the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 May 2026
Assessing spatially distributed snow simulations with MEB-Crocus in subalpine forests through modelling experiments
Abstract. The growing use of physics-based snow models at sub-kilometric resolution for scientific and operational applications calls for spatially distributed model evaluations. Such are particularly challenging in forested areas, where suitable ground truth data is largely lacking. This study assesses the first spatially distributed simulations of the forest snow scheme MEB-Crocus at 250 m resolution through comparison with a benchmark model, FSM2oshd. MEB-Crocus will be integrated in Météo-France's operational snow modelling chain in the near future. Its canopy implementation is based on principles typical for land surface models intended for coarse-resolution large-scale applications, while the canopy implementation in FSM2oshd was specifically developed and validated for simulations in alpine terrain and at sub-kilometric resolution. FSM2oshd has already been successfully evaluated against spatial observations and vegetation parameters were upscaled from hyper-resolution snow-vegetation simulations, providing confidence for using it as a benchmark. A suite of modelling experiments with varying combinations of vegetation datasets and grid cell tiling was performed to enable assessment of different aspects of MEB-Crocus. In the default operational configuration, MEB-Crocus was found to overestimate snow water equivalent (SWE) at peak of winter especially at elevations where forest is present, but to simulate shorter snow cover durations than FSM2. Land cover datasets and process parametrizations accounted for a similar share of model discrepancies. With identical forest structure information, differences in canopy snow processes were the main driver of model discrepancies, with MEB-Crocus generally overestimating peak SWE especially in denser forests. Simulations with MEB-Crocus including recent enhancements to the parametrizations of canopy snow interception and unloading led to strongly reduced differences in peak SWE. Insights from these model comparisons inform future model development efforts and encourage the evaluation of spatially distributed models across a range of forest structures, topographic settings, and meteorological conditions.
- Preprint
(1705 KB) - Metadata XML
-
Supplement
(143 KB) - BibTeX
- EndNote
Status: open (until 03 Jul 2026)
- RC1: 'Comment on egusphere-2026-1464', Benjamin Bouchard, 22 Jun 2026 reply
-
CC1: 'Comment on egusphere-2026-1464', Alex Cebulski, 30 Jun 2026
reply
Mazzotti et al. present novel results from a comparison of two operational snow-hydrological models. The model of interest, Crocus, was experimented with using different land cover datasets and parameterizations. Differences in model outputs, relative to the benchmark model FSM2oshd, are used to attribute the sensitivity of Crocus to these modelling decision. The study provides an important contribution to hydrological science by quantifying the relative influence of land cover data and model parameterizations on model sensitivity. I have some comments for the authors to consider, regarding justifying the benchmark model, better describing the parameterizations, quantifying model differences, and some reworking of the text.
General Comments
1. Evaluation of the benchmark model
While those familiar with snow-hydrology modelling will generally know FSM2 to have impressive model scores, the average reader might need some evidence to back up the use of FSM as a benchmark model. I think some minor reworking to the introduction or methods section could help support the use of FSM as a benchmark.
2. More details on the model parameterizations
It would be helpful to know how the snow interception and ablation processes were represented in the models. Since the study attributes some of the difference in model performance to these parameterizations, interpretation of the results would be improved by providing some context on these equations, especially since the study referenced with these equations is not yet published. I see some of these equations are in the supplement, but are incomplete (i.e., the initial snow interception parameterization is missing). I would recommend listing the equations for the MEBCroXT_all setup in the manuscript, if length permits, and then describing how these differ from MEBCro_def and FSM. Based on this, in the discussion it would be interesting to link the interpretation of your results (i.e., Crocus to overestimate SWE in denser forest) to the process representation differences between the models. Related to this it would be revealing to have a figure diagnosing canopy snow ablation terms (i.e., cumulative sublimation, melt/drip) for the various models to better get at process representation/diagnosis. Some additional length needed for this description could be gained from removing some of the descriptions of statistics and other model details.
3. Transferability of results to other sites and climates.
The model was tested in Switzerland, however, the plan is to operationalize the model in France. How might the change in location of future applications of the Crocus model influence model performance? Are there differences in land cover type and climate that might influence model performance or transferability of parameters. Moreover, how might future climatic conditions influence model performance. For example, as forest snowfall regimes transition from being dominated by intercepted snow sublimation losses to more unloading and drip in a warmer climate. I know this isn't the main topic of this study, but it would be great to hear the authors thoughts on these ideas through the addition of a few sentences in the discussion.
4. Quantifying model differences
Much of the paper centres around comparing different model results. Currently this is done quantitatively, quite nicely in the figures, but in the text I think it would be helpful to have some additional statistics (i.e., percentage difference) to support claims and/or be more specific regarding phrasing such as "approximately equal" see L492 below and others throughout.
