the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 May 2024
Spatio-temporal snow data assimilation with the ICESat-2 laser altimeter
Abstract. The satellite laser altimeter ICESat-2 provides accurate surface elevation observations across the globe. Where a high-resolution DEM is available, we can use these measurements to retrieve snow depth profiles even in areas where snow amounts are poorly constrained, despite being of great societal interest. However, the adoption of these retrievals remains low since they are very sparse in space (the satellite measures along profiles) and in time (the revisit is 3 months). Data assimilation methods can exploit snow observations to constrain snow models and provide gap-free snow map time series. Assimilation of observations like snow cover is established, but there are currently no methods to assimilate sparse ICESat-2 snow depth profiles. We propose an approach that spatially propagates information using – instead of the classic geographical distance – an abstract distance measured in a feature space defined by topographical parameters and the melt-out climatology.
We demonstrate this framework for a small experimental catchment in the Spanish Pyrenees through three experiments. We assimilate different observations in an intermediate-complexity snow model: fractional snow cover retrievals from Sentinel-2, snow depth profiles from ICESat-2 located in the proximity of the catchment, or both snow cover and depth in a joint assimilation experiment. Results show that assimilating ICESat-2 snow depth profiles successfully updates the neighboring unobserved catchment, improving the simulated average snow depth compared to the prior run. Moreover, adding the snow depth profiles to fractional snow-covered area observations leads to an accurate reconstruction of the snow depth spatial distribution, improving the skill score by 22 %.
-
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
(2058 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2058 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-1404', Anonymous Referee #1, 10 Jul 2024
- AC1: 'Reply on RC1', Marco Mazzolini, 06 Sep 2024
-
RC2: 'Comment on egusphere-2024-1404', Anonymous Referee #2, 12 Jul 2024
This study investigates using ICESat-2 satellite data to improve FSM2 simulations. The authors employ the DA proposed by Alonso-Gonzales et al. (2023) with a spatial propagation of the sparse data points from ICESat-2 that tries to compensate the fact that ICESat-2 data are acquired in profiles with many temporal and spatial gaps. They perform three experiments to assess the effectiveness of assimilating different data types snow cover area, from Sentinel-2, snow depth, from ICESat-2 or both. The reported findings indicate that by incorporating snow cover area data alongside snow depth from ICESat-2 led to the most accurate snowpack simulations.
The authors leverage a data assimilation system (MuSA) from a previous work (Alonso-González et al., 2023). Moreover, the methodology for spatializing the ICESat-2 data (a point of significant interest) builds upon concepts presented in the previous work, but without substantial further development. While the initial findings are interesting, their persuasiveness could be strengthened through further analysis. This deeper exploration would allow the research to culminate in a more robust and impactful paper. To strengthen the paper focus and impact, I recommend the following revisions:
- While the current findings are interesting, further analysis could significantly enhance their persuasiveness. Consider incorporating other inputs data that only ERA-5 (also derived from in-situ) and also move to other (larger) catchment where distributed HS are available (e.g., ASO data if ICESat-2 data are available or Dischma in Switzerland) to solidify the results.
- Revisit the scientific questions the paper aims to address. Sharpening these questions will guide the research and ensure the experiments directly address them.
- Section 3.5 appears to hold the core of the paper contribution. Dedicating more space and development to this section would allow for a more thorough exploration and potentially lead to more impactful conclusions.
Detailed comments
- The introduction of the paper could benefit from being condensed and sharpened. Focus on presenting the key scientific questions, the research aims to answer, and clearly outlining the paper main novelty (difference with previous works). This will ensure the experiments directly target those questions and guide the research direction. The core innovation of the paper lies in applying the DA method (from Alonso-Gonzales et al. 2023) to ICESat-2 data. The unique approach for propagating spatial and climatological information holds significant promise. However, further development of this methodology is necessary, particularly regarding the justification for using data outside the area of interest for analyzing snow accumulation and redistribution (see next points).
- Figure 3, potentially the paper core novelty, requires a more detailed explanation. From the scatterplot, it appears there might be a weak correlation between snow depth and CSMD (and TPI24). However, the relationship between snow depth and Sx200 seems less clear, potentially indicating no significant correlation. At least this is my understanding with the provided text. If this is not correct, I suggest a clearer description to enhance reader comprehension explicitly guide them through the correct explanation.
- Please revise the text from L276-282 to make more clear (and less compressed).
