General comments

In this study, the authors have undertaken a considerable effort to refine precipitation analysis in mountainous regions, a challenging yet crucial area of research. A high-resolution ensemble precipitation analysis is developed by combining data from rain gauges, radar observations and a numerical weather prediction model (NWP). Initially, the authors address artifacts inherent in a combined gauge and radar-derived precipitation product (ANTILOPE), with the aim of mitigating issues like ground cluttering in a preprocessing phase, enhancing the accuracy of the precipitation fields in complex terrain. Subsequently, they employ three distinct approaches, two leveraging data from an ensemble NWP system, to generate an ensemble precipitation analysis using the preprocessed precipitation fields. In summary, the authors strive to improve precipitation products in mountainous areas, aiming also to correctly characterizing the uncertainty in the estimates. Through these developments, improved precipitation datasets can significantly enhance the accuracy of high-resolution snow simulations. Therefore, this work holds importance for various applications including water resource management, natural hazard mitigation, and climate impact assessments.

While the study demonstrates significant strengths, particularly in the quality of results and highly information illustrations of radar measurement issues, certain aspects warrant refinement before an eventual publication. In particular, the methods for characterizing the errors in the ANTILOPE dataset and the approach for correcting the spatial artifacts in that particular product requires a much clearer explanation. The authors tend to provide somewhat loosely coupled information about the methods in a few bullet points, albeit somewhat similar to a puzzle with a few missing pieces. To enhance clarity, the authors could employ hypothetical numerical examples illustrating key concepts. Additionally, the authors should also outline the assumption underlying their methods (especially those described in section 3.1.1 to 3.1.4). Finally, the authors may also want to discuss how undercatch by precipitation gauges will impact snow simulations using the precipitation datasets developed in this study, and discuss potential mitigation strategies.

Overall, the authors present a highly relevant study that forms an important contribution to the snow modelling community in particular and cryospheric sciences at large. The choices of methods appear sound, and the results are interesting and promising. To increase the value of the study further, please consider improving the description of the methods in particular, and also complement the results with more information. Please also see the detailed comments outlined below.

Specific comments

Line 2: What does “their” in “to capture their complexity” refer to?

Lines 6-9: The combination of the two sentences seems to suggest that the radar product is unbiased, in contrast to the precipitation fields from the biased NWP products. Is this correct?

Lines 13-14: Please rephrase “radar and gauges precipitation estimation products” to “precipitation estimates by radar and gauges” or similar.

Line 19: Does “ski resort” provide any relevant information here?

Line 24: Maybe change “human activities” to “practical applications”?

Lines 45-48: Please also mention that the precipitation measurements are affected by systematic undercatch in case of solid precipitation.

Lines 58-78: These two paragraphs lack a clear statement about the research gap of current studies, and what research is needed for filling this gap.

Lines 80-83: Perhaps move these lines to the end of the paragraph.

Lines 90-91: Consider moving the sentence “Evaluation data is only available for the period from 1 December 2021 to 30 April 2022” to section 2.4 describing the evaluation data.

Line 90-94: Consider adding information here stating that 24 h sums of all precipitation products up to 08:00 CET were analyzed in this study (if this is the case).

Section 2.4 and 2.5: The authors have chosen to first present grid-based products based on observations (section 2.1 to 2.3), followed by the point evaluation measurements (section 2.4), and finally the gridded data from the NWP (section 2.5). To me, it seems more logical to first present all gridded products (currently sections 2.1-2.3 and 2.5), followed by the evaluation data (currently section 2.4). Thus, consider swapping order of section 2.4 and 2.5.

Figure 2: What is the blueish horizontal bar in the left panel?

Figure 2: The text in the figure is very small. Please try to enlarge where possible.

Figure 2, caption: Consider changing the color of the line in Figure 1 showing the radar beam from Pic Blanc to Grandes Rousses massif from gray to a more prominent color.

Heading 2.2: Consider changing to “radar-rain gauge combination” which seems to be a more common terminology (Ochoa-Rodriguez et al., 2019).

Line 130: I wonder whether the usage of “1x1 km^2” is correct. Maybe “1 by 1 km” is better. Or even “…is available at a 1 km resolution…”. If adapting another convention, please change throughout the paper.

