Process verification of a hydrological model using a temporal parameter sensitivity analysis

Pfannerstill, M.; Guse, B.; Reusser, D.; Fohrer, N.

doi:https://doi.org/10.5194/hess-19-4365-2015

Articles | Volume 19, issue 10

https://doi.org/10.5194/hess-19-4365-2015

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

https://doi.org/10.5194/hess-19-4365-2015

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

Articles | Volume 19, issue 10

Research article

|

29 Oct 2015

Research article |

| 29 Oct 2015

Process verification of a hydrological model using a temporal parameter sensitivity analysis

M. Pfannerstill, B. Guse, D. Reusser, and N. Fohrer

Download

Final revised paper (published on 29 Oct 2015)
Preprint (discussion started on 05 Feb 2015)
Supplement to the preprint

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC C520: 'Review of “Temporal parameter sensitivity guided verification of process dynamics” by Pfannerstill et al., 2015', Shervan Gharari, 10 Mar 2015
- AC C1606: 'Response on interactive comment of Shervan Gharari', Matthias Pfannerstill, 19 May 2015
RC C1196: 'Review', Anonymous Referee #2, 22 Apr 2015
- AC C1615: 'Response on interactive comment of anonymous referee 2', Matthias Pfannerstill, 19 May 2015

Peer-review completion

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

ED: Reconsider after major revisions (20 May 2015) by Erwin Zehe

Dear Mr. Pfannestil,
I had close look at your manuscript, before I checked the reviewer comments and your responses. Both reviewers came up with a rather critical assessment. The most critical point, which is consistently made within both reviews, is the partly circular reasoning of using TEDPASAll to reconfirm that parameters of the fast ground water domain are indeed earlier sensitive as those of the slow domain.
Reviewer II, an respected expert in the field of hydrological modelling and parameter estimation, even recommended rejection due to this circular reasoning. Although I wouldn’t go that far, I think the study needs to be thoroughly revised, before it might potentially be accepted. The main challenge is to break out from this circular or at least plausible reasoning within the revised manuscript. A possible way how this could be achieved is to corroborate that the improved model does indeed perform superior; for instance by employing the TIGER method (time series of grouped errors, proposed by Reusser et al. (2009) in HESS.

In a nutshell this method detects temporal occurrence of model error types: such as peak too low, or recession too slow, to identify model deficiency based on their repetitive occurrence. In this particular case I would expect that prior to adding the fast groundwater domain, the model should show systematic errors (recession too slow) in the early stage of the recession phase. After the model tuning this error should vanish or at least be reduced. In combination with TEDPAS this would corroborate that the proposed approach/framework does indeed allow a targeted improvement of a model with respect to perform more consistent with observations (maybe also GW levels if available). Of course you are free to decide to follow another route within the revision. Of key importance to corroborate the value of your approach is to show that the model performs better compared to observations. The fact that the current manuscript relies more on plausibility of the argument, in the sense that you find what you expect in terms of reaction times, contributes largely to the circular character of the manuscript in the present form.

Best regards,

Erwin Zehe

Hide

AR by Matthias Pfannerstill on behalf of the Authors (31 Jul 2015) Author's response Manuscript

ED: Referee Nomination & Report Request started (12 Aug 2015) by Erwin Zehe

RR by Anonymous Referee #2 (28 Aug 2015)

Suggestions for revision or reasons for rejection

In this new version, the manuscript has undergone some major and necessary improvements, which has eliminated the major issue of circular reasoning. My reading of the current version is essentially an implementation of the TEDPAS sensitivity analysis on an implementation of the SWAT model for a watershed in Germany. This is a useful case study, and may be worth publishing. The manuscript is generally well written, and the presentation of figures and tables is of good quality. However, I identify still some weaknesses, which I would encourage the authors to address:

- The introduction can do with a better contextualization of TEDPAS in the contemporary literature. In particular, methods such as DYNIA and TIGER (as suggested by resp. the other reviewer and the editor) are relevant and even if not implemented can be used to sketch a broader picture of temporal sensitivity analysis.

- I can understand the angle of hypothesis testing used in the paper, but I think that it still sits a bit uncomfortable. The problem with many hypotheses including those put forward in the manuscript is that they are vague and either too straightforward or too hard to test. In this case, I think that one would struggle to find a hydrologist that would disagree with any of those listed in Table 1. Another problem with the hypothesis discourse is that it makes the language of the introduction rather heavy handed and difficult to read. It might be worth considering the hypotheses simply as key pieces of knowledge about the catchment processes, and TEDPAS as a powerful method to make sure that that those are represented in the model? A more fundamental problem with the hypotheses is that they are qualitative. It would have been ideal if any hard data were available, for instance the start and end time of tile drainage contribution. The lack of such data makes building a strong argument difficult, and as a result the discussion is rather fuzzy. But I understand that this is outside the scope of the study.

- The temporal sensitivity plots in figure 3 and 4 are useful, but I wondered whether it would be possible to add a stacked area plot showing the different contributions to streamflow, especially related to the flows mentioned in the hypotheses. I think that this could be a very intuitive visualization of how the model simulates the magnitude and timing of these flows (e.g., when tile flow contribution starts, peaks, and stops). As such, it may also help with a qualitative evaluation of the model.

Hide

RR by Shervan Gharari (21 Sep 2015)

Suggestions for revision or reasons for rejection

Review of “Temporal parameter sensitivity guided verification of process dynamics” by Pfannerstill et al., 2015.
The revised manuscript has been improved significantly and it is much clearer now. To what I understood from the paper the work is as follow:
- The authors took a model and calibrated it.
- The authors studied and listed the way catchment seems to function in their view.
- The authors compared the way they perceived the catchment is working with calibrated model via sensitivity of the model parameters.
However in my point of view the paper can be written with better language. I would suggest the authors to make the paper more fluent. If I were instead of the authors, I would rewrite many parts of the paper starting from the title; instead of the current title the authors can use a more revealing title such as “Verification of processes [in model XXX] using a temporal parameter sensitivity analysis” or “On the role of parameter sensitivity analysis for the process verification [for a model XXX]”. Just a simple example which I can point out is page 4 line 213. The authors wrote “The sum of all partial variances cannot be higher than one by definition. However, it can be smaller than one due to parameter interactions. This is the case for the sensitivity of one parameter that is affected by other parameters”. These sentences can be shortened and rewritten in the following ways “The sum of all partial variances cannot be higher than one by definition.” Or “Because of parameter interactions the sum of all partial variances fluctuates between 0 and 1 while it cannot be more than 1 by definition.”
In my point of view the paper reads much better scientifically now and deserve to be published. However as I mentioned I think the paper will benefit a lot by smoothing the use of English language. As I am not a native speaker and my English skills might be poor I leave it to the Editor to decide.
With Kind regards
Shervan Gharari

Hide

ED: Publish subject to minor revisions (Editor review) (24 Sep 2015) by Erwin Zehe

AR by Matthias Pfannerstill on behalf of the Authors (04 Oct 2015) Author's response Manuscript

ED: Publish as is (07 Oct 2015) by Erwin Zehe

Short summary

To ensure reliable model results, hydrological processes have to be represented adequately in models. We present a framework that uses a temporal parameter sensitivity analysis and observed hydrological processes in the catchment to verify hydrological models. The framework is exemplarily applied to verify the groundwater structure of a hydrological model. The results show the appropriate simulation of all relevant hydrological processes in relation to processes observed in the catchment.