Have you ever seen the rain? Observing a record convective rainfall with national and local monitoring networks and opportunistic sensors

Petersson Wårdh, Louise; Hosseini, Hasan; van de Beek, Remco; Andersson, Jafet C. M.; Hashemi, Hossein; Olsson, Jonas

doi:10.5194/egusphere-2025-2820

Preprints

https://doi.org/10.5194/egusphere-2025-2820

Preprints

30 Jun 2025

| 30 Jun 2025

Have you ever seen the rain? Observing a record convective rainfall with national and local monitoring networks and opportunistic sensors

Louise Petersson Wårdh, Hasan Hosseini, Remco van de Beek, Jafet C. M. Andersson, Hossein Hashemi, and Jonas Olsson

Abstract. Short-duration extreme rainfall can cause severe impacts in built environments and flood mitigation measures require high-resolution rainfall data to be effective. It is a particular challenge to observe convective storms which are expected to intensify with climate change. However, rainfall monitoring networks operated by national meteorological and hydrological services generally have limited ability to observe rainfall at sub-hourly and sub-kilometre scale. This paper investigates the capability of second- and third-party rainfall sensors to observe a highly localized convective storm that hit southwestern Sweden in August 2022. Specifically, we compared the observations from professional weather stations, C-band radar, X-band radar, Commercial Microwave Links and Personal Weather Stations to get a full impression of the sensors’ strengths and weaknesses in the context of convective storms. The results suggest that second- and third-party networks can contribute with important information on short-duration extreme rainfall to national weather services. The second-party network assisted in quantifying the magnitude and spatial variability of the event with high precision. The third-party network could contribute to the understanding of the duration and spatial distribution of the storm, but underestimated the magnitude compared with the reference sensors.

Received: 16 Jun 2025 – Discussion started: 30 Jun 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 1880 KB)

Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
Preprint (1880 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Journal article(s) based on this preprint

21 Jan 2026

Have you ever seen the rain? Observing a record convective rainfall with national and local monitoring networks and opportunistic sensors

Louise Petersson Wårdh, Hasan Hosseini, Remco van de Beek, Jafet C. M. Andersson, Hossein Hashemi, and Jonas Olsson

Atmos. Meas. Tech., 19, 461–483, https://doi.org/10.5194/amt-19-461-2026,https://doi.org/10.5194/amt-19-461-2026, 2026

Short summary

Louise Petersson Wårdh, Hasan Hosseini, Remco van de Beek, Jafet C. M. Andersson, Hossein Hashemi, and Jonas Olsson

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-2820', Anonymous Referee #1, 22 Jul 2025

This manuscript aims to assess the capacity for second and third-party sensors to observe short duration extreme rainfall via comparisons to the national rainfall monitoring network in Sweden. While this study found that even the national rainfall monitoring network was unable to fully capture the results of a strong convective storm, it was shown that second and third party sensors can provide accurate representations of that data. This is an important addition to accurate weather monitoring as these additional sensors can assist with creating detailed representations of significant storms within a given area. The authors highlight several limitations to the integration of second and third party sensors into conventional weather monitoring systems and encourage the need for more research in this area. I support the publication of this work following minor edits described below. This work highlights crucial elements for establishing robust weather monitoring protocols and should provide the authors and other researchers will a strong starting point for future research and improvements in this area.
There are, however, a few areas where this manuscript could be improved. Most of the comments before focus on clarity of terminology and how the discussion relates back to the results presented.
Line 161: I was unsure what area this portion of the text was referring to in Figure 1. Is this just a composite of SMHI stations? Clarity could be improved here so readers understand the monitoring locations.
Line 194: Can you define overshooting in the text? What does it mean in this context?
Line 227-229: This sentence would benefit from more clarity and further explanation.
Line 272 and 274: I appreciate the clear explanation of all equations and understand that several of the parameters were obtained from other sources. Could you briefly elaborate what the G and offset variables represent for Equation 6 as well as a and b for Equation 7?
Line 392: Is it typical to see a trend with higher accumulated depth at inland regions like Baramossa? A reference here may be appropriate to establish that your results are typical and expected.
Figure 4: It’s difficult to tell the difference between SMHI Hov and SMHI Laholm D. The line types are very similar. This figure would be more clear with a different line choice for one of those data sets.
Line 494: Could you elaborate in the text on why the plateau period was considering unsuitable for comparison?
Figure 10: Is each panel a-e correspond to one minute? For example: panel a is 17:22-17:23 and panel b is 17:23-17:24. If so, this is not clear based on the information provided in the figure caption.
Figure 11a: Are you expecting a linear trend with this comparison? If so, reporting an R² value would be useful.
Figure 12: The figure caption, please specify that the r_s is relating the data to XWR data.
The discussion overall is thorough and touches on many good points. However, is there a reason the results and discussion sections are separate? At times this made assessing the claims made in the discussion challenging due to needing to locate the appropriate data in the Results section. If you choose to leave the Discussion as a separate section, please refer the reader to relevant tables and figures when emphasizing trends observed in the data. I have noted a few specific lines below, but I encourage you to go back through this section for any crucial information that would benefit from being redirected back to a table or figure.
Lines 589-591, Lines 598-599, Line 604, Line 612-616, Lines 628-630, Lines 644-645, Lines 660-662.

Citation: https://doi.org/10.5194/egusphere-2025-2820-RC1
- AC1: 'Reply on RC1', Louise Petersson Wårdh, 21 Aug 2025
  
  We thank the reviewer for the helpful comments that will support clarifications and further enhancement of the manuscript. Please find answers to the comments below.
  Line 161: Thank you for pointing this out. The text is referring to the C-band radar located 6 km South of the study area, marked with a white dot in Map 1 in Figure 1. The radar location is unfortunately missing from Map 2 in Figure 1 and will be added.
  
  Line 194: Overshooting in this context means that the radar beam shoots above the precipitation cloud and hence does not record it. It is a common source of error in radar data. We will add some standard literature as reference for further reading on the topic. The purpose of this section is to see if there is risk of overshooting by the X-band radar in the area of interest. The same analysis is done in lines 168 – 171 for the C-band radar. As the radar beam we are using travels on 200-300 m height at the area of interest for CWR and 300-1200 m for XWR, and convective precipitation in the summer months in Sweden typically originate from much higher altitudes, the risk of overshooting is very small. We will add sources to support these numbers.
  
  Line 227-229: The intention of this sentence is to add clarity, but if it only confuses things, it is maybe better removed. For sub-hourly rainfall data, the correlation between time series recorded by nearby sensors can be low even if they record similar total rainfall depths (for example Spearman correlation 0.18 between XWR and the municipal gauge, see Results). As convective rainfall is highly variable in space and time the observations per time step can be very different at nearby locations. If you use correlation as metric this will indicate poor performance, when, in fact, the sensors simply may experience different rainfall intensities even if they are closely located. By shifting the time series in time, you can account for the fact that a certain rainfall intensity may be observed by the radar before it reaches a rain gauge on the ground, for example.
  
  Line 272 and 274:
  Equation 6: Reflectivity data from radars are stored and distributed as integers between 0 and 255 to enable smaller storage size, following European standards (Michelson et al., 2014). To convert these integers back to reflectivity (dBZ) you apply the coefficients G (gain) and offset.
  Equation 7: a and b are well-established empirical constants that Marshall and Palmer (1948) found when establishing the relationship between size distribution of raindrops (DSD) in a radar pulse volume to the rainfall rate.
  
  Line 392: This is not a very general feature for Sweden, but the statement referred to the specific study area. The pattern can be seen in maps of 30-year mean annual precipitation here: https://www.smhi.se/klimat/klimatet-da-och-nu/normalkartor/normal/arsnederbord-normal. We will clarify this part of the text in the revised manuscript.
  
  Figure 4: Thank you for pointing this out. The line types will be better differentiated in the next version.
  
  Line 494: Given the sudden constant records of rainfall rate observed in the CML time series, which is unexpected and clearly not representative for the actual rainfall rate, it was excluded from the analysis.
  
  Figure 10: Thank you for pointing this out. It is correct that, for example, panel a is 17:22-17:23 and panel b is 17:23-17:24. This will be clarified in the next version.
  
  Figure 11a: Thank you for pointing this out. A linear trend is expected because, ideally, the two sensors (CML and XWR at the CML location) should record the same or very similar rainfall intensities. R² will be reported in the next version.
  
  Figure 12: Thank you for this valid and helpful comment, it will be added in the next version.
  
  About the Discussion section: We were indeed discussing whether to keep the results and discussion sections together or apart before arriving at this solution. It is understood that the discussion, if kept separate, will need more cross-references to the text. We will await the second review to decide which approach to choose.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2820-AC1
- AC3: 'Reply on RC1', Louise Petersson Wårdh, 24 Oct 2025
  
  After receiving comments from reviewer #2, we decided to keep the results and discussion section separate as is but improve the cross-referencing to the results section.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2820-AC3
RC2:
'Comment on egusphere-2025-2820', Hidde Leijnse, 26 Sep 2025
General comments
This paper describes a thorough analysis of in intense convective storm that occurred in southern Sweden. Data from the national precipitation monitoring network (gauges and C-band radar) are compared to second-party (X-band radar and a municipal gauge) and third-party (a commercial microwave link and rain gauges on personal weather stations) for this storm. The analyses presented give insight into the performance of the different sensors, and the discussion section provides clear conclusions about the strong and weak aspects of the different sensors. I think the paper is interesting and relevant for Atmospheric Measurement Techniques. I do think that the paper needs some work in improving the readability, providing some additional information that is important for interpreting the results, and checking some of the data and figures that are presented.
The most important aspect that I think requires attention is that there seems to be an inconsistency between the X-band radar data presented in Figures 7 ,8, and 9 (see the specific point about this below). It is also not clear how the time shift that was apparent in the municipal gauge was taken into account (I'm assuming here that the other sensors have more accurate clocks), and I'm missing something in the discussion about the importance of accurate time recording of the data in order for it to be useful. And some details provided in the paper about data processing are not relevant and could be removed, while other aspects of the datasets and their processing are missing. Details are provided in specific comments provided below (I realize that there are quite many comments; please take these as an encouragement to further improve this interesting paper).

Specific comments
line 87: CMLs measure path-integrated attenuation, not rainfall. I guess that is what you're trying to say with this sentence: that the assumption is that this is directly related to the path-averaged rainfall intensity.

lines 88-89: this assumption has been investigated by Berne et al. (2007), Leijnse et al. (2008, 2010), and De Vos et al. (2018) in high-resolution simulation frameworks. It would be interesting to see how your results for this case compare to what is presented in this literature.

line 94: consider using “people” instead of “occupants”.

line 107-108: you state that “cross-referencing radar observations with path-integrated rainfall estimated by CML” is a gap in the field of high-resolution rainfall monitoring. A large portion of the work cross-references radar and CML rainfall estimates, so I don't fully understand this statement. Could you make the statement more specific about which gap exactly is bein addressed here?

lines 109-114: I think that the research questions as they are formulated here cannot be answered by the analysis of a single convective event, because it probably depends greatly on the path of the storm and the local topology of the sensors in that path. Could the research questions be reformulated to take this into account?

