the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 12 May 2025
Simulating snow properties and Ku-band backscatter across the forest-tundra ecotone
Abstract. Sophisticated snowpack models are required to provide accurate estimation of snowpack properties across the forest-tundra ecotone where in situ measurements are sparse. As snowpack properties strongly influence radar scattering signals, accurate simulation is crucial for the success of spaceborne radar missions to retrieve snow water equivalent (SWE). In this study, we evaluate the ability of default and Arctic versions of Crocus embedded within the Soil, Vegetation and Snow version 2 (SVS2-Crocus) land surface model to simulate snowpack properties (e.g. depth, density, SWE, specific surface area; SSA) across a 40-km transect of the Northwest Territories, Canada, using two winter seasons (2021–22 & 2022–23) of in situ measurements. An ensemble of simulated snowpack properties (120 members from default and Arctic SVS2-Crocus) was used in the Snow Microwave Radiative Transfer (SMRT) model to simulate Ku-band (13.5 GHz) backscatter. Modelled backscatter using multi-layer SVS2-Crocus snowpack simulations were compared to backscatter using a simplified 3-layer radar-equivalent snowpack. Results highlight that Arctic SVS2-Crocus wind-induced compaction modifications were spatially transferable across the forest-tundra ecotone, reducing the RMSE of surface density by an average of 29 %. Basal vegetation modifications were less effective in simulating low-density basal snow layers at all sites (2022 & 2023; default RMSE: 67 kg m-3; Arctic RMSE: 69 kg m-3) but were necessary to simulate a physically representative Arctic density profile. SVS2-Crocus underestimated SSA leading to high errors in the simulation of snow backscatter (2022 & 2023; default RMSE 3.5 dB; Arctic RMSE: 4.8 dB). RMSE of backscatter was reduced by implementing a minimum SSA value (8.7 m2 kg-1; 2022 & 2023; default RMSE: 1.5 dB; Arctic RMSE: 1.5 dB). A radar-equivalent snowpack was effective in retaining the scattering behaviour of the multi-layer snowpack (RMSE < 1 dB) providing a means to estimate SWE with increased computational efficiency.
Competing interests: Some authors are members of the editorial board of The Cryosphere.
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.- Preprint
(2779 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 17 Jul 2025)
-
RC1: 'Comment on egusphere-2025-1498', Anonymous Referee #1, 13 Jun 2025
reply
General
The study compares simulated snow parameters using SVS2-Crocus to in situ measurements at a study area in the NWT, Canada. Two versions of the model are applied; the default model and an Arctic-specific modification. Several locations with differing vegetation and snow conditions are analyzed. Furthermore, a forward model is used to simulate microwave backscatter from SVS2-Crocus outputs, comparing these to backscatter simulations using the in situ data directly. A microwave-effective snowpack concept, which aggregates the data to three representative layers, is used. The paper is of interest to the scientific community as it has become clear that some type of fusion of modelled and remote sensing information is required for successful retrieval of SWE from spaceborne radar. This concerns, in particular, approaches using Ku-band SAR backscatter. This paper takes some steps to attempt to quantify how an advanced coupled land-surface -snow process model reproduces natural snowpacks in a challenging environment, and what are the implications on (simulated) backscatter. As such, the study is worthy to consider for publication.
The study is also generally well written although some specifics in the methodology are hard to grasp and require several readings. For example, from the abstract it was not at first obvious that actual observations of backscatter are not used, only simulations. Furthermore, it is hard to discern where exactly the three layer aggregates (radar equivalent snowpack) where used: in simulations based on SVS2-Crocus, in simulations from snowpit data, or perhaps both? Terminology referring to the sites also changes occasionally, sometimes referring to the biome (tundra, forest), sometimes to the names defined in Figure 1. These are the main examples, but the complexity of the diverse model settings makes the paper somewhat hard to follow and requires particular care in describing the experimental setups.
My main concern, however, is related to the usefulness of the backscatter simulation setup itself. The SMRT model is treated as a black box, testing which kind of numbers come out with each version of the input data. It seems that the pit data forcing is used as the “truth”, with SVS2-Crocus -based simulations representing deviations (“errors”) from this. No real effort is placed on which parameters actually induce these differences, beyond testing different approaches for tuning the optical grain size/SSA. Since the study is based on only simulated backscatter, one could expect a thorough sensitivity study on different parameters, or something similar. Perhaps using the 120 ensemble members in SVS2-Crocus makes this a challenge; however, you could then consider dropping the ensemble approach, and use individual, controlled simulations?
Further, some confusion is created by first choosing to fix a scaling parameter in the forward model (the polydispersity factor K) at unity, only to re-introduce another scaling parameter with basically the same end result, as adjusting K would have had. I realized only after some time that the idea is to try to scale the SVS2-Crocus SSA to observations; however, the choice of the scaling factors tested seems arbitrary. Of some interest could be to try to derive your own optimal scaling for 1) SSA 2) density 3) possibly another parameter, such as snow depth, and with the (average) optical scaling reconduct the simulation exercise.
Please see major comments below for specifics.
