Verifying the relationships among the variabilities of summer rainfall extremes over Japan in the d4PDF climate ensemble, Pacific sea surface temperature, and monsoon activity

Lee, Shao-Yi; He, Sicheng; Takemi, Tetsuya

doi:https://doi.org/10.5194/nhess-25-2225-2025

Articles | Volume 25, issue 7

https://doi.org/10.5194/nhess-25-2225-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/nhess-25-2225-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 25, issue 7

Research article

|

08 Jul 2025

Research article |

| 08 Jul 2025

Verifying the relationships among the variabilities of summer rainfall extremes over Japan in the d4PDF climate ensemble, Pacific sea surface temperature, and monsoon activity

Shao-Yi Lee, Sicheng He, and Tetsuya Takemi

Download

Final revised paper (published on 08 Jul 2025)
Preprint (discussion started on 04 Jun 2024)
Supplement to the preprint

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2024-1304', Anonymous Referee #1, 12 Jul 2024

This manuscript investigates the interannual variations of the observed precipitation extremes over the western Japan during June and July with the three datasets, i.e., the Radar-AMeDAS during 2006–2022, the 54 rain gauge data during 1952–2022, and the 5km mesh 10-member RCM simulations covering the 59-year period 1952–2010. Differences in the data periods result in an interpretation of the results being difficult and ambiguous. Since the correlation coefficients are discussed and not climate values, data from the same time period should be used. Furthermore, the short 17-year time period of the Radar-AMeDAS DATA is questionable in discussing clustering calculations and inter-annual variations. There is no rationale for comparing Radar-AMeDAS and RCM simulations because the model is forced with the observed SSTs, and therefore should be compared to data between the same periods. However, in this case, the common period is only 5 years, 2006-2010. Therefore, clustering and subsequent analysis should be performed on the rain-gauge data for the period 1952-2010, the same period as the d4PDF.
I recommend to re-submit the manuscript. One choice would limit the analysis to an evaluation of model performance by comparing long-term rain-gauge and d4PDF simulations.

Citation: https://doi.org/10.5194/egusphere-2024-1304-RC1
- AC1: 'Reply on RC1', Shao-Yi Lee, 22 Aug 2024
  
  Dear reviewer, thank you for your comments. If the manuscript is sent to revision, it may take another 6 weeks, but we would like to share our progress with you.
  We agree with your comment and indeed find it difficult to interpret our results due to the differences in time period. Radar-AMeDAS was used despite its short 17-year period because of its high-resolution spatial coverage, but as you have pointed out there is only a common period of 5 years. As you have suggested, we have performed a clustering on rain-gauges for the 1952-2010 period, identical to the simulation period. Furthermore, this was done for the ~130 rain-gauges over a larger region over Japan, compared to ~50 in western Japan previously. Below is the result for 99th upper percentile hourly rainfall, and a minimum cluster size setting of 3. This resulted in 16 classes, but also quite a number of unclassified rain-gauges. (Due to the image limitation, the quality is not so good.) From here, we should be able to compare identical regions in the simulations compared to the regions covered by the rain-gauges.
  
  Citation: https://doi.org/10.5194/egusphere-2024-1304-AC1
RC2:
'Comment on egusphere-2024-1304', Anonymous Referee #2, 01 Aug 2024

This paper investigates relationships between precipitation extremes over western Japan and major SST modes over the Pacific based on observational data, and it tries to explain the relationships through the modulation of monsoon activity. Then, the paper verifies the representation of the obtained relationships in d4PDF data. I found it difficult to evaluate this paper due to several issues listed below during the review. Therefore, I would like to reassess the significance of this paper after the authors have addressed these issues.
First, this paper seems to have incorrect notations in the figure numbers and similar references listed below. Authors should carefully proofread the manuscript before submitting it.

- Lines 350: Supp. Figs S3-3a -> Supp. Figs S1-3a ?

- Lines 351-355: Similar mistakes as above.

- Supplementary Material Section 1: Supplementary Figure S2-0 -> Supplementary Figure S1-0 ?

- Supplementary Material Section 1: Similar mistakes in other figure captions in this section.

- The caption of Supplementary Table S2-1: “Meteorological stations used, in three columns, with names in English and Japanese. Years listed with the shaded station are those with insufficient data and not considered.” It seems that this explanation does not fully correspond with Table S2-1.

- Supplementary Material Section 2: Figure S2-2 -> Supplementary Figure S2-1 ?

- Supplementary Material Section 4: Table S3-1 -> Supplementary Figure S4-1 ?

- Supplementary Material Section 4: Placing Figure S4-5 just below Figure S4-4 would be better.