Figure 1: it would be instructive to have different symbols on the map for the SMHI automatic gauge and the SMHI manual gauges.

Figure 1: could the location of the C-band radar be put on map 2 as well, so that it is clear where the radar is relative to the study area?

Figure 1: would it be possible to add the predominant wind direction to the figure? This could help in interpreting differences between the municipal gauge and the SMHI gauges.

line 155: it is mentioned here that “gauge-adjusted Plan Position Indicator” is used. However, in the subsequent paragraphs there is no information about how the gauge adjustment was carried out. Because this is very relevant for interpreting the results, please add information about this in the paper. Furthermore, I get the impression here that a precipitation product is used in this paper, but from Section 4.4.1 it seems that it is actually a much more raw product. Please clarify this in the paper. And if this is indeed a more raw reflectivity product, then I would suggest to use the radar data on its native (spherical) grid, i.e., not the 2x2 km composite because this would give much more spatial detail.

line 155: consider adding “reflectivity” (or even “horizontal reflectivity”) between “radar” and “composite” to make clear what kind of PPI is used.

lines 157-160: here you explain Z in terms of the DSD (you could also have chosen to present it in the form of the radar equation). To me it seems more relevant to present the resulting Z-R relation instead, preferably the one that was used to retrieve rainfall intensity for this study.

lines 161-162: the degree to which a radar contributes to a given point in a composite depends on the employed compositing method. So consider adding “and the compositing method is based on the closest radar” (or something similar, depending on the compositing method that is used) after “during the selected event”.

line 163: here it is mentioned that the resolution of the composite is 2x2 km. Given the very close range and the expected high spatial variability of the precipitation, using the radar data on its native (spherical) grid would give much more spatial detail of the storm. Can you indicate the reason for using the composite in the paper?

lines 171-173: here you discuss partial beam blockage. Could you indicate the severity of the beam blockage? And could the location of the vegetation causing this blockage be indicated on map 2 of Fig. 1? This is relevant for the interpretation of the results.

line 190: dual-pol attenuation correction is mentioned here, and the reader is referred to the literature for details on the method. I think it would be relevant here to add a few words (or a sentence) about which method is used (referring to the literature for details of the method is fine).

line 191 and Figure 3: “half-beam” and “half beamwidth” are mentioned here. What is meant by this? Do you mean half-power beam width? If so, please modify this. And if this is the case, then consider to use the same terminology for the C-band radar in Fig. 2.

lines 198-204: what is the frequency of this link? Please also indicate this in the paper.

Section 4.1: the evaluation metrics are discussed here. You choose to use the Spearman rank correlation over the more traditional Pearson correlation. I'm assuming that the reason for this is the insensitivity to outliers. Please indicate the main reason for using this correlation parameter. And you're also using the RMSE to complement the analysis, which I fully support. Consider using a normalized version of the RMSE (normalizing by the mean reference precipitation, like you do for PBIAS in Eq. (4)).

Section 4.1: on what time resolution are the metrics computed? Is it the native resolution of the sensor with the coarsest resolution? Or do you use a common time resolution for all sensors? The resolution can have a large impact on the values of the metrics, so it is important to know, and to take into account in interpreting the results.

lines 225-227: a time shift is discussed if correlations are low. I think this is a good idea. However, it is not clear from the paper when exactly this time shift is applied (“If very low correlations (close to 0) were found” is too vague). And it is also not clear how the time shift is determined. This should be included in the paper. Is the same time shift also applied when computing the RMSE? Please include that in the paper, too.

Section 4.4: I found myself scrolling back to Section 3 very often when reading Section 4.4. I therefore think it would greatly improve the readability of the paper if the contents of this section are moved to Section 3, where the different datasets are introduced.

lines 269-272: I think this text is not relevant for the paper. These are technical details of how to extract Zfrom the radar files and can be omitted from the paper in my view.

lines 277-281: it is not relevant that the resulting data are stored as geoTIFF files. You could also just summarize this paragraph by saying that CWR time series at a 5-minute resolution were created at the locations of the municipal rain gauge an the eight PWS.

lines 294-295: it is indicated here that there were missing values in the X-band radar data during the most intense part of the storm. This in itself is relevant information about the usefulness of such data: if data are missing when they are most crucial then the radar becomes much less useful. Therefore it would be very useful to know why these data are missing. And are entire 1-minute data missing, or are there gaps in the data around the most intense part of the storm? Please elaborate on this in the paper.

line 295: linear interpolation is mentioned here. Is it spatial or temporal linear interpolation. Please mention this explicitly.

lines 295-298: what is the reason for regridding onto a 500-m grid here? I think you can lose a lot of detail by doing this (given the native resolution of 250 m in range and approximately 350 m in azimuth at 40 km range).

Section 4.4.3: CML processing is first discussed in terms of a short literature overview of methods, after which the reader is referred to the appendix for information about which methods have actually been used in this paper. I think the literature overview of the methods (lines 301-319) should be moved to the appendix (it is valuable information, but it distracts from the main topic of the paper). The methods that are used in this paper should then be briefly described.

line 329: a difference in total rainfall depth of 3 mm is judged to be small here. The reader has not yet come across numbers of total accumulation as measured by the CML, so it is difficult to judge whether this is indeed the case. I suggest either adding the total accumulation for the links here, or expressing the difference as a percentage of the total accumulation.

line 335: it is stated that “XWR bins were sampled each 250 m”. However, in Section 4.4.2 (lines 295-297) you explain that the “volumetric data was gridded into a cartesian grid [with a resolution of] 500 meters”. So sampling it at 250 m doesn't make sense to me. The effective spacing of the data is then between 500 m and 700 m, depending on the orientation of the grid with respect to the CML. I think this should be made clear in the paper. And an explanation should be added about the reasons for this way of sampling the XWR data.

line 344: how relevant for this paper is the fact that the data were available as csv files and then converted to netCDF? I think this could be removed.

Table 1: if these numbers are presented here, then the meaning of all of the variables should be made clear in the text. Alternatively, you could remove the table, and only provide the values of m_match and m_int(the fact that you're using the values from de Vos et al. (2019) is already in the text on lines 375-376).

Table 2: is there a specific reason for taking the absolute value of the accumulated difference? Leaving the sign would be more instructive (although it can of course be read from the PBIAS column).

Section 5.2.1: I think the relevance of this section is minimal. The topic of the paper is on extreme precipitation, and the duration of precipitation is mostly determined by low-intensity precipitation. So I think most of this section (including Table 3) can removed, thereby improving the readability of this paper.

lines 440-442: a time lag of 25 minutes is necessary to optimize the correlation between the municipal gauge and the C-band radar. Such a lag is far beyond what would be expected from e.g. the time needed for the raindrops to fall from the radar measurement volume to the gauge. What can explain this difference. It is now stated the the C-band radar data should be shifted in time, but could it be that the clock on the municipal gauge was off? Please elaborate on this.

Figures 6, 8, 9, and 12: please express the RMSE in mm h^-1. This helps the reader compare to the y-axis that is used in the figures.

lines 447-449: the duration of the event is not really relevant here. The gauge recorded 75.4 mm in 54 minutes, so it recorded the same amount of precipitation in 60 minutes. I would suggest removing this sentence, and mentioning that 75.4 mm was recorded by the gauge within 60 minutes here.

lines 450-451 and 457-459: the difference between the X- and C-band radars are very large, both in spatial structure and in total accumulation. Could beam blockage indeed explain this (see my point about adding information in Figure 1 before)? Or are there other reasons that could cause such underestimates and differences in spatial structure?

lines 464-465: the lag that optimizes the correlation between XWR and the municipal gauge time series is 10 minutes. This is very different from the 25 minutes reported for the C-band weather radar. Does that mean that there is a time difference between the clocks of the C-band and the X-band radars? Or could it be that the time lag that optimizes the correlation is not the true time lag. Assuming that both radar clocks are correct, I think that a single time lag should be determined based on the comparison with both radars.

lines 466-469: in the discussion of Fig. 8 it is stated that the X-band radar underestimates the peak and overestimates the low intensities. I don't think you can draw that conclusion based on Fig. 8a. I see both over- and underestimation during heavy rain (i.e., everything above 10 mm h^-1). And I see that the X-band radar reports quite heavy rain before and after the most intense part reported by the municipal gauge (17:10-18:10). Especially the peak just before 17:00 (more than 10 mm h^-1) is significant (note that anything above 1 mm h^-1 cannot be called drizzle). The gauge reports nothing at that time. I think a more thorough discussion of this figure should be included in the paper.

Figure 9a: How is the spearman correlation coefficient computed given the fact that there are many CML data points that have exactly the same value (how do you determine the rank of these points)? I find it hard to believe that the correlation between CML and mean XWR is 0.9 given the large differences in the time series. Please check this number.

Figure 9b: when I look at the mean rainfall depth recorded by the X-band radar, and I compare this to the map in Fig. 7, these figures are inconsistent. Fig. 9 shows more than 60 mm, whereas the maximum rainfall depth on the link path in Fig. 7 is in the 50-60 mm class. Given the fact that parts of the link are in areas that received only 10-20 mm of rain according to Fig. 7, the numbers cannot be correct. The same holds for Fig. 8b where the graph indicates almost 80 mm of rain, but Fig. 7 shows between 50 and 60 mm. Please check the data behind Figures 7, 8, and 9, and correct where necessary.

Figure 9b: the gray area seem to be the accumulations based on the time series of the 10^th and 90^th But this is not really the spread in accumulations among the pixels over the CML. To show that, you would have to make the graphs of the accumulation for each pixel separately, and then take the 10^thand 90^th percentiles. I would expect that the spread would be more limited then (still significant, judging from Fig. 7). Please correct this in this figure.

lines 482-483: my conclusion here would be that the fact that the metrics are so good is pure coincidence. So I strongly suggest to remove any conclusions about how good the CML performs for this case because you can't really conclude that based on the data.

Figure 10: I really like this figure as it nicely demonstrates the enormous space-time variability of rainfall intensities for this storm and hence the challenge of sampling it well. Would it be possible to add the times (i.e., “17:22”, “17:23”, etc.) to each of the panels? That would increase the readability of the figure.

