the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Jul 2022
Assimilation of Transformed Water Surface Elevation to Improve River Discharge Estimation in a Continental-Scale River
Menaka Revel
Xudong Zhou
Dai Yamazaki
Shinjiro Kanae
Abstract. Quantifying continental-scale river discharge is essential to understanding the terrestrial water cycle but is susceptible to errors caused by a lack of observations and the limitations of hydrodynamic modeling. Data assimilation (DA) methods are increasingly used to estimate river discharge in combination with emerging river-related remote sensing products (e.g., water surface elevation [WSE], water surface slope, river width, and flood extent). However, directly comparing simulated WSE to satellite altimetry data remains challenging (e.g., because of large biases between simulations and observations or uncertainties in parameters), and large errors can be introduced when satellite observations are assimilated into hydrodynamic models. In this study we performed direct, anomaly, and normalized value assimilation experiments to investigate the capacity of DA to improve river discharge within the current limitations of hydrodynamic modeling. We performed hydrological DA using a physically-based empirical localization method applied to the Amazon Basin. We used satellite altimetry data from ENVISAT, Jason 1, and Jason 2. Direct DA was the baseline assimilation method and was subject to errors due to biases in the simulated WSE. To overcome these errors, we used anomaly DA as an alternative to direct DA. We found that the modeled and observed WSE distributions differed considerably (e.g., differences in amplitude, seasonal flow variation, and a skewed distribution due to limitations of the hydrodynamic models). Therefore, normalized value DA was performed to improve discharge estimation. River discharge estimates were improved at 24 %, 38 %, and 62 % of stream gauges in the direct, anomaly, and normalized value assimilations relative to simulations without DA. Normalized value assimilation performed best for estimating river discharge given the current limitations of hydrodynamic models. Most gauges within the river reaches covered by satellite observations accurately estimated river discharge, with Nash-Sutcliffe efficiency (NSE) > 0.6. The amplitudes of WSE variation were improved in the normalized DA experiment. Furthermore, in the Amazon Basin, normalized assimilation (median NSE = 0.50) improved river discharge estimation compared to open-loop simulation with the global hydrodynamic model (median NSE = 0.42). River discharge estimation using direct DA methods was improved by 7 % with calibration of river bathymetry based on NSE. The direct DA approach outperformed the other DA approaches when runoff was considerably biased, but anomaly DA performed best when the river bathymetry was erroneous. The uncertainties in hydrodynamic modeling (e.g., river bottom elevation, river width, simplified floodplain dynamics, and the rectangular cross-section assumption) should be improved to fully realize the advantages of river discharge DA through the assimilation of satellite altimetry. This study contributes to the development of a global river discharge reanalysis product that is consistent spatially and temporally.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(6646 KB)
-
Supplement
(1687 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(6646 KB) -
Supplement
(1687 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Menaka Revel et al.
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2022-412', Anonymous Referee #1, 20 Jul 2022
General comment:
In this research, the authors performed data assimilation (DA) experiments to explore the capacity to improve daily river discharge within current limitations of global hydrodynamic modeling. For this purpose, the water surface elevation (WSE) from satellite altimetry was assimilated in a configuration of three experiments, the direct (absolute values), the anomalies and the normalized anomalies. The authors also evaluated the capability of these DA experiments in some scenarios, for instance when some parameters/forcing (river bathymetry and runoff) of the hydrodynamic model are biased, as well as conditions when river bathymetry is calibrated. The results showed that, in general, the normalized DA performance was the best, improving the daily discharge estimates in up to almost 60% of the stations evaluated, compared to the simulation without DA. These results considering the current limited conditions of the global hydrodynamic models (e.g. without calibration).
The major contribution of this research is the evaluation of these experiments and scenarios, providing adequate knowledge and insights in terms of how DA techniques could be used to improve discharge estimates, which fits with the perspectives of SWOT missions for example. In general, this work is worth publishing in the Hydrology and Earth System Sciences journal, however, it needs “moderate revisions”. Some suggestions for revisions are as follows.
Moderate comments:
- The article is well written and follows a logical order. Although it is a bit long, an average reader can follow the reading, however at a certain point there are more experiments than initially described. For example, the evaluation of DA under biased runoff and river bathymetry conditions; DA under calibrated river bathymetry conditions; DA using the runoff forcing of a bias-corrected model. That is why I recommend the authors to describe more explicitly these experiments in section 2.6.
- Regarding the selection of virtual stations (VSs) for assimilation or validation, the justification is a bit vague, even though this may be important for the performance of the experiments, so I recommend improving this point.
- In sections 3.1.1, 3.1.2 and 3.1.3 take care with the description of the time series in figures 4, 5 and 6, respectively. There is a confusion between the description of the gauges results. For example, line 315 describes the Santos Dumont station on the Purus river, however the series in Figure 4d are from a gauge on the Juruá river. This confusion occurs for the stations Gaviao (Juruá) and Manacapurú (Amazon) in Figure 5, and in all the gauges in Figure 6. This was probably an involuntary error in the preparation of the figures, please correct.
- Experiments that assimilate absolute (direct) values, anomalies and normalized anomalies are referred to by the acronyms Exp. 1, Exp. 2 and Exp. 3 respectively, however throughout the manuscript both nomenclatures are used. I suggest that only one be adopted to improve the readability of the text. Even so that the information can be quickly abstracted by the reader these experiments could be called DIR_DA, ANOM_DA and NORM_DA for example.
- Since the authors have used a localization method in the DA scheme, I suggest reinforcing the discussion on how this might affect discharge estimates due to assimilation of WSE within or outside the influence coverage of the VSs.
- Could you discuss a bit about to what do you attribute the lower efficiency of flow estimates in the upper Solimoes River ? efficiency of the CaMa-Flood model ? selection of VSs ? localization ? large uncertainties in the VSs data in that region ?
Specific comments (Line-by-line comments):
Introduction:
- 35: It is more accurate to say, "River discharge records can be used...".
- 42: I would say that also these simulations (of GHMs) have been used to complement observed records.
- 48: If we go deeper we could say that these forcing factors can also be rainfall and climatic variables.
- 50: I would say: "Given the current limitations of GHMs and in-situ measurements, ...".
- 57: This sentence mentioning the SWOT mission seems a bit loose, you should rework it to integrate it with what you want to mention above.
- 62: Typo: "combining" instead of "combing".
- 83: This statement describes information repeated in the previous one, perhaps you could combine them.
Methodology:
- 124: In this sentence you can already start reporting on the period of DA experiments (2009-2014).
- 215: Why wasn't the SURFEX-TRIP model outputs used since it also belongs to WRR2?
- 236: To reference these annual average rainfall values you can cite Builes-Jaramillo & Poveda, 2018; Espinoza et al., 2009. (https://doi.org/10.1029/2017WR021338 and https://doi.org/10.1002/joc.1791)
- 237: Please specify what you mean by large number of observations, perhaps this is valid for remote sensing observations because it is a large basin with strong hydrological signals, hence the citation of Fassoni-Andrade et al. 2021.
- 249: You could elaborate a little more on this sentence. Why these virtual stations could affect the estimates using assimilation? this exclusion of 3% was by a visual analysis of the series only? these stations are located in some particular place in the Amazon, maybe rivers with a small width?
- Sections 2.7.1 and 2.7.2 could be merged, as it could confuse the reader. The main objective of this research is to evaluate the performance in simulating daily discharge but here also the performance of WSE will be evaluated. This merged section could be called "observational data" since the altimetry data has also been used for validation.
Results:
- 294-296: This sentence seems to be repetitive with the previous one, you could merge them.
- 302: It would be appropriate to refer to Figure 4b in this sentence.
- 309: The time series for the Santos Dumont station is not shown in Figure 4d. Instead, a station on the Juruá River is shown. See my major comments above.
- 325: The information in parentheses should go in the methodology section.
- 330: “WSE performance decreased…” instead of “WSE decreased…”.
- 332: The Gavião and Manacapuru gauges do not correspond to Figures 5c and d.
- 351-352: None of these described gauges correspond to figures 6c, d and e.
- 412: I think there is a typo, please delete "3.2.1.".
- 8: It is not possible to distinguish gauges inside or outside the coverage area of the altimetric satellites. Could you differentiate them somehow?
- Table 3. I have noticed that some values in this table do not correspond exactly to those described. For instance, in the first column (All and r) in the table, the values are 0.74, 0.85 and 0.84 for experiments 1, 2 and 3 respectively. While in the description the values are 0.73, 0.84 and 0.83 (L. 431, L. 415 and L. 439 respectively).
- 452: As shown in Figure 9, the BIAS values are only positive, so I recommend describing somewhere (probably methodology) that the index is an absolute value of BIAS.
- Figure 9: It is a bit difficult to differentiate the VSs that were used for assimilation and validation. Perhaps it could be improved by changing the symbology from "o" to "*", increasing the size of the maps by reducing the space between them and decreasing a little the size of the station symbols so that they do not overlap too much. This is just a suggestion.
- Section 3.3: Please detail how in this experiment you have generated the realizations of the set for assimilation.Was it with the same perturbation as for the WRR2 models?
- 476-478: The end of this sentence sounds strange, I suggest to redo it or delete this last part from "...,direct DA (Exp 1)...."
Conclusions:
- 624: Typo, it's HTESSEL not HTEESSEL (same for L.541, 542 and 544).
Citation: https://doi.org/10.5194/egusphere-2022-412-RC1 -
AC1: 'Reply on RC1', Menaka Revel, 18 Aug 2022
Referee #1
General comment:
In this research, the authors performed data assimilation (DA) experiments to explore the capacity to improve daily river discharge within current limitations of global hydrodynamic modeling. For this purpose, the water surface elevation (WSE) from satellite altimetry was assimilated in a configuration of three experiments, the direct (absolute values), the anomalies and the normalized anomalies. The authors also evaluated the capability of these DA experiments in some scenarios, for instance when some parameters/forcing (river bathymetry and runoff) of the hydrodynamic model are biased, as well as conditions when river bathymetry is calibrated. The results showed that, in general, the normalized DA performance was the best, improving the daily discharge estimates in up to almost 60% of the stations evaluated, compared to the simulation without DA. These results considering the current limited conditions of the global hydrodynamic models (e.g. without calibration).
The major contribution of this research is the evaluation of these experiments and scenarios, providing adequate knowledge and insights in terms of how DA techniques could be used to improve discharge estimates, which fits with the perspectives of SWOT missions for example. In general, this work is worth publishing in the Hydrology and Earth System Sciences journal, however, it needs “moderate revisions”. Some suggestions for revisions are as follows.
Reply:
We would like to thank the referee for his kind remarks. We will address all the comments in the revised manuscript, and comprehensive explanations are provided below.
Moderate comments:
1. The article is well written and follows a logical order. Although it is a bit long, an average reader can follow the reading, however at a certain point there are more experiments than initially described. For example, the evaluation of DA under biased runoff and river bathymetry conditions; DA under calibrated river bathymetry conditions; DA using the runoff forcing of a bias-corrected model. That is why I recommend the authors to describe more explicitly these experiments in section 2.6.
Reply:
Thank you very much for the recommendation. We will include brief descriptions of the additional experiments in section 2.6 considering the length of the manuscript.
2. Regarding the selection of virtual stations (VSs) for assimilation or validation, the justification is a bit vague, even though this may be important for the performance of the experiments, so I recommend improving this point.
Reply:
We would like to express our gratitude to referee #1. We simply separated VSs to assimilation and validation 80%, and 20%, respectively. We will revise the description to reflect the selection of the VS for assimilation and validation
3. In sections 3.1.1, 3.1.2 and 3.1.3 take care with the description of the time series in figures 4, 5 and 6, respectively. There is a confusion between the description of the gauges results. For example, line 315 describes the Santos Dumont station on the Purus river, however the series in Figure 4d are from a gauge on the Juruá river. This confusion occurs for the stations Gaviao (Juruá) and Manacapurú (Amazon) in Figure 5, and in all the gauges in Figure 6. This was probably an involuntary error in the preparation of the figures, please correct.
Reply:
We thank referee #1 for recognizing the mistake. It was indeed an involuntary error in preparing the figures. We will correct the cross-reference with the corresponding figure and the text. Furthermore, we will correct the manuscript according to the referee's comment. Basically, we will replace Figures 4, 5, and 6 to match the descriptions in Sect. 3.1.1, 3.1.2, and 3.1.3, respectively.
4. Experiments that assimilate absolute (direct) values, anomalies and normalized anomalies are referred to by the acronyms Exp. 1, Exp. 2 and Exp. 3 respectively, however throughout the manuscript both nomenclatures are used. I suggest that only one be adopted to improve the readability of the text. Even so that the information can be quickly abstracted by the reader these experiments could be called DIR_DA, ANOM_DA and NORM_DA for example.
Reply:
We appreciate the suggestion from referee #1. We will adopt a better naming convention for the experiments in the revised manuscript such as DIR, ANO, and NOM.
5. Since the authors have used a localization method in the DA scheme, I suggest reinforcing the discussion on how this might affect discharge estimates due to assimilation of WSE within or outside the influence coverage of the VSs.
Reply:
We would like to thank the great suggestion by the referee. The localization method we used in this study is an adaptive localization method (Revel et al., 2019) which is far different from conventional localization methods which use fixed square-shaped local patches. The comparison between those methods can be found in Revel et al., (2019). The adaptive localization method recognizes the highly correlated areas and removes less correlated areas (e.g., small downstream tributaries). We will highlight the effect of adaptive localization in the revised manuscript.
6. Could you discuss a bit about to what do you attribute the lower efficiency of flow estimates in the upper Solimoes River ? efficiency of the CaMa-Flood model ? selection of VSs ? localization ? large uncertainties in the VSs data in that region ?
