Which global reanalysis dataset represents better in snow cover on the Tibetan Plateau?

Yan, Shirui; Pu, Wei; Chen, Yang; Hou, Yaliang; Li, Xuejing; Xing, Yuxuan; Wu, Dongyou; Cui, Jiecan; Zhou, Yue; Wang, Xin

doi:https://doi.org/10.5194/egusphere-2024-82

Preprints

https://doi.org/10.5194/egusphere-2024-82

Preprints

23 Feb 2024

| 23 Feb 2024

Which global reanalysis dataset represents better in snow cover on the Tibetan Plateau?

Shirui Yan, Wei Pu, Yang Chen, Yaliang Hou, Xuejing Li, Yuxuan Xing, Dongyou Wu, Jiecan Cui, Yue Zhou, and Xin Wang

Abstract. The extensive snow cover across the Tibetan Plateau (TP) regions has a major influence on the climate and water supply for over one billion downstream inhabitants. However, an adequate evaluation of snow cover fraction (SCF) variability over the TP simulated by global multiple reanalysis datasets has yet to be undertaken. In this study, we examined eight global reanalysis SCF datasets using the Snow Property Inversion from Remote Sensing (SPIReS) product spanning the period 2001–2020. The results reveal that the HMASR generated the best SCF simulations because of its outstanding spatial and temporal accuracy. The GLDAS and CFSR demonstrated acceptable SCF accuracy with respect to spatial variability, but struggled to reproduce the annual trend. Pronounced SCF overestimations were found when using the ERA5, ERA5L, and JRA55, but SCF was underestimated by MERRA2, and CRAL generated poor spatial pattern. Overall biases were related to the combined effect of precipitation forcing, temperature forcing, snow data assimilation, and SCF parameterization methods, with the dominant factor changing across datasets. In ERA5 and ERA5L, temperature and snowfall bias exhibited significant correlations with SCF bias over most TP areas, therefore having a greater impact on the accuracy of SCF in terms of spatial variability and temporal evolution. On the other hand, the impact of snow assimilation was possibly more pronounced in MERRA2 and CRAL. Although parameterization methods can improve SCF simulation accuracy, their influence was weaker than those of other factors, except for JRA55. To further improve the accuracy of SCF simulation, an ensemble average method was developed. The ensemble based on HMASR and GLDAS generated the most accurate SCF spatial distribution, whereas the ensemble containing ERA5L, CFSR, CRAL, GLDAS, ERA5, and MERRA2 proved optimal for capturing the annual trend.

Received: 10 Jan 2024 – Discussion started: 23 Feb 2024

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.

Download & links

Preprint (PDF, 10128 KB)

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 (10128 KB)

Supplement (3350 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Journal article(s) based on this preprint

10 Sep 2024

Which global reanalysis dataset has better representativeness in snow cover on the Tibetan Plateau?

Shirui Yan, Yang Chen, Yaliang Hou, Kexin Liu, Xuejing Li, Yuxuan Xing, Dongyou Wu, Jiecan Cui, Yue Zhou, Wei Pu, and Xin Wang

The Cryosphere, 18, 4089–4109, https://doi.org/10.5194/tc-18-4089-2024,https://doi.org/10.5194/tc-18-4089-2024, 2024

Short summary

Shirui Yan, Wei Pu, Yang Chen, Yaliang Hou, Xuejing Li, Yuxuan Xing, Dongyou Wu, Jiecan Cui, Yue Zhou, and Xin Wang

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2024-82', Anonymous Referee #1, 05 Mar 2024
Yan et al. analyze snow cover fraction from various global reanalyses and one specific snow reanalyses for High Mountain Asia, and relate this to remotely sensed snow cover fraction and a regional meteorological dataset. The study has some potential, however, in the current stage of the manuscript I am not willing to do a full detailed review until basic scientific principles are adhered to (see major points).

Major points:

Introduction has many inaccuracies, and I stopped at around L70 (see also minor points below).

Methods are incomplete to evaluate the manuscript: study area, hydrological basins, temporal period, how was snowfall derived from TPMFD, how were the datasets on different resolutions merged, definition of hydrological year (if any), and more.

I’m not sure about the research aims. If HMASR is the only reanalysis from all studied that assimilates MODIS observations, and MODIS is used for evaluation, then the expected results are rather obvious.

Minor points:

L50: I would assume glaciers are more sensitive to long-term climate changes. Snow cover is extremely sensitive to year-to-year variability. And because of this strong interannual variability, less to long-term changes.

L54: Beniston et al. 2018 deals with European cryosphere, not TP.

L60: SPIReS is not a product, but a spectral unmixing method. And Stillinger et al. 2023 recommend spectral unmixing in general, not SPIReS in particular.

L69: There are no “global” meteorological agencies. Just regional, at best supranational, which produce the reanalyses.

At this point I grew tired of the introduction: Please check the accuracy of your statements. And in particular against the references you cite.

Table 1 and related text: Please explain why you defined the SCF parametrization for ERA5, ERA5L, MERRA and JRA as you did.

L255: Taylor diagrams do not evaluate spatial correlation.

Data and methods: Missing study area and period of analysis.

L265: Unclear how you applied the CI, since it requires a common grid, and all your datasets come with different resolutions.

Results: Where do the basins come from? Please introduce in methods.

