Statistical forecast of seasonal discharge in Central Asia using observational records: development of a generic linear modelling tool for operational water resource management

Apel, Heiko; Abdykerimova, Zharkinay; Agalhanova, Marina; Baimaganbetov, Azamat; Gavrilenko, Nadejda; Gerlitz, Lars; Kalashnikova, Olga; Unger-Shayesteh, Katy; Vorogushyn, Sergiy; Gafurov, Abror

doi:https://doi.org/10.5194/hess-22-2225-2018

Articles | Volume 22, issue 4

https://doi.org/10.5194/hess-22-2225-2018

© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 4.0 License.

Special issue:

Sub-seasonal to seasonal hydrological forecasting

https://doi.org/10.5194/hess-22-2225-2018

© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 22, issue 4

Research article

|

11 Apr 2018

Research article |

| 11 Apr 2018

Statistical forecast of seasonal discharge in Central Asia using observational records: development of a generic linear modelling tool for operational water resource management

Heiko Apel, Zharkinay Abdykerimova, Marina Agalhanova, Azamat Baimaganbetov, Nadejda Gavrilenko, Lars Gerlitz, Olga Kalashnikova, Katy Unger-Shayesteh, Sergiy Vorogushyn, and Abror Gafurov

Download

Final revised paper (published on 11 Apr 2018)
Preprint (discussion started on 21 Jun 2017)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC1: 'Peer review of “Statistical forecast of seasonal discharge in Central Asia for water resources management: development of a generic linear modelling tool for operational use”', Anonymous Referee #1, 28 Jun 2017
- AC2: 'Reply to reviewer comment RC1', Heiko Apel, 21 Sep 2017
SC1: 'General commentary', Samuel Dixon, 28 Jun 2017
- AC1: 'Reply to short comment of Samuel G. Dixon', Heiko Apel, 21 Sep 2017
RC2: 'Review of hess-2017-340', Anonymous Referee #2, 18 Jul 2017
- AC3: 'Reply to reviewer comment RC2', Heiko Apel, 21 Sep 2017

Peer-review completion

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

ED: Publish subject to revisions (further review by Editor and Referees) (01 Oct 2017) by Ilias Pechlivanidis

AR by Heiko Apel on behalf of the Authors (09 Nov 2017) Author's response Manuscript

ED: Referee Nomination & Report Request started (26 Nov 2017) by Ilias Pechlivanidis

RR by Anonymous Referee #2 (12 Dec 2017)

RR by Anonymous Referee #3 (16 Jan 2018)

Suggestions for revision or reasons for rejection

Overall opinion

I will provide a relatively brief review here as this paper has already been reviewed and revised in considerable detail.

The paper describes streamflow prediction in thirteen river basins in Central Asia using combinations of a range of *observed* predictors. I like the idea that is tested and the overall ‘simplicity’ of the model (though judging by the number of predictor combinations, I doubt that ‘simplicity’ is necessarily an appropriate term!). I also like the fact that the model is designed to be implemented operationally by the Central Asian Hydromet services.

The method seems reasonable, and makes sense, but is a little confusing at first because it is not immediately apparent that it is based solely on observational (historical) data. Given the manuscript title, I was expecting to read about streamflow forecasting using climate model outputs of precipitation/temperature, or large-scale climate indices. However, the predictions are based solely on historical data (antecedent meteorological measurements and streamflow).
I think the use observational records (and particularly the use of antecedent discharge) needs to be stated more clearly in the abstract and in the methods, because it explains why the forecasting skill is so high.

In fact, I would also recommend changing the title to something more explicit like ‘Statistical forecast of seasonal discharge in Central Asia using observational records: development of a generic linear modelling tool for operational water resources management’.

It would be interesting also if the authors could provide some comparison of the skill of their method with other forecasting approaches (including GCM/RCM-based methods) in the same region, over different lead times. Such a comparison would be particularly welcome in the last sections, which provide little perspective from the existing literature.

Overall, I would suggest making the methodology and choice of data as clear as possible in the text (see minor comments).

I would also suggest re-reading very carefully to improve the language.

Page 1
L.22. Suggest making this more explicit, e.g. ‘the development of a generic tool for deriving statistical forecasts of seasonal river discharge using only observational meteorological and hydrological records. The generic model is kept as simple as possible (i.e., does not include any forecast data)’.

Page 2.
L.5. ‘The improved skill of the model ensemble’. I find this language (model ensemble) a little confusing, because to me it suggests that you are using GCM forecasts..
L.5. ‘very narrow predictive uncertainty bands’: --> suggest just ‘narrow’
L.22. Run-on sentence (there are a lot of long sentences here at the start..). Suggest changing ‘exposed to the westerly air flows being a …’ --> ‘exposed to the westerly air flows, which are a…’
L.23. This what? This spatial variation?
L.23. ‘with snow melt to be’ --> ‘where snowmelt is’
L.28 ‘They are also among those river basins with the highest share’ --> ‘These mountains also have a very high fraction ….’
L.31. Full stop missing

Page 3
L.2. THE vegetation period
L.7. based on the empirical --> based on empirical
L.21. This paragraph needs an introductory sentence, before listing all the examples. What is the main idea of the paragraph?
L.34. Predictive skill (no s)

Page 4
L.13. Tools for automated image acquisition (no the)
L.23. ‘which areas’ --> ‘, whose combined drainage areas’
L.24. Long sentence. For clarity, I would suggest finishing the first sentence and starting a new one, e.g. “automatic predictor selection. The predictors are based on…”
L.25. “And additionally leverage by the” should be “and additionally leverage the”
L.27. This statement (“it is argued that … “) needs some references (e.g. Schär et al. 2004, Slater et al. 2017)

Page 5
L.2. Add a comma after forecast models
L.2. They cover --> the catchments cover (better to repeat the object at the start of a sentence)
L.3. remove ‘over catchments at’
L.6. Add a comma after three orders of magnitude
L.18 ‘For those catchments stations nearby’ --> ‘For those catchments, nearby stations’. I hope this is discussed later as a potential limitation.

