Development of Snow Cover Frequency Maps from MODIS Snow Cover Products
Round 1
Reviewer 1 Report
This paper developed snow cover frequency maps based on 22-year MODIS snow cover products. The authors’ method for removing the “false snow” on MODIS snow cover frequency maps have been proved effective, and the produced snow cover frequency maps could be potentially used in snow climatology. However, there are still some issues needed to be justified clearly.
1) Line 106, what product does MOD10C1F CGF represent?
2) This paper used a 5×5 window and 33% threshold to remove the “false snow”. However, it is confused for Line 120-122.
3) For the spatial filter, why used 33% threshold to remove the “false snow”? If this threshold is too high, it is more likely to exclude some rare or infrequent snow events. It is suggested to add sensitivity analysis for this threshold.
4) Compared with other snow cover frequency products or snow probability products, what new information can be extracted or what new application can be implemented from the snow cover frequency maps produced in this paper?
Author Response
1) Line 106, what product does MOD10C1F CGF represent?
That was a naming error that has been corrected. “MOD10C1F CGF” was corrected to MOD10C1_CGF.
2) This paper used a 5×5 window and 33% threshold to remove the “false snow”. However, it is confused for Line 120-122.
The description of the snow cover frequency algorithm was confusing. Other reviewers had similar comments regarding the algorithm description. We added a flow diagram of the snow cover frequency algorithm to clarify how it is applied, and we revised the description of the algorithm in the Snow Cover Frequency Computation section to clarify points raised by reviewers.
3) For the spatial filter, why used 33% threshold to remove the “false snow”? If this threshold is too high, it is more likely to exclude some rare or infrequent snow events. It is suggested to add sensitivity analysis for this threshold.
We agree with this concern. We had evaluated thresholds of 25%, 33% and 50% on removing ‘false snow’. We selected the 33% threshold as the best compromise between removing too much snow or not removing enough ‘false snow’. We added an explanation in the Snow Cover Frequency Computation section.
4) Compared with other snow cover frequency products or snow probability products, what new information can be extracted or what new application can be implemented from the snow cover frequency maps produced in this paper?
Snow cover frequency maps were produced for every day of the year at a synoptic scale. We are not aware of other snow frequency maps for every day of the year at this scale. The snow frequency maps could be used to investigate changes in snow cover climatology factors. We think that that discussion of the algorithm, the filtering of ‘false snow’, and the results at a relatively coarse resolution of ~ 5 km provides insights for developing higher spatial resolution snow cover frequency maps from MODIS, VIIRS or other optical remote sensing instruments. We suggested making snow cover frequency maps from the MOD10A1F cloud-gap-filled product at 500 m resolution and from VIIRS data at 375 m resolution in the Conclusions. Other snow cover frequency algorithms and products may be compared to the MODIS snow cover frequency product to demonstrate improved products. There is a tool in Google Earth Engine that can be used to create derived products from the MOD10A1 product [12,8]
Author Response File: 
Author Response.docx
Reviewer 2 Report
Journal: Remote Sensing
Manuscript title: Development of Snow Cover Frequency Maps from MODIS Snow Cover Products
Authors: George Riggs, Dorothy Hall, Carrie Vuyovich and Nicolo DiGirolamo
Review:
The manuscript is a good short technical note that attempts to solve the problem of "cloud-snow confusion" in MODIS maps. The Authors compare MODIS maps with NOAA's historical probability of snow cover. They also indicate the causes of snow cover detection errors. The article is based on correct assumptions and contributes to improving the quality of maps of snow cover and detailing our knowledge of its changes. So, I recommend the manuscript to be published in the journal Remote Sensing. However, I would suggest Authors to consider following comments:
- add, please, information, how many (what percentage?) were there cases of "false snow" in the investigated period?
- Figure 3: I suggest adding a map c) showing the difference in value between a) and b).
- the comparison of MODIS maps with NOAA's historical probability was done on one example only (25. December). It would be better to add other such examples if possible.