HMASR covers only partly of the 2001-2020 period you used for the other datasets. Why not make a common period? As it is, you introduce artifacts with different periods.
Citation: https://doi.org/10.5194/egusphere-2024-82-RC1
- AC2: 'Reply on RC1', Shirui Yan, 05 May 2024
  
  Dear Prof. Anonymous Referee #1,
  Thank you very much for your constructive suggestions, which will significantly enhance the scientific quality and language expression of our manuscript. In response to your feedback, we have made extensive revisions to the language and methodology of the manuscript. A detailed point-by-point response to your comments and the modifications made to the manuscript will be provided in the supplementary file. We look forward to the revised manuscript meeting your standards and eagerly await your further review.
  Yours faithfully,
  Wei Pu,
  College of Atmospheric Sciences, Lanzhou University, Gansu, China
  5 May, 2024
  
  Citation: https://doi.org/10.5194/egusphere-2024-82-AC2
RC2:
'Comment on egusphere-2024-82', Anonymous Referee #2, 21 Mar 2024

Which global reanalysis dataset represents better in snow cover on the Tibetan Plateau?
This paper evaluates snow cover fraction (SCF) of eight reanalysis datasets over the Tibetan Plateau during 2001-2020. The SPIReS EO product is used as the reference dataset and products are evaluated based on their bias, Taylor skill score and its component metrics, and trends by means of a consistency index. The impact of biases in temperature and precipitation on forcing is assessed through comparisons of the reanalysis forcing with a reference dataset (TPMFD). An interesting analysis of SCF parameterization is also provided as are tests to identify the optimal combination of SCF products to include in an ensemble. The manuscript tries to conclude on the impact of snow data assimilation in SCF product performance.
A comparison of SCF products over the TP is an important endeavor. The manuscript is promising and contains useful and interesting analysis. However, I find the authors struggle to present their results in a way that builds logically to their stated conclusions, specifically as it relates to the role of snow data assimilation. The authors have a tendency to make general statements that do not flow logically from the evidence presented in the text. And at times the evidence presented contradicts these general statements.
The methods are not adequately described (see ‘primary concerns - Methods’). The authors struggle to interpret and communicate information contained in the Taylor plots. This may be related to the lack of clarity in how the metrics in the plots were calculated and what they represent. The text, particularly the results, could be better organized and ideas clearly separated. I encourage the authors to make better use of paragraphs. Sometimes, the references don’t support the statement as written -i.e. either not the appropriate reference or summary of the study contains inaccuracies. This may be due language issues.
Primary concerns
I am unconvinced by the statement that the reanalysis datasets incorporating snow assimilation always outperform those without snow assimilation [lines 348-351 and elsewhere]. The results, as presented, do not support this firm conclusion but rather suggest a more nuanced finding. In terms of skill scores, the top two products use snow data assimilation. However, JRA55 which also assimilates snow information has a comparable skill score to ERA5, ERA5L and CRAL which do not assimilate snow information. On the other hand, GLDAS which does not assimilate snow information has comparable skill scores to HMASR and CFSR. These findings are presented in the text and seem to contradict the general statement. I suggest the authors provide a more nuanced discussion and revisit how the ideas flow and build towards a specific conclusion. For example, later the authors tie the poor performance of JRA55 to SCF parameterization. Is it also possible that assimilation of poor snow data (JRA55) can degrade performance? Does the strong performance of GLDAS indicate that performance comparable to the best products that assimilate snow data can be achieved without snow data assimilation? Further, the above results pertain to SCF accuracy and not to the representation of trends where ERA5 and ERA5L (no snow data assimilation) performed best.
Methods –
The methods, specifically the statistical approaches, need to be clarified to adequately review the results presented in the manuscript.
Did you use anomaly fields or raw SCF values in deriving your Taylor plots? Please clarify.
‘Spatial correlation’ – was this calculated using all times and locations which would be more of a bulk correlation rather than a spatial or pattern correlation. Please clarify. Using the full time series is fine, you just need to keep this in mind when interpreting the results and remove ‘spatial’ from talk of correlation.
Standard deviation ratio – unclear what you mean by ‘degree of similarity in dispersion patterns’. Note that if you used the full time series then the variability captured by the STDR would be related to the variability in both time and space. Again, it is perfectly fine to use the full 20 year anomaly time series but the terminology and definitions need to be adjusted to reflect what was calculated.
Depending how the above metrics were calculated, the wording and interpretations in the results section may need to be revised to accurately reflect the methods.
Additionally, approaches to investigate the impact of meteorological forcings and SCF parameterization should be outlined in the methods section. Currently Methods is limited to presenting the metrics used to compare a product with a reference and does not lay out how the authors plan to meet the objectives stated in the introduction. For example, ~L372-375 you mention that you investigate SCF bias by looking at snowfall, temperature, and parameterization thresholds but you do not clearly lay out how you conducted this investigation. Instead, it jumps right into results. This would be ok if the methods were outlined in the methods section but they are not. What methods were used? Some of what is in the results could be moved to the methods.
For precipitation forcing, did you use snowfall or total precipitation? Please clarify.
Additional comments
Suggest separating analysis of impact of meteoritical forcing and SCF parameterization
L286-288: How did you select these two products as being the ‘worst’? If the comparison is relative to SPIRES which has a TP SCF average of 0.13 then MERRA2 at 0.05 is only 0.08 away from SPIRES whereas ERA5 (0.41), ERA5L and JRA55 all appear to have larger absolute biases (with respect to SPIRES reference) compared to MERRA2. I can’t judge from the plot where CFSR falls. I think what you are pointing out is the product with the largest positive bias (ERA5) and the product with the largest negative bias (MERRA2).
Include mention of the STDR of the various products, not just the correlation on the Taylor plots. The skill score is more than just the correlation, there is some mention of this on Line 300.
The linkages between the maps (Figures 1a and 2a) and absolute bias to the STDR is tricky because it’s not clear how it was calculated. i.e. does the STDR reflect the variability in both time and space or just space?
L300 – add a few words about what an STDR close to 1 means.
Description of product performance could be a bit more nuanced: Although HMSAR is good, and ranks at or near the top in all assessments, it appears to over(under) estimate in the east(west) compared to the SPIRES dataset. These biases probably average out when looking at the TP as a whole. Further, HMSAR has too little spatial variability (low STDR) compared to the reference while GLDAS has a more appropriate amount of ‘variability’ with STDR of close to 1.
L306-307: unclear what you mean by ‘poor Taylor performance’. From Fig2b&c it that E5 captures the amount of variability well enough (STDR) but its large bias results in a high RMSD (and low skill score). I’d try to be more specific about which elements you are talking about.
L329-330: unclear what you mean by ‘scattered on the Taylor diagram panels’. Please be more specific. i.e. there is a large spread in STDR, RMSD is higher compared to the westerly basins, correlations are fairly consistent between 0.5 and 0.6 (this is just by eye).
L331-332: Do you mean the regional average SCF?
L333-334: Greater deviations compared to what? Unclear here and not obvious from the plots.
L367: start new paragraph with ‘Variations’.
L~415 – MERRA2 and CRAL seem to have strong correlations with temperature and precipitation in the west. I’m not convinced the evidence provided is sufficient to conclude that it is solely the absence of snow assimilation. Maybe use slightly more nuanced language.
L439: cut ‘leading to a better Taylor performance’ and instead just reference the skill scores.
L510-512: You just described how ERA5 and eRA5L have the best agreement in terms of trends and those products don’t assimilate snow data. So, the low performance of CRAL might not only be because of lack of data assimilation. At least, not if I follow the logic as presented in your paper. It might be that I am missing something or that the flow of ideas needs to be revised to support your conclusion.
Section 4.1 – You discuss in terms of good and bad but what do you mean by this? How much can a parameterization alter the SCF and the accompanying SS and CI? Is this the same for all products? When you say it doesn’t change the general performance do you mean the rankings of the products or the SS/CI values or both?
Section 4.2 – not surprising that HMSAR+GLDAS produces the best skill scores because they had the best skill scores already. None of the other products had equivalent but opposite biases of these two products that would even out and improve the overall performance.
L61: more advanced compared to the standard Modis SCF algorithm?
L73: Unclear which reanalysis dataset you are referring to. Reanalysis datasets in general?
L84-86: unclear. Do you mean reanalysis datasets that assimilate IMS and/or ground data?
L139: specify ‘over mountain areas in the western United States’.
L195: weekly

