 m. Consequently, backscatter contributions of the subnivean ground and snow/ground interaction play also a role.
Line 304-305: The Sentinel-1 orbit accuracy is very high so that orbit errors do not play any role. For estimating the impact of SAR speckle, estimates for the speckle-related uncertainty would be useful.
Line 317-318: “ ... S1 snow retrievals agree best with lidar snow depth measurements in regions with snow packs deeper than 1.5 m …” this refers to the local snow depth relative to the mean value, not to the magnitude of snow depth. This should be stated. See comment line 261.
Line 321-323: Ku-band and X-band are mentioned here. Why not L-band, for which the InSAR phase delay is applicable for SWE retrievals also in deep snow?
Line 324 ff, Section 4.2: The analysis of the CR time series suffers from the spatial disparity between point measurements (SNOTEL) and spatial average S1 data of large test sites extending over different elevation zones (see comment line 240ff).
Line 373 to 376: A conclusive time series analysis of CR is lacking. See the comment above.
Line 375: Please provide numbers for “significant relationship”.
Appendix B:
Line 459, Table B1: Units?
Line 466, 467: “Since C simply scales values in the final step of the retrieval, this parameter can be optimized efficiently and should be adjusted first when applying this technique at a new site”. This indicates that for any site an individual calibration of the scaling parameters is needed to obtain useful results. To this end representative and reliable snow depth data from other sources are needed. Furthermore, due to interannual changes in permittivity and structural properties of the snow cover ground, the value of the scaling parameter may change from year to year. This questions the feasibility of the retrieval approach for regular applications over extended areas.
Line 484, Fig. B1: Please check the sign for the % change in snow depth in Figs a, b, c. Would not C = 0.0 result in zero snow depth (-100%), rather than in +100% ?
Page 28, Fig.B2: Figs. B2a and B2c show low RMSE for the wet snow threshold -1 dB and high RMSE for the threshold -3dB. As a lower wet snow threshold reduces the number of misclassifications for dry snow, the opposite behaviour is expected.
Citation: https://doi.org/10.5194/egusphere-2024-1018-RC2 - AC1: 'Reply on RC2', Zachary Hoppinen, 03 Jul 2024
-
RC3: 'Comment on egusphere-2024-1018', Benoit Montpetit, 27 May 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-1018/egusphere-2024-1018-RC3-supplement.pdf
-
AC3: 'Reply on RC3', Zachary Hoppinen, 03 Jul 2024
We thank Dr. Montpetit for their insightful comments. The comments were thorough and helpful and have certainly improved the writing and analysis. The suggestion to include delta VV and VH in the boxplots was especially helpful and added to our analysis. We have highlighted changes made to the manuscript below in response to the reviewer’s suggestions.
-
AC3: 'Reply on RC3', Zachary Hoppinen, 03 Jul 2024
-
RC4: 'Comment on egusphere-2024-1018', Anonymous Referee #4, 30 May 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-1018/egusphere-2024-1018-RC4-supplement.pdf
- AC2: 'Reply on RC4', Zachary Hoppinen, 03 Jul 2024
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-1018', Anonymous Referee #1, 10 May 2024
review of paper egusphere-2024-1018, “Evaluating Snow Depth Retrievals from Sentinel-1 Volume Scattering over NASA SnowEx Sites”
Hoppinen et al.
General
A novel method to retrieve snow depth from C-band Sentinel-1 observations was previously introduced by Lievens et al. (2019). While potentially very valuable as a novel source of snow information, the study has also caused a degree of debate in the community due to the conventionally assumed insensitivity of radar backscatter at this frequency to snow accumulation. Although new theoretical analysis suggests that such sensitivity may be possible in certain cases indirectly through the sensitivity of cross-polarized backscatter to anisotropic snow structures, an independent validation of the method beyond some case studies has been largely lacking. The study by Hoppinen et al. attemps just this, providing a potentially valuable contribution to the community, also by giving access to an open-source software replicating the snow depth retrieval method.
The study itself is well written and clear. I find the presentation of results convincing and thorough. I recommend publication of the paper, after considering the following few minor suggestions.
