the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Feb 2025
CoSWAT-WQ v1.0: a high-resolution community global SWAT+ water quality model
Abstract. Quantifying the global extent of anthropogenic impacts on freshwater quality remains challenging due to limited monitoring data, especially in low and middle-income regions. To address this gap and improve our understanding of surface water quality, we introduce CoSWAT-WQ, a large-scale water quality model developed to simulate river water quality constituents of Total Nitrogen (TN) and Total Phosphorus (TP), across global and regional freshwater systems. CoSWAT-WQ, an adaptation of the Soil and Water Assessment Tool (SWAT), is run at a global scale, providing high-resolution simulations that capture spatial and daily temporal dynamics of riverine nutrient loads. Here, we describe the model's inputs, setup structure, processes, and evaluate its performance by comparing model outputs to in-situ water quality observations and other global nutrient models. CoSWAT-WQ achieves comparable ranges of river nutrient loads in comparison to other global nutrient models. Additionally, a normalized root mean square error (nRMSE) < 1 was achieved with in-situ observations at more than 80 % of the gauging stations for TN and TP concentrations. However, there was a general weak underestimation of observed concentrations and variability as seen with low Kling–Gupta efficiency (KGE) values for selected stations. Despite its limitations, the model enables the simulation of river TN and TP constituents at a global scale while keeping local relevance. CoSWAT-WQ’s modular setup allows coupling with sectoral models addressing lake systems, agricultural runoff, and aquatic biodiversity, thereby broadening its applicability for cross-sectoral assessments. The model outputs offer valuable data that can inform ecological risk assessments, human health evaluations, and policy decisions on global freshwater quality management.
- Preprint
(1639 KB) - Metadata XML
-
Supplement
(2353 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
CC1: 'How well can we predict water quality at the global scale?', Tobias Krueger, 05 Mar 2025
Dear authors,
I would like to respectfully raise a number of concerns regarding our ability – as a hydrological modelling community – to predict water quality at the global scale. Some are generic; some are more specific to this manuscript.
Representing the relevant processes at the global scale and having data to parameterise them
Land management is known to be very important for modulating pollution transfers. But how to even define those sets of rules (decision tables) for scheduling management at the 2x2km grid scale or HRU scale (page 5)? How do decisions made at the farm scale or smallholder scale aggregate? How to represent heterogeneity?
What information exists to estimate tile drainage globally (figure 2)?
How do the uncertainties in global data (which are often generated themselves from models), e.g. on point sources, plant and harvest dates, and fertilizer and manure use rates, affect model predictions (page 6)?
Choosing an appropriate tool for what we want to predict
When it is said on page 13 that the primary objective of global water quality models is not to predict exact daily concentrations, then maybe a model that aims to represent processes at that scale is not appropriate. Monthly aggregated data are not predicted well by the daily model either, see below.
When the aim is instead to identify spatial and temporal pollution hotspots within global river networks (page 13), there may be other data-based methods that are better suited for this purpose.
Model calibration
The model in this study is said to be uncalibrated (page 6). How were parameter values determined instead?
The lack of sufficient observational data, strong spatial biases in available observations and uncertainties in input datasets (page 7) are put forward as obstacles to model calibration. What do these obstacles say about our ability to run a model at all?
It is argued that the model is theoretical applicable in regions with no calibration data without a significant loss in performance (page 7). However, the hydrological modelling literature has repeatedly shown that even physically well-founded models require calibration to make them work in particular places. Among other things, this is because the processes that are represented are “effective” processes at the scale of application, either because physical theory is moved between scales or scale-bound abstractions are used in the first place (the MUSLE is an obvious example in this study).
When it is said that uncalibrated physically-based models are often preferred for global change assessment on page 7, I am unclear whether this is because they are uncalibrated or because they are physically-based, and why exactly either of these features makes them preferrable. Keeping in mind the limits of physically-based models discussed above.
Model intercomparison
If all models in an intercomparison are affected by the aforementioned shortcomings in process representation and data, can a model intercomparison really enhance confidence in large-scale water quality models as argued on page 7?
In this study, the log-scaling of the predictions in figure 5 make the discrepancies between the two models appear smaller than they are on the original scale, which are often an order of magnitude. When recalculating the R2 statistic on the back-transformed data, R2 decreases from 0.71 to about 0.41 for TP, for example. I am unclear why the same data appear twice with two different symbols in the graph.
I suggest that a model intercomparison would benefit from a deeper exploration of the mechanics of the individual models. What might explain the differences or similarities in predictions?
Model comparison with observations
The model does not fit the monthly data well (as seen by the KGE values and figure 8). Data uncertainties are mobilised to explain mismatches but not instances of better fits (page 10). However, if the data are uncertain then the better fits might equally well be due to chance.
