Improving large-scale river routing models for climate studies: the impact of ESA long-term CCI discharge products on correcting multi-model hydrological simulations

Sadki, Malak; Noual, Gaëtan; Munier, Simon; Pedinotti, Vanessa; Verma, Kaushlendra; Albergel, Clément; Biancamaria, Sylvain; Andral, Alice

doi:https://doi.org/10.5194/hess-2024-328

Preprints

https://doi.org/10.5194/hess-2024-328

Preprints

18 Nov 2024

| 18 Nov 2024

Status: this preprint is currently under review for the journal HESS.

Improving large-scale river routing models for climate studies: the impact of ESA long-term CCI discharge products on correcting multi-model hydrological simulations

Malak Sadki, Gaëtan Noual, Simon Munier, Vanessa Pedinotti, Kaushlendra Verma, Clément Albergel, Sylvain Biancamaria, and Alice Andral

Abstract. Large scale hydrological models like CTRIP and MGB are essential for simulating river dynamics and supporting large-scale climate studies. Their accuracy can be significantly improved through satellite data assimilation. This study leverages 20 years of high-resolution discharge data (2000–2020) from the ESA Climate Change Initiative (CCI) to enhance CTRIP and MGB models via ensemble Kalman Filter frameworks (HyDAS and HYFAA). Applied to the Niger and Congo basins, the models assimilate discharge data derived from altimetry and multispectral imagery, alongside water surface elevation (WSE) anomaly data, to evaluate their impact on model performance.

Discharge assimilation was more effective than WSE anomaly assimilation, as it provided a more direct input for improving model accuracy. Temporal data density was the key factor in reducing bias and enhancing the simulation of seasonal flow patterns, with spatial coverage and data quality also playing important roles. In the Niger Basin, the assimilation of denser discharge data resulted in a significant bias reduction, which should improve the representation of long-term climate trends. Furthermore, the higher temporal resolution allowed for better capture of flow variability, which is crucial for both seasonal climate studies and short-term predictions, such as extreme hydrological events.

The study also emphasizes the trade-offs between data resolution and quality, particularly in the Congo Basin. Future advancements include merging altimetry and multispectral discharge data, improving the discharge retrieval algorithms using SWOT data, and refining data assimilation techniques to improve climate studies and river system modeling in complex, climate-impacted basins.

Received: 15 Oct 2024 – Discussion started: 18 Nov 2024

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Malak Sadki, Gaëtan Noual, Simon Munier, Vanessa Pedinotti, Kaushlendra Verma, Clément Albergel, Sylvain Biancamaria, and Alice Andral

Status: final response (author comments only)

RC1:
'Comment on hess-2024-328', Menaka Revel, 22 Dec 2024
Improving large-scale river routing models for climate studies: the impact of ESA long-term CCI discharge products on correcting multi-model hydrological simulations
by Sadki Malak, Noual Gaëtan, Munier Simon, Pedinotti Vanessa, Verma Kaushlendra, Albergel Clément, Biancamaria Sylvain, and Andral Alice
The authors compared two data assimilation frameworks namely CTRIP-HyDAS and MGB-HYFAA. They performed a few experiments using remotely sensed discharge and water surface elevations from multiple nadir altimetry satellite missions to assimilate into the models in Congo and Niger basins. They assimilated altimetry WSE anomalies (dH), altimetry-based discharge (Qalti), and multispectral image-based discharge (Qmultispec) into two models namely CTRIP and MGB. They further investigate the different observation errors for data assimilation. The authors concluded that assimilating discharge data can effectively improve the discharge simulation of the model. Furthermore, they discussed that the observation error, observation frequency, and spatial extent can influence the accuracy of the assimilated values.
In the era of satellite data, comparing data assimilation methods across large river basins is of immense value. With the launch of satellites such as Surface Water and Ocean Topography (SWOT), the application of advanced data assimilation techniques becomes even more critical.
However, the presentation quality of the manuscript can be further enhanced by improving figure quality, reducing the number of figures, summarizing certain paragraphs, or moving them to the supplementary materials or appendix.
Some comments have been provided to enhance the manuscript.
Major Comments:
I strongly recommend revising the title of the manuscript. It needs to reflect the content of the manuscript. For example, "correcting" is too broad as the manuscript is about assimilating data into a hydrological/hydrodynamic model. In addition, even though the authors mention "climate studies" in the title, the authors seldom discuss how these methods be used in climate studies.

