Dear authors,

Thank you for your revised version of this manuscript.

One of the reviewers rated it very unfavorably and recommended rejection. I have now carefully worked though all comments and the paper and find it in a state with many open questions with respect to the validity of the findings and open issues with respect to the reviewers’ comments.

I believe the topic of this study to be important and relevant and am convinced that you have a very good dataset. At the same time the manuscript requires important work not only on its presentation, but likely on the analyses, data processing, and interpretation.

The manuscript goes back to you, flagged for Major Revisions and the issues described below could equally justify Reject with an encouragement to resubmit. Therefore, please be sure that you can thoroughly address the points below if you decide to resubmit, and give the manuscript a very good editing for structure and rigor.

Kind regards,
Stephan Gruber

(1) Quality control: The argumentation in lines 202-206 is not clear with respect to how this test was performed. If up to 65 days are missing, how are the start/end dates computed to within 2 days? Depending on the timing of the gap that cannot be done. If the averaged result over all stations only changes by 2 days, than the spatial coverage may still be polluted by heavily flawed values. This argument needs to make sure that for each station included, the values derived do not change by more than a certain threshold.

(2) Urbanization: Similar to the quality control with respect to missing days, the effect of urbanization should be controlled for at the level of individual stations. Aggregation to areas such as the one shown in the new Figure 8 does not necessarily inform what happens at the stations. Given the size of the aggregation domain, there could be unchanged rural stations, even in the polygons labeled “high”. On line 405 and following, “This is because…” you invoke a causality that I do not see justified from the evidence presented.

(3) As Reviewer 2 pointed out, actual values are needed in order to provide context for interpreting anomalies. The relevance of a 5-day change is different for a freezing duration of 2 weeks as compared with one of 4 months. Include maps of the actual values and a plot showing the trends as a function of e.g. freezing duration. The paper in review you mention in your response has not been mentioned in your reference list or in the cover letter – it would be very unfortunate to have the values and their changes surgically separated into differing publication.

(4) A better conceptual framework (and sketch?) is required for discussing what GST is and what drives it. If you relate GST changes to climate, urbanization, etc and you state that changes in snow cover are negligible, it would be useful to first describe the system under investigation and what drivers control it.


Other comments:
Throughout: (a) replace “altitude” with “elevation”; (b) consider “unfrozen” instead of “freeze-free” (e.g., L83); (c) consider replacing “soil freeze” with “soil freezing” (e.g., L85); (d) better indicate the unit for Figures 2 etc. are the numbers days, or anomalies w.r.t. a certain time period; (e) the English needs substantial editing; (f) insignificant means that something is negligible, non-significant however means that something does not show a statistical relation with a p-value you accept as satisfactory. This is a big difference, please correct in your text.

Measurement method: If the thermometer is buried half and exposed half (L153) how can it be read-out daily in regions with a snow cover without causing a significant disturbance? It is not sufficient to believe that (L163-L166) that station relocation was not significant. Please indicate the number of stations without relocation (also on the map) and test if the subset without relocation produced a different result.

L215 and Reviewer 2 Q10: It still seems to me that what you describe is called the “coefficient of determination”. Your answer to the Reviewer does not actually explain that. Please explain in the methods section, how you use this and what insight is derived from it.

L34: “coupled” is a very vague term. Do you mean “related to” or “caused by”? Be more specific.

L148: Only acknowledge the SRTM in the figure legend, this is not a method influencing your results.

Figure 6: Rescale the horizontal axis of the bottom right part from 0-2km.

Figure 8 and 9: Do you mean “median” or “medium”?

The first bullet of the summary seems contradictory (5 days or 10 days?) and may miss a word after “soil” on L441.

Dear authors,

Thank you very much for your substantial and thorough revisions to this manuscript. After renewed review, one referee recommends acceptance as is and one asks for minor revisions. I concur with the request for minor revisions and have included below a list of items to be addressed, also those raised by the reviewer. EC1 and EC2 border major revisions as they may change the results of your analysis somewhat.

EC1: How do you justify removing outliers (cf. RC3, below)? The hot summer of 2003 in Europe was 5 sigma out from the mean and a very real and relevant event. Consider not removing outliers or finding a cleaning procedure that will check if neighboring station also have outliers.

EC2: Why do you only keep stations with significant change for your analysis (L122)? I assume this is the reason why numbers a lot smaller than 636 are reported in Sections 3.2–5 and in Figures 3–6. Your results and abstract state that you used 636 stations and that certain changes in ground freezing occur. Are those numbers really informative if you exclude the time series with non-significant trends? In the extreme, you could have only one significant station and then conclude on an overall trend. Please consider revising this methodology (and adding non-significant stations as e.g., black dos to Figures 3–6) or explain again if I misunderstood you.

EC3: In your conclusion, you present the result of auxiliary analyses performed only in your discussion (L433: “…, the North …. few decades.”). This should not be presented as your major investigation or result as it did not receive the same level of analysis as the freezing regimes. Do you really need the NAO and AO analysis to discuss your results?

EC4: Please address these small items: L177 “…, primarily due to…” Delete rest of sentence or move to discussion. L76: “… at THE ground surface…” L77: “ground-surface temperature and mean…” (no comma) L79: “Temperature monitoring was conducted…” L82: “ Although the measurement standard states…” (the, delete s)

RC1. The kriging procedure, Fig. 2: The soil freezing critically depends on a number of factors, such as altitude, as investigated in detail by the authors. The map, however, is simply an interpolation between the individual stations, without considering e.g. the real topography, and changes in other factors. If all stations were for instance located in valley systems in an otherwise mountainous region, the “average” soil freezing of the region would not be adequately represented. I agree that Fig. 2 is a nice illustration, but the authors should critically evaluate (and discuss in the manuscript) the possibility of false representations and interpolation artifacts in certain regions. For instance, is it reasonable to believe that the three dark blue spots in Fig. 2C are really localized areas with more pronounced ground freezing, or is it simply the effect of the particular setting of the station in the center of each region?

RC 2. For the last date, the rate has a plus-sign in the abstract, and a minus-sign later (l.182). In the abstract (and l. 182 again), it says that “the last date was advanced…”, which at least I misunderstood in the beginning. Just state that the last date occurred earlier, or something similar, then the direction of the change is clear.

RC3. L.131 ff: What’s the reason for rejecting “outlier” years? Is it reasonable to assume that such an outlier is due to bad data quality? Or could it not just as well be natural variability which should be included when calculating trends? In the worst case, excluding outlier years could even make a trend statistically significant, which otherwise would not could not be statistically secured.

RC4. Sect. 4.2: It is excellent that the authors investigate a potential influence of urbanization. However, it does not become immediately clear why the authors chose to do this somehow involved analysis and not simply excluded stations which are in urban areas at present. The text seems to suggest that there is a digitized product of urban areas, so it should be straight-forward to filter out the stations that are now in urban areas, and simply calculate the trends with these stations removed. The authors should clearly state why they chose to do the analysis this way, and why “detailed investigation of this issue is far beyond this study”.

Kind regards,
Stephan Gruber