Defining high-flow seasons using temporal streamflow patterns from a global model

Lee, D.; Ward, P.; Block, P.

doi:https://doi.org/10.5194/hess-19-4689-2015

Articles | Volume 19, issue 11

https://doi.org/10.5194/hess-19-4689-2015

© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.

https://doi.org/10.5194/hess-19-4689-2015

© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 19, issue 11

Research article

|

27 Nov 2015

Research article |

| 27 Nov 2015

Defining high-flow seasons using temporal streamflow patterns from a global model

D. Lee, P. Ward, and P. Block

Download

Final revised paper (published on 27 Nov 2015)
Preprint (discussion started on 30 Apr 2015)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC C1603: 'Review of "A Global Approach to Defining Flood Seasons"', Anonymous Referee #1, 19 May 2015
- AC C2224: 'Response to anonymous reviewer 1', Donghoon Lee, 24 Jun 2015
RC C1848: 'Review ‘A Global Approach to Defining Flood Seasons’ (hess-2015-140) by D. Lee, P. Ward, and P. Block', Anonymous Referee #2, 29 May 2015
- AC C2417: 'Response to anonymous reviewer 2', Donghoon Lee, 07 Jul 2015
- AC C2816: 'Revised manuscript', Donghoon Lee, 24 Jul 2015
RC C2137: 'Manuscript hess-2015-140', Anonymous Referee #3, 19 Jun 2015
- AC C2418: 'Response to anonymous reviewer 3', Donghoon Lee, 07 Jul 2015
- AC C2818: 'Revised manuscript', Donghoon Lee, 24 Jul 2015

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

ED: Reconsider after major revisions (28 Jul 2015) by Ross Woods

AR by Donghoon Lee on behalf of the Authors (03 Aug 2015) Author's response Manuscript

ED: Referee Nomination & Report Request started (04 Aug 2015) by Ross Woods

RR by Anonymous Referee #1 (04 Aug 2015)

RR by Anonymous Referee #2 (18 Aug 2015)

Suggestions for revision or reasons for rejection

2nd Review of the manuscript ‘Defining Flood Seasons Globally using Temporal Streamflow Pattern' (hess-2015-140) by D. Lee, P. Ward, and P. Block

General Comments:
The manuscript has been revised thoroughly by the authors and some of the previous comments of the reviewers. This improved the clarity of what the authors had actually been doing, but also raised some more points to comment.
Overall, there are still some key aspects that need to be addressed before the paper can be published (together with some other minor aspects mentioned below).
1. The title of the paper has been changed to ‘Defining Flood Seasons Globally using Temporal Streamflow Patterns’.
However, with the clarifications and explanations on the methodology provided by the authors after the first review, this title does not seem appropriate!
The methodology presented enables the identification of the ‘high flow’ seasons and NOT the ‘flood season’! So the using ‘flood seasons’ in the title is misleading.
For many regions in the world, the ‘real floods’ do actually not occur during the high flow season (in the manuscript called ‘flood season.’)
This is actually also corroborated in several of their results. E.g. Figure 9, in which (although difficult to tell from the colour code used in that plot) more than 1/3 to 1/2 of all pixels fall in the evaluation class of ‘low’ to ‘poor’ (based on their own ‘subjective’ classification scheme (P7 L 17-18)).
Therefore, the paper is far from ‘defining flood seasons globally’!
Additionally, I would suggest, that the title should indicate that the ‘global’ scale of the paper comes from model output, as the actual data available for checking the results has strong spatial biases.

2. The authors highlight their new technique to define ‘flood seasons’. However, no formal testing of the new technique is performed, in which the technique would be compared to other already well established techniques that aim to define the flood season.
Generally, the method is similar to a ‘peak over threshold approach’ but instead of considering independent peaks, all daily flows above a volume based threshold are used to define the ‘flood season’ (which is actually the ‘high flow season’, or the ‘high flow spell’).
Therefore, the method simply identifies the month with the highest number of days above a flow quantile (here the upper 95% percentile). It is not clear why, the authors call it a new ‘volume and magnitude’ based technique. Additionally, a proper testing of the sensitivity of the results obtained to the threshold selected is essential when presenting a ‘new method’.