Reply:We would like to thank referee #1 for raising the question. The WSE simulations using CaMa-Food of these two locations show a reasonable agreement with the satellite altimetry observations, therefore direct DA method performed better. On the other hand, anomaly and normalized value DA have a limitation that the assimilated WSE is dependent on the statistics (i.e., mean and standard deviation) of open loop simulation. If statistics may not represent the actual values of mean WSE and the standard deviation of WSE, the assimilated WSE can be also different from the actual values in the anomaly and normalized assimilation method. When the relationship between the WSE and discharge (i.e., rating curve) is well represented in the model, the anomaly and normalized DA may not correctly estimate the river discharge if the statistics are biased. We will add some explanation to the manuscript regarding this.
Specific comments (Line-by-line comments):
Introduction:
1. 35: It is more accurate to say, "River discharge records can be used...".
Reply: We will revise the sentence as per referee #1’s comment.
2. 42: I would say that also these simulations (of GHMs) have been used to complement observed records.
Reply: We agree with referee #1 that the simulations of GHSs have been used to complement observations rather than compensate. Hence, we will revise the text to reflect the referee's comment.
3. 48: If we go deeper we could say that these forcing factors can also be rainfall and climatic variables.
Reply: We concur with referee #1 that the forcing factors are primarily rainfall and climatic variables. Therefore, we will modify the text to include rainfall and climatic variables.
4. 50: I would say: “Given the current limitations of GHMs and in-situ measurements, …”.
Reply: We admit referee #1’s suggestion that both GHMS and in-situ measurements have their own limitations. It will be updated in the revised manuscript.
5. 57: This sentence mentioning the SWOT mission seems a bit loose, you should rework it to integrate it with what you want to mention above.
Reply: The sentence will be revised to reflect the referee #1’s comment.
6. 62: Typo: "combining" instead of "combing".
Reply: We will correct it.
7. 83: This statement describes information repeated in the previous one, perhaps you could combine them.
Reply: By thanking referee #1, the paragraph will be modified.
Methodology:
1. 124: In this sentence you can already start reporting on the period of DA experiments (2009-2014).
Reply: We would like to thank Referee #1 for the nice suggestion. But we try to dedicate this section (2.1) solely to describing the assimilation framework. So, we will introduce the period of the experiments in section 2.6.
2. 215: Why wasn't the SURFEX-TRIP model outputs used since it also belongs to WRR2?
Reply: We would like to thank referee #1 for the question. We did not consider outputs from the SURFEX-TRIP model because those are not compatible with the CaMa-Flood model. The SURFEX-TRIP model consists of the capillary rise in the runoff variable where CaMa-Flood is not capable of dealing with such runoff data. Therefore, we will add some descriptions to the text.
3. 236: To reference these annual average rainfall values you can cite Builes-Jaramillo & Poveda, 2018; Espinoza et al., 2009. (https://doi.org/10.1029/2017WR021338 and https://doi.org/10.1002/joc.1791)
Reply: We would like to express our gratitude to referee #1. We will use the above studies when describing the annual average rainfall in the Amazon basin and the text will be modified considering the above studies.
4. 237: Please specify what you mean by large number of observations, perhaps this is valid for remote sensing observations because it is a large basin with strong hydrological signals, hence the citation of Fassoni-Andrade et al. 2021.
Reply: We thank referee #1 for raising this issue. Yes, indeed we refer to remote sensing data in the Amazon basin. So, we will revise the text of the manuscript.
5. 249: You could elaborate a little more on this sentence. Why these virtual stations could affect the estimates using assimilation? this exclusion of 3% was by a visual analysis of the series only? these stations are located in some particular place in the Amazon, maybe rivers with a small width?
Reply: We would like to express our gratitude to referee #1 for the important question. We did some extensive analysis of these largely biased VS which cannot be compared with the MERIT DEM (Yamazaki et al., 2017). This exclusion was done by comparing the mean WSE of satellite altimetry observation with the MERIT DEM, not on visual analysis. Our analysis revealed that most of the “biased VS” were in narrow rivers at relatively high elevations. We will include some descriptions of these biased VS in the revised manuscript.
6. Sections 2.7.1 and 2.7.2 could be merged, as it could confuse the reader. The main objective of this research is to evaluate the performance in simulating daily discharge but here also the performance of WSE will be evaluated. This merged section could be called "observational data" since the altimetry data has also been used for validation.
Reply: We appreciate the comment on merging Sections 2.7.1 and 2.7.2. We will revise it in the revised manuscript.
Results:
1. 294-296: This sentence seems to be repetitive with the previous one, you could merge them.
Reply: We would like to thank referee #1 for the suggestion and we will revise the text.
2. 302: It would be appropriate to refer to Figure 4b in this sentence.
Reply: Thank you very much for pointing out this. We will refer to figure 4d in the sentence.
3. 309: The time series for the Santos Dumont station is not shown in Figure 4d. Instead, a station on the Juruá River is shown. See my major comments above.
Reply: Thank you very much for the comment. We will revise Figure 4d to be compatible with the description in the text.
4. 325: The information in parentheses should go in the methodology section.
Reply: Thank you for the kind suggestion. We will include that in the methodology section.
5. 330: “WSE performance decreased…” instead of “WSE decreased…”.
Reply: Thank you very much. We will revise it in the revised manuscript.
6. 332: The Gavião and Manacapuru gauges do not correspond to Figures 5c and d.
Reply: Thank you. We will revise figures 5c and d to match the description in Section 3.1.2.
7. 351-352: None of these described gauges correspond to figures 6c, d and e.
Reply: Thank you for raising this error. We will replace the correct figures for figures 6c, d, and e and the description will be modified to correspond to figures 6c, d, and e in Section 3.1.3.
8. 412: I think there is a typo, please delete "3.2.1.".
Reply: Thank you for pointing out it. We will delete "3.2.1.".
9. 8: It is not possible to distinguish gauges inside or outside the coverage area of the altimetric satellites. Could you differentiate them somehow?
Reply: Thank you for the great suggestion. We will modify figure 8 to distinguish the gauges inside and outside satellite altimetry coverage.
10. Table 3. I have noticed that some values in this table do not correspond exactly to those described. For instance, in the first column (All and r) in the table, the values are 0.74, 0.85 and 0.84 for experiments 1, 2 and 3 respectively. While in the description the values are 0.73, 0.84 and 0.83 (L. 431, L. 415 and L. 439 respectively).
Reply: Thank you for cross-checking the description with the data provided in the tables. The values in the table are correct values so we will revise the text according to Table 3. We will revise the L. 431, L. 435, and L. 439 with the values of 0.74, 0.85, and 0.84.
11. 452: As shown in Figure 9, the BIAS values are only positive, so I recommend describing somewhere (probably methodology) that the index is an absolute value of BIAS.
Reply: Thank you for pointing out this. We will add “absolute bias” to the methodology section when introducing “BIAS” term as follows:
12. Figure 9: It is a bit difficult to differentiate the VSs that were used for assimilation and validation. Perhaps it could be improved by changing the symbology from "o" to "*", increasing the size of the maps by reducing the space between them and decreasing a little the size of the station symbols so that they do not overlap too much. This is just a suggestion.
Reply: Thank you very much for your precious suggestion. We will revise figure 9 and other similar figures to better represent symbols.
13. Section 3.3: Please detail how in this experiment you have generated the realizations of the set for assimilation.Was it with the same perturbation as for the WRR2 models?
Reply: Thank you for asking for clarification. This section used the same assimilation outputs from the direct, anomaly, and normalized DA experiments. The perturbations are similar to WRR2 model outputs.
14. 476-478: The end of this sentence sounds strange, I suggest to redo it or delete this last part from "...,direct DA (Exp 1)...."
Reply: Thank you. We will revise the sentence in the revised manuscript.
Conclusions:1. 624: Typo, it's HTESSEL not HTEESSEL (same for L.541, 542 and 544).
Reply: Thank you, We will correct them to HTESSEL.
Reference:1. Revel, Ikeshima, Yamazaki and Kanae: A Physically Based Empirical Localization Method for Assimilating Synthetic SWOT Observations of a Continental-Scale River: A Case Study in the Congo Basin, Water, 11(4), 829, doi:10.3390/w11040829, 2019.
2. Yamazaki, D., Ikeshima, D., Tawatari, R., Yamaguchi, T., O’Loughlin, F., Neal, J. C., Sampson, C. C., Kanae, S. and Bates, P. D.: A high-accuracy map of global terrain elevations, Geophys. Res. Lett., 44(11), 5844–5853, doi:10.1002/2017GL072874, 2017.Citation: https://doi.org/10.5194/egusphere-2022-412-AC1 -
RC3: 'Reply on AC1', Anonymous Referee #1, 30 Aug 2022
I have reviewed in detail the authors' responses to my comments in the first round of revisions. It appears that the authors were able to resolve and address all the corrections. I will be waiting for the final version of the main document (file I have not yet received) before I can make a decision. Thanks
Citation: https://doi.org/10.5194/egusphere-2022-412-RC3 -
AC3: 'Reply on RC3', Menaka Revel, 13 Sep 2022
We would like to express our gratitude to referee #1 for his continued interest in our study. We look forward to submitting the revised manuscript after the open discussion phase.
Citation: https://doi.org/10.5194/egusphere-2022-412-AC3
-
AC3: 'Reply on RC3', Menaka Revel, 13 Sep 2022
-
RC3: 'Reply on AC1', Anonymous Referee #1, 30 Aug 2022
-
RC2: 'Comment on egusphere-2022-412', Anonymous Referee #2, 30 Jul 2022
The authors investigated the assimilation of satellite altimetry to improve discharge. The article is excellent. The authors presented 3 types of altimetry assimilation, which are 1. direct, 2. anomaly, and 3. normalized assimilation, and concluded that to improve discharge it is better to just assimilate the water surface dynamics (3.) given that the model is relatively accurate. If the model is completely corrupted, the authors concluded that anomaly and/or direct assimilation are more effective.
My main complaint about this manuscript is in section 3.1 (specifically 3.1.1, 3.1.2, and 3.1.3) where the figures (fig 4, 5 and 6) showed hydrographs that were different from the locations discussed in the text. Besides that, I have just some small comments that are specified below:
Line 40. I think it is better to change GHM definition only to Global Hydrodynamic Models instead of Hydrological due to some features the authors discuss further such as "runoff as a forcing factor", "discretized river", "surface water dynamics", etc.. Line 77 and 519 could be GHM instead of global hydrodynamic models.
Line 52. The authors could also mention Laser altimetry. The ICESat missions are also very used in academic research. Maybe instead of a radar pulse, can be a radar/light pulse or even an electromagnetic pulse.
As an alternative to the semi-variogram analysis to determine the spatial dependency weights, the authors could have used "backwater lengths in rivers" studied by Samuel (1989). It can give an idea of which river reaches are affected by WSE variations at the VS locations. It would be a good idea to compare both approaches in future studies (not now).
Samuels, P. G. (1989). Backwater lengths in rivers. Proceedings of the Institution of Civil Engineers, 87(4), 571–582. https://doi.org/10.1680/iicep.1989.3779
Line 156. using a power law dependent on what? Width? Upstream drainage area? Is it the same power law parameters for the whole basin?
Line 199. Something went wrong with the font size of some words. Line 351. Line 438. Line 637.
Line 229. So, the mean and standard deviation were calculated based on the open loop simulation?
Line 231 to 239. Some of your readers might be unfamiliar with the Amazon Basin. It would be interesting to write a short and objective section about this basin, presenting a DEM map at least (a mean Precipitation map would be nice too).
Line 275. How do you measure the relative sharpness and the difference in reliability? Line 383 should be here.
Several wrong references in section 3.1.
Line 309. The authors said that the Santos Dumont gauge is in the Purus River, but in Figure 4 it says Jurua River.
Line 332. "Figure 5c–e displays hydrographs of the Jurua (Gaviao), Amazon (Manacapuru), and Negro (Serrinha) rivers" but in Figure 5 it says Manicore on the Madeira River, Aruma on the Purus River, and Sao Felipe on the Negro River.
Line 351. "The lower panels of Figure 6 illustrate flow dynamics along the Amazon mainstem (Sao Paulo De Olivenca; Figure 6c) and Japura (Vila Bittencourt; Figure 6d) and Negro (Curicuriari; Figure 6e) rivers." but in Figure 6 it is written "Hydrographs recorded at Humaita on the Madeira River, Santos Dumont on the Jurua River, and Canutama on the Purus River are presented on panels c, d, and e, respectively."
Line 321. Saying that the "direct DA generally improved flow dynamics" is very optimistic. Based on these results, I'd probably say that the direct DA maintained or even degraded the general performance, at least for discharge.
Line 340. Once again, I think it is an optimistic conclusion. In the last sentence, the authors just said: "although NSE and ISS values worsened slightly." So how can the authors say afterward that "discharge estimates improved moderately"? I don't think that improvements in the correlation coefficient are enough for such a statement given that the NSE has become worst. But I reckon that seasonality got better as correlation got higher. Maybe the authors should clarify what they try to achieve with DA assimilation.
Figure S2 should be in the main manuscript. It could be together with Figure 7 as 7c, 7d, and 7e. Figures 7a and 7b don't need to be so large.
Line 427. "However, the direct DA experiments efficiently improved sharpness, thereby increasing confidence in the assimilated river discharge." I would say "FALSELY increasing confidence" as the authors just observed that the reliability drops more than 50% for direct DA experiments. What is the point of being narrower if the observation falls out of the confidence interval? I think the authors should be careful with that.
On tables 3 and 4 it would be nice to see the Open Loop and the CaMa VIC BC performances for comparison.