Line 139: Rephrase from “the errors’ magnitude” to “the magnitude of the errors”.

Section 2.4: Please provide the number of stations including their minimum, mean and maximum altitude.

Lines 159-162: Add a reference to a study about undercatch issues with precipitation gauges such as Rasmussen et al. (2012).

Line 160: Change from “under-catchment” to “undercatch”.

Lines 167-170: Why were precipitation sums calculated as a 24 h sum up to 06:00 UTC and not up to 08:00 CET as with the other products? What influence does this have on the analysis presented in the study?

Line 177: What region does “metropolitan France” refer to?

Line 188: Change from “half those” to “half of those”.

Line 188: Change from “in between” to “between the two peaks”.

Lines 188-190: Please rephrase this sentence since it currently does not read well. Maybe even spilt into two sentences.

Lines 205-206: Please use either “true” or “real” precipitation, and align this with the notation used for the variables. Likely, “true” is a better terminology than “real” precipitation.

Equation 1: Consider renaming the variable representing accumulated precipitation (RR) to an acronym that cannot be confused with the variable representing the ratios (R). The current usage of variable names is confusing.

Line 215: Change from “gauges locations” to “location of gauges”.

Line 222: Change from “far from pixel i of a distance d_ik” to “with a distance d_ik separating point i and k”.

Line 225: What does this ratio express? Please provide a meaning to the reader for easier understanding of the methods.

Section 3.1.1 to 3.1.4: The explanation of these methods needs to be improved. Potentially a hypothetical numerical example will help the reader to understand the methods better illustrating the core concept of the method using (a) one location including a gauge observation, (b) one location where the ANTILOPE is free of artifacts, and (c) a final location where ANTILOPE is affected by ground cluttering.

Lines 291-306: What do the authors mean with a spatial observation error and a dynamic error? Is the first constant in space while the second varies in space?

Lines 321-322: Reformulate. Likely the reference does suggest to reduce the number of assimilated observations in general, but not to reduce the number of particles to 16 and the number of observations to 1 in particular, as I understand the current text.

Line 353: What does “improved predictive added value” mean?

Line 355: Consider removing “local” from the sentence.

Line 361: What is the difference between systematic errors and biases?

Lines 367-368: Change from “the estimated and observed 24 h precipitation values time series” to “the estimated and observed time series of 24 h precipitation records” or similar.

Line 386: Repetition of “when only solid precipitation is considered”.

Line 387-390: Consider swapping order of these two sentences.

Lines 396-398: Where are these results?

Line 402: The first sentence on this line should probably be combined with the previous sentence.

Section 4: For the reader, it is likely easier if single sections presents results with a focus on individual figures. As an example, results from Figure 8 and 9 are presented in section 4.2.1 on lines 407-409, and again in section 4.2.2 on lines 411-417. In this example, the heading of the first section could be changed to “Impact of pre-processing procedure on spatial precipitation patterns” and the second section to “Skill of the precipitation estimates” (or similar) to allow for a focus of individual figures in single sections. Of course, cross references are allowed, but the current presentation of the results is confusing to me.

Figure 8: Please find a more descriptive label for the vertical axis.

Figure 9, caption: Rephrase to “Ratio between estimated and observed accumulated precipitation for the different methods assessed in this study” or similar.

Lines 420-421: Change from “(along the black line Hopson, 2014)” to “(along the one-to-one line) as described by Hopson et al. (2014)” or similar.

Line 424: What does “quite comparable” mean? Please avoid fuzzy terms.

Line 426: “To a lesser extent” than what?

Figure 10, caption: Please split the last sentence into two sentences to improve readability.

Line 448: Remove “estimation” in the part “precipitation estimation products”.

Line 449: Consider changing from “more competitive” to “provide better results”.

Line 465-467: Please provide the number of stations above 2000 m.a.s.l.

Line 473: Consider changing from “suffers from some” to “has”.

Line 483: Consider changing from “It will compare” to “In such a study, we will”.

Lines 475-493: In my opinion, the authors overuse bullet list making the text difficult to read. Please consider reducing the use of bullet points throughout the paper.

Line 504: “More spatially homogeneous errors” than what and why?

Line 519-520: Note that a precipitation product with shortcomings was assimilated and not direct observations. Please discuss the implications of this approach for the final results.