Minor remarks:
- line 47: "start, duration, and end of the snow season..." add please "snow cover depth", because it is equally important in snow cover climatology studies;
- Figure 4: insert the red rectangle in b) as well.
- line 317: “22-year data record”. 2001-2020 is 20 years.
Author Response
The manuscript is a good short technical note that attempts to solve the problem of "cloud-snow confusion" in MODIS maps. The Authors compare MODIS maps with NOAA's historical probability of snow cover. They also indicate the causes of snow cover detection errors. The article is based on correct assumptions and contributes to improving the quality of maps of snow cover and detailing our knowledge of its changes. So, I recommend the manuscript to be published in the journal Remote Sensing. However, I would suggest Authors to consider following comments:
- add, please, information, how many (what percentage?) were there cases of "false snow" in the investigated period?
That is a challenging question. We added a paragraph describing how we have estimated the occurrence of ‘false snow’ in the MODIS snow cover products to Section 3.1 Uncertainties. Typically, ‘false snow’ is in the range of 0-3% in any MODIS swath. We assume that range also applies to the daily global MOD10C1 and MOD10C1_CGF products. Thanks for asking for this information, it is good to include.
- Figure 3: I suggest adding a map c) showing the difference in value between a) and b).
It would be interesting to show a difference map, but we are unable to create one. We only have the NOAA historical probability map, Figure 3 (b), as an image. The purpose of the comparison of the MODIS snow cover frequency to the NOAA historical probability is to provide qualitative evidence of the accuracy of MODIS compared to a weather station snow measurements map.
- the comparison of MODIS maps with NOAA's historical probability was done on one example only (25. December). It would be better to add other such examples if possible.
We agree that more than a single example would be better. We present only a single day example to illustrate how the filtering of ‘false snow’ in the MODIS snow cover frequency algorithm can result in snow events being missed. Some snow events can be missed, or their observed duration decreased because of cloud cover. We present the research as a short technical note so we only included one example. A comment on why only one example was evaluated was added to the end of the section.
Minor remarks:
- line 47: "start, duration, and end of the snow season..." add please "snow cover depth", because it is equally important in snow cover climatology studies;
We agree that snow cover depth is equally important in the study of snow cover climatology. However, snow cover depth cannot be measured by the MODIS instrument, so in order to remain focused on factors that can be observed by MODIS we have not included snow cover depth in the discussion.
- Figure 4: insert the red rectangle in b) as well.
Instead of inserting the red rectangle in b) we deleted the red rectangle in a). We made that deletion because we think that without that rectangle a reader can better focus on the entire region of snow cover in northern Georgia. Also, the maps are not at the same scale, so we are limited to a qualitative visual comparison. We replaced the image and revised the figure caption.
- line 317: “22-year data record”. 2001-2020 is 20 years.
That sentence was confusing, we meant to indicate that we used 20 years of the MODIS data record that is now about 22 years in length and should continue for a few more years. We revised the sentence to describe the 20-year period used.
Reviewer 3 Report
1. The main question addressed by the research.
The paper developed a Snow Cover Frequency Maps using a ~ 5 km resolution MODIS snow cover map (MOD10C1). The maps provide information on how frequently snow cover has been observed, on every day of a year over the study period from 2001 to 2020 in the Northern Hemisphere.
2. The originality or relevance of the topic in the field.
This paper is an extension of previous work using the daily global MOD10C1 snow cover product at 5 km resolution that investigated developing a CGF product. The Authors added filtering methods to the MOD10C1_CGF maps to alleviate 'false snow' to enable study of snow climatology.
3. The contribution to the scientific literature.
These MODIS snow cover frequency maps will be compared to snow cover metrics created by other researchers using MODIS data and NOAA IMS data.
4. I only suggest a few revisions before publishing the manuscript.
The first concerns the addition of a flow chart to improve the readability of the methodology.