The recognition on page 13 that a comparison with observations and other global nutrient models can identify which parts of the model can be improved but that it is difficult to determine specific areas for improvement to me points to a promising research programme.
The recognition on page 13 that hydrology plays a particularly important role in model results whereas the discussion focuses on nutrient-related processes points into the same direction. The hydrological model setup is in a parallel discussion at EGUsphere.
I agree with the assessment on page 14 that model validation and calibration to improve accuracy and reliability are needed.
I hope these comments are helpful for improving global water quality assessments.
Kind regards,
Tobias Krueger
Citation: https://doi.org/10.5194/egusphere-2025-703-CC1 -
AC3: 'Reply on CC1', Albert Nkwasa, 26 Jun 2025
We sincerely thank Dr. Tobias Krueger for taking the time to read our manuscript and share his insightful concerns. We appreciate his thoughtful feedback and the opportunity to engage with the important questions he raises about global-scale water quality modelling.
Please see attached a PDF of our detailed point-by-point response, where we also indicate changes/alterations that we intend to make in the manuscript (in blue).
-
AC3: 'Reply on CC1', Albert Nkwasa, 26 Jun 2025
-
RC1: 'Comment on egusphere-2025-703', Anonymous Referee #1, 09 Apr 2025
The manuscript presents the new CoSWAT-WQ model, which enhances our understanding of global water quality problems related to nutrients, particularly through its high spatial and temporal resolution. This represents an important step in the field of water quality research and paves the way for future model developments. I read the manuscript with great interest. It is generally well-written and presents a new global water quality model with a remarkably high resolution. This model, indeed, holds potential for scientists, policymakers and other stakeholders in the field of water quality. However, currently, the limited methodological detail makes it challenging to fully assess the scientific approach and methods applied. Moreover, the manuscript could highlight the novel insights gained from the model results rather than solely focusing on model evaluation. Improvements in the Methods, Results, and Discussion sections are needed before considering this manuscript for publication. I have explained my concerns in the comments below.
- One of the main novelties of the new CoSWAT-WQ model is its high spatial and temporal resolution at the global scale. Yet, this also comes with some concerns:
- The authors present the model as a daily time step model. Yet, the discussion includes the following: “Given the approximations inherent in model structure, uncertainties in input data, and the complexity of nutrient transport and transformation dynamics, the primary objective of global water quality models is not to predict exact daily concentrations (UNEP, 2016). Instead, they aim to identify major spatial and temporal pollution hotspots within global river networks. For this purpose, the CoSWAT-WQ model performs adequately in comparison to other global nutrient models for river nutrient loads but with room for improvement in comparison to the concentration observation data”. I would suggest the authors clarify their justification for a daily time-step model and the implications of this choice, especially at the global scale. This particularly holds for tropical and subtropical regions where scheduling using heat units may result in incorrect cropping seasons, as noted in Nkwasa et al., (2022). This raises questions such as “how well does the model capture daily patterns?” and “how should the reader interpret the model outputs?” Another option is to revise the aim of the study and clearly link the purpose of the model and how this fits the daily time step.
- I acknowledge that model evaluations at the global scale are challenging. Hence, I appreciate the authors efforts in evaluating the model by comparing the model results with the IMAGE-GNM model (Figures 4-5) and monitoring data from GEMStat (Figure 8). For the comparison with IMAGE-GNM, the authors compared simulations for the world’s 30 largest rivers. To enhance the relevance of this comparison, I suggest including information such as the percentage of the global total drainage area that these rivers cover or their share related to total nutrient exports. In Figure 5, it is difficult to interpret the different symbols used (circle for IMAGE-GNM and triangle for CoSWAT-WQ). The use of symbols seems unnecessary as each model is included on a different axis. In addition, I wonder whether the co-authors have considered comparing their model results to other large-scale water quality models (e.g. mQM for Europe and SWAT+ for Africa). Regarding the comparison to monitoring data from GEMStat (the temporal patterns in Figure 8), it becomes clear that the model may underestimate extreme events and shows a slightly advancing pattern (e.g. earlier peaks) compared to the monitoring data. Next to GEMStat, there are other databases available. For example, Jones et al. (2024) compiled a comprehensive dataset of water quality monitoring data. Including this dataset in the evaluation can further strengthen the evaluation efforts.
- The Method section seems to provide only a limited description of several important aspects of the model. Below, I outline specific areas where further detail would improve clarity and reproducibility.
- In Lines 114-116, the authors state that the timing of fertilizer and manure applications, as well as irrigation and biomass removal, can be scheduled based on calendar days or heat units, as described in Nkwasa et al. 2022. To fully understand the method, readers may need to consult previous SWAT model documentation. Given that this manuscript emphasizes the novelty of its spatial and temporal resolutions, it would be beneficial to include a brief overview of the downscaling procedures – either directly in the text or as part of a conceptual framework. For example, a brief description on how the timing of manure and fertilizer applications (which