Figure 11b: this negative correlation is expected if the exponent of the CML rainfall retrieval relation (a) as expressed in Eq. (A2) is smaller than 1 (see Leijnse et al., 2010; Eq (10); note that in this equation the exponent b of the retrieval relation is defined differently (b=1/a)). What is the value of a used in this paper?

line 537: “expect” should be “except”.

lines 579-581: it is suggested here that partial beam blockage may be the cause of the severe underestimation of the rainfall accumulations by the C-band weather radar. This could definitely be the case. However, there is not enough information in the paper to be able to conclude this. There are also other sources of error could have resulted in these underestimated, even at close range (see e.g. van de Beek et al., 2016). Please add the relevant information to the paper (the SMHI, 2020 reference doesn't give the relevant information).

lines 600-602: it is stated that the size of the bucket in the tipping bucket gauge (0.2 mm) could be the cause of the low correlation. However, given the 5-minute time sampling used for computing this correlation, a single tip per 5 minutes would correspond to 2.4 mm h^-1. The intensities recorded by the X-band radar are much higher than that, even in the “calmer periods of the event”, so this can't be a significant factor in the low observed correlation coefficient.

lines 608-611: the effect of partial beam blockage is discussed here. I don't think that near-ground observations are particularly prone to beam blockage as a rule. And the larger sampling volume of the X-band radar will not suffer less from beam blockage. The siting of the radar is a much more important factor in this. Please adapt the discussion in the paper to reflect this.

line 612-613: I think that the statement that the CML correlated well with the X-band radar data is not something that should be highlighted here because of the plateau that the CML experiences (see also my earlier point about this).

line 628: X-band variability is partly attributed here to attenuation, but it is mentioned in Section 3.2 (lines 189-190) that attenuation has been corrected for. Please elaborate on this.

lines 632-634: the underestimation of both the C-band radar and CML with respect to the X-band radar are stated to be related to the shorter wavelength of the X-band radar. I don't think this is true. I strongly doubt that the wavelength of the CML is longer than that of the X-band radar (please indicate the frequency of the CML in the paper). And given the very short range of the C-band radar and the typical sensitivity of operational weather radars, this radar should have no trouble in detecting even very light precipitation. So I don't think that the wavelength of the different sensors can explain the observed differences.

line 637: I think the differences in durations between the rain gauge on the one hand and the CML and radars on the other can't be attributed to wind drift. The other two reasons that are given seen very valid to me. So consider removing the wind drift as a possible cause for this.

lines 658-674: the station outlier quality control filter is discussed here. It doesn't seem to perform well for this storm, flagging valid data, and failing to flag data of questionable quality. You mentioned earlier (on lines 378-382) that you modified two of the parameters of the SO algorithm. To what extent do you think this has influenced these results? Please also discuss this here.

lines 680-681: the excellent coverage of X-band radars in southern Sweden is discussed here. What fraction of the land surface of southern Sweden is actually covered by X-band radars? It would be interesting for readers to know this.

lines 687-688: rain/no rain detection is mentioned here. I don't think this is relevant for this paper (which is about intense convective precipitation). Furthermore, I don't think you can really conclude this based on the results presented here. It is very difficult knowing what the truth is with very light rainfall, given the sensors that have been used in this study. So I would suggest removing this from the paper.

Section 7: One thing that I think should be stressed more in the conclusions is that you've clearly shown that this storm is extremely variable in space (and that this holds for most storms that produce so much precipitation), and that all sensors that are added could potentially provide valuable information about parts of the storm that are otherwise not properly sampled. Could you stress this even more in the conclusions? This really shows why it is so important to continue research on second- and third-party data.

Appendix A: CML attenuation is discussed in this appendix, and it looks to me like attenuation (expressed in dB) and specific attenuation (expressed in dB km^-1) are intermixed. Please correct any issues related to this, and include units whenever introducing a new variable.

References
van de Beek, C. Z., Leijnse, H., Hazenberg, P., and Uijlenhoet, R.: Close-range radar rainfall estimation and error analysis, Atmos. Meas. Tech., 9, 3837–3850, doi:10.5194/amt-9-3837-2016, 2016.

Berne, A. and Uijlenhoet, R.: Path-averaged rainfall estimation using microwave links: Uncertainty due to spatial rainfall variability, Geophys. Res. Lett., 34, L07403, doi:10.1029/2007GL029409, 2007.

Leijnse, H., Uijlenhoet, R., and Stricker, J. N. M.: Microwave link rainfall estimation: Effects of link length and frequency, temporal sampling, power resolution, and wet antenna attenuation, Adv. Water Resour., 31, 1481–1493, doi:10.1016/j.advwatres.2008.03.004, 2008.

Leijnse, H., Uijlenhoet, R., and Berne, A.: Errors and uncertainties in microwave link rainfall estimation explored using drop size measure- ments and high-resolution radar data, J. Hydrometeorol., 11, 1330–1344, doi:10.1175/2010JHM1243.1, 2010.

de Vos, L. W., Raupach, T. H., Leijnse, H., Overeem, A., Berne, A., and Uijlenhoet, R.: High-resolution simulation study exploring the potential of radars, crowdsourced personal weather stations, and commercial microwave links to monitor small-scale urban rainfall, Water Resour. Res., 54, 10 293–10 312, doi:10.1029/2018WR023393, 2018.
Citation: https://doi.org/10.5194/egusphere-2025-2820-RC2
- AC2:
  'Reply on RC2', Louise Petersson Wårdh, 24 Oct 2025
  Thank you for taking the time to make this thorough review to improve the paper. We hope that the responses below will bring clarity to the arisen questions.
  Specific comments
  Thank you for noticing this – “rainfall” will be replaced with “attenuation”.
  
  We will refer to these papers and reflect on how the results compare with this literature in the discussion.
  
  We will replace “occupants” with “people”.
  
  We are specifically referring to line 89-93 and the results in section 5.2.4, especially the visualization in Fig. 10 and the analysis in Fig. 11. However, we acknowledge that this has been addressed in Leijnse et al. (2008, 2010) and other studies and we will therefore remove it from the list of research gaps and rephrase line 89-93.
  
  These are guiding research questions and we don’t claim to answer them completely once and for all with this study. We will relaxe the framing of the questions even further to emphasize this.
  
  This will be changed in the figure.
  
  The location of the C-band radar will be added to map 2.
  
  We are not sure if the reviewer asks for the predominant wind direction over all in this region or for the studied event specifically. In any case, the answer in both cases is southwest. This will be added to the text.
  
  The gauge adjustment technique is described in Michelson & Koistinen (2000) and we will outline the method briefly here. It is mentioned in lines 164-165 that the composite is distributed as reflectivity data, we will move this a bit earlier in the text for clarity. The composite must indeed be processed as described in section 4.4.1. Please refer to response #13 below for more details on the selection of this composite. There is no precipitation product available for CWR, as is the case with XWR in this study.
  
  The text will be changed to “radar horizontal reflectivity composite"
  
  We introduce the fundamental concept of DSD here and the applied Z-R relation in section 4.4.1 (methods).
  
  The text will be updated as suggested in the comment.
  
  The main reason for using this composite in the paper is reproducibility; it is the only SMHI radar product that is distributed openly. Secondly, the purpose of the paper is to benchmark second- and third-party sensors with a “conventional monitoring network”, not with the “best available radar product”. The applied composite is used operationally by the forecasting service at SMHI and distributed to external partners that use the radar sequences in decision making. We therefore consider this product to be the best representation of a “conventional monitoring network” in Sweden.
  
  When studying accumulated total depth recorded by this radar for the years 2022-2023, an internal evaluation at SMHI showed that the radar underestimated the accumulated depth compared with the SMHI rain gauges located to the north of Båstad with about 80% (see attached figure – it did not work to upload images, so please refer to the PDF-file under “supplement”). The beam blockage affects a circular sector of 60 degrees to the North of the radar location. We don’t find it appropriate to add the location of the vegetation to Figure 1 as the reader is not introduced to the problem of beam blockage by the radar at that point in the text. The affected area is shown in the reference SMHI (2020). This webpage also includes a north-south cross section (note that the area of interest in the study is located straight to the north of the radar location) which shows the elevation and vegetation in the area in relation to the radar beam, clearly indicating that the main lobe is blocked by vegetation. These figures will be added to the appendix.
  
  This method estimates attenuation as an approximately linear function of the specific differential phase shift, which depends on phase rather than signal intensity and is therefore less sensitive to attenuation (Kumjian 2013). We will mention this in the text and refer to Hosseini et al. (2020) for further details.
  
  Thank you for this question. We noticed an error in Fig. 3. For consistency, we will update both Fig. 2 and Fig. 3 so that we illustrate the beamwidth for both radars. To clarify, these are the nominal beamwidths provided by the manufacturers, sometimes referred to as half-power or –3 dB beamwidths.
  
  The frequency is 23.1 GHz, this will be added to the manuscript.
  
  We indeed think Spearman rank is a better choice in the context of convective storms due to the large variability and range of observations during such events. It also captures non-linear relationships. We considered to use normalized RMSE but will instead change the unit to mm/h in all plots (see comment 35 – thanks for pointing this out). This will allow intercomparisons both between different plots and with the y-axis in each plot.
  
  The metrics are indeed calculated on the resolution of the sensors with the coarsest resolution – PWS and CWR – which is 5 minutes. So, the metrics are calculated on 5 minutes resolution, except the long-term analysis, which is daily, and section 5.2.4 (XWR and CML analysis along CML path) which is on 1 minute. This will be added to the text and clarified where needed in the manuscript.
  
  Thanks for pointing this out. We will add a section in Methods called “Time lags” or the like where the procedure will be described. In addition, the lags have been recalculated and new (more reasonable) numbers will be reported- this affects comment #34 and #38 as well. See response #34 for more details on the new results. The time lags were calculated on 5 minutes resolution, as this is the resolution of the CWR data. Time lags from -10 (shifting radar 10 timesteps back in time compared with the municipal gauge) to 10 timesteps (same procedure but ahead in time) were considered and for each lag the Spearman rank correlation was calculated over the event duration as defined by the municipal gauge. The best correlation value was then selected. It should be noted that the selection of event duration has a great impact on these results. Including just one extra time step can give very different correlation values, often because it introduces more zero-rainfall-measurements and hence spuriously increases the correlation – we therefore find it important to keep section 5.2.1 in the paper to be clear on the event duration for each sensor (see comment 33). In fact, time lags were only applied to the radar data. The idea of applying time lags appeared when “low” correlations between the radars and municipal gauge were found. For the opportunistic sensors, the lowest correlation found was 0.44 compared with reference (PWS 4). For the radars, which we would have expected to perform better, the original calculations showed a correlation of 0.07 (CWR) and 0.18 (XWR) compared with reference. However, these results were based on incorrect duration of the storm (see response # 34). Rather than setting an arbitrary correlation threshold for the application of time lags, we will just state that time lags were applied to the radar data in the new section. The RMSE (and all other metrics) is calculated on the original time series, not the shifted ones. This shows how the sensors perform before applying lags which we think is more in line with the scope of the paper. This will be clarified in the results section.
  
  We agree and will incorporate the content of section 4.4 to section 3 instead.
  
  We think it is relevant to report which coefficients that were used to convert reflectivity to rain rate for reproducibility of the results and will therefore keep these lines in the text.
  
  The text will be updated accordingly.
  
  We agree that this is important to highlight this to understand the applicability of XWR for monitoring of convective storms. There are individual bins missing data around the intense part of the storm, not the full domain. Investigations suggest that this not due to a limitation of the sensor itself but imposed by post-processing in the pre-calculated rain rate that we used for the analysis. The data was compressed to 8-bit format and therefore any value above 255 (mm/h in this case) was lost. VeVa (see line 186) has since then changed the compression to 16-bit which allows for observations up to 650 mm/h without losing data, but the error is unfortunately present in 2022 data. We will elaborate on this in the paper.
  
