General Comments

Caro and co-authors present a new dataset of snow depth observations across the Southern Andes, compiled from several independent sources in Chile and Argentina. The observations are quality-controlled and homogenized through a systematic procedure. The authors demonstrate the scientific value of the dataset by analysing the relationship between snow depth observations and elevation in each of the three best-monitored river basins in the dataset. The article is within the scope of ESSD.

This study is a very welcome contribution, as snow depth data in the Andes are scarce and sometimes very difficult to find and process. The overview of data availability provided by the authors is useful for understanding the current limitations in spatial and temporal coverage. The new dataset will serve as a reference for hydro-meteorological studies in the region.

I am confident that this work can be accepted soon, but I would suggest a major revision at this stage so that the authors have the opportunity to clarify a few methodological issues and refine the presentation. Please see my comments below.

1.    Methods

The presentation of the methods is confusing in some parts, and I could not follow all the details, particularly the logic and explanations behind the new PCI index. Please see my specific comments below.

2.    Snow depth – elevation gradients

The conclusions about snow depth variations with elevation need to be cautious, as they are not necessarily generalizable from the available stations. Although the results presented here are based on this excellent database, the stations are still restricted to a few accessible locations, and many areas remain unmonitored. I refer especially to the discussion of altitudinal patterns of snow accumulation with breaks at specific elevations, for example in L49–50 and L534–542. 

3.    Text length

The text is lengthy in some sections. Please check for repetitions and consider shortening some paragraphs. I have left a few suggestions in this direction.

4.    Updating the database

And a general question: I didn’t understand whether this is a static database, or whether you are planning to have regular updates.

Specific comments and text suggestions

-    41: I think that the mean is a better indicator here. The median neglects the magnitude of extreme values as it is calculated using only the central values. This comment is valid in several parts of the paper. 

-    56-57: Although this applies only to mid-latitude regions. In fact, you show later that snow accumulation in the northern stations occurs mostly during the austral winter.

-    58: Can you add a reference for this? Is this meant in general or just for the Southern Andes?

-    64: Please specify briefly here the elevation limits of the Mediterranean Andes. Is this above a certain elevation? The rivers of central Chile exhibit an important pluvial component to lower elevations. Or maybe specify that you refer to mountain rivers.

-    84: Chile and Argentina?

-    102-103: percentile ranges of what? Do you mean that 50% of the stations are between 32 and 34 degrees? 

-    103:104: Can you provide examples o these extrapolations? Or do you mean in non-academic work?

-    104:105: The water supply sentence sounds like a repetition from the first paragraph.

-    107-108: Although Cornwell et al. (2016) did find evidence of a peak in SWE around 5000 m.

-    131: Check the paper structure. It reads strange that there is only 1.1. I suggest that Study area moves to number 2. 

-    132: commonly -> can be 

-    132-133: Is this classification used in those studies? I thought that Caro et al. (2021) use the Tropical-Dry-Wet classification from Lliboutry, don’t they? Maybe Sarricolea et al. (2017) is a better reference here? 

-    135: delete “northern”

-    147: “comparatively” to what?

-    149: can you briefly explain what does “persistent” mean here? Having snow for a minimum number of months during the year?

-    150-151: “Chilean ...” can be removed

-    151-154: Repetition. Can be removed

-    161: At what time resolution are the original data provided?

-    162-163: Maybe “three” river basins?

-    163: Delete “including”? These are all the stations used in the article

-    172: “initial station-level screening” Is this a visual screening based on expert judgment?

-    172: Is this initial screening a type of Step 0?

-    172: “Prior to the data cleaning” Does the “data cleaning” corresponds to steps 2.1 to 2.5?

-    173: Remove “criteria used to evaluate”

-    174-177: I think this would fit better at the end of this sub-section. It is a nice summary of the final dataset.

-    181-182: what is “sufficient temporal coverage”? Did you set a minimum time length?

-    184-186: Is this referring to steps 2.1-2.5? Please refer to the corresponding number of each step to be clear about what are you referring to.

-    188: “This step”, what number?

