Dear Chunyu Dong, we highly appreciate your time to review the paper and your positive report and suggestions to improve the manuscript. Find the responses to your comments with our planned actions below:

Regarding the major points:

1. We agree the grouping into regionalisation and evaluation of the model was simplified by using the SNOTEL dataset for regionalisation and the others for evaluation. The reason for this was to evaluate the model with fully independent datasets, rather than splitting the SNOTEL dataset in two groups. We thought this was the best option because we later apply the regionalised model to a fully independent snow depth dataset to generate the NH-SWE. Furthermore, we were not able to find data from more areas, considering the model evaluation requires continuous daily measurements of snow depth along with measurements (not necessarily continuous) of SWE or snow density. We will emphasize and further justify this in the revised manuscript.
2. We agree with the reviewer that Figure 3 gives that impression. We will expand the figure by adding other snow seasons from the same location, as well as some other locations so that the reader can see the effect of interannual variability on the results.
3. Indeed calibrating only ρ0 and ρmax is a limitation of our modelling approach. However the other parameters did not show any clear sensitivity to the climate variables explored. A comprehensive sensitivity analysis has already been completed by Winkler et al. 2021, which identified the importance of the two density parameters compared to the others, which is why we have focused on these. This reasoning is only briefly mentioned in the manuscript, so we will further justify it and discuss the limitations and uncertainties of this.

Regarding the minor point: We will increase the font size of tables and figures for the revised manuscript, and make two-column figures for those that look too small.

Dear Dr. Christoph Marty, thank you for taking the time to read the manuscript and for your positive comments and suggestions to improve it. Please find our responses with planned actions below:

Regarding the two main uncertainties:

• We acknowledge that SNOTEL is subject to errors as described in the literature. Avanzi et al. (2014) mainly describe issues with the hourly data, and Hill et al. (2019) also find biases and errors in daily data. We used daily data and applied the same quality control as described in Hill et al. (2019). The process described in our manuscript (L365) shows that the ∆SNOW model considers uncertainty in the snow depth measurements when the daily snow depth change is smaller than 2.4 cm. However, some measurement errors will be larger and not detected, so we will add a description of the uncertainty associated with this process in Section 6.2.
• We agree with the uncertainty associated with manual profiles that might not be taken at the exact same spot. We will discuss this and we will also specify which SWE data is from manual profiles in Section 2 (Data sources).

Regarding the specific comments:

Table 1: Thanks, we will correct.

We will add the SLF acknowledgement along with the Meteoswiss data.

L183 and L336: We will specify and add where appropriate that the original parameters were obtained for the European Alps only

L240/41: This is perhaps oddly phrased. We intended to help the reader interpreting Fig. 4, by hypothetically taking two of the many points in Fig 4b with a similar maximum snow depth (x-axis), and realising that the warmer climate (color-axis) has a higher snow density (y-axis). We will make it clear.

L291: We will correct it.

L293: We will add this.

Table 4: You are right, we forgot to specify how we estimated peak SWE for the datasets with biweekly measurements of SWE such as the GCOS-SWE dataset. We took the highest biweekly SWE measurement in the snow season and compared it with modelled SWE on that same date. This might be an unfair comparison for the model but the only way to compare the highest measured SWE value with modelled SWE for the datasets without daily measurements of SWE. This adds value to the evaluation and the biases found for those datasets are similar to the ones found in the datasets with daily SWE measurements. The timing of snowmelt onset was not estimated though, hence the gap in Table 4 for snowmelt onset for GCOS-CH. We will add this.

Figure 9: We will check this and explain it in the text. Both peak SWE and snow cover duration are higher than the surrounding stations for those two dots, so they are indeed probably high elevation stations.

L455: For stations that were not gap-filled, we only modelled SWE for those years with continuous daily measurements of snow depth. If there were none, the stations was not used for modelling SWE. We will add this.

L494: The previous paragraphs were on the gap-filling method, but we also apply a quality control of the gap-filling after that, as explained in the paragraph. We will make this clearer.

Dear referee, we appreciate you find the study and the dataset valuable, and we thank you for the comments to improve the paper. Find below our answers with our planned actions:  

1. Thank you for pointing out these relevant references that we had missed, we will appropriately introduce them in the text. Regarding the use of the Bias Correction and Quality Control SNOTEL data (BCQC SNOTEL), we have explored if this would potentially improve the calibration and regionalisation of the model. We agree that the BCQC SNOTEL dataset provides an improved quality with respect to the raw SNOTEL dataset. However, we had already applied a quality control to the SNOTEL dataset (same as in Hill et al. 2019, see Line 87-88 in our manuscript), and we only used SNOTEL stations and years with a continuous record of daily snow depth. This control has already highly reduced the frequency of poorer quality data. Nevertheless, we have downloaded the BCQC SNOTEL dataset and we have compared it with the raw daily SNOTEL data that we used. In the figure below left panel, the x-axis is the filtered SWE data used in our manuscript, and in the y-axis the SWE data from BCQC SNOTEL. Indeed, there are many points that differ a lot, but the scatter density in the 1:1 line is over 10 000, thus the overall variation is very minor and the difference between the two is smaller than 1 mm for over 99% of the points, as can be seen in the cdf plot in the right panel. We therefore believe that using BCQC SNOTEL would not change our results or our model regionalisation significantly given our quality control version is already remarkably similar. We will discuss in the manuscript the uncertainty associated with the few potentially biased or erroneous SNOTEL data.

1. (bis) The rationale for using the SNOTEL dataset only for regionalisation was to keep independent datasets for model regionalisation and model evaluation, since this is ultimately what we did to generate the NH-SWE dataset (i.e., we regionalised the model with SNOTEL, evaluated it with 8 independent datasets (Table 1), and applied it to our independent collection of Northern Hemisphere snow depth datasets (Table 1)). If we used all the HS-SWE data to regionalise the model, we would potentially lose confidence in the application to an independent dataset, as we would no longer have a technically independent evaluation. In addition, the SNOTEL dataset does contain a wide range of hydroclimates to allow generation of a regional parameter set that can be tested for broader applicability with the independent dataset. As stated by Sun et al. (2019): “The 246 SNOTEL sites represent diverse hydroclimate conditions in the western United States: the Pacific Northwest's relatively mild and wet winter; the Intermountain West's continental climate; the Rocky Mountains' cold, snowy winter; California's Mediterranean climate; and the Southwest's arid climate.” We also used the sites from high latitudes in Alaska. We will make this rationale clearer in the text.
2. We agree that the evaluation lacks spatial context, and we are happy to improve that. The tables and figures only provide mean values for different datasets, so we will add maps with the error of the assessed metrics and add a discussion interpreting the results, as suggested.
3. Sorry we realised the table caption says median while in reality we are providing the mean. We will correct it.