The description of the data assimilation (“2.3.2 The Ensemble Kalman Filter (EnKF)”) is more focused on the data assimilation method used in CTRIP-HyDAS despite the manuscript addressing both CTRIP-HyDAS and MGB-HYFAA. The authors should include an explanation of the methods used in MGB-HYFAA. Is MGB-HYFAA based on Wongchuig-Correa et al., (2020); Wongchuig et al., (2023, 2024). It appears that MGB-HYFAA uses the Local Ensemble Kalman Filter (LEnKF), whereas CTRIP-HyDAS employs the Local Ensemble Transformation Kalman Filter (LETKF). Additionally, it is unclear whether the same localization approach is used in data assimilation frameworks. The authors should clarify whether CTRIP-HyDAS and MGB-HYFAA apply identical observation localization methods. Furthermore, the differences between CTRIP-HyDAS and MGB-HYFAA in the context of data assimilation should be discussed. Summarizing the key features of HyDAS and HYFAA—such as routing schemes, data assimilation methods, and localization approaches—would greatly improve clarity.

The manuscript is overly lengthy and contains 26 figures, which is excessive for a research article. To improve readability and focus, the authors are encouraged to condense the content and move some sections to supplementary materials or an appendix. For instance, the evaluation of satellite-based discharges and the open-loop performance of the models could be relocated to supplementary materials. This would streamline the main text, ensuring that the most critical findings and discussions remain central, while detailed analyses and additional data are still accessible for interested readers.

Based on the explanation in Section 2.4.1, Table 2, and Figure 8, it appears that the authors have used only satellite observations corresponding to GRDC locations. Why didn’t the authors utilize all the available satellite altimetry data across the entire basin? The strength of satellite data lies in its spatial distribution, even if it is limited in temporal resolution. While Qalti and Qmultisepc generate discharge data only at locations where in-situ observations are available, WSE has the potential to provide coverage across the entire basin. To fully exploit the potential of satellite altimetry, the authors could incorporate all available satellite data across the basin.

Furthermore, the authors could explore and discuss the added value of spatially distributed observations for assimilation. Relying solely on satellite data at GRDC gauges may fail to leverage the advantages of data assimilation localization (as noted in Line 637), particularly since no additional observations would influence the assimilation in nearby river reaches.

The hashed effect observed in the assimilated products seems to result from the sparseness of observations and the use of observation localization. From the description in Lines 504–512 and Figure 17, it is unclear whether all observations within the 'local patch' (as shown in Figure 7) are utilized. It would be valuable to investigate how many observations can be accommodated within a single local patch. If the local patch is large enough to include a substantial number of observations, the hashed effect could be minimized. Therefore, the authors are encouraged to analyze the effect of local patch size and the number of observations within each patch to determine their impact on the results.

The authors have generally stated that discharge is the preferred variable to assimilate, citing mismatches in the rating curves as a reason. However, this explanation lacks depth. The authors should delve deeper into the specific circumstances under which discharge is a more suitable variable for assimilation. They should provide a more detailed analysis, outlining the advantages and limitations of assimilating discharge versus other variables such as water surface elevation (WSE), particularly in the context of large-scale river models. Moreover, offering guidance on how to select the most appropriate variables for data assimilation would significantly enhance the manuscript. This guidance could include considerations such as data availability, model structure, observation accuracy, and the specific objectives of the assimilation. Such an exploration would not only strengthen the current study but also serve as a valuable resource for researchers working with large-scale river models.