5. Conclusions
The concluding remarks in Section 5 regarding the attribution of land cover and parameterizations to model sensitivity differs from those given in the abstract. Some reworking of Section 5 could better emphasize these important results. This information is also hard to interpret in Figure 12, so some work on the specific comments listed below related to this could help better emphasize these results.
Specific comments
L47-49 - Much of this paragraph focuses on spatial distribution, but I think the key for this paper is getting both the spatial and temporal distributions. So some reworking of this paragraph could be helpful to mention both dimensions.
L58 - Some recent large-scale acquisitions now also available in western Canada (Bisset et al., 2026).
Bisset, R., Floyd, W. C., Menounos, B., Bishop, A., Marchenko, S., Fisk, G., Cebulski, A., Marshall, P., Heathfield, D., Beffort, S., White, R., Arriola, S. G., & Viner, N. (2026). Snow Water Storage Within Eight Pacific Coastal Watersheds in British Columbia (Canada) Inferred From Four Years of Airborne Lidar Data. Water Resources Research, 62(5), e2025WR040837. https://doi.org/10.1029/2025WR040837
L76-77 - I think its fine to use FSM2oshd as a benchmark as it permits this interesting model evaluation. However, I think other readers that are not familiar with snow modelling might need more convincing. I think some text here could be warranted to quantify the existing performance of FSM2oshd to justify the use as a benchmark.
L80 - I see Q1 was discussed in 4.1 which I think is an informative discussion, but it is not mentioned again in the conclusions, is this needed as a specific research question?
L127 - Since a main theme of this paper focuses on how process representations influence the results, I think a description of the mass balance parameterizations for intercepted snow is warranted here.
L172-181 - See general comment #1.
L189 - The Quéno et al. (2024) study is not referenced again, consider whether this ref is needed here.
L284-290 - These lines introduce the figures but do not provide interpretation of the results, consider whether this paragraph is really needed.
L193 - Some more detail on the actual tree species would be useful here. Snow interception model development studies are limited to specific species. So additional information on the tree species here could help interpret the results.
L212 - Change word <Thank> to <Thanks>. Also consider numbering strategy here, starting with 1 might be more appropriate.
L243 - Is <defined as in Mazzotti et al. (2023)> needed here?
L259 - How is drainage of rainfall or canopy snow drip handled in the subcanopy snowpack? Is it included in the melt term? Please make sure this mass balance is valid.
L269-271 - <However, given the many dimensions...model results is indispensable.> Is this sentence necessary? I think the next sentence still reads okay without it.
L330 - First sentence missing figure reference?
L336 - Check wording on <reveal the two models to simulate>.
L492 - <Vegetation data and process parametrizations contribute in approximately equal>. They do not look equal in Figure 12, can you clarify? A percent difference calculation could be helpful here and throughout this paragraph to support claims on the degrees of model sensitivity.
L493-495 / Figure 12 - It would be helpful to have lettering on the panels of Fig. 12 as in the other figures. The bottom left plot seems out of place, based on the model descriptions this plot would represent vegetation dataset and parameterization differences, and yet the column name suggests it is vegetation effects. The change between the <MEBCro_def-MEBCro_all> and <MEBCro_def-MEBCroXT_all> comparisons seems to suggest there is a large sensitivity in changing the parameterizations, which is not in agreement with the final sentence referenced here. Please clarify. It might be useful to add in a <MEBCro_all-MEBCroXT_all> comparison as well to better illustrate the parameterization change sensitivity, although this might just be expected to be similar to the <MEBCro_all-FSM_all> comparison?
L520-524 - See major comment #2. Can you support this claim with a process diagnosis for the different models showing different levels of throughfall, unloading, sublimation, and melt/drip? Was there any attribution of a chance in the initial loading of snow in the canopy as well? I agree without measurements, direct assessment of the parameterizations is difficult, but it is still possible to do a process diagnosis which I think would be interesting.
L524-525 - There is a new study from Cebulski et al., (2026, hydrological processes) where this new approach shows improvements in continental, subarctic, and especially maritime needleleaf forests where melt/drip dominates.
Cebulski, A. C., Pomeroy, J. W., & Floyd, W. C. (2026). An Evaluation of New Snow Interception and Ablation Parameterisations in Continental, Subarctic and Maritime Needleleaf Forests. Hydrological Processes, 40(6), e70573. https://doi.org/10.1002/hyp.70573
L543 - I think its worth mentioning here that sub-canopy snow accumulation measurements are rarely undertaken (and more are needed) and also not possible from regional-scale remote sensing methods. This can also help justify the use of the benchmark model.