Line 531: What does “despite major shortcomings” refer to?

Line 535: What does “the advanced” mean in this context?

Line 537: Consider changing from “local 24 h precipitation” to “24 h precipitation sums” or similar.

Line 540: Is the ARMOE system “the most advanced high-resolution numerical weather prediction model” of all NWPs in the world?

Line 543: Consider changing from “The authors” to “In this study, we”.

Line 546: Reformulate. Change from “They also applied a correction algorithm” to “A correction algorithm was applied” or similar.

Line 549: “Local” seems to refer to station observations. Please clarify.

Line 550: Please give one or two examples of possible refinements in the conclusion.

Technical comments

Line 37: A misplaced comma.

Line 44: Rephrase to “precipitation inputs provided”.

Line 59-61: There seems to be an error with the usage of parenthesis here.

Line 80: Remove “preliminary”.

Line 148: Change from “Fig. 1” to “Figure 1”.

Line 379: Wrong units: “g m^-2”.

Line 380: “This Figure”. Misplace capital letter.

Line 383: Inconsistent usage of units: “kg/m2/24h”. Please also correct many of the figure titles.

Line 387: Likely refers to appendix B and not A.

Line 393: A white space is missing.

Line 395: Error in figure reference.

Line 401: Likely wrong reference to section 5.

Line 416: Change from “Figure 9” to “Figure 9b”.

Line 440: Consider changing from “Figure 11” to “Figure 11e and f”.

Line 460: Wrong reference.

Line 510: Wrong reference format.

Line 583: Remove “has” from the sentence.

References

Ochoa-Rodriguez, S., Wang, L. P., Willems, P., & Onof, C. (2019). A Review of Radar-Rain Gauge Data Merging Methods and Their Potential for Urban Hydrological Applications. Water Resources Research, 55(8), 6356-6391. https://doi.org/10.1029/2018wr023332 

Rasmussen, R., Baker, B., Kochendorfer, J., Meyers, T., Landolt, S., Fischer, A. P., Black, J., Thériault, J. M., Kucera, P., Gochis, D., Smith, C., Nitu, R., Hall, M., Ikeda, K., & Gutmann, E. (2012). HOW WELL ARE WE MEASURING SNOW? The NOAA/FAA/NCAR Winter Precipitation Test Bed. Bulletin of the American Meteorological Society, 93(6), 811-829. https://doi.org/10.1175/Bams-D-11-00052.1 

General comments and one general question

This article deals with the quantitative estimation of precipitation by radar and its uncertainties. This remains a major issue, given the difficulty of making such estimates, particularly in mountainous areas. In their article, the authors present both a method for 1) identifying, quantifying and correcting any shortcomings in the Antilope product - a product combining kriged rainfall data with radar data - to generate rainfall quantities estimates fields on an hourly time step and 2) producing ensemble precipitation analyses using a) only the rainfall estimates corrected in the previous step and using their associated uncertainties, b) with the additional use of NWP data through a Particle filter and c) idem but with an Ensemble Kalman filter.

The introduction recalls the importance of monitoring snow cover in mountainous areas and the state of the art in terms of modelling and observation. It explains the subject of the article: the generation of ensemble analyses of rainfall totals from different sources of information (network of rain gauges, radar data, NWP outputs) that make it possible to quantify the uncertainties in spatial rainfall estimates with a view to facilitating the assimilation of rainfall-related data into snow cover models. The region studied is the French Alps and the study period is one year.

The 2nd section describes the datasets used and/or evaluated:

1) the PANTHERE radar product, which corresponds to the operational radar rainfall totals estimates in France in real time. This is not indicated in the article (-> it could be added / mentioned in the text): this product already contains an initial adjustment of the radar estimates with rain gauge data using a correction factor (not spatialised) that considers the ratios between radar rainfall totals and rain gauge totals over the previous hours;

2) the combination product (I would use the term "combination" rather than "fusion") of ANTILOPE radar and rain gauge data;

3) rain gauge data spatialised by the kriging method;

4) rain gauge data (from ski resorts) used exclusively for evaluating the methods presented in section 3.2 and

5) rainfall quantity estimates from NWP.

Section 3 presents the error estimation methods for the Antilope product and the methods used to generate daily ensemble precipitation analyses.