The manuscript does not address the impact of the forcing data on the final assimilation results. CTRIP uses ERA5 forcing data, while MGB employs GSMAP/IMERG and ERA5. Given that precipitation may differ between ERA5 and GSMAP/IMERG, it is important to assess how this discrepancy influences the assimilation performance. If this analysis is beyond the scope of the manuscript, the authors should consider using the same forcing data for both models to ensure a fair comparison.

The explanation provided in lines 513-525 and Figure 19 is somewhat vague and incomplete. The authors attempt to explain that the accuracy of the assimilation depends entirely on the uncertainty of the observations. However, the authors should first clarify the concepts of increment/innovation and residuals. Additionally, in the case of dH assimilation, a 0.4m observation error is compared to the dH uncertainty of the models, whereas in discharge assimilation, 30% of Qalti is compared with the modeled discharge. For discharge assimilation, model uncertainty is likely to be larger than dH uncertainty. Therefore, it is crucial to examine how a 0.4m variation in water level could affect the variation in discharge. Furthermore, the authors note that outliers at the peak were disregarded in the EnKF due to high observation errors, but they should also discuss how the EnKF would perform if an outlier occurred during the low-flow season. Additionally, the filter divergence and inflation applied may influence such behavior in the EnKF. Therefore, it is recommended that the authors further investigate this issue.

Overall, the authors should enhance the presentation quality of the manuscript. The results section could be better structured by grouping similar ideas together, such as discussing discrepancies in the rating curves in a cohesive manner. Additionally, all figures should be improved, including more detailed captions that are self-explanatory. It would also be beneficial to include statistical measures such as NSE and KGE in the time series of discharge and water level. Furthermore, the authors should improve the citations and references; for instance, references like “?Biancamaria et al.,” and “Filippucci et al. (in preparation)” should be clarified and properly formatted.

Moderate Comments:
L4: HyDAS and HYFAA were not introduced. It is hard to think that the reader outside the field of river DA would know these acronyms.

L24: The authors would ideally need to start the discussion about the river DA in a broader sense such as by referring to relevant studies (e.g., Clark et al., 2008; Feng et al., 2021; Michailovsky et al., 2013; Paiva et al., 2013; Revel et al., 2023; Wongchuig et al., 2019). Then authors can narrow down the explanation to CTRIP and MGB DA.

L88: The rationale for selecting these two basins could benefit from further explanation, such as considerations of data availability, a lack of previous studies in the region, or other relevant factors. The motivation behind this choice should be emphasized more clearly.

L216-222: In the paragraph, the authors presented the calculation of the “length of the localization” but how about the observation localization weight, w?

Figures 9 and 10: Figures 9 and 10 can be combined. As Figure 9 shows the KGE and its components, those can be combined into Figure 10 by showing the numeric value of each metric.

Section 4.1.3 would be more appropriate in the discussion section. Therefore, it is recommended to move it under “4.3 Discussion.”

L483: The authors need to examine more about the mismatch between the model’s rating curve and satellite-based rating curves such as parameter error, model physics, etc.

L488-495: The authors should further investigate the spatial density, temporal frequency, and accuracy of the observations. This is crucial for future studies, real-time applications, etc. The authors could also provide recommendations on data accuracy, spatial density, and other factors to improve the assimilation of discharge or water surface elevation (WSE).

Figure 18: It would be better to compare the in-situ rating curves here, as the MGB model indicates that the locations shown in the figure exhibit hysteresis in the rating curve. The Qalti/dH does not seem to capture this hysteresis. Additionally, the authors should discuss the reasons for the differences in the rating curves between Obs_CCI and MGB.

L542-543: The authors can use the CCI discharge comparison with the in-situ observations (Section 2.4.2) to further reinforce the hypothesis of “The uncertainty for WSE ...”. But that hypothesis contradicts the explanation given in the L551-554. So, the authors need to clearly explain the conditions and circumstances when discharge assimilation works well compared to dH assimilation. It will be helpful to investigate more about model parameters as well as the model physics when discussing the “preferred assimilation variable”.