-    184-188: In what temporal resolution are these proceedings applied? 

-    194: “a median window length of five days was adopted” But you use different windows for each station, right? Or is there a relation with the macrozone? Please explain better the selection of these windows. How do you select the value for each station?

-    197: Why 5 cm?

-    200: Please introduce the “validated offset” earlier in the text

-    202: Is the Campbell sensor accuracy valid for all stations?

-    203: Please present the information about the hourly observations earlier in the text

-    191-207: I suggest the authors to re-structure this paragraph. I think that all the information is there, but the explanations are convoluted.

-    210: The use of the median is again not fully informative. See my comment 3 on Table 1.

-    210-211: What criteria did you use to select the window size?

-    208: Are you using daily or hourly data at this point?

-    216: “k was calibrated”, but what is the target in this calibration?

-    217: “the median value” across stations or basins?

-    223-224: How did you select these thresholds? Based on expert judgement?

-    228: “the median physical limits” do the limits depend on each station?

-    232: “cleaned SD”, is this the final clean dataset shown in Figure 2? Then the arrow should come from Step 2 and not Step 4?

-    238: what type of external sources?

-    239-240: is the dataset going to be regularly updated? See one of my main comments

-    245: observations at the same site, right? Maybe specify more explicitly

-    251: “We developed and implemented a new index of physical consistency” Just to be more explicit

-    252: Here and later, there are several “?” symbols in the equations

-    259-260: I don’t understand this part. Why is the temperature threshold defined as the 95th percentile? Wouldn’t be better to use an absolute number between 0 and 2 C? Don’t you obtain values of T well above zero C using the 95th percentile? Or even below zero?

-    263-264: So the only difference between the numerator and denominator is the condition of T<Train? Is that the purpose of the PCI index? To detect unplausible events with Delta-S>0 and Pr>1 but with T>Train? Wouldn’t be more logical to use all the SD changes in the denominator (and not only those with Pr>`mm)? Then, the raw, noisy datasets will have low PCIs.

-    275-276: “The procedure effectively identified and removed…” To what procedure are you referring to? Step 2?

-    282: “we selected precipitation observations” for what? For the classification of years?

-    288-289: “In contrast, mean air temperature…” This sounds trivial, I suggest removing

-    290: “competing effects” why are they exactly competing effects? Both positive precipitation gradients and negative temperature gradients affect in the same direction, don’t they?

-    296: Add the reference of the SRTM dataset

-    297-304: What is the purpose of this procedure using the snow persistence?

-    302: Maybe remove the sentence about SRTM. I would say it is logical that you used the same DEM.

-    303: “elevation contour”

-    306: As you are analysing the distribution of stations, and not actual observations of snow depth, I suggest changing the title to : “Spatial distribution of snow depth stations” or similar. 

-    337: delete “effective”

-    354: “Key places” why are these places key? Maybe change to “selected”

-    366: “across the Andes” or just the Mediterranean Andes?

-    380: availability -> quality?

-    380-434: This section is a bit long providing many details about each corrected station. Can the text be summarized? Specially because the authors comment on several details that can’t be seen in any figure.

-    388: Same problem for the median here. Better use the mean?

-    425-426: Here (and in other parts of this subsection) the use of the median can also be misleading. The median of the reductions could have changed little but the median will not capture a station where the reduction was very high.

-    442-444: Please reword these lines

-    450: why the “spread of values”?

-    457-459: Isn’t that a consequence of using Pr>1mm in the calculation of N, in the PCI formula? See my previous comment on 263-264.

-    477-478: “but from the effective correction data issues…” in what step is that? Please include it in the next

-    483-484: I would delete the first part, start with “We selected…”

-    491: Here and throughout this sub-section, what is this median SD? The median across what? years?

-    518-522: I agree about wind erosion and limited snow deposition, and sublimation. Also, the work of Viale and Nuñez (2011) shows that precipitation has a maximum below the crest of the Andes.   

-    531: “In addition, a more robust…” or something similar to reduce the length of the text.

-    537-539: Keep in mind that the analysis is made only for a few stations. The text is written as if the pattern was general. See one of my main comments.