Citation: https://doi.org/10.5194/egusphere-2024-82-RC2
- AC1: 'Reply on RC2', Shirui Yan, 05 May 2024
  
  Dear Prof. Anonymous Referee #2,
  Thank you very much for your positive feedback and valuable suggestions, which have greatly contributed to improving the quality of our manuscript. A detailed point-by-point response to your comments and the modifications made to the manuscript will be provided in the supplementary file. We look forward to the revised manuscript meeting your standards and eagerly await your further review.
  Yours faithfully,
  Wei Pu,
  College of Atmospheric Sciences, Lanzhou University, Gansu, China
  5 May, 2024
  
  Citation: https://doi.org/10.5194/egusphere-2024-82-AC1
RC3:
'Comment on egusphere-2024-82', Anonymous Referee #3, 24 Mar 2024

This study comprehensively evaluated the snow cover fraction (SCF) of eight reanalysis datasets over the Tibetan Plateau based on selected remote sensing products. The authors found that each dataset demonstrates distinct characteristics in describing the SCF. Combining interpolation of atmosphere variables and parameterizations, the authors systematically investigate potential causes of SCF disagreements among these datasets. Additionally, an ensemble algorithm was developed to optimize SCF. In addition to being appropriate for The Cryosphere, this article offers significant implications for future research on data guidance. I could recommend publishing the manuscript in The Cryosphere after essential revisions.
Major comments:
1) Language
I strongly agree with reviewer #1 that the language needs to be significantly improved. I’m not a native speaker, so I leave the work to the authors.
2) Reference datasets
The data assessment and manuscript quality are largely based on the SPIReS dataset. It would be helpful if the authors could provide further clarification regarding this dataset's representativeness. Why is SPIReS chosen as a reference here? At the very least, a summary review of RS products is needed. It would also be nice to add SPIReS (and TPMFD) to Table 1.
3) Manuscipt structure
Sec. 3 Results
Please make the results sharp and use numbers to promote your results. Generally, the results section lacks qualitative descriptions and remains subjective. Please leave the discussion to the discussion section and present only results here.
Sec. 5 Conclusions
Generally, the conclusions are very specific and lack an overall assessment of the performance of state-of-the-art reanalyses. Furthermore, the conclusion and summary are largely similar to the abstract.
Specific comments:
L22: Please define ERA5L
L71: Please add relvelent reference here.
L109: from 2 to 26 cm
L173: HTESSE“L”
L225: Please provide the full name of the CRA-Land dataset, similar to the other reanalysis datasets mentioned in the manuscript.
L240: Important infomration of TPMFD dataset is not availbale. What variables are used in this data, what is the resolution and temporal coverage, how is snowfall derived, etc. This could be easily done by adding TPMFD to table 1.
L372: Jiang et al., 2020 reported to the signiﬁcant simulation biases for SCF over the TP. Please consider citing here.
L404: Remove “This suggests the presence of another significant factor that is responsible for the overestimation of SCF in JRA55”
L413: change “the important role of ” to “the importantance”
Table 1: Please also add the reference dataset here.
Figure 2: "Tibetan Plateau region" is used in many figure captions. Please remove the “region” and revise throughout the manuscript.
Figure 4: Instead of using both, I suggest using Autumn/Winter/Spring/Summer OR SON/DJF/MAM/JJA.
Figure 6: To improve the compariability, I would suggest changing the y-axis range of subpolt b the same.
Figure 10, the abbreviation for the reanalyzed dataset is not given.

