Improvement of the SWAT model for event-based flood simulation on a sub-daily timescale

Yu, Dan; Xie, Ping; Dong, Xiaohua; Hu, Xiaonong; Liu, Ji; Li, Yinghai; Peng, Tao; Ma, Haibo; Wang, Kai; Xu, Shijin

doi:https://doi.org/10.5194/hess-22-5001-2018

Articles | Volume 22, issue 9

https://doi.org/10.5194/hess-22-5001-2018

© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 3.0 License.

https://doi.org/10.5194/hess-22-5001-2018

© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 22, issue 9

Research article

|

27 Sep 2018

Research article |

| 27 Sep 2018

Improvement of the SWAT model for event-based flood simulation on a sub-daily timescale

Dan Yu, Ping Xie, Xiaohua Dong, Xiaonong Hu, Ji Liu, Yinghai Li, Tao Peng, Haibo Ma, Kai Wang, and Shijin Xu

Download

Final revised paper (published on 27 Sep 2018)
Preprint (discussion started on 17 Aug 2017)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC1: 'review of the manuscript by Yu et al.', Anonymous Referee #1, 21 Sep 2017
- AC1: 'Point by point responses to Referee #1', Dan Yu, 17 Nov 2017
RC2: 'Review 1', Anonymous Referee #2, 22 Sep 2017
- AC2: 'Point by point responses to Referee #2', Dan Yu, 17 Nov 2017

Peer-review completion

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

ED: Reconsider after major revisions (further review by editor and referees) (20 Dec 2017) by Markus Weiler

AR by Dan Yu on behalf of the Authors (15 Mar 2018) Author's response Manuscript

ED: Referee Nomination & Report Request started (16 Mar 2018) by Markus Weiler

RR by Anonymous Referee #1 (23 Apr 2018)

Suggestions for revision or reasons for rejection

Review of the manuscript „Improvement of the SWAT model for event-based flood forecasting on a subdaily time scale“ by Yu et al.

In this manuscript, Yu et al. aimed at improving flood peak modelling with the SWAT model. To achieve this, they suggest a modification of the SWAT model. For this, they have implemented a sub-daily event modeling and a spatial-distributed calculation of the unit hydrograph into the SWAT model.

Major comments:

In this study, it is shown how to improve the SWAT model for flood peaks. This is a justification for a study. However, I still do not see a strong link to flood forecasting in this article. You may argue that reasonable flood peak simulations are required for flood forecasting (a point that can be raised e.g. in the discussion), but to use flood forecasting in the title, I expect a higher contribution to this topic.

I am still not convinced that spatial heterogeneity is adequately considered. The high number of rainfall gauges might allow a higher number of subbasins. I see that the number of subbasins increased from 4 to 21 compared to the first version. However, I think that some statements of the spatial heterogeneity within the subbasins would be helpful to justify the still relatively low number of subbasins. This is in particular relevant since the SWAT model is based on the HRU concept (spatially not located hydrotops) and that a focus of the article is the calculation of the time of concentration. A higher spatially explicit discretization would certainly be helpful in calculating the time of concentration realistically.

I am not satisfied with the parameter selection and the method presented for sensitivity analysis and model calibration (see detailed comments below).

I have one comment to the calculation of objective functions for SWAT-EVENT and the comparison with the classical SWAT model. To calibrate the SWAT, the entire time series was used. For SWAT-EVENT, only events were used. It is then not very surprising that an optimization of a smaller time period leads to better results within this time period. A fair comparison would be to calibrate the SWAT model also only for the event period. Maybe I understand something wrong, but currently it seems to be that the better result in SWAT-EVENT is (at least partly) related to the selection of a smaller time period. In all cases, the study would benefit from showing the difference between SWAT and SWAT-EVENT by using the same event time period.

Comments:

P. 1, L.15-16: I agree that SWAT is a good water quality model, but needs time for several simulations due to its complexity. Thus, there is an explanation missing why you have selected the SWAT model for flood studies. A simpler but also semi-distributed model which is only focused on hydrology might be more useful.

P.2., L.9: For flood forecasting purpose, it is required to simulate on time and not only the past. At least in the future perspective, on-time simulation is needed. Please refer to this point (e.g. in the discussion).