-    554: “that do not provide observations during normal precipitation years” This sounds as the stations would never capture normal years, but I guess the authors refer only to the analysed years.

-    583: “Of the 118…” -> “Of an original total of 118 …” or similar

-    589: “significantly reduced” can you provide the percentage here?

-    596: why is it relevant that these basins had a single station?

-    601: Begin with “Only three river basins…” if you would like to reduce text.

-    601-607: please add some caution to this paragraph stating that the elevation patterns are inferred from a (still) limited set of stations that do not represent the entire catchment. This is valid not only for Maule-Itata but also for Elqui and Maipo.

Figures

-    Fig 1: Mention the 3 highlighted basins in the caption. Why are they highlighted?

-    Fig 1: Can you use softer colours for the base maps? Maybe using transparency. As it is, it is difficult to see all the details of the stations.

-    Fig 2: I think it would be helpful for the reader to show here the 5 sub-steps of Step 2.

-    Fig 2: Specify in the figure that five logos above correspond to the data providers

-    Fig 3: “Colors indicate” what colours?

-    Fig 4: Can you use softer colours for the base maps? Maybe using transparency. As it is, it is difficult to see all the details of the stations.

Tables

-    Table 1: Basin area from Table 2 fits better here

-    Table 1: Given the small number of stations per basin, I think that the mean is a better metric that the median here. If you have, for example, 3 stations in the catchment with elevations E1, E2 and E3, the median elevation will always be E2, and the information about the elevation of the other two stations is lost. Same for latitude.

-    Table 2: I suggest changing to the mean here too

References

Sarricolea, P., Herrera-Ossandon, M., and Meseguer-Ruiz, Ó.: Climatic regionalisation of continental Chile, J. Maps, 13, 66–73, https://doi.org/10.1080/17445647.2016.1259592, 2017.

Viale, M. and Nuñez, M. N.: Climatology of winter orographic precipitation over the subtropical central Andes and associated synoptic and regional characteristics, J. Hydrometeorol., 12, 481–507, https://doi.org/10.1175/2010JHM1284.1, 2011.

Caro, A., Condom, T., and Rabatel, A.: Climatic and Morphometric Explanatory Variables of Glacier Changes in the Andes (8–55°S): New Insights From Machine Learning Approaches, Front. Earth Sci., 9, 1–21, https://doi.org/10.3389/feart.2021.713011, 2021.

General comments

Caro and co-authors present a valuable daily snow depth dataset for the Southern Andes of Chile and Argentina, compiled from several institutional networks and processed through a quality-control and harmonization procedure. The manuscript addresses a relevant observational gap in a region where in situ snow observations are scarce, heterogeneous, and not always straightforward to access or reuse. The dataset has clear potential for hydrological modeling, seasonal streamflow forecasting, satellite and reanalysis evaluation, and broader cryosphere studies in the Andes.

Overall, I consider this contribution well suited to ESSD. The manuscript is timely and useful, and the effort to integrate data from Chile and Argentina is particularly welcome. I also appreciate the authors’ attempt to go beyond a simple data compilation by documenting a quality-control workflow and by presenting an initial application related to snow depth–elevation patterns.

At the same time, I think the manuscript would benefit from a major revision before publication. My main concerns are not about the relevance of the dataset, but rather about the level of documentation needed for a data publication. In particular, I encourage the authors to improve the consistency between the manuscript and the dataset metadata, provide a more transparent description of the quality-control procedure, clarify the role of manual filtering, strengthen the discussion of uncertainty, and present the snow depth–elevation analysis more cautiously. These improvements would make the dataset easier to evaluate, reproduce, and reuse by future users. I see the dataset as an important contribution, but several aspects should be clarified to fully meet the expectations of an ESSD data description article. Please see my comments below.

1) The dataset has strong potential for reuse, but the manuscript should provide a clearer description of the files available in the repository. I recommend adding a concise file manifest, either in the main text or as supplementary material. This could include file names, variables, units, station identifiers, temporal coverage, time zone, daily aggregation method, missing-value codes, quality-control flags, and metadata fields. This additional documentation would help users understand and apply the dataset without needing to infer details from the manuscript or contact the authors. It would also strengthen the paper as a data publication rather than only as a description of the processing workflow.