- The paper relies solely on ERA-5 data for atmospheric forcing. While ERA-5 is a valuable product, acknowledging the existence of other models with potentially significant output variability (up to 100%) would strengthen the main message of the paper that ICESat-2 data can be useful and in which situation. A discussion on why ERA-5 was chosen over other options would be beneficial. However, for a more robust understanding of the proposed method I suggest incorporating data from at least a couple of additional models is recommended. This comparative analysis would highlight the method sensitivity to different forcing data. In particular, given the small target catchment area ( and the high target resolution of 20m), exploring the use of spatially distributed data from nearby in-situ stations could be highly valuable. This would provide a more realistic scenario: not sure the first choice to simulate 5ha at a resolution of 20m in an experimental catchment in a European mountain range is starting from a 30km ERA-5 data.
- The results of experiment D are puzzling (at least to me in the present form). While Figure 3 (and related text) suggests that ICESat-2 data captures the relationship between SD distribution, topography, and climatology using this data alone in experiment D appears to yield inaccurate snow patterns, whereas it helps in experiment J when used together with Sentinel-2, what is the main mechanism behind this behavior?
- A crucial evaluation metric for DA methods is computational time. The paper should explicitly report and analyze this metric, ideally providing a detailed profile for each operational step.
- Fig 7 can you add the drone maps? Beside demonstrating a significant improvement in an accuracy score, the scientific community is starting to become interested in understanding how realistic snow distributions become when observations at high resolution are assimilated into models. The paper could benefit from a stronger emphasis on this aspect.
- While Figure 7 presents the CRPS for various scenarios, a deeper analysis could help isolate the contribution of ICESat-2 data. The similarity between the CRPS with ICESat-2 and the prior suggests limited influence of ICESat-2. However, the simultaneous improvement in J (potentially reflecting the contribution of ICESat-2) is intriguing and puzzling at the same time. Can you better comment on this?
- L430: the fact that SCA assimilation doesn't improve the simulation during the accumulation period could be attributed to the fact that the area of interest is having 100% snow cover?
- L439: this does not seem true to me (at least in the present form).
- L442: this is in contradiction with L253 where it is stated that the snow depth is strongly governed by topography, which is also the main hypothesis why the proposed approach has been applied. This ping pong effects makes it challenging to understand the overall benefit of ICESat-2 data (and the validity of the presented results).
- L452: speculative, please revise it.
- L484: 20 m is not hyper resolution.
- References are generally ordered alphabetically.
Citation: https://doi.org/10.5194/egusphere-2024-1404-RC2 - AC2: 'Reply on RC2', Marco Mazzolini, 06 Sep 2024
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-1404', Anonymous Referee #1, 10 Jul 2024
- AC1: 'Reply on RC1', Marco Mazzolini, 06 Sep 2024
-
RC2: 'Comment on egusphere-2024-1404', Anonymous Referee #2, 12 Jul 2024
This study investigates using ICESat-2 satellite data to improve FSM2 simulations. The authors employ the DA proposed by Alonso-Gonzales et al. (2023) with a spatial propagation of the sparse data points from ICESat-2 that tries to compensate the fact that ICESat-2 data are acquired in profiles with many temporal and spatial gaps. They perform three experiments to assess the effectiveness of assimilating different data types snow cover area, from Sentinel-2, snow depth, from ICESat-2 or both. The reported findings indicate that by incorporating snow cover area data alongside snow depth from ICESat-2 led to the most accurate snowpack simulations.
The authors leverage a data assimilation system (MuSA) from a previous work (Alonso-González et al., 2023). Moreover, the methodology for spatializing the ICESat-2 data (a point of significant interest) builds upon concepts presented in the previous work, but without substantial further development. While the initial findings are interesting, their persuasiveness could be strengthened through further analysis. This deeper exploration would allow the research to culminate in a more robust and impactful paper. To strengthen the paper focus and impact, I recommend the following revisions:
- While the current findings are interesting, further analysis could significantly enhance their persuasiveness. Consider incorporating other inputs data that only ERA-5 (also derived from in-situ) and also move to other (larger) catchment where distributed HS are available (e.g., ASO data if ICESat-2 data are available or Dischma in Switzerland) to solidify the results.
- Revisit the scientific questions the paper aims to address. Sharpening these questions will guide the research and ensure the experiments directly address them.
