The manuscript of Groos et al. presents a dataset on ground temperature data: At 5 + 14 locations in the Ethiopian highlands, ground temperature was monitored at three depths (5) or near the surface (14 stations) for nearly three years. The manuscript contains a first analysis of the data with focus on the occurance of frost, annual and seasonal temperature variations.

I am not an expert in the area of ground temperature dynamics. Thus, my comments are mostly restricted to the overall content of the paper, its structure and technical details.

That being said, the authors convincingly lay out the potential use of the data set. Moreover, the logistic and technical challenges of the resprective measurements in this remote area underline its uniqueness.

However, with effectively only three station providing true temeprature profiles (albeit 3 depths anly), and timespan of just 3 years and many data gaps (although reasonably-well filled), the data set is not excessively rich. Therefore, I strongly suggest to include the concommittant measurements at the meteorological stations into the dataset. The paper often relates to them (e.g. ll), and even includes their analysis (p. 12)., as they greatly enhance its value or seems even mandatory (as stated in the Discussion) for its analysis. I was even somewhat surprised not to find them. As for meteo-data of secondary provider that cannot be included, at least the respective reference would of great help to the potential user.

The core data of the publication are available from a Zenodo repository. However, they seem incomplete in some aspects, while redundant in others. I recommend some restructuring:

- Naming the folders "raw data" (i.e. log files) and "processed data" would seem more intuitive to me.

- Some files (e.g. Hourly_Ground_Temperatures_Corrected.csv) have a deviating formatting of the date column. Highly impractical for automatic use / scripts

- Information_Sheet_Data_Correction.ods: For each logger, add a column "flag" with indicator(combinations) for "interpolated to full hour", "missing/filtered", "interpolated/gap filled based on logger nn"

This potentially makes the other files (except, perhaps, Information_Sheet_Data_Interpolation) obsolete.

- The duplication of *.txt and *.ods seems unnecessary. I suggest using txt for the actual data, ods for the meta-data

- Please add a GIS file (or at least table with the station coordinates), the information of Table 1 plus further site attributes (vegetation, etc.)

- Please add some reference to the ESSD-manuscript (e.g. with an overall readme.txt and/or in the Zenodo desciption field)

The overview data analysis in the paper provide an adequate first insight into the data. As this is not the core of a data publication, I recommend not extending them. Please consider shortening by excluding the analysis of teh effect of slope and orientation. However, I made some suggestions how to make improve some points, if these are considered essential. Some conclusions drawn should probably be formulated more carefully, as they do not seem to be fully backed up by the analysis.

The formal quality of the paper is high, with logical structure, adequate style and (mostly) helpful figures. As a technical suggestion I strongly urge to decrease the file size of the resulting PDF - 20 MB for a text document is inacceptable. I assume this comes from photos with excessively high resolution .

The dataset and paper merits publication, especially when supplemented with the meteo-data.

However, I recommend moderate revision before.

Further details in the annotated PDF.

This paper outlines a unique dataset on ground temperatures recorded over a 3 year period on the high plateau of the Bale Mountains. The area is underrepresented in terms of past climate data records and the data is of clear use in understanding the current environment of tropical high mountains. It is also of interest as a dataset collected in remote circumstances and I think therefore that this paper and the data are of value and should be published. There are some concerns however. The data is somewhat messy, contains a lot of gaps, and the time period is not extensively long. The value would be significantly enhanced therefore if it was to be combined with attendant meteorological measurements (particularly air temperature) which have apparently also been measured in many of the same locations at similar hourly resolution.

Since this is a data paper the communication of the dataset and its organisation are important. I am a little confused by the numbering of the locations/loggers and the rather unsystematic organisation of this aspect. I know several loggers were stolen and there are no observations as a result. I would recommend ignoring these and numbering the sensors with data starting with TM1 and GT1, and somehow creating a more logical order (maybe high elevation to low elevation sites in order?). Loggers that were lost don’t really need numbers in the final dataset. Figure 3 can then be ordered in the correct order (starting with TM1 and GT1)… You could even number the loggers at the same location but different depths GT1h (2 cm), GT1m (10 cm) and GT1l (50 cm) -  i.e, high, mid, low (or something similar, a, b and c) to make it more obvious which are in groups of three. A more intelligent numbering system would make the dataset easier to navigate and make it appear less a collection of disparate sub-experiments.