References
Jiang, Y., Chen, F., Gao, Y., He, C., Barlage, M., Huang, W., 2020. Assessment of Uncertainty Sources in Snow Cover Simulation in the Tibetan Plateau. JGR Atmospheres 125, e2020JD032674. https://doi.org/10.1029/2020JD032674

Citation: https://doi.org/10.5194/egusphere-2024-82-RC3
- AC3: 'Reply on RC3', Shirui Yan, 05 May 2024
  
  Dear Prof. Anonymous Referee #3,
  We greatly appreciate your positive feedback and constructive suggestions, which have significantly contributed to the improvement of our manuscript. Additionally, we carefully reviewed each comment and made corresponding adjustments and improvements to the manuscript. A detailed point-by-point response to your comments and the modifications made to the manuscript will be provided in the supplementary file. We look forward to the revised manuscript meeting your standards and eagerly await your further review.
  Yours faithfully,
  Wei Pu,
  College of Atmospheric Sciences, Lanzhou University, Gansu, China
  5 May, 2024
  
  Citation: https://doi.org/10.5194/egusphere-2024-82-AC3

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2024-82', Anonymous Referee #1, 05 Mar 2024
Yan et al. analyze snow cover fraction from various global reanalyses and one specific snow reanalyses for High Mountain Asia, and relate this to remotely sensed snow cover fraction and a regional meteorological dataset. The study has some potential, however, in the current stage of the manuscript I am not willing to do a full detailed review until basic scientific principles are adhered to (see major points).

Major points:

Introduction has many inaccuracies, and I stopped at around L70 (see also minor points below).

Methods are incomplete to evaluate the manuscript: study area, hydrological basins, temporal period, how was snowfall derived from TPMFD, how were the datasets on different resolutions merged, definition of hydrological year (if any), and more.

I’m not sure about the research aims. If HMASR is the only reanalysis from all studied that assimilates MODIS observations, and MODIS is used for evaluation, then the expected results are rather obvious.

Minor points:

L50: I would assume glaciers are more sensitive to long-term climate changes. Snow cover is extremely sensitive to year-to-year variability. And because of this strong interannual variability, less to long-term changes.

L54: Beniston et al. 2018 deals with European cryosphere, not TP.

L60: SPIReS is not a product, but a spectral unmixing method. And Stillinger et al. 2023 recommend spectral unmixing in general, not SPIReS in particular.

L69: There are no “global” meteorological agencies. Just regional, at best supranational, which produce the reanalyses.

At this point I grew tired of the introduction: Please check the accuracy of your statements. And in particular against the references you cite.

Table 1 and related text: Please explain why you defined the SCF parametrization for ERA5, ERA5L, MERRA and JRA as you did.

L255: Taylor diagrams do not evaluate spatial correlation.

Data and methods: Missing study area and period of analysis.

L265: Unclear how you applied the CI, since it requires a common grid, and all your datasets come with different resolutions.

Results: Where do the basins come from? Please introduce in methods.

HMASR covers only partly of the 2001-2020 period you used for the other datasets. Why not make a common period? As it is, you introduce artifacts with different periods.
Citation: https://doi.org/10.5194/egusphere-2024-82-RC1
- AC2: 'Reply on RC1', Shirui Yan, 05 May 2024
  
  Dear Prof. Anonymous Referee #1,
  Thank you very much for your constructive suggestions, which will significantly enhance the scientific quality and language expression of our manuscript. In response to your feedback, we have made extensive revisions to the language and methodology of the manuscript. A detailed point-by-point response to your comments and the modifications made to the manuscript will be provided in the supplementary file. We look forward to the revised manuscript meeting your standards and eagerly await your further review.
  Yours faithfully,
  Wei Pu,
  College of Atmospheric Sciences, Lanzhou University, Gansu, China
  5 May, 2024
  
  Citation: https://doi.org/10.5194/egusphere-2024-82-AC2
RC2:
'Comment on egusphere-2024-82', Anonymous Referee #2, 21 Mar 2024