3. Additionally, the editor’s requests ‘What are the advantages to defining the new measure PM (and FS), in relation to existing published measures of flood seasonality? What shortcomings did the previous existing published measures have, and to what extent do your new measures overcome those shortcomings? have not been addressed!
The authors only compare their method to other approaches, by cross-correlating the identified peak months with other techniques.
Although the correlations applied are similar (Table 1) (which would be expected by using the same dataset and techniques that aim to identify similar features of the flow regime), a correlation between the different classification techniques cannot be used to justify the superior performance of their new method!
Correlation should never be confused with causation!
Instead, differences in the obtained correlation with any of the other classification techniques could actually mean that these techniques are superior in their performance in capturing the peaks!
I urge the authors to follow the well-established research approach of first testing all these techniques and then select the most appropriate technique for the rest of the analyses (which might be the new method, but maybe the older techniques perform better (one cannot tell from the current manuscript)). Instead of coming up with a new method first and then not thoroughly evaluating if the new method actually is better!

Specific Comments:
Section Abstract:
P2 L3-4: Please rephrase, as the sentence currently gives the idea that only the new approach of defining flood seasons is ‘objective’ and not the other methods. I think the current approach is as ‘objective’ as the other methods. So I would restrain from using the word ‘objective’ here.
P2 L9: I disagree that the defined flood seasons represent well the actual flood records from DFO. This is only achieved when the minor secondary flood seasons are included later in the manuscript! Please rephrase.
P2 L12-15: This is a false claim. The identified seasons (which are the peak month +- 1 month) do certainly not help to improve the understanding of flood frequency, trends and interannual variability. Please remove.
P4 L3-12: In this paragraph, the shortcomings of the previous studies are presented. High emphasis is placed on the issue of clustering. However, for most of the studies this is not the main aim, instead they also show very distinct seasonality patterns. Addition, the other studies are criticised for ‘not being representative of local scale conditions ‘ and that the current study is addressing ‘basin and even grid cells’. However, the analysis of this study is not ‘local’ either and the approach used to define a sub-basin’s months of flood peak (P 9 L10-17), local conditions are also lumped and lost as well. Therefore I suggest, rephrasing the entire paragraph and discussing the differences in the methods (how the flood seasons are defined) instead, with a focus on the outcomes of the flood season and not how the results are being applied to cluster regions in previous studies (which is only the second research step in most of the studies).
P4 L27: Please specify what ‘relaxing the criteria’ means. What options have been assessed?
P 6 Section 3.1: The first paragraph that briefly reviews existing methods and explains the similarities and differences with the new approach, is still very confusing.
Please re-write the section again in a more structured manner. Especially, please clarify what is meant by streamflow volume and magnitude (water level?)
P7 L 12: After reading the previous section it is still not clear why the PAMF ‘inherently contains magnitude and volume properties’.
P7 L 20- P8 L2: The description of other methods should go a to a paragraph before the ‘new method’ is presented
P8 L3-17: See ‘General comment Number 3’. Cross correlation between different techniques can never ‘indicate some success’ of one method compared to the other! Correlation is not causation!
P10 L 22-25: I suggest moving the discussion of Figure 7 further down in the document and discuss Figure 4, 5 and 6 first. Additionally, change ‘United States and Canada’, to ‘North America’ to be consistent with the labels in Figure 7.
P 10 L27-30: It is mentioned that low PAMF values are computed for the US and Europe and this is attributed to be due ‘at least in part, to reservoirs and dams along the Mississippi, Missouri and Danube’. This might true for the observed flows, however, with the modelled streamflow this should not be the case.
However, from Figure 5 for example on can see that the model obtains even lower PAMF values in Europe! So certainly, the human impact does not play a role for the PAMF value to be low!
As the analyses has a global focus, an in depth discussion (with more focus on spatial location) on the obtained differences in PM and PAMF is needed!
Additionally, for Europe and the US, only a few stations are actually located on the strongly anthropogenic impacted Rivers mentioned before. Therefore, the poor performance shown with the PAMF might also be a shortcoming of the method and needs to be discussed further!
P 11 L1-20: The discussion focusses too much on the areas where performance of the flood season can be considered acceptable. Areas with stronger differences such as Australia and South America are currently ignored and have to be discussed as well!
Additionally, it is highlighted that 40% of the models and the data share the same peak months. This is a quite low performance; however, the authors are not critical about this low outcome at all.
Overall, I think in discussion the results, the authors should aim for a more balanced assessment of the good and less successful outcomes of their method!
P 12 L1: I strongly object the authors claim that there is a ‘striking similarity’ between the DFO and the modelled season and that this ‘further supports the model’s ability to appropriately identify the PM spatially’. Please rephrase.
P13 L27: Please rephrase ‘streamflow magnitude and volume characteristics of floods’ to something that explains the method better.
P14 L 31-P15 L 2 : Please rephrase, as there are only 40% of the peak months that are appropriately identified correctly, which is not an ‘indication of strong agreement between model and observed flood season’ and the flood records of the DFO are also not ‘well represented’, to be more realistic about the outcomes of the study.
P15 L 24-25: Please rephrase, as the model does not ‘enable the complete flood season identification globally’. There are many locations on the globe where, there are problems and low performance as indicated by low PAMF values. Therefore, the method is not globally applicable. It would be better if it would be highlighted where the flood season can be expected to be well represented by the model!

