SnowCloudMetrics: Snow Information for Everyone

Round 1

Reviewer 1 Report

General comments

The aim of the paper is: 1) to introduce and demonstrate the SnowCloudMetrics portal, and 2) to validate the snow products, offered by the portal. The first goal is a kind of “advertisements” – only describe another source of snow information (focused on hydrological applications). Much more interesting is the validation part, which attempt to determined how reliable is the MODIS MOD10A1 snow product in the part of the U.S.

In my opinion, Authors fail to provide a valuable validation results. First, Authors decide to use the snow water equivalent (SWE) data for validating the snow cover extent. This is probably the most challenging attempt, as those two parameters differs substantially in the way, they are observed. Theoretically SWE greater than zero implies a snow cover presence. In practice, methods used for reporting SWE are not sensitive enough to detect a shallow snow cover. On the other hand, few millimeters of snow can change the reflectance dramatically – optical systems will detect such snow easily. The source of ground data for the study is the SNOTEL system. However Authors provide no information on the snow pillow sensitivity. Impact of the SNOTEL’s SWE accuracy on the results is not discussed. Consequently, it is very difficult to conclude about the reliability of the validation results. Since the actual data on snow presence is not known, using the term ‘error’ should be avoided (‘discrepancy’ may be more suitable).

Second, the validation is far too general. Measures used to compare SNOTEL and MODIS are limited to RMSE, and comparison between variations and std. deviations. Although there are many ways to report agreement between two samples, probably the most basic – still very informative – is a confusion matrix. Computation of such statistic by the Authors would allow them to provide a reliability measure for user/producer perspective, and also provide measure to compare Authors’ results with other studies on the same issue. MODIS snow products were validated many times in the past, but Authors do not refer to/ discuss those results. It surprised me a little. Moreover, results only present the statistics averaged over a state. Being able to collocate a SNOTEL station with a MODIS 500 m pixel, it would be interesting to explore how the differences are at station level. Are the differences related to MODIS viewing angle? Time of the snow season (more ephemeral and shallow snow cover at the beginning and end of the season)? How do they vary with increasing SWE? How with increasing MODIS fractional snow cover? What does MODIS miss when clouds are preset? Was the strategy to deal with a cloudy MODIS IFOVS accurate, based on SNOTEL data? Those are only few questions not answered by the validation.

To be clear – the motivation for developing the described web service is fulle understandable. The fact the service is used by some entities (mentioned in the Discussion) provides additional arguments for developing such tools. Authors undoubtedly have skills to manipulate satellite data using cloud-based infrastructure. The paper itself is well written (in terms of language), and easy to follow. My main concern is the quality of validation, and the fact, that providing another web service is not a great novelty alone.

Specific comments

118: Authors use only MODTS/terra data (MOD10A1). Why not to include Aqua? Authors also use MODIS collection 006 data, while collection 061 is the most recent.
https://nsidc.org/sites/nsidc.org/files/technical-references/C6.1_MODIS_Snow_User_Guide.pdf
Impact of collection change on the final results (most notably – on the validation) should be discussed.

122 and Equation 1: consider using words ('total', 'number', 'count', or similar) instead of symbols ('#').

Fig. 1: the figure is rather trivial. I found it non informative. SCF and SDD concepts are very easy to understand without graphics.

148: Why to use a 5-day period? Why not to refer to the first/last day with snow within a snow season?

152: ‘assumes an October 1 to September 30 WY’ – I assume this results from the USGS definition. However on the northern hemisphere mid/late summer is the actual time of snow extent minimum (https://nsidc.org/cryosphere/sotc/snow_extent.html). August seems more suitable for beginning a snow season. Otherwise Oct. 1 will start the hydrological year ~1-2 months after the actual snow season started in many locations.

164: ‘We use this cloud mask to adjust both the SCF and SDD algorithms” – how exactly it was adjusted? The information about used strategies should be described here.

165: What about other possibility: snow-free, followed by cloud cover, followed by snow-free? This the case MODIS will miss the very first/last snow in the snow season. How often it happens? It can be tested with SNOTEL and MODIS data.

176: (100 mm yr-1) should be in one line.

203: Provide at least 2-3 sentences on how SNOTEL observed the snow cover. Snow pillow is not a standard procedure implemented at the WMO’s stations, hence may not be familiar to many readers outside the U.S. Provide an extensive information on SNOTEL accuracy and sensitivity – those information are crucial for the validation aspects.

226-230: ‘We exported the SCF datasets to GeoTIFF image using the GEE export function.’, and ‘we then created a SCF GeoTIFF image’ in line 302. In my opinion statements of that kind are not necessary (basic technical details of no importance for understanding the research).

230-234: Those are rather general information, and known for years.

238-239: ‘contains mostly no data values because it remains primarily snow covered all year.’ I agree, but using identical symbol to describe “no snow at all” and “snow all the year” is very misleading.

240: “We focus our SDD results on the Northern Hemisphere because of the nature of the SDD algorithm in GEE” – I understand the authors do not wish to investigate the snow cover on the southern hemisphere. However blaming the algorithm for the choice is not justified: Authors developed the algorithm, so omission of the south hemisphere is their decision (not the algorithm).

