Frequency Trend Analysis of Heavy Rainfall Days for Germany

Round 1

Reviewer 1 Report

The manuscript is improved significantly according to the review reports, and uncertainty analysis of the data is performed in Section 2.2. However, few questions are still raised after reviewing the manuscript.

1. I found that Fig. 7 in this version and Fig. 5 in last version (April, 2020) is different in spatial distribution of trends while Fig. 6 in this version and Fig. 4 in last version is the same. Please make sure the correctness of data processing.
2. The resubmitted manuscript is added more uncertainty analysis of the data. Please add a research flowchart from data preprocessing to trend analysis.
3. L387-L398, authors mentioned the application of the nationwide rain radar to identify heavy-rainfall events. However, the radar products also have their problems, like the uncertainty of Z-R relation, effects of terrain shadowing and radar sensitivity limitations. Be careful to apply radar data.
4. Figure 15 shows that the extreme rainfall of Dedelow is reasonable based on the observation of radars. In early years, would the extreme rainfall data be deleted as a outliner by DWD? The data cleaning would influence long-term trend of rainfalls. please check data again.

Author Response

Reviewer 1

We appreciate your helpful comments.

1. I found that Fig. 7 in this version and Fig. 5 in last version (April, 2020) is different in spatial distribution of trends while Fig. 6 in this version and Fig. 4 in last version is the same. Please make sure the correctness of data processing.

Response 1: The previous Fig. 5 showed the regional majority of trend directions whereas the current Fig. 7 shows the regional trend mean (“For each region and CLINO period, the average of Kendall’s tau was calculated…”, lines 163-164). They both use the same database.

2. The resubmitted manuscript is added more uncertainty analysis of the data. Please add a research flowchart from data preprocessing to trend analysis.

Response 2: Thank you for the suggestion. We added a flow chart as new Figure A1 in Appendix A (line 461) and refer to it in line 90.

3. L387-L398, authors mentioned the application of the nationwide rain radar to identify heavy-rainfall events. However, the radar products also have their problems, like the uncertainty of Z-R relation, effects of terrain shadowing and radar sensitivity limitations. Be careful to apply radar data.

Response 3: We agree with you that each method has its own uncertainties. Nonetheless, it is the only data product which covers all of Germany which is a huge step forward, e.g. for erosion studies, and this is what we wanted to highlight.

4. Figure 15 shows that the extreme rainfall of Dedelow is reasonable based on the observation of radars. In early years, would the extreme rainfall data be deleted as a outliner by DWD? The data cleaning would influence long-term trend of rainfalls. please check data again.

Response 4: True, there is an unknown impact of data revision and post-processing by the Meteorological Service which has continuously been adjusted along the technical development. From our own experience, sub-daily and daily data partly differ which has to be considered as part of the QA of the data provider (e.g. to remove spurious signals). How this affects trends is beyond the scope of our study which is based on published data which passed some QA. Such a study would require an in-depth comparison of alternative approaches to correct raw data.

Please note that radar data is not used to confirm station data. In contrast, station data is used to calibrate radar data. As Dedelow is not run by DWD, this station is not used for calibration. Instead it could be used as ground truth for the gridded radar data.

Author Response File: Author Response.docx

Reviewer 2 Report

The reviewer highly commends the authors as there is little to no grammatical error found in the paper. However, there are some points that the authors need to address:

1. The title can be revised to “Frequency Trend Analysis of Heavy Rainfall Days for Germany”
2. Line 416-418 “The direction and strength of multi-decadal trends of heavy-rainfall days however vary in space and time. After 1951, stable positive trends occurred in southern and parts of northern Germany, but stable negative trends in Central Germany.”

-What do you think is the reason behind these two opposing trends? Is the Central Germany and Northern Germany belong to different zone types? What other factors might have affected the trend behavior of the two? These things should be properly discussed in the paper.

3. Line 419-420 “Despite the frequent changes in the location of stations and in the reference time for daily sums, the trends are reliable.”

-How did you say so? What made it reliable? There should be a validating/verification test.

4. It is stated in Line 224-225 that “..individual stations, however, location shifts can affect both the strength and the direction of trends. The impact was spatially highly variable and lacked a clear pattern..”