P.2., L.18: Here, a justification for the model selection is missing. There are a huge amount of spatial-distributed model. At least a few words on that are required.

P.2., L. 18: Please cite the SWAT model here with the original reference.

P.2, L.26: What is meant here with „by research involving model parameters“? Please be precise. Model parameters themselves are not a research topic. Parameter sensitivity, optimization, screening etc would be a topic.

P.3., L. 18: How is flood forecasting included?

P.3, L.20: All SWAT model versions include an auto-calibration routine. This is not specific for SWAT2005. I can understand that you prefer to use SWAT2005 because you have reasons for that. But I still do not see an advantage of SWAT2005 compared to more recent versions.

P.4., L.27: How can you consider 138 gauges for 21 subbasins using Thiessen average rainfall? It is unclear how spatial rainfall variability is included keeping in mind that the SWAT model is using one rainfall station per subbasin. And an accurate rainfall estimation might be crucial for flood forecasting. Please give more details on this method.

P.5, L.8: Again information is required how to consider „timely flood warning“.

P.6, L. 21: A core term in calculating the time of concentration is the average slope length, which is the average of the slope length within a subbasin. This parameter rather critical. Its spatial variability within a subbasin may increase with increasing catchment size. Thus, again this is a point which should be considered in the discussion of spatial heterogeneity. I assume that a larger number of subbasins and a better representation of the slope length might result in a more realistic modeling.

P.7, L.16: Certainly, it is not required to calibrate 26 model parameters to model floods with the SWAT model. I recommend using less model parameters based on hydrological knowledge.

P.7, L.17: It is reasonable to use only the sensitive model parameters. However, it would be great to know how this decision was made. Thus, how were model parameters identified as sensitive?

P.7, L.21: I cannot follow the justification for an automatic calibration. Certainly it is not possible to calibrate 26 parameters manually in a realistic time period. However, it is certainly possible to reduce the number of parameters to an acceptable number.

P.8, L.21: It would be interesting to see a sensitivity analysis specific for SWAT-EVENT.

P.9, L.15: It is rather strange to highlight that the performance is very good according to ENS, but to ignore that the performance is only satisfactory according to PBIAS refered to the same reference. Please consider all thresholds or none of them.

P.12, L.18-21: The selection of objective functions is certainly a trade-off in optimizing different parts of the hydrograph. A focus on high flows might lead to lower performance for normal or low flows. It should be made clear that the selection of the objective function was made to fulfill the goals of the study and that it was accepted to have a lower performance in other phases of the hydrograph.

Table A1: Several parameters are certainly not relevant in modelling flood peaks such as BIOMIX, SOL_ALB, TLAPS, etc. Please reduce the number of parameters based on hydrological knowledge in advance.

Table A1: Please explain the meaning of -25 and 25 for snow melt temperature (SMTMP).

Table A1: The slope length is used first to describe the geographical features of the subbasins and then it is calibrated. Is this approach valid?

Table A2: The differences in model parameters between the SWAT model and the SWAT-EVENT version is very high. From hydrological point of view, it does not make sense to use very different parameter sets of the same model in the same catchment. This means that the hydrological processes are not well represented/understood.

Table 4: All flood events are summer events. Do you really need to consider snow model parameters?

Figure 2a: can be removed since land use and soil types are not discussed in detail.

Other comments:

P.2., L. 16: Small letters for Beven and Kirkby

P.2., L.22: HRU

P.6, Eq.2-6: Not all terms of these equations are clearly visible.

P.11, L.5: concluded

Figure 8: basin instead of bsin (x-axis)

Hide

ED: Reconsider after major revisions (further review by editor and referees) (03 May 2018) by Markus Weiler

AR by Dan Yu on behalf of the Authors (16 Jul 2018)

ED: Referee Nomination & Report Request started (28 Jul 2018) by Markus Weiler

RR by Anonymous Referee #1 (22 Aug 2018)

ED: Publish subject to minor revisions (review by editor) (23 Aug 2018) by Markus Weiler

AR by Dan Yu on behalf of the Authors (02 Sep 2018) Author's response Manuscript

ED: Publish as is (03 Sep 2018) by Markus Weiler

AR by Dan Yu on behalf of the Authors (04 Sep 2018)