Line 541. HTESSEL not HTEESSEL.
Line 624. Which experiment is that? The one in section 4.2.? Or the one with VIC BC (section 3.3)?
Citation: https://doi.org/10.5194/egusphere-2022-412-RC2 -
AC2: 'Reply on RC2', Menaka Revel, 18 Aug 2022
Referee #2
The authors investigated the assimilation of satellite altimetry to improve discharge. The article is excellent. The authors presented 3 types of altimetry assimilation, which are 1. direct, 2. anomaly, and 3. normalized assimilation, and concluded that to improve discharge it is better to just assimilate the water surface dynamics (3.) given that the model is relatively accurate. If the model is completely corrupted, the authors concluded that anomaly and/or direct assimilation are more effective.
Reply:
We would like to express our gratitude to referee #2 for his thoughtful comments. We will address all the comments in the revised manuscript and detailed responses to the comments are given below.
1. My main complaint about this manuscript is in section 3.1 (specifically 3.1.1, 3.1.2, and 3.1.3) where the figures (fig 4, 5 and 6) showed hydrographs that were different from the locations discussed in the text. Besides that, I have just some small comments that are specified below:
Reply:
We would like to thank referee #2 for identifying this issue. We found that some panels of figures 4, 5, and 6 are not matching with the description. Therefore, we will revise figures 4, 5, and 6 to be compatible with the description in Sections 3.1.1, 3.1.2, and 3.1.3.
2. Line 40. I think it is better to change GHM definition only to Global Hydrodynamic Models instead of Hydrological due to some features the authors discuss further such as "runoff as a forcing factor", "discretized river", "surface water dynamics", etc.. Line 77 and 519 could be GHM instead of global hydrodynamic models.
Reply:
We would like to express our gratitude to referee #2 for his valuable suggestion. We agree with referee #2 that our introduction should be more focused on global hydrodynamic models. We will revise the manuscript according to the suggestion.
3. Line 52. The authors could also mention Laser altimetry. The ICESat missions are also very used in academic research. Maybe instead of a radar pulse, can be a radar/light pulse or even an electromagnetic pulse.
Reply: We would like to thank referee #2 for suggesting introducing laser missions. So, we will revise the text in the revised manuscript to address the referee's comments.
4. As an alternative to the semi-variogram analysis to determine the spatial dependency weights, the authors could have used "backwater lengths in rivers" studied by Samuel (1989). It can give an idea of which river reaches are affected by WSE variations at the VS locations. It would be a good idea to compare both approaches in future studies (not now).
Samuels, P. G. (1989). Backwater lengths in rivers. Proceedings of the Institution of Civil Engineers, 87(4), 571–582. https://doi.org/10.1680/iicep.1989.3779
Reply: We would like to express our gratitude to referee #2 for the suggestion. We will consider the suggestion in our future studies.
5. Line 156. using a power law dependent on what? Width? Upstream drainage area? Is it the same power law parameters for the whole basin?
Reply: We appreciate referee #2 for raising this question. The power law depends on a prior annual average river discharge and uses a single set of parameters (i.e., a and b) for the whole basin (a=0.1, b=0.5). The power law is shown in Equation 1 in the supplementary document. We will revise the text according to this comment.
6. Line 199. Something went wrong with the font size of some words. Line 351. Line 438. Line 637.
Reply: We would like to thank referee #2 for identifying those mistakes. We will correct all those kinds of font errors.
7. Line 229. So, the mean and standard deviation were calculated based on the open loop simulation?Reply: Yes, the mean and standard deviation were calculated based on the long-term open loop simulation (2000-2014).
8. Line 231 to 239. Some of your readers might be unfamiliar with the Amazon Basin. It would be interesting to write a short and objective section about this basin, presenting a DEM map at least (a mean Precipitation map would be nice too).Reply: We like to thank referee #2 for the great suggestion. We will be happy to include a description and a figure of the Amazon basin but considering the length of and the number of figures in the manuscript, we will add them to the supplementary material.
9. Line 275. How do you measure the relative sharpness and the difference in reliability? Line 383 should be here.Reply: We believe Line 279-283 describes the same ideas as Line 383. We will modify the text to enhance the understanding of the Interval Skill Score (ISS).
10. Several wrong references in section 3.1.Line 309. The authors said that the Santos Dumont gauge is in the Purus River, but in Figure 4 it says Jurua River.
Line 332. "Figure 5c–e displays hydrographs of the Jurua (Gaviao), Amazon (Manacapuru), and Negro (Serrinha) rivers" but in Figure 5 it says Manicore on the Madeira River, Aruma on the Purus River, and Sao Felipe on the Negro River.
Line 351. "The lower panels of Figure 6 illustrate flow dynamics along the Amazon mainstem (Sao Paulo De Olivenca; Figure 6c) and Japura (Vila Bittencourt; Figure 6d) and Negro (Curicuriari; Figure 6e) rivers." but in Figure 6 it is written "Hydrographs recorded at Humaita on the Madeira River, Santos Dumont on the Jurua River, and Canutama on the Purus River are presented on panels c, d, and e, respectively."
Reply: We would be grateful to the referee for identifying the mistake. We revised Figures 4, 5, and 6 corresponding to the description of sections 3.1.1., 3.1.2., and 3.1.3.
11. Line 321. Saying that the "direct DA generally improved flow dynamics" is very optimistic. Based on these results, I'd probably say that the direct DA maintained or even degraded the general performance, at least for discharge.
Reply: We would be thankful to the referee for the detailed comments on the text. We agree with the referee that the Direct DA method somewhat degraded the performance of discharge estimations in some locations. But the discharge was improved in some gauge locations in the Amazon basin such as Santo Antonio Do Ica and Sao Paulo de Olivenca. Therefore, we will revise the text to deliver the ideas suggested by referee #2.
12. Line 340. Once again, I think it is an optimistic conclusion. In the last sentence, the authors just said: "although NSE and ISS values worsened slightly." So how can the authors say afterward that "discharge estimates improved moderately"? I don't think that improvements in the correlation coefficient are enough for such a statement given that the NSE has become worst. But I reckon that seasonality got better as correlation got higher. Maybe the authors should clarify what they try to achieve with DA assimilation.
Reply: We would like to express our gratitude to referee #2. Our goal is to improve the overall performance of the river discharge hence higher NSE the better discharge estimated would be. But we try to look at the positives of each method and come to a broader conclusion as none of the methods performed perfectly in all the scenarios. Ideally, it is better if the direct DA performs better than others as assimilation will not depend on prior statistics of the open loop simulation. We basically discuss the median performance in the discussion of NSE and ISS here. But there is a large variation in the statistics as shown in figure 7, table 2 and figure S2. A reasonable number of gauges improved their performance by anomaly DA. An overall moderate number of gauges improved their discharge estimations. Therefore, we will modify the text to highlight the percentage of improved gauges.
13. Figure S2 should be in the main manuscript. It could be together with Figure 7 as 7c, 7d, and 7e. Figures 7a and 7b don't need to be so large.
Reply: We would like to appreciate the suggestion by referee #2. We will include all or some of the panels of Figure S2 in Figure 7.
14. Line 427. "However, the direct DA experiments efficiently improved sharpness, thereby increasing confidence in the assimilated river discharge." I would say "FALSELY increasing confidence" as the authors just observed that the reliability drops more than 50% for direct DA experiments. What is the point of being narrower if the observation falls out of the confidence interval? I think the authors should be careful with that.
Reply: We would like to thank referee #2 for pointing out this. We agree with the referee that “falsely” increasing the confidence will not be beneficial. In this sentence, we mean that when data assimilation is performed in direct values, the spread of final assimilated values will be narrower than in other methods because the assimilation was performed in direct DA. But not in anomaly or normalized value DA. Of course, the reliability should be higher. We will revise the sentence to convey the idea more clearly.
15. On tables 3 and 4 it would be nice to see the Open Loop and the CaMa VIC BC performances for comparison.Reply: Thank you very much for the nice suggestion. We will include the median statistics of the Open Loop and the CaMa VIC BC in tables 3 and 4 even though CaMa VIC BC cannot be assessed for sharpness.
16. Line 541. HTESSEL not HTEESSEL.Reply: Thank you for recognizing the mistake. We modified all the instances which have HTESSEL in the revised manuscript.
17. Line 624. Which experiment is that? The one in section 4.2.? Or the one with VIC BC (section 3.3)?
Reply: Thank you for recognizing the mistake. It should be VIC BC. We will correct the text in the revised manuscript.
Citation: https://doi.org/10.5194/egusphere-2022-412-AC2
-
AC2: 'Reply on RC2', Menaka Revel, 18 Aug 2022
-
RC4: 'Comment on egusphere-2022-412', Anonymous Referee #3, 31 Aug 2022
This manuscript explores different strategies to assimilate water surface elevation (WSE) derived from satellite altimetry into the CaMa-Flood global hydrodynamic model. During the recent years, a large number of studies have demonstrated the potential benefits of assimilating WSE for various purposes, such as parameter estimation or river dynamics modeling improvement. One of the main difficulties relies on the fact that WSE from altimetry provides the water elevation based on a reference (geoid) that can differ from the model reference (because of DEM errors for example) and the induced bias may highly degrade the assimilation performances. An alternative consists in considering WSE anomalies instead of absolute values. Another one, introduced in this study, also normalizes WSE anomalies to account for possible errors in the signal amplitude. Here the authors explore those three strategies and their respective performances in reproducing river discharge over the Amazon Basin. This scientific question is highly relevant and the study is well conducted, which makes it worth publishing, especially in the context of the forthcoming SWOT mission.
Overall, the manuscript is well written and well organized. The methodology is clearly stated and the results are quite convincing, although the latter should be revisited to correct some mistakes and clarify some points, as explained thereafter.
Major remarks
1. The quality of figures 4 to 6 is quite bad and colors are hard to identify (in the maps and in the time series). Also the time series subplots do not correspond to the description in the results section (3.1.1, 3.1.2 and 3.1.3). Hence it is not possible understand, confirm or refute the description of these plots, as well as the conclusions drawn (L309-323, L332-342, L351-362).
2. It is not clear to me which variables are included in the state vector x. I understand from L509 that the prognostic variable of CaMa-Flood is water storage. On the other hand, it is stated (L188) that the state vector includes river discharge, WSE, flooded area, flood height and storage. Could you clarify this point? Also, in the latter case, shouldn’t the observation operator H contain only zeros except for the column corresponding to WSE? Moreover, are all state variables (river discharge, WSE, etc.) converted to anomaly and normalized values as written in L193?
3. More importantly, I think that the analysis of DA performances when runoff forcing or bathymetry are biased (section 4.2) requires a bit more explanations.
- Model simulations are affected by biases (errors in absolute values) and by errors in dynamics. Decreasing the river bathymetry (by lowering the river bottom elevation) would lower the absolute WSE without impacting the flow dynamics, except if bank overflow occurs (flooding). If there is no flooding, I would have expected large impacts on the direct DA performances but no impact on anomaly and normalized value DA. Given that, do the results of Fig. 11b mean that the degradation of those two experiments is due to bank overflow? In addition, what can explain the very poor performances of normalized value DA method? Maybe some example time series could help better understand these results, as done for the previous experiments (perhaps as a supplement).
- Is there any possible explanation of the better normalized value DA performances with runoff bias and bathymetry error compared to performances with only bathymetry errors (panels b and d)?
- Finally, considering only one runoff (HTESSEL) to generate the ensemble reduces the dynamics variations between the members. Could this be a reason of the poor DA performances, especially with the normalized value DA method?
Each of the three DA methods can outperform the other two depending on the configuration, making the choice of the DA method quite difficult for further studies. I think providing more insight in these experiments might help readers better understand the pros and cons of each method.
Minor remarks
Fig. 1. In panel b, upper left square, it should be “Altimetry Auxillary Data”.
L134-137. “VSs with considerable variation in mean WSE compared to the MERIT Hydro (Yamazaki et al., 2017, 2019) elevation (expressed as riverbank height) were filtered through comparison of mean observations and riverbank heights.” What could be the cause(s) of such errors? Maybe the answer is given in L506-508.
L157. Is the river width from remote sensing available for every reaches of the river network? If not, how is it determined?
Fig. 2. I would suggest to add a legend in the time series, and maybe add error bars in observed WSE (from HydroWeb).
Eq. (2). Since H is linear, maybe it is better to write Hxk instead of H(xk).
L288. Sharpness and reliability are not defined.
L285. Nash and Sutcliffe (1970) and Kling and Gupta (2009) are cited several times, which is not necessary.
L313. It is written (but I cannot verify it) that the 95 % ensemble spread is improved until mid-2010, when the ENVISAT satellite was available. But this satellite is supposed to be available until 2012 (Tab. 1). Also, could you explain what an improvement in ensemble spread is? Is it a reduction of the spread?
L321. Considering numbers in L300-302, I would not say that “direct DA generally improved flow dynamic simulation to a moderate extent”: concerning river discharge, 8 % of gauges show an improvement while 43 % show a degradation.
L326. What could be the impact of choosing different time periods for observed and simulated WSE when computing the long term mean (and std)? For example if a multi-year drought is accounted for in one period and not in the other.
L336. It should be “improvements in r and ISS” not “in Dr and rISS”.
L348. Considering the quality of the figure and the color range, decreases in discharge correlation is not that evident in the Amazon mainstem.
L365. It is stated that the assimilation has very little influence outside the area of satellite observations. First, does the satellite coverage area correspond to the reaches downstream any VS? Second, shouldn’t the localization method used here allow to correct river discharge upstream VS?