Which global reanalysis dataset represents better in snow cover on the Tibetan Plateau?
This paper evaluates snow cover fraction (SCF) of eight reanalysis datasets over the Tibetan Plateau during 2001-2020. The SPIReS EO product is used as the reference dataset and products are evaluated based on their bias, Taylor skill score and its component metrics, and trends by means of a consistency index. The impact of biases in temperature and precipitation on forcing is assessed through comparisons of the reanalysis forcing with a reference dataset (TPMFD). An interesting analysis of SCF parameterization is also provided as are tests to identify the optimal combination of SCF products to include in an ensemble. The manuscript tries to conclude on the impact of snow data assimilation in SCF product performance.
A comparison of SCF products over the TP is an important endeavor. The manuscript is promising and contains useful and interesting analysis. However, I find the authors struggle to present their results in a way that builds logically to their stated conclusions, specifically as it relates to the role of snow data assimilation. The authors have a tendency to make general statements that do not flow logically from the evidence presented in the text. And at times the evidence presented contradicts these general statements.
The methods are not adequately described (see ‘primary concerns - Methods’). The authors struggle to interpret and communicate information contained in the Taylor plots. This may be related to the lack of clarity in how the metrics in the plots were calculated and what they represent. The text, particularly the results, could be better organized and ideas clearly separated. I encourage the authors to make better use of paragraphs. Sometimes, the references don’t support the statement as written -i.e. either not the appropriate reference or summary of the study contains inaccuracies. This may be due language issues.
Primary concerns
I am unconvinced by the statement that the reanalysis datasets incorporating snow assimilation always outperform those without snow assimilation [lines 348-351 and elsewhere]. The results, as presented, do not support this firm conclusion but rather suggest a more nuanced finding. In terms of skill scores, the top two products use snow data assimilation. However, JRA55 which also assimilates snow information has a comparable skill score to ERA5, ERA5L and CRAL which do not assimilate snow information. On the other hand, GLDAS which does not assimilate snow information has comparable skill scores to HMASR and CFSR. These findings are presented in the text and seem to contradict the general statement. I suggest the authors provide a more nuanced discussion and revisit how the ideas flow and build towards a specific conclusion. For example, later the authors tie the poor performance of JRA55 to SCF parameterization. Is it also possible that assimilation of poor snow data (JRA55) can degrade performance? Does the strong performance of GLDAS indicate that performance comparable to the best products that assimilate snow data can be achieved without snow data assimilation? Further, the above results pertain to SCF accuracy and not to the representation of trends where ERA5 and ERA5L (no snow data assimilation) performed best.
Methods –
The methods, specifically the statistical approaches, need to be clarified to adequately review the results presented in the manuscript.
Did you use anomaly fields or raw SCF values in deriving your Taylor plots? Please clarify.
‘Spatial correlation’ – was this calculated using all times and locations which would be more of a bulk correlation rather than a spatial or pattern correlation. Please clarify. Using the full time series is fine, you just need to keep this in mind when interpreting the results and remove ‘spatial’ from talk of correlation.
Standard deviation ratio – unclear what you mean by ‘degree of similarity in dispersion patterns’. Note that if you used the full time series then the variability captured by the STDR would be related to the variability in both time and space. Again, it is perfectly fine to use the full 20 year anomaly time series but the terminology and definitions need to be adjusted to reflect what was calculated.
Depending how the above metrics were calculated, the wording and interpretations in the results section may need to be revised to accurately reflect the methods.
Additionally, approaches to investigate the impact of meteorological forcings and SCF parameterization should be outlined in the methods section. Currently Methods is limited to presenting the metrics used to compare a product with a reference and does not lay out how the authors plan to meet the objectives stated in the introduction. For example, ~L372-375 you mention that you investigate SCF bias by looking at snowfall, temperature, and parameterization thresholds but you do not clearly lay out how you conducted this investigation. Instead, it jumps right into results. This would be ok if the methods were outlined in the methods section but they are not. What methods were used? Some of what is in the results could be moved to the methods.
For precipitation forcing, did you use snowfall or total precipitation? Please clarify.
Additional comments
Suggest separating analysis of impact of meteoritical forcing and SCF parameterization
L286-288: How did you select these two products as being the ‘worst’? If the comparison is relative to SPIRES which has a TP SCF average of 0.13 then MERRA2 at 0.05 is only 0.08 away from SPIRES whereas ERA5 (0.41), ERA5L and JRA55 all appear to have larger absolute biases (with respect to SPIRES reference) compared to MERRA2. I can’t judge from the plot where CFSR falls. I think what you are pointing out is the product with the largest positive bias (ERA5) and the product with the largest negative bias (MERRA2).
Include mention of the STDR of the various products, not just the correlation on the Taylor plots. The skill score is more than just the correlation, there is some mention of this on Line 300.
The linkages between the maps (Figures 1a and 2a) and absolute bias to the STDR is tricky because it’s not clear how it was calculated. i.e. does the STDR reflect the variability in both time and space or just space?
L300 – add a few words about what an STDR close to 1 means.
Description of product performance could be a bit more nuanced: Although HMSAR is good, and ranks at or near the top in all assessments, it appears to over(under) estimate in the east(west) compared to the SPIRES dataset. These biases probably average out when looking at the TP as a whole. Further, HMSAR has too little spatial variability (low STDR) compared to the reference while GLDAS has a more appropriate amount of ‘variability’ with STDR of close to 1.
L306-307: unclear what you mean by ‘poor Taylor performance’. From Fig2b&c it that E5 captures the amount of variability well enough (STDR) but its large bias results in a high RMSD (and low skill score). I’d try to be more specific about which elements you are talking about.
L329-330: unclear what you mean by ‘scattered on the Taylor diagram panels’. Please be more specific. i.e. there is a large spread in STDR, RMSD is higher compared to the westerly basins, correlations are fairly consistent between 0.5 and 0.6 (this is just by eye).
L331-332: Do you mean the regional average SCF?
L333-334: Greater deviations compared to what? Unclear here and not obvious from the plots.
L367: start new paragraph with ‘Variations’.
L~415 – MERRA2 and CRAL seem to have strong correlations with temperature and precipitation in the west. I’m not convinced the evidence provided is sufficient to conclude that it is solely the absence of snow assimilation. Maybe use slightly more nuanced language.
L439: cut ‘leading to a better Taylor performance’ and instead just reference the skill scores.
L510-512: You just described how ERA5 and eRA5L have the best agreement in terms of trends and those products don’t assimilate snow data. So, the low performance of CRAL might not only be because of lack of data assimilation. At least, not if I follow the logic as presented in your paper. It might be that I am missing something or that the flow of ideas needs to be revised to support your conclusion.
Section 4.1 – You discuss in terms of good and bad but what do you mean by this? How much can a parameterization alter the SCF and the accompanying SS and CI? Is this the same for all products? When you say it doesn’t change the general performance do you mean the rankings of the products or the SS/CI values or both?
Section 4.2 – not surprising that HMSAR+GLDAS produces the best skill scores because they had the best skill scores already. None of the other products had equivalent but opposite biases of these two products that would even out and improve the overall performance.
L61: more advanced compared to the standard Modis SCF algorithm?
L73: Unclear which reanalysis dataset you are referring to. Reanalysis datasets in general?
L84-86: unclear. Do you mean reanalysis datasets that assimilate IMS and/or ground data?
L139: specify ‘over mountain areas in the western United States’.
L195: weekly