-Since the previous and the current locations are known, why not get the relationship between the two (something line geographic weighted regression or the like?) Are the authors certain that a change in location has no effect?

5. Line 344 also states that “Compared to the distances between rainfall stations, these location shifts are normally small.”

-How small are these shifts? In what range?

6. Line 194-195 “..result of an increase in summer and partly in winter..”

-increase of what? Please specify.

7. Line 198-200 “..shift from more negative trends to more positive trends in summer (Figure 3a), and an inverse shift in winter (Figure 3b) resulting in the almost balanced annual trends. In general, the trends are more variable for the 10-mm threshold than for the higher thresholds.”

-What do you think is the reason behind the variability in 10mm? how about those lower than 10mm?

8. Please elaborate the conclusion part. 

Author Response

Reviewer 2

Thank you for your critical and supportive comments.

1. The title can be revised to “Frequency Trend Analysis of Heavy Rainfall Days for Germany”

Response 1: Good suggestion. We agreed and changed the title accordingly.

2. Line 416-418 “The direction and strength of multi-decadal trends of heavy-rainfall days however vary in space and time. After 1951, stable positive trends occurred in southern and parts of northern Germany, but stable negative trends in Central Germany”.

-   What do you think is the reason behind these two opposing trends? Is the Central Germany and Northern Germany belong to different zone types? What other factors might have affected the trend behavior of the two? These things should be properly discussed in the paper.

Response 2: The complex causes of climate oscillations and changes in extreme rainfall at different time-scales are beyond the scope of our paper. They have been discussed in the context of large-scale variations in teleconnection pattern, specifically the North-Atlantic Oscillation (NAO), the El Niño/Southern Oscillation (ENSO), and the Atlantic Multidecadal Oscillation (AMO) (Casanueva et al. 2014, López-Parages et al. 2014, Zolina et al. 2008). So, different changes in such pattern may in general explain the spatial and seasonal trend changes.

Łupikasza et al. (2011) point out that the large-scale atmospheric patterns are modified by local conditions (e.g. wind exposure, precipitation shadow) resulting in different trends, e.g. in eastern Germany and southern Poland. However, the large spatial trend variability of heavy-rainfall indices in the Czech Republic could not be clearly related to location and elevation (Hanel and Kyselý 2015).

So, more research is needed to assess to which degree these drivers affect different parts of Germany. We changed the final sentence in the Conclusions accordingly: “Future research should further explore how data inconsistencies affect trends of (different) rainfall indices in different regions and assess the drivers of trend variability.” (lines 432-434)

3. Line 419-420 Despite the frequent changes in the location of stations and in the reference time for daily sums, the trends are reliable.

-How did you say so? What made it reliable? There should be a validating/verification test.

Response 3: We derived this conclusion from our statistical analyses which are (and can only be) an approximation as for each station only one time-series exists and high-resolution data is increasingly available since mid-1990s. It is noteworthy that this statement is not generally valid for individual stations, so we changed the conclusion to “…the trends can be considered reliable for regional to national studies.” (line 425)

4. It is stated in Line 224-225 that ”…individual stations, however, location shifts can affect both the strength and the direction of trends. The impact was spatially highly variable and lacked a clear pattern…”

-    Since the previous and the current locations are known, why not get the relationship between the two (something line geographic weighted regression or the like?) Are the authors certain that a change in location has no effect?

Response 4: The reason is that the time-series do not overlap. For each station, we only have a single time-series. Therefore, we estimated the effect on trends by comparing nearby stations which are further away than the typical location change. As we state: our results showed that location changes can influence trends, even trend directions. So, location change must definitely be taken into consideration in small-scale trend studies. However, for the whole or parts of Germany, the trend distribution was not significantly affected (cf. Response 3).

5. Line 344 also states that ”Compared to the distances between rainfall stations, these location shifts are normally small.”

-     How small are these shifts? In what range?

Response 5: In general, site changes are spatially limited as e.g. the WMO regulations must be observed. We provide average numbers in Table 1 and line 139 (“As the average location shift is smaller than the average difference to neighboring stations (>11000 m)”. Of course, the shifts and distances between stations vary in space and time.