Fig. 7. In the caption of a, it should be “probability distribution” instead of “cumulative distribution”. Also, it could be helpful to plot the vertical line at 0. Same remark for Fig. S5.
L385. “A large reduction in sharpness was observed in the direct assimilation experiment (Exp 1), mainly because the assimilation was conducted directly.” I do not see the link here, could you expand a bit more?
L408. For river discharge, sharpness is also considered (L415).
L530. Indeed, a huge potential from SWOT is expected in this kind of study. But how to deal with the need to compute long term mean and std for the derivation of anomalies and normalized values?
L536. Water height in the river is approximately 50 % lower, not WSE.
Fig. 12. The range of the y-axis could be reduced.
Minor remarks in supplementary material
Fig. S1. Square and circle are inverted in the legend.
L45. It should be “Exp 2a and Exp3a”.
L57. What is Exp 3b?
Fig. S7. It is hard to see the effect of DA on low flows. Maybe consider a log-log scatter plot?
Citation: https://doi.org/10.5194/egusphere-2022-412-RC4 -
AC3: 'Reply on RC3', Menaka Revel, 13 Sep 2022
We would like to express our gratitude to referee #1 for his continued interest in our study. We look forward to submitting the revised manuscript after the open discussion phase.
Citation: https://doi.org/10.5194/egusphere-2022-412-AC3
-
AC3: 'Reply on RC3', Menaka Revel, 13 Sep 2022
-
AC4: 'Comment on egusphere-2022-412', Menaka Revel, 13 Sep 2022
Referee #3
This manuscript explores different strategies to assimilate water surface elevation (WSE) derived from satellite altimetry into the CaMa-Flood global hydrodynamic model. During the recent years, a large number of studies have demonstrated the potential benefits of assimilating WSE for various purposes, such as parameter estimation or river dynamics modeling improvement. One of the main difficulties relies on the fact that WSE from altimetry provides the water elevation based on a reference (geoid) that can differ from the model reference (because of DEM errors for example) and the induced bias may highly degrade the assimilation performances. An alternative consists in considering WSE anomalies instead of absolute values. Another one, introduced in this study, also normalizes WSE anomalies to account for possible errors in the signal amplitude. Here the authors explore those three strategies and their respective performances in reproducing river discharge over the Amazon Basin. This scientific question is highly relevant and the study is well conducted, which makes it worth publishing, especially in the context of the forthcoming SWOT mission.
Overall, the manuscript is well written and well organized. The methodology is clearly stated and the results are quite convincing, although the latter should be revisited to correct some mistakes and clarify some points, as explained thereafter.
Reply:
We would like to convey our acknowledgment to referee #3 for his informative comments. We will address all the comments in the revised text, and responses are provided below.
Major remarks
1. The quality of figures 4 to 6 is quite bad and colors are hard to identify (in the maps and in the time series). Also the time series subplots do not correspond to the description in the results section (3.1.1, 3.1.2 and 3.1.3). Hence it is not possible understand, confirm or refute the description of these plots, as well as the conclusions drawn (L309-323, L332-342, L351-362).
Reply:
We would be grateful to referee #3 for raising this issue. When creating these figures, a mistake occurred. Therefore, we will revise figures 4-6 to correspond to the descriptions in sections 3.1.1, 3.1.2, and 3.1.3.
2. It is not clear to me which variables are included in the state vector x. I understand from L509 that the prognostic variable of CaMa-Flood is water storage. On the other hand, it is stated (L188) that the state vector includes river discharge, WSE, flooded area, flood height and storage. Could you clarify this point? Also, in the latter case, shouldn’t the observation operator H contain only zeros except for the column corresponding to WSE? Moreover, are all state variables (river discharge, WSE, etc.) converted to anomaly and normalized values as written in L193?
Reply:
We would like to express our gratitude to referee #3 for the insightful comment. We found that the using same H (observational operator) in both equations 2 and 3 is confusing. Equation 1 is about the CaMa-Flood model time evaluation and equation 2 is the conceptual relationship of CaMa-Flood state variables with the observations. So, the x (simple x) vector consists of CaMa-Flood state variables. Equation 3 is the analysis equation for data assimilation, there we used WSE only for data assimilation. Therefore, we think H in equation 3 is a subset of H in equation 2. Hence, we will revise the H in equation 2 to be (curly H) and will revise the text to explain that we used only WSE in the LETKF analysis equation.
3. More importantly, I think that the analysis of DA performances when runoff forcing or bathymetry are biased (section 4.2) requires a bit more explanations.
- Model simulations are affected by biases (errors in absolute values) and by errors in dynamics. Decreasing the river bathymetry (by lowering the river bottom elevation) would lower the absolute WSE without impacting the flow dynamics, except if bank overflow occurs (flooding). If there is no flooding, I would have expected large impacts on the direct DA performances but no impact on anomaly and normalized value DA. Given that, do the results of Fig. 11b mean that the degradation of those two experiments is due to bank overflow? In addition, what can explain the very poor performances of normalized value DA method? Maybe some example time series could help better understand these results, as done for the previous experiments (perhaps as a supplement).
Reply:
We would like to thank referee #3 for the valuable comment and suggestion. In these experiments, we try to assess the performance of DA methods in simplified error conditions in forcing and parameters of the model. We agree with referee #3 that including corrupted bathymetry will lower the absolute WSE. Hence, it affects WSE assimilation in direct DA.
But in the cases of anomaly and normalized value DA, the degradation of discharge accuracy is mainly due to the errors in the statistics (i.e., mean and standard deviation). These statistics were computed using long-term open-loop simulations. When the river bathymetry is corrupted the mean open-loop WSE will also be biased, and the standard deviation will be different. Hence the assimilation result will also be inaccurate.
- Is there any possible explanation of the better normalized value DA performances with runoff bias and bathymetry error compared to performances with only bathymetry errors (panels b and d)?
Reply:
We would like to express our gratitude to referee #3 for raising this question. With current experiments, the statistics of the normalized value DA method should be much lower because of the lowering of bathymetry as well as reducing the runoff. Hence, we need to investigate the experimental setting again whether there is an error in preparing Figure 11. We will re-check Figure 11 and add some explanation about such a trend in these experiments.
- Finally, considering only one runoff (HTESSEL) to generate the ensemble reduces the dynamics variations between the members. Could this be a reason of the poor DA performances, especially with the normalized value DA method?
Reply:
We would like to express our sincere thanks to referee #3. We don’t think the poor performance of the normalized value DA method in corrupted bathymetry or runoff error cases is due to low variation in perturbations. It is mostly due to the bias and differences in statics.
4. Each of the three DA methods can outperform the other two depending on the configuration, making the choice of the DA method quite difficult for further studies. I think providing more insight in these experiments might help readers better understand the pros and cons of each method.
Reply:
We would like to appreciate referee #3 for the great comment. We will add some more recommendations and improve the current description to enhance these ideas.
Minor remarks
1. In panel b, upper left square, it should be “Altimetry Auxillary Data”.
Reply: Thank you for your recognition of the error. We will correct the text accordingly.
2. L134-137. “VSs with considerable variation in mean WSE compared to the MERIT Hydro (Yamazaki et al., 2017, 2019) elevation (expressed as riverbank height) were filtered through comparison of mean observations and riverbank heights.” What could be the cause(s) of such errors? Maybe the answer is given in L506-508.
Reply: We would like to thank referee #3 for raising this question. From our analysis, we found that most erroneous VSs are in narrower rivers at high elevations. There can be several reasons for these, 1. Non-nadir direction observations, 2. Errors in post-processing of VS (e.g., geoid conversion), etc.
3. Is the river width from remote sensing available for every reaches of the river network? If not, how is it determined?
Reply: We thank referee #3 for the question. Remote sensing river widths were used in the river reaches with river width > 300m (Yamazaki et al., 2014). A power law relationship of the average river discharge was used to estimate the river width for the other smaller river reaches. We will revise the text to reflect these ideas.
4. Fig. 2. I would suggest to add a legend in the time series, and maybe add error bars in observed WSE (from HydroWeb).
Reply: Thank you very much for the suggestion. We will revise Fig 2 according to referee #3’s suggestion.
5. Eg. (2). Since H is linear, maybe it is better to write Hxk instead of H(xk).
Reply: Thank you for the suggestion. Here H is more of an operator for converting simulated variables to observable variables. Not all the simulated variables are possible to observe. So, we would like to keep the equation as it is, but we will use two symbols for equations 2 and 3 because H in equation 3 is a subset of H in equation 2.
6. Sharpness and reliability are not defined.
Reply: Thank you for the suggestion. We will define sharpness and reliability in the methodology section.
7. Nash and Sutcliffe (1970) and Kling and Gupta (2009) are cited several times, which is not necessary.
Reply: Thank you. We will remove those citations.
8. It is written (but I cannot verify it) that the 95 % ensemble spread is improved until mid-2010, when the ENVISAT satellite was available. But this satellite is supposed to be available until 2012 (Tab. 1). Also, could you explain what an improvement in ensemble spread is? Is it a reduction of the spread?
Reply: We appreciate referee #3 for raising this issue. ENVISAT nominal period is 2002-2010 after 2010 ENVISAT track was changed. Therefore, ENVISAT data after 2010 was not used in HydroWeb. Hence, we will revise Table 1, we will add only the nominal periods used in HydroWeb for each satellite.
An improvement in ensemble spread is referred to as the reduction of spread or reduction of sharpness. We think “improvement in ensemble spread” is better be revised as an improvement in sharpness. Hence, we will revise the manuscript according to referee #3’s comments
9. Considering numbers in L300-302, I would not say that “direct DA generally improved flow dynamic simulation to a moderate extent”: concerning river discharge, 8 % of gauges show an improvement while 43 % show a degradation.
Reply: We would like to express our gratitude to referee #3 for pointing out this. It is evident that many gauges degraded their accuracy of river discharge by the direct DA method. However, some gauges such as Santo Antonio Do Ica and Sao Paulo de Olivenca improved their discharge accuracy. Hence, we will revise the text according to the comment of referee #3 to reflect that the river discharge estimated of some gauges show improvement.
10. What could be the impact of choosing different time periods for observed and simulated WSE when computing the long term mean (and std)? For example if a multi-year drought is accounted for in one period and not in the other.
Reply: We would like to thank referee #3 for the question. There can be not much effect of using different time periods for statistics for simulation and observations if the statistics are representative statistics of the long-term values. But if the statics are largely different from the observed statistics as shown in the biased runoff experiment case, the accuracy of estimated discharge by the anomaly and normalized value DA can be hampered. We checked sensitivity on the simulation WSE statistics where we found that the statistics calculated using simulation of 5 years or more are reasonably reproduced river discharges.
11. It should be “improvements in r and ISS” not “in Dr and rISS”.
Reply: Thanking referee #3, we will revise the text.
12. Considering the quality of the figure and the color range, decreases in discharge correlation is not that evident in the Amazon mainstem.
Reply: We would like to thank referee #3 for the comment. We will improve the quality of the figures in the revised manuscript. We agree with referee #3 that the correlation in the Amazon mainstream is quite good. This may be because model parameters (e.g., bank full height, river bathymetry, etc.) are still very good on these river reaches.
13. It is stated that the assimilation has very little influence outside the area of satellite observations. First, does the satellite coverage area correspond to the reaches downstream any VS? Second, shouldn’t the localization method used here allow to correct river discharge upstream VS?
Reply: We would like to express our gratitude to referee #3 for the great question. Here, we define the satellite coverage area as the river reaches located downstream of the most upstream VS in each tributary (green circles in Figure S1 indicate GRDC locations that were in the satellite coverage area). Hence, the satellite coverage areas cover downstream of all the VSs. Although using the localization method, the WSE can be updated without using any local observation. But in the upper reaches (narrow rivers with small catchment areas), the size of the local patches is small (Revel et al., 2019). Hence, assimilation efficiency can be lower in these upper reaches.
14. Fig. 7. In the caption of a, it should be “probability distribution” instead of “cumulative distribution”. Also, it could be helpful to plot the vertical line at 0. Same remark for Fig. S5.
Reply: We would like to thank referee #3 for the suggestion. We will revise the caption of Fig. 7 and Fig S5 according to referee #3’s suggestion.
15. “A large reduction in sharpness was observed in the direct assimilation experiment (Exp 1), mainly because the assimilation was conducted directly.” I do not see the link here, could you expand a bit more?
Reply: Thank you very much for asking for calcification. Sharpness reduction means the reduction of the ensemble spread. When the assimilation was performed in real values the ensemble spread become smaller than it was performed in anomaly or normalized values. We will revise the text to reflect this meaning in the revised manuscript.
16. For river discharge, sharpness is also considered (L415).
Reply: Thank you very much. We indeed calculated sharpness for river discharge as well even though we did not include it in Fig 8. The median sharpness values were shown in Table 3.
17. Indeed, a huge potential from SWOT is expected in this kind of study. But how to deal with the need to compute long term mean and std for the derivation of anomalies and normalized values?
Reply: We would like to thank referee #3. In this study, we did not assess the possibilities of using anomaly or normalized DA methods for the real-time forecast or usage with short-term observation records. However, a representative estimate for SWOT observations can provide reasonable discharge estimates. We will assess the sensitivity of the observational statistics which can be used for the real-time/short-term forecast in our future studies.
18. Water height in the river is approximately 50 % lower, not WSE.
Reply: Thank you very much for pointing out this. We will revise the text accordingly.
19. Fig. 12. The range of the y-axis could be reduced.
Reply: Thank you very much. We will revise Fig 12.
Minor remarks in supplementary material
1. Fig. S1. Square and circle are inverted in the legend.
Reply: Thank you very much. We will correct it.
2. It should be “Exp 2a and Exp3a”.
Reply: Thank you for pointing this out. We will revise it.