L545 - Remove double brackets from citation.
L568 - <yielding larger peak SWE than FSM2oshd> What processes contributed to this difference?. Fix word <fore>.
Several Figures - Several figures are quite blurry. Please increase the quality / resolution.
Thank you for the opportunity to review this manuscript. The comments above are yours to consider.
--
Alex Cebulski
Postdoctoral Fellow
School of Resource and Environmental Management
Simon Fraser University
8888 University Dr., Burnaby, B.C., V5A 1S6Citation: https://doi.org/10.5194/egusphere-2026-1464-CC1
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 164 | 63 | 11 | 238 | 31 | 15 | 11 |
- HTML: 164
- PDF: 63
- XML: 11
- Total: 238
- Supplement: 31
- BibTeX: 15
- EndNote: 11
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
In their manuscript entitled “Assessing spatially distributed snow simulations with MEB-Crocus in subalpine forests through modelling experiments”, Mazzotti et al. compared different configurations of MEB-Crocus and FSM2oshd over a forested domain of eastern Switzerland with complex topography running distributed experiments for 7 winter years. This work reveals important discrepancies between two land-surface schemes that the authors attribute to each model’s canopy snow process parameterization and to the quality of the canopy structure database used in the modeling setup. Overall, this manuscript is clear, well-structured and written in a concise and straightforward manner, which makes it easy to read. I especially enjoyed going through the Discussion which is truly insightful. The scientific content is solid, and the methodological framework is also robust. I only have a few general comments and a few more specific and technical recommendations that, I believe, would benefit this manuscript, which is already of good quality.
General comments:
1. The framing of FSM2oshd.
While the title suggests the prevailing role of MEB-Crocus in this work, the manuscript itself rather presents FSM2oshd as a key component of this work, more than just ‘any’ benchmark model to assess MEB-Crocus. We see it already in the Abstract (ll. 15-19), in the first research question (ll. 81-82) and in some other instances in the Results section where FSM2oshd is sometimes presented in front of FMS2-Crocus (e.g.: ll. 301, 308, 326-327, 334, and others). From my point-of-view, this work is a comparison of three canopy-snow schemes: FSMoshd (which is has proven to be successful for representing canopy-snow processes a high resolution), MEB-Crocus (which is set to be used within Météo-Frances operational framework) and MEB-CrocusXT (which is an improved version of MEB-Crocus for canopy-snow parameterization). I think that framing the Title and the research questions/objectives of the study accordingly, and adjusting slightly the structure of the Result section would suit better the narrative of your manuscript.
2. Where are the equations?
I must admit that I was surprised not to see a single equation in a manuscript centered exclusively around modeling results. Fair enough – MEB-Crocus (Boone et al., 2017; Napoly et al., 2020 and Lafaysse et al., 2025) and FSM2oshd (Mazzoti et al., 2020; Essery et al., 2025) have already been described in previous papers which are cited accordingly. Yet, given that the respective process parameterizations of each model partly explain some of the discrepancy between simulations, I encourage the authors to include the main equations governing canopy-snow processes in the main manuscript (perhaps in an Appendix). I believe that this would help in the interpretation of some of the observed discrepancies.
As for the parameterization of MEB-CrocusXT, although this is likely going to be described in detail in an upcoming manuscript (Pauze et al. in prep) and is for now only presented in a Supplementary Materials, I recommend moving this content to an Appendix of the main manuscript so the reader can refer to it more readily.
3. Presenting showdowns in the Methods.
Each subsection of the Results presents a comparison between FSMoshd and MEB-Crocus (or MEB-CrocusXT) under different configurations (open, forest, both). These step-by-step “showdowns” are essential to the narrative of the manuscript and each of them have a specific purpose. Therefore, I suggest introducing these comparisons in a dedicated subsection of the Methods which, I think, would make the Result section even more straightforward. This implies, for instance, rephrasing partly and moving ll. 281-285, 318-320, 350-353, 417-419 to the Methods.
Distribution of the grid cell classes.
If I am not mistaken, results are aggregated into 48 classes (6 elevation bands and 8 aspects). A major part of your results is analyzed based on the simulated SWE in those classes. However, as you show on Figure 1 (right of top row), the distribution of grid-cell elevation is not equal for each elevation range, and I would assume the same for the aspect ranges. I suggest presenting in a polar plot the distribution of all the 48 classes. I think that it would bring extra precision to the interpretation of the results with respect to the influence of topography on modeling results.