2) The quality-control procedure is one of the central contributions of the manuscript. The description is generally informative, but some parts remain difficult to reproduce from the information currently provided. The authors mention a Python-based algorithm and provide a code example, but it is not fully clear whether the complete workflow, parameters, and processing sequence are available. I encourage the authors to provide the full quality-control code, or at least a more complete reproducibility package, including the parameter values used by basin or station, the processing order, and the expected input and output formats. If some raw data cannot be redistributed because of institutional restrictions, the authors could still provide the code, parameter tables, and example files. This would make the workflow more transparent and would increase confidence in the dataset.

3) The manuscript indicates that a supplementary manual filtering step was applied based on expert judgment. This is understandable for heterogeneous snow depth records, especially when working with multiple institutions and sensor types. However, the current description does not provide enough detail for users to assess how much of the final dataset depends on automated filtering versus manual decisions. I suggest including a station-level summary of manual interventions. This could indicate the number or percentage of values removed manually, the affected periods, and the main reason for removal, such as sensor drift, unrealistic spikes, summer artifacts, maintenance periods, baseline shifts, or other issues. If possible, the published dataset should retain quality-control flags indicating why a value was removed or modified. This would make the dataset more transparent and allow users to apply alternative criteria if needed.

4) The reduction in data availability after quality control is an important result and deserves a slightly more detailed discussion. In some basins, especially in the Wet Andes, the percentage of removed data appears relatively high. This may be justified, but the manuscript should provide more evidence that these removed observations correspond to unreliable data rather than to physically plausible but irregular snow conditions. This is particularly relevant in maritime or rain-snow transition environments, where shallow snowpacks, short snow accumulation episodes, rain-on-snow events, and rapid melt can occur. I suggest adding before-and-after examples for a few representative stations, especially those with large reductions in data availability. These examples would help readers understand the effect of the quality-control procedure and would support the interpretation that the final dataset is more reliable.

5) The Physical Consistency Index (PCI) is a useful idea and provides an interesting way to evaluate whether snow accumulation events are compatible with precipitation and air temperature. However, the current explanation could be clearer. In particular, the index is applied only to selected stations in the Elqui and Maipo basins. I suggest explaining more explicitly why these basins were selected and why the same validation could not be extended to other regions, especially the Wet Andes, where the quality-control procedure seems to remove a relatively large amount of data. It would also be helpful to discuss the limitations of the PCI. As currently defined, the index focuses on positive snow depth changes during days with measurable precipitation. This means that some potentially problematic events, such as positive snow depth changes without recorded precipitation, are not directly included in the denominator. This may be reasonable because of gauge undercatch, wind redistribution, and local variability, but the implications should be clearly stated.

6) The temperature threshold used in the PCI is defined as a station-specific 95th percentile. I understand the motivation for using a site-specific threshold, but the physical interpretation of this choice is not completely clear. I recommend that the authors report the actual threshold values obtained for each station and discuss whether they are consistent with plausible rain-snow transition temperatures. A simple sensitivity test would also be useful. For example, the authors could compare the PCI results using the current percentile-based threshold and one or more fixed thresholds such as 0 °C, 1 °C, or 2 °C. This would help demonstrate whether the conclusions are robust to the choice of temperature criterion.

7) The additional metrics used to complement the PCI are useful, but their interpretation could be improved. The Noise Ratio is described mainly as an indicator of the removal of spurious positive snow depth events when values are greater than 1. However, at least one case appears to show a value below 1, which is interpreted as signal recovery. I suggest clarifying the expected interpretation of NR > 1, NR = 1, and NR < 1. If the metric can indicate both removal of noisy events and recovery of plausible accumulation events after baseline correction, this should be stated more explicitly. A small conceptual example could help readers understand the metric.

