the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 Mar 2026
Snowpack trends across Canada's largest river basin using three decades of manual snow surveys and ERA5-Land data
Abstract. In the Mackenzie River basin (MRB), end-of-season snowpack volume is an essential indicator for spring flood potential, wildfire risk, and hydroelectric power generation, and it underpins cold region hydrological research and monitoring. This study presents a trend analysis of 31 years of previously unpublished manual snow survey data from Canada's Northwest Territories (NWT) and incorporates snow survey data from other jurisdictions within the MRB to determine how the end-of-season snow water equivalent (SWE) has changed. We then use this independent SWE dataset to evaluate SWE trends from ERA5-Land reanalysis data for its suitability as an additional data source for assessing interannual variability in SWE. Spearman rank correlation analysis demonstrated moderately strong agreement in SWE trend variability between datasets, with a stronger agreement for sites in the NWT. We found that climate variables were the dominant drivers of residuals. Over the 31-year period, SWE trends exhibited substantial spatial variability within and between sub-basins and ecozones. No consistent latitudinal or elevational trends emerged, highlighting the complex, landscape-dependent influence of warming air temperatures on snowpack accumulation. These findings are relevant to decision makers who need an improved understanding on how SWE trends are changing at a basin scale. These results demonstrate that the combined use of manual snow surveys and gridded reanalysis datasets can be used to strengthen long-term snow monitoring and research in the Northwest Territories.
Status: closed
- RC1: 'Comment on egusphere-2026-1002', Anonymous Referee #1, 01 May 2026
-
RC2: 'Comment on egusphere-2026-1002', Anonymous Referee #2, 09 Jul 2026
The authors present an analysis of 30-year long observed in-situ snow water equivalent (SWE) data from a very large river basin (Mackenzie river basin; MRB) in Canada together with SWE data from a simulated gridded ERA5 land reanalysis data set. The main aim of the manuscript is to show 30-year snow trends and to evaluate the SWE grids in ERA5 land data against the available in-situ observations. A moderately strong correlation between these two datasets appears. Generally non-significant trends and no clear spatial trend patterns emerge from their analysis. Interestingly, no clear latitudinal north-south pattern seems to appear in the trend estimates.
Long-term trends in snow-conditions is an interesting and important subject, especially when applying in-situ data from such a large area as the MRB. Thus, there is research potential in the material presented. However, in my opinion, the manuscript is rather short and many of the conclusions and results are not sufficiently underpinned by well-quantified and justified facts and discussions, and remain of the qualitative type, e.g. “increased confidence in ERA5”. Many details behind the methods and analysis results, as well as the main contribution and novelty in this manuscript to the scientific community remain unclear to me. Therefore, I regrettably cannot recommend its publication in The Cryosphere.
Specific comments:
Snow and climate trends for MRB/Canada detected in previous studies should be briefly reviewed/mentioned in the “Introduction” section.
L74-75: please explain clearer how end-of-season SWE can in practice be used to predict the likelihood of ice-jam break-ups.
A technical description of the ERA5 land gridded data set is almost entirely missing in the text (only 7 lines of text in section 2.1.2). As this is a central data set used in the manuscript, it should be more thoroughly covered (how is it simulated/made, what is the spatial resolution, which atmospheric and snow observations are assimilated into it, pros and cons, etc.). Specifically on L90-97: it should be made clearer what snow data (if any) has been assimilated to produce ERA5 land SWE grids. Have satellite images or snow depth observations been used to adjust snow conditions and SWE in ERA5 land, or is this gridded data set produced without any assimilation of in-situ data, as the authors claim?
Section 2.2. is very brief and some more technical information about the SWE measurement devices would be good to include (e.g. sampling tube diameter).
The authors should give some more detailed technical descriptions of the trend and other methods used (e.g. Random forest, now only giving a reference to a R-package).
More details on the typical snow conditions in the MRB should be mentioned. Now the manuscript, which is about SWE, does not mention or display any absolute SWE values in mm-units. I wonder e.g. what the typical snow depth ranges are in the different parts of the catchment? In addition, climate change is discussed in the ms. but no quantitative numbers of the temperature and precipitation changes e.g. during the last 30 years are given.
