High-Resolution Monitoring of the Snow Cover on the Moroccan Atlas through the Spatio-Temporal Fusion of Landsat and Sentinel-2 Images

Round 1

Reviewer 1 Report

Comments on the “High-Resolution Monitoring of the Snow Cover on the Moroccan Atlas through the Spatio-Temporal Fusion of Landsat and Sentinel-2 Images”

 

The email for all the authors should be added after the affiliation

 

The reference would better not do as the subject in a sentence. Such as, Page 3, first paragraph.

 

A deeper discussion should be added about the different method comparison.

 

When we fuse the different satellite data, the noise would be added. Therefore, how do you address this challenge?

 

Compared with previous method, I find that the work does not seem to be very innovative.

 

The cloud on Landsat satellite images is extremely easy to confuse with snow, and it is difficult to differentiate using both NDSI and SVM, so I think the author needs to further improve the article method.

The different sensor will show difference of spectrum, how do you address the difference between the different sensor? I think the author should give a more deeper discussion about their main method, as we can see the NDSI method is general for the remote sensing researchers. Therefore I give the major decision.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The reviewer would like to thank the authors for this thoughtful manuscript. This work has good potential. The authors are requested to put in some additional efforts to improve the quality of this manuscript. 

 

Introduction

The authors are requested to highlight the importance of monitoring snow cover for different areas of applications and cite the following articles highlighting the interdisciplinary implementation such as natural disasters and hydrological modelling.

-Tsai, Y.L.S., et al., 2019. Remote Sensing of Snow Cover Using Spaceborne SAR: A Review. Remote Sensing.

-Qiao, H., et al., 2021. A New Geostationary Satellite-Based Snow Cover Recognition Method for FY-4A AGRI. IEEE JSTARS.

 

Radiometric Normalization

The authors are requested to explain how they normalize the radiometric differences between the Sentinel-2 and Landsat-8 sensors while computing the NDSI. What are the steps to make NDSIs from different sensors comparable?

 

Snow Cover Under Forest Canopy

The authors are requested to cite recent articles discussing the role of forests in limiting the accuracy of snow monitoring using optical satellites. 

-Kostadinov et al., “Watershed-scale mapping of fractional snow cover under conifer forest canopy using lidar”, RSE, 2019.

-Muhuri et al., “Performance Assessment of Optical Satellite-Based Operational Snow Cover Monitoring Algorithms in Forested Landscapes”, IEEE JSTARS, 2021. 

Hence, the authors are requested to mention that forests are both crucial in determining the accuracy of snow detection in such environments. Furthermore, the authors are requested to draw similarities in their findings with the accuracy of retrieving snow cover at different forest canopy densities, if there is any in their study area. What methods do they implement for this?

 

Fractional Snow Cover

The authors are requested to include the following article employing high-resolution optical data for fSCA determination. 

-Gascoin et al., “Estimating Fractional Snow Cover in Open Terrain from Sentinel-2 Using the Normalized Difference Snow Index”, Remote Sensing, 2020.

Please discuss the scope of including the simple technique presented by Gascoin et al., 2020 in the methodology proposed in this investigation.

 

Forest Density Maps

The authors are requested to cite and discuss the forest density map that is widely used and can help in determining the percentage of fractional snow cover in vegetated areas. 

-Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Thau, D., Stehman, S.V., Goetz, S.J., Loveland, T.R. and Kommareddy, A., 2013. High-resolution global maps of 21st-century forest cover change. Science, 342(6160), pp.850-853. doi:10.1126/science.1244693.

 

Figures

The authors are requested to work on the presentation of the figures and avoid using unnecessary graphics as in Fig. 11. The figures presented in the manuscript are not coherent in terms of scale, font styles, and other graphical treatment. 

