Potential application of hydrological ensemble  prediction in forecasting floods and its components  over the Yarlung Zangbo River basin, China

Liu, Li; Xu, Yue Ping; Pan, Su Li; Bai, Zhi Xu

doi:https://doi.org/10.5194/hess-23-3335-2019

Articles | Volume 23, issue 8

https://doi.org/10.5194/hess-23-3335-2019

© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/hess-23-3335-2019

© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 23, issue 8

Research article

|

14 Aug 2019

Research article |

| 14 Aug 2019

Potential application of hydrological ensemble prediction in forecasting floods and its components over the Yarlung Zangbo River basin, China

Li Liu, Yue Ping Xu, Su Li Pan, and Zhi Xu Bai

Download

Final revised paper (published on 14 Aug 2019)
Supplement to the final revised paper
Preprint (discussion started on 20 Apr 2018)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC1: 'Potential application of hydrological ensemble prediction in forecasting flood and its components over the Yarlung Zangbo River Basin, China', Renata Romanowicz, 09 May 2018
- AC1: 'Revision notes for “Potential application of hydrological ensemble prediction in forecasting flood and its components over the Yarlung Zangbo River Basin, China”', Li Liu, 13 Jul 2018
RC2: 'Comments on “ Potential application of hydrological ensemble prediction in forecasting flood and its components over the Yarlung Zangbo River Basin, China”', Ye Tian, 05 Jun 2018
- AC2: 'Revision notes for “Potential application of hydrological ensemble prediction in forecasting flood and its components over the Yarlung Zangbo River Basin, China”', Li Liu, 13 Jul 2018

Peer-review completion

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

ED: Reconsider after major revisions (further review by editor and referees) (30 Aug 2018) by Hannah Cloke

AR by Li Liu on behalf of the Authors (10 Oct 2018)

ED: Referee Nomination & Report Request started (30 Oct 2018) by Hannah Cloke

RR by Anonymous Referee #2 (29 Nov 2018)

RR by Anonymous Referee #3 (18 Dec 2018)

Suggestions for revision or reasons for rejection

The manuscript (version 3) presents the development and performance of flood forecast system for the mountainous basin of Yarlung Zangbo River (China). The runoff is modelled using Variable Infiltration Capacity (VIC) model and for the input ECMWF meteorological ensembles are used. Authors have defined two research goals (1) to investigate the ability of ensemble predictions, to forecast flood volume and its components (base flow, snow melt runoff, rain runoff) for mountainous region; (2) analysis of the approach in which model is evaluated for an ensemble of parameters, determined as Pareto optimal solution for several fit measures. The case study and applied methods of parameter identification are interesting and for sure valuable from a scientific point of view. Unfortunately, despite these advantages, in my opinion, authors failed achieving any of their goals:

1) How the “ability” of ensemble predictions to forecast can be defined? It could be expected that some weather projections predict observed flood peak and other not. At which point the hypothesis can be falsified? What is a level of CRPS, CRPSS or MAE at which authors would agree that ensemble predictions are unable to forecast a flood peak? In my opinion it always is possible to insist that the approach is “able” - better or worse.
The problem should be rather analysed in terms of suitability, where the ensemble predictions are compared with other methods. Then it would be possible quantifying the performance of different methods using i.e. CRPSS. Such an approach can be found i.e. in Pappenberger et al. (2005) who analysed ensemble along with deterministic predictions.

2) Hydrogram separation is probably the weakest part of the study. According to the text, glaciers are important part of the basin system. However, the applied model does not account for this component. In Page 5, line 16-18 authors assume that it can be described as snow accumulation. This is a very rough assumption and its limitations can be seen in Fig 3 (authors are aware of this fact, page 8, lines 25-26), where flows, probably shaped by glacier outflow, are not explained. What is the point analysing forecasted streamflow components section 4.3), if the essential element of the base flow is missing? The conclusion to this remark could be the sentence from the Discussion section, referring to the forecasted components (Page 14, lines 18-19):
“Therefore, in this study the simulation driven by observations becomes an alternative to act as proxy although it is difficult to determine whether such proxy is believable or not.” If it is hard to determine that we can believe the method or not, it is not a scientific approach.

