the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Feb 2025
Benchmarking Snow Fields of ERA5-Land in the Northern Regions of North America
Abstract. Reanalysis products provide new opportunities for assessments of historical Earth System states. This is crucial for snow variables, where ground-based observations are sparse and incomplete, and remote sensing measurements still face limitation. However, because reanalysis data are model-based, their accuracy must be evaluated before being applied in impact and attribution studies. In this study, we assess the accuracy of ERA5-Land's snow cover, snow depth, and Snow Water Equivalent (SWE) across monthly, seasonal, and annual scales, within the ecological regions of Canada and Alaska, regions that are characterized by prolonged seasonal snow cover. Using MODIS satellite snow cover observations and the gridded snow depth/SWE analysis data from the Canadian Meteorological Centre, we conduct a consistent benchmarking of ERA5-Land’s snow fields to (1) identify discrepancies at both gridded and regional scales, (2) evaluate the reproducibility of spatial structure of snow variables, and (3) uncover potential spatial patterns of discrepancies in ERA5-Land's snow statistics. Our results highlight significant discrepancies, particularly for snow depth and SWE, where ERA5-Land tends to grossly overestimate long-term mean values and interannual variability, while underestimating trends, i.e., moderating positive trends and exaggerating negative ones. The discrepancies in SWE, however, are primarily driven by biases in snow depth rather than snow density. Therefore, we advise against the direct use of ERA5-Land's snow depth and SWE in Canada and Alaska. While snow cover and snow density may still be useful for impact and attribution studies, they should be applied with caution and potential bias corrections particularly at local and smaller scales.
- Preprint
(2089 KB) - Metadata XML
-
Supplement
(1001 KB) - BibTeX
- EndNote
Status: open (until 03 Aug 2025)
-
CC1: 'Comment on egusphere-2024-4150', Colleen Mortimer, 27 May 2025
reply
This paper seeks to understand the performance of snow estimates from ERA5-Land over Canada. Canada-specific analysis of the performance of snow products is lacking in the broad scientific literature; this work could help fill this knowledge gap.
The authors evaluate 3 snow-related variables from the ERA5-Land reanalysis: snow water equivalent (SWE), snow depth (SD), and snow cover fraction (SCF). For each of these three variables, they use a single reference dataset intended to represent "truth". For SCF, the uncertainty in the MODIS dataset may be low enough that a single reference product is sufficient for the evaluation. However, for snow depth and SWE, there is much more uncertainty in historical estimates. At present some of this uncertainty is irreducible, and so previous work has demonstrated the value in using an ensemble of datasets for evaluating SWE (Mudryk et al. 2015; 2025; Mortimer et al. 2020). Ensembles are helpful in providing a range of reasonable values against which outliers can be screened (especially for climatological snow mass and trends (Mudryk et al 2015, 2025)).
The author’s decision to rely on CMC as the only reference data for SD and SWE is further complicated because ERA5-Land is optimized for SWE whereas CMC is optimized for SD. Discrepancies between it and CMC may stem either from errors in the SWE values, or in the parameterizations, and the analysis presented does not distinguish which source of error is contributing to the discrepancy. Although the CMC product provides monthly SWE it is not really a SWE product. Instead, climatological snow density values from a lookup table, which don't evolve over the time series, are used to go from SD to SWE. Therefore, the CMC SWE product should not be considered as a reference 'truth'. On the other hand, SWE is the prognostic variable directly simulated in ERA5-Land, while SD is estimated using snow density parameterizations so discrepancies between it and CMC are expected. While the CMC product does assimilate ground observations, these are not available over the entire country and therefore the SD values in the CMC product represent a mixture of information from both ground observations the snow model. This means that away from locations with assimilated data, the assessed differences between CMC and ERA5-Land will just represent differences in the snow models used to produce each product.
We encourage the authors to identify a more appropriate set of SWE products to use in their evaluation and to discuss the limitations of ERA5-Land's SD estimations. Additional data could include other reanalysis datasets and/or in situ data (e.g. NorSWE (Mortimer and Vionnet, 2025; https://zenodo.org/records/15263370) for SWE, global SYNOP network or GHCN-D [https://www.ncei.noaa.gov/products/land-based-station/global-historical-climatology-network-daily] for SD) as used in Mortimer et al 2024 and Mudryk et al 2025. The strategies employed to understand the reproducibility of spatial structure are interesting, if expanded, could provide useful insight about the strengths and limitations of ERA5-Land.
Finally, we have a few minor comments about the treatment of the CMC product and the analysis regions.