The second concerns the addition of some other detailed analysis, such as that carried out in the USA, in territories with significant spatial variations of snow cover such as Europe.
Author Response
- The main question addressed by the research.
The paper developed a Snow Cover Frequency Maps using a ~ 5 km resolution MODIS snow cover map (MOD10C1). The maps provide information on how frequently snow cover has been observed, on every day of a year over the study period from 2001 to 2020 in the Northern Hemisphere.
2. The originality or relevance of the topic in the field.
This paper is an extension of previous work using the daily global MOD10C1 snow cover product at 5 km resolution that investigated developing a CGF product. The Authors added filtering methods to the MOD10C1_CGF maps to alleviate 'false snow' to enable study of snow climatology.
3. The contribution to the scientific literature.
These MODIS snow cover frequency maps will be compared to snow cover metrics created by other researchers using MODIS data and NOAA IMS data.
- I only suggest a few revisions before publishing the manuscript.
The first concerns the addition of a flow chart to improve the readability of the methodology.
The description of the snow cover frequency algorithm was confusing. Other reviewers had similar comments regarding the algorithm description. We added a flow diagram of the snow cover frequency algorithm to clarify how it is applied, and we revised the description of the algorithm in the Snow Cover Frequency Computation section to clarify points raised by reviewers.
The second concerns the addition of some other detailed analysis, such as that carried out in the USA, in territories with significant spatial variations of snow cover such as Europe.
We agree that analysis of other locations should be done. We present only a single day example to raise concerns about how the filtering of ‘false snow’ in the MODIS snow cover frequency algorithm, and the ~ 5 km resolution, can cause some snow events to be missed. Also, some snow events can be missed, or their observed duration decreased because of cloud cover. We have looked at other snow events and observed similar results from the snow cover frequency algorithm, as presented for the example in the paper. A relatively lengthy discussion would be need for each event, and because we present the research as a short technical note, we have only included one example. A comment on why only one example was evaluated was added to the end of the Discussion section.
Reviewer 4 Report
The authors present a new data product from MODIS - global snow cover maps for each day of year, derived from 20 years of data. The product is potentially of interest as a climatological background for snow cover extent. The presentation is sound and the results in line with expected output. As the paper is submitted as a technical note, the lack of quantitative validation results is acceptable, although this also defers true quality assessments to later studies. In general, while some clarifications and revisions in the manuscript are in order, I see no obstacle to publication once these are appropriately addressed.
Comments:
- The discussion notes plans to quantitatively evaluate the product, yet I feel that in situ data are not really highlighted. Snow course data are available up kilometer lengths (at least in the Nordics), and aggregated in situ snow measurements as used by e.g. Chen et al. (2021*) would serve the upcoming validation well.
*https://doi.org/10.5194/essd-2021-279
- First two paragraphs of the manuscript (ln 25-39) contain some unnecessary repetition, please revise.
- The cautionary filtering procedures (discarding <10% snow extent, the 5x5 box filter requiring 1/3 of snow) seem to suggest that there is a sensitivity bound for lowest snow frequency that is detectable. Is it possible to simulate what this would typically be?
- Figure 1 suggests that there is a Sun Zenith Angle cutoff in the algorithm, please specify that in text. The figure color also suggests that snow frequency drops a bit close to this cutoff line, is that correct? If so, what is the cause?
- Ln 106, the commas around "MOD10C1F CGF" feel disjointed, and the first "F" seems out of place.
Author Response
Comments:
- The discussion notes plans to quantitatively evaluate the product, yet I feel that in situ data are not really highlighted. Snow course data are available up kilometer lengths (at least in the Nordics), and aggregated in situ snow measurements as used by e.g. Chen et al. (2021*) would serve the upcoming validation well.
*https://doi.org/10.5194/essd-2021-279
Thank you for that reference. I found that the preprint article with that doi has been withdrawn, however the dataset is available at Zenodo. I could not find a revised version of the paper. A challenge to using that dataset is that it is 8-day temporal resolution, not daily resolution, and it is binary snow map.