 

Conclusion

 

The authors are requested to highlight the key contributions/findings from this investigation. At the moment the conclusion is not detailed enough. 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Article: High-Resolution Monitoring of the Snow Cover on the Moroccan Atlas through the Spatio-Temporal Fusion of Landsat and Sentinel-2 Images

Authors: Mostafa Bousbaa, Abdelaziz Htitiou, Abdelghani Boudhar, Youssra El jabiri, Haytam El youssfi, Hafsa Bouamri, Hamza Ouatiki, and Abdelghani Chehbouni

Summary: This manuscript presents a comparison of three data fusion models that merge Landsat-8 and Sentinel-2 satellite imagery to monitor NDSI and snow covered area in the Moroccan Atlas mountains. The study region is an important water source for a large population, but is sparsely instrumented and lacks reliable, spatially distributed measurements of snow cover which are important for streamflow forecasting. The authors compare the performance of two established data fusion models (ESTARFM and FSDAF) as well as a third model (Pre-classification FSDAF) that utilizes a supervised classification step instead of unsupervised classification in the previously established model. Results show that the Pre-classification FSDAF model performs well at the data fusion task, and that the fused Landsat/Sentinel snow cover product captures snowfall/melt events at finer temporal scales than what is possible using solely Landsat or Sentinel data. References are extensive and represent both classic and recent relevant literature.

General comments: A significant number of critical details and justifications are missing from the manuscript, primarily in the description of the methodology. Specific instances are listed in the next section, but perhaps the most pertinent involves the method used to determine the optimal image pairs for data fusion (Section 2.6):

·       The fact that the Similarity Index (SI – Equation 4) computed for an image with respect to itself is not constant (and also relatively low – see Figure 5) raises questions about how useful this metric is. For example, the SI between 2021-04-18 and 2018-05-12 is 0.02, which is higher than the SI for five images when compared to themselves. This is quite counterintuitive as I would expect a derived metric to behave similarly to the correlation coefficient, in that the highest/”best” values would be achieved by computing the metric using the same image for both inputs, and would in that case always be consistent (e.g. always equal 1).

·       The way the equations for reflective difference, correlation coefficient, and similarity index (Equations 2-4) are formulated, there is no guarantee that the image pair with the highest similarity index is also the image pair with the highest correlation coefficient, but the authors state that the chosen pair maximizes both metrics. The authors must address how they would reconcile this scenario. Alternately, if there is a physical reason that this hypothetical scenario could never arise, it must be justified in the manuscript.

Additionally, one of the main contributions of the study is the Pre-classification FSDAF model, which uses Support Vector Machine (SVM) supervised classification to replace unsupervised classification in the original FSDAF model and slightly improve upon the NDSI. Important details regarding the SVM methodology are missing – what reference data was used to train the model? Was any parameter tuning required to achieve acceptable results? How was the SVM classification accuracy assessed, especially as compared to the unsupervised classification used in the original configuration? Additionally, the improvements achieved by the Pre-classification FSDAF compared to the original FSDAF seem relatively minor (RMSE decrease of 0.003 and R^2 increase of 0.01). There is significant opportunity to expand upon these results in the Discussion – for example, what effect does this improvement bring with respect to over/underestimating snow covered area, the timing of snow events, effects on streamflow forecasting, others? Providing more context in this regard would help readers understand the significance of the study.

One hypothetical reader who could benefit from this paper is a scientist hoping to apply these data fusion techniques to repeat this snow cover analysis at a different study site. This study would be significantly enhanced if the code to produce the fused data product were released in some open-source manner along with the manuscript (the Discussion mentions Google Earth Engine for example). Access to the source code would also help readers to clarify small methodological details that may not make it into the final paper.

Specific comments:

·       Abstract:

o   Suggestion to remove “poor” from sentence starting “In addition, poor atmospheric conditions…”

o   “…Pre-classification FSDAF model generates more precise fused NDSI images” – should this be “most precise” referring to all three models?

·       Introduction

o   Most paragraphs in this section contain phrases or sentences that would benefit from English language editing. Some specific instances and potential revisions are noted below, but these examples should not be taken as a comprehensive list:

§  “…providing a regular source of drinking water for irrigation…” – Confusing as written. Suggestion to list drinking water as its own resource instead of lumped into the other uses.

§  “…fills dams…” – fills reservoirs?

§  “However, although the significance of the snow-covered area for water resource management in these regions…” – Suggest to change to “Despite the significance…”

§  “…two satellites in polar orbit emplaced in the same synchronous orbit…” – typo?

§  “…due to the late availability of Sentinel-2 images.” – unclear what “late availability” means here.