Figures:
Figure 8: I suggest merging Figure 8 and Figure 9 into one Figure with 2 panels to allow a better interpretation of the results. Having Figure 8 and 9 together helps to interpret the reliability of the months defined in Figure 8 ).
Figure 9: From the current way of plotting, it is difficult to distinguish the different reliability classes as defined on page 7. For example, based on the classes defined beforehand, central Europe and most of Australia has a poor reliability. I suggest showing the PAMF not as gradual colours but actually show the colours according to the reliability classes defined beforehand so the results can be interpreted accordingly. This should also be better discussed in the text.

Hide

ED: Reconsider after major revisions (27 Aug 2015) by Ross Woods

AR by Donghoon Lee on behalf of the Authors (01 Oct 2015) Author's response Manuscript

ED: Referee Nomination & Report Request started (15 Oct 2015) by Ross Woods

RR by Anonymous Referee #2 (29 Oct 2015)

Suggestions for revision or reasons for rejection

3rd review of the manuscript hess-2015-140 by D. Lee, P. Ward, and P. Block

General Comments:
The manuscript has been revised thoroughly in the 2nd revision and it is clear from the author’s responses that they did their best to incorporate all the comments of the reviewers and editor.
The clarity of the document improved considerable but there are still a few points/comments that I suggest to be considered before the final publication of the document.
I therefore recommend accepting the document with minor revisions.

Specific Comments:
Section Abstract:
P1 L3: I suggest adding in parenthesis that the high-flow season is 3 months long
P1 L11: I suggest detailing that the ‘high-resolution high-flow seasons’ are spatially high resolution (due to the gridded model)

Section 1 Introduction:
P4 L13-14 & : Spell out PM, HS, and PAMF and put the abbreviations in parenthesis behind

Section 2:
P5 L15-16: From the paper that is cited here, it is not obvious how the model has been validated to extreme discharges (which are of key interest to this paper). Ward et al. seem to have only evaluated areas that are > 125000 km2 (which are ~55 stations), which is quite different from the size of the basins/grid cells used in this study.
Additionally, it is noted here that the model has a ‘strong performance’, however, the paper cited shows that especially in dryer areas the model overestimated by > 50%.
I therefore suggest adding a short discussion on the reliability of the model to smaller basins and drier areas, instead of highlightening a ‘strong model performance’.

Section 3:
P6 & 7 General comment on POT:
Please add a sentence/paragraph indicating that the threshold for POT can either s be based on threshold that defines ‘specific average number of floods’ (e.g. as in Cunderlik et al. 2004) or be based on a volumetric threshold (which can either be on a year to year basis or over the entire series)(e.g. Lang et al. 1999, in which it is already stated that the POT ‘requires the analysis of both , the magnitude and the time of arrival of a peak’).