L197-199: it remains unclear what advantage the normalizations against the two winters gives here.
In section 3 the text readability suffers from too much numbers and place names. Some better structure in this chapter would be beneficial, e.g. by adding a table/figure.
L218; 282: please explain what “simple difference” means?
L284-288: it sounds rather trivial that precipitation and precipitation as snow would be the strongest site-specific predictors for SWE in the random forest analysis.
L291-304: repetitive and somewhat trivial text in the section Discussion and Conclusions. Separation of the Discussion and Conclusions sections is recommended.
L297: percent or percent-point units here, please check?
Figure 3: The difference between colored points with and without a black circle is not explained in the caption.
L311-313, 371: please give a practical example of “coupled use” of ERA5 and manual snow surveys for e.g. Goverment agencies, based on the results of this study.
L314-325: This conclusion, that the “site locations and sampling design captured variability in snow conditions” needs some more explanation and quantification to be justified. Explain/list which subgrid-variability metrics were exactly tested here?
L331-333: please expand this discussion; in what way exactly do the trends in these previous studies differ from the current study?
L333-335: overall and trivial sentence.
L340: these results are interesting, but reference should be given to the overall statement “there is some observational evidence”. In addition, please explain and quantify the recent observed climate change in the region with some numbers.
L344: I suggest to discuss here the study by Gottlieb, A. R. & Mankin, J. S. 2024: Evidence of human influence on Northern Hemisphere snow loss. Nature 625, 293–300. doi: 10.1038/s41586-023-06794-y
L351-353: unclear sentence (stating basically that rain impacts the increase in precipitation?)
L354-356: it seems trivial that snow melt affects SWE, but how large is the sublimation effect? Please find a proper reference here to quantify this.
L357: please mention what has been the magnitude of warming the last 30 years in this basin?
L362-371: The most significant result in the manuscript is, according to the authors (L364) ”the moderately strong correlation of SWE trends between the ERA5-Land and manual snow survey datasets”. The authors should discuss more if this is something one would normally expect from a “top performing” (L305) reanalysis data set such as ERA5. In addition, the exact evaluation (goodness of fit) numbers are not given for relative SWE-value comparison (only reference to good agreement for two winter seasons, in Figures S13 and S14 in Supplementary material).
Citation: https://doi.org/10.5194/egusphere-2026-1002-RC2
Status: closed
-
RC1: 'Comment on egusphere-2026-1002', Anonymous Referee #1, 01 May 2026
This study aims to evaluate SWE estimates from the ERA5-Land reanalysis for hydrological studies and trend analysis in the MRB. The authors characterize basin and ecozone SWE using unpublished NWT data and CanSWE observations from ECCC. The study’s novelty, methodological clarity, and consistency in analysis periods need to be strengthened before publication can be considered. Several sections would benefit from clearer justification of methodological choices, improved organization, and additional references. In particular, the authors should address the following:
Major comments
1. The introduction would benefit from adding more references, especially regarding prior validation of SWE in ERA5-Land (e.g., studies like Kanda et al (2025). and Kouki et al.(2024)). More importantly, the novelty and motivation are not clearly established. As mentioned in the manuscript, similar benchmarking and trend analyses have already been conducted (e.g., Mudryk et al.), so the authors must explain clearly what the new contributions are here.2. The methods section needs clearer rationale for several important choices. Further, if approaches are based on previous studies, appropriate references should be added. In particular:
o Greater clarity about the time periods used: the trend analysis appears to include recent winters (1995-2025), while predictors for the random forest are based on 1991–2020. This discrepancy should be explained.
o Do the ERA5-Land, CanSWE, and GNWT datasets overlap temporally, and are GNWT data are included within CanSWE or are they independent? This needs to be made explicit. Also, it would also help to specify the total number of stations, their record lengths, and how many GNWT stations include multi-point snow surveys. It is also mentioned that only forested and upland land cover types were GNWT was selected. Were the same filtering criteria applied to CanSWE sites and ERA5L grids?
o There are also inconsistencies in how SWE metrics are defined. The rationale for using different definitions of maximum or end-of-season SWE across datasets (e.g., February–May for ERA5-Land, maximum SWE from CanSWE, post–March 15 for GNWT) should be explained.