Minor comments
- Abstract line 8: 2020-2021
- Abstract line 9 and throughout the paper; “poor agreement”. While I agree with the authors that the agreement is indeed poor, I would still suggest another way of pointing this out. Also, I’m not sure anyone has quantified what is “poor”… e.g. just stating the R value and nRMSE should be sufficient for readers to make their own conclusions. You can always cite requirements placed on e.g. nRMSE, as you have done in several places. Also you could cite values obtained with PMW (e.g. Mortimer et al., 2022). In the abstract you could just say the achieved accuracy “is considerably lower than requirements placed for remotely sensed observations of SD” or something similar.
- Introduction, line 36 “passive microwave measurements saturate” only applies to the 37 GHz frequency typically used in retrievals. Please reword.
- Introduction, around line 60. Here, it would be appropriate to briefly acknowledge the diverse methods SAR could be used to retrieve SD/SWE: 1) repeat-pass InSAR at low frequencies to obtain deltaSWE 2) single-pass InSAR (DEM differencing) to obtain SD 3) volume scattering approach to obtain SWE directly. Please add also a few appropriate references. You can then tie Lievens et al. more or less to the volume scattering approach, giving a motivation for section 1.1. Please also recap some of the difficulties associated with the volume scattering approach, namely the separation of ground and snow backscattering contributions, as well as the influence of snow microstructure.
- Section 1.1, line 75. Surface scattering contributions for dry snow should be very small compared to volume scatter and ground backscatter. This could be good to point out, perhaps with a reference?
- section 1.1 line 77 “Some SAR-based methods…” brings up the question of what other methods there are. See previous comment #4, please reword.
- section 1.2, lines 111 and 115, maybe elsewhere: suggest to change “in Lievens et al.” to “by Lievens et al.”
- Section 2.2 lines 163-164. Sentence seems out of place/complementary of what comes on lines 165-166. Remove?
- section 2.2 line 174 & elsewhere. I guess the parameters A, B and C are not actually dimensionless? delta SD should come out as meters?
- Section 2.2 line 187. I find the rather limited correlation surprising, it is a pity that apparently this could not be pursued further. I’m also a bit lost why the intercomparison between the products produces a R value of 0.64, when comparisons to lidar data are apparently very similar between the two… can you elaborate still on the comparisons to lidar data? Did this represent e.g. a subset of the product intercomparison? Maybe even some map comparisons between C-SNOW and your retrievals could be in order e.g. in the Appendix?
- Figure 3 panel b: I guess x-axis label could be just “Snow depth” since both lidar and S1 SDs are presented.
Citation: https://doi.org/10.5194/egusphere-2024-1018-RC1 - AC4: 'Reply on RC1', Zachary Hoppinen, 03 Jul 2024
-
RC2: 'Comment on egusphere-2024-1018', Anonymous Referee #2, 21 May 2024
Comment on “Evaluating Snow Depth Retrievals from Sentinel-1 VolumeScattering over NASA SnowEx Sites” by Z. Hoppinen et al.
General Comments:
This article reports on the evaluation of an algorithm for retrieval of snow depth based on C-band backscatter intensity images of the Sentinel-1 mission. The first version of the retrieval algorithm, presented by Lievens et al. (2019), applied an empirical change detection method using temporal changes of the cross- to co- polarization backscatter ratio (VH/VV) to compute a snow index that is rescaled by means of empirical parameters in order to obtain the snow depth. A modified version of the snow depth algorithm was applied for snow depth retrievals over the European Alps (Lievens et al., 2022). This version of the algorithm employs the VH/VV ratio and the VV backscatter intensity, accounts for the fractional forest cover, and uses several empirical scaling and weighting parameters. A critical constituent of the algorithm is the determination of empirical scaling factors that relate the backscatter intensity to snow depth.
A comprehensive independent validation of the algorithm has been lacking by now. The work by Hoppinen at al. addresses this open issue, evaluating the performance of Sentinel-1 snow depth maps derived by means of the 2022 version of the algorithm. Reference data for performance assessment are available from nine high-resolution airborne lidar acquisitions over extended study sites in the Western US, an excellent data set for algorithm validation. The scaling factors are derived from a subset of these data by optimizing the correlation coefficient and minimizing the mean absolute error.