The high redundancy of the phrase "higher temporal resolution of Qmultispec" is evident throughout Section 4.3.2. The authors should summarize the content more concisely and present the key points clearly.

Minor Comments:
L20: Please provide the references to CTRIP and MGB

L43-52: Why authors did not introduce Wongchuig et al., (2024)?

Figure 1: It was felt that Figure 1, Figure 2, and Figure 8 can be summarized into one figure with two panels.

Figure 3 & 4: The authors need to include some more description on the captions of those figure.

Figure 4: In the caption of the figure “((Fleischmann et al., 2018))” should be “(Fleischmann et al., 2018)”

L173: The authors could combine Sections 2.2 and 2.3 considering CTRIP-HyDAS and MGB-HYFAA.

L219: the title of Section 2.3.2 can be revised as a broader title such as “Data assimilation method” rather than EnKF because the actual paragraph is presenting about LETKF.

L216: “The localization” should be “The observations localization weight”. In DA literature “localization” has a broader meaning than “w”.

L217: “length of the localization” is commonly referred to as the local patch in DA studies which is the area used for collecting observations for assimilation for a given river pixel.

L220: What is the threshold used for defining the “length of the localization”?

Figure 7: What does the color bar mean? observations localization weight (w)?

L232: What is the meteorological forcing perturbated in making the ensemble?

Figure 9: The authors better reorganize this figure more carefully by adding figure panel numbers and titles like “Niger” and “Congo”.

L253: Typo “?Biancamaria et al.,”?

L273: Check whether this method citation is correct “Filippucci et al. (in preparation)”.16.

L275: Have any outlier removal methods been applied to the assimilated satellite data? Based on Figure 10, applying an outlier removal method could improve the accuracy of data assimilation.

Figure 11: Some abbreviations, such as "Ppt" and "Pref," are not clear or defined in the caption. These should be clarified or spelled out for better understanding.

L335: The authors should introduce how anomalies were calculated in “water surface elevation anomaly (dH)”.

Figure 12: It is better to show the KGE and NSE values in each figure.

Figure 13: Why does this figure not include CTRIP_OL

Figures 15 and 16: These figures need further improvements, such as making the vertical axes comparable between the (1) and (2) columns and increasing the size of the horizontal axis labels for better readability.

L501-503: The authors should discuss relative improvement/degradation from the open-loop.

L533: Please revise the sentence structure to emphasize the idea.

Figure 17: Please confirm whether “Obs_CCI” refers to both WSE from satellite altimetry and altimetry-based discharge.

Figure 22: The authors need to explain more about the figure in the caption.

L557: What are the conditions to satisfy this such as “when the discharge estimations were accurate enough”.

Figure 21: Typo “CCIP”

References: The references need to be corrected for example Biancamaria, S., et al. (2024). Water Surface Elevation Time Series Derived from Multiple Nadir Altimetry Missions: A Long-Term Dataset for Hydrological Applications. Hydrology and Earth System Sciences, In press.; Cassé, C., et al. (2015). Modeling of the Congo River basin hydrology: The CRBH model – Version 1.0. *Geoscientific Model Development, 8*(8), 2315–2333. https://doi.org/10.5194/gmd-8-2315-2015