§  “While an application such as snow cover monitoring…” – delete “While”

§  “On the other side, fusion models can be used…” – suggestion to change to “In addition, fusion models can be used…”

§  “…and its pairwise technique nature.” – unclear

§  “…are consistent and comparable if it is pure.” – unclear

o   Suggested addition: “…major difficulties for in situ snow cover mapping and monitoring.”

o   Suggested addition: “As an alternative, optical satellite remote sensing can provide…”

o   Suggested revision: “However, the low resolution of MODIS is not sufficient to catch the snow cover spatial variability and it limits the ability to link the presence or duration of snow to behavior…” to “However, the spatial resolution of MODIS is not sufficient to capture snow cover spatial variability, which limits our ability to link the presence or duration of snow cover…”

§  Additionally it is not clear what “behavior” means in this context – please clarify.

o   “…16-day temporal resolution is generally not suitable for monitoring snow cover in mountainous regions…” – what is the assessment of the 30 m spatial resolution for snow cover monitoring? References 23 and 24 are studies about crop type and NDVI monitoring, but it would be good to see a couple references here about snow cover or NDSI monitoring with 30 m Landsat data.

o   “This makes STARFM appropriate for only coherent regions like large area crop fields [18,43].” – the word “coherent” is confusing to me in this context as it has a specific meaning in the radar remote sensing literature (related to the consistency of land surface characteristics over a given timespan). If the word has a similar meaning here, I would think that a large crop field would exhibit some radical changes pre- and post-harvest, and also some during the growing season. But perhaps these changes may occur on similar timescales as seasonal snow cover in some areas? I checked references [18, 43] and found no mention of the word “coherent.” Please clarify or revise as necessary.

o   “…[44] proposed the improved STARFM (ESTARFM)…” – suggestion to change “improved” to “Enhanced” to match the cited work.

o   “…improves the prediction accuracy…” – so far in this paragraph these models have been described as weighting or interpolation methods, so “prediction accuracy” in this sentence is confusing. What is being predicted here? Please revise as this will also be an important consideration clarity for the Methods section.

o   “…and the coherence criterion considers…” – should this be the “consistency” criterion to match the previous sentence?

o   “Since in this research, we have used the S2 and L8 data, which have significantly similar radiometric and geometric characteristics, for this reason, the use of the second criterion does not give us any differentiation between the pairs of base images used. It is, therefore, important in our case to propose an alternative approach to find the optimal input reference pair of images, which is a key step before the implementation of STF techniques. For this purpose, we proposed to use both the consistency and similarity criteria between the low-resolution image at the baseline date and the one at the merge date.” – If the geometric and radiometric characteristics of the S2 and L8 platforms are too similar for the consistency criterion to work as intended, I fail to see how comparing two L8 images would solve this issue. Surely the radiometric and geometric characteristics of two L8 images would be identical, rendering the criterion even less useful. Please clarify.

§  Additionally, I think that including this information in the Methods section (possibly Section 2.6) may improve the readability of the Introduction.

o   “Sections 2.1 and 2.2 present…” – suggestion to condense the subsections of 2.1-2.7 into Section 2.

 

·       Materials and Methods

o   Figure 1 - this figure is very unclear in its present form and must be revised.

§  There does not seem to be a logical flow between the three panels, in part because it is difficult to understand the relationship between the second and third panels.

§  The first panel shows an unnecessarily large area. What is the purpose of showing the locations of satellite tiles over coastal Mauritania?

§  Why are the legend notations for rectangular satellite tiles such irregular shapes? Do all legend entries even appear in the panels? For example I cannot locate a green outline in the maps that the legend indicates as “Study area”.

§  Elevation color legend is too compressed to convey any helpful information.

§  Suggestion to display scale bars at round intervals – 320 Km is somewhat unhelpful. Scale bar in third panel is so dark as to be almost unreadable.

o   Subsection 2.2: suggestion to change title to “Remotely-sensed data acquisition” or “Satellite data acquisition”. Also check remainder of document for consistency. I suggest not using the phrase “remote sensing image[s]”

o   First sentence of 2.2: suggestion to list S2 and L8 in the order you describe them in the following paragraphs.

o   “…and embarks the OLI push-broom sensor…” – Unclear/typo

o   “…ten to five days…” – suggestion to change to “five to ten”

o   “However, not all images taken could be utilized due to the presence of massive cloud cover in the study area.” – Surely this is also an issue with the L8 imagery? Please comment.

o   “13 overlapping spectrum bands at wavelengths from 0.4 to 2.2 micrometers: visible and near-infrared at 10 meters, red edge and near-infrared at 20 meters, and atmospheric bands at a spatial resolution of 60 meters” – please indicate a difference here between spectral and spatial resolutions. Unclear as written.

