This manuscript deals with the quantitative estimation of rainfall in the south-west Indian Ocean, more particularly on Réunion Island where the orographic enhancement of rainfall can be very significant. A polarisation diversity X-band research radar was installed and tested. The authors state that this is the first search radar of this type to be deployed in this region of the world. The authors chose as a case study the passage of cyclone Batsirai in February 2022. They propose to compare the rainfall estimates obtained by radar using data from 6 rain gauges available on the island as a reference. Two radar rain intensity estimators were tested: the first based on the radar reflectivity factor Z, the second on the specific differential phase Kdp. For the first estimator, which is very sensitive to attenuation by precipitation, two attenuation correction methods were applied and evaluated: the method of Hitschfeld and Bordan (1954) and the so-called phi-linear method of Bringi et al. (1990). The authors confirmed the results of the literature on the subject: the fact that the second correction method is more reliable than the first and that the estimate with Kdp is more efficient for intense precipitation than the Z estimator. A third method, directly deriving the rainfall intensity from Kdp is also used and compared.

This article does not provide any new knowledge per se, but reports on the first quantitative rainfall estimates using polarisation diversity search radar in the south-west Indian Ocean. Section 2 is interesting because it provides a detailed description of the differential phase shift pre-processing.

In my opinion, some points in the article need to be corrected or clarified, in particular certain equations.  It is a pity that the operational radar data available from the National Meteorological Centre is not used (or that its use is not discussed) in this study. In addition, the study only covers one case study, which does not provide a robust assessment. The case chosen corresponds to a very specific cyclone situation. This is both a strength (the type of event is poorly documented, particularly the interaction with the terrain) and a weakness (there are doubts about the quality of the reference intensities provided by the rain gauges under these conditions, which makes it more difficult to compare radar and rain gauge estimates). 

Questions and suggestions:

• Page 2, Line 43: Are the S-band radar data usable? Could they have been used?
• Page 2, Line 53: The reference to Krämer and Verworn is only a conference paper, and therefore not peer-reviewed. Does it really provide information that would not be available in peer-reviewed articles published in peer-reviewed journals?
• Page 2, Line 58: define the variable (total differential phase shift) and make a reference to section 2.2.2.1
• Page 2, Line 62: Isn't it too ambitious and sufficiently robust to reproduce this R-KDP law on a single case study?
• Page 5, line 100-105, table 2: It is preferable to specify how beamwidth is defined: 3 dB? 6 dB?
• Page 7, figure 2: first sub-figure "Crête", the value of the vertical distance between radar beam and rain gauge should be 350 m and not 659 m. 659 m" should be replaced by "350 m" to be consistent with Table 2. 
• Page 8, Lines 170, 172 and figure 3: make the information consistent. The text indicates phi_dp<0.85 whereas the figure indicates phi_dp >0.85. However, it is consistent for and SNRH >3 in both the text and the figure.
• Page 8, Line 180: separate "to" and "derive".
• Page 10, Ligne 231: « db » to be replaced by « dB »
• Page 10, Line 231: give the unit of Z here (is it dBZ or mm-/m3 ?)
• Page 10, line 230-232: The equation and the values of c and d are not mentioned in the proposed reference of Berne and Uijlenjoet (2006). Please cite a reference that use this expression and these values. Moreover, usually, these coefficients c and d correspond to a relationship between specific attenuation (dB/km) and reflectivity factor expressed in mm6/m3 (and not in dBZ). Please mention clearly the units of each variable in all the equations.
• Page 10, Line 235: I think relationship (3) is wrong. The prefactor c and the exponent d would be applied twice, which seems wrong to me.
• Page 10, Line 245: Equation (6) is exactly the same as equation (4), so is it necessary to duplicate it?
• Page 10, Line 258: "mie" -> "Mie".
• Page 14, Line 331-334: It is clear that only one event (cyclone Batsirai in early February 2022) is used to determine this R-Kdp relationship. This is very little to ensure a minimum of robustness. Moreover, it gives the impression that rain gauge data are used both to establish this relationship and to assess it. This raises the question of the independence between the rain gauge data used to establish the Kdp-R relationship and those used to evaluate the quantitative estimates made using this relationship: has the dataset been split in two to carry out these two operations?
• Page 15, figure 7: Put the units of the variables on the axes.
• Page 16, table 3: What is the time step used? “Hourly” is in the title of the paper and mentioned for fig. 9 but should be also indicated in the text of section 3 and in the title of table 3 if it is this same timestep that is used.
• Page 18, line 398: "the HB method exhibited a negative correlation coefficient, suggesting that radar rainfall estimates and rain gauge measurements were not related”. It is rather the fact that the correlation coefficient is close to 0 that indicates an absence of link.
• Page 19, figure 9: What do you mean by " box " in the legend? Could you indicate which quantiles represent the limits of the boxes and whiskers in the boxplots?
• Page 20, line 443: We can also imagine that there is a lot of wind, which must lead to a lot of errors in rain gauge measurements. Is the effect of the wind corrected? Do you have any idea about error might result from wind of the cyclone?

