The authors proposed a comprehensive framework to process GNSS raw data under complex environment conditions to retrieve snow depth, and based on this, the authors produced a GNSS-based long-term snow depth data set over China from 80 stations. The topic is very interesting from a perspective different from traditional microwave remote sensing retrievals. As the authors’ statement, this data set has a unique spatial resolution between point-scale and coarse grid-scale. The new data set is valuable to the science community from this point. I also have confidence in this GNSS-IR technique, which could be a helpful and complementary tool for producing more snow depth products with high spatial-temporal resolution using extended global GNSS networks, particularly from GNSS sites in polar regions or even on scientific expedition vessels.

I recommend this work for publication after revisions.

Several comments are listed below:

1. Vegetation and terrain are two significant issues that affect snow depth estimation. The authors only discussed vegetation in Section 5.1. How about terrain effects? I recommend adding in-depth discussions relevant to this issue.
2. Around Line 185: “4) For high- and medium-quality sites, the model for deriving daily reflector height is established, and the raw snow depth for each GNSS satellite, each quadrant, and each GNSS frequency is subsequently calculated as the difference value of the referenced height in Step 3) and the height of this step”. I am confused about the descriptions of “height.” Which height was used as the referenced height? The authors should revise the texts to clarify this issue.
3. Line 350: “The 8-day MODIS NDVI is also involved as a quality flag in the data set to show the vegetation conditions of the site initially”. The authors only gave this vegetation flag. How to use this flag? I recommend adding a few sentences to describe.
4. Should Section 4.4 be a separate section? I am afraid it is improper to put the data set descriptions inside Section 4.
5. Figure 17 x-label is not correct? Should it be the current number +10?
6. I am aware that the authors tried their best to reuse the data from the current GNSS networks. There are still limitations concerning the raw data. I look forward to seeing a longer-time series of snow depth products from more sites and systems (such as from China’s Beidou). I also encourage the authors to find some way (e.g., making a website?) to maintain and share the methods and data sets to broader users?

I think this is an excellent paper describing a new snow depth dataset.  My comments below - tagged by page number/line -  are meant to improve the readability and value of the paper. 

Some of my comments are directed to the figure captions. I think the goal should be to allow people to read/look at the figures without reading the paper.  So this means they haev to explicitly say how many sites are in the figures and so on.

My main technical comment is that the authors describe whether they set an azimuth mask for each site. I do like that they are investigating new ways to evaluate the QC for their sites, but we really need to know a little bit more about how they did that (peak ratios e.g.).  I agree that setting the bare soil values is complicated - using NDVI is a good way to get a handle on whether the vegetation is dead and thus a better proxy for bare soil.

I would also ask the authors to add a few more sentences about how they dealt with the non-repeating ground tracks for Glonass satellites. 

I do echo the comment of the previous reviewer. For this dataset to be truly useful, it needs to be easily found. For soil moisture, I could point to the International Soil Moisture Netwrok. Even though my soil mositure project has ended, the ISMN has provided a way for researchers to use our soil moisture data without contacting me directly.  I do not know the best place for snow data, particularly for the international community.

First page

A new snow depth data set over northern China by developing a comprehensive framework using the complex GNSS station network

I would suggest a slightly different title. We don’t use complex in quite this way.Yes, the dataset is complicated - but really the issue is that the sites were not located in an ideal way for snow sensing.  Maybe :

“A new snow depth data set over northern China derived using GNSS interferometric reflectometry from a continuously operating GNSS network.”