o   “In advance of the following sections…” – this sentence needs revision for clarity.

o   “After the choice of dates when S2 coincides with L8 over the study area from 2015 to 2021, we used the temporal profile of the average NDSI values evolution for each image of the S2 series to find out the specific dates when snow fell and select the cloud-free ones to obtain a chain of snowy and cloud-free L8 and S2 image pairs.” – more details are necessary here. It seems that the S2 timeseries allow the determination of when snow fell within a 5-day window – how do you then determine the “specific dates” of snow events? How do cloudy images factor into this calculation? This paragraph discusses choosing image pairs only based on relative dates – how does this relate to the previous discussion of the similarity and consistency criteria presented earlier?

o   Figure 2

§  Suggestion to label y-axis as “S2 NDSI values”

§  Suggestion to change x-axis labels to English abbreviations

o   “…achieved by computing the normalized difference snow index (NDSI)” – this abbreviation should be defined the first time it is referenced.

o   Last paragraph in Section 2.3 beginning with “There is an agreement in the literature…” – more details are necessary here. References [17] and [58] discuss NDSI thresholds for snow covered pixels for MODIS and Landsat imagery. In this study, are all pixels with NDSI > 0.4 treated as entirely snow covered? If so, a reference or other justification is necessary to apply this threshold value to S2 imagery. If not, what calculation is used to transition between NDSI and snow covered area? How is sub-pixel spatial variability considered? Are there special considerations for the High Atlas study region, and how are they justified?

o   “…have considerably different main bases…” – unclear

o   “These models have a fundamental step of finding similar, homogeneous, and neighboring low-resolution pixels…” – how are similar, homogeneous, and neighboring defined? Is it possible to include equations, thresholds, etc. for these terms? The lack of clarity here contributes to additional confusion in Steps 1-4 listed in Section 2.5.1.

o   “…a collection of sparse-resolution images…” – unclear – should this be “coarse-resolution”?

o   “Before executing the FSDAF model, the high- and low-resolution images must be co-registered and corrected to the identical physical quantity” – unclear what “identical physical quantity” means in this context.

o   “(1) reprojection of the coordinates of the Landsat-8 image on the Sentinel-2 image” – suggestion to use the introduced abbreviations L8 and S2 for consistency.

o   “(4) then clipping of all images (Make sure all the input data with the same size)” – revise for clarity

o   Section 2.5.2, Steps 1-6 are missing significant information/term definitions and require revision:

§  “(1) Classify the NDSI Sentinel-2 image at time t1: In this model, an unsupervised classification algorithm…” – What was the purpose of this classification, snow presence/absence? Or were more than two classes specified? Which algorithm was used and was any model optimization required to achieve acceptable results? How was the classification accuracy assessed?

§  “(2) Estimate the temporal variation of every cluster in the Landsat NDSI from time t1 to time t2” – based on the fact that an S2 image was classified in Step 1, should this Step 2 refer to S2 NDSI or is Landsat correct? Also, how is temporal variation defined and estimated?

§  “(3) …and compute the residuals at every coarse pixel” – how are residuals computed? This may be more obvious if it was clear how temporal variation was defined and estimated.

§  “TPS interpolator” – please define this abbreviation

§  (5) – please define the weighted function and “temporal predictive remainders”

o   Section 2.5.3, SVM paragraph – was any parameter tuning required to achieve acceptable results? How was the SVM classification accuracy assessed? This is a supervised classification method, so what reference data was utilized for training the model?

o   “…they are not appropriate for our study.” – Why not?

o   (Note: also mentioned in the General Comments) “According to which the optimal pair for fusion is the one with the highest similarity index and the highest correlation coefficient with the low-resolution image at the desired prediction date (Figure 4)” – I have two issues here:

§  The way the equations for reflective difference, correlation coefficient, and similarity index are formulated, there is no guarantee that the image pair with the highest similarity index is also the image pair with the highest correlation coefficient For example, a pair with an extremely low reflective difference and median correlation coefficient could theoretically result in the highest similarity index. How would you reconcile this scenario? Alternately, if there is some physical reason that  this hypothetical scenario could never arise, it must be justified in the manuscript.