Citation: https://doi.org/10.5194/egusphere-2024-82-RC2
- AC1: 'Reply on RC2', Shirui Yan, 05 May 2024
  
  Dear Prof. Anonymous Referee #2,
  Thank you very much for your positive feedback and valuable suggestions, which have greatly contributed to improving the quality of our manuscript. A detailed point-by-point response to your comments and the modifications made to the manuscript will be provided in the supplementary file. We look forward to the revised manuscript meeting your standards and eagerly await your further review.
  Yours faithfully,
  Wei Pu,
  College of Atmospheric Sciences, Lanzhou University, Gansu, China
  5 May, 2024
  
  Citation: https://doi.org/10.5194/egusphere-2024-82-AC1
RC3:
'Comment on egusphere-2024-82', Anonymous Referee #3, 24 Mar 2024

This study comprehensively evaluated the snow cover fraction (SCF) of eight reanalysis datasets over the Tibetan Plateau based on selected remote sensing products. The authors found that each dataset demonstrates distinct characteristics in describing the SCF. Combining interpolation of atmosphere variables and parameterizations, the authors systematically investigate potential causes of SCF disagreements among these datasets. Additionally, an ensemble algorithm was developed to optimize SCF. In addition to being appropriate for The Cryosphere, this article offers significant implications for future research on data guidance. I could recommend publishing the manuscript in The Cryosphere after essential revisions.
Major comments:
1) Language
I strongly agree with reviewer #1 that the language needs to be significantly improved. I’m not a native speaker, so I leave the work to the authors.
2) Reference datasets
The data assessment and manuscript quality are largely based on the SPIReS dataset. It would be helpful if the authors could provide further clarification regarding this dataset's representativeness. Why is SPIReS chosen as a reference here? At the very least, a summary review of RS products is needed. It would also be nice to add SPIReS (and TPMFD) to Table 1.
3) Manuscipt structure
Sec. 3 Results
Please make the results sharp and use numbers to promote your results. Generally, the results section lacks qualitative descriptions and remains subjective. Please leave the discussion to the discussion section and present only results here.
Sec. 5 Conclusions
Generally, the conclusions are very specific and lack an overall assessment of the performance of state-of-the-art reanalyses. Furthermore, the conclusion and summary are largely similar to the abstract.
Specific comments:
L22: Please define ERA5L
L71: Please add relvelent reference here.
L109: from 2 to 26 cm
L173: HTESSE“L”
L225: Please provide the full name of the CRA-Land dataset, similar to the other reanalysis datasets mentioned in the manuscript.
L240: Important infomration of TPMFD dataset is not availbale. What variables are used in this data, what is the resolution and temporal coverage, how is snowfall derived, etc. This could be easily done by adding TPMFD to table 1.
L372: Jiang et al., 2020 reported to the signiﬁcant simulation biases for SCF over the TP. Please consider citing here.
L404: Remove “This suggests the presence of another significant factor that is responsible for the overestimation of SCF in JRA55”
L413: change “the important role of ” to “the importantance”
Table 1: Please also add the reference dataset here.
Figure 2: "Tibetan Plateau region" is used in many figure captions. Please remove the “region” and revise throughout the manuscript.
Figure 4: Instead of using both, I suggest using Autumn/Winter/Spring/Summer OR SON/DJF/MAM/JJA.
Figure 6: To improve the compariability, I would suggest changing the y-axis range of subpolt b the same.
Figure 10, the abbreviation for the reanalyzed dataset is not given.