Page 6
Table 1: ‘ist the mean annual precipitation sum’ --> ‘is the total mean annual precipitation of the meteorological station…’. Here and elsewhere in the manuscript, ‘total’ should be used (instead of ‘sum’).

Page 7.
L.4. for the presented study --> for the present study
L.7. Sentence is too long and needs to be split into 2 or 3 sentences to make sense.

Page 8
L.2. For all catchmentS
L.17. Consider removing thus and ‘to a large extent’ --> ‘The analysis of the inter-annual variability broadly maps the … differences of the catchments considered in this study’

Page 10
L.9. Here you use ‘snow melt’ in 2 words; elsewhere it is written in one word. Please revise for consistency throughout.
L.11. ‘Data about the’ --> Snow depth and snow water equivalent data

Page 11
L2. ‘An indicator about’ --> an indicator of

Page 27
L11. I think you mean Figure 8?

Page 31
L.24. ‘the separation of the largest share’: unclear what this means; needs rephrasing.

Figures
Figure 4. Prediction month needs to be mentioned in the caption (Jan-June). Grey background (in this figure and the other ones) is a bit distracting – consider removing.
Figure 5 is interesting and suggests very high skill. I’m surprised that the skill is shown as time-series rather than scatter plots, as it makes it hard to visualise the biases
Figure 8. The text is too small to read, even at full size.

Annexes
Annex 2. I find this list difficult to read; it would have been clearer in scientific notation.

References
Schär, C., Vasilina, L., Pertziger, F. and Dirren, S., 2004. Seasonal runoff forecasting using precipitation from meteorological data assimilation systems. Journal of Hydrometeorology, 5(5), pp.959-973
Slater, L.J., Villarini, G., Bradley, A.A. and Vecchi, G.A., 2017. A dynamical statistical framework for seasonal streamflow forecasting in an agricultural watershed. Climate Dynamics, pp.1-17. doi:10.1007/s00382-017-3794-7

Hide

RR by Anonymous Referee #4 (24 Jan 2018)

Suggestions for revision or reasons for rejection

According to the answers to reviewers provided by the authors and the improved version of the manuscript, I can state that most of the reviewers’ concerns have been worked out and the quality of the manuscript has been distinctly improved. However, there are still some points that need to be addressed before considering it ready for publication.

First of all, the overlapping between predictors and the predictand pointed out in section 3.1 should be better explained from my point of view. It is not acceptable that the predictors and the predictand are not fully independent. I also understand that the system operators need to be given bulk numbers for the entire vegetation period (April – September). An easy way to work this out, in my opinion, would be to distinguish between what is forecasted and which result is given to the system operators. Rather than predicting the entire discharge for the April – September period, the MLR models developed for May 1st and June 1st should aim at predicting the discharge for the remaining of the period. Once this remaining discharge is forecasted, it can be added to the discharge already recorded during the season (discharge in April and discharge in April and May respectively) to obtain the forecasted discharge for April – September season. In this way, the independence between predictors and predictand is respected while the system operators are given the required number.

Furthermore, the reason behind the preliminary predictor list (before significance level calculations) given in section 3 (pages 10 and 11) is not provided. Authors’ should explain why they chose these predictors and time lags (expert judgement, data analysis, previous studies, etc.). Moreover, the MLRs which have not passed the significance tests (it is indicated that in some months less than 20 models did it) should be excluded from the following calculations.

Finally, I consider necessary to include the R-squared coefficients computed in May and June for just the remaining period of the vegetation season in Table 2 and Figure 4. They can appear together with the ones calculated for the whole period. I agree on the authors’ point of view regarding the current practices. However, providing solely the R-squared coefficient for the whole season may lead to the conclusion that the MLRs’ performance increases across the vegetation period, while this increase in the R-squared value may be caused simply by the fact that one or two months are already known and, consequently, the uncertain part decreases. Providing R-squared values for the remining period would be more adequate to rank the quality of the forecasting models and find out which of them should be improved.

Minor concerns:
1. Page 12, lines 27-30: I would make a reference to Annex 4 and summarize the results of the tests described here rather in section 4 (page 15, lines 8-14).
2. Page 13, line 6: Which annex are you referring to?

Hide

ED: Publish subject to minor revisions (review by editor) (04 Feb 2018) by Ilias Pechlivanidis

AR by Heiko Apel on behalf of the Authors (13 Feb 2018) Author's response Manuscript

ED: Publish as is (27 Feb 2018) by Ilias Pechlivanidis

Short summary

Central Asia crucially depends on water resources supplied by snow melt in the mountains during summer. To support water resources management we propose a generic tool for statistical forecasts of seasonal discharge based on multiple linear regressions. The predictors are observed precipitation and temperature, snow coverage, and discharge. The automatically derived models for 13 different catchments provided very skilful forecasts in April, and acceptable forecasts in January.