3. What is Exp 3b?
Reply: We would like to thank referee #3. Exp 3b should be removed. We will revise it.
4. Fig. S7. It is hard to see the effect of DA on low flows. Maybe consider a log-log scatter plot?
Reply: Thank you very much for the comment. We will revise Fig. S7 accordingly.
Reference:
- Revel, Ikeshima, Yamazaki and Kanae: A Physically Based Empirical Localization Method for Assimilating Synthetic SWOT Observations of a Continental-Scale River: A Case Study in the Congo Basin, Water, 11(4), 829, doi:10.3390/w11040829, 2019.
- Yamazaki, D., O’Loughlin, F., Trigg, M. A., Miller, Z. F., Pavelsky, T. M. and Bates, P. D.: Development of the Global Width Database for Large Rivers, Water Resour. Res., 50(4), 3467–3480, doi:10.1002/2013WR014664, 2014.
Citation: https://doi.org/10.5194/egusphere-2022-412-AC4
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2022-412', Anonymous Referee #1, 20 Jul 2022
General comment:
In this research, the authors performed data assimilation (DA) experiments to explore the capacity to improve daily river discharge within current limitations of global hydrodynamic modeling. For this purpose, the water surface elevation (WSE) from satellite altimetry was assimilated in a configuration of three experiments, the direct (absolute values), the anomalies and the normalized anomalies. The authors also evaluated the capability of these DA experiments in some scenarios, for instance when some parameters/forcing (river bathymetry and runoff) of the hydrodynamic model are biased, as well as conditions when river bathymetry is calibrated. The results showed that, in general, the normalized DA performance was the best, improving the daily discharge estimates in up to almost 60% of the stations evaluated, compared to the simulation without DA. These results considering the current limited conditions of the global hydrodynamic models (e.g. without calibration).
The major contribution of this research is the evaluation of these experiments and scenarios, providing adequate knowledge and insights in terms of how DA techniques could be used to improve discharge estimates, which fits with the perspectives of SWOT missions for example. In general, this work is worth publishing in the Hydrology and Earth System Sciences journal, however, it needs “moderate revisions”. Some suggestions for revisions are as follows.
Moderate comments:
- The article is well written and follows a logical order. Although it is a bit long, an average reader can follow the reading, however at a certain point there are more experiments than initially described. For example, the evaluation of DA under biased runoff and river bathymetry conditions; DA under calibrated river bathymetry conditions; DA using the runoff forcing of a bias-corrected model. That is why I recommend the authors to describe more explicitly these experiments in section 2.6.
- Regarding the selection of virtual stations (VSs) for assimilation or validation, the justification is a bit vague, even though this may be important for the performance of the experiments, so I recommend improving this point.
- In sections 3.1.1, 3.1.2 and 3.1.3 take care with the description of the time series in figures 4, 5 and 6, respectively. There is a confusion between the description of the gauges results. For example, line 315 describes the Santos Dumont station on the Purus river, however the series in Figure 4d are from a gauge on the Juruá river. This confusion occurs for the stations Gaviao (Juruá) and Manacapurú (Amazon) in Figure 5, and in all the gauges in Figure 6. This was probably an involuntary error in the preparation of the figures, please correct.
- Experiments that assimilate absolute (direct) values, anomalies and normalized anomalies are referred to by the acronyms Exp. 1, Exp. 2 and Exp. 3 respectively, however throughout the manuscript both nomenclatures are used. I suggest that only one be adopted to improve the readability of the text. Even so that the information can be quickly abstracted by the reader these experiments could be called DIR_DA, ANOM_DA and NORM_DA for example.
- Since the authors have used a localization method in the DA scheme, I suggest reinforcing the discussion on how this might affect discharge estimates due to assimilation of WSE within or outside the influence coverage of the VSs.
- Could you discuss a bit about to what do you attribute the lower efficiency of flow estimates in the upper Solimoes River ? efficiency of the CaMa-Flood model ? selection of VSs ? localization ? large uncertainties in the VSs data in that region ?
Specific comments (Line-by-line comments):
Introduction:
- 35: It is more accurate to say, "River discharge records can be used...".
- 42: I would say that also these simulations (of GHMs) have been used to complement observed records.
- 48: If we go deeper we could say that these forcing factors can also be rainfall and climatic variables.
- 50: I would say: "Given the current limitations of GHMs and in-situ measurements, ...".
- 57: This sentence mentioning the SWOT mission seems a bit loose, you should rework it to integrate it with what you want to mention above.
- 62: Typo: "combining" instead of "combing".
- 83: This statement describes information repeated in the previous one, perhaps you could combine them.
Methodology:
- 124: In this sentence you can already start reporting on the period of DA experiments (2009-2014).
- 215: Why wasn't the SURFEX-TRIP model outputs used since it also belongs to WRR2?
- 236: To reference these annual average rainfall values you can cite Builes-Jaramillo & Poveda, 2018; Espinoza et al., 2009. (https://doi.org/10.1029/2017WR021338 and https://doi.org/10.1002/joc.1791)
- 237: Please specify what you mean by large number of observations, perhaps this is valid for remote sensing observations because it is a large basin with strong hydrological signals, hence the citation of Fassoni-Andrade et al. 2021.
- 249: You could elaborate a little more on this sentence. Why these virtual stations could affect the estimates using assimilation? this exclusion of 3% was by a visual analysis of the series only? these stations are located in some particular place in the Amazon, maybe rivers with a small width?
- Sections 2.7.1 and 2.7.2 could be merged, as it could confuse the reader. The main objective of this research is to evaluate the performance in simulating daily discharge but here also the performance of WSE will be evaluated. This merged section could be called "observational data" since the altimetry data has also been used for validation.
Results:
- 294-296: This sentence seems to be repetitive with the previous one, you could merge them.
- 302: It would be appropriate to refer to Figure 4b in this sentence.
- 309: The time series for the Santos Dumont station is not shown in Figure 4d. Instead, a station on the Juruá River is shown. See my major comments above.
- 325: The information in parentheses should go in the methodology section.
- 330: “WSE performance decreased…” instead of “WSE decreased…”.
- 332: The Gavião and Manacapuru gauges do not correspond to Figures 5c and d.
- 351-352: None of these described gauges correspond to figures 6c, d and e.
- 412: I think there is a typo, please delete "3.2.1.".
- 8: It is not possible to distinguish gauges inside or outside the coverage area of the altimetric satellites. Could you differentiate them somehow?
- Table 3. I have noticed that some values in this table do not correspond exactly to those described. For instance, in the first column (All and r) in the table, the values are 0.74, 0.85 and 0.84 for experiments 1, 2 and 3 respectively. While in the description the values are 0.73, 0.84 and 0.83 (L. 431, L. 415 and L. 439 respectively).
- 452: As shown in Figure 9, the BIAS values are only positive, so I recommend describing somewhere (probably methodology) that the index is an absolute value of BIAS.
- Figure 9: It is a bit difficult to differentiate the VSs that were used for assimilation and validation. Perhaps it could be improved by changing the symbology from "o" to "*", increasing the size of the maps by reducing the space between them and decreasing a little the size of the station symbols so that they do not overlap too much. This is just a suggestion.
- Section 3.3: Please detail how in this experiment you have generated the realizations of the set for assimilation.Was it with the same perturbation as for the WRR2 models?
- 476-478: The end of this sentence sounds strange, I suggest to redo it or delete this last part from "...,direct DA (Exp 1)...."
Conclusions:
- 624: Typo, it's HTESSEL not HTEESSEL (same for L.541, 542 and 544).
Citation: https://doi.org/10.5194/egusphere-2022-412-RC1 -
AC1: 'Reply on RC1', Menaka Revel, 18 Aug 2022
Referee #1
General comment:
In this research, the authors performed data assimilation (DA) experiments to explore the capacity to improve daily river discharge within current limitations of global hydrodynamic modeling. For this purpose, the water surface elevation (WSE) from satellite altimetry was assimilated in a configuration of three experiments, the direct (absolute values), the anomalies and the normalized anomalies. The authors also evaluated the capability of these DA experiments in some scenarios, for instance when some parameters/forcing (river bathymetry and runoff) of the hydrodynamic model are biased, as well as conditions when river bathymetry is calibrated. The results showed that, in general, the normalized DA performance was the best, improving the daily discharge estimates in up to almost 60% of the stations evaluated, compared to the simulation without DA. These results considering the current limited conditions of the global hydrodynamic models (e.g. without calibration).
The major contribution of this research is the evaluation of these experiments and scenarios, providing adequate knowledge and insights in terms of how DA techniques could be used to improve discharge estimates, which fits with the perspectives of SWOT missions for example. In general, this work is worth publishing in the Hydrology and Earth System Sciences journal, however, it needs “moderate revisions”. Some suggestions for revisions are as follows.
Reply:
We would like to thank the referee for his kind remarks. We will address all the comments in the revised manuscript, and comprehensive explanations are provided below.
Moderate comments:
1. The article is well written and follows a logical order. Although it is a bit long, an average reader can follow the reading, however at a certain point there are more experiments than initially described. For example, the evaluation of DA under biased runoff and river bathymetry conditions; DA under calibrated river bathymetry conditions; DA using the runoff forcing of a bias-corrected model. That is why I recommend the authors to describe more explicitly these experiments in section 2.6.
Reply:
Thank you very much for the recommendation. We will include brief descriptions of the additional experiments in section 2.6 considering the length of the manuscript.
2. Regarding the selection of virtual stations (VSs) for assimilation or validation, the justification is a bit vague, even though this may be important for the performance of the experiments, so I recommend improving this point.
Reply:
We would like to express our gratitude to referee #1. We simply separated VSs to assimilation and validation 80%, and 20%, respectively. We will revise the description to reflect the selection of the VS for assimilation and validation
3. In sections 3.1.1, 3.1.2 and 3.1.3 take care with the description of the time series in figures 4, 5 and 6, respectively. There is a confusion between the description of the gauges results. For example, line 315 describes the Santos Dumont station on the Purus river, however the series in Figure 4d are from a gauge on the Juruá river. This confusion occurs for the stations Gaviao (Juruá) and Manacapurú (Amazon) in Figure 5, and in all the gauges in Figure 6. This was probably an involuntary error in the preparation of the figures, please correct.
Reply:
We thank referee #1 for recognizing the mistake. It was indeed an involuntary error in preparing the figures. We will correct the cross-reference with the corresponding figure and the text. Furthermore, we will correct the manuscript according to the referee's comment. Basically, we will replace Figures 4, 5, and 6 to match the descriptions in Sect. 3.1.1, 3.1.2, and 3.1.3, respectively.
4. Experiments that assimilate absolute (direct) values, anomalies and normalized anomalies are referred to by the acronyms Exp. 1, Exp. 2 and Exp. 3 respectively, however throughout the manuscript both nomenclatures are used. I suggest that only one be adopted to improve the readability of the text. Even so that the information can be quickly abstracted by the reader these experiments could be called DIR_DA, ANOM_DA and NORM_DA for example.
Reply:
We appreciate the suggestion from referee #1. We will adopt a better naming convention for the experiments in the revised manuscript such as DIR, ANO, and NOM.
5. Since the authors have used a localization method in the DA scheme, I suggest reinforcing the discussion on how this might affect discharge estimates due to assimilation of WSE within or outside the influence coverage of the VSs.
Reply:
We would like to thank the great suggestion by the referee. The localization method we used in this study is an adaptive localization method (Revel et al., 2019) which is far different from conventional localization methods which use fixed square-shaped local patches. The comparison between those methods can be found in Revel et al., (2019). The adaptive localization method recognizes the highly correlated areas and removes less correlated areas (e.g., small downstream tributaries). We will highlight the effect of adaptive localization in the revised manuscript.
6. Could you discuss a bit about to what do you attribute the lower efficiency of flow estimates in the upper Solimoes River ? efficiency of the CaMa-Flood model ? selection of VSs ? localization ? large uncertainties in the VSs data in that region ?
Reply:We would like to thank referee #1 for raising the question. The WSE simulations using CaMa-Food of these two locations show a reasonable agreement with the satellite altimetry observations, therefore direct DA method performed better. On the other hand, anomaly and normalized value DA have a limitation that the assimilated WSE is dependent on the statistics (i.e., mean and standard deviation) of open loop simulation. If statistics may not represent the actual values of mean WSE and the standard deviation of WSE, the assimilated WSE can be also different from the actual values in the anomaly and normalized assimilation method. When the relationship between the WSE and discharge (i.e., rating curve) is well represented in the model, the anomaly and normalized DA may not correctly estimate the river discharge if the statistics are biased. We will add some explanation to the manuscript regarding this.
Specific comments (Line-by-line comments):
Introduction:
1. 35: It is more accurate to say, "River discharge records can be used...".
Reply: We will revise the sentence as per referee #1’s comment.
2. 42: I would say that also these simulations (of GHMs) have been used to complement observed records.
Reply: We agree with referee #1 that the simulations of GHSs have been used to complement observations rather than compensate. Hence, we will revise the text to reflect the referee's comment.
3. 48: If we go deeper we could say that these forcing factors can also be rainfall and climatic variables.
Reply: We concur with referee #1 that the forcing factors are primarily rainfall and climatic variables. Therefore, we will modify the text to include rainfall and climatic variables.
4. 50: I would say: “Given the current limitations of GHMs and in-situ measurements, …”.
Reply: We admit referee #1’s suggestion that both GHMS and in-situ measurements have their own limitations. It will be updated in the revised manuscript.
5. 57: This sentence mentioning the SWOT mission seems a bit loose, you should rework it to integrate it with what you want to mention above.
