See attached pdf.

Review of „Enhancing the usability of weather radar data for the statistical analysis of extreme precipitation events”

The manuscript deals with the important and timely topic of determining design storms with return periods of up to 100 years from rather short time series of precipitation data from radar observations. The authors present a method to statistically extend time series of weather radar rainfall estimates by combining regional frequency analyses with subsequent bias correction. The results show improvement over the sampling approach by Goudenhoofdt et al. (2017) that is used as basis for their method, but uncertainties, e.g. a bias in the radar data for design storms with large return periods, still remain.

The study fits in the scope of HESS and is of interest to the research community. I have already reviewed an earlier version of the manuscript and the authors have taken some of my suggestions into account. However, some major concerns remain and new questions came up, that I listed below. I recommend major revisions of the manuscript before publishing it. I’d be happy to discuss my suggestions with the authors in the open discussion and clear up possible misunderstandings.

Major comments:

1. A major concern is the minimum distance of the radar cells that are considered to statistically extend the time series of the cell of interest. As far as I understand the cells have to be at least 4 km apart. The authors mention that the typical size of a convective cell in Germany is 40 km for hourly events according to Lengfeld et al. 2019 (p.5, l.150 in this manuscript). Therefore, the minimum distance of 4 km seems a bit too small to me, especially when considering also daily events that have a much larger typical spatial extent. The authors mention that more or less the same amount of events have been taken from all pixels, but did the authors perform any kind of independence check for the time series from the cells that are combined to a long time series, e.g. the correlation of the time series?
2. The aim of the study is to determine design storms with a return period of 100 years. Therefore, a method to extend rather short time series (19 years) from radar data by using additional data from similar regions is presented and compared to a station-based interpolate product. It makes sense to have a time series of more or less the length of the return period for the radar data. Therefore, a length of 95 years has been chosen which equals a combination of 5 pixels. But there is no information on the length of the station-based products that are used as references here. To my knowledge KOSTRA is contains 60 years of data. How reliable are the estimations of design storms with a return period of 100 years from KOSTRA? How many years of data are included in BW-stat? I was wondering how fair the comparison is when using data sets with different lengths. It would be beneficial to the manuscript if the authors add some information about this.
3. Although the authors extended Section 2.2 about the reference data sets some information are still missing (e.g. how many stations are considered, length of the time series, interpolation methods, etc.) A more detailed description of methods used in BW-stats and KOSTRA as well as the differences in the statistical approaches to determine design storms from those data sets is also desirable. The method for BW-Stat is briefly described in section 2.4. Maybe it would be better to have a general section about the methods first and then describe the data sets and their differences. Important information about the methods are missing that are crucial to understand the results and differences between the datasets.
4. A more detailed description of the sampling process, the generation of the ensemble members, the bootstrapping method and the bias correction is needed to allow for better understanding of the results and of the choices made by the authors.
5. Some findings are mentioned in the result section, but not sufficiently discussed in the discussion section. E.g. Why the spatial pattern in BW-stat is following the behaviour of RAD-BC for a return period of 1 year (p.10, l.295). Extending the discussion section and analysing the results in more depth is necessary to enhance the quality of the manuscript.

Minor comments:

p.1, l.6-8: This sentence sounds odd to me. Please rephrase.

p.1, l.17: A bracket seems to be missing here.

p.2, l.49: times series --> time series

p.2, l.62: …might not sufficient… --> …might not be sufficient…

p.4, l.107: What exactly are the methodological differences the authors mention here?

p.4, l.126: A word seems to be missing in this sentence.

p.5, l.127: Is the limitation to values between the 5^th and 95^th percentile really necessary? How large the outliers? Please justify this decision.

p.12, l.371: …the also the… --> ….also the…

p.12, l.372: approached --> approaches

General comments:

The authors provide a quite novel approach to estimate design rainfall from weather radar in Baden-Württemberg (BW). The main idea is the pooling of data from radar pixels in the proximity of the target cell to increase the sample size beyond the 19 years available record length. The radar data are bias corrected and compared against two station-based data sets, the German design storm standard KOSTRA and a regional data set from BW.

Although the approach is quite heuristic with several arbitrary assumptions and decisions (e.g., search radius, local estimation method, interpolation of parameters over durations, independence assumptions, etc.) it is practically pragmatic and statistically satisfactory. There are some major issues which need further attention and discussion. The first is the selection of events within a small search radius with the assumption of spatial independence. Second, the many arbitrary assumptions need to be better justified. Third, there is a need for better and more formal description of the methods.

Altogether the idea is good, the results are interesting and plausible. The text is significant and reads quite well although the English could be improved. I would recommend publication after major revision.

Specific comments:

1. Title: The title suggests a postprocessing of radar data for a later statistical analysis. However, you have done already the analyses. I would recommend to adjust the title e.g. something like “A pooling approach for design storm estimation using weather radar data – a case study for BW”
2. Lines 31: The non-stationarity is not considered in the approach. Of course, with only 19 years of observations this is hardly feasible. However, at least a brief discussion or an outlook should be included.
3. Line 89: “Reassembling via running sums” becomes not clear. Usually, the highest temporal resolution of 5 min is used to build the extreme values series for all durations by calculating sums over a moving window with width equal to the duration and moving step of 5 min? Is that the procedure used here?
4. Lines 146 ff: I think I could finally figure out how the sampling locations are selected but the description is weak. Please, reformulate and explain better. There are several arbitrary assumptions: why have you selected the normal distribution, how did you define its parameters, how did you select the 0.8 threshold, etc.? These need to be justified and discussed.  
5. Line 157: Include equation for normalisation.
6. Line 168: The “sub-sampling is not adapted for different event durations”; I guess you mean by that, that the same locations have been sampled independent of durations?
7. Line 189: What about spatial independence? Is this minimum separation time of 48 hours between events applied on the whole compiled data set from all 5 locations together? Only that way a spatial independence can be assumed. On the other hand, in that case considering the small search radius I would assume, that the sample from the five locations is not really comparable with a real 100-year sample; it probably will contain less extreme events and finally lead to an underestimation, which partly may explain the results.
8. Line 197 ff: In the independent fitting of distributions for different durations order relations problems may occur. This is accounted for in DWA (2012) by smoothing the parameters over the durations, which is a bit “old-fashion”. Please, explain more in detail which method has been applied here and discuss also alternatives.
9. Line 217 ff: Please provide equations for the quantile mapping approach.
10. Fig. 3: Are the probability distributions compiled from all stations/locations together? If yes how many stations are included?
11. Line 290: What is meant by spatial correlation analyses? Do you refer to correlations between rainfall and elevation? If yes, this is a cross-correlation but not a spatial correlation, which is usually used to quantify spatial persistence by correlation-distance relationships like the variogram, which by the why could have been employed for a more objective selection of the neighbourhood for sampling.
12. Fig. 5: Same question as for Fig. 3.
13. Line 315: Why are you using an 80% confidence interval here; usually a 90% interval between 5% and 95% quantiles is used?
14. Discussion/ conclusion: The new product has been compared against 2 reference data sets, but no strict validation has been carried out as usually desired. This is of course difficult since the truth is not known. However, often the long-term observations (>30 years) are applied as truth in a cross-validation mode. The application of this is also difficult here since the RADKLIM data set itself is a merged product involving these stations which makes this infeasible. At least a discussion of this problematic is required and optimal would be some suggestions for further research.