§  Jumping ahead a bit to Figure 5, the fact that the Similarity Index for an image with itself is not constant (and also relatively low) makes me question how useful this metric is. For example, the SI between 2021-04-18 and 2018-05-12 is 0.02, which is higher than the SI for 5 images when compared to themselves. This is quite counterintuitive as I would expect a derived metric to behave similarly to the correlation coefficient, in that the highest/”best” values would consistent (e.g. always equal 1) and would be achieved by computing the metric using the same image for both inputs.

o   In Equations 2 and 3, does subscript j denote individual pixels? Please clarify.

o   “According to equation (4), a higher SI value results in a lower reflectance difference and a larger correlation coefficient.” – I think I understand what you are getting at here but your description of this equation is not always true. Perhaps: “lower reflectance differences and larger correlation coefficients (or combinations of the two) result in higher SI values.”

o   Figure 4

§  At this point in the Methods section there are many uncertain aspects and this figure does not help with clarification. Does the Sentinel-2 entry date image relate in any way to the Landsat-8 Fusion Image? Perhaps this figure would be more enlightening with a different layout, but in its current form it does not help my understanding.

o   Table 1

§  Is “Number of Pairs” more of an index number (e.g. Pair #1, Pair #2, etc) than the actual number of image pairs? If so, consider removing this column entirely or change column heading at minimum.

§  Prediction Date columns are unnecessary as the same image is used for all pairs.

o   Equation 5

§  It is somewhat confusing to refer to images A and B and then introduce subscripts x and y to refer back to those images. Is it possible to revise this equation to use A and B subscripts to refer back to their respective images?

§  m variable appears in the equation but is not defined.

o   After Equation 7: “…represent the NDSI value of the i-th element in both the estimated…” – suggestion to change to “i-th pixel”

 

·       Result Analysis (suggestion to rename this section “Results”)

o   “These results give us an idea about the choice of the optimal image pairs, and as we chose previously to predict the image of the date April 4, 2016, we can see that the two pairs of March 25, 2018, and January 28, 2021, have the highest similarity indices (SI = 0.016 and SI = 0.013 respectively) and the largest correlation coefficients (R = 0.91 and R = 0.82 respectively) which allows us to use them for the spatiotemporal fusion models.” – In Figure 5, why show the entire table of correlation/SI values when the only values used in the analysis are the comparison with the 2016-04-04 pair? This led to some confusion during my initial read-through. Suggestion to remove the extraneous values from Fig 5 and present only the relevant values for comparison. In addition, for Fig 5:

§  Please revise to colorblind-friendly coloring schemes. Additionally, I am accustomed to green and red in the same color scheme indicating “good” and “bad”, respectively, but I think that a high correlation or similarity index would be positive outcomes.

§  Left panel legend: should this be “Pearson Correlation”?

o   Figure 6

§  This figure is messy and confusing. Instead of two panels separated by FSDAF and ESTARFM, I suggest four panels where each panel shows a separate metric and has lines that denote the different model configurations. Is there a reason to exclude Pre-classification FSDAF results from this figure?

o   Figure 7

§  These panels are so small that they look essentially identical to me. Figures 8 and 9 are more effective because they present the same information over smaller areas, which makes the subtle changes between models more apparent. I suggest removing Figure 7 entirely, in which case I also suggest removing the discussion in the previous paragraphs about the visual/qualitative analysis.

o   Figures 8/9

§  These are more effective than Figure 7 but still look quite similar to my untrained eye. One possible change to highlight the areas of difference is to include maps of NDSI differences – e.g. for each model, present the data of (model NDSI – S2 NDSI). This could be instead of or in addition to the raw NDSI values as they are currently presented. This would be an interesting comparison with the topography – for example does the model perform better/worse on ridgetops/valleys?

o   “…but there are also significant differences in the prediction quality between the methods used.” – Unless statistical tests with associated significance values are employed later, I suggest rephrasing this to e.g. “notable differences” or “important differences”.

o   Figure 10

§  What do the two lines in each subplot denote? Please add a label or caption description and increase the line widths.