Minor comments
- How were the limitations listed in Brown and Brasnett 2010 Section 3.2.1 Warnings and Notices addressed?
- How was permanent land ice accounted for?
- Given the snow densities are based on snow classes, did you consider using snow classes instead of ecoregions?
References
Mudryk, L. R., Derksen, C., Kushner, P. J., and Brown, R.: Characterization of Northern Hemisphere Snow Water Equivalent Datasets, 1981–2010, J. Climate, 28, 8037–8051, https://doi.org/10.1175/JCLI-D-15-0229.1, 2015.
Mudryk, L., Mortimer C., Derksen, C., Elias-Chereque, A., Kushner, P.: Benchmarking of SWE products based on outcomes of the SnowPEx+ Intercomparison Project, The Cryosphere, 19, 201–218, https://doi.org/10.5194/tc-19-201-2025, 2025.
Mortimer, C., Mudryk, L., Derksen, C., Luojus, K., Brown, R., Kelly, R., and Tedesco, M.: Evaluation of long-term Northern Hemisphere snow water equivalent products, The Cryosphere, 14, 1579–1594, https://doi.org/10.5194/tc-14-1579-2020, 2020.
Mortimer, C. and Vionnet, V.: Northern Hemisphere in situ snow water equivalent dataset (NorSWE, 1979–2021), Earth Syst. Sci. Data Discuss. [preprint], https://doi.org/10.5194/essd-2024-602, in review, 2025.
Sincerely,
Colleen Mortimer and Lawrence Mudryk
Citation: https://doi.org/10.5194/egusphere-2024-4150-CC1 -
RC1: 'Comment on egusphere-2024-4150', Steven Fassnacht, 27 Jun 2025
reply
General
This paper assesses thew snow information from the ERA5 Land dataset at a 9 km, hourly resolution, in comparison to Moderate Resolution Imaging Spectroradiometer (MODIS) SC data at a 5 km, monthly resolution and the Canadian Meteorological Centre’s (CMC) SD/SWE product at a 24 km, daily/monthly resolution. The ERA5 dataset is being used more and more, and this is an important assessment. There is a mismatch of spatial and temporal resolutions Between the ERA5L and evaluation datasets, and the authors use data harmonization. The final assessment is 1 March 2000 to 31 December 2020 at 25 km on an average monthly time step for October to June (monthly, seasonal, annual). Twenty-one ecological regions across Canada and Alaska were considered.
Overall, this is a lot of work and is the basis for a reasonable paper. However, the paper needs reorganization (e.g., Methods moved from Results and Discussion, improved figures, better explanation on how to interpret figures). There are issues with the “truth” data that are used. Those are not discussed. The Discussion is lacking and needs to circle back to the Introduction.
While it is understandable that the coarsest spatial and temporal resolutions are used to have consistency, the coarsens the datasets being assessed (ERA5L) by 25 times spatially and 720 times temporally. This removes some of the nuance of the finer resolution ERA5L data. The snowpack does not vary drastically on a daily basis, so coarsening to that resolution is acceptable. However, using a monthly time step and averaging the ERA5L data is a problem. The implications need to be discussed further.
While it is acceptable to use the standard deviation (Std Dev – I don’t have sigma handy on my keyboard) to assess interannual variability, this is biased when a trend is present. Specifically, SD could be large because there is a lot of interannual variability, or because there is a large trend. Consider detrending the time series to better assess the interannual variability. If you don’t address this, at least discuss the implications.
The figures are somewhat understandable but need work. I cannot distinguish some of the colors from one another, i.e., there are sets of 3 regions with almost the same color as overall there are only about 6 colors. Perhaps use a sub-set of the 21 regions Further, the rainbow color ramp is difficult for some people who cannot with visual impairment. The individual panes in each figure tend to be small, especially when 12 panes are present (Figures 2 to 5). Make each pane large, homogenize the x-axes range, and remove some of the white space and repeated axes labels. Probably add a letter to each figure pane.
At the beginning of each section in the Results, some methods are introduced in the first paragraph. These should be moved to the Methods section. There is some explanation on how to read the figures, but this seems incomplete. For example, when I look at Figure 2, is there a good shape, i.e., ERA5L is close to the “truth?” I assume that a perfect correlation between ERA5L and “truth” is a vertical line along RD* = 0. It is difficult to interpret what the different curves mean. Referring to Figure 3, the authors use CV (SD / mean) > 1 (vs. < 1), but this cannot be distinguished in Figure 3.