This is a suggestion - not any kind of required change.

page 3, line 67 I think you should explicitly say in the US (the reader should not need to go to the website to find out that SCAN and SNOTEL are only in the western US).

page 4, line 92 For typical GNSS sites the spatial footprint is ~1000 m^2.  (rather than saying recognized).

page 4 last couple sentences my copy has the commas in 9,000,000 and 4,200,000 as ‘ rather than , 

page 6, line 138

and the stations have turned into a certain amount since 2012.

better to say “ and the station build phase was completed in 2012. “ ???

page 7

I think it is sufficient to say you used the broadest ephemeris. They should be the same everywhere and it should not matter where you got them from.

page 7, last line.  the RINEX files may truly only have 10 degree data - but was that imposed when the file was made or when the GPS station was set up?  It is pretty unusual to set that at the RINEX creation stage (not saying it wasn’t done, just asking to be clear). 

Page 8

Shouldn’t there be references for these data, SMAP, NDVI data etc?

Add a reference for Lomb Scargle Periodogram

As part of your quality control, I think you have set a mask for each site,but you don’t really say it.  Why not make it clear?  You should not bother calculating reflector heights if they are not useful for snow sensing.

page 9

mean Peak-to-Noise Rate (PNR)

I think you mean peak to noise ratio, not rate

page 11

line 219 are analyzed in Table 1,

You really mean to say that they are listed - not analyzed

Even though, the SNR model has been verified to have higher accuracy than the L4 and F2C models (Liu et al., 2021).

remove “Even though”

It is worth mentioning that, for GPS and GLONASS satellite, the reflector height is given per satellite, quadrant, and frequency band, while for BDS satellite, the reflector height is given by quadrant only because the BDS MEO satellite changes its trajectory day by day.

GLONASS does not have a daily repeatable ground track. What do you do to account for this?

Figure 13

The caption of Figure 4 should be expanded to describe the subplots, especially the ones with data in them. I know what is in the figures, but most people would not.

I also suggest using grid lines in the plots.

page 14

The majority of the CEA antennas are settled down on a standard rooftop, with the GNSS receivers being put in the accompanying small house. It explains why most of the CEA sites are not suitable for snow depth retrieval.

I suggest you say “set upon a rooftop.”  Settled is used for something that starts at one height and slowly changes, like snow settling.

page 16

Figure 6 shows the rule being applied to six individual sites with various surroundings,

You say this before you define the rule. I think you really need to define the rule first.

Figure 6

How do we know that these are good retrievals? Maybe some of the periodograms have poorly defined peaks?

page 16

the three sites at the bottom of Figure 6, i.e., uqwl, qhdl, and qhbm, show continuously changed hX values. It

indicates that it is unreliable to determine a true hX by the Lomb-Scargle spectrum due to complex environmental 280 conditions.

I know what you are trying to say here, but I think it is simply a matter that you are computing reflector heights at sites where you should not bother. The sites are surrounded by too much clutter.

When you are using your rules to find good sites, do you prepare and save azimuth masks?  It is not clear to me that you do.

page 18

0 ~ 12 am or 12 ~ 24 pm within one specific day.

You should say 0-12 UTC and 12-24 UTC. 

Figure 7 is this for one site? Which one?

Figure 9 add the number of sites this represents in the caption. My recollection is that you don’t have that many that observe both GPS and Glonass.

Figure 10- Couldn’t part of the difference between L1 and L2 be due to the phase centers not being in the same place? Did you assume the phase centers were the same? How many sites are shown in these figures?

Figure 11 add the 1 to 1 (diagonal) line as you had in Figure 10.

How many sites are represented in each figure? This information should be in the caption or in the figure.

page 23

In addition, the peak of the PMW snow trend for each snow season moves to the right,

Say instead that the peak is later in the season rather than moves to the right

Figure 12 Explicitly say in the caption how many sites are shown.

page 24

This result indicates that the laser measurements in operational meteorological observations are not always reliable.

Could say “this result is a reminder that operational laser measurements of snow depth are not always reliable.

page 30

First, the minimum elevation angle of GNSS satellites should be set to 5° ~ 15° to preserve the multipath effect as much as possible because only data with low elevation angles can show the surface reflection.