259: Details about how snow pillow-based measurements work should appear earlier, in the Data/Methods section, not in the Results.

261-265: Key information is missing: SNOTEL sensitivity to the snow presence.
262: Was the validation with SNOTEL performed at 500 m/pixel or at the aggregated 2 km/pixel? Please, clarify.

Figure 4: Please, do not split panels between pages.

Fig 4: patter seen on SNOTEL and MODIS data are nearly the same. It is difficult to note what are the differences. I suggest to replace the plot with other one, showing the differences between SNOTEL and MODIS (a visual form of 1, or scatter plot with station-level data).

277: ‘aggregated by state,’ – it would be far more interesting to see what is happening at the station level, and explore the observed discrepancies. Averaging cancels number of local disagreements. I guess users will be interested in smaller spatial domains than a state, perhaps even few pixels across. Station level validation would be the most informative and usable for them.

277: What does it mean ‘the GEE algorithm performs better in some regions’? As far as I understand, the algorithm within the GEE is not a scientific algorithm, but only a chain of procedures to manipulate the existing products. Is it the procedure exactly the same for each single pixel? If so, then “better”/ “worse” can apply only to MODIS snow detection algorithm, and/or SNOTEL snow detection algorithm.

387: On the other hand, MODIS is limited by clouds, and only allows for snow detection when IFOV is cloud-free. Snow can be present on the ground for few days before being detected from orbit. SNOTEL is probably not affected such way? I assume SNOTEL can detect ephemeral snow, missed by MODIS during cloudy days. How does it impact the results?

404-405: I do not agree with the argument against the Landsat. Actually MODIS/Terra and Landsat-7 fly together in a close orbital formation called ‘the morning constellation’. Landsat passes the same area MODIS did but 45 minutes later. For a common part of swaths both sensors (MODIS and ETM+) collects data very useful for assessing MODIS cloud detection.

415: ‘New global snow metrics generated using GEE’ – frequency of snow cover (SCF) and the day of beginning/and of snow season (also length of the snow season), are not new concepts.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This topic is interesting and important. I, personally, have visited the website and found the work is quite helpful for potential users to find the information they need, without struggling too much with the dataset itself. I have some comments below

(1)   It is possible that the authors can add a potential function to show the time series of SCF/SDD over predefined geographic regions?

(2)   One key issue is to separate the cloud from snow, although the authors refer to some publication, it will be very helpful to 1) give a short summary how the cloud is separated from snow; 2) what are the potential problem, especially how the potential cloud contamination may affect your SCF/SDD calculation

(3)   In the evaluation part, the authors made some efforts to explain the difficulties of spatial/temporal challenge for the validation, some more details are needed. For instance, is one ground-site collocated with the satellite pixel covering that site? As the authors mentioned, the site is quite intensive, why not try to analyze several sites together, in this way, the author can find cases when the snow has similar temporal/spatial behavior.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The authors developed a SnowCloudMetrics  portal for on-demand production and delivery of snow  information including snow cover frequency (SCF) and snow disappearance date (SDD) using 16 Google Earth Engine (GEE). it is interesting for snow scientists, water resource managers.

Overall, I find that the paper lacks detail in some areas that are particularly relevant to this study's original contributions. the SCF and  SDD is not a new  concept for snow evaluation. the cloud cover in the MODIS product  is an important factor to limit the precision of these two parameters, the cloud problem is not well resolved in the manuscript.

if anyone with a web browser can really create customizable snow information from the SnowCloudMetrics  portal, this may be an significance of this manuscrip.

what is the value of NDSI to determine the presence or absence of snow? it is is not clearly defined.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

General Comments:

In the manuscript “SnowCloudMetrics: Snow Information for Everyone”, the authors present SnowCloudMetrics, a snow information web portal that produces two global-to-local scale snow metrics, snow cover frequency (SCF) and snow disappearance date (SDD). Users can explore the snow information via a GEE map interface and download scripts for access to tabular and image data in non-proprietary formats for additional analyses. The SnowCloudMetrics can benefit snow scientists, water resource managers, climate scientists, and snow related industries providing SCF and SDD information tailored to their needs, especially in data sparse regions. In general, this paper is well written. The methodology, evaluation, and case analysis are stated clearly in detail. I recommend its publication with minor revision.

Minor comments:

Lines 156-158, “ If the entire year has snow present, or the alternate case when no snow is found during the water year, the pixel is masked, and no value is designated for SDD.”  If the authors can use different colors to mask the pixels with “entire year has snow present” and pixels “without snow during the water year”, the visualization of the maps may look better.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The developed a SnowCloudMetrics portal for SCF) and SDD using Google Earth Engine (GEE).  it is likely interesting for snow scientists, water resource managers.  Thank you for responsing my comments/suggestions.

however, through my further review of the paper, I found that there was really little innovation innovative in the theory , method and results.  if the developed SnowCloudMetrics portal's function is just to calculate these two metrics ( SCF and SDD) based on existing methods, I don't think it can be published as an academic paper.  I suggest authors consider the publication of software Copyrights or patents. 

I am thus sorry to discourage you with the submission of a revised manuscript, although the modification time is very short.

I am sorry for not being more positive. I hope that the  comments will help you to move forward with your work.