Figures 6 and 7 seem to be meant to explain the previous results. However, they would be better placed in the actual Results section. Similar to above, there are methods presented at the beginning of several of the Discussion sub-sections.
The Discussion does not explain the work outside of the work itself. There are no citations and the authors do not use the literature to explain their findings. As I state above, there are spatial and temporal issues and potential problems with the “truth” datasets. For example, SD/SWE are derived using the Brown et al. (2003) model, but that model has some large assumptions, such as the rain-snow threshold of +2C (this varies with climate and other factors), and the new snow density (the Hedstrom-Pomeroy equation is wrong based on the data that it is fit to). While not stated in Supplement S2 so it is unclear whether it is used in the paper, Brown 2003 do not account for precipitation undercatch and apply a 20% reduction in precipitation to account for sublimation and blowing snow. This is very important across much of the study domain (less so regions 12, 13, 15, 17, 20? – I can’t tell exactly from the colors). The actual difference between precipitation, undercatch, sublimation and blowing snow varies spatially and temporally.
Specific
- Line 33: While there is no complete agreement in snow variables names, we typically call Snow Cover Snow Covered Area, with the short form of SCA. SD is acceptable for snow depth, but d_subscript_s is often used.
- Line 39: SWE and the others are “variables,” not “parameters,” as they change over both space and time. In the next sentence you use “variables.” I recommend using the term “variables” here.
- Lines 42-55: This paragraph is relevant, but it seems somewhat superfluous, as it gives information that is already known to the cryospheric community. It is short but is somewhat of a “throw away” paragraph – consider being more succinct.
- Line 43: I question if in-situ measurements are the most “accurate?” There are various papers that talk about the point to area problem with field measurements.
- Lines 109-110 and Supplement S1: this is a good addition to explain how the Hydrology-Tiled ECMWF Scheme for Surface Exchanges over Land model works to represent the snow variables in ERA5L. Thank you!
- Lines 119-120 and Supplement S2: this is a good addition to explain the “Brown 2003” model.
- Lines 130-131 and Supplement S3: this is a good addition to explain the monthly MODIS time series data.
- Line 135: here you use the work “altitude,” but in the previous part of the sentence you say “elevation.” Elevation is the correct term. Also, be consistent.
- Lines 140, 141, etc.: Data is a plural word, so it should read “The data “are” publicly …”
- Lines 174-179: put these 21 ecological regions in a table. Also give us the area of each. The general location could be helpful, as several are blended together in Figure 1a (and are thus indistinguishable).
- Figure 1 can be improved. Figures 1c and 1d should have the same sized y-axis as they both go from 0 to 100%.
- Line 196: It is the Theil-Sen slope, not just Sen. Also, the two (self) citations provided do not speak specifically to the Mann-Kendall test or Theil-Sen slope – use appropriate citations here.
- While equations 1 and 2 are very simple, they are acceptable since X is a statistic and not a variable.
- Line 211: do you “consider” or “use” the scaled version. Here and in other locations in the text, the language is tenuous.
- Line 222: good idea to consider “brevity,” but are the result presented at the monthly scale? Figures 2, 3, 5, 6, 7 present annual results.
- Lines 224-235: This paragraph is mostly methods and should be moved to that section. More explanation on how to read Figure 2 would be useful – what shape of lines is good, i.e., ERA5L is close to the “truth?”
- Figure 2: this figure is difficult to understand, partly because the individual figures are small. I assume that the right figure is D and the let 3 are RD*? Consider use the same x-axis scales for all 12 figures so that the reader can visually compare the results (at least the same for all RD and for all D). As per my comment about the colors in Figure 1, I cannot tell regions apart. Consider how you can improve upon this – perhaps 10 representative figures. Since ECDF = 0.5 and RD* of 0 are the centre? Perhaps a horizontal dotted line across ECDF = 0.5 would help.
- Figure 3: The caption should read second moment (Std Dev) “versus” the first moment
- Figure 4: apparently the spatial structure can be some other than the 4 states listed? For example, SC for region 1 is along the dashed line for all four statistics. Does that mean that they are the same, i.e., complete agreement?
I stopped examining specifics at this point, as the general reworking of the paper is necessary before the details can be examined.
Citation: https://doi.org/10.5194/egusphere-2024-4150-RC1
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 189 | 74 | 12 | 275 | 29 | 10 | 18 |
- HTML: 189
- PDF: 74
- XML: 12
- Total: 275
- Supplement: 29
- BibTeX: 10
- EndNote: 18
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1