References:
Clark, M. P., Rupp, D. E., Woods, R. A., Zheng, X., Ibbitt, R. P., Slater, A. G., Schmidt, J. and Uddstrom, M. J.: Hydrological data assimilation with the ensemble Kalman filter: Use of streamflow observations to update states in a distributed hydrological model, Adv. Water Resour., 31(10), 1309–1324, doi:10.1016/j.advwatres.2008.06.005, 2008.
Feng, D., Gleason, C. J., Lin, P., Yang, X., Pan, M. and Ishitsuka, Y.: Recent changes to Arctic river discharge, Nat. Commun., 12(1), 1–9, doi:10.1038/s41467-021-27228-1, 2021.
Michailovsky, C. I., Milzow, C. and Bauer-Gottwein, P.: Assimilation of radar altimetry to a routing model of the Brahmaputra River, Water Resour. Res., 49(8), 4807–4816, doi:10.1002/wrcr.20345, 2013.
Paiva, R. C. D., Collischonn, W., Bonnet, M. P., De Gonçalves, L. G. G., Calmant, S., Getirana, A. and Santos Da Silva, J.: Assimilating in situ and radar altimetry data into a large-scale hydrologic-hydrodynamic model for streamflow forecast in the Amazon, Hydrol. Earth Syst. Sci., 17(7), 2929–2946, doi:10.5194/hess-17-2929-2013, 2013.
Revel, M., Zhou, X., Yamazaki, D. and Kanae, S.: Assimilation of transformed water surface elevation to improve river discharge estimation in a continental-scale river, Hydrol. Earth Syst. Sci., 27(3), 647–671, doi:10.5194/hess-27-647-2023, 2023.
Wongchuig-Correa, S., Cauduro Dias de Paiva, R., Biancamaria, S. and Collischonn, W.: Assimilation of future SWOT-based river elevations, surface extent observations and discharge estimations into uncertain global hydrological models, J. Hydrol., 590(March), 125473, doi:10.1016/j.jhydrol.2020.125473, 2020.
Wongchuig, S., Kitambo, B., Papa, F., Paris, A., Fleischmann, A. S., Gal, L., Boucharel, J., Paiva, R., Oliveira, R. J., Tshimanga, R. M. and Calmant, S.: Improved modeling of Congo’s hydrology for floods and droughts analysis and ENSO teleconnections, J. Hydrol. Reg. Stud., 50, doi:10.1016/j.ejrh.2023.101563, 2023.
Wongchuig, S., Paiva, R., Siqueira, V., Papa, F., Fleischmann, A., Biancamaria, S., Paris, A., Parrens, M. and Al Bitar, A.: Multi-Satellite Data Assimilation for Large-Scale Hydrological-Hydrodynamic Prediction: Proof of Concept in the Amazon Basin, Water Resour. Res., 60(8), 1–34, doi:10.1029/2024WR037155, 2024.
Wongchuig, S. C., de Paiva, R. C. D., Siqueira, V. and Collischonn, W.: Hydrological reanalysis across the 20th century: A case study of the Amazon Basin, J. Hydrol., 570(November 2018), 755–773, doi:10.1016/j.jhydrol.2019.01.025, 2019.
Citation: https://doi.org/10.5194/hess-2024-328-RC1
- AC1: 'Reply on RC1', Malak Sadki, 11 Feb 2025
  
  We would like to thank the reviewer for the detailed and insightful comments on the manuscript, please find a response to each comment in the .pdf file.
  
  Citation: https://doi.org/10.5194/hess-2024-328-AC1
RC2:
'Comment on hess-2024-328', Anonymous Referee #2, 26 Dec 2024