- Supplementary Table S6-1a, S6-1b, S6-2a, and S6-2b: What does “ENSO-NC” mean in the rightmost column? There is no explanation of this term in the table caption and the manuscript.

- Line 9 in the caption of Supplementary Table S6-2a: 99.9th percentile hourly rainfall -> 99th percentile daily rainfall

- Line 174: dJF -> DJF

- Line 175: Djf -> DJF
Second, I think Table 2 is the most crucial result in this paper; however, I could not understand what type of observational data the presented results are based on and which period they covered. Please note this information in the table caption. In addition, I could not understand the results easily because the results of observation and d4PDF are displayed in layers with complex notations. I would like to ask the authors to present the observations and d4PDF separately.
Third, I am concerned about the difference in the periods of the Radar-AMeDAS data and the d4PDF data in interpreting the results. Since the overlapping period between the Radar-AMeDAS data and the d4PDF data is short, it would be better to focus on comparing the ground observation data with the d4PDF data. There is no problem with using the Radar-AMeDAS data as supplementary data. Another choice is that using AMeDAS data would provide more spatially dense observational information since the late 1970s.
Fourth, in this paper, many figures and tables are presented in the supplementary section and are cited in the main text. However, I do not believe all these figures and tables are necessary to reach the paper's conclusions. Presenting numerous results with little significance only wastes the reader's time. Please carefully select the figures and tables to be included in the paper. Associated with this point, it seems that the analysis of Ph99.9 and Pd99 has a minor role in this paper, so I think this analysis could be omitted.
[Other comments]

1. In this paper, the SST mode four is interpreted as the Pacific Decadal Variability mode. Please show　a temporal correlation between this mode and a well-known climate index, such as the IPO or PDO index, which would be available on a website.

2. Line 533: observations -> JRA55

Citation: https://doi.org/10.5194/egusphere-2024-1304-RC2
- AC2: 'Reply on RC2', Shao-Yi Lee, 22 Aug 2024
  
  Dear reviewer, thank you for your comments. If the manuscript is sent for revision, it will be another 6 weeks, so we would like to update you on the current progress.
  First of all, we would like to apologise for incorrect references/notations and general poor readability of the manuscript. It went through 3 major rewritings and the writing has become confused, so we will try to correct the errors and improve its readability.
  A large part of the confusing presentation may be trying to compare d4PDF with both radar and rain-gauges. As you have pointed out, the overlapping period of radar-AMeDAS and d4PDF is short. This issue was also raised by reviewer 1. We will focus on comparing rain-gauges with d4PDF for identical periods, and this should improve the interpretability of the results.
  After excluding radar-AMeDAS, then improving the d4PDF data-processing method (described later), it is possible to cluster 5km d4PDF over a larger area of Japan in one step. (Previously, the raw RCM data was loaded into GrADs which interpolated it onto a lon-lat grid, which resulted in a larger volume of data. Now, the grid-points from the smaller native RCM grid are used. The grid-points inside a convex hull over major Japanese islands are cropped, then any significant correlation with nearby points compressed as a sparse matrix.) The results of clustering the 99th upper percentile hourly rainfall are shown below. This can be compared with the results based on rain-gauges. In both figures, the minimum cluster size is set to 3. The interannual variability of rainfall can be compared between d4PDF and rain-gauges by selecting points covered by the rain-gauge clusters.
  
  With regards to whether SST 4 can be interpreted as PDO, we have calculated a time series of PDO index (https://www.ncei.noaa.gov/access/monitoring/pdo/) for 1952-2010. The mean of 5 seasons centred on JJA of the year was taken, using a similar treatment to SST, where each PCA sample was 5 seasons concatenated. Below is the comparsion of the two. Black is the PDO index. Red is standardised score of SST mode 4. The Spearman correlation between the two is 0.55, statistically significant at alpha=5%. It is not an extremely strong correlation but better than the correlation between PDO index and other 3 modes, which are of magnitude 0.31 and not statistically significant. There seems to be a lot of higher frequency signal still in SST mode 4, although the lower frequency seems to generally match the PDO index. This may be why the correlation is only 0.55. We can also do a direct correlation between the rainfall extremes and the PDO index.
  
  Citation: https://doi.org/10.5194/egusphere-2024-1304-AC2

Peer review completion

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

ED: Reconsider after major revisions (further review by editor and referees) (02 Sep 2024) by Ricardo Trigo

AR by Shao-Yi Lee on behalf of the Authors (11 Oct 2024) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (22 Oct 2024) by Ricardo Trigo

RR by Anonymous Referee #2 (20 Nov 2024)

RR by Anonymous Referee #3 (07 Jan 2025)

Suggestions for revision or reasons for rejection

Review of the manuscript “Verifying the relationships among the variabilities of summer rainfall extremes over Japan in the d4PDF climate ensemble, Pacific Sea surface temperature, and monsoon activity” by Shao-Yi Lee, Sicheng He and Tetsuya Takemi.

