the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Evaluation of annual maximum snow depth data estimation from the European-wide reanalysis C3S MTMSI (Copernicus Climate Change Service – Mountain Tourism Meteorological and Snow Indicators) against in-situ observations
Abstract. Large snow load events are a major hazard for both human societies, in particular buildings and transport safety, and natural ecosystems. National and European frameworks provide guidelines and standards in order to take into account extreme snow load hazard in infrastructure design. However, there is a lack of reference data for their implementation. This is even more challenging in the context of climate change, which modifies the frequency and intensity of major snow load events. In the context of the Framework Partnership Agreement on Copernicus User Uptake, we have developed a pan-European extreme value analysis of annual snow load maximum based on the Mountain Tourism Meteorological and Snow Indicators (MTMSI) dataset available on the Copernicus Climate Change Service. This dataset includes reanalysis data, based on the UERRA (Uncertainties in Ensembles of Regional Reanalyses) reanalysis and snow cover simulations, and past and future climate projections based on regional climate simulations. Here we describe the evaluation of the MTMSI reanalysis component in terms of annual snow depth maxima against multiple in-situ observation datasets. Results are provided at the NUTS-3 (Nomenclature des unités territoriales statistiques) scale used in MTMSI, for multiple elevations, over a large area stretching from the European Alps to the Scandinavian countries. We highlight satisfactory skills of MTMSI annual snow depth maxima on most NUTS-3, based on the Kling-Gupta Efficiency metric, correlation, and bias scores. We identify some areas where MTMSI does not adequately portray in-situ observation of snow depth maxima, located in the Alps, and coastal areas of the Netherlands, Norway, Sweden, and Croatia. This study thus provides background information for assessing the relevance of this pan-European dataset in terms of annual snow depth maxima, relevant for annual snow mass and snow load maxima based on complementary information based on snow cover model output. The MTMSI annual maximum snow depth reanalysis dataset is available through the following link: https://doi.org/10.5281/zenodo.15181401 (Kamir et al., 2025).
- Preprint
(5209 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 13 Aug 2025)
-
RC1: 'Comment on essd-2025-225', Michael Matiu, 30 Jun 2025
reply
Kamir et al. evaluate maximum snow depth from MTMSI, a reanalysis-based snow indicator dataset over NUTS regions, with respect to in-situ snow depth observations. The paper has high quality graphics and is well readable. Their analysis is able to guide extreme snow loads identification and future assessments, which is of societal relevance.
The study has a few strengths:
-
The evaluation covers a large geographic area in Europe, from the Mediterranean to Scandinavia.
-
The study is interesting and worthy of publication because it looks at extreme values, while most previous studies only look at averages. Since the behavior can be different comparing means or extremes, it is highly relevant, also because impacts are much higher for extremes than means.
However, also a few issues as outlined below. Finally, I’m not sure if the study lies within the scope of ESSD, since the authors do not produce/describe a dataset, but instead just extract values from an existing one. But this is for the editor to decide.
Also please disregard the formal manuscript rating in the editorial system, since the points there refer to a novel dataset, and thus 2) significance and 4) data quality cannot be meaningfully answered for this study.
Major points:
-
While the evaluation of extremes is relevant, the analysis performed in the study is at times superficial. I acknowledge the complicated structure of the MTMSI dataset, which makes it challenging.
-
The elevational analysis, for example, is challenging to understand in the current form. It is unclear how many stations/regions at which elevations in which locations were used/considered. Maybe it would be useful to distinguish between plain and mountain NUTS. Also the analysis needs to somehow consider the latitudinal gradient.
-
Station data are used multiple times for the different NUTS/elevation groups. While I understand the author’s needs to cover as much as possible of the MTMSI dataset, this still feels like inflating the analysis or the number of observation pairs. There is some discussion at the end, but only based on one example. This might deserve some more thinking.
-
Negative bias at high elevations (Fig 6) is not discussed in detail.
-
-
Some comparison of extremes versus means would be highly beneficial. Also to put the previous studies in context. While this could be done just discussing the numbers of previous studies with the ones here, alternatively the analysis, or parts thereof, could be repeated for means.
(some of the above might be repeated/further explained below)
Minor points:
-
L11 “satisfactory” means? Some number would be helpful
-
L29ff, this paragraph is more a description of methods, not introduction
-
L53 Monteiro and Morin did not use remotely sensed snow depth, only in-situ snow depth (and remote sensing snow cover fraction)
-
L55 again, “satisfying” is vague. … I see you use this phrase a lot, if possible please add a quantitative number for clearness.
-
L78 based on what criteria were plain and mountain NUTS3 distinguished?
-
Why did you not use the NH-SWE dataset, which is basically ECAD with quality checks. https://essd.copernicus.org/articles/15/2577/2023/
-
Related: The study needs some explanation/discussion why there have not been used tools such as the delta_snow model or HS2SWE, which convert daily time series of snow depth to SWE and vice versa without further input. Snow depth is a good proxy for SWE, but still, the sensitivity of results to the chosen approach could use some further analysis and/or discussion.
-
-
L178-181 Unclear, please reformulate or expand.
-
Evaluation metrics:
-
Why did you not consider a measure for spread, like RMSE or MAE?
-
Besides absolute bias, I recommend investigating also the relative bias, which is often a more useful metric for zero-bounded variables.
-
Since you are looking at extremes, why did you not consider extreme value theory, or metrics based on GEV distributions, return levels, or similar?
-
-
Fig 5a, better if you switch red and blue colours, for easier visual perception (red drier, blue wetter)
-
L227 you mean lower instead of larger?