  It is temporal linear interpolation, which will be clarified in the text.
  
  All XWR time series in the paper are sampled directly from the polar bins as described in lines 288 to 292. The data was resampled to a cartesian grid for the following reasons: 1) create the map in Figure 7, 2) to get a spatial overview of the total depth of the event and (importantly) 3) to allow for comparison with the CWR composite in Figure 5. We believe a resolution of 500 m serves these purposes. Although a higher spatial resolution of (for example) 250 m would have given even more detail it would also increase the difference even more to the CWR resolution. We will update the text to clarify this.
  
  We will move the literature review to the appendix and more content on the CML processing method will be moved from the appendix to this section.
  
  We will express the difference in percent instead. The difference is 3% (the total depth recorded by the sub-links was 57 and 60 mm, respectively).
  
  As outlined in response 26, all XWR time series in the paper are sampled from the polar bins as described in lines 288 to 292, not from the 500 m grid. That grid was created for the reasons outlined in response 26, mainly for visualization purposes in Fig 7.
  
  The text will be updated to reflect the comment.
  
  The table will be removed and the m_match and m_int values will be described in the text.
  
  The absolute value of the accumulated difference will be added to the table and all plots.
  
  The topic of the paper is not only extreme precipitation per se, but how the observations of such events differ between different types of sensors. Studying the duration of the same event observed by different sensors gives insights on 1) the sensitivity of the sensors 2) the temporal movement of the storm (which, as can be concluded from table 3, is from the south-west to the north-east in this case). In addition, we think it is an interesting finding that the CML and radars record similar durations. The duration also directly affects the calculation of return periods which we think is an important part of the discussion section. Furthermore, the evaluation metrics are computed for the duration of the event as recorded by each reference sensor (to exclude measurements of zero rainfall). For the given reasons, we suggest to keep this section in the paper.
  
  After checking the calculations of time lags in detail they were found to be incorrect and we have now corrected the calculations. Firstly, the event start at the municipal gauge was erroneously set to 17:15 in the calculations, when it in fact started at 17:05. As the metrics (and lags) are calculated for the duration of the municipal gauge when it serves as reference sensor, this will affect the calculation of time lags and metrics in Fig. 6 and Fig 8 as well. Secondly, the radar time series were mistakenly sliced to the event duration of the municipal gauge before shifting them in time, meaning that some iterations was based on only a few time steps, including the optimum value (25 minutes) that was based on only 5 timesteps. Instead, the correlation should of course be calculated over the event duration (17:05-18:10 at the municipal gauge) while shifting the tested time series step by step. After this correction, the Spearman correlation is 0.4 for CWR compared with reference without time lag, which could be increased to 0.83 when shifting the CWR time series 10 minutes. We find this to be reasonable values.
  
  RMSE will be expressed as mm/h in the figures.
  
  The text will be updated to mention 60 minutes duration only.
  
  We are confident that the difference in total accumulation is caused by beam blockage in this case, see responses to comment 14 and 47. The cause of differences in spatial structure is uncertain. The northeastern part of the study area has missing echo when comparing CWR with XWR – this might be a radar shadow behind the convective cell – but it is still very close to the radar location. Apart from this, it is hard to tell how much the two data sets should align given that they have different spatial resolution. It might be better to resample XWR to 2km to allow for more consistent visual comparison with CWR. If anything, these differences at least highlight the need for high-resolution data to observe convective storms.
  
  Following up on the response to comment #34, the same corrections were made for the XWR data. The Spearman correlation is 0.56 for XWR compared with reference without time lag, which could be increased to 0.7 when shifting the XWR time series 5 minutes. After these corrections we do not see a reason to believe that there is any significant difference between the clocks. Nevertheless, we will highlight the importance of accurate time recording in the discussion section.
  
  We will rephrase as follows: “In Fig. 8a, there is a tendency that XWR underestimated the overall peak rainfall intensity and overestimated lower rainfall intensities”. Note that we previously used the word “seems”, which we do not think indicate a strong statement. The overall peak 17:30-17:40 recorded by the municipal gauge is indeed underestimated by XWR, and if we use the municipal gauge as reference, then the peak recorded by XWR at 16:55 when the gauge reports nothing is by definition an overestimation – the same holds for 18:15-18:35. However, there are indeed a couple of timesteps 17:45-18:10 when XWR reports higher intensities than the gauge, hence we will use the word “tendency”.
  
  Recalculation of the Spearman rank gives the same result. It is correct that Spearman rank should not be applied for time series with many multiple values (ties) but we have applied ties-correction in the calculation (Emond et al., 2002). But as pointed out, it is probably also correct that the good metrics is a pure coincidence as the “plateau value” recorded by the CML happens to be in the middle range of the fluctuation of XWR observations, so the total accumulated depths are similar between CML and XWR reference. When computing the metrics for “non-plateau” values only the correlation is still 0.86 which suggest that CML and XWR record similar temporal variations during the storm. The absolute difference is then -15.5 mm, PBIAS 52 % and RMSE 4.1 mm/h for the non-plateau period, so a lot worse than with the “plateau-period” included. However, the non-plateau timesteps are outside the most intense part of the storm and therefore not the main focus of the analysis. We will keep the metrics as they are but use a more nuanced language, remove any formulations about that these metrics are “good” and discuss the fact that it is a coincidence in this case that the “plateau-intensity” leads to a similar total accumulation as the reference sensor.
  
  See response #26. The time series in Figure 8 and 9 are derived from XWR polar bins and Fig. 7 is gridded data which inevitably introduces averaging, especially for data with high spatial variability. This is explained in lines 455-457 but we will try to clarify this even further.
  
  Thanks for pointing this out, we had accidentally plotted instantaneous spread instead of cumulative spread. The figure will be corrected accordingly.
  
  We agree and will update the text accordingly.
  
  Thank you. This was actually included in an earlier version of the figure and will be added again.
  
  The value applied for alpha (for 23 GHz, vertical polarization) is 0.96, i.e. b = 1.04. We will refer to Leijnse et al., (2010) to support the findings.
  
  Thank you, this will be changed.
  
  See response to comment 14. Based on internal evaluations and experience with this particular radar, we strongly believe that the siting of the radar (and consequently vegetation beam blockage) is the most contributing factor to the severe underestimation. See response #14 for more details. We will add a clause on other possible sources of the error and refer to the suggested paper.
  
  Thank you for pointing this out, we will remove it as a possible cause for the low correlation with the municipal gauge.
  
  Thank you for pointing this out, we will remove this sentence.
  
  We agree and will update the discussion accordingly.
  
  There might be local attenuation in the data even after application of attenuation correction as all methods have limitations and uncertainties. We will add a clause like “… due to possible uncertainties in the attenuation correction” or the like.
  
  It is correct that the frequency of the XWR (9.4 GHz) is of course lower than the CML (23.1 GHz) so this statement will be removed. Our investigations show that the CWR indeed has trouble detecting light precipitation at this elevation angle (higher angles perform better). However, we think it is out of the scope of the paper to elaborate on this, so we will just delete this sentence.
  
  Thank you, wind drift will be removed as a possible cause for the observed discrepancy.
  
  The original values of the parameters m_match and m_intwere 200 and 4032, which means that 200 rainy time steps in the last 2 weeks (for 5-minute data, which is used here) are required to compute Station Outlier flags (as we know, correlations should not be calculated on dry time steps). With these settings, all stations were flagged as “-1” (“filter cannot be applied”) for the period of interest because there were not enough rainy timesteps in the last 2 weeks. This will be clarified in lines 379-380. Instead, we require 100 wet timesteps in the last four weeks to apply the filter. The result of this is that no timesteps are flagged as “-1” in the period of interest but the SO algorithm can be applied at all time steps. The quality control is then based on less data (100 wet timesteps instead of 200) but it is a trade-off between having any quality control performed at all, and having it performed based on less reference data. So, the failure of the SO-algorithm cannot be attributed to the change of parameter settings but rather to the spatial variability of the storm. We will incorporate these reflections in the discussion section.
  
  We will change vague references like “southern Sweden” and “southwestern Sweden” to the more well-defined area “Skåne county” in section 1, 2, 3.2 and 6. The word “excellent” in the text refers to the combination of XWR and rain gauges operated by local authorities, not the XWR coverage itself. There are two X-band radars in Skåne (and all of Sweden) – their coverage is shown in Figure 1 in Hosseini et al (2023). This is approximately 9000 km², corresponding to 80 % of Skåne county and 2 % of all of Sweden.
  
  The intention of this statement is to highlight that third-party sensors can assist in understanding the spatial variability of a storm, while they may not always provide accurate information on intensity/depth at all points of interest. This phrasing is already used in the conclusion section and will be added here for clarity.
  
  Section 7 will be amended to put more emphasis on the added value of incorporating these sensor types as complementary observations in the context of extreme rainfall.
  
  Thank you for noticing this, we will revise this part of the appendix and clarify the units. The unit of equation A2 was incorrect.
  
  References
  Emond, Edward J., and David W. Mason. "A new rank correlation coefficient with application to the consensus ranking problem." Journal of Multi‐Criteria Decision Analysis 11.1 (2002): 17-28
  Kumjian, M. R. (2013). Principles and applications of dual-polarization weather radar. Part I: Description of the polarimetric radar variables. Journal of Operational Meteorology, 1.
  Hosseini, S. H., Hashemi, H., Berndtsson, R., South, N., Aspegren, H., Larsson, R., Olsson, J., Persson, A., & Olsson, L. (2020). Evaluation of a new X-band weather radar for operational use in south Sweden. Water Science and Technology
  Hosseini, S. H., Hashemi, H., Larsson, R., & Berndtsson, R. (2023). Merging dual-polarization X-band radar network intelligence for improved microscale observation of summer rainfall in south Sweden. Journal of Hydrology, 617, 129090. https://doi.org/10.1016/j.jhydrol.2023.129090
  SMHI. (2020). Trädtoppar orsakar blockeringar för väderradar Ängelholm—SMHI. https://www.smhi.se/nyheter/nyheter/2020-11-20-tradtoppar-orsakar-blockeringar-for-vaderradar-angelholm
  de Vos, L. W., Leijnse, H., Overeem, A., & Uijlenhoet, R. (2019). Quality Control for Crowdsourced Personal Weather Stations to Enable Operational Rainfall Monitoring. Geophysical Research Letters, 46(15), 8820–8829. https://doi.org/10.1029/2019GL083731
  
  Citation: https://doi.org/10.5194/egusphere-2025-2820-AC2

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-2820', Anonymous Referee #1, 22 Jul 2025

