The manuscript presents a dataset comprised of a series of thermal maps/images and RGB images collected from a single observation station looking across a snow-covered mountain area in the French Alps. The paper presents perhaps the most rigorous and best example of calibration routines for thermal imaging over snow surfaces (and perhaps for any land surface thermal imaging) in the current literature. The authors address most of the major sensor limitations of microbolometer thermal sensors; including sensor drift, internal temperature, bias correction, emissivity, etc. This is a technically very challenging task and there are many valuable insights in the workflow presented that will be of great use to the research community. The datasets themselves are likely to provide a useful resource for snow studies and calibration of future satellite missions (if the dataset continues to be collected contemporaneously with these), alongside being a useful dataset in their own right for studies of snow surface energy balance, heterogeneity, etc. Furthermore, the manuscript is well written, and easy to follow.

However, I do have one major comment on the paper. While the authors account for most of the error sources of microbolometer sensors they do not address the potential impact of variable viewing angles and variable distance to target on their measurements/dataset. Studies have shown these can both potentially significantly impact thermal camera measurements. The impact of these on the datasets is likely not apparent in the error assessment against the more accurate IR sensors on the AWS because these are all located at roughly similar distances from the camera and have similar viewing angles. As a result, I believe the reported absolute error values of 0.67K (spring 2023) is probably only representative of areas with similar distance and viewing angle to the AWS stations and is likely worse in other areas of the study site, these limitations should ideally be investigated to confirm or negate their presence. If present the datasets could be improved by addressing these issues, however this is likely to be fairly labour intensive and the dataset and manuscript in their current form are still at the forefront of research in this field. To produce the most accurate and thus useful dataset I suggest the authors identify and address these error sources, however given the high quality of the paper and datasets already I would be satisfied if their potential impacts and consequent limitations were discussed within the paper, at a minimum.

More details on the issues mentioned above and specific comments are included in the attached pdf. 

This paper presents a very thorough description of this dataset of thermal infrared imagery, how it was collected, calibrated, and sources of error quantified. The rich dataset will no doubt prove useful for validating the retrieval of land and snow surface temperature from satellite remote sensing observations. The detailed descriptions of instrument setup, lab calibration, and ancillary datasets used will also provide a good roadmap for others to set up similar observations in other locations. There are two things that will make this dataset particularly value, it is of high quality (or rather, sources of error have been investigated and quantified), and it is a rich dataset, providing high temporal and spatial resolution observations within two snow seasons (and sufficient ancillary datasets such as local meteorological variables and internal camera temperatures). However, there are two main points that should be addressed in the paper to better describe or quantify sources of error or uncertainty before it is published: 1) angular emissivity, and 2) the parallax effect on large trees in the image projection.

1) Emissivity:

The authors note the high, near-blackbody, emissivity of snow surfaces with some variation due to grain size, but neglect to address the angular variability of snow emissivity (see Dozier & Warren 1982, which the authors did cite). The difference between observed temperature using a blackbody assumption and the true surface temperature of snow is likely much larger than the errors from other sources that the authors did investigate (e.g. window transmissivity). High spatial resolution DEMs of the study site were used in the projection of the imagery, therefore I suggest that the authors use these same DEMs to estimate snow emissivities given the varying view angles across the study site. Alternatively, an additional map could be provided with the dataset to show the variation in view angle across the study area so that a user of the data can apply their own angular emissivity corrections across the image.

2) Image projection:

Inspecting the TIR imagery, I noticed an area in which there were significant parallax effects due to the images of trees being projected onto a large area of adjacent terrain (see attached Figure 1). It appears that this line of trees (behind a snow fence from the camera’s point of view) occupies a slight rise in terrain before the terrain dips into a ravine or stream valley below. When the TIR images are projected using a bare-ground DEM, the sides of these trees are therefore draped across a large swath of terrain, making it appear that this terrain is the same temperature as that of the sides of these trees. Though the parallax effect causes this same “lay over” of trees elsewhere in the imagery, those instances are of much less concern since the terrain behind the trees rises uphill (therefore the area of terrain they are projected onto is much smaller). Ideally, regions of the image where the land surface is not visible (such as behind large trees) should be masked out like was done for the hill slopes hidden from view, and/or a digital surface model should be used for projection (one that includes the “surface” of large vegetation). Otherwise, it would be sufficient that the paper should at least mention this issue in the images, and call out the one particular region where the effect is most significant. Users of the imagery data can then mask out this region themselves. (And a related minor note, the road passes this same location as this stand of trees such that if an image was taken just as a car was driving by, the projection stretches the car across the terrain creating a narrow streak of very warm temperature in the image. See attached Figure 2. This is understandably not avoidable, but would be worth mentioning in the paper as an artifact of the image projection.)

Lastly, I have one minor comment: Section 4.2 describes quantifying the transmissivity of the germanium window of the TIR camera housing. Are there specifications from the manufacturer of this window that these results can be compared to?

This dataset is of good quality and will be useful for future work. The paper sufficiently describes the setup and methods used such that not only is it useful in interpreting the dataset (except for the two areas identified that should be expanded on), but it is also useful in providing a guide on how to design these sorts of data collections. I hope the authors can address the two points I raised to improve the interpretation and use of their dataset for others.