-
For a better understanding from your readership, I suggest adding a map of the mean maxima over the NUTS regions, so the readers can also put the bias values in perspective. In addition to also showing the relative bias (see commment above).
-
Fig 6: are there spatial pattern to this? I would recommend to split at least by latitude bands, since 1000m in the Alps is very different to 1000m in north Scandinavia.
-
Figure 7 would be better suited further up, even in methods or beginning of results, to explain the approach. Also, it would be interesting to see the time series of the not-so-good NUTS/elevation pairs. … ok, I see, this comes later with Fig 10.
-
Sec 3.3 unclear; also why only correlation is shown and not bias. Is this related to regions > 2 stations?
-
Sec 4.1: Vague, since MTMSI has already been evaluated on similar but slightly different variables, please put your results in more quantitative comparison.
Citation: https://doi.org/10.5194/essd-2025-225-RC1 -
-
RC2: 'Comment on essd-2025-225', J. Ignacio López-Moreno, 04 Aug 2025
reply
The manuscript presents the validation of annual maxima snow depth from a snow product covering large parts of Europe. The manuscript is well-written, the methodology is clear and well-suited to the purpose of the study, and the limitations of the dataset are honestly discussed. Even if the data set shows evident problems at high elevation, and only annual maxima is a snow parameter a bit limited for many research and applied uses, all data about snow is welcome for the community, especially when covering a very large area of Europe. Thus, I recommend the publication of the work in ESSD after considering a few changes to be considered for a revised version.
-1- If not the title, the abstract should show the period covered by the dataset.
-2- In methods, I wonder if it would be more interesting to use a relative bias than the bias itself (or present both). In part, the relative low bias at low elevation can be explained by low snow depth values. If bias is divided by the average value, and expresend in % or 0-1 units it could provide a better representation of the validity of the dataset.
-3- Along theresults section there are many qualitative references about the accuracy estimators instead of giving the values or ranges of values (to say that a r value of 0.6 is "satisfactory" is very relative..).
- 4- Figure 2 shows stations in Catalonia, but is not mentioned anymore. At some point is mentioned Andorra (small but relevant in terms of snow and snow loads), but it is not shown in the figures (i.e. could be zoomed in figures 4 and 5).
- 5- In discussion, I have the feeling that large errors at high elevation are softened arguing that most of constructions are at low elevation. But many critical infraestructures are also at high elevations and is where snow loads often represent a big problem. Just to mild the assessment.
Hoping my comments will result useful to authors,
Citation: https://doi.org/10.5194/essd-2025-225-RC2 -
RC3: 'Comment on essd-2025-225', Anonymous Referee #3, 05 Aug 2025
reply
The paper evaluates an existing annual maxima snow depth product, from the C3S MTMSI, against in situ snow depth observations to ensure its validity, going a step further than in previous assessments. The original dataset appears to be used for various purposes, including infrastructure design, and therefore, I find the comparison carried out extremely useful to ensure that the extreme snow values used are trustworthy for this purpose. In addition, the study covers a large range of latitudes within Europe, providing the evaluation with an additional strength.
I am not sure that the paper fits in the scope of the journal since it does not present a new dataset but rather carries out an evaluation over an existing one. In any case, I have some concerns before its final publication:
- I miss in the abstract a reference to the length of the dataset. Additionally, the reader would appreciate a brief description of the NUTS-3 spatial resolution. Personally, after reading the manuscript, I realized that it is an administrative regionalization, but it was unclear to me before seeing Figure 1.
- The methodology used by Morin et al. (2021) to derive C3S MTMSI is explained twice, both in the introduction and the method section. I think this is repetitive.
- Could you elaborate a bit more about the error/uncertainty you are assuming when choosing just a single annual value from each in situ station?
- The way in which the annual maxima snow depth pairs are chosen for representing the elevation bands in the NUTS-3 areas is not fully clear. What happens if there is more than one situ observation per elevation band? And the opposite, only one observation per NUTS-3 region in a mountainous site?
- The representation of the results of the assessment of annual maxima snow depth by elevation band are not sufficiently clear. I understand that it is difficult to spatially show these results, which are a combination of both NUTS-3 and elevation. I suggest to show the effect of the elevation band on the annual maxima snow depth performance by mountain range/area. I understand high elevations are mainly located in the Alps, but do they have the same performance in low elevation from Germany, the Netherlands, or Scandinavia?
- KGE is a nice metric to evaluate high values in time series. In general, -0.41 can be considered the threshold of acceptable performance, as you have highlighted in Figure 4. However, since the higher the KGE, the better the performance, I would recommend using a more contrasted color ramp palette to understand exactly which KGE values correspond to each of the blues. In addition, since KGE is a metric composed of three other metrics, I would explore the performance using these three other metrics. You are already using the correlation coefficient, but not the relative errors (beta in the manuscript KGE equation) and the relative error on the deviation (gamma, in the manuscript KGE equation). I would recommend using them rather than the absolute error.
- In the discussion section, I would comment further the bias found at the higher elevation and also include a sentence referring to the error made by the in situ snow depth measurements.
Citation: https://doi.org/10.5194/essd-2025-225-RC3
Data sets
The European-wide Mountain Tourism Meteorological and Snow Indicators (MTMSI) dataset : annual snow depth maxima E. Kamir et al. https://doi.org/10.5281/zenodo.15181402
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
209 | 43 | 25 | 277 | 16 | 26 |
- HTML: 209
- PDF: 43
- XML: 25
- Total: 277
- BibTeX: 16
- EndNote: 26
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1