3) The concept of the ensemble of Pareto optimal parameters is the most interesting, but not addressed properly. Authors have identified a N-set of parameters that provides the tradeoffs between four fit measures: the Nash–Sutcliffe efficiency and relative bias for whole flow time series and limited to 10% of highest flows. However, the performance of the forecast system is analysed only in the respect of a single measure: accumulated flood volumes. For such a single criterion it is quite obvious that a single compromise simulation should be better. The advantage of the parameter ensemble could be found, if other measures, similar to those in calibration, were also included.

In the reviewer’s opinion, authors should revise their research design, focusing maybe on the forecast quality, in the respect of identified parameters (Pareto-sets), instead of stream flow components.

Other comments

A) The description of Snowmelt Tracking Algorithm, provided in Section 3.3 is unclear and not supported with references to Li et al. (2017). The original Snowmelt Tracking Algorithm Li et al. 2017) acquires internal variables of a runoff model:
"In this study we develop explicit quantification of the historical and future fQ , snow over the western U.S. We quantify fQ , snow by tracking the fate of snowmelt in modeled hydrologic fluxes" - Li et al. (2017).
In my opinion all assumptions used in eq. (5-7) should be justified and explained by authors. It is unclear, why it is possible to assume, that infiltration (f_{i,snow}) and runoff ratios (f_{R,snow}) are equal (Eq. 5). Also Eq. 7 is unclear. What is a definition of fi,snow? Equiation 5 suggests that it is a ratio: fi,snow = \frac{I_snow}{R}. Why in Eq. 7 it is multiplied by i (infiltration?)? The same applies to F_{w, snow}.
Please, provide also dimensions for variables: R, B, W and ETt.

B) Page 3, Lines 10-15 are unclear and please consider rewriting. Parameters ensembles of a hydrological model and parameter sets refer here to a wider problem of the hydrological model uncertainty. Mentioned techniques are just possible solutions. I suggest defining this problem as an uncertainty analysis and provided these references as examples. Please note, that in the present form the text is confusing, as it is hard to distinguish between "forecast ensemble" (meteorological model uncertainty) and "parameter ensemble" (for hydrological models' uncertainty).

C) Page 4, Lines 16-17: Please note, that the supporting references are more than 4 years old.

D) Page 5, Line 7, what data was used in cross-validation - from these 27 meteorological stations? If so, it should not be written, that lapse rates were estimated using available meteorological data?

E) Section 4.2. It is unclear for how a flood forecasting system was tested. Did authors use a receding horizon or forecast was issued for chosen events? Terms "first floods" maximum floods etc. should be defined.

F) Page 13, Line 21: “Firstly, N-simulations generally behave better when the trade-offs in multi-objectives are significant.” does the study supports this finding?

Pappenberger, F., Beven, K. J., Hunter, N. M., Bates, P. D., Gouweleeuw, B. T., Thielen, J., & De Roo, A. P. J. (2005).
The cascading model uncertainty from medium range weather forecasts (10 days) through a rainfall-runoff model to flood inundation predictions within the European Flood Forecasting System (EFFS). Hydrology and Earth System Sciences Discussions, 9(4), 381-393.

Hide

ED: Publish subject to minor revisions (review by editor) (15 Jan 2019) by Hannah Cloke

AR by Li Liu on behalf of the Authors (30 Jan 2019) Author's response Manuscript

ED: Publish as is (27 Jun 2019) by Erwin Zehe

AR by Li Liu on behalf of the Authors (07 Jul 2019) Manuscript

Short summary

The ensemble flood forecasting system can skillfully predict annual maximum floods with a lead time of more than 10 d and has skill in forecasting the snowmelt-related components about 7 d ahead. The accuracy of forecasts for the annual first floods is inferior, with a lead time of only 5 d. The snowmelt-induced surface runoff is the most poorly captured component by the system, and the well-predicted rainfall-related components are the major contributor to good performance.