The authors have hypothesized an eventual relationship between Summer (June-July) monsoon rainfall extremes in Japan (at the hour, day and pentad timescales) and some North-Pacific SST seasonal anomaly modes (ENSO; PDO, pseudo-trend), which has been shown to be non-existent or non-statistically significant. This is concomitant, as well, with unclear or non-existent relationship (according to authors and literature) between ENSO and average monsoon rain in Japan. To represent extremes, the authors have used both rain-gauges and the dynamically downscaled D4PDF dataset. In order to aggregate stations and model grid-points, a rather complex cluster analysis is performed, though not using standard geostatistical techniques. The representativeness of the D4PDF dataset to simulate rainfall extremes is somehow ambiguous. For instance, no systematic comparison is made between the distribution of observed extremes and the distribution of simulated extremes. It was done only indirectly through the comparison between extreme rainfall values and climatic indices.
Authors have performed a lot of work, trying many methodological possibilities (e.g. clustering) to present the inexistence of direct control of the extremes by the analyzed climatic indices.
Despite that non-result, the authors have explored (only at the discussion section) the relationship between rainfall extremes and indices of the summer monsoon (Baiu). Moreover, the modulation of monsson indice statistics (average and standard deviation) by the SST Pacific modes has been studied. Those relationships seem robust, new and interesting.
Giving the above considerations, the manuscript can be publishable after concretizing two main tasks (major points):
1) Rewrite some parts of the main manuscript.
2) Explore further the results linking extremes with monsoon and Pacific indices providing physical reasons.

Below are presented the ordered list of corrections asked:

1) Line 15. Are the modes sorted by decreasing explained variance? Clarify. In line 18 ‘higher modes’ refer to the previous order of modes?

2) Line 17-20. Rewrite the phrase of lines 17-20 in a much clear way by splitting it into two sentences.

3) Line 95 Clarify the simulation period of the ‘100-member historical-warming (HPB) climate ensemble’ as well as the temporal and spatial resolution of the referred GCM.

4) Lines 99-100. No systematic comparison is made between the distribution of observed extremes and the distribution of simulated extremes by the 4DPDF dataset. A synthetic study of the representativeness of 4DPDF extremes shall be made.

5) Line 142 and wherever needed: Change ‘Principle Component Analysis’ to ‘Principal Component Analysis’, throughout all the manuscript.

6) Line 144. The linear long trend of SST was not removed to get the anomalous SST. It was obtained farther as a mixed mode (TREND+) or pseudo-trend. Authors could be more direct by correlating the linear SST trend (for instance averaged throughout all Japan) with extremes. Comment on that.

7) Line 146 Authors have performed extended-PCA (extended Principal Component Analysis) and not a simple PCA. This must be referred explicitly and for clarity, since an extended vector merging 5 delayed (5 trimesters centered in JJA) of spatially distributed values have been taken, from which the covariance and its eigen-decomposition was computed. This is like a MSSA (Multi Singular Spectrum Analysis) with embedding dimension 5 and trimestral sampling.

8) Line 168 The spatial clustering of temporal extremes, particularly the rainfall (e.g. Ma et al. 2020 and references therein) has been studied by different authors, namely by using geostatistical techniques. The reference to those works must be included in the manuscript. The considered metric (distance) to cluster rain-gauges and grid points, used in cluster analysis HDBSCAN depends uniquely on the temporal similarity (Spearman rank correlation) between time-series, being equivalent to the F-madogram. However, for spatially distributed data, a geometrical term, weighing the point-wise distance (e.g. Euclidean) must be added to the statistical distance. The omission of the geometrical term leads to fragmented, topologically complex clusters (even not simply connected, i.e. with ‘holes’). This apparently was remedied by adopting ad-hoc clustering rules. The authors are asked to comment on that.

Yingzhao Ma, Mengqian Lu, Cameron Bracken, Haonan Chen, 2020. Spatially coherent clusters of summer precipitation extremes in the Tibetan Plateau: Where is the moisture from?, Atmospheric Research, Volume 237, 104841, ISSN 0169-8095,https://doi.org/10.1016/j.atmosres.2020.104841.

9) Line 175. Authors use the procedure ‘percentiles were calculated for each individual ensemble member, then the ensemble mean was taken’. This is an alternative to take quantiles of a super-sample (collecting all sub-samples). Please comment on that.

10) Line 184 Say here the value of CC significant at alpha=0.05.