This manuscript aims to assess the capacity for second and third-party sensors to observe short duration extreme rainfall via comparisons to the national rainfall monitoring network in Sweden. While this study found that even the national rainfall monitoring network was unable to fully capture the results of a strong convective storm, it was shown that second and third party sensors can provide accurate representations of that data. This is an important addition to accurate weather monitoring as these additional sensors can assist with creating detailed representations of significant storms within a given area. The authors highlight several limitations to the integration of second and third party sensors into conventional weather monitoring systems and encourage the need for more research in this area. I support the publication of this work following minor edits described below. This work highlights crucial elements for establishing robust weather monitoring protocols and should provide the authors and other researchers will a strong starting point for future research and improvements in this area.
There are, however, a few areas where this manuscript could be improved. Most of the comments before focus on clarity of terminology and how the discussion relates back to the results presented.
Line 161: I was unsure what area this portion of the text was referring to in Figure 1. Is this just a composite of SMHI stations? Clarity could be improved here so readers understand the monitoring locations.
Line 194: Can you define overshooting in the text? What does it mean in this context?
Line 227-229: This sentence would benefit from more clarity and further explanation.
Line 272 and 274: I appreciate the clear explanation of all equations and understand that several of the parameters were obtained from other sources. Could you briefly elaborate what the G and offset variables represent for Equation 6 as well as a and b for Equation 7?
Line 392: Is it typical to see a trend with higher accumulated depth at inland regions like Baramossa? A reference here may be appropriate to establish that your results are typical and expected.
Figure 4: It’s difficult to tell the difference between SMHI Hov and SMHI Laholm D. The line types are very similar. This figure would be more clear with a different line choice for one of those data sets.
Line 494: Could you elaborate in the text on why the plateau period was considering unsuitable for comparison?
Figure 10: Is each panel a-e correspond to one minute? For example: panel a is 17:22-17:23 and panel b is 17:23-17:24. If so, this is not clear based on the information provided in the figure caption.
Figure 11a: Are you expecting a linear trend with this comparison? If so, reporting an R² value would be useful.
Figure 12: The figure caption, please specify that the r_s is relating the data to XWR data.
The discussion overall is thorough and touches on many good points. However, is there a reason the results and discussion sections are separate? At times this made assessing the claims made in the discussion challenging due to needing to locate the appropriate data in the Results section. If you choose to leave the Discussion as a separate section, please refer the reader to relevant tables and figures when emphasizing trends observed in the data. I have noted a few specific lines below, but I encourage you to go back through this section for any crucial information that would benefit from being redirected back to a table or figure.
Lines 589-591, Lines 598-599, Line 604, Line 612-616, Lines 628-630, Lines 644-645, Lines 660-662.

Citation: https://doi.org/10.5194/egusphere-2025-2820-RC1
- AC1: 'Reply on RC1', Louise Petersson Wårdh, 21 Aug 2025
  
  We thank the reviewer for the helpful comments that will support clarifications and further enhancement of the manuscript. Please find answers to the comments below.
  Line 161: Thank you for pointing this out. The text is referring to the C-band radar located 6 km South of the study area, marked with a white dot in Map 1 in Figure 1. The radar location is unfortunately missing from Map 2 in Figure 1 and will be added.
  
  Line 194: Overshooting in this context means that the radar beam shoots above the precipitation cloud and hence does not record it. It is a common source of error in radar data. We will add some standard literature as reference for further reading on the topic. The purpose of this section is to see if there is risk of overshooting by the X-band radar in the area of interest. The same analysis is done in lines 168 – 171 for the C-band radar. As the radar beam we are using travels on 200-300 m height at the area of interest for CWR and 300-1200 m for XWR, and convective precipitation in the summer months in Sweden typically originate from much higher altitudes, the risk of overshooting is very small. We will add sources to support these numbers.
  
  Line 227-229: The intention of this sentence is to add clarity, but if it only confuses things, it is maybe better removed. For sub-hourly rainfall data, the correlation between time series recorded by nearby sensors can be low even if they record similar total rainfall depths (for example Spearman correlation 0.18 between XWR and the municipal gauge, see Results). As convective rainfall is highly variable in space and time the observations per time step can be very different at nearby locations. If you use correlation as metric this will indicate poor performance, when, in fact, the sensors simply may experience different rainfall intensities even if they are closely located. By shifting the time series in time, you can account for the fact that a certain rainfall intensity may be observed by the radar before it reaches a rain gauge on the ground, for example.
  
  Line 272 and 274:
  Equation 6: Reflectivity data from radars are stored and distributed as integers between 0 and 255 to enable smaller storage size, following European standards (Michelson et al., 2014). To convert these integers back to reflectivity (dBZ) you apply the coefficients G (gain) and offset.
  Equation 7: a and b are well-established empirical constants that Marshall and Palmer (1948) found when establishing the relationship between size distribution of raindrops (DSD) in a radar pulse volume to the rainfall rate.
  
  Line 392: This is not a very general feature for Sweden, but the statement referred to the specific study area. The pattern can be seen in maps of 30-year mean annual precipitation here: https://www.smhi.se/klimat/klimatet-da-och-nu/normalkartor/normal/arsnederbord-normal. We will clarify this part of the text in the revised manuscript.
  
  Figure 4: Thank you for pointing this out. The line types will be better differentiated in the next version.
  
  Line 494: Given the sudden constant records of rainfall rate observed in the CML time series, which is unexpected and clearly not representative for the actual rainfall rate, it was excluded from the analysis.
  
  Figure 10: Thank you for pointing this out. It is correct that, for example, panel a is 17:22-17:23 and panel b is 17:23-17:24. This will be clarified in the next version.
  
  Figure 11a: Thank you for pointing this out. A linear trend is expected because, ideally, the two sensors (CML and XWR at the CML location) should record the same or very similar rainfall intensities. R² will be reported in the next version.
  
  Figure 12: Thank you for this valid and helpful comment, it will be added in the next version.
  
  About the Discussion section: We were indeed discussing whether to keep the results and discussion sections together or apart before arriving at this solution. It is understood that the discussion, if kept separate, will need more cross-references to the text. We will await the second review to decide which approach to choose.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2820-AC1
- AC3: 'Reply on RC1', Louise Petersson Wårdh, 24 Oct 2025
  
  After receiving comments from reviewer #2, we decided to keep the results and discussion section separate as is but improve the cross-referencing to the results section.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2820-AC3
RC2:
'Comment on egusphere-2025-2820', Hidde Leijnse, 26 Sep 2025
General comments
This paper describes a thorough analysis of in intense convective storm that occurred in southern Sweden. Data from the national precipitation monitoring network (gauges and C-band radar) are compared to second-party (X-band radar and a municipal gauge) and third-party (a commercial microwave link and rain gauges on personal weather stations) for this storm. The analyses presented give insight into the performance of the different sensors, and the discussion section provides clear conclusions about the strong and weak aspects of the different sensors. I think the paper is interesting and relevant for Atmospheric Measurement Techniques. I do think that the paper needs some work in improving the readability, providing some additional information that is important for interpreting the results, and checking some of the data and figures that are presented.
The most important aspect that I think requires attention is that there seems to be an inconsistency between the X-band radar data presented in Figures 7 ,8, and 9 (see the specific point about this below). It is also not clear how the time shift that was apparent in the municipal gauge was taken into account (I'm assuming here that the other sensors have more accurate clocks), and I'm missing something in the discussion about the importance of accurate time recording of the data in order for it to be useful. And some details provided in the paper about data processing are not relevant and could be removed, while other aspects of the datasets and their processing are missing. Details are provided in specific comments provided below (I realize that there are quite many comments; please take these as an encouragement to further improve this interesting paper).

Specific comments
line 87: CMLs measure path-integrated attenuation, not rainfall. I guess that is what you're trying to say with this sentence: that the assumption is that this is directly related to the path-averaged rainfall intensity.

lines 88-89: this assumption has been investigated by Berne et al. (2007), Leijnse et al. (2008, 2010), and De Vos et al. (2018) in high-resolution simulation frameworks. It would be interesting to see how your results for this case compare to what is presented in this literature.

line 94: consider using “people” instead of “occupants”.

line 107-108: you state that “cross-referencing radar observations with path-integrated rainfall estimated by CML” is a gap in the field of high-resolution rainfall monitoring. A large portion of the work cross-references radar and CML rainfall estimates, so I don't fully understand this statement. Could you make the statement more specific about which gap exactly is bein addressed here?

lines 109-114: I think that the research questions as they are formulated here cannot be answered by the analysis of a single convective event, because it probably depends greatly on the path of the storm and the local topology of the sensors in that path. Could the research questions be reformulated to take this into account?

Figure 1: it would be instructive to have different symbols on the map for the SMHI automatic gauge and the SMHI manual gauges.

Figure 1: could the location of the C-band radar be put on map 2 as well, so that it is clear where the radar is relative to the study area?

Figure 1: would it be possible to add the predominant wind direction to the figure? This could help in interpreting differences between the municipal gauge and the SMHI gauges.

line 155: it is mentioned here that “gauge-adjusted Plan Position Indicator” is used. However, in the subsequent paragraphs there is no information about how the gauge adjustment was carried out. Because this is very relevant for interpreting the results, please add information about this in the paper. Furthermore, I get the impression here that a precipitation product is used in this paper, but from Section 4.4.1 it seems that it is actually a much more raw product. Please clarify this in the paper. And if this is indeed a more raw reflectivity product, then I would suggest to use the radar data on its native (spherical) grid, i.e., not the 2x2 km composite because this would give much more spatial detail.

line 155: consider adding “reflectivity” (or even “horizontal reflectivity”) between “radar” and “composite” to make clear what kind of PPI is used.

lines 157-160: here you explain Z in terms of the DSD (you could also have chosen to present it in the form of the radar equation). To me it seems more relevant to present the resulting Z-R relation instead, preferably the one that was used to retrieve rainfall intensity for this study.

lines 161-162: the degree to which a radar contributes to a given point in a composite depends on the employed compositing method. So consider adding “and the compositing method is based on the closest radar” (or something similar, depending on the compositing method that is used) after “during the selected event”.

line 163: here it is mentioned that the resolution of the composite is 2x2 km. Given the very close range and the expected high spatial variability of the precipitation, using the radar data on its native (spherical) grid would give much more spatial detail of the storm. Can you indicate the reason for using the composite in the paper?

lines 171-173: here you discuss partial beam blockage. Could you indicate the severity of the beam blockage? And could the location of the vegetation causing this blockage be indicated on map 2 of Fig. 1? This is relevant for the interpretation of the results.

line 190: dual-pol attenuation correction is mentioned here, and the reader is referred to the literature for details on the method. I think it would be relevant here to add a few words (or a sentence) about which method is used (referring to the literature for details of the method is fine).

line 191 and Figure 3: “half-beam” and “half beamwidth” are mentioned here. What is meant by this? Do you mean half-power beam width? If so, please modify this. And if this is the case, then consider to use the same terminology for the C-band radar in Fig. 2.

lines 198-204: what is the frequency of this link? Please also indicate this in the paper.

Section 4.1: the evaluation metrics are discussed here. You choose to use the Spearman rank correlation over the more traditional Pearson correlation. I'm assuming that the reason for this is the insensitivity to outliers. Please indicate the main reason for using this correlation parameter. And you're also using the RMSE to complement the analysis, which I fully support. Consider using a normalized version of the RMSE (normalizing by the mean reference precipitation, like you do for PBIAS in Eq. (4)).