Specific and technical comments:
ll. 17-19. I don’t feel like this sentence is necessary in the abstract. I would rather expand on the sentence presented on ll.19-20 which provides information on the methodology.
l.22. Should it be FSM2oshd (instead of FSM2 only) ?
l.25. I would specify in one short sentence what are the enhancements to the parameterizations.
ll. 46-67. In the second paragraph of the Introduction, modeling or observation scales as well as resolutions are often mentioned. (i.e. ll. 46, 51, 57, 59, 60, 62. 62). I suggest to give examples of order of magnitudes of the scales (large scale versus landscape scale vs operational resolution, etc.)
l. 61 and all other corresponding instances. Make sure to refer to the final publication of Haagmans et al (2025).
l. 73 and l. 77. Maybe it is just me, but I found that refereeing to upcoming sections of the manuscript in the Introduction is slightly peculiar. Can you replace that by existing publications or publicly available reports?
l.127, 232, 418. Whether or not you include the detail of the MEB-CrocusXT parameterization in the manuscript, I recommend replacing the reference to Pauze et al paper in preparation by the work that was presented at EGU26 (https://doi.org/10.5194/egusphere-egu26-6561).
l. 149. Please specify how many ‘few (numerical) layers’ is.
l. 153. Although, it is mentioned later (l. 201) I would specify the resolution of COSMO here since you mention the resolution of the model runs right before.
Figures. Maybe it is something that will be corrected by the editing team, but make sure that the quality of all figures (resolution-wise) is good enough. Currently, the resolution of most figures is underachieving.
Figure1. I would add the lat-long tickmarks on one of the maps.
Figure 3. It took me some time to get that the SWE (black curve) was the same on each frame. Please make sure that this is clearer in the revised manuscript. Also, I would recommend specifying the panel ID (e.g. a), b)…) on the figure. This also applies to other figures.
Figure 4. On the timeseries, I would recommend to plot the median instead of the mean, which is more coherent with the envelope that shows the interquartile range. Also, consider showing a time series of the SWE difference between the two experiments. I think that this would allow to grasp more easily the seasonal dynamics (accumulation vs ablation) rather than inferring it from the polar plots above.
l. 299. Just to be sure, is it 13% of the average peak SWE as simulated by MEBCrocus? Maybe it would be worth specifying.
l. 330 (Figure 5). Why don’t you show all years? Is there some different behaviors of the models in other years? If you decide to show specific years, I recommend presenting complete time series in Supplementary Materials. This applies to the following subsections of the results.
Figure 6. Was it intended not to present results from simulations configures with a LAI between 3.5 and 4.5?
ll. 375-377. I think that this is a good reason to present other WYs (at least in Sup. Materials).
Figure 9. I have some trouble reading Figure 9c). This is maybe just me, but I find the solid, the dashed and the dotted line very similar to one another for a given color. Moreover, the legend only presents the solid line. I recommend using specific colors to each line (Perhaps other colors than purple and yellow, which are already used to present WY18-19 and WY19-29 on Figure 7).
Figure 10. I think that ‘XT’ is missing in the name of the MEBCro simulations in the legend of 10d. Please double-check.
ll. 425-427. I was surprised by the strong increase in discrepancies in snow duration considering the better match in peak SWE and ablation rate on Figure 10b. The explanation provided here is however satisfying to me. Nonetheless, I feel that this may be due to the definition of the start of the snow season day (SSD) and the snow disappearance day (SDD) which rely on thresholds of 10 mm of SWE (see ll. 251-253). Can you elaborate on the importance of the choice of that threshold in the context of increasingly frequent ephemeral snowpacks at low elevations and how it affects the interpretation of the results.
ll. 466-468. The treatment of the snow cover fraction in distributed land surface models is a major challenge for any modelers aiming to better represent cold region water and energy balance beyond point-scale. In fact, the snow cover fraction in LSM impacts the heat transfer with the atmosphere (or the overlying canopy) through the albedo or the surface temperature as much as the heat fluxes to the ground. I appreciate that you address this important factor in their discussion. I would be interested to learn more about your ideas to improve this critical component in MEBCrocus (or even FSM2!). Can you elaborate on ways to improve the sub-grid representation of the snow cover fractions in the context of this work?
ll.774-777. This reference appears twice in the bibliography. Please revise the corresponding entries in the manuscript.
As already mentioned, this was a pleasure to read your manuscript. I hope that my suggestions, if judged relevant, will contribute to enhance this work (which is already quite strong). If the authors have any questions regarding my comments, or if they would like to discuss in more detail some of the challenges inherent to land surface models applied to cold environments, they should feel free to reach out.
Benjamin Bouchard
benjamin.bouchard@ec.gc.ca