6. Line 194-195 ”…result of an increase in summer and partly in winter…”

-increase of what? Please specify.

Response 6: It refers to “annual increase of heavy-rainfall days” at the beginning of the sentence. To keep the sentence short, we replaced the indefinite article: “…between 1971 and 2000 is the result of their increase in summer and partly in winter.” (lines 199-200)

7. Line 198-200 ”…shift from more negative trends to more positive trends in summer (Figure 3a), and an inverse shift in winter (Figure 3b) resulting in the almost balanced annual trends. In general, the trends are more variable for the 10-mm threshold than for the higher thresholds.”

- What do you think is the reason behind the variability in 10 mm? how about those lower than 10mm?

Response 7: As we have not examined these questions, we can only speculate that the 10 mm threshold is often just exceeded. Our analyses do not allow conclusions for days with less than 10 mm. Moberg and Jones (2005) found both an increase in moderately heavy-rainfall days as well as (albeit insignificantly) in the length of dry spells. Parallel trends in wet and dry spells were found e.g. for the Netherlands by Zolina et al. (2013). In-depth analyses of changes in atmospheric circulation pattern and weather conditions would be required to discuss these questions properly which was beyond the scope of our study. So, we only refer briefly to the general physics of heavy precipitation in lines 375-381 and address the need for future research in the Conclusions (cf. Response 2).

8. Please elaborate the conclusion part.

Response: We adjusted parts of the conclusion, cf. Responses 2 and 3.

Author Response File: Author Response.docx

Reviewer 3 Report

Paper presents variability and trends of extreme precipitations in Germany. Authors collected very high number of daily precipitation records from the period 1951-2019 and some longer. Rainfall erosivity was analysed using 1-10 minute data from the period 1955-2019 from only one station. It is an interesting paper, worth publishing. However, there are some shortcomings which should be removed before publishing.

line 91: It is: “The period 1901–2019 was sub-divided into ten periods of 30 years, overlapping by 10 years” These 10 periods overlap 20 years and are shifted by 10 years.

lines 115-116: It is: “Therefore, we applied the non-parametric Mann-Whitney test to all combinations (n=18)”, What 18 combinations?

line 181: It is: “the peak intensity in 30 minutes” Do you mean the highest average 30-minutes intensity?

lines 196-197: It is: “Over the last seven decades, there has been a slight shift from more negative trends to more positive trends in summer (Figure 3a)”, however in the last CLINO period positive and negative trends were balanced.

lines 289-294: It is: “In contrast to the negligible trend in heavy rainfall days – independent of the reference time - (Figure 13), the magnitude of rainfall events increased more significantly (Figure 14). The moving average of EI30 changed from 45 N h-1 (1955-1975) to 65 N h-1 (1955-2019). The strongest change occurred after 1990 (red trend line) with an average value of 80 N h-1 for the last 30 years. The annual values generally ranged widely, from about 11 to >300 N h-1. However, the three highest values happened after 2000.” What is the averaging period? The 21-year average (1955-197) and 65-year average (1955-2019) should not be compared. An average value of 80 N h-1 for the last 30 years is not change that occurred after 1990.

fig.2 & 3 For the 30 mm threshold, the median and quartiles are not visible on a black background.

fig 7. 1951-2019, 1951-2010, 1961-2010 are not CLINO periods

fig. 8-12 the moving average on the graphs should be marked in the middle year of the period, not the end year

fig 12 x is missing in the equation for the blue line.

fig14. What does moving average mean in this case? What is the averaging period?

Author Response

Reviewer 3

Thank you for the meticulous review.

1. line 91: It is: ”The period 1901-2019 was sub-divided into ten periods of 30 years, overlapping by 10 years“ These 10 periods overlap 20 years and are shifted by 10 years.

Response 1: Thank you. The sentence was changed to “…ten periods of 30 years. These overlapping periods are shifted by 10 years. They correspond…” [lines 91-92].

2. lines 115-116: It is: ”Therefore, we applied the non-parametric Mann-Whitney test to all combinations (n=18)” What 18 combinations?