o The process of point-to-grid matching would benefit from clearer explanation. On L134 it is not clear what p-value is being referred to, or how it relates to matching. What is the justification for using a 50km threshold for matching to a basin? What is the sensitivity of the results to this choice?
o Precipitation indices are mentioned as key predictors (L284), but the indices are not defined in the methods.3. The statistical and logical basis for using the random forest model to infer physical drivers of SWE trends needs to be justified in detail. Section 2.4 is essentially presented in note form, leaving the reader with no real understanding of what is being done, or why. No references supporting the methodological choice are provided, nor is any information about model hyperparameter tuning, cross-validation etc. Furthermore, the results from the randomforest model are discussed extremely briefly in Section 3.4, and no tables or figures are provided as supporting evidence. This part of the analysis felt very underdeveloped. A fundamental science question that is unaddressed is what is the physical meaning of differences in trends between the reanalysis and point surveys (defined here as “residuals”)? What is the role for measurement and/or modelling uncertainty, and how are they factored into this analysis? And ultimately, how do these results help understand the spatial variability in trends, including opposite sign trends in different watersheds?
4. L227: Ecozone averaged SWE was higher for Cordillera ecozones. Would the analysis become easier to follow if the study area was stratified by elevation and/or land cover types? The role of elevation should be more explicitly considered, as reduced ERA5-Land performance in certain ecozones may be linked to elevation effects reported in previous studies (e.g., Kanda et al.2025).
5. There is little discussion of uncertainties in either in situ measurements or ERA5-Land data, and no uncertainty estimates are presented. What are the confidence intervals on the trend estimates for individual basins? Given the small sample sizes and high variability, I assume that the error bars are substantial, perhaps even including zero (thus making individual trends not statistically significant).
6. While ERA5-Land is highlighted as useful for spatial gap filling (e.g., Vionnet et al., 2025), potential biases and their implications should also be discussed.
7. A large number of figures and tables are included as an Appendix to the main manuscript; however, I believe that these belong more in a Supplementary section that is separate from the main manuscript.
Minor Comments:
• Figure 1 could be improved by distinguishing between data sources (e.g., CanSWE vs GNWT sites).
• Some sections could be reorganized for clarity (e.g., merging section 2.1.1 with 2.2?).
• The treatment of winters (e.g., inclusion of 2021–2022 and 2023–2024 but not 2022–2023) also needs justification (lines 198).
• The results section would be easier to follow if aligned more closely with the research questions outlined earlier.
• Some statements require clarification, for example, “10% of the network” (L236) should specify which network is being referred to.Citation: https://doi.org/10.5194/egusphere-2026-1002-RC1 -
RC2: 'Comment on egusphere-2026-1002', Anonymous Referee #2, 09 Jul 2026
The authors present an analysis of 30-year long observed in-situ snow water equivalent (SWE) data from a very large river basin (Mackenzie river basin; MRB) in Canada together with SWE data from a simulated gridded ERA5 land reanalysis data set. The main aim of the manuscript is to show 30-year snow trends and to evaluate the SWE grids in ERA5 land data against the available in-situ observations. A moderately strong correlation between these two datasets appears. Generally non-significant trends and no clear spatial trend patterns emerge from their analysis. Interestingly, no clear latitudinal north-south pattern seems to appear in the trend estimates.
Long-term trends in snow-conditions is an interesting and important subject, especially when applying in-situ data from such a large area as the MRB. Thus, there is research potential in the material presented. However, in my opinion, the manuscript is rather short and many of the conclusions and results are not sufficiently underpinned by well-quantified and justified facts and discussions, and remain of the qualitative type, e.g. “increased confidence in ERA5”. Many details behind the methods and analysis results, as well as the main contribution and novelty in this manuscript to the scientific community remain unclear to me. Therefore, I regrettably cannot recommend its publication in The Cryosphere.
Specific comments:
Snow and climate trends for MRB/Canada detected in previous studies should be briefly reviewed/mentioned in the “Introduction” section.
L74-75: please explain clearer how end-of-season SWE can in practice be used to predict the likelihood of ice-jam break-ups.