- Section 3.5 appears to hold the core of the paper contribution. Dedicating more space and development to this section would allow for a more thorough exploration and potentially lead to more impactful conclusions.
Detailed comments
- The introduction of the paper could benefit from being condensed and sharpened. Focus on presenting the key scientific questions, the research aims to answer, and clearly outlining the paper main novelty (difference with previous works). This will ensure the experiments directly target those questions and guide the research direction. The core innovation of the paper lies in applying the DA method (from Alonso-Gonzales et al. 2023) to ICESat-2 data. The unique approach for propagating spatial and climatological information holds significant promise. However, further development of this methodology is necessary, particularly regarding the justification for using data outside the area of interest for analyzing snow accumulation and redistribution (see next points).
- Figure 3, potentially the paper core novelty, requires a more detailed explanation. From the scatterplot, it appears there might be a weak correlation between snow depth and CSMD (and TPI24). However, the relationship between snow depth and Sx200 seems less clear, potentially indicating no significant correlation. At least this is my understanding with the provided text. If this is not correct, I suggest a clearer description to enhance reader comprehension explicitly guide them through the correct explanation.
- Please revise the text from L276-282 to make more clear (and less compressed).
- The paper relies solely on ERA-5 data for atmospheric forcing. While ERA-5 is a valuable product, acknowledging the existence of other models with potentially significant output variability (up to 100%) would strengthen the main message of the paper that ICESat-2 data can be useful and in which situation. A discussion on why ERA-5 was chosen over other options would be beneficial. However, for a more robust understanding of the proposed method I suggest incorporating data from at least a couple of additional models is recommended. This comparative analysis would highlight the method sensitivity to different forcing data. In particular, given the small target catchment area ( and the high target resolution of 20m), exploring the use of spatially distributed data from nearby in-situ stations could be highly valuable. This would provide a more realistic scenario: not sure the first choice to simulate 5ha at a resolution of 20m in an experimental catchment in a European mountain range is starting from a 30km ERA-5 data.
- The results of experiment D are puzzling (at least to me in the present form). While Figure 3 (and related text) suggests that ICESat-2 data captures the relationship between SD distribution, topography, and climatology using this data alone in experiment D appears to yield inaccurate snow patterns, whereas it helps in experiment J when used together with Sentinel-2, what is the main mechanism behind this behavior?
- A crucial evaluation metric for DA methods is computational time. The paper should explicitly report and analyze this metric, ideally providing a detailed profile for each operational step.
- Fig 7 can you add the drone maps? Beside demonstrating a significant improvement in an accuracy score, the scientific community is starting to become interested in understanding how realistic snow distributions become when observations at high resolution are assimilated into models. The paper could benefit from a stronger emphasis on this aspect.
- While Figure 7 presents the CRPS for various scenarios, a deeper analysis could help isolate the contribution of ICESat-2 data. The similarity between the CRPS with ICESat-2 and the prior suggests limited influence of ICESat-2. However, the simultaneous improvement in J (potentially reflecting the contribution of ICESat-2) is intriguing and puzzling at the same time. Can you better comment on this?
- L430: the fact that SCA assimilation doesn't improve the simulation during the accumulation period could be attributed to the fact that the area of interest is having 100% snow cover?
- L439: this does not seem true to me (at least in the present form).
- L442: this is in contradiction with L253 where it is stated that the snow depth is strongly governed by topography, which is also the main hypothesis why the proposed approach has been applied. This ping pong effects makes it challenging to understand the overall benefit of ICESat-2 data (and the validity of the presented results).
- L452: speculative, please revise it.
- L484: 20 m is not hyper resolution.
- References are generally ordered alphabetically.
Citation: https://doi.org/10.5194/egusphere-2024-1404-RC2 - AC2: 'Reply on RC2', Marco Mazzolini, 06 Sep 2024
Peer review completion
Journal article(s) based on this preprint
Model code and software
MuSA: v2.1 Esteban Alonso-González, Marco Mazzolini, and Kristoffer Aalstad https://zenodo.org/records/11147258
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 708 | 296 | 74 | 1,078 | 48 | 56 |
- HTML: 708
- PDF: 296
- XML: 74
- Total: 1,078
- BibTeX: 48
- EndNote: 56
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Cited
1 citations as recorded by crossref.
Marco Mazzolini
Kristoffer Aalstad
Esteban Alonso-González
Sebastian Westermann
Désirée Treichler
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2058 KB) - Metadata XML