P6 L22: I recommend already adding the reference here to papers by Lang et al. (1999.) and the Institute of Hydrology, 1999 (mentioned at a later stage on page 8)
P6 L23: Please remove ‘, prior to a specific date ’, as this is not correct, but I presume this snippet accidentally remained there during the revisions.
P6 L25: replace ‘miss’ with ‘do not capture’, as the word missing invokes a drawback which is actually a feature of the method.
P 7 L4-8 Based on the general comment on POT above, I fail to understand why it is highlighted that the POT approach is over the entire series and not on a year-to-year basis.
As the Section P6 L10 to P7 L 10 has been rewritten several times, it is confusing and I therefore suggest rewriting the entire paragraph incorporating the above comments.
P 7 L10-13: As there is a thorough analysis of results of the different methods later in the paper I suggest to remove the sentence of the selection of the threshold here, as it otherwise pre-empts the detailed analysis later.
P 7 L 14-15: Please further elaborate, why you decide to use the month before and after (which is subjective), instead of using the adjacent months with the highest number of days above the threshold, which would be able to capture also asymmetric peak flow distribution around the PM. (This request had been already mentioned in my previous review but was not addressed by the authors yet.)
P 7 L22: The AMF and therefore the PAMF as an evaluation measure is strongly dependent on how the year is defined (i.e. calendar year versus hydrological year). Please discuss briefly how this might affect the scores of PAMF.
P 7 L25: Did you mean ‘percentage’ instead of the ‘percent of time’?
P8 L12-14: Not clear what is meant here. Elaborate/further explain why this is the case.
P8 L27-30: It is not clear if the table shows all cases or only the cases in which the PMS differ. Please explain further.
P9 L8-11: See comments on P6 L25 and P 7 L4-8 and change accordingly.

Section 4:
P 10 L 13 & 31: Based on my experience reservoirs do not exclusively result in low PMAF values (i.e. spread of the peak flow values across the months), they can also cause the opposite (concentration of the peak values into a specific month). This all depends on the operation rules!
To avoid that this strong assumption (as already mentioned by the authors in L 14) is confused with a statement of generality, I suggest that it should be made clear again in Line 31 the low PMAF this is based on an assumption and that reservoirs can also have the opposite effect.
P 12 L 11: What does ‘around Finland’ mean? South Finland around the Baltic Sea? Please specify.
P 13 L 11-19: Please add to your discussion of the regions that have low PAMF values a sentence that critically discusses the low model reliability in South America and particularly Europe, as most of the grid cells there have low or poor performance.
P 13 L 26 and Figure 9: It is not clear what the Figure is showing. Is it the month with the maximum number of floods events in the DFO or the mean month of all events on record? Please specify in the text and figure caption

Section 5:
No suggestions; Well analysed and discussed.

Section 6
Discussion in section 4 and 6 have improved considerably.
I congratulate the authors on these sections.
No further comments.

Figures:
Fig 4 and 6: For the ease of the reading/interpreting the results, please indicate in the text or the caption of these figures that negative values mean that the simulated high flow season is earlier and that positive values mean that the simulated HS is later.
Fig 5 and Fig 11: For the ease of the reading/interpreting the results, please indicate in the text that refers to these figures or in the captions what the percentages of PMAF mean. I.e. 0-40% = poor, 40-60% low, as it is already done in Figure 8.
Fig9. See comment above for P 13 L 26

Typography:
Please make sure that you replace the full stop inside the parenthesis with the full stop outside of the parenthesis (i.e. use ‘).’ instead of ‘.)’ )

Hide

ED: Publish subject to minor revisions (Editor review) (29 Oct 2015) by Ross Woods

AR by Donghoon Lee on behalf of the Authors (06 Nov 2015) Author's response Manuscript

ED: Publish as is (09 Nov 2015) by Ross Woods

Short summary

This paper presents a global approach to defining high-flow seasons by identifying temporal patterns of streamflow. Simulations of streamflow from the PCR-GLOBWB model are evaluated to define dominant and minor high-flow seasons globally, and verified with GRDC observations and flood records from Dartmouth Flood Observatory.