A technical description of the ERA5 land gridded data set is almost entirely missing in the text (only 7 lines of text in section 2.1.2). As this is a central data set used in the manuscript, it should be more thoroughly covered (how is it simulated/made, what is the spatial resolution, which atmospheric and snow observations are assimilated into it, pros and cons, etc.). Specifically on L90-97: it should be made clearer what snow data (if any) has been assimilated to produce ERA5 land SWE grids. Have satellite images or snow depth observations been used to adjust snow conditions and SWE in ERA5 land, or is this gridded data set produced without any assimilation of in-situ data, as the authors claim?
Section 2.2. is very brief and some more technical information about the SWE measurement devices would be good to include (e.g. sampling tube diameter).
The authors should give some more detailed technical descriptions of the trend and other methods used (e.g. Random forest, now only giving a reference to a R-package).
More details on the typical snow conditions in the MRB should be mentioned. Now the manuscript, which is about SWE, does not mention or display any absolute SWE values in mm-units. I wonder e.g. what the typical snow depth ranges are in the different parts of the catchment? In addition, climate change is discussed in the ms. but no quantitative numbers of the temperature and precipitation changes e.g. during the last 30 years are given.
L197-199: it remains unclear what advantage the normalizations against the two winters gives here.
In section 3 the text readability suffers from too much numbers and place names. Some better structure in this chapter would be beneficial, e.g. by adding a table/figure.
L218; 282: please explain what “simple difference” means?
L284-288: it sounds rather trivial that precipitation and precipitation as snow would be the strongest site-specific predictors for SWE in the random forest analysis.
L291-304: repetitive and somewhat trivial text in the section Discussion and Conclusions. Separation of the Discussion and Conclusions sections is recommended.
L297: percent or percent-point units here, please check?
Figure 3: The difference between colored points with and without a black circle is not explained in the caption.
L311-313, 371: please give a practical example of “coupled use” of ERA5 and manual snow surveys for e.g. Goverment agencies, based on the results of this study.
L314-325: This conclusion, that the “site locations and sampling design captured variability in snow conditions” needs some more explanation and quantification to be justified. Explain/list which subgrid-variability metrics were exactly tested here?
L331-333: please expand this discussion; in what way exactly do the trends in these previous studies differ from the current study?
L333-335: overall and trivial sentence.
L340: these results are interesting, but reference should be given to the overall statement “there is some observational evidence”. In addition, please explain and quantify the recent observed climate change in the region with some numbers.
L344: I suggest to discuss here the study by Gottlieb, A. R. & Mankin, J. S. 2024: Evidence of human influence on Northern Hemisphere snow loss. Nature 625, 293–300. doi: 10.1038/s41586-023-06794-y
L351-353: unclear sentence (stating basically that rain impacts the increase in precipitation?)
L354-356: it seems trivial that snow melt affects SWE, but how large is the sublimation effect? Please find a proper reference here to quantify this.
L357: please mention what has been the magnitude of warming the last 30 years in this basin?
L362-371: The most significant result in the manuscript is, according to the authors (L364) ”the moderately strong correlation of SWE trends between the ERA5-Land and manual snow survey datasets”. The authors should discuss more if this is something one would normally expect from a “top performing” (L305) reanalysis data set such as ERA5. In addition, the exact evaluation (goodness of fit) numbers are not given for relative SWE-value comparison (only reference to good agreement for two winter seasons, in Figures S13 and S14 in Supplementary material).
Citation: https://doi.org/10.5194/egusphere-2026-1002-RC2
Data sets
Northwest Territories snow survey data and methodology, 1965-2024 E. Riley et al. https://doi.org/10.46887/2025-005
Model code and software
Snow Package: Snow water equivalent analysis and trend detection. Gregory E. https://github.com/EmmaGRiley/NWTHydroclimate/tree/main/packages/snow
Viewed
Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 705 | 0 | 2 | 707 | 0 | 0 |
- HTML: 705
- PDF: 0
- XML: 2
- Total: 707
- BibTeX: 0
- EndNote: 0
Viewed (geographical distribution)
Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This study aims to evaluate SWE estimates from the ERA5-Land reanalysis for hydrological studies and trend analysis in the MRB. The authors characterize basin and ecozone SWE using unpublished NWT data and CanSWE observations from ECCC. The study’s novelty, methodological clarity, and consistency in analysis periods need to be strengthened before publication can be considered. Several sections would benefit from clearer justification of methodological choices, improved organization, and additional references. In particular, the authors should address the following:
Major comments
1. The introduction would benefit from adding more references, especially regarding prior validation of SWE in ERA5-Land (e.g., studies like Kanda et al (2025). and Kouki et al.(2024)). More importantly, the novelty and motivation are not clearly established. As mentioned in the manuscript, similar benchmarking and trend analyses have already been conducted (e.g., Mudryk et al.), so the authors must explain clearly what the new contributions are here.