§  I can barely see the contour lines indicating scatter plot density. I think this information is conveyed reasonably well with the coloring of the points, but please add a colorbar for this purpose. If contours are left in the line width should also be increased.

o   “Although this result is very useful in choosing the Pre-Classification FSDAF, it is not as satisfactory…” – does the “useful result” refer to visual analysis of the scatter plots? If so, suggestion to remove this and the related previous discussion as it contains no rigorous analysis. If not, please clarify.

o   Figure 11

§  Isn’t this information included in the subpanels of Figure 10? I suggest removing this Figure 11 entirely.

o   Section 3.3

§  “…by applying different thresholds of NDSI snow indices (0.3, 0.35 and 0.4) were compared to the reference classification map.” – is the reference classification map the 10m Sentinel-2 NDSI image, where NDSI> 0.4 implies a snow covered pixel? If so, it appears that the main point of Figure 13 is to show that when the L8 imagery used to drive the FSDAF model and S2 imagery used for the reference image use the same NDSI threshold, the SCA calculations have the closest match. This analysis has the potential to be very revealing but in its current form the conclusions are rather intuitive. Is it possible to extend this, e.g. by varying the NDSI on the reference image? Otherwise, why adjust the NDSI value in the first place?

o   Figure 12

§  This is an interesting figure – Consider adding numerical labels to dotted lines and please label subpanels A, B, C.

o   “…the fusion of the S2 and L8 data allowed for perfect and continuous monitoring…” – “perfect” is not an appropriate descriptor here.

o   Figures 14/15

§  Figure 15 is a much improved version of Figure 14. I suggest merging these two figures together with two panels in the style of Figure 15, where one panel shows NDSI and the other shows snow covered area.

o   “However, in terms of snow area estimate, the fused series of L8 and S2 data with a high spatial resolution (10 m) provides an accurate estimation of the snow surface for each map during the hydrological year (SCA (Combined) = 2091.82 km2) which cannot be properly assessed using only the time series of L8 (SCA (L8) = 506.35 km2) or S2 (SCA (S2) = 1650.10 km2).” – The estimates are clearly different between the three data products, but claiming that the combined product is more accurate requires some sort of validation data that is not present in this study.

·       Discussion

o   “…the Pre-classification model uses a supervised SVM classifier to classify the high-resolution input image (NDSI S2), which is the first step before the implementation of the fusion algorithm to obtain the fraction of each class (snow and snow-free) in each coarse pixel (L8 pixels)” – Was SVM was used to classify snow vs snow-free pixels? NDSI is a continuous variable so I think that would be a regression task, not classification.

o   “…our research is not the case to use one of them…” – revise for clarity

o   (Note: also mentioned in the General Comments) “The high performance of the Pre-Classification FSDAF model compared to the other two models is mainly due to the use of the supervised SVM classifier to classify the high- resolution input image, which is a critical step before fusion implementation.” – In my view this overstates the contribution of the Pre-classification FSDAF. Both FSDAF configurations are improvements over the ESTARFM model, but the improvements between standard and Pre-classification FSDAF are very small in comparison. This section of the discussion would be improved with more context – for example, predicted NDSI improves because RMSE decreases by 0.003 or Square R increases by 0.01. What effect does this have in terms of over/underestimating snow covered area, the timing of snow events, effects on streamflow forecasting, etc?

o   “The accuracy of spatiotemporal data fusion (STDF) is primarily impacted by…” – Maintaining consistency with the previously-defined STF abbreviation will improve readability here.

·       References

o   Check for extraneous information in references, e.g. [2] and [25] seem to have some user tags that have made it into the citations.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 4 Report

In this manuscript the authors examine the effectiveness of the combined use of optical sensors through image fusion techniques for capturing snow dynamics and producing detailed and dense Normalized Difference Snow Index (NDSI) time series within a semi-arid context. In general, the manuscript is well written, the topic is interesting and the most important references are provided, so it could be accepted for publication after major revisions.

 

 

General comments :

I think it is very important to add additional information in the discussion related to the advantages of having selected Landsat and Sentinel-2 images for this study compared to images from other sensors such as MODIS, that although the authors indicate in the introduction the problems related to its low spatial resolution to model the spatial variability of the snow cover, its daily temporal resolution is a very important factor to take into account because, as indicated in the manuscript, sometimes the temporal resolution of Sentinel-2 using both satellites it may be insufficient to model the snow cover, especially in periods with abundant cloudiness.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I appreciate the effort made by the authors

Author Response

The authors would like to thank the reviewer for the careful and thorough reading of our article and for the thoughtful comments and constructive suggestions, which improved the quality of this manuscript.

Reviewer 3 Report

Please see attachment.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

The authors have responded and have taken into account all my suggestions, so I consider that the manuscript can be accepted for publication.

Author Response

The authors would like to thank the reviewer for the careful and thorough reading of our article and for the thoughtful comments and constructive suggestions, which improved the quality of this manuscript.