Review of “Rain gauges and X-band radar hourly comparison under complex orographic conditions in Reunion Island during the passage of the cyclone Batsirai

General thoughts and comments:

The paper provides a description of a quantitative precipitation estimation data processing chain for an X-band radar deployed on Reunion Island. The most novel aspect of the paper is the location, with this being the first deployment of an X-band radar in the South-West Indian Ocean (SWIO) region. The paper then compares QPE from the radar to 6 rain gauges for a 4 day event, with dual-polarisation based techniques producing QPEs which compare better with rain gauges than the older, H-B technique of attenuation correction.

One of the key features missing from the paper, is the inclusion of an example PPI or two showing the radar coverage of the cyclone passage. A main advantage of weather radars is the wide area observations they provide and it is difficult to understand how the X-band is benefitting Reunion without some representative figures.

Another limitation is the implementation of R(Kdp). The authors derive a relationship based on an empirical fit to the data in the case study, which is very limited in scope. A quick comparison to the R(Kdp) relations detailed for X-band radars in Ryzhkov and Zrnic (2019) suggests the coefficients derived fall outside the expected range, with an a of 8.062 comparing to values between 14 and 20 and a b of 0.49 comparing to a range between 0.73 and 0.85. The authors do not comment on these differences and whether they are a result of the derivation methodology, which we have to assume uses aggregated radar data (6-min frequency) to compare with hourly gauge totals potentially for a single range gauge site (based on a n value of 82 in Figure 7, which matches the n value for Grande-Coude in table 3) or due to the unique drop-size distributions observed during the cyclone. At a minimum the authors should expand their description of the empirical derivation to explain how the temporal resolution issues are accounted for and exactly which gauge data is used. The authors should also consider how R(Kdp) performs when using a representative, disdrometer derived, relation from the literature that is as close as possible to their case study (tropical cyclone) as that would be more applicable. If disdrometer observations are available from Reunion they could provide a third avenue to improving the use of Kdp in this study. The current implementation also raises the question of what would happen if R(Z) was derived in the same way. How would this differ from the 300, 1.5 relation applied already? The authors should justify why an empirical derivation is acceptable for one technique but not the other when it is clear both could be assessed in the same way.

The pre-processing of differential phase shift is interesting and thorough, yet it is not clear which aspects are novel in comparison to the cited paper by Padmanabhan (2024) which applies processing to the same radar data. The finding that the system differential phase is often azimuth and elevation dependent and requires careful assessment is useful to the wider radar community. The authors could emphasise this more. To further highlight this point they could also improve Figure 5. I’d suggest changing from line plots to either boxplots (for each azimuth) or a min-max interval with a median/mean + iq range/std deviation plotted too, this would make it clearer how much the offset varied in each sector. To make this more visible the authors could then select 3 elevations to plot which are representative of the whole volume, and include individual plots as supplementary information if they believe them to be relevant. Again an inclusion of a representative PPI showing the impact of this azimuthal variation would be beneficial.