Section 4.1: on what time resolution are the metrics computed? Is it the native resolution of the sensor with the coarsest resolution? Or do you use a common time resolution for all sensors? The resolution can have a large impact on the values of the metrics, so it is important to know, and to take into account in interpreting the results.

lines 225-227: a time shift is discussed if correlations are low. I think this is a good idea. However, it is not clear from the paper when exactly this time shift is applied (“If very low correlations (close to 0) were found” is too vague). And it is also not clear how the time shift is determined. This should be included in the paper. Is the same time shift also applied when computing the RMSE? Please include that in the paper, too.

Section 4.4: I found myself scrolling back to Section 3 very often when reading Section 4.4. I therefore think it would greatly improve the readability of the paper if the contents of this section are moved to Section 3, where the different datasets are introduced.

lines 269-272: I think this text is not relevant for the paper. These are technical details of how to extract Zfrom the radar files and can be omitted from the paper in my view.

lines 277-281: it is not relevant that the resulting data are stored as geoTIFF files. You could also just summarize this paragraph by saying that CWR time series at a 5-minute resolution were created at the locations of the municipal rain gauge an the eight PWS.

lines 294-295: it is indicated here that there were missing values in the X-band radar data during the most intense part of the storm. This in itself is relevant information about the usefulness of such data: if data are missing when they are most crucial then the radar becomes much less useful. Therefore it would be very useful to know why these data are missing. And are entire 1-minute data missing, or are there gaps in the data around the most intense part of the storm? Please elaborate on this in the paper.

line 295: linear interpolation is mentioned here. Is it spatial or temporal linear interpolation. Please mention this explicitly.

lines 295-298: what is the reason for regridding onto a 500-m grid here? I think you can lose a lot of detail by doing this (given the native resolution of 250 m in range and approximately 350 m in azimuth at 40 km range).

Section 4.4.3: CML processing is first discussed in terms of a short literature overview of methods, after which the reader is referred to the appendix for information about which methods have actually been used in this paper. I think the literature overview of the methods (lines 301-319) should be moved to the appendix (it is valuable information, but it distracts from the main topic of the paper). The methods that are used in this paper should then be briefly described.

line 329: a difference in total rainfall depth of 3 mm is judged to be small here. The reader has not yet come across numbers of total accumulation as measured by the CML, so it is difficult to judge whether this is indeed the case. I suggest either adding the total accumulation for the links here, or expressing the difference as a percentage of the total accumulation.

line 335: it is stated that “XWR bins were sampled each 250 m”. However, in Section 4.4.2 (lines 295-297) you explain that the “volumetric data was gridded into a cartesian grid [with a resolution of] 500 meters”. So sampling it at 250 m doesn't make sense to me. The effective spacing of the data is then between 500 m and 700 m, depending on the orientation of the grid with respect to the CML. I think this should be made clear in the paper. And an explanation should be added about the reasons for this way of sampling the XWR data.

line 344: how relevant for this paper is the fact that the data were available as csv files and then converted to netCDF? I think this could be removed.

Table 1: if these numbers are presented here, then the meaning of all of the variables should be made clear in the text. Alternatively, you could remove the table, and only provide the values of m_match and m_int(the fact that you're using the values from de Vos et al. (2019) is already in the text on lines 375-376).

Table 2: is there a specific reason for taking the absolute value of the accumulated difference? Leaving the sign would be more instructive (although it can of course be read from the PBIAS column).

Section 5.2.1: I think the relevance of this section is minimal. The topic of the paper is on extreme precipitation, and the duration of precipitation is mostly determined by low-intensity precipitation. So I think most of this section (including Table 3) can removed, thereby improving the readability of this paper.

lines 440-442: a time lag of 25 minutes is necessary to optimize the correlation between the municipal gauge and the C-band radar. Such a lag is far beyond what would be expected from e.g. the time needed for the raindrops to fall from the radar measurement volume to the gauge. What can explain this difference. It is now stated the the C-band radar data should be shifted in time, but could it be that the clock on the municipal gauge was off? Please elaborate on this.

Figures 6, 8, 9, and 12: please express the RMSE in mm h^-1. This helps the reader compare to the y-axis that is used in the figures.

lines 447-449: the duration of the event is not really relevant here. The gauge recorded 75.4 mm in 54 minutes, so it recorded the same amount of precipitation in 60 minutes. I would suggest removing this sentence, and mentioning that 75.4 mm was recorded by the gauge within 60 minutes here.

lines 450-451 and 457-459: the difference between the X- and C-band radars are very large, both in spatial structure and in total accumulation. Could beam blockage indeed explain this (see my point about adding information in Figure 1 before)? Or are there other reasons that could cause such underestimates and differences in spatial structure?

lines 464-465: the lag that optimizes the correlation between XWR and the municipal gauge time series is 10 minutes. This is very different from the 25 minutes reported for the C-band weather radar. Does that mean that there is a time difference between the clocks of the C-band and the X-band radars? Or could it be that the time lag that optimizes the correlation is not the true time lag. Assuming that both radar clocks are correct, I think that a single time lag should be determined based on the comparison with both radars.

lines 466-469: in the discussion of Fig. 8 it is stated that the X-band radar underestimates the peak and overestimates the low intensities. I don't think you can draw that conclusion based on Fig. 8a. I see both over- and underestimation during heavy rain (i.e., everything above 10 mm h^-1). And I see that the X-band radar reports quite heavy rain before and after the most intense part reported by the municipal gauge (17:10-18:10). Especially the peak just before 17:00 (more than 10 mm h^-1) is significant (note that anything above 1 mm h^-1 cannot be called drizzle). The gauge reports nothing at that time. I think a more thorough discussion of this figure should be included in the paper.

Figure 9a: How is the spearman correlation coefficient computed given the fact that there are many CML data points that have exactly the same value (how do you determine the rank of these points)? I find it hard to believe that the correlation between CML and mean XWR is 0.9 given the large differences in the time series. Please check this number.

Figure 9b: when I look at the mean rainfall depth recorded by the X-band radar, and I compare this to the map in Fig. 7, these figures are inconsistent. Fig. 9 shows more than 60 mm, whereas the maximum rainfall depth on the link path in Fig. 7 is in the 50-60 mm class. Given the fact that parts of the link are in areas that received only 10-20 mm of rain according to Fig. 7, the numbers cannot be correct. The same holds for Fig. 8b where the graph indicates almost 80 mm of rain, but Fig. 7 shows between 50 and 60 mm. Please check the data behind Figures 7, 8, and 9, and correct where necessary.

Figure 9b: the gray area seem to be the accumulations based on the time series of the 10^th and 90^th But this is not really the spread in accumulations among the pixels over the CML. To show that, you would have to make the graphs of the accumulation for each pixel separately, and then take the 10^thand 90^th percentiles. I would expect that the spread would be more limited then (still significant, judging from Fig. 7). Please correct this in this figure.

lines 482-483: my conclusion here would be that the fact that the metrics are so good is pure coincidence. So I strongly suggest to remove any conclusions about how good the CML performs for this case because you can't really conclude that based on the data.

Figure 10: I really like this figure as it nicely demonstrates the enormous space-time variability of rainfall intensities for this storm and hence the challenge of sampling it well. Would it be possible to add the times (i.e., “17:22”, “17:23”, etc.) to each of the panels? That would increase the readability of the figure.

Figure 11b: this negative correlation is expected if the exponent of the CML rainfall retrieval relation (a) as expressed in Eq. (A2) is smaller than 1 (see Leijnse et al., 2010; Eq (10); note that in this equation the exponent b of the retrieval relation is defined differently (b=1/a)). What is the value of a used in this paper?

line 537: “expect” should be “except”.

lines 579-581: it is suggested here that partial beam blockage may be the cause of the severe underestimation of the rainfall accumulations by the C-band weather radar. This could definitely be the case. However, there is not enough information in the paper to be able to conclude this. There are also other sources of error could have resulted in these underestimated, even at close range (see e.g. van de Beek et al., 2016). Please add the relevant information to the paper (the SMHI, 2020 reference doesn't give the relevant information).

lines 600-602: it is stated that the size of the bucket in the tipping bucket gauge (0.2 mm) could be the cause of the low correlation. However, given the 5-minute time sampling used for computing this correlation, a single tip per 5 minutes would correspond to 2.4 mm h^-1. The intensities recorded by the X-band radar are much higher than that, even in the “calmer periods of the event”, so this can't be a significant factor in the low observed correlation coefficient.

lines 608-611: the effect of partial beam blockage is discussed here. I don't think that near-ground observations are particularly prone to beam blockage as a rule. And the larger sampling volume of the X-band radar will not suffer less from beam blockage. The siting of the radar is a much more important factor in this. Please adapt the discussion in the paper to reflect this.

line 612-613: I think that the statement that the CML correlated well with the X-band radar data is not something that should be highlighted here because of the plateau that the CML experiences (see also my earlier point about this).

line 628: X-band variability is partly attributed here to attenuation, but it is mentioned in Section 3.2 (lines 189-190) that attenuation has been corrected for. Please elaborate on this.

lines 632-634: the underestimation of both the C-band radar and CML with respect to the X-band radar are stated to be related to the shorter wavelength of the X-band radar. I don't think this is true. I strongly doubt that the wavelength of the CML is longer than that of the X-band radar (please indicate the frequency of the CML in the paper). And given the very short range of the C-band radar and the typical sensitivity of operational weather radars, this radar should have no trouble in detecting even very light precipitation. So I don't think that the wavelength of the different sensors can explain the observed differences.

line 637: I think the differences in durations between the rain gauge on the one hand and the CML and radars on the other can't be attributed to wind drift. The other two reasons that are given seen very valid to me. So consider removing the wind drift as a possible cause for this.

lines 658-674: the station outlier quality control filter is discussed here. It doesn't seem to perform well for this storm, flagging valid data, and failing to flag data of questionable quality. You mentioned earlier (on lines 378-382) that you modified two of the parameters of the SO algorithm. To what extent do you think this has influenced these results? Please also discuss this here.

lines 680-681: the excellent coverage of X-band radars in southern Sweden is discussed here. What fraction of the land surface of southern Sweden is actually covered by X-band radars? It would be interesting for readers to know this.

lines 687-688: rain/no rain detection is mentioned here. I don't think this is relevant for this paper (which is about intense convective precipitation). Furthermore, I don't think you can really conclude this based on the results presented here. It is very difficult knowing what the truth is with very light rainfall, given the sensors that have been used in this study. So I would suggest removing this from the paper.

Section 7: One thing that I think should be stressed more in the conclusions is that you've clearly shown that this storm is extremely variable in space (and that this holds for most storms that produce so much precipitation), and that all sensors that are added could potentially provide valuable information about parts of the storm that are otherwise not properly sampled. Could you stress this even more in the conclusions? This really shows why it is so important to continue research on second- and third-party data.