Figure 1 needs to show all the weather stations (two are off the bottom edge of the map) and I think some contours would help show the hypsometry more clearly – the current shading is somewhat confusing and concentrates on landforms rather than elevation bands… perhaps 500 m contour interval?

There are two sets of calibration data between the high and low cost loggers in the lab and in the field. Again it would be best if this was clearly accessible and perhaps made available in a sub-directory, since this is important information. Having said this I am somewhat concerned that the field calibration is of limited use, since one logger was installed slightly lower than the other (page 10: line 32). This seems like an error. The details of the lab calibration in the text are vague (page 5, line 31). It says “several hours” and does not say under what controlled temperatures for example. If there is too much information for the main text put it in a text file with the calibration data.

I think the data itself might be clearer in 3 directories, a) raw data, b) corrected data and c) complete time series with gaps filled. The current structure is a little confusing. Put the readme files concerning each stage in the relevant directory.

I also have some comments about the analyses and findings.

Much of the data analysis concerns comparing ground temperatures with equivalent meteorological variables measured at Tuluka (3848 m). Since this AWS is not adjacent to most of the sensors a more logical choice would be Tullu Dimtu (4377 m) which apparently also has an AWS. Can you explain why this station is not chosen?. If it has similar data I would suggest using this site. Having said this, Figures 4 and 5 are very useful. Vertical lines separating each season (NDJF, MAM, JASO) would help the reader see the seasonal changes each year much more clearly. I am also wondering whether June is a short (but cloudy) break in the two wet seasons since it does appear that there is a short dry period around this time (any comments?). Maybe this is a fourth season?

The analysis of slope aspect is good (I would keep it) but it is not just the timing of the peak soil temperature that is changed due to aspect (page 12, line 14). The peak is much subdued on the north facing slope because the sky is cloudy during June when the sun is at its most northern point in the sky. The cloudless period coincides with when the sun is near its most southern trajectory. Thus, this explains the much higher readings recorded on the south face.

The lapse rate relationships in Figure 6 appear to be skewed by an outlier which is much warmer than expected given the elevation (I guess it is TM12 since its elevation is just below 3800 m), particularly in NDJF (when it is often sunny). I suspect therefore that this site is south facing (or has a distinct microclimate) and I would drop it from the lapse rate calculation.

Some specific comments:

Page 2, line 1. It is not always true that high mountains in the tropics receive more precipitation than adjacent lowlands (see Kilimanjaro for example) and precipitation often decreases on the highest summits. This statement is a bit misleading.

Page 2: line 12: elevation-dependent warming (and places elsewhere)

Page 2: line 16: define longer records (I think the original context was >20 years)

Page 5: line 21: TM04 was used as calibration but then no longer used in the field. Any reason? This whole section is a bit vague on detail.

Table 1: It strikes me that the elevation range of 3493 m to 4377 m (marketed in the abstract) is rather optimistic since the lowest station was only recorded for a year, and without this station the range is only ~600 m.

Page 9: line 14 ff: It would be good to have the regression equations listed somewhere in the metadata files, rather than just r2 values and RMSE. This enables someone else to replicate your work.

Page 10: lines 23-28. This paragraph seems out of place… it is about reliability of method and should come after everything else about the dataloggers – or in the method section.

Page 11, line 6: where does the 2°C figure come from (data source?)

Page 11, line 23: Also the solar angle is lowest in Dec/Jan – with a maximum elevation of only around 60° at the December solstice – yet it is overhead in Apr/Aug.

Page 11, line 27: This is so much higher than the sites shown in Figure 4 (which are also at 2 cm) and must be a result of specific soil properties or the datalogger becoming exposed to radiation at the surface? Can you comment?

Page 11, line 30: cold air ponding is an interesting hypothesis but do you have any evidence? i.e. from air temperatures?