8) The manuscript refers to the ±1 cm accuracy of the Campbell SR50A sensor when justifying some thresholds. However, the dataset appears to include observations from different sensor types and possibly different sensor generations or replacements. I recommend clarifying whether all stations used sensors with comparable accuracy. If not, the authors should discuss how sensor-specific uncertainty may affect the quality-control thresholds. This point does not necessarily require a complete redesign of the method, but it would be useful to acknowledge whether common thresholds were applied across different sensors and what implications this may have for shallow snow or near-zero snow depth conditions.

9) The analysis of snow depth–elevation patterns is a valuable illustration of the dataset’s potential. However, I suggest presenting the results more cautiously. The available stations are not uniformly distributed across elevation, latitude, exposure, or accessibility. Therefore, the observed elevation patterns should be described as patterns represented by the available station network rather than as definitive basin-scale gradients. This is particularly important where the manuscript identifies snow depth maxima at specific elevations. These findings are interesting, but they may depend on station location, selected representative years, local topography, wind redistribution, and the uneven distribution of observations. I recommend softening the wording in the abstract, discussion, and conclusions to avoid overgeneralization.

10) The workflow includes a fourth step described as cross-variable validation. It would be helpful to clarify whether this step modifies the final dataset or whether it only evaluates the quality-controlled product. If Step 4 is used only for validation, I suggest presenting it as an independent evaluation step rather than as part of the cleaning procedure itself. This clarification would also improve Figure 2. The figure could show more clearly which steps produce the final clean dataset and which steps are used to evaluate its physical consistency.

Specific comments

- Abstract: Please check the consistency of station numbers by Andean zone and ensure that these values match Table 1 and the repository metadata.

- Abstract: The improvement in PCI “from 87% to 95% at some stations” is useful, but it would be clearer to specify the stations or basins involved, or to present the result in more general terms.

- Abstract: Please clarify how the 23% reduction in data availability was calculated. Is this based on the median across stations, across basins, or across all daily observations?

- Lines 160–170: Please specify the original temporal resolution of each data source and explain how daily values were calculated.

- Lines 172–183: The criteria used to exclude stations should be described more explicitly. For example, was there a minimum number of months or years required?

- Lines 184–188: Please clarify whether the thresholds were selected by basin, by station, or through a combination of both.

- Lines 191–207: The zero-level correction is important but somewhat difficult to follow. I suggest restructuring this paragraph and adding a simple example.

- Lines 202–207: Please clarify whether the Campbell SR50A accuracy applies to all stations or only to part of the network.

- Lines 216–224: Please explain how the scaling factor k and the absolute spike thresholds were selected.

- Lines 225–231: Please clarify whether near-zero values are set to zero or removed, and whether this operation is flagged in the dataset.

- Lines 245–266: The PCI method would benefit from a clearer explanation of the denominator and of the events that are excluded from the calculation.

- Lines 259–260: Please report the actual ATrain values and discuss their physical plausibility.

- Lines 267–276: Please clarify the interpretation of NR values lower than 1.

- Lines 278–286: Please explain why the selected precipitation stations are representative for classifying dry, normal, and wet years.

- Lines 294–296: Please justify the 70% monthly data availability threshold or briefly discuss its sensitivity.

- Lines 297–304: The snow persistence and snowline analysis should be introduced more clearly before the technical description.

- Lines 385–431: Some detailed station-level explanations could be moved to the appendix to improve readability of the main text.

- Lines 567–579: The data and code availability section should be expanded to describe the complete processing workflow and repository structure.

- Figure 1: Consider using a lighter basemap or improving symbol contrast, as some stations are difficult to distinguish.

- Figure 2: Please show the substeps of the quality-control procedure and clarify whether Step 4 modifies the dataset or only validates it.

- Figure 3: The figure contains useful information but is dense. A higher-resolution version or simplified version could improve readability. Or mention if the web version will be available in higher resolution in case any reader requires it.

- Figure 6: Please clarify whether D, N, and W refer only to the selected representative years or to broader precipitation-year categories.

- Table 2: Consider including ranges or totals in addition to medians, especially for basins with few stations.

- Code availability: A short example script is helpful, but the manuscript would be stronger if the complete processing chain were available.