It’s my pleasure to review hess-2024-328 “Improving large-scale river routing models for climate studies: the impact of ESA long-term CCI discharge products on correcting multi-model hydrological simulations” by Malak et al. The authors conducted several numerical experiments using both CTRIP-HyDAS and MGB-HYFAA models to evaluate the impact of assimilating different CCI discharge products on improving discharge simulations in both Niger and Congo river basins. Although the research shows potential for improving large-scale river discharge simulations via DA, current presented structure of the paper and the figures make the paper less readable, and the design of DA experiments can be also enhanced. Accordingly, major revision is recommended. My comments are as follows.
1.The current layout and structure of the paper need significant improvement.
(1) The introduction of data and model should be divided into two separate sections.
(2) It’s strange to include subsection 2.4.2 in the introduction of data and model.
(3) It’s strange to include the discussion (i.e. subsection 4.3) in the section of Results.
(4) The layout should be improved, for instance, there are a large number of blank spaces between pages.
2.The quality of presented figures need significant improvement.
(1) There are too many figures presented in the paper, which I think the authors should try to reduce the number of figures to highlight the main results. For instance, I don’t find the necessary to include the Figures 3-6, and Figures 1-2 and 8 can be merged into one figure.
(2) The quality of most figures should be improved, because it’s difficult to read the text, numerical values and legend presented in most figures.
3.It’s not clear why the authors use both CTRIP and MGB models. Given the fact that the MGB model was already calibrated against in-situ discharge time series, I don’t think it’s fair to compare its performance to the CTRIP that was not calibrated yet. In addition, the input of precipitation for the two models are also different.
4.The design of DA experiments can be further enhanced.
(1) The authors can consider implementing only the MGB model for the DA experiment, and I think the authors can consider two cases, the MGB model with and without calibration, to investigate the impact of calibration on assimilating satellite-based product.
(2) The authors can also consider assimilating the in-situ discharge data from the stations where both in-situ and satellite-based data are available, and the remaining stations can be used for the validation. As such, the subsection 2.4.2 can be included in the section of Results, and the impact of uncertainty of the satellite-based data can be further investigated compared to the performance of in-situ observations.
5.The Introduction can be further improved to review current progress of assimilating discharge data (including those research using in-situ observations) and relevant DA methods.

Citation: https://doi.org/10.5194/hess-2024-328-RC2
- AC2: 'Reply on RC2', Malak Sadki, 11 Feb 2025
  
  We would like to thank the reviewer for the helpful and interesting comments on the manuscript, please find a response to each comment in the .pdf.
  
  Citation: https://doi.org/10.5194/hess-2024-328-AC2

Malak Sadki, Gaëtan Noual, Simon Munier, Vanessa Pedinotti, Kaushlendra Verma, Clément Albergel, Sylvain Biancamaria, and Alice Andral

Viewed

Total article views: 918 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
780	113	25	918	23	48

HTML: 780
PDF: 113
XML: 25
Total: 918
BibTeX: 23
EndNote: 48

Views and downloads (calculated since 18 Nov 2024)

Month	HTML	PDF	XML	Total
Nov 2024	93	11	6	110
Dec 2024	67	18	3	88
Jan 2025	18	6	1	25
Feb 2025	29	8	4	41
Mar 2025	18	4	2	24
Apr 2025	19	6	0	25
May 2025	22	3	25
Jun 2025	28	3	4	35
Jul 2025	26	11	0	37
Aug 2025	94	12	1	107
Sep 2025	345	27	1	373
Oct 2025	21	7	0	28

Cumulative views and downloads (calculated since 18 Nov 2024)

Month	HTML	PDF	XML	Total
Nov 2024	93	11	6	110
Dec 2024	67	18	3	88
Jan 2025	18	6	1	25
Feb 2025	29	8	4	41
Mar 2025	18	4	2	24
Apr 2025	19	6	0	25
May 2025	22	3	25
Jun 2025	28	3	4	35
Jul 2025	26	11	0	37
Aug 2025	94	12	1	107
Sep 2025	345	27	1	373
Oct 2025	21	7	0	28

Viewed (geographical distribution)

Total article views: 875 (including HTML, PDF, and XML) Thereof 875 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 16 Oct 2025

Short summary

This study explores how 20 years of remote-sensed discharge data from the ESA CCI improve large-scale hydrological models, CTRIP and MGB, through data assimilation. Using an EnKF framework across the Niger and Congo basins, it shows how assimilating denser temporal discharge data reduces biases and improves flow variability, enhancing accuracy. These findings underscore the role of long-term discharge data in refining models for climate assessments, water management, and forecasting.


Total:	0
HTML:	0
PDF:	0
XML:	0