Reply: The sentence will be revised to reflect the referee #1’s comment.
6. 62: Typo: "combining" instead of "combing".
Reply: We will correct it.
7. 83: This statement describes information repeated in the previous one, perhaps you could combine them.
Reply: By thanking referee #1, the paragraph will be modified.
Methodology:
1. 124: In this sentence you can already start reporting on the period of DA experiments (2009-2014).
Reply: We would like to thank Referee #1 for the nice suggestion. But we try to dedicate this section (2.1) solely to describing the assimilation framework. So, we will introduce the period of the experiments in section 2.6.
2. 215: Why wasn't the SURFEX-TRIP model outputs used since it also belongs to WRR2?
Reply: We would like to thank referee #1 for the question. We did not consider outputs from the SURFEX-TRIP model because those are not compatible with the CaMa-Flood model. The SURFEX-TRIP model consists of the capillary rise in the runoff variable where CaMa-Flood is not capable of dealing with such runoff data. Therefore, we will add some descriptions to the text.
3. 236: To reference these annual average rainfall values you can cite Builes-Jaramillo & Poveda, 2018; Espinoza et al., 2009. (https://doi.org/10.1029/2017WR021338 and https://doi.org/10.1002/joc.1791)
Reply: We would like to express our gratitude to referee #1. We will use the above studies when describing the annual average rainfall in the Amazon basin and the text will be modified considering the above studies.
4. 237: Please specify what you mean by large number of observations, perhaps this is valid for remote sensing observations because it is a large basin with strong hydrological signals, hence the citation of Fassoni-Andrade et al. 2021.
Reply: We thank referee #1 for raising this issue. Yes, indeed we refer to remote sensing data in the Amazon basin. So, we will revise the text of the manuscript.
5. 249: You could elaborate a little more on this sentence. Why these virtual stations could affect the estimates using assimilation? this exclusion of 3% was by a visual analysis of the series only? these stations are located in some particular place in the Amazon, maybe rivers with a small width?
Reply: We would like to express our gratitude to referee #1 for the important question. We did some extensive analysis of these largely biased VS which cannot be compared with the MERIT DEM (Yamazaki et al., 2017). This exclusion was done by comparing the mean WSE of satellite altimetry observation with the MERIT DEM, not on visual analysis. Our analysis revealed that most of the “biased VS” were in narrow rivers at relatively high elevations. We will include some descriptions of these biased VS in the revised manuscript.
6. Sections 2.7.1 and 2.7.2 could be merged, as it could confuse the reader. The main objective of this research is to evaluate the performance in simulating daily discharge but here also the performance of WSE will be evaluated. This merged section could be called "observational data" since the altimetry data has also been used for validation.
Reply: We appreciate the comment on merging Sections 2.7.1 and 2.7.2. We will revise it in the revised manuscript.
Results:
1. 294-296: This sentence seems to be repetitive with the previous one, you could merge them.
Reply: We would like to thank referee #1 for the suggestion and we will revise the text.
2. 302: It would be appropriate to refer to Figure 4b in this sentence.
Reply: Thank you very much for pointing out this. We will refer to figure 4d in the sentence.
3. 309: The time series for the Santos Dumont station is not shown in Figure 4d. Instead, a station on the Juruá River is shown. See my major comments above.
Reply: Thank you very much for the comment. We will revise Figure 4d to be compatible with the description in the text.
4. 325: The information in parentheses should go in the methodology section.
Reply: Thank you for the kind suggestion. We will include that in the methodology section.
5. 330: “WSE performance decreased…” instead of “WSE decreased…”.
Reply: Thank you very much. We will revise it in the revised manuscript.
6. 332: The Gavião and Manacapuru gauges do not correspond to Figures 5c and d.
Reply: Thank you. We will revise figures 5c and d to match the description in Section 3.1.2.
7. 351-352: None of these described gauges correspond to figures 6c, d and e.
Reply: Thank you for raising this error. We will replace the correct figures for figures 6c, d, and e and the description will be modified to correspond to figures 6c, d, and e in Section 3.1.3.
8. 412: I think there is a typo, please delete "3.2.1.".
Reply: Thank you for pointing out it. We will delete "3.2.1.".
9. 8: It is not possible to distinguish gauges inside or outside the coverage area of the altimetric satellites. Could you differentiate them somehow?
Reply: Thank you for the great suggestion. We will modify figure 8 to distinguish the gauges inside and outside satellite altimetry coverage.
10. Table 3. I have noticed that some values in this table do not correspond exactly to those described. For instance, in the first column (All and r) in the table, the values are 0.74, 0.85 and 0.84 for experiments 1, 2 and 3 respectively. While in the description the values are 0.73, 0.84 and 0.83 (L. 431, L. 415 and L. 439 respectively).
Reply: Thank you for cross-checking the description with the data provided in the tables. The values in the table are correct values so we will revise the text according to Table 3. We will revise the L. 431, L. 435, and L. 439 with the values of 0.74, 0.85, and 0.84.
11. 452: As shown in Figure 9, the BIAS values are only positive, so I recommend describing somewhere (probably methodology) that the index is an absolute value of BIAS.
Reply: Thank you for pointing out this. We will add “absolute bias” to the methodology section when introducing “BIAS” term as follows:
12. Figure 9: It is a bit difficult to differentiate the VSs that were used for assimilation and validation. Perhaps it could be improved by changing the symbology from "o" to "*", increasing the size of the maps by reducing the space between them and decreasing a little the size of the station symbols so that they do not overlap too much. This is just a suggestion.
Reply: Thank you very much for your precious suggestion. We will revise figure 9 and other similar figures to better represent symbols.
13. Section 3.3: Please detail how in this experiment you have generated the realizations of the set for assimilation.Was it with the same perturbation as for the WRR2 models?
Reply: Thank you for asking for clarification. This section used the same assimilation outputs from the direct, anomaly, and normalized DA experiments. The perturbations are similar to WRR2 model outputs.
14. 476-478: The end of this sentence sounds strange, I suggest to redo it or delete this last part from "...,direct DA (Exp 1)...."
Reply: Thank you. We will revise the sentence in the revised manuscript.
Conclusions:1. 624: Typo, it's HTESSEL not HTEESSEL (same for L.541, 542 and 544).
Reply: Thank you, We will correct them to HTESSEL.
Reference:1. Revel, Ikeshima, Yamazaki and Kanae: A Physically Based Empirical Localization Method for Assimilating Synthetic SWOT Observations of a Continental-Scale River: A Case Study in the Congo Basin, Water, 11(4), 829, doi:10.3390/w11040829, 2019.
2. Yamazaki, D., Ikeshima, D., Tawatari, R., Yamaguchi, T., O’Loughlin, F., Neal, J. C., Sampson, C. C., Kanae, S. and Bates, P. D.: A high-accuracy map of global terrain elevations, Geophys. Res. Lett., 44(11), 5844–5853, doi:10.1002/2017GL072874, 2017.Citation: https://doi.org/10.5194/egusphere-2022-412-AC1 -
RC3: 'Reply on AC1', Anonymous Referee #1, 30 Aug 2022
I have reviewed in detail the authors' responses to my comments in the first round of revisions. It appears that the authors were able to resolve and address all the corrections. I will be waiting for the final version of the main document (file I have not yet received) before I can make a decision. Thanks
Citation: https://doi.org/10.5194/egusphere-2022-412-RC3 -
AC3: 'Reply on RC3', Menaka Revel, 13 Sep 2022
We would like to express our gratitude to referee #1 for his continued interest in our study. We look forward to submitting the revised manuscript after the open discussion phase.
Citation: https://doi.org/10.5194/egusphere-2022-412-AC3
-
AC3: 'Reply on RC3', Menaka Revel, 13 Sep 2022
-
RC3: 'Reply on AC1', Anonymous Referee #1, 30 Aug 2022
-
RC2: 'Comment on egusphere-2022-412', Anonymous Referee #2, 30 Jul 2022
The authors investigated the assimilation of satellite altimetry to improve discharge. The article is excellent. The authors presented 3 types of altimetry assimilation, which are 1. direct, 2. anomaly, and 3. normalized assimilation, and concluded that to improve discharge it is better to just assimilate the water surface dynamics (3.) given that the model is relatively accurate. If the model is completely corrupted, the authors concluded that anomaly and/or direct assimilation are more effective.
My main complaint about this manuscript is in section 3.1 (specifically 3.1.1, 3.1.2, and 3.1.3) where the figures (fig 4, 5 and 6) showed hydrographs that were different from the locations discussed in the text. Besides that, I have just some small comments that are specified below:
Line 40. I think it is better to change GHM definition only to Global Hydrodynamic Models instead of Hydrological due to some features the authors discuss further such as "runoff as a forcing factor", "discretized river", "surface water dynamics", etc.. Line 77 and 519 could be GHM instead of global hydrodynamic models.
Line 52. The authors could also mention Laser altimetry. The ICESat missions are also very used in academic research. Maybe instead of a radar pulse, can be a radar/light pulse or even an electromagnetic pulse.
As an alternative to the semi-variogram analysis to determine the spatial dependency weights, the authors could have used "backwater lengths in rivers" studied by Samuel (1989). It can give an idea of which river reaches are affected by WSE variations at the VS locations. It would be a good idea to compare both approaches in future studies (not now).
Samuels, P. G. (1989). Backwater lengths in rivers. Proceedings of the Institution of Civil Engineers, 87(4), 571–582. https://doi.org/10.1680/iicep.1989.3779
Line 156. using a power law dependent on what? Width? Upstream drainage area? Is it the same power law parameters for the whole basin?
Line 199. Something went wrong with the font size of some words. Line 351. Line 438. Line 637.
Line 229. So, the mean and standard deviation were calculated based on the open loop simulation?
Line 231 to 239. Some of your readers might be unfamiliar with the Amazon Basin. It would be interesting to write a short and objective section about this basin, presenting a DEM map at least (a mean Precipitation map would be nice too).
Line 275. How do you measure the relative sharpness and the difference in reliability? Line 383 should be here.
Several wrong references in section 3.1.
Line 309. The authors said that the Santos Dumont gauge is in the Purus River, but in Figure 4 it says Jurua River.
Line 332. "Figure 5c–e displays hydrographs of the Jurua (Gaviao), Amazon (Manacapuru), and Negro (Serrinha) rivers" but in Figure 5 it says Manicore on the Madeira River, Aruma on the Purus River, and Sao Felipe on the Negro River.
Line 351. "The lower panels of Figure 6 illustrate flow dynamics along the Amazon mainstem (Sao Paulo De Olivenca; Figure 6c) and Japura (Vila Bittencourt; Figure 6d) and Negro (Curicuriari; Figure 6e) rivers." but in Figure 6 it is written "Hydrographs recorded at Humaita on the Madeira River, Santos Dumont on the Jurua River, and Canutama on the Purus River are presented on panels c, d, and e, respectively."
Line 321. Saying that the "direct DA generally improved flow dynamics" is very optimistic. Based on these results, I'd probably say that the direct DA maintained or even degraded the general performance, at least for discharge.
Line 340. Once again, I think it is an optimistic conclusion. In the last sentence, the authors just said: "although NSE and ISS values worsened slightly." So how can the authors say afterward that "discharge estimates improved moderately"? I don't think that improvements in the correlation coefficient are enough for such a statement given that the NSE has become worst. But I reckon that seasonality got better as correlation got higher. Maybe the authors should clarify what they try to achieve with DA assimilation.
Figure S2 should be in the main manuscript. It could be together with Figure 7 as 7c, 7d, and 7e. Figures 7a and 7b don't need to be so large.
Line 427. "However, the direct DA experiments efficiently improved sharpness, thereby increasing confidence in the assimilated river discharge." I would say "FALSELY increasing confidence" as the authors just observed that the reliability drops more than 50% for direct DA experiments. What is the point of being narrower if the observation falls out of the confidence interval? I think the authors should be careful with that.
On tables 3 and 4 it would be nice to see the Open Loop and the CaMa VIC BC performances for comparison.
Line 541. HTESSEL not HTEESSEL.
Line 624. Which experiment is that? The one in section 4.2.? Or the one with VIC BC (section 3.3)?
Citation: https://doi.org/10.5194/egusphere-2022-412-RC2 -
AC2: 'Reply on RC2', Menaka Revel, 18 Aug 2022
Referee #2
The authors investigated the assimilation of satellite altimetry to improve discharge. The article is excellent. The authors presented 3 types of altimetry assimilation, which are 1. direct, 2. anomaly, and 3. normalized assimilation, and concluded that to improve discharge it is better to just assimilate the water surface dynamics (3.) given that the model is relatively accurate. If the model is completely corrupted, the authors concluded that anomaly and/or direct assimilation are more effective.
Reply:
We would like to express our gratitude to referee #2 for his thoughtful comments. We will address all the comments in the revised manuscript and detailed responses to the comments are given below.
1. My main complaint about this manuscript is in section 3.1 (specifically 3.1.1, 3.1.2, and 3.1.3) where the figures (fig 4, 5 and 6) showed hydrographs that were different from the locations discussed in the text. Besides that, I have just some small comments that are specified below:
Reply:
We would like to thank referee #2 for identifying this issue. We found that some panels of figures 4, 5, and 6 are not matching with the description. Therefore, we will revise figures 4, 5, and 6 to be compatible with the description in Sections 3.1.1, 3.1.2, and 3.1.3.
2. Line 40. I think it is better to change GHM definition only to Global Hydrodynamic Models instead of Hydrological due to some features the authors discuss further such as "runoff as a forcing factor", "discretized river", "surface water dynamics", etc.. Line 77 and 519 could be GHM instead of global hydrodynamic models.