Response 2: Thank you for spotting the error left from an earlier state of our study. We have 10 CLINO periods and three thresholds, i.e. 30 combinations. This was corrected to “…Mann-Whitney test to all 30 combinations of thresholds and CLINO periods in order…” [line 117-118]. Likewise, we changed line 216 to “The majority of the Mann-Whitney tests, 24 out of 30, implied only…”

3. line 181: It is: ”the peak intensity in 30 minutes” Do you mean the highest average 30-minutes intensity?

Response 3:   Yes. We use now the original wording by Wischmeier USDA-Handbook 282: “…the maximum 30-min intensity…” (line 182).

4. lines 196-197: It is: “Over the last seven decades, there has been a slight shift from more negative trends to more positive trends in summer (Figure 3a)”, however in the last CLINO period positive and negative trends were balanced.

Response 4:  True. The currently balanced trend distribution means that the number of heavy rainfall days remains at the elevated level reached during the last decades. So, we added the sentence “However, the balanced summer trends for the current CLINO period show that the number of heavy rainfall days does not further increase for the whole of Germany.” [lines 202-205]

5. lines 289-294: It is: “In contrast to the negligible trend in heavy rainfall days independent of the reference time - (Figure 13), the magnitude of rainfall events increased more significantly (Figure 14). The moving average of EI30 changed from 45 N h-1 (1955-1975) to 65 N h-1 (1955-2019). The strongest change occurred after 1990 (red trend line) with an average value of 80 N h-1 for the last 30 years. The annual values generally ranged widely, from about 11 to >300 N h-1. However, the three highest values happened after 2000.” What is the averaging period? The 21-year average (1955-197) and 65-year average (1955-2019) should not be compared. An average value of 80 N h-1 for the last 30 years is not change that occurred after 1990.

Response 5: We clarified this in the description of Eq. 4 lines 186-190. We replaced the term “moving average” by “long-term average”. Such multi-decadal average annual EI30 are used in soil erosion modelling (”For the erosivity (EI₃₀), the starting year is fixed (m=1) to show the long-term average as used in the USLE/DIN 19708.”, lines 189-190). The period length for averaging EI30 is not fixed, so we rephrased the comparison: “The long-term average of EI₃₀ changed from 45 N h^-1 (mid-1970s) to 65 N h^-1 (2019).” (lines 296-297), i.e. if someone used a long-term average value this would be around 45 N h^-1 while today, it might be 65 N h^-1 or, considering only the last 3 decades, 80 N h^-1. The last value is a remarkable increase of the soil erosion risk compared to previous periods. The linear slope after 1990 is also steeper than the slope over the whole measurement period (Fig. 14).

6. fig.2 & 3 For the 30 mm threshold, the median and quartiles are not visible on a black background.

Response 6: We followed your suggestion and made the colours lighter to improve readability, including in (new) Figure A3 (previously A2) in Appendix C.

7. fig 7. 1951-2019, 1951-2010, 1961-2010 are not CLINO periods

Response 7:  Correct. The periods were indeed derived from multiple CLINO periods. We removed the misleading “CLINO” in the figure caption.

8. fig. 8-12 the moving average on the graphs should be marked in the middle year of the period, not the end year

Response 8: We used here the Excel functionality only to visualize the changes over 30 years (and 5 years for comparison). Therefore, we prefer to keep it as is.

9. fig 12 x is missing in the equation for the blue line.

Response 9: Thank you, the equation was corrected.

10. fig14. What does moving average mean in this case? What is the averaging period?

Response 10:  The term was misleading and was replaced by “long term average” (lines 296-297, cf. Response 5). It refers to Eq. 4 (cf. lines 186-190).

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

This study applied three daily-rainfall thresholds to multi-decadal data on daily precipitaton of the DWD for analyzing the spatial and temporal variability of changes in the occurrence of heavy rainfall. It is a cricual issue to climate changes. However, some main questions are raised after reviewing the manuscript.