Section 4 presents the results obtained from the evaluation of existing precipitation products and ensemble precipitation analyses.

Section 5 is a discussion on precipitation estimation products, observation errors and the choice of the data assimilation method.

The subject of the article is a response to the need to correct radar estimates which, despite a sophisticated processing chain, are still marred by shortcomings, particularly in mountain regions where the interaction of radar beams with the terrain produces ground clutter and beam blocking effects, and where the quantitative estimation of precipitation on the ground is also made more difficult by the high altitude of the radars, which consequently survey at very high altitudes. At these altitudes, the characteristics of the precipitation can be very different from those on the ground.

One point that could be discussed, albeit briefly, is the uncertainty of the estimates obtained from the network of rain gauges, whether from the operational network or from the ski resorts. Like radar data, rain gauge data can be subject to measurement error, particularly in strong winds and/or snowy conditions. So, the different scores presented for the various radar products and rainfall ensembles take as their reference values from ground-based rain gauges, which can sometimes be accompanied by significant uncertainty. Another remark that is a little outside the scope of this article is that in general the kriging with external drift used to merge rainfall and radar data provides both an estimate of the rainfall values and also an estimate of the associated uncertainty that could be used. Perhaps this information could have been considered for this study.

--> What do the authors think?

Specific comments and questions

Page 5: Figure 2 is a very good illustration of the problem encountered in mountainous areas. However, I'm not sure that the coloured band at 3500m is easy to understand and I wonder whether it could be misleading or misinterpreted. Is it an estimate of rainfall at a constant altitude of 3500 m or is it the cumulative total on the ground along the transect between the Moucherotte radar and the Pic Blanc? The chosen radar (Moucherotte in this case) is perhaps not the best one for this illustration, as there is a strong under-estimate near the Pic Blanc "upstream" / before the highest summits, whereas these strong under-estimates should rather be the result of masking by the relief "downstream", i.e. after the relief. In all likelihood, this area of underestimation is more likely to be the result of the Colombis radar being masked by the Ecrins massif (between La Meije and Colombis radar)? In fact, what's hard to highlight from this coloured band is that it's a combination of data from several radars (or rather, at each point it's data from a single elevation selected from a single radar). Morover, the indication "radar information not used" and "use if corrected radar information" only apply here if only the Moucherotte radar was used.

-> Perhaps it would be better to remove this coloured band to avoid misinterpretations or misunderstandings?

Page 5, Line 109: I propose to replace "measured" by "estimated"

Page 10, section 3.1.2: the ratios are estimated on an annual basis. ANTILOPE uses PANTHERE products, which combine data from different radars with elevation angles that can change over time depending on weather conditions and the availability of radars in real time.

-> Is this a problem and/or is this effect compensated/corrected in the dynamic stage of section 3.1.4?

Page 11, equation 8: Is it possible to give physical meanings to the two different terms in this equation?

Page 12, line 273: About the weight cik: Can it tend towards infinity if Uk tends towards 0?

Eq 10, Pk would be to be defined in the text

Page 13: What exactly does Cii correspond to? I'm having trouble convincing myself that this could be a special case of Cik defined on the previous page.

Line 379: I propose to put kg/m2 instead of g/m2

Line 399, title of 4.2: add the word "ensemble" in the title?

Line 255. Given the equation for U and the elements in Appendix 1, do you confirm that U have the unit kg/m²? If confirmed, I propose to mention it in the text and in the colorbar of fig. 5d

Page 9: Figure 3a: what do the shapes used (triangles, circles) correspond to?

Page 12: Figure 4a: How can the large differences between the background colour and the local rainfall ratio value be explained? Is this just an effect of the weighting carried out in 3.1.2?

Page 14: Figure 5 suggests that there is little difference between a) and c) and that, consequently, the correction provided by knowledge of b) is weak, whereas the structure of the image b) seems to show very high variability and a strong capacity to correct the raw image a).

-> Isn't this surprising? As a result, figure 5e) gives the impression that it is the WMA that contributes most to the correction of a)

-> Is this what we should understand?

Page 14, this figure 5 could be enlarged to make it easier to read the axes and colorbars of the individual figures.

Line 460: The symbol “?” might be replaced by a reference or deleted