the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Optimizing WRF-Hydro Calibration in the Himalayan Basin: Precipitation Influence and Parameter Sensitivity Analysis
Abstract. Lack of in-situ observations and reliable climate information in large part of the Himalayas poses a significant challenges for assessing water resources vulnerability accurately. Further, the reliance on only a few coarse resolution gridded datasets with considerable uncertainty complicates this problem. However, integrated hydrometeorological modeling systems, like WRF-Hydro, have the potential to provide information in such ungauged or poorly observed regions, yet they require careful calibration. Here, we calibrate and assess the fidelity of WRF-Hydro in simulating the hydrological regime of the Beas basin. Selected WRF-Hydro model parameters are calibrated using the PEST framework, using eight simulations with two meteorological forcings from two WRF realisations. Model calibration improves the accuracy of the basin discharge simulation, however the choice of precipitation forcing is also critically important. We propose an ensemble weighting scheme to optimize an intra-annual tradeoff between streamflow under- and overestimation in different WRF-Hydro configurations. This study demonstrate the efficacy of using coupled (offline) WRF and WRF-Hydro for providing climate change impact-relevant information in data-sparse basins.
- Preprint
(5351 KB) - Metadata XML
-
Supplement
(1947 KB) - BibTeX
- EndNote
Status: open (until 14 May 2024)
-
RC1: 'Comment on egusphere-2024-587', Anonymous Referee #1, 05 Apr 2024
reply
General comments
The paper entitled “Optimizing WRF-Hydro Calibration in the Himalayan Basin: Precipitation Influence and Parameter Sensitivity Analysis“ introduces the topic of deterioration of future fresh water resources in the Himalayan mountain range. The authors evaluate the use of atmospheric model WRF and hydrological model WRF-Hydro as numerical tools to assess water resources in the observations-sparse region of the Beas basin. Specifically, the authors: (1) evaluate different techniques of the PEST software to calibrate the hydrological model WRF-Hydro , (2) perform a sensitivity analysis of the parameters of the WRF-Hydro model, (3) evaluate the precipitation impact of two WRF model parameterization schemes on streamflow simulated by the WRF-Hydro model, and (4) suggest an ensemble averaging method for the WRF schemes to improve the WRF-Hydro simulations of streamflow. This paper has the potential to be a substantial contribution to the scientific community of water resources studies because it suggests methods and provides findings for improved hydrological simulations in data-sparce regions. It has a good structure regarding the presentation of methods and results, which can however be improved through clarifications in the methodology and modifications and further explanations in the presentation of the results, discussion and conclusions.
Major comments
Comments on the topic of the sensitivity of parameters and calibration algorithms.
#1 The authors present the parameter sensitivity objective first in the presentation of the objectives of the study (L114-115) and the sensitivity analysis results last in the order of presenting the results of the paper. However, a reasonable sequence of an atmospheric-hydrological modelling study usually begins, both in Methods and Results sections, with setting up the models and the presentation of the sensitivity analyses, before the actual simulations. The authors could make this adjustment, which would make the follow-up presentation of atmospheric – hydrologic simulations easier, considering the use of different calibration algorithms in the sensitivity and the calibration.
# 2 The parameter sensitivity and calibration methods comprise an important part of the paper, yet they are not well introduced in Section 2.5. Some sentences that explain the two methods are mixed in the results (e.g. L505-513 and L521-527). In particular, the SVD method is first mentioned in L306 (Methods) and LSQ method in L409 (Results). Are sensitivity and calibration performed simultaneously? Are SVD and LSQ part of the regularisation mode? How regularization is defined compared to the estimation mode? What are the meanings of the LSQ and SVD abbreviations? The supplement contains some information, but the main paper should contain information as well because the calibration algorithms are evaluated throughout the paper. Please make all these concepts clear in the method section so that they are clear when the reader comes to the results.
# 3 Are SVD and LSQ used in previous studies? If yes, how they compare to the current study? This could be added in the discussion section.
# 4 What is the conclusion on the comparison of the estimation and regularization mode in PEST? Is the performance improvement worth to use the additional computation resources required by the regularization mode? Was there a significant improvement with the regularization mode? This can be added in discussion or conclusions.
