the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 30 Jun 2025
Have you ever seen the rain? Observing a record convective rainfall with national and local monitoring networks and opportunistic sensors
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.
- Preprint
(1880 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-2820', Anonymous Referee #1, 22 Jul 2025
-
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
-
AC1: 'Reply on RC1', Louise Petersson Wårdh, 21 Aug 2025
-
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 mmatch and mint(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 10th and 90th 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 10thand 90th 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
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 648 | 53 | 18 | 719 | 13 | 24 |
- HTML: 648
- PDF: 53
- XML: 18
- Total: 719
- BibTeX: 13
- EndNote: 24
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
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 rs 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.