11) Line 202-205. Authors say: ‘At each timestep, the maximum rainfall in each set was selected. For example, in a cluster of three points, this would be the point with the highest rainfall and the other two points would be discarded. This was based on the concept that a set included locations which experienced rainfall from the same event, and the maximum rainfall of that event was sampled by an observer that could “see” the entire location rather than only one specific point.’ This procedure seems ad-hoc and not well supported statistically. The maximum of the extremes within a cluster is taken, instead for instance the cluster average of maxima. What is the representativeness of the maximum (among the extremes) within the cluster?

12) Line 220. Authors say: ‘Since there were over a hundred rain-gauges, we would like to group them into regions that persisted across the three time resolutions’. The obtained clusters depend on the time resolution (hour, day or pentad), what seems expected, since the associated meteorological systems have different spatio-temporal scales. Authors shall provide a clear justification for their choice. For instance, if instead we had taken the time-scales: hour, 6 hours, 3 days, would the clustering output be the same? Ideally, a uniform (or time-scale integrated) criterium should be adopted. Comment on that.

13) Line 220-221. Authors have performed cluster analysis for each of the three analyzed extremes and then they have grouped the respective clusters into common regions. Explain the rationale of that.

14) Line 241 Authors use both the symbol R and CC for the Spearman correlation. Choose only one symbol.

15) Lines 275 The seasonal anomalous Jlat, Flat, QU and QV were obtained as residuals from a sinusoidal and logistic fit of daily values. The result can be biased due to choice of the fit. Why not computing explicitly the daily-basis seasonal cycle from averages along the 61 days with the same Julian day (or within a smoothing window of 2-3 days)?

16) Line 285 Emphasize again that extended-PCA is applied, not PCA, to help the reading. The extended-PCA (Ext-PCA) analysis in the North-Pacific, shown by authors, have omitted other relevant modes of SST Pacific variability like the Northern Pacific Gyre Oscillation (NPGO) (DiLorenzo et al. 2008) that can be also candidates as drivers of extremes. Moreover, the Ext-PCA looks for variability all along the year (all trimesters) and not uniquely focused in the summer season. For instance, a PCA of SST in (JJA) would capture uniquely the difference between summers, eventually producing more suited modes driving the summer rain in Japan. Comment on that.

Di Lorenzo E., Schneider N., Cobb K. M., Chhak, K, Franks P. J. S., Miller A. J., McWilliams J. C., Bograd S. J., Arango H., Curchister E., Powell T. M. and P. Rivere, 2008: North Pacific Gyre Oscillation links ocean climate and ecosystem change. Geophys. Res. Lett., 35, L08607, doi:10.1029/2007GL032838.

17) Line 323 The so-called TREND+ mode seems to contain interannual variability. It is not clear how much is attributed to the trend and global warming. It seems to be a mixed mode including the linear trend (maybe nor the largest part), the PDV (or more commonly the PDO: Pacific Decadal Oscillation) and the NPGO.

18) Line 476. The relationship between monsoon parameters and rainfall extremes is done uniquely for daily extremes. Explain the omission of the other scales.

19) Lines 481-483 Authors say: ‘The spatial patterns of correlation were similar between d4PDF and rain-gauges for the 𝜇FLat, 𝜇JLat, and 𝜇QV, showing correlation along the Sea of Japan coast and anti-correlation along the Pacific coast (Figs 8e, 8g, 9e)’. This statement is based on very few rain-gauges, while in some others the agreement is neutral (weak correlations). For other parameters the similarity is quite inexistent, so any apparent similarity seems to be unfair.

20) Line 503. Authors refer to ‘scores of the Pacific SST’. Do you mean PCs (ENSO+, ENSO-, TREND+ etc.)? Rewrite and clarify. A Table with correlations between PCs and monsoon indices should be presented with analysis of significance and discussed afterwards. No physical explanation is provided for the correlations found at the end of the Discussion Sec. 4. Improvement needed.

Referee Report: PDF

Hide

ED: Reconsider after major revisions (further review by editor and referees) (10 Jan 2025) by Ricardo Trigo

AR by Shao-Yi Lee on behalf of the Authors (08 Mar 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (13 Mar 2025) by Ricardo Trigo

RR by Anonymous Referee #3 (19 Mar 2025)

ED: Publish as is (21 Mar 2025) by Ricardo Trigo

AR by Shao-Yi Lee on behalf of the Authors (27 Mar 2025)

Short summary

The authors performed verification on the relationships between extreme monsoon rainfall over Japan and Pacific sea surface temperature variability in the “database for Policy Decision-making for Future climate changes” (d4PDF). Observations showed widespread weak relationships between hourly extremes and the warming mode but reversed relationships between daily extremes and the decadal variability mode. Biases in d4PDF could be explained by the monsoon's slower movement over Japan in the model.