The paper is a valuable contribution to the topic of SAR-based approaches for mapping snow depth and snow mass. It addresses a very relevant question in this context. The data analysis, results and conclusions are well described and conclusive. However, there are still some issues to be checked and clarified, addressed below. In particular Section 1.1 on the theoretical background needs major revision. Besides, information on the available Sentinel-1 coverage (orbits, repeat coverage) would be of interest.
Specific Comments:
Section 1.1 SAR volume scattering snow depth retrieval theory:
This section refers to the radar signal interaction with snow. The description of processes having an impact on the C-band backscatter signal for snow over ground is incomplete and lacks quantitative information. Volume scattering cannot be considered as a stand-alone process for deriving physical snow parameters from backscatter intensity. In support of interpretation and discussion of the results of the study, I recommend including a concise description of the main contributions to the observed backscatter signal (including not only snow, but also ground and vegetation) and their impact on retrievals of snow depth. Because large parts of the study sites are covered by forest, the impact of forests on C-band signal propagation should be addressed. Figure 1 needs to be revised. The information to be conveyed by this figure is unclear. It implies that the incoming radar signal is reflected within the snow volume as a main source and the signal increases directly with increasing snow depth. Besides, the figure shows incoming and reflected beams in bistatic configuration.
Line 12 (Abstract): the term “cross-polarization backscatter ratio” should be defined at its first usage.
Line 13 (Abstract): Please provide a number for “significant correlation”
Line 99, 100: “as new snow increases the cross-polarized energy that is backscattered toward the sensor” This is not in accordance with experimental data and theory. Due to the low C-band scattering albedo of fresh snow the backscatter signal of a medium below with higher scattering albedo (e.g. coarse-grained metamorphic snow, refrozen snow) is attenuated when propagating through fresh snow.
Line 125: Please check these two references: “Frerebeau et al., 2023; Lebrun et al., 2020”; both refer to “Dose Rate Estimation from in-Situ Gamma-Ray Spectrometry” and not to SAR.
Line 126: “European Space Agency, 2021” missing in the reference list
Line 130: The speckle-related uncertainty of the selected grid size would be of interest.
Line 133 to 135: The backscatter intensity at different incidence angles is not an “artifact” but contains relevant information related to physical properties of a medium which is suppressed of data with different incidence angles are merged.
Line 175 to 177: According to this information a single S1 image per lidar acquisition data set (a subset of the total S1 data set) is used for deriving optimized scaling parameters. However, the snow depth retrieval algorithm is not based on single images but on changes of backscatter intensity in time over an extended period (Appendix A).
Line 208, 209: From the histograms in Fig. 3b, the agreement between the medians of these three sites is not obvious (due to the log-scaling). Besides, Banner 2020 and Fraser 2020 show a rather high negative bias for average snow depth (Table 2).
Line 209: The acronyms “ICC”, and in Fig 3b “LCC”, refer probably to Little Cottonwood?
Fig. 3a: The linear trendline (on which the correlation coefficient is based) should be included, as its slope is an indication for the S1 sensitivity in respect to snow depth.
Line 240 to 244 and Fig.7: The SNOTEL snow depth times series should be compared to the Sentinel-1 CR data of areas in the vicinity of the SNOTEL stations rather than to the site-wide mean cross ratio. Surface elevation has a major impact on the state of the snowpack and its backscatter properties.
Line 245 to 248 and Fig. 8: The two classes with deep snow (2.5-3, 3+) showing a distinct rise in delta-CR comprise only 1.2 % of the total sample. Hardly a suitable basis for a statistically significant conclusions regarding the retrieval performance for deep snow.
Page 14, Fig. 4 caption: the labels for FC and elevation in the figures and caption are mixed up.
Page 15, Fig. 5: These histograms are probably also log-scaled.
Line 252: The correlation coefficient is not a suitable parameter for assessing the performance of the spatial distribution.
Line 255: “Frasier” typo
Line 261-262: Fig. 6a shows a decrease of the relative error with lidar snow depth, but the absolute error increases with snow depth for snow depth > 1 m. Also, in Fig. 5a the S1 and lidar histogram for snow depth > 2 m show the largest disparity. This is not a clear evidence for improved performance for deep snow.