It strikes me that an analysis of frost incidence at each site would be really interesting. Perhaps a histogram showing the number of hours below freezing and its seasonal distribution at each site would be a useful graph. This is especially important, given the context given for the research which is about permafrost and peri-glacial landforms.

Page 12, line 14: the sun is not in its zenith in Jan/Feb…. it is overhead in the southern hemisphere. It may have a high local angle of incidence on the south facing slope, but that is not the same thing.

Page 12, line 28: the five sites which are being continued need identification in Figure 1, Table 3.

Page 13, line 22: the mean annual air temp is how much lower than soil temp?…. some figures would be useful here

Page 13, line 28: not just the timing, but the amplitude of the seasonal cycle

Page 14, line 14: elevation-dependent warming (as earlier)

The conclusions are a bit similar to the abstract. I note that ground frost is again mentioned here as a major finding, yet there is no analysis of this aspect (frost frequency).

I hope that these suggestions will improve the organisation and communication of the findings.

The manuscript “A multiannual ground temperature dataset covering sixteen high elevation sites (3493–4377 m a.s.l.) in the Bale Mountains, Ethiopia” by  Groos et al presents a dataset from 29 ground temperature loggers that were installed at 16 sites and operate since 2017. The data was measured from 2 to 50cm depth, using single-channel miniloggers, UTL-3 Scientific Dataloggers providing high resolution and accuracy temperatures, and tempmate.®-B2 providing low accuracy temperarures. The measurements were obtained at hourly intervals. The data that is made available in the repository Zeondo is relevant for research and applications, mainly due to the scarcity of ground temperature observations from mountains and highlands in Africa. It has the potential to be used for soil climate characterization, geomorphic dynamics analysis, as well as to locally validate climate modeling or thermal remote sensing imagery. However, the dataset shows several relevant gaps that occurred due to logger malfunction. These gaps were filled by simple statistical techniques based on correlation. Both the original and the full infilled series are presented. The series were characterized using statistical measures, including some widely used indexes in environmental studies, but the results are not analysed in depth. The manuscript is, in general, well organized, clear and the language is precise. I must, however, note that English is not my native Language. 

Despite the overall clarity if the manuscript, I think that there are some significant issues that the authors must address to improve the manuscript and the applicability of the dataset before it may be accepted for publication.

The coding system used to identify the data is cumbersome. It is based solely on the logger IDs (given as numbers, which are in some cases, non-consecutive spatially). I strongly suggest using an ID system allowing to identify the location, depth of measure and the type of logger. This would facilitate data analysis.

The characterization of the soil at the different sites where the instruments were installed is highly needed and the manuscript needs this information before its acceptance. This issue is of most importance, both for the analysis of the datasets, but also because data from different sites are used to estimate missing temperature values.

The procedures used for data-filling must be described in more detail. In particular, all correlations, p-values and also scatterplots of soil temperatures between the sites (reference vs corrected) should be presented. The limitations of the procedures should be also described, especially in what concerns to the distance between sites and the potential impacts of the types of vegetation cover, soils and topography in the data. It seems that some depths were infilled with regression using temperature profiles from other sites. How were soil type and moisture differences considered for the procedure? These are factors that may strongly influence temperature change with depth and dynamics and they are especially relevant in environments with ground frost.

I also have some more specific comments:

- In the title, the data period (start-end) should be indicated.

- The abstract should be improved. For instance, in line 9, the authors mention that the series provide new insights, but what they say is too vague. They should concretize some of the results of the temperature analysis that they have done.

- In the last paragraph of page 3, It should be mentioned the type of climate according to Köppen classification, and it should be mentioned why the wet season is bimodal. It is confusing to mention that there are two rainy seasons and only one dry season. It could be mentioned that this is a transitional regime and that the two peaks in precipitation follow the movement of the sun and it can also be indicated the importance of the Intertropical Convergence.

- In page 14, line 8, it doesn’t make sense the mention of the endemic giant root rat. The reference should be contextualized.

- Figure 1 could be improved if the type of font/type/size of letters discriminate the information that they relate to.

-  in fig 3, as mentioned before the acronyms of the loggers should be modified in order to simplify the reading.