References
Jiang, Y., Chen, F., Gao, Y., He, C., Barlage, M., Huang, W., 2020. Assessment of Uncertainty Sources in Snow Cover Simulation in the Tibetan Plateau. JGR Atmospheres 125, e2020JD032674. https://doi.org/10.1029/2020JD032674

Citation: https://doi.org/10.5194/egusphere-2024-82-RC3
- AC3: 'Reply on RC3', Shirui Yan, 05 May 2024
  
  Dear Prof. Anonymous Referee #3,
  We greatly appreciate your positive feedback and constructive suggestions, which have significantly contributed to the improvement of our manuscript. Additionally, we carefully reviewed each comment and made corresponding adjustments and improvements to the manuscript. A detailed point-by-point response to your comments and the modifications made to the manuscript will be provided in the supplementary file. We look forward to the revised manuscript meeting your standards and eagerly await your further review.
  Yours faithfully,
  Wei Pu,
  College of Atmospheric Sciences, Lanzhou University, Gansu, China
  5 May, 2024
  
  Citation: https://doi.org/10.5194/egusphere-2024-82-AC3

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Reconsider after major revisions (further review by editor and referees) (06 May 2024) by Chris Derksen

AR by Shirui Yan on behalf of the Authors (07 May 2024) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (08 May 2024) by Chris Derksen

RR by Anonymous Referee #1 (13 May 2024)

Suggestions for revision or reasons for rejection

The authors put a lot of effort in the revision of their manuscript by addressing all raised issues.

Major points:
• L261: Good that you brought everything to a common grid. However, bilinear interpolation is not suited to upscale data over multiple orders of magnitude. Especially for HMASR at 500m resolution. The recommended procedure is to simply aggregate the data to approximately a similar resolution (e.g, from a 0.1 to 0.5 grid by averaging 5 by 5 pixels), and then using the bilinear interpolation just to match the final grids (origin, extent, and/or slight resolution mismatch). (you can see this effect clearly in Fig1a, where SPIRES is much too noisy)
• Capturing sign and significance of trends over 17 years is not the best assessment of temporal agreement. It would be great to see some time series plots of the variation of SCF with water years by dataset, potentially also by basin. This could be easily summarized with values of bias and correlations, to give you some more information on temporal agreement between datasets.

Minor points:

• L314: parentheses three (two) is weird. Please spell out.
• Figure 5 does not match with previous ones: MERRA2 seems to have little bias, while in the previous plots it showed a high negative SCF bias.
• Figure 9 is a bad choice for visualization, since it’s extremely hard to compare opposite bars. Consider a table or simple bar/point chart. Also bias as metric would be interesting, to see if there is a relationship with snow height. Did you assess this by basin? Maybe different parametrizations might work better in different areas (shallow or variable snowpack as in the ITP vs more deep snowpack as in Amu and Indus, etc.)
• L701: “from the internet” is vague, consider providing a doi or url for each.

Hide

RR by Anonymous Referee #2 (16 May 2024)

Suggestions for revision or reasons for rejection

Which global reanalysis dataset represents better snow cover on the Tibetan Plateau? Major revision 1

Please refer to the edited manuscript (egusphere-2024-82-referee-report.pdf) for detailed editorial suggestions.

I thank the authors for carefully responding to my comments and those of the other reviewers. As stated in my initial review, comparison of SCF products over the TP is an important endeavour. This manuscript attempts to provide such an evaluation and includes some interesting analysis of the impact of SCF parameterization (going from SD/SWE to SCF). However, even with the additions made to the manuscript I continue to struggle to follow the logic and get lost in the weeds of the results section. The authors struggle to interpret and present their results in a succinct and logical manner. It’s not clear to me if this issue stems from language or from a lack of understanding of the statistical tests used.

The text could be significantly reduced. I appreciate it is challenging to summarize the performance of several different reanalysis products over various spatial and temporal (annual and seasonal) domains. Nevertheless, it is important to synthesize the results to digestible amount of text and figures.

The authors removed many of their broad conclusions that were not supported by evidence which is appreciated. However, it was still difficult to review the quality of some of the interpretations and analysis due to the paper structure. I found myself having to examine the figures, go back to the text, and try to reconcile the two and figure out what the authors are trying to say. This was also the case in the first round of revisions, but I was hopeful that by the second round it would be easier. For this reason, I did not provide too many comments on the content and validity of the conclusions.

The authors added descriptions of the methods which are appreciated. However, it would help readability if the methods-related text in the results section was moved to and integrated with this new methods section. While it can be helpful to remind the readers of the objectives and the methods, I found there was a bit too much explanation in the results section and the text loses focus. Further, there are still elements of the methods that aren’t clearly laid out. I have provided a suggested template in the detailed comments.

I feel there is good work in this paper, and it should be published at some point. Unfortunately, in its current form I do not feel it would be a useful contribution to the community. I really want to see this manuscript succeed and hope the authors are able to re-work the text.

I have provided detailed editorial suggestions directly on the revised manuscript. I hope the authors can build on some of these suggestions to produce a revised manuscript that is able to communicate this interesting work to the broader community.

Specific comments
-Please refer to the edited manuscript (egusphere-2024-82-referee-report.pdf) for further suggestions. The absence of detailed suggestions doesn’t mean revisions are not recommended. Not all sections received the same level of attention.
-Suggest trying to separate pure results from discussion and interpretation to help the flow of the manuscript.
-Just say correlation rather than writing R.

