Overall, the paper is well presented based on outstanding field and model works.  The data and metadata as well as methedology presented in the paper are very helpful to the geoscientists and engineering in cold regions. We all know that data sharing in permafrost temperature study has been rather difficult. These ground temperature data are thus invaluable in evaluating the thermal state of permafrost and for validating many gecryological, hydrological, ecological and land-surface processes models, and for engineering design and construction in elevational permafrost regions.

The structures of the paper are well thought out and basically follow the ESSD mandates. However, the authors are encouraged to tell more on the methods of air and ground temperature measurements and their evolutionary paths, since different measurements methods can result in false trends in climate or permafrost changes. For example, your FDD or TDD or your MAGT@DZAA is based on ground or air temperature measurements, and the methods have been advancing rapidly. In the same time, the authors should be more explicit on the criterion selection as why 15 m can be regarded as the DZAA, for which it is evidently illogical. In the meantime, positive MAGT does not necessarily means absence of permafrost because of extensive and increasing presence of supra-permafrost subaerial talik, especially to the east of the QTEC from Golmud-Lhasa and along the engineering lines. Thus, a criteron of subzero MAGT for judging the occurrence of permafrost may underestimate the permafrost extent. That means, you have to be cautious of areas with rapidly or chronically degrading permafrost. Lithology and soil moisture contents are key in defining local or regional DZAA, ALT and MAGT. Thus, using a given depth of either 10 or 15 m as DZAA seems not so reasonable. Thus, if you chose 15 m as the DZAA, you'd better convince readers that your choice is acceptable.

In addition, ESSD papers should try to avoid over-interpret the patterns or trends of data. It is supposed to tell the integral story of the data structure and functions. Please make the paper concise and on the point, avoiding unnessary details as possible.

Other minor issues regarding editing of the MS are provided on the marked MS in the attached document. This is a very quick editing. It is up to authors to ensure the presenting quality to suffice the ESSD standards based on meticulous efforts.

Review of Zou et al., 2024: Permafrost temperature baseline at 15 meters depth in the Qinghai-Tibet Plateau (2010–2019)

General comments:

Zou et al. present a dataset that extrapolates ground temperatures over the QTP at the depth of zero annual amplitude (here determined to be at 15 m depth). They use a support vector regression to predict ground temperatures based on nine environmental predictors. They justify this approach by claiming that this method has been shown to be superior to other supervised learning algorithms such as random forest in one study (Ran et al., 2021). While the dataset is novel in the sense that no ground temperatures at 15 m depth have been predicted with this method in the QTP, I have a few concerns about the methods used to create the dataset and the fact that a similar dataset exists on a pan-Arctic scale for the entire permafrost region through the permafrost cci ground temperature dataset. Dismissing this dataset solely on the grounds of it not reaching as deep as the dataset presented in this study is not sufficient in my opinion. Especially considering the fact that the authors claim that the DZAA ranges from 10 to 15 m in central Asia and therefore would partially be covered by the permafrost cci product. Furthermore, the R2 value of the prediction is below 0.5, meaning that less than half of the variance in ground temperature can be explained by the model. This suggests that the model could potentially be improved or a different model should be tested to see if the predictions accuracy can be increased.

I do not want to dismiss the work that the authors have put into this dataset, however I am unsure if it offers a significant contribution to the scientific community in its current state. I have a few suggestions on how to enhance the impactfulness of the paper, but I am unsure if it then still fits the scope of ESSD. 

1. The SVR method has been tested by Ran et al., 2021 and found to be sufficient for their purposes. However, their R2 was 0.71 as compared to 0.48 in this study. Further, they have tested various different supervised learning algorithms to conclude that SVR is the best model to use, which is lacking in the present manuscript. Hence, I would suggest the authors also perform a test for the other models in question that can be used for this task to get a better idea of their individual performance. 
2. Currently, the dataset is presented as a stand-alone dataset to be published in ESSD. However, the overlap with the existing permafrost cci ground temperature dataset can not be denied. My suggestion may significantly change the scope of the paper, but I wonder if it would make more sense to use the borehole data used in this study to assess how useful the ground temperatures could be to inform e.g., boundary conditions of permafrost models in the QTP. As the authors describe, the boreholes are equipped with thermistor strings, which probably means that measurements are available at several depths. This would serve as a basis to compare the borehole data directly to the ground temperature dataset at 10 m depth. A comparison to the existing dataset could then be a better motivation to conduct your own supervised learning method to improve the accuracy. However, if the R2 is similar or higher when directly compared to the existing data (permafrost cci), there may not be a need for this since depth extrapolations of temperatures below the DZAA could be achieved with geothermal heat flux and simpler heat conduction models.

Regardless of the scope of the final manuscript, I think a comparison to the existing datasets is crucial, considering the model in this study explains a relatively low amount of variance in the data.

Specific comments:

1. L56: What kind of datasets are you talking about here? Either delete the last part of the sentence or give an overview (for example in a table) about the datasets you are talking about here.
2. L75: Do I understand correctly that you implemented a procedure to fill temporal gaps in 78% of the data based on 22% of the observations? Please clarify.
3. L77-82: From what I understand, you used 51 sites to calculate a linear trend to fill the gaps in the remaining 180 sites by assuming they all experience the same warming trend. However, your Fig. 2a clearly shows that warming trends are very different for cold vs. warm permafrost. I think applying a single warming trend that is based on 22% of the data is very problematic here. If I misunderstood this part, please clarify. Otherwise I am doubtful of the reliability of this preprocessing step.
4. Eq 1. An R2 of 0.45 does not create a lot of trust into your interpolation method (see comment above). 
5. L107: I am not very familiar with SVR, but is a 90/10 a typical split for this method? I was expecting a 80/20 or even a 70/30 split since you do not have a very large dataset. Can you provide the model performances with different splits? And how high is the risk for overfitting with the 90/10 split?
6. L132: “high accuracy” is inappropriate here. How do you determine it is “high”? The indicators you are describing are not creating a lot of confidence.
7. Fig. 3: Please add a label for the red line either in the figure or in the caption.
8. Fig. 4: Maybe I have missed it in the text with all the numbers, but did you say that you are masking all values > 0°C? It looks like the final dataset only shows values < 0°C. Is that because you do not have confidence in non-frozen conditions? Are you assuming that there is no permfrost in regions with T > 0°C? Please clarify this throughout your results section.  
9. Section 3.2.2: This section is very difficult to read. Would it be possible to put all those numbers into a table, refer to it in the text and focus on the conceptual characteristics only? 
10. Fig 7.: What are the units in the figure legend? I assume °C?
11. L259-261: This sentence is very confusing and I am not able to follow it. Please see Biskaborn et al., 2019, which you are already citing, for an example on how to describe the difference between warming of “cold” and “warm” permafrost and how it relates to latent heat consumption.