the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 05 Mar 2026
Integrating reservoirs and lakes in the CoSWAT global hydrological model
Abstract. Global water models are essential tools for assessing water resource challenges in the context of climate change, land use changes, and human activities. The CoSWAT Global Water model is a global application of the Soil and Water Assessment Tool (SWAT +). It is a high-resolution tool designed to simulate water systems using a basin-oriented structure. The CoSWAT model currently lacks a realistic representation of reservoirs and lakes, limiting its ability to adequately represent basins where these water bodies play a significant hydrological role. The scarcity and limited accessibility of global reservoir operation or lake outflow data make it challenging to represent these water bodies with the current tools that SWAT+ supports, particularly at a global scale. In this study, we address this limitation by combining commonly used reservoir and lake modelling schemes from other global water models and the capabilities of SWAT+. Moreover, to model irrigation reservoirs, we implemented an approach that combines global datasets with a topological method to estimate irrigation demand for each reservoir. With these new implementations, the CoSWAT model was restructured for selected regions worldwide, where validation of reservoir or lake storage, inflow, and outflow was performed, and the impact of these implementations on streamflow performance was assessed. The results show that the model captures storage dynamics with reasonable performance, comparable to other state-of-the-art global models, and demonstrate a general improvement (70 % of evaluated stations) in streamflow representation following the integration of these water bodies. The new methodological advancements represent a substantial improvement for the CoSWAT global model, enabling more robust and realistic assessments of inland water systems at the global scale.
- Preprint
(2823 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 30 Apr 2026)
- RC1: 'Comment on egusphere-2026-881', Anonymous Referee #1, 11 Mar 2026 reply
-
RC2: 'Comment on egusphere-2026-881', Anonymous Referee #2, 20 Mar 2026
reply
This manuscript presents improvements applied to the global CoSWAT modelling approach, focused on improving the representation of human actions, which emphasis on reservoirs. The manuscript is clear and well structured, and the methods are appropriate and include a comparison with other global models. The main limitation of the study is related to the no calibration of the model, but it is acknowledged and proposed for further work. Improving the inflow calibration will surely lead to better performance for reservoir variables and streamflow, but considering the scale of work, the current advances presented in this manuscript already suppose a novelty and an improvement.
I consider the manuscript suits the journal and it is already in a very good shape, so I can only provide few suggestions/comments to improve it if the authors agree.
Line 44. Fix the phrase
Line 96 and 103. You have more recent examples and using SWAT+ (e.g., https://link.springer.com/article/10.1007/s11269-024-04071-9).
Line 98. The fact that the model has not been calibrated may also explain that.
Line 230. Current storage is also an important factor for this method.
Line 277. Is there not available data to initialize the reservoirs more realistically? Starting with the reservoirs almost full may overestimate storage during a long period.
Line 280. These thresholds are questionable.
Figure 5. Inflow simulation and its overestimation clearly affects the simulation of outflow and storage. In part, because the method you are using consider the inflow magnitude to determine the release. I wonder if the decision of running the warm-up period with all the reservoirs as natural lakes is in part responsible for this, especially for cascading reservoirs.
Line 438. I would consider changing ‘remains similar’ with ‘was slightly reduced’.
Citation: https://doi.org/10.5194/egusphere-2026-881-RC2
Data sets
Integrating reservoirs and Lakes in the CoSWAT model J. P. Teran https://doi.org/10.5281/zenodo.18733431
Model code and software
Soil and Water Assessment Tool Plus (SWAT+) J. G. Arnold et al. https://doi.org/10.5281/zenodo.18727784
CoSWAT-Framework C. J. Chawanda and J. P. Teran https://doi.org/10.5281/zenodo.18746453
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 167 | 56 | 17 | 240 | 15 | 19 |
- HTML: 167
- PDF: 56
- XML: 17
- Total: 240
- BibTeX: 15
- EndNote: 19
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Main comments:
The authors present an interesting study integrating reservoirs and lakes into the CoSWAT global hydrological model. The motivation for this work is clear, and the proposed model improvements appear promising. However, several aspects require further clarification and discussion before publication. The manuscript would benefit from additional analysis of basins where streamflow or storage deviates from observations, and the presentation of some figures could be improved for clarity. A clearer discussion of the model’s limitations, including potential sources of uncertainty in representing reservoirs and lakes at the global scale, is needed.
Detailed comments:
Abstract
Line 27: Provide more specific information on how reservoirs and lakes affect streamflow and other hydrological variables, including quantitative performance metrics (e.g., stations with improved streamflow, KGE, PBIAS). Shorten the methods section to focus more on results.
Introduction
Line 49: List models like ParFlow that include explicit reservoir and lake representation.
An example is the research of Benjamin, https://doi.org/10.5194/hess-29-245-2025, A scalable and modular reservoir implementation for large-scale integrated hydrologic simulations
Line 44-45:"Drying and wetting trends are intensifying…, replace “as is” with “accompanied by” for clarity.
Line 51: The sentence includes too many pauses and should be rewritten for clarity.
2 Methodology
2.1 Global Datasets: Clarify the target resolution to which all data were resampled.
Line 145-157: Only highlight the nine major basins instead of listing all basins in detail. Consider merging with Section 2.3 for clarity.
Figure 1: Add the north arrow and scale bar.
2.3 Reservoir/Lake integration into model network
Figure 2: According to Section 2.3, Figure 2b represents the overall workflow and should appear first. Rename figures to ensure Figure 2b is the first (overall workflow), and 2a follows. Like Figure 2b become 2a, and 2a become 2b.
