Reconstruction of Snow Depth Data at Moderate Spatial Resolution (1 km) from Remotely Sensed Snow Data and Multiple Optimized Environmental Factors: A Case Study over the Qinghai-Tibetan Plateau

Round 1

Reviewer 1 Report

This document includes the review comments on the manuscript titled, “Reconstruction of Snow Depth Data at Moderate Spatial Resolution (1km) from Remotely Sensed Snow Data and Multiple Optimized Environmental Factors: A Case Study Over the Qinghai-Tibetan Plateau”, by Wei et al.

Downscaling of the low resolution the passive microwave remote sensing data (Passive Microwave Snow Depth dataset, PSD) to 1km based on the MODIS snow cover fraction (SCF) data.  The PSD and the MODIS SCF were combined into the fused snow depth (FSD) data at the 1km resolution.  Then, the linear multivariate snow depth reconstruction (LMSDR) model was developed using the ground-observed snow depth data at 100 in-situ stations.  The LMSDR estimates the snow depth by a linear regression relationship to FSD, longitude, latitude, slope, surface roughness, surface relief, and snow cover days (SCD). The results were checked by the 10 independent in-situ snow depth data.  Main reason in the discrepancy was uneven distribution of the in-situ snow measurement sites.

Aa I did not find any technical flaw; this technique should work in the Qinghai-Tibetan Plateau although the transferability to other region may be questionable. After a few minor revisions listed below, I found some merit in this manuscript for publication in this journal.

Clarification questions and suggestions:

1. Table 2: What these X_1 through X_7 are? You selected 6 sources.  What is the seventh? (It should be FSD…)
2. Figure 6: What the timing of the snow distribution in Sep 11, 2011 and Mar 31, 2012? The time range is too long.  Is this season average snow depth or annual total snow precipitation? 
3. I do not think Fig 5 is effective.
4. How to ensure the stationarity of the model parameters? The stationarity of the regression parameters is a major limitation of this technique.  It should be mentioned somewhere.

 

General questions:

Why snow depth instead of snow water equivalence (SWE) is selected?

Do you have any idea why the linear multivariate regression model showed slightly better result than the other models?

 

Author Response

please find the attached file

Author Response File: Author Response.docx

Reviewer 2 Report

This research provides an important path towards more effective moderate-resolution snow depth modeling in mountainous regions. The authors apply robust methodologies to produce a novel 1-km snow product.

 

Major comments:

• Introduction is fantastic.
• In discussion during analysis of errors, the authors should include a more detailed analysis of the spatial distribution of model error. As is, error discussion is focused on statistics calculated over large regions (Sec. 5.1) or snow characteristics (Sec. 5.2). Spatial trends in error are discussed in Sec. 5.3 (L430-468), but provide only one measure of error distribution (L436-437) and highlight 6 specific in-situ sites. Why not provide a map with in-situ sites plotted as points quantifying model error per location? Or maybe histogram of error metrics?
• 5 - I'm not sure radial/circular plots are a more effective method of presenting results than a standard x-y bar or point plot. It's more difficult to compare error stats between groups. At the very least, need to increase the size of the points or replace with orthogonal lines.
• Discussion/conclusion: You mention that more in-situ observations would help improve your model (L533-534). Do you think incorporating finer-resolution datasets to describe sub-pixel heterogeneity could be used to improve your model as well? Could other machine learning methods be used for estimating snow depth, such as random forest, neural nets, etc?
• Discussion: Please mention how transferable your model is - can it be scaled to other regions?

 

Minor comments:

• Writing needs some cleaning up in terms of English and wording, but manuscript is still quite readable. I highlight a couple typos in the abstract below.
• L25, typo: "on board Terra..."
• L32, typo: "at moderate resolution (1km) for 16 consecutive hydrological years".
• L350, reword: "The LMSDR model performed worst over grassland (RMSE = 2.28 cm), but..."
• 5 - should state the number of validation sites per interval/group in each subplot
• 6 - Define legend scale "SD" as snow depth in figure caption. SD sometimes used as acronym for standard deviation. Remind reader of the source/genesis of each dataset in caption, along with spatial resolution.
• L433 - Define what "match" means when comparing two continuous variables. Is it within a certain percentage?

Author Response

please find the attached file

Author Response File: Author Response.docx

Reviewer 3 Report

Major comments:

The abstract lacks any significant results.  The abstract should give the area-average snow depth for the QTP (not the Himalaya) for the month of greatest snow depth (February), and also say what percent of the QTP area is free of snow during that month.

Give a brief description of how snow depth can be inferred from passive microwave, considering the dependence of emissivity on snow density, temperature, liquid content, and grain size as well as depth.

Melting snow has a dramatically different microwave emissivity than cold snow.  How is this incorporated into the method?

Specific comments:

Line 17.  “mountainous regions”.  The title instead says “plateau”.  It’s important to distinguish the QTP from the Himalaya.

Line 25.  Change borad to board.

Line 27.  “passive snow depth data”.  Does this mean microwave?

Line 33.  Define “snow cover day”.

Line 38.  “fusion” means “melting”.  By “fusion snow depth product” do you mean the change of snow depth during melting?

Line 59.  “QTP forms the largest portion of the cryosphere in the mid-latitude regions”.  This is not correct.  In winter, the area of snow-covered land in North America and Northern Asia (Siberia) greatly exceeds the area of the QTP.

Line 115.  “fused snow depth”.  Because of the ambiguity of “fused”, a different adjective would be better, maybe “merged”.

Line 131.  Change “Lantsang” to “Lantsang (Mekong)”.

Line 157.  Show an example plot of the snow-depletion curve.

Line 159.  “SCD”.  The reader may have difficulty keeping in mind the distinction between SCD and SDC.

Line 173.  How do the automatic stations measure snow depth?  Do they accurately report snow depths of zero (snow-free)?

Line 185.  “soil particle size distribution data”.  How are these data used?

Line 215.  “SDCi” is not in Equation 1.

Equation 2.  Why is the subscript i needed on u-sub-i?

Line 226.  Change “exponential” to “power-law”.  Y=x^b is power-law; y=b^x is exponential.

Line 236.  “more than 140,000”.  I assume this means “less than 150,000”.

Line 261.  “longitude”.  The dependence of the model on longitude applies only to the QTP; it will not be applicable generally.  It must be partly a proxy for elevation (high in the west, low in the east).

Line 295 “SDC” and Line 298 “SCD”.  Which is it?

Line 302.  What is PEM?

Line 332-333.  “western QTP snow is generally deeper than eastern QTP”.  What is the evidence for this?  Figure 6 shows the opposite, with greater depth in the east.

Line 439.  Are these six stations located in the six boxed regions in Figure 6?  If not, please mark their locations in Figure 6.

Figure 1.  Define SD and its units.  What month?  Are there any stations with SD=0?  If so, show them also.

Figure 3.  Say what month this is for.  Use a different color for lakes, to distinguish lakes from SD=100. 

Figure 5.  What is plotted here?  I’m guessing it is the error, in cm.  Is that correct?

Figure 5b.  Change “2000-2005” to “2000-2500”.

Figure 6.  Tell the date.

Author Response

please find the attached file

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors addressed my previous comments adequately.