Agree that in situ data would be useful for evaluation primarily in validating the MOD10_L2 product, we and other researchers have evaluated the MOD10_L2 or the MOD10A1 products with in situ data. For the MOD10C we plan to do comparative evaluation with other well-known snow cover products such as the IMS or the Copernicus snow cover product.
- First two paragraphs of the manuscript (ln 25-39) contain some unnecessary repetition, please revise.
We removed some of the repetition from the Introduction.
- The cautionary filtering procedures (discarding <10% snow extent, the 5x5 box filter requiring 1/3 of snow) seem to suggest that there is a sensitivity bound for lowest snow frequency that is detectable. Is it possible to simulate what this would typically be?
The description of the snow cover frequency algorithm was confusing. Other reviewers had similar comments regarding the algorithm description. We added a flow diagram of the snow cover frequency algorithm to clarify how it is applied, and we revised the description of the algorithm in the Snow Cover Frequency Computation section to clarify points raised by reviewers.
We had evaluated thresholds of 25%, 33% and 50% on removing ‘false snow’. We selected the 33% threshold as the best compromise between removing too much snow or not removing enough ‘false snow’. We added an explanation in the Snow Cover Frequency Computation section.
- Figure 1 suggests that there is a Sun Zenith Angle cutoff in the algorithm, please specify that in text. The figure color also suggests that snow frequency drops a bit close to this cutoff line, is that correct? If so, what is the cause?
We added a brief explanation of the cause to the 3.1 Uncertainties section
That is the day/night terminator. The distinction between day/night is made in the MOD10_L2 product algorithm. A solar zenith angle ≥ 70° is night. In the MOD10C1 product there is a band of night in the boreal regions that moves with the seasons. We treated night as no data in the snow cover frequency algorithm. The snow frequency drop at the terminator is caused by geolocation uncertainty of observations through the product levels and different projections. Geolocation uncertainty means that the terminator is not mapped at the same location on the same day in every year, that results in a fuzzy terminator line in the snow cover frequency maps over the 20-year study period. Below is a block of snow frequency data from 30 January Figure 1 (a) to illustrate the fuzzy terminator line, the count of years of snow cover, at the terminator, the 0 lines are no data to the north of the terminator and move south down the lines across the terminator region.
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0
4 4 4 4 4 4 4 4 4 4
5 6 5 5 5 5 5 5 5 5
6 6 7 7 7 7 6 6 6 6
8 8 8 9 9 8 8 8 8 8
9 9 8 10 9 9 10 9 9 9
11 10 12 9 10 10 9 9 10 10
12 12 12 13 12 12 13 14 14 13
12 13 13 14 14 15 15 14 15 15
14 16 16 16 16 16 17 16 16 14
18 16 18 17 18 17 16 17 16 18
18 18 18 17 17 16 16 17 17 18
20 20 20 20 19 19 20 20 19 20
20 20 19 19 19 19 19 20 20 20
19 19 19 19 19 19 19 19 18 18
20 18 19 19 18 19 20 20 20 20
20 20 20 20 19 18 19 20 19 20
- Ln 106, the commas around "MOD10C1F CGF" feel disjointed, and the first "F" seems out of place.
That was a naming error that has been corrected. “MOD10C1F CGF” was corrected to MOD10C1_CGF.
Round 2
Reviewer 1 Report
The authors made good responses to reviewers' comments. I think this paper is ready for publication.
Reviewer 2 Report
Journal: Remote Sensing
Manuscript title: Development of Snow Cover Frequency Maps from MODIS Snow Cover Products
Authors: George Riggs, Dorothy Hall, Carrie Vuyovich and Nicolo DiGirolamo
Review after corrections:
I am fully satisfied with the answers of the manuscript Authors to my comments. So, I recommend the manuscript to be published in the journal Remote Sensing in present form.