Appendix A: CML attenuation is discussed in this appendix, and it looks to me like attenuation (expressed in dB) and specific attenuation (expressed in dB km^-1) are intermixed. Please correct any issues related to this, and include units whenever introducing a new variable.

References
van de Beek, C. Z., Leijnse, H., Hazenberg, P., and Uijlenhoet, R.: Close-range radar rainfall estimation and error analysis, Atmos. Meas. Tech., 9, 3837–3850, doi:10.5194/amt-9-3837-2016, 2016.

Berne, A. and Uijlenhoet, R.: Path-averaged rainfall estimation using microwave links: Uncertainty due to spatial rainfall variability, Geophys. Res. Lett., 34, L07403, doi:10.1029/2007GL029409, 2007.

Leijnse, H., Uijlenhoet, R., and Stricker, J. N. M.: Microwave link rainfall estimation: Effects of link length and frequency, temporal sampling, power resolution, and wet antenna attenuation, Adv. Water Resour., 31, 1481–1493, doi:10.1016/j.advwatres.2008.03.004, 2008.

Leijnse, H., Uijlenhoet, R., and Berne, A.: Errors and uncertainties in microwave link rainfall estimation explored using drop size measure- ments and high-resolution radar data, J. Hydrometeorol., 11, 1330–1344, doi:10.1175/2010JHM1243.1, 2010.

de Vos, L. W., Raupach, T. H., Leijnse, H., Overeem, A., Berne, A., and Uijlenhoet, R.: High-resolution simulation study exploring the potential of radars, crowdsourced personal weather stations, and commercial microwave links to monitor small-scale urban rainfall, Water Resour. Res., 54, 10 293–10 312, doi:10.1029/2018WR023393, 2018.
Citation: https://doi.org/10.5194/egusphere-2025-2820-RC2
- AC2:
  'Reply on RC2', Louise Petersson Wårdh, 24 Oct 2025
  Thank you for taking the time to make this thorough review to improve the paper. We hope that the responses below will bring clarity to the arisen questions.
  Specific comments
  Thank you for noticing this – “rainfall” will be replaced with “attenuation”.
  
  We will refer to these papers and reflect on how the results compare with this literature in the discussion.
  
  We will replace “occupants” with “people”.
  
  We are specifically referring to line 89-93 and the results in section 5.2.4, especially the visualization in Fig. 10 and the analysis in Fig. 11. However, we acknowledge that this has been addressed in Leijnse et al. (2008, 2010) and other studies and we will therefore remove it from the list of research gaps and rephrase line 89-93.
  
  These are guiding research questions and we don’t claim to answer them completely once and for all with this study. We will relaxe the framing of the questions even further to emphasize this.
  
  This will be changed in the figure.
  
  The location of the C-band radar will be added to map 2.
  
  We are not sure if the reviewer asks for the predominant wind direction over all in this region or for the studied event specifically. In any case, the answer in both cases is southwest. This will be added to the text.
  
  The gauge adjustment technique is described in Michelson & Koistinen (2000) and we will outline the method briefly here. It is mentioned in lines 164-165 that the composite is distributed as reflectivity data, we will move this a bit earlier in the text for clarity. The composite must indeed be processed as described in section 4.4.1. Please refer to response #13 below for more details on the selection of this composite. There is no precipitation product available for CWR, as is the case with XWR in this study.
  
  The text will be changed to “radar horizontal reflectivity composite"
  
  We introduce the fundamental concept of DSD here and the applied Z-R relation in section 4.4.1 (methods).
  
  The text will be updated as suggested in the comment.
  
  The main reason for using this composite in the paper is reproducibility; it is the only SMHI radar product that is distributed openly. Secondly, the purpose of the paper is to benchmark second- and third-party sensors with a “conventional monitoring network”, not with the “best available radar product”. The applied composite is used operationally by the forecasting service at SMHI and distributed to external partners that use the radar sequences in decision making. We therefore consider this product to be the best representation of a “conventional monitoring network” in Sweden.
  
  When studying accumulated total depth recorded by this radar for the years 2022-2023, an internal evaluation at SMHI showed that the radar underestimated the accumulated depth compared with the SMHI rain gauges located to the north of Båstad with about 80% (see attached figure – it did not work to upload images, so please refer to the PDF-file under “supplement”). The beam blockage affects a circular sector of 60 degrees to the North of the radar location. We don’t find it appropriate to add the location of the vegetation to Figure 1 as the reader is not introduced to the problem of beam blockage by the radar at that point in the text. The affected area is shown in the reference SMHI (2020). This webpage also includes a north-south cross section (note that the area of interest in the study is located straight to the north of the radar location) which shows the elevation and vegetation in the area in relation to the radar beam, clearly indicating that the main lobe is blocked by vegetation. These figures will be added to the appendix.
  
  This method estimates attenuation as an approximately linear function of the specific differential phase shift, which depends on phase rather than signal intensity and is therefore less sensitive to attenuation (Kumjian 2013). We will mention this in the text and refer to Hosseini et al. (2020) for further details.
  
  Thank you for this question. We noticed an error in Fig. 3. For consistency, we will update both Fig. 2 and Fig. 3 so that we illustrate the beamwidth for both radars. To clarify, these are the nominal beamwidths provided by the manufacturers, sometimes referred to as half-power or –3 dB beamwidths.
  
  The frequency is 23.1 GHz, this will be added to the manuscript.
  
  We indeed think Spearman rank is a better choice in the context of convective storms due to the large variability and range of observations during such events. It also captures non-linear relationships. We considered to use normalized RMSE but will instead change the unit to mm/h in all plots (see comment 35 – thanks for pointing this out). This will allow intercomparisons both between different plots and with the y-axis in each plot.
  
  The metrics are indeed calculated on the resolution of the sensors with the coarsest resolution – PWS and CWR – which is 5 minutes. So, the metrics are calculated on 5 minutes resolution, except the long-term analysis, which is daily, and section 5.2.4 (XWR and CML analysis along CML path) which is on 1 minute. This will be added to the text and clarified where needed in the manuscript.
  
  Thanks for pointing this out. We will add a section in Methods called “Time lags” or the like where the procedure will be described. In addition, the lags have been recalculated and new (more reasonable) numbers will be reported- this affects comment #34 and #38 as well. See response #34 for more details on the new results. The time lags were calculated on 5 minutes resolution, as this is the resolution of the CWR data. Time lags from -10 (shifting radar 10 timesteps back in time compared with the municipal gauge) to 10 timesteps (same procedure but ahead in time) were considered and for each lag the Spearman rank correlation was calculated over the event duration as defined by the municipal gauge. The best correlation value was then selected. It should be noted that the selection of event duration has a great impact on these results. Including just one extra time step can give very different correlation values, often because it introduces more zero-rainfall-measurements and hence spuriously increases the correlation – we therefore find it important to keep section 5.2.1 in the paper to be clear on the event duration for each sensor (see comment 33). In fact, time lags were only applied to the radar data. The idea of applying time lags appeared when “low” correlations between the radars and municipal gauge were found. For the opportunistic sensors, the lowest correlation found was 0.44 compared with reference (PWS 4). For the radars, which we would have expected to perform better, the original calculations showed a correlation of 0.07 (CWR) and 0.18 (XWR) compared with reference. However, these results were based on incorrect duration of the storm (see response # 34). Rather than setting an arbitrary correlation threshold for the application of time lags, we will just state that time lags were applied to the radar data in the new section. The RMSE (and all other metrics) is calculated on the original time series, not the shifted ones. This shows how the sensors perform before applying lags which we think is more in line with the scope of the paper. This will be clarified in the results section.
  
  We agree and will incorporate the content of section 4.4 to section 3 instead.
  
  We think it is relevant to report which coefficients that were used to convert reflectivity to rain rate for reproducibility of the results and will therefore keep these lines in the text.
  
  The text will be updated accordingly.
  
  We agree that this is important to highlight this to understand the applicability of XWR for monitoring of convective storms. There are individual bins missing data around the intense part of the storm, not the full domain. Investigations suggest that this not due to a limitation of the sensor itself but imposed by post-processing in the pre-calculated rain rate that we used for the analysis. The data was compressed to 8-bit format and therefore any value above 255 (mm/h in this case) was lost. VeVa (see line 186) has since then changed the compression to 16-bit which allows for observations up to 650 mm/h without losing data, but the error is unfortunately present in 2022 data. We will elaborate on this in the paper.
  
  It is temporal linear interpolation, which will be clarified in the text.
  
  All XWR time series in the paper are sampled directly from the polar bins as described in lines 288 to 292. The data was resampled to a cartesian grid for the following reasons: 1) create the map in Figure 7, 2) to get a spatial overview of the total depth of the event and (importantly) 3) to allow for comparison with the CWR composite in Figure 5. We believe a resolution of 500 m serves these purposes. Although a higher spatial resolution of (for example) 250 m would have given even more detail it would also increase the difference even more to the CWR resolution. We will update the text to clarify this.
  
  We will move the literature review to the appendix and more content on the CML processing method will be moved from the appendix to this section.
  
  We will express the difference in percent instead. The difference is 3% (the total depth recorded by the sub-links was 57 and 60 mm, respectively).
  
  As outlined in response 26, all XWR time series in the paper are sampled from the polar bins as described in lines 288 to 292, not from the 500 m grid. That grid was created for the reasons outlined in response 26, mainly for visualization purposes in Fig 7.
  
  The text will be updated to reflect the comment.
  
  The table will be removed and the m_match and m_int values will be described in the text.
  
  The absolute value of the accumulated difference will be added to the table and all plots.
  
  The topic of the paper is not only extreme precipitation per se, but how the observations of such events differ between different types of sensors. Studying the duration of the same event observed by different sensors gives insights on 1) the sensitivity of the sensors 2) the temporal movement of the storm (which, as can be concluded from table 3, is from the south-west to the north-east in this case). In addition, we think it is an interesting finding that the CML and radars record similar durations. The duration also directly affects the calculation of return periods which we think is an important part of the discussion section. Furthermore, the evaluation metrics are computed for the duration of the event as recorded by each reference sensor (to exclude measurements of zero rainfall). For the given reasons, we suggest to keep this section in the paper.
  
  After checking the calculations of time lags in detail they were found to be incorrect and we have now corrected the calculations. Firstly, the event start at the municipal gauge was erroneously set to 17:15 in the calculations, when it in fact started at 17:05. As the metrics (and lags) are calculated for the duration of the municipal gauge when it serves as reference sensor, this will affect the calculation of time lags and metrics in Fig. 6 and Fig 8 as well. Secondly, the radar time series were mistakenly sliced to the event duration of the municipal gauge before shifting them in time, meaning that some iterations was based on only a few time steps, including the optimum value (25 minutes) that was based on only 5 timesteps. Instead, the correlation should of course be calculated over the event duration (17:05-18:10 at the municipal gauge) while shifting the tested time series step by step. After this correction, the Spearman correlation is 0.4 for CWR compared with reference without time lag, which could be increased to 0.83 when shifting the CWR time series 10 minutes. We find this to be reasonable values.
  