Comments on the comparison of the WRF and WRF-Hydro configurations
# 5 In terms of total amounts, can the authors present the precipitation and streamflow totals in the same units (e.g. mm) for all experiments? It could also be more informative to compute and present the percent bias of modeled streamflow relative to the observed streamflow. This information can be added in the existing tables.
# 6 The authors state, “The performance of the calibrated model depends strongly on the season” (L403). They also assume that “one might expect a seasonally specific calibration to further improve model performance during that season” (L405-406). However, no seasonally dependent evaluation metrics are presented to justify this statement in sections 3.2.1 – 3.2.2. Could the authors present some metrics to justify the additional experiments for the summer months?
# 7 In all hydrographs, especially in Figure7 and Figure 8, after WRF-Hydro is calibrated for the summer months only, and after the ensemble estimations are made, a substantial mismatch between observations and simulations is seen for streamflow for the months of September until December. What is this underestimation attributed to? Is there a source for streamflow not accounted in the model? Is there a physical process that is missing or is the precipitation forcing inadequate? What explanation can the authors give?
Other comments
L218-225: It is unclear how the LULC update is made.
L297: How is “a change in the hydrograph” defined? Is there a particular measure used?
L612-614: Are there previous studies examining the results of ensemble averaging on precipitation and the impact of it on streamflow?
L618-621: Which land use classes used by the land surface component of WRF and WRF-Hydro are there in the Beas basin? What is the fractional coverage? This type of information can be added in the study area description and could be related to the discussion on the sensitivity of parameters.
L651: How would the use of OVROUGHRT and RETDEPRT change the results in the current study?
Equations 1-4: The equations presented in this section fit better in the Methods rather than the results. Could you generalize the equations and move them to Methods?
Minor comments:
L58: The introduction of the Beas basin is abrupt to readers unfamiliar with the study area. What is the importance of the Beas basin relative to the broader region? Please rephrase or move the sentence to a later paragraph when the focus is on the specific basin.
L55-57 and L69-71 are contradictory. If the glaciers disappear by the end of the century by 90%, how will the water stored in glaciers meet the water demands at least by the end of the 21st century? Is there high uncertainty reported in other studies? If yes, this could be mentioned.
L290: “…having each parameter perturbed to its default value..”. Do you mean perturbed from the default value with the range of possible values specified in Table 2?
Table 2: Please provide the units of the listed parameters. If possible, please add the description of these parameters.
What does (2009) stand for with KGE and what are the units of RMSE in Tables 4-7?
L306-309: Please define parameter identifiability?
L616: “LSQR has the advantage of computational speed”, compared to what?
Citation: https://doi.org/10.5194/egusphere-2024-587-RC1 -
RC2: 'Comment on egusphere-2024-587', Anonymous Referee #2, 20 Apr 2024
reply
This paper presents the application of the PEST algorithm to calibrate the WRF-Hydro hydrologic modeling system in the Beas basin in the Himalayan region. While the paper’s goals could be within the scope of the journal, I found the paper hard to understand in many parts. Therefore, I recommend rejection at this stage. In particular:
The objectives are not properly explained, and little details are provided about the hydrologic and WRF atmospheric model, the datasets, and the calibration approach. The origination is also quite confusing. For example, what is the link between Sections 1.2 and 2.5?
In many circumstances, the authors refer to other papers (largely, Dixit et al., 2023) to justify the assumptions made and the data used in this manuscript, while they should have still properly explained the implications of these assumptions.
More generally, in my opinion, testing the effectiveness of a calibration method (here, PEST) should be done in regions where there is a wealth of data. What is learned there could then, hopefully, be translated to other areas. The authors explicitly say that the precipitation data are inaccurate in their study region. So, what is the value of this effort? This should have carefully been addressed.
Finally, the level of English should be significantly improved.
Citation: https://doi.org/10.5194/egusphere-2024-587-RC2
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
203 | 82 | 11 | 296 | 17 | 6 | 7 |
- HTML: 203
- PDF: 82
- XML: 11
- Total: 296
- Supplement: 17
- BibTeX: 6
- EndNote: 7
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1