The rain gauge analysis is limited in its applicability by the very small sample size, both in terms of the number of gauges and the length of time used. While it has some value in supporting the pre-existing scientific consensus it could benefit from either the inclusion of significantly more data from the X-band radars year long deployment on Reunion or by reframing the paper more as a case-study analysis of the cyclone itself and what the wider (in a spatial sense) dual-polarisation variables help explain about the cyclone and its interaction with the orography, which would seem to have great potential given the novel aspects of the deployment. Incorporating data from the S-band radars mentioned in the paper would also add to such a case study analysis.

Ryzhkov, A.V., Zrnic, D.S., 2019. Polarimetric Measurements of Precipitation, in: Ryzhkov, A.V., Zrnic, D.S. (Eds.), Radar Polarimetry for Weather Observations, Springer Atmospheric Sciences. Springer International Publishing, Cham, pp. 373–433. https://doi.org/10.1007/978-3-030-05093-1_10

Specific Comments:

L33: Is there a reference for the 12h, 24h and 72h precipitation records? Can you include the actual record values too.

L76. Including an indication of the total number of volume scans used would be useful.

Figure 1: Can you add a 75km radar range ring to show the data extent. Perhaps you could also add a specific radar PPI here given the blank space to show the coverage during the cyclone passage.

Table 2: Can you update to include actual values used during the data collections, for example the azimuth rotation speed, the PRF and the pulse width. These are much more useful for considering the data here than the range of possibilities available for any deployment.

Table 2: Almost all the data included here is also shown in Figure 2 - I’d suggest you don’t need both, and Figure 2 seems to be a better visualization for the reader.

L181: Is any disdrometer data available for Reunion? Could this be used here to derive specific R(Z) and R(Kdp) relations?

L189: What exactly is a maxdisplay plot? Have you manually assessed each of these to identify stratiform times or applied an algorithm. A little more detail is needed to understand your methodology.

L239: In this section the H-B implementation is described, and one of the main limitations of the method is its tendency to explode to unrealistic PIA values. In the literature many studies therefore cap the maximum PIA that can be applied. Did you do this here? If not, why not? Also how applicable are the c and d values used for this region, would you expect them to vary with the DSD?

L253: Here you make reference to many alpha values without citing what atmospheric conditions/DSDs they were derived from. Can a more focused approach be taken? One of the limitations of linear phi is that alpha can vary along the ray, especially in the presence of hail (so called hot-spots). Is that a concern here?

L267: What constitutes a “high gradient value”? Can you give specific thresholds that are used?

Figure 5: See my comments in the general section. I think this is a very valuable result but the presentation can be improved to avoid the horizontal lines you see in many of the plots. Consider a hexbin or a boxplot or something else that can depict the range of values at each azimuth more clearly. Also why do you restrict this to 3 hours in particular? Does it change if using a different 3 hour window?

Figure 6: You could add an intermediate ray trace, after the removal of system phase and unfolding to show the variability before and after smoothing.

Figure 7 and surrounding section: This requires a lot more detail as to how you compare 6 minute Kdp with hourly gauge totals, a description of exactly what data is going onto the plot and why. For example is this from all gauges or just one. Are there thresholds applied to rain-rate or kdp?

L418: I would consider a NB of 0.5 to be more than a “slight tendency” to overestimate as it indicates a 50% overestimation by the radar on average. More discussion of this and the reasons for it would be beneficial.

L422: I’d argue that the NB results don’t indicate that R(Kdp) is better at least for group 1, where all 3 gauges have a lower NB using R(Z) than R(Kdp). 

L467: Again only correlation indicates R(Kdp) is an improvement, with NB being inconclusive. The picture is much more complicated than R(Kdp) being a significant improvement over R(Z) and this is probably the result of several factors such as the variability of attenuation, changes in VPR and changes in the accuracy of Kdp estimation at different intensities of rainfall. It’s interesting here that you don’t consider the use of a hybrid approach as is often implemented where R(Kdp) is used for higher rain-rates where it benefits from reduced estimations noise, less DSD variability and immunity to attenuation but where R(Z) is used at lower rain-rates when attenuation is lower, Kdp is harder to estimate and the R(Kdp) relation has more DSD variability.

L486: “just under a year” This raises the question why the paper only focuses on one 4 day case here. Is there a complementary case study that could work alongside it, or just the potential to expand to include all the data to get a more representative set of statistics.