  RMSE will be expressed as mm/h in the figures.
  
  The text will be updated to mention 60 minutes duration only.
  
  We are confident that the difference in total accumulation is caused by beam blockage in this case, see responses to comment 14 and 47. The cause of differences in spatial structure is uncertain. The northeastern part of the study area has missing echo when comparing CWR with XWR – this might be a radar shadow behind the convective cell – but it is still very close to the radar location. Apart from this, it is hard to tell how much the two data sets should align given that they have different spatial resolution. It might be better to resample XWR to 2km to allow for more consistent visual comparison with CWR. If anything, these differences at least highlight the need for high-resolution data to observe convective storms.
  
  Following up on the response to comment #34, the same corrections were made for the XWR data. The Spearman correlation is 0.56 for XWR compared with reference without time lag, which could be increased to 0.7 when shifting the XWR time series 5 minutes. After these corrections we do not see a reason to believe that there is any significant difference between the clocks. Nevertheless, we will highlight the importance of accurate time recording in the discussion section.
  
  We will rephrase as follows: “In Fig. 8a, there is a tendency that XWR underestimated the overall peak rainfall intensity and overestimated lower rainfall intensities”. Note that we previously used the word “seems”, which we do not think indicate a strong statement. The overall peak 17:30-17:40 recorded by the municipal gauge is indeed underestimated by XWR, and if we use the municipal gauge as reference, then the peak recorded by XWR at 16:55 when the gauge reports nothing is by definition an overestimation – the same holds for 18:15-18:35. However, there are indeed a couple of timesteps 17:45-18:10 when XWR reports higher intensities than the gauge, hence we will use the word “tendency”.
  
  Recalculation of the Spearman rank gives the same result. It is correct that Spearman rank should not be applied for time series with many multiple values (ties) but we have applied ties-correction in the calculation (Emond et al., 2002). But as pointed out, it is probably also correct that the good metrics is a pure coincidence as the “plateau value” recorded by the CML happens to be in the middle range of the fluctuation of XWR observations, so the total accumulated depths are similar between CML and XWR reference. When computing the metrics for “non-plateau” values only the correlation is still 0.86 which suggest that CML and XWR record similar temporal variations during the storm. The absolute difference is then -15.5 mm, PBIAS 52 % and RMSE 4.1 mm/h for the non-plateau period, so a lot worse than with the “plateau-period” included. However, the non-plateau timesteps are outside the most intense part of the storm and therefore not the main focus of the analysis. We will keep the metrics as they are but use a more nuanced language, remove any formulations about that these metrics are “good” and discuss the fact that it is a coincidence in this case that the “plateau-intensity” leads to a similar total accumulation as the reference sensor.
  
  See response #26. The time series in Figure 8 and 9 are derived from XWR polar bins and Fig. 7 is gridded data which inevitably introduces averaging, especially for data with high spatial variability. This is explained in lines 455-457 but we will try to clarify this even further.
  
  Thanks for pointing this out, we had accidentally plotted instantaneous spread instead of cumulative spread. The figure will be corrected accordingly.
  
  We agree and will update the text accordingly.
  
  Thank you. This was actually included in an earlier version of the figure and will be added again.
  
  The value applied for alpha (for 23 GHz, vertical polarization) is 0.96, i.e. b = 1.04. We will refer to Leijnse et al., (2010) to support the findings.
  
  Thank you, this will be changed.
  
  See response to comment 14. Based on internal evaluations and experience with this particular radar, we strongly believe that the siting of the radar (and consequently vegetation beam blockage) is the most contributing factor to the severe underestimation. See response #14 for more details. We will add a clause on other possible sources of the error and refer to the suggested paper.
  
  Thank you for pointing this out, we will remove it as a possible cause for the low correlation with the municipal gauge.
  
  Thank you for pointing this out, we will remove this sentence.
  
  We agree and will update the discussion accordingly.
  
  There might be local attenuation in the data even after application of attenuation correction as all methods have limitations and uncertainties. We will add a clause like “… due to possible uncertainties in the attenuation correction” or the like.
  
  It is correct that the frequency of the XWR (9.4 GHz) is of course lower than the CML (23.1 GHz) so this statement will be removed. Our investigations show that the CWR indeed has trouble detecting light precipitation at this elevation angle (higher angles perform better). However, we think it is out of the scope of the paper to elaborate on this, so we will just delete this sentence.
  
  Thank you, wind drift will be removed as a possible cause for the observed discrepancy.
  
  The original values of the parameters m_match and m_intwere 200 and 4032, which means that 200 rainy time steps in the last 2 weeks (for 5-minute data, which is used here) are required to compute Station Outlier flags (as we know, correlations should not be calculated on dry time steps). With these settings, all stations were flagged as “-1” (“filter cannot be applied”) for the period of interest because there were not enough rainy timesteps in the last 2 weeks. This will be clarified in lines 379-380. Instead, we require 100 wet timesteps in the last four weeks to apply the filter. The result of this is that no timesteps are flagged as “-1” in the period of interest but the SO algorithm can be applied at all time steps. The quality control is then based on less data (100 wet timesteps instead of 200) but it is a trade-off between having any quality control performed at all, and having it performed based on less reference data. So, the failure of the SO-algorithm cannot be attributed to the change of parameter settings but rather to the spatial variability of the storm. We will incorporate these reflections in the discussion section.
  
  We will change vague references like “southern Sweden” and “southwestern Sweden” to the more well-defined area “Skåne county” in section 1, 2, 3.2 and 6. The word “excellent” in the text refers to the combination of XWR and rain gauges operated by local authorities, not the XWR coverage itself. There are two X-band radars in Skåne (and all of Sweden) – their coverage is shown in Figure 1 in Hosseini et al (2023). This is approximately 9000 km², corresponding to 80 % of Skåne county and 2 % of all of Sweden.
  
  The intention of this statement is to highlight that third-party sensors can assist in understanding the spatial variability of a storm, while they may not always provide accurate information on intensity/depth at all points of interest. This phrasing is already used in the conclusion section and will be added here for clarity.
  
  Section 7 will be amended to put more emphasis on the added value of incorporating these sensor types as complementary observations in the context of extreme rainfall.
  
  Thank you for noticing this, we will revise this part of the appendix and clarify the units. The unit of equation A2 was incorrect.
  
  References
  Emond, Edward J., and David W. Mason. "A new rank correlation coefficient with application to the consensus ranking problem." Journal of Multi‐Criteria Decision Analysis 11.1 (2002): 17-28
  Kumjian, M. R. (2013). Principles and applications of dual-polarization weather radar. Part I: Description of the polarimetric radar variables. Journal of Operational Meteorology, 1.
  Hosseini, S. H., Hashemi, H., Berndtsson, R., South, N., Aspegren, H., Larsson, R., Olsson, J., Persson, A., & Olsson, L. (2020). Evaluation of a new X-band weather radar for operational use in south Sweden. Water Science and Technology
  Hosseini, S. H., Hashemi, H., Larsson, R., & Berndtsson, R. (2023). Merging dual-polarization X-band radar network intelligence for improved microscale observation of summer rainfall in south Sweden. Journal of Hydrology, 617, 129090. https://doi.org/10.1016/j.jhydrol.2023.129090
  SMHI. (2020). Trädtoppar orsakar blockeringar för väderradar Ängelholm—SMHI. https://www.smhi.se/nyheter/nyheter/2020-11-20-tradtoppar-orsakar-blockeringar-for-vaderradar-angelholm
  de Vos, L. W., Leijnse, H., Overeem, A., & Uijlenhoet, R. (2019). Quality Control for Crowdsourced Personal Weather Stations to Enable Operational Rainfall Monitoring. Geophysical Research Letters, 46(15), 8820–8829. https://doi.org/10.1029/2019GL083731
  
  Citation: https://doi.org/10.5194/egusphere-2025-2820-AC2

Peer review completion

AR – Author's response | RR – Referee report | ED – Editor decision | EF – Editorial file upload

AR by Louise Petersson Wårdh on behalf of the Authors (21 Nov 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (23 Nov 2025) by Gianfranco Vulpiani

RR by Hidde Leijnse (04 Dec 2025)

ED: Publish subject to minor revisions (review by editor) (04 Dec 2025) by Gianfranco Vulpiani

AR by Louise Petersson Wårdh on behalf of the Authors (11 Dec 2025) Author's response Author's tracked changes Manuscript

ED: Publish as is (13 Dec 2025) by Gianfranco Vulpiani

AR by Louise Petersson Wårdh on behalf of the Authors (15 Dec 2025)

Journal article(s) based on this preprint

21 Jan 2026

Have you ever seen the rain? Observing a record convective rainfall with national and local monitoring networks and opportunistic sensors

Louise Petersson Wårdh, Hasan Hosseini, Remco van de Beek, Jafet C. M. Andersson, Hossein Hashemi, and Jonas Olsson

Atmos. Meas. Tech., 19, 461–483, https://doi.org/10.5194/amt-19-461-2026,https://doi.org/10.5194/amt-19-461-2026, 2026

Short summary

Louise Petersson Wårdh, Hasan Hosseini, Remco van de Beek, Jafet C. M. Andersson, Hossein Hashemi, and Jonas Olsson

Viewed

Total article views: 1,074 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
878	156	40	1,074	26	45

HTML: 878
PDF: 156
XML: 40
Total: 1,074
BibTeX: 26
EndNote: 45

Views and downloads (calculated since 30 Jun 2025)

Month	HTML	PDF	XML	Total
Jun 2025	27	2	0	29
Jul 2025	104	23	8	135
Aug 2025	85	11	4	100
Sep 2025	416	15	6	437
Oct 2025	70	23	8	101
Nov 2025	51	28	7	86
Dec 2025	59	35	6	100
Jan 2026	66	19	1	86

Cumulative views and downloads (calculated since 30 Jun 2025)

Month	HTML	PDF	XML	Total
Jun 2025	27	2	0	29
Jul 2025	104	23	8	135
Aug 2025	85	11	4	100
Sep 2025	416	15	6	437
Oct 2025	70	23	8	101
Nov 2025	51	28	7	86
Dec 2025	59	35	6	100
Jan 2026	66	19	1	86

Viewed (geographical distribution)

Total article views: 1,053 (including HTML, PDF, and XML) Thereof 1,053 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 21 Jan 2026

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (1880 KB)
Metadata XML

Short summary

Extreme rainfall can cause severe damage, especially in cities. However, national meteorological institutes have difficulties to observe such events. In this study we show that rainfall observations collected by local actors, such as municipalities and even citizens, can contribute to better rainfall observations. Sweden’s official monitoring network could not capture the event under study, whereas the complementary sensors contributed to a better understanding of the magnitude of the event.


Total:	0
HTML:	0
PDF:	0
XML:	0

Have you ever seen the rain? Observing a record convective rainfall with national and local monitoring networks and opportunistic sensors

Journal article(s) based on this preprint

Interactive discussion

Interactive discussion

Peer review completion

Suggestions for revision or reasons for rejection

Journal article(s) based on this preprint

Viewed

Viewed (geographical distribution)