Line 267: “SAR signals primarily interact with layers within the snowpack rather than individual snow grains”. Please check this statement. The scattering elements within layers are grains and grain clusters.
Line 267-271: Experimental data on propagation losses show for C-band power penetration length in dry seasonal snow typical numbers in excess of 10 m. Consequently, backscatter contributions of the subnivean ground and snow/ground interaction play also a role.
Line 304-305: The Sentinel-1 orbit accuracy is very high so that orbit errors do not play any role. For estimating the impact of SAR speckle, estimates for the speckle-related uncertainty would be useful.
Line 317-318: “ ... S1 snow retrievals agree best with lidar snow depth measurements in regions with snow packs deeper than 1.5 m …” this refers to the local snow depth relative to the mean value, not to the magnitude of snow depth. This should be stated. See comment line 261.
Line 321-323: Ku-band and X-band are mentioned here. Why not L-band, for which the InSAR phase delay is applicable for SWE retrievals also in deep snow?
Line 324 ff, Section 4.2: The analysis of the CR time series suffers from the spatial disparity between point measurements (SNOTEL) and spatial average S1 data of large test sites extending over different elevation zones (see comment line 240ff).
Line 373 to 376: A conclusive time series analysis of CR is lacking. See the comment above.
Line 375: Please provide numbers for “significant relationship”.
Appendix B:
Line 459, Table B1: Units?
Line 466, 467: “Since C simply scales values in the final step of the retrieval, this parameter can be optimized efficiently and should be adjusted first when applying this technique at a new site”. This indicates that for any site an individual calibration of the scaling parameters is needed to obtain useful results. To this end representative and reliable snow depth data from other sources are needed. Furthermore, due to interannual changes in permittivity and structural properties of the snow cover ground, the value of the scaling parameter may change from year to year. This questions the feasibility of the retrieval approach for regular applications over extended areas.
Line 484, Fig. B1: Please check the sign for the % change in snow depth in Figs a, b, c. Would not C = 0.0 result in zero snow depth (-100%), rather than in +100% ?
Page 28, Fig.B2: Figs. B2a and B2c show low RMSE for the wet snow threshold -1 dB and high RMSE for the threshold -3dB. As a lower wet snow threshold reduces the number of misclassifications for dry snow, the opposite behaviour is expected.
Citation: https://doi.org/10.5194/egusphere-2024-1018-RC2 - AC1: 'Reply on RC2', Zachary Hoppinen, 03 Jul 2024
-
RC3: 'Comment on egusphere-2024-1018', Benoit Montpetit, 27 May 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-1018/egusphere-2024-1018-RC3-supplement.pdf
-
AC3: 'Reply on RC3', Zachary Hoppinen, 03 Jul 2024
We thank Dr. Montpetit for their insightful comments. The comments were thorough and helpful and have certainly improved the writing and analysis. The suggestion to include delta VV and VH in the boxplots was especially helpful and added to our analysis. We have highlighted changes made to the manuscript below in response to the reviewer’s suggestions.
-
AC3: 'Reply on RC3', Zachary Hoppinen, 03 Jul 2024
-
RC4: 'Comment on egusphere-2024-1018', Anonymous Referee #4, 30 May 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-1018/egusphere-2024-1018-RC4-supplement.pdf
- AC2: 'Reply on RC4', Zachary Hoppinen, 03 Jul 2024
Peer review completion
Journal article(s) based on this preprint
Data sets
Sentinel-1 Derived Snow Depths and SnowEx Lidar Netcdfs Zachary Hoppinen, Ross T. Palomaki, George Brencher, Devon Dumire, Eric Gagliano, Adrian Marziliano, Jack Tarricone, and Hans-Peter Marshall https://zenodo.org/records/10913396
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
658 | 270 | 36 | 964 | 14 | 15 |
- HTML: 658
- PDF: 270
- XML: 36
- Total: 964
- BibTeX: 14
- EndNote: 15
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1
Ross T. Palomaki
George Brencher
Devon Dunmire
Eric Gagliano
Adrian Marziliano
Jack Tarricone
Hans-Peter Marshall
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(19119 KB) - Metadata XML