2. The methods section needs clearer rationale for several important choices. Further, if approaches are based on previous studies, appropriate references should be added. In particular:
o Greater clarity about the time periods used: the trend analysis appears to include recent winters (1995-2025), while predictors for the random forest are based on 1991–2020. This discrepancy should be explained.
o Do the ERA5-Land, CanSWE, and GNWT datasets overlap temporally, and are GNWT data are included within CanSWE or are they independent? This needs to be made explicit. Also, it would also help to specify the total number of stations, their record lengths, and how many GNWT stations include multi-point snow surveys. It is also mentioned that only forested and upland land cover types were GNWT was selected. Were the same filtering criteria applied to CanSWE sites and ERA5L grids?
o There are also inconsistencies in how SWE metrics are defined. The rationale for using different definitions of maximum or end-of-season SWE across datasets (e.g., February–May for ERA5-Land, maximum SWE from CanSWE, post–March 15 for GNWT) should be explained.
o The process of point-to-grid matching would benefit from clearer explanation. On L134 it is not clear what p-value is being referred to, or how it relates to matching. What is the justification for using a 50km threshold for matching to a basin? What is the sensitivity of the results to this choice?
o Precipitation indices are mentioned as key predictors (L284), but the indices are not defined in the methods.
3. The statistical and logical basis for using the random forest model to infer physical drivers of SWE trends needs to be justified in detail. Section 2.4 is essentially presented in note form, leaving the reader with no real understanding of what is being done, or why. No references supporting the methodological choice are provided, nor is any information about model hyperparameter tuning, cross-validation etc. Furthermore, the results from the randomforest model are discussed extremely briefly in Section 3.4, and no tables or figures are provided as supporting evidence. This part of the analysis felt very underdeveloped. A fundamental science question that is unaddressed is what is the physical meaning of differences in trends between the reanalysis and point surveys (defined here as “residuals”)? What is the role for measurement and/or modelling uncertainty, and how are they factored into this analysis? And ultimately, how do these results help understand the spatial variability in trends, including opposite sign trends in different watersheds?
4. L227: Ecozone averaged SWE was higher for Cordillera ecozones. Would the analysis become easier to follow if the study area was stratified by elevation and/or land cover types? The role of elevation should be more explicitly considered, as reduced ERA5-Land performance in certain ecozones may be linked to elevation effects reported in previous studies (e.g., Kanda et al.2025).
5. There is little discussion of uncertainties in either in situ measurements or ERA5-Land data, and no uncertainty estimates are presented. What are the confidence intervals on the trend estimates for individual basins? Given the small sample sizes and high variability, I assume that the error bars are substantial, perhaps even including zero (thus making individual trends not statistically significant).
6. While ERA5-Land is highlighted as useful for spatial gap filling (e.g., Vionnet et al., 2025), potential biases and their implications should also be discussed.
7. A large number of figures and tables are included as an Appendix to the main manuscript; however, I believe that these belong more in a Supplementary section that is separate from the main manuscript.
Minor Comments:
• Figure 1 could be improved by distinguishing between data sources (e.g., CanSWE vs GNWT sites).
• Some sections could be reorganized for clarity (e.g., merging section 2.1.1 with 2.2?).
• The treatment of winters (e.g., inclusion of 2021–2022 and 2023–2024 but not 2022–2023) also needs justification (lines 198).
• The results section would be easier to follow if aligned more closely with the research questions outlined earlier.
• Some statements require clarification, for example, “10% of the network” (L236) should specify which network is being referred to.