Reply:
We would like to express our gratitude to referee #2 for his valuable suggestion. We agree with referee #2 that our introduction should be more focused on global hydrodynamic models. We will revise the manuscript according to the suggestion.
3. Line 52. The authors could also mention Laser altimetry. The ICESat missions are also very used in academic research. Maybe instead of a radar pulse, can be a radar/light pulse or even an electromagnetic pulse.
Reply: We would like to thank referee #2 for suggesting introducing laser missions. So, we will revise the text in the revised manuscript to address the referee's comments.
4. As an alternative to the semi-variogram analysis to determine the spatial dependency weights, the authors could have used "backwater lengths in rivers" studied by Samuel (1989). It can give an idea of which river reaches are affected by WSE variations at the VS locations. It would be a good idea to compare both approaches in future studies (not now).
Samuels, P. G. (1989). Backwater lengths in rivers. Proceedings of the Institution of Civil Engineers, 87(4), 571–582. https://doi.org/10.1680/iicep.1989.3779
Reply: We would like to express our gratitude to referee #2 for the suggestion. We will consider the suggestion in our future studies.
5. Line 156. using a power law dependent on what? Width? Upstream drainage area? Is it the same power law parameters for the whole basin?
Reply: We appreciate referee #2 for raising this question. The power law depends on a prior annual average river discharge and uses a single set of parameters (i.e., a and b) for the whole basin (a=0.1, b=0.5). The power law is shown in Equation 1 in the supplementary document. We will revise the text according to this comment.
6. Line 199. Something went wrong with the font size of some words. Line 351. Line 438. Line 637.
Reply: We would like to thank referee #2 for identifying those mistakes. We will correct all those kinds of font errors.
7. Line 229. So, the mean and standard deviation were calculated based on the open loop simulation?Reply: Yes, the mean and standard deviation were calculated based on the long-term open loop simulation (2000-2014).
8. Line 231 to 239. Some of your readers might be unfamiliar with the Amazon Basin. It would be interesting to write a short and objective section about this basin, presenting a DEM map at least (a mean Precipitation map would be nice too).Reply: We like to thank referee #2 for the great suggestion. We will be happy to include a description and a figure of the Amazon basin but considering the length of and the number of figures in the manuscript, we will add them to the supplementary material.
9. Line 275. How do you measure the relative sharpness and the difference in reliability? Line 383 should be here.Reply: We believe Line 279-283 describes the same ideas as Line 383. We will modify the text to enhance the understanding of the Interval Skill Score (ISS).
10. Several wrong references in section 3.1.Line 309. The authors said that the Santos Dumont gauge is in the Purus River, but in Figure 4 it says Jurua River.
Line 332. "Figure 5c–e displays hydrographs of the Jurua (Gaviao), Amazon (Manacapuru), and Negro (Serrinha) rivers" but in Figure 5 it says Manicore on the Madeira River, Aruma on the Purus River, and Sao Felipe on the Negro River.
Line 351. "The lower panels of Figure 6 illustrate flow dynamics along the Amazon mainstem (Sao Paulo De Olivenca; Figure 6c) and Japura (Vila Bittencourt; Figure 6d) and Negro (Curicuriari; Figure 6e) rivers." but in Figure 6 it is written "Hydrographs recorded at Humaita on the Madeira River, Santos Dumont on the Jurua River, and Canutama on the Purus River are presented on panels c, d, and e, respectively."
Reply: We would be grateful to the referee for identifying the mistake. We revised Figures 4, 5, and 6 corresponding to the description of sections 3.1.1., 3.1.2., and 3.1.3.
11. Line 321. Saying that the "direct DA generally improved flow dynamics" is very optimistic. Based on these results, I'd probably say that the direct DA maintained or even degraded the general performance, at least for discharge.
Reply: We would be thankful to the referee for the detailed comments on the text. We agree with the referee that the Direct DA method somewhat degraded the performance of discharge estimations in some locations. But the discharge was improved in some gauge locations in the Amazon basin such as Santo Antonio Do Ica and Sao Paulo de Olivenca. Therefore, we will revise the text to deliver the ideas suggested by referee #2.
12. Line 340. Once again, I think it is an optimistic conclusion. In the last sentence, the authors just said: "although NSE and ISS values worsened slightly." So how can the authors say afterward that "discharge estimates improved moderately"? I don't think that improvements in the correlation coefficient are enough for such a statement given that the NSE has become worst. But I reckon that seasonality got better as correlation got higher. Maybe the authors should clarify what they try to achieve with DA assimilation.
Reply: We would like to express our gratitude to referee #2. Our goal is to improve the overall performance of the river discharge hence higher NSE the better discharge estimated would be. But we try to look at the positives of each method and come to a broader conclusion as none of the methods performed perfectly in all the scenarios. Ideally, it is better if the direct DA performs better than others as assimilation will not depend on prior statistics of the open loop simulation. We basically discuss the median performance in the discussion of NSE and ISS here. But there is a large variation in the statistics as shown in figure 7, table 2 and figure S2. A reasonable number of gauges improved their performance by anomaly DA. An overall moderate number of gauges improved their discharge estimations. Therefore, we will modify the text to highlight the percentage of improved gauges.
13. Figure S2 should be in the main manuscript. It could be together with Figure 7 as 7c, 7d, and 7e. Figures 7a and 7b don't need to be so large.
Reply: We would like to appreciate the suggestion by referee #2. We will include all or some of the panels of Figure S2 in Figure 7.
14. Line 427. "However, the direct DA experiments efficiently improved sharpness, thereby increasing confidence in the assimilated river discharge." I would say "FALSELY increasing confidence" as the authors just observed that the reliability drops more than 50% for direct DA experiments. What is the point of being narrower if the observation falls out of the confidence interval? I think the authors should be careful with that.
Reply: We would like to thank referee #2 for pointing out this. We agree with the referee that “falsely” increasing the confidence will not be beneficial. In this sentence, we mean that when data assimilation is performed in direct values, the spread of final assimilated values will be narrower than in other methods because the assimilation was performed in direct DA. But not in anomaly or normalized value DA. Of course, the reliability should be higher. We will revise the sentence to convey the idea more clearly.
15. On tables 3 and 4 it would be nice to see the Open Loop and the CaMa VIC BC performances for comparison.Reply: Thank you very much for the nice suggestion. We will include the median statistics of the Open Loop and the CaMa VIC BC in tables 3 and 4 even though CaMa VIC BC cannot be assessed for sharpness.
16. Line 541. HTESSEL not HTEESSEL.Reply: Thank you for recognizing the mistake. We modified all the instances which have HTESSEL in the revised manuscript.
17. Line 624. Which experiment is that? The one in section 4.2.? Or the one with VIC BC (section 3.3)?
Reply: Thank you for recognizing the mistake. It should be VIC BC. We will correct the text in the revised manuscript.
Citation: https://doi.org/10.5194/egusphere-2022-412-AC2
-
AC2: 'Reply on RC2', Menaka Revel, 18 Aug 2022
-
RC4: 'Comment on egusphere-2022-412', Anonymous Referee #3, 31 Aug 2022
This manuscript explores different strategies to assimilate water surface elevation (WSE) derived from satellite altimetry into the CaMa-Flood global hydrodynamic model. During the recent years, a large number of studies have demonstrated the potential benefits of assimilating WSE for various purposes, such as parameter estimation or river dynamics modeling improvement. One of the main difficulties relies on the fact that WSE from altimetry provides the water elevation based on a reference (geoid) that can differ from the model reference (because of DEM errors for example) and the induced bias may highly degrade the assimilation performances. An alternative consists in considering WSE anomalies instead of absolute values. Another one, introduced in this study, also normalizes WSE anomalies to account for possible errors in the signal amplitude. Here the authors explore those three strategies and their respective performances in reproducing river discharge over the Amazon Basin. This scientific question is highly relevant and the study is well conducted, which makes it worth publishing, especially in the context of the forthcoming SWOT mission.
Overall, the manuscript is well written and well organized. The methodology is clearly stated and the results are quite convincing, although the latter should be revisited to correct some mistakes and clarify some points, as explained thereafter.
Major remarks
1. The quality of figures 4 to 6 is quite bad and colors are hard to identify (in the maps and in the time series). Also the time series subplots do not correspond to the description in the results section (3.1.1, 3.1.2 and 3.1.3). Hence it is not possible understand, confirm or refute the description of these plots, as well as the conclusions drawn (L309-323, L332-342, L351-362).
2. It is not clear to me which variables are included in the state vector x. I understand from L509 that the prognostic variable of CaMa-Flood is water storage. On the other hand, it is stated (L188) that the state vector includes river discharge, WSE, flooded area, flood height and storage. Could you clarify this point? Also, in the latter case, shouldn’t the observation operator H contain only zeros except for the column corresponding to WSE? Moreover, are all state variables (river discharge, WSE, etc.) converted to anomaly and normalized values as written in L193?
3. More importantly, I think that the analysis of DA performances when runoff forcing or bathymetry are biased (section 4.2) requires a bit more explanations.
- Model simulations are affected by biases (errors in absolute values) and by errors in dynamics. Decreasing the river bathymetry (by lowering the river bottom elevation) would lower the absolute WSE without impacting the flow dynamics, except if bank overflow occurs (flooding). If there is no flooding, I would have expected large impacts on the direct DA performances but no impact on anomaly and normalized value DA. Given that, do the results of Fig. 11b mean that the degradation of those two experiments is due to bank overflow? In addition, what can explain the very poor performances of normalized value DA method? Maybe some example time series could help better understand these results, as done for the previous experiments (perhaps as a supplement).
- Is there any possible explanation of the better normalized value DA performances with runoff bias and bathymetry error compared to performances with only bathymetry errors (panels b and d)?
- Finally, considering only one runoff (HTESSEL) to generate the ensemble reduces the dynamics variations between the members. Could this be a reason of the poor DA performances, especially with the normalized value DA method?
Each of the three DA methods can outperform the other two depending on the configuration, making the choice of the DA method quite difficult for further studies. I think providing more insight in these experiments might help readers better understand the pros and cons of each method.
Minor remarks
Fig. 1. In panel b, upper left square, it should be “Altimetry Auxillary Data”.
L134-137. “VSs with considerable variation in mean WSE compared to the MERIT Hydro (Yamazaki et al., 2017, 2019) elevation (expressed as riverbank height) were filtered through comparison of mean observations and riverbank heights.” What could be the cause(s) of such errors? Maybe the answer is given in L506-508.
L157. Is the river width from remote sensing available for every reaches of the river network? If not, how is it determined?
Fig. 2. I would suggest to add a legend in the time series, and maybe add error bars in observed WSE (from HydroWeb).
Eq. (2). Since H is linear, maybe it is better to write Hxk instead of H(xk).
L288. Sharpness and reliability are not defined.
L285. Nash and Sutcliffe (1970) and Kling and Gupta (2009) are cited several times, which is not necessary.
L313. It is written (but I cannot verify it) that the 95 % ensemble spread is improved until mid-2010, when the ENVISAT satellite was available. But this satellite is supposed to be available until 2012 (Tab. 1). Also, could you explain what an improvement in ensemble spread is? Is it a reduction of the spread?
L321. Considering numbers in L300-302, I would not say that “direct DA generally improved flow dynamic simulation to a moderate extent”: concerning river discharge, 8 % of gauges show an improvement while 43 % show a degradation.
L326. What could be the impact of choosing different time periods for observed and simulated WSE when computing the long term mean (and std)? For example if a multi-year drought is accounted for in one period and not in the other.
L336. It should be “improvements in r and ISS” not “in Dr and rISS”.
L348. Considering the quality of the figure and the color range, decreases in discharge correlation is not that evident in the Amazon mainstem.
L365. It is stated that the assimilation has very little influence outside the area of satellite observations. First, does the satellite coverage area correspond to the reaches downstream any VS? Second, shouldn’t the localization method used here allow to correct river discharge upstream VS?
Fig. 7. In the caption of a, it should be “probability distribution” instead of “cumulative distribution”. Also, it could be helpful to plot the vertical line at 0. Same remark for Fig. S5.
L385. “A large reduction in sharpness was observed in the direct assimilation experiment (Exp 1), mainly because the assimilation was conducted directly.” I do not see the link here, could you expand a bit more?
L408. For river discharge, sharpness is also considered (L415).
L530. Indeed, a huge potential from SWOT is expected in this kind of study. But how to deal with the need to compute long term mean and std for the derivation of anomalies and normalized values?
L536. Water height in the river is approximately 50 % lower, not WSE.
Fig. 12. The range of the y-axis could be reduced.
Minor remarks in supplementary material
Fig. S1. Square and circle are inverted in the legend.
L45. It should be “Exp 2a and Exp3a”.
L57. What is Exp 3b?
Fig. S7. It is hard to see the effect of DA on low flows. Maybe consider a log-log scatter plot?
Citation: https://doi.org/10.5194/egusphere-2022-412-RC4 -
AC3: 'Reply on RC3', Menaka Revel, 13 Sep 2022
We would like to express our gratitude to referee #1 for his continued interest in our study. We look forward to submitting the revised manuscript after the open discussion phase.
Citation: https://doi.org/10.5194/egusphere-2022-412-AC3
-
AC3: 'Reply on RC3', Menaka Revel, 13 Sep 2022
-
AC4: 'Comment on egusphere-2022-412', Menaka Revel, 13 Sep 2022
Referee #3
This manuscript explores different strategies to assimilate water surface elevation (WSE) derived from satellite altimetry into the CaMa-Flood global hydrodynamic model. During the recent years, a large number of studies have demonstrated the potential benefits of assimilating WSE for various purposes, such as parameter estimation or river dynamics modeling improvement. One of the main difficulties relies on the fact that WSE from altimetry provides the water elevation based on a reference (geoid) that can differ from the model reference (because of DEM errors for example) and the induced bias may highly degrade the assimilation performances. An alternative consists in considering WSE anomalies instead of absolute values. Another one, introduced in this study, also normalizes WSE anomalies to account for possible errors in the signal amplitude. Here the authors explore those three strategies and their respective performances in reproducing river discharge over the Amazon Basin. This scientific question is highly relevant and the study is well conducted, which makes it worth publishing, especially in the context of the forthcoming SWOT mission.