I don’t understand this sentences.  If you had said “set to 5 degrees” I would agree because you said you are setting a minimum, which should be a single number. I do not know why you mention 15.  Despite the fact that these organizations unfortunately used an elevation mask of 10 degrees, your paper demonstrates that stations with an elevation minimum of 10 degrees can be used (though not optimal).

Third, the cycle slip of GNSS observation can severely reduce the data quantity available for snow depth retrieval.

I don’t understand why a cycle slip would by itself reduce the data quantity for SNR?

page 31

Also, the snow depth results on snowy days could, to some extent, affect the accuracy.

What do you mean here? I don’t disagree with you - but if you say it, you have to explain why you say it.

General comments:

The authors present a new snow depth dataset using a network of GNSS stations in northern China.  The data were derived using a varied of established methods depending on data quality and instruments available at the different stations. They furthermore propose a method to automatically control the data quality.

The quality of the obtained snow depth data is evaluated comparing the different GNSS snow derivation methods with each other and to in situ manual measurements and passive microwave data.

The presented dataset is a valuable contribution to the research community and the paper should be accepted for publications after the following points are taken into consideration.

The RMSD (Figure 9 and 10) and STE (Figure 11) should be given also as relative errors (%), allowing a better comparison to other studies. This is particular important since the snow depth in the studied area is particularly low compared to other snow-covered areas in the world. Moreover, data points with < 5cm were included in analysis and seems to be the majority of the datapoints (See figure 10). However, later it was stated that the obtained results are not reliable for snow depth < 5 cm (line 493). I see a contradiction here. Due to the high density of points with very low snow depth the reported deviations can be misleading. Consider also providing the analysis only for datapoints above a certain threshold. It would be also useful to add a regression line to the scatter plots in figures 9 to 11.

It would be nice if more recent data would be included in the dataset.

The site info are given only in .doc an .xls formats. It would be good to have that information also in non-proprietary format as text file or csv.

Detailed comments:

Line 153-155: I don’t understand what the authors mean with: “ … we preserved the high-quality and medium-quality sites as much as possible…”

Figure 5: Please indicate which sites are considered high quality or low quality in figure caption.

Line 276: Please explain why the site bgfc is has stable h0 only at specific orientations.

Line 284:  a value range of 0.5 m is still pretty large if it is considered that the snow depth to be measure is generally lower than 0.3 m.  Can you comment on this point?

Line 316: Specify the length of the moving window.  10 datapoints, hours, days, …?

Figure 7: The oscillations/noise are much bigger than the reported errors. Can you comment on this?

Lines 321-340: Is the volumetric soil moisture not changing over the year? How is this accounted in the data derivation. What is the effect of soil freezing?

Line 354:  Please indicate relative to which quantity the standard error is calculated?

Figure 12: I find quite unusual to compare the seasonal evolution of the snow depth only for the mean of the 17 sites since in this way oscillations and outliers are probably smoothed away.  It would be good to see also the comparison of the 3 methods for single sites, which would give an indication of the validity of the data in the single cases.

Figure 13: It would be good to indicate the standard deviation or RMSE for the mean and max values.

Figure 15 and section 4.3: It is true that the GNSS dataset deliver a higher data rate than the other methods. However due to the lack of reference data at the same rate it is impossible to judge the quality of the 2h dataset. In fact there are several discontinuities in the GNSS derived snow depth (sharp decrease and increase of snow depth) that are normally not seen in snow depth data from a snow storm. It is in fact not possible to judge what changes are due to the real snow depth evolution and which changes are due to artefacts that could be due to i.e GNSS satellite configuration or snow deposited on the GNSS antenna.  It is unfortunate that the in-situ snow depth data are not available at higher rate which is normally possible for laser snow depth instruments.

Line 550: It is not clear what the authors mean with this sentence.

Figure 19: I’m missing a yes or no on the diagram arrows indicating in which direction is taken after a decision.

Conclusion: Please indicate also the (relative) error of the GNSS data compared to the in-situ and PMW data. Not only the internal consistency between the different GNSS systems.