Major comments
- Abstract: by reading only the abstract, it is not fully clear that you do not actually use observations of backscatter: “Modelled backscatter…were compared to backscatter using…” please make it more explicit from the start that this is a simulation exercise.
- Abstract, line 25: “leading to high errors…” this is misleading again, as an “error” in simulated backscatter implies you have measured the actual one. Rather, this is a root-mean-square difference between two simulations with different model forcing. I would suggest to change the terminology throughout the manuscript, also changing RMSE(rror) to RMSD(eviation)
- Introduction: this section is very nicely written
- …except that in the last sentence, you should already make clear which (Crocus outputs, field data, or both) are converted from multi-layer to the radar-equivalent 3-layer setup. Now this is not clear at all.
- Figure 1 is very good and informative. However, can you add a scale for the snow depths? I guess the average snow depth can be indicated just as easily as the relative depth?
Throughout the paper, please choose which name you use when referring to sites. Now, sometimes “forested sites” and “tundra” are used, sometimes “Havikpak” etc. You could also use always both to be explicit e.g. “Upper plateau (tundra)”
- p9, eqs 1 and 2 and associated text. As pointed out in the above, it is unclear from the text why the K parameter is fixed here, while later on you choose to scale the SSA. I realize this is due to the experimental setup of comparing “simulations vs. simulations” , and this should be clearly mentioned. You should display the scaling factor you use in eq. 2 (and maybe forget about K, since essentially you do not use it).
I also fail to see why the scaling factor of 0.63 derived in a paper from 2011 could be relevant. I would suggest another approach: 1) derive your optimal scaling factors required to match SVS2-Crocus to field data 2) do this at least for SSA, density and snow depth 3) analyze which is the most important factor to scale (which has the largest impact) for each surface type. As an optimal case, you could test scaling all three+ parameters to match the field data, and see how much RMSD remains from variability in the ensembles.
Again, it is not clear to which data the n-layer -> 3-layer conversion is applied. Apparently to SVS2-Crocus at least, but is it applied to field data as well? Or are filed data always simulated “as is”?
Whole section 4.0 on results. It is quite tedious to read endless RMSE values in the text, and the point is quickly lost. This really gets out of hand e.g. on p. 16. Please tabulate the results, refer in the text to the tables, highlighting only the most important results.
- Figure 4: you could comment in the text that the variability of Arctic SVS2-Crocus snow depth is much reduced compared to the Default run for most sites. Why is this?
It is also not clear against which SD data the RMSEs are calculated against. Magnaprobe, pit, or both?
Same question as above for SWE. are the RMSEs against Pit or SWE Tube values?
For density, can you not calculate a reference value from SWE tubes? Do you have the snow depth from the SWE core site recorded?
- Figure 8. The text refers to SSA, but optical grain diameter is shown. Please use one or the other.
- Figure 9: the figure sort of captures why applying just a random scaling factor does not tell us much. Please consider, as suggested, making equivalent scatter plots by optimally scaling different variables. This will tell you at least which parameters carry the most weight, possibly informing where future modeling efforts should focus on.
- p19, lines 377-380. The two methods are already described in the Methods section, not necessary to repeat here. As indicted before, I do disagree with the usefulness of the methods.
- p19 line 385 “simulation of weaker backscatter that is more representative of measurements” Which measurements? This again seems to refer to measurements of backscatter.
- p19 line 394. “A radar equivalent snowpack…” reading between the lines, I realized from this sentence that the pit data were used “as is” and the radar-equivalent approach was not applied to them. However, taken out of the context of the rest of the paper, this sentence again seems to imply that you had some backscatter measurements to compare with (“…replicate the scattering behaviour of the multi-layered snowpack”). It should be made clear throughout that you are comparing simulations with other simulations.
- p19 line 398 “vertically averaged scaling factor” What is this?
Minor comments
- P8, line 194. Please clarify how exactly the division into WS and DHS was made (values).
- Figure 7: please switch the order of Arctic and Default in the panels. Logically, since default is the starting point, it should be displayed on the right (also to match Figure 4). explain acronyms WS and DHS also in the figure caption.
- p20 line 406-407 “to assess the reliability of Ku-band backscatter SWE retrievals”. This statement may be the underlying reason of the study, but you do not really go into retrieval so it is also misleading. Please reword, e.g. “assess the reliability of forward model simulations driven by SVS2-Crocus, a crucial factor considering inversion of SWE from backscatter observations” or something similar.
- p23 line 505 large grains result in increasing exponential correlation length, not the opposite.
Citation: https://doi.org/10.5194/egusphere-2025-1498-RC1
Data sets
Meteorological forcing data, SVS2-Crocus and SMRT simulated output and measurements of snowpack properties (2022-2023) Georgina Woolley https://doi.org/10.5281/zenodo.15091024
Model code and software
SVS2-Crocus Model Code Georgina Woolley et al. https://doi.org/10.5281/zenodo.15091095
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 115 | 25 | 7 | 147 | 4 | 6 |
- HTML: 115
- PDF: 25
- XML: 7
- Total: 147
- BibTeX: 4
- EndNote: 6
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1