Overall, the manuscript is well written and well organized. The methodology is clearly stated and the results are quite convincing, although the latter should be revisited to correct some mistakes and clarify some points, as explained thereafter.
Reply:
We would like to convey our acknowledgment to referee #3 for his informative comments. We will address all the comments in the revised text, and responses are provided below.
Major remarks
1. The quality of figures 4 to 6 is quite bad and colors are hard to identify (in the maps and in the time series). Also the time series subplots do not correspond to the description in the results section (3.1.1, 3.1.2 and 3.1.3). Hence it is not possible understand, confirm or refute the description of these plots, as well as the conclusions drawn (L309-323, L332-342, L351-362).
Reply:
We would be grateful to referee #3 for raising this issue. When creating these figures, a mistake occurred. Therefore, we will revise figures 4-6 to correspond to the descriptions in sections 3.1.1, 3.1.2, and 3.1.3.
2. It is not clear to me which variables are included in the state vector x. I understand from L509 that the prognostic variable of CaMa-Flood is water storage. On the other hand, it is stated (L188) that the state vector includes river discharge, WSE, flooded area, flood height and storage. Could you clarify this point? Also, in the latter case, shouldn’t the observation operator H contain only zeros except for the column corresponding to WSE? Moreover, are all state variables (river discharge, WSE, etc.) converted to anomaly and normalized values as written in L193?
Reply:
We would like to express our gratitude to referee #3 for the insightful comment. We found that the using same H (observational operator) in both equations 2 and 3 is confusing. Equation 1 is about the CaMa-Flood model time evaluation and equation 2 is the conceptual relationship of CaMa-Flood state variables with the observations. So, the x (simple x) vector consists of CaMa-Flood state variables. Equation 3 is the analysis equation for data assimilation, there we used WSE only for data assimilation. Therefore, we think H in equation 3 is a subset of H in equation 2. Hence, we will revise the H in equation 2 to be (curly H) and will revise the text to explain that we used only WSE in the LETKF analysis equation.
3. More importantly, I think that the analysis of DA performances when runoff forcing or bathymetry are biased (section 4.2) requires a bit more explanations.
- Model simulations are affected by biases (errors in absolute values) and by errors in dynamics. Decreasing the river bathymetry (by lowering the river bottom elevation) would lower the absolute WSE without impacting the flow dynamics, except if bank overflow occurs (flooding). If there is no flooding, I would have expected large impacts on the direct DA performances but no impact on anomaly and normalized value DA. Given that, do the results of Fig. 11b mean that the degradation of those two experiments is due to bank overflow? In addition, what can explain the very poor performances of normalized value DA method? Maybe some example time series could help better understand these results, as done for the previous experiments (perhaps as a supplement).
Reply:
We would like to thank referee #3 for the valuable comment and suggestion. In these experiments, we try to assess the performance of DA methods in simplified error conditions in forcing and parameters of the model. We agree with referee #3 that including corrupted bathymetry will lower the absolute WSE. Hence, it affects WSE assimilation in direct DA.
But in the cases of anomaly and normalized value DA, the degradation of discharge accuracy is mainly due to the errors in the statistics (i.e., mean and standard deviation). These statistics were computed using long-term open-loop simulations. When the river bathymetry is corrupted the mean open-loop WSE will also be biased, and the standard deviation will be different. Hence the assimilation result will also be inaccurate.
- Is there any possible explanation of the better normalized value DA performances with runoff bias and bathymetry error compared to performances with only bathymetry errors (panels b and d)?
Reply:
We would like to express our gratitude to referee #3 for raising this question. With current experiments, the statistics of the normalized value DA method should be much lower because of the lowering of bathymetry as well as reducing the runoff. Hence, we need to investigate the experimental setting again whether there is an error in preparing Figure 11. We will re-check Figure 11 and add some explanation about such a trend in these experiments.
- Finally, considering only one runoff (HTESSEL) to generate the ensemble reduces the dynamics variations between the members. Could this be a reason of the poor DA performances, especially with the normalized value DA method?
Reply:
We would like to express our sincere thanks to referee #3. We don’t think the poor performance of the normalized value DA method in corrupted bathymetry or runoff error cases is due to low variation in perturbations. It is mostly due to the bias and differences in statics.
4. Each of the three DA methods can outperform the other two depending on the configuration, making the choice of the DA method quite difficult for further studies. I think providing more insight in these experiments might help readers better understand the pros and cons of each method.
Reply:
We would like to appreciate referee #3 for the great comment. We will add some more recommendations and improve the current description to enhance these ideas.
Minor remarks
1. In panel b, upper left square, it should be “Altimetry Auxillary Data”.
Reply: Thank you for your recognition of the error. We will correct the text accordingly.
2. L134-137. “VSs with considerable variation in mean WSE compared to the MERIT Hydro (Yamazaki et al., 2017, 2019) elevation (expressed as riverbank height) were filtered through comparison of mean observations and riverbank heights.” What could be the cause(s) of such errors? Maybe the answer is given in L506-508.
Reply: We would like to thank referee #3 for raising this question. From our analysis, we found that most erroneous VSs are in narrower rivers at high elevations. There can be several reasons for these, 1. Non-nadir direction observations, 2. Errors in post-processing of VS (e.g., geoid conversion), etc.
3. Is the river width from remote sensing available for every reaches of the river network? If not, how is it determined?
Reply: We thank referee #3 for the question. Remote sensing river widths were used in the river reaches with river width > 300m (Yamazaki et al., 2014). A power law relationship of the average river discharge was used to estimate the river width for the other smaller river reaches. We will revise the text to reflect these ideas.
4. Fig. 2. I would suggest to add a legend in the time series, and maybe add error bars in observed WSE (from HydroWeb).
Reply: Thank you very much for the suggestion. We will revise Fig 2 according to referee #3’s suggestion.
5. Eg. (2). Since H is linear, maybe it is better to write Hxk instead of H(xk).
Reply: Thank you for the suggestion. Here H is more of an operator for converting simulated variables to observable variables. Not all the simulated variables are possible to observe. So, we would like to keep the equation as it is, but we will use two symbols for equations 2 and 3 because H in equation 3 is a subset of H in equation 2.
6. Sharpness and reliability are not defined.
Reply: Thank you for the suggestion. We will define sharpness and reliability in the methodology section.
7. Nash and Sutcliffe (1970) and Kling and Gupta (2009) are cited several times, which is not necessary.
Reply: Thank you. We will remove those citations.
8. It is written (but I cannot verify it) that the 95 % ensemble spread is improved until mid-2010, when the ENVISAT satellite was available. But this satellite is supposed to be available until 2012 (Tab. 1). Also, could you explain what an improvement in ensemble spread is? Is it a reduction of the spread?
Reply: We appreciate referee #3 for raising this issue. ENVISAT nominal period is 2002-2010 after 2010 ENVISAT track was changed. Therefore, ENVISAT data after 2010 was not used in HydroWeb. Hence, we will revise Table 1, we will add only the nominal periods used in HydroWeb for each satellite.
An improvement in ensemble spread is referred to as the reduction of spread or reduction of sharpness. We think “improvement in ensemble spread” is better be revised as an improvement in sharpness. Hence, we will revise the manuscript according to referee #3’s comments
9. Considering numbers in L300-302, I would not say that “direct DA generally improved flow dynamic simulation to a moderate extent”: concerning river discharge, 8 % of gauges show an improvement while 43 % show a degradation.
Reply: We would like to express our gratitude to referee #3 for pointing out this. It is evident that many gauges degraded their accuracy of river discharge by the direct DA method. However, some gauges such as Santo Antonio Do Ica and Sao Paulo de Olivenca improved their discharge accuracy. Hence, we will revise the text according to the comment of referee #3 to reflect that the river discharge estimated of some gauges show improvement.
10. What could be the impact of choosing different time periods for observed and simulated WSE when computing the long term mean (and std)? For example if a multi-year drought is accounted for in one period and not in the other.
Reply: We would like to thank referee #3 for the question. There can be not much effect of using different time periods for statistics for simulation and observations if the statistics are representative statistics of the long-term values. But if the statics are largely different from the observed statistics as shown in the biased runoff experiment case, the accuracy of estimated discharge by the anomaly and normalized value DA can be hampered. We checked sensitivity on the simulation WSE statistics where we found that the statistics calculated using simulation of 5 years or more are reasonably reproduced river discharges.
11. It should be “improvements in r and ISS” not “in Dr and rISS”.
Reply: Thanking referee #3, we will revise the text.
12. Considering the quality of the figure and the color range, decreases in discharge correlation is not that evident in the Amazon mainstem.
Reply: We would like to thank referee #3 for the comment. We will improve the quality of the figures in the revised manuscript. We agree with referee #3 that the correlation in the Amazon mainstream is quite good. This may be because model parameters (e.g., bank full height, river bathymetry, etc.) are still very good on these river reaches.
13. It is stated that the assimilation has very little influence outside the area of satellite observations. First, does the satellite coverage area correspond to the reaches downstream any VS? Second, shouldn’t the localization method used here allow to correct river discharge upstream VS?
Reply: We would like to express our gratitude to referee #3 for the great question. Here, we define the satellite coverage area as the river reaches located downstream of the most upstream VS in each tributary (green circles in Figure S1 indicate GRDC locations that were in the satellite coverage area). Hence, the satellite coverage areas cover downstream of all the VSs. Although using the localization method, the WSE can be updated without using any local observation. But in the upper reaches (narrow rivers with small catchment areas), the size of the local patches is small (Revel et al., 2019). Hence, assimilation efficiency can be lower in these upper reaches.
14. Fig. 7. In the caption of a, it should be “probability distribution” instead of “cumulative distribution”. Also, it could be helpful to plot the vertical line at 0. Same remark for Fig. S5.
Reply: We would like to thank referee #3 for the suggestion. We will revise the caption of Fig. 7 and Fig S5 according to referee #3’s suggestion.
15. “A large reduction in sharpness was observed in the direct assimilation experiment (Exp 1), mainly because the assimilation was conducted directly.” I do not see the link here, could you expand a bit more?
Reply: Thank you very much for asking for calcification. Sharpness reduction means the reduction of the ensemble spread. When the assimilation was performed in real values the ensemble spread become smaller than it was performed in anomaly or normalized values. We will revise the text to reflect this meaning in the revised manuscript.
16. For river discharge, sharpness is also considered (L415).
Reply: Thank you very much. We indeed calculated sharpness for river discharge as well even though we did not include it in Fig 8. The median sharpness values were shown in Table 3.
17. Indeed, a huge potential from SWOT is expected in this kind of study. But how to deal with the need to compute long term mean and std for the derivation of anomalies and normalized values?
Reply: We would like to thank referee #3. In this study, we did not assess the possibilities of using anomaly or normalized DA methods for the real-time forecast or usage with short-term observation records. However, a representative estimate for SWOT observations can provide reasonable discharge estimates. We will assess the sensitivity of the observational statistics which can be used for the real-time/short-term forecast in our future studies.
18. Water height in the river is approximately 50 % lower, not WSE.
Reply: Thank you very much for pointing out this. We will revise the text accordingly.
19. Fig. 12. The range of the y-axis could be reduced.
Reply: Thank you very much. We will revise Fig 12.
Minor remarks in supplementary material
1. Fig. S1. Square and circle are inverted in the legend.
Reply: Thank you very much. We will correct it.
2. It should be “Exp 2a and Exp3a”.
Reply: Thank you for pointing this out. We will revise it.
3. What is Exp 3b?
Reply: We would like to thank referee #3. Exp 3b should be removed. We will revise it.
4. Fig. S7. It is hard to see the effect of DA on low flows. Maybe consider a log-log scatter plot?
Reply: Thank you very much for the comment. We will revise Fig. S7 accordingly.
Reference:
- Revel, Ikeshima, Yamazaki and Kanae: A Physically Based Empirical Localization Method for Assimilating Synthetic SWOT Observations of a Continental-Scale River: A Case Study in the Congo Basin, Water, 11(4), 829, doi:10.3390/w11040829, 2019.
- Yamazaki, D., O’Loughlin, F., Trigg, M. A., Miller, Z. F., Pavelsky, T. M. and Bates, P. D.: Development of the Global Width Database for Large Rivers, Water Resour. Res., 50(4), 3467–3480, doi:10.1002/2013WR014664, 2014.
Citation: https://doi.org/10.5194/egusphere-2022-412-AC4
Peer review completion
Post-review adjustments
Journal article(s) based on this preprint
Menaka Revel et al.
Data sets
HydroDA v1.0 Revel, M., X. Zhou, S. Kanae, D. Yamazaki https://doi.org/10.4211/hs.08e1b18aa9f240758dd13d9ac875621f
Model code and software
HydroDA 1.0 Revel, M. https://github.com/MenakaRevel/HydroDA/releases
Menaka Revel et al.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 484 | 194 | 24 | 702 | 50 | 5 | 5 |
- HTML: 484
- PDF: 194
- XML: 24
- Total: 702
- Supplement: 50
- BibTeX: 5
- EndNote: 5
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(6646 KB) - Metadata XML
-
Supplement
(1687 KB) - BibTeX
- EndNote
- Final revised paper