1. This study focused on the trends of extreme rainfall days in Germany. Therefore, some papers related to the trend analysis of extreme rainfalls in Europe should be reviewed in the Introduction, like "Review of trend analysis and climate change projections of extreme precipitation and flood s in Europe," (Madsen et al., 2014)
2. This study used very long-term data, multi-decadal data, to analyze the trends of heavy rainfall days. Therefore, QA of the data is the key issue highly related to the trend analysis. Some description about the data QA should be added in the Materials, including following questions.
   1. Is the data addendum performed by the DWD? Because data addendum is hard to re-realize the extreme value, the data addendum would affect the trend analysis.
   2. How is the measurement quality of each station? Because some measurement would be upgraded during the very long-term period, the same value of a station in different periods has different uncertainty or bias, especially for the lowland station Lindenberg in which the number of heavy rainfall days is small.
   3. The uncertainty of the data in different periods could be added in figures.
3. The extreme rainfall evets are highly related to climate oscillation, and these fluctuations may be quasiperiodic, usually occurring within seasonal, interannual, multiple years, inter-decadal, multi-decadal, century, or longer time scales. The periods of moving average for trend analysis are 5 and 30 years in this study. Why did the study select these two periods? More explaination should be added for enhancing the links between resutls of trend analysis and the climate oscillation.
4. In the Methods, authors should explain how to compute the moving average of rain erosivity in Figures 11, 12, and 13.

Reviewer 2 Report

Review to water 778072

The paper deals with the computation of trends in the number of days of high rainfall in Germany, relatively to historical rainfall records. At the same time the authors present a short analysis of trends in rainfall erosivity.

In general I see the following general issues:

1. The material presented is generally confusing, and I think some points of the methodology and the analysis present relevant flaws (see specific comments).
2. Most of the figures and tables have not been prepared with the needed care, and are difficult to read.
3. The analysis of erosivity is quite short and preliminary and I do not think that paper benefits from it, so it just may be omitted
4. English needs improvement in several points.

In conclusion, the material presented is not mature. Hence I regret to write that in the present form I suggest rejection for the manuscript. Perhaps with more work and a better presentation, it may be resubmitted and reconsidered. I hope that the authors find the above and the below comments useful for this purpose.

Specific comments 

- L 78-79 the inconsistencies related to the time window on which daily rainfall is computed can have non negligible effects on the computed trends. The authors should prove that this is not the case. This can be done using some high temporal rainfall recent records and comparing rainfall of a given station aggregated on the two time windows.

- L 105-106 this sentence is very unclear. In general it is unclear which are the final data that the authors analyze. Which raingauges? Which analysis are conducted with reference to the 1901-present period and which to 1951-present period?

- Fig.1 must be improved. For instance: enlarge the map and indicate the name of some of the pluviometers/ombrometers (for instance the three for a more detailed analysis) and show terrain elevation from a digital terrain model

- L 120: why there is no call to Fig. A2 for the Naturraume?

- Sect 2.3 The part on erosivity is too short and seems of no use for the manuscript

- Fig. 4 is difficult to read. Could not get what the legend is indicating

- The whole methodology section is unclear and needs a thorough revision.

- Fig. 5. Why do periods change in the legend? It is unclear why results refer to different periods and in which sense

- L 183: To search trends when only few (2-4) events occur per year, makes no sense from a statistical point of view. This should be fixed: perhaps just exclude raingauges where the number of events is to low. Or define a threshold of rainfall that varies from station to station (for instance a given level of non-exceedance probability), and not just fixed for all locations to 10 / 20 or 30 mm/day.

- Fig.  6 – 11 This plots are difficult to read: too many lines. Reconsider presenting the results in a better way

- Fig. 7: seems there is “a shift” rather then a “linear trend”. Could be something happening to the station (Change in the location of the pluviometer). Check

- Figures and table in the appendix need to be prepared with more care.

- Some points in the results  and the conclusions are not supported by the results:  L203-204, L 306-307 and

Minor points/Technical corrections

- Rainfall is uncountable, so the use of “rainfalls” is uncorrect. Perhaps change to “rainfall events” or “rain days” in the several points where “rainfalls” is used.  

- Authors use several times the word “exemplarily” in an appropriate manner. They mean that they just present a short “example” of a possibly wider analysis. But the meaning of “exemplarily” has little to do with that.

- In the annotated pdf the authors can find some other technical corrections which I suggest to apply to improve the paper

Comments for author File: Comments.pdf