Methods
-Move line 337 end to line 329 to methods.
-Some suggested text for the methods section to more clearly lay out the study:
L261: For each 0.5x0.5 grid cell we calculate the climatological SCF over the full period and seasonally for each of the reanalysis datasets and SPIRES (e.g. Figure 1a). Absolute bias and correlation (Pearson correlation coefficient) were calculated from these values at both the basin and regional (TP) scales. [please correct me if I am wrong here and you used the time series of annual SCF] Spatial maps and summary values of the climatological SCF and its bias with respect to the SPIRES dataset are presented. Additionally, Taylor Diagrams are used to provide additional information regarding the RMSE and standard deviation ratio (STDR) . The climatological SCF values for each grid cell were used as input to calculate the Taylor Diagram’s component metrics - correlation, RMSE, and Standard Deviation Ratio (STDR). [cleaned text from line 271 277 goes here ending with (closer to 1).]. The component metrics (correlation, RMSE, STDR) were summarized by the Taylor Skill Score (SS) as follows:
SS = .... [the SS equation goes here] can probably omit the rest of the text on line 280-282]
Similar to the absolute bias, Taylor Diagrams and SS were calculated for the full TP and each of the nine sub-basins for the full period and seasonally.

Methods Section 2.3.2 – suggest splitting into 3 separate short sections. This structure could help re-frame the results.
- Analysis of SCF bias sources
- Analysis of SCF parameterization
- Generation of ensemble datasets

Results
- Condense and streamline
- Separate results and discussion related to SCF parameterization from the text concerning temperature and precipitation forcing. For example, L436-437 you mention that met forcing can only explain some of the biases and then you go into a discussion of parameterization. Instead, I suggest you mention that SCF parameterization could be a compounding factor and that it will be discussed in Section #.
- Lines 446-460 could be moved to discussion and/or the parameterization parts moved to a results piece related to parameterization thresholds.
- Consider moving seasonal SCF state and performance to follow immediately the annual/full period results before discussing the bias attribution. Bias attribution for seasonal case can follow or be integrated with the full period results.
- There are still difficulties in explaining and interpreting the results and metrics and some confusion between the time and space dimensions.

Discussion
I did not provide specific or detailed comments on the discussion or conclusions as there is already enough to be re-worked in the first three sections.

Conclusions
The authors expanded the conclusions at the request of one of the reviewers. While a good attempt, the new text needs to be cleaned and carefully edited (condensed). I did not provide detailed comments on the conclusions but hope the authors can build on suggestions provided elsewhere in the text.

Referee Report: PDF

Hide

RR by Anonymous Referee #3 (20 May 2024)

ED: Publish subject to revisions (further review by editor and referees) (20 May 2024) by Chris Derksen

AR by Shirui Yan on behalf of the Authors (30 Jun 2024) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (03 Jul 2024) by Chris Derksen

RR by Anonymous Referee #2 (05 Jul 2024)

ED: Publish subject to technical corrections (07 Jul 2024) by Chris Derksen

AR by Shirui Yan on behalf of the Authors (14 Jul 2024) Author's response Manuscript

Journal article(s) based on this preprint

10 Sep 2024

Which global reanalysis dataset has better representativeness in snow cover on the Tibetan Plateau?

Shirui Yan, Yang Chen, Yaliang Hou, Kexin Liu, Xuejing Li, Yuxuan Xing, Dongyou Wu, Jiecan Cui, Yue Zhou, Wei Pu, and Xin Wang

The Cryosphere, 18, 4089–4109, https://doi.org/10.5194/tc-18-4089-2024,https://doi.org/10.5194/tc-18-4089-2024, 2024

Short summary

Shirui Yan, Wei Pu, Yang Chen, Yaliang Hou, Xuejing Li, Yuxuan Xing, Dongyou Wu, Jiecan Cui, Yue Zhou, and Xin Wang

Supplement

https://doi.org/10.5194/egusphere-2024-82-supplement

Shirui Yan, Wei Pu, Yang Chen, Yaliang Hou, Xuejing Li, Yuxuan Xing, Dongyou Wu, Jiecan Cui, Yue Zhou, and Xin Wang

Viewed

Total article views: 729 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
513	178	38	729	52	21	42

HTML: 513
PDF: 178
XML: 38
Total: 729
Supplement: 52
BibTeX: 21
EndNote: 42

Views and downloads (calculated since 23 Feb 2024)

Month	HTML	PDF	XML	Total
Feb 2024	95	40	4	139
Mar 2024	120	46	8	174
Apr 2024	66	18	7	91
May 2024	91	33	10	134
Jun 2024	60	15	5	80
Jul 2024	33	8	3	44
Aug 2024	23	13	1	37
Sep 2024	25	5	0	30

Cumulative views and downloads (calculated since 23 Feb 2024)

Month	HTML	PDF	XML	Total
Feb 2024	95	40	4	139
Mar 2024	120	46	8	174
Apr 2024	66	18	7	91
May 2024	91	33	10	134
Jun 2024	60	15	5	80
Jul 2024	33	8	3	44
Aug 2024	23	13	1	37
Sep 2024	25	5	0	30

Viewed (geographical distribution)

Total article views: 747 (including HTML, PDF, and XML) Thereof 747 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 18 Sep 2024

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (10128 KB)
Metadata XML

Short summary

The snow cover over the Tibetan Plateau (TP) plays a role in the climate and hydrological systems, yet there exist uncertainties in snow cover fraction (SCF) estimations within reanalysis datasets. This study utilized the Snow Property Inversion from Remote Sensing (SPIReS) product to assess the accuracy of eight global reanalysis SCF datasets over the TP. Factors contributing to uncertainties were analyzed, and an ensemble averaging method was employed to provide optimized SCF simulations.


Total:	0
HTML:	0
PDF:	0
XML:	0