Line 168-171: Reword to describe “integration of reservoirs and lakes into the model network” rather than “a new lake/reservoir resolution procedure.”
Line 176: how to burn-in of lake/reservoir elevations into DEM?
Line 186-188: Clarify why certain variables such as pvol, evol, parea, eaera, shp_col1/shp_col2, which represent reservoir information, point to hydrology.res rather than reservoir.con? Is it not more reasonable to use reservoir.con for reservoir information? Since area refers to the water body surface, is it not more consistent to use hydrology.res for water surface area? Also, why are max/min storage and outflow rates missing?
2.4 New lake and reservoir simulation scheme
Line 207: provide more details for decision table (e.g.reference).
Line 208: Clarify which two parameterization methods are used (e.g., Doll et al., 2003 and H06 scheme).
Line 208 & 735: Introducing arrays, loops, variable initialization, represents preparatory steps rather than true modifications of the source code. Only changes to formulas/functions should be considered innovations. The third and fifth item in Table A-1 representing a novel contribution.
Line 253: Combine sentences for better flow.
Figure 3: In subplot (a), the blue area is not labeled, and the brown “Groundwater” layer appears minimal or missing. Please add a north arrow and scale bar.
2.6 Simulation setup, model comparison, and evaluation
Line 274 & 295: Ensure uniform resolution for the nine selected basins and all input datasets.
Figure 4: Add the north arrow and scale bar.
3.2 Reservoir and lake storage, inflow and outflow evaluation
Figure 5: Why do KGE histograms in the first subplot show a minimum value, but others do not?
Figure B-1: Why is the spatial distribution presented as a block-type map? Most of the Mississippi and South American basins show KGE values below -0.4. The authors should explain the cause of this phenomenon/
Line 340: Only 34% of basins achieve positive KGE values. What factors contribute to this low performance?
Line 368: provide more details for these explanations (e.g., figure of a few soil moisture profiles supporting the statement (Figure B-2c).
Line 374-375: Explain why Berryessa Lake shows underestimation from 1999-2004.
Line 382: The inclusion of reservoirs in Lake Oahe–Mississippi River appears to decrease streamflow performance, particularly in 2004–2009 when storage is poorly simulated. What causes this, and why do inflow and outflow metrics show large deviations (inflow: KGE = -2.03, PBIAS = -96.9%; outflow: KGE = -1.84, PBIAS = -109.3%)?
3.3 Comparison with other global models
Figure 7: Clarify why CoSWAT’s distribution is inconsistent across models. Are the violin plots comparing metrics between CoSWAT and other models or between models and observation data? Also, in subplot (b), the model colors differ. Why are dark and light colors not consistently labeled in the legend to indicate metric ranges? clarify color representations in the legend.
Line 395-396: Provide references for the reservoir-related models mentioned.
3.4 Streamflow evaluation
Figure 8: Subplot (a) is redundant with Figure B-1a; Subplot (b): Distinguish categories 1-4 with colors or labels.
Figure 9: Recommend indicating the area of the six sub-basins and adding inset maps showing sub-basin shapes and the locations of reservoirs and lakes in upstream/downstream positions.
Line 451-452: Analyze the reasons for these discrepancies in the Mississippi and Central European regions and propose potential improvements? Figure 8 also seems to reflect this issue.
Line 482-483: Clarify whether the overestimation is caused specifically by the reservoir implementation.
Line 485: Justify the choice of the six representative basins.
Line 495-496: Add figures showing reservoir configurations in the six basins with and without reservoirs for better comparison.
What is the basis for this observed phenomenon (e.g., which figure or data)? provide a deeper analysis of the reason (eg. the modeling mechanisms, the formulation and parameterization choices)
Line 499-500: Clarify the basis for the statement in Figure 9.
4 Discussion
4.1 Integration of lakes, reservoirs, and irrigation
Line 518-521: If the irrigation module is included, validation should be provided, or remove the discussion of irrigation and focus on reservoirs and lakes.
4.2 Model performance of simulated reservoirs and lakes
Line 549-550: Discuss why Nasser Lake and Lake Oahe show overestimation and underestimation in certain years.
Line 565-568: Given that the model performance does not improve in some cases after introducing reservoirs and lakes, can the authors suggest potential improvements? For example, could the number of reservoirs be adjusted based on basin characteristics (hydropower, flood control, irrigation)? In basins not dominated by hydropower, could some reservoirs or lakes be reduced? Additionally, have the authors considered evaluating the impact of reservoirs and lakes on other hydrological variables beyond streamflow (e.g., soil moisture, ET)? Also, following the suggestions provided for Section 3.4, add the discussion about these issues.
4.3 Impacts on streamflow representation
Line 575: Clarify the origin of the 70% and 42% values. Are they based on the sum of categories described in Section 3.4?
4.4 Implications and future work
Line 599-600: Provide examples of dedicated lake models for coupling with CoSWAT.
Line 613: Consider adjusting reservoir numbers based on basin characteristics in future work.
5 Conclusion
The model should not be called a "network-resolution approach" since it integrates reservoir and lake areas rather than changing network resolution.
Line 625-627: Add quantitative indicators (e.g., performance metrics) to the conclusion.
Line 628: The statement about improved low-flow control is not supported by the results, as some representative basins (Section 3.2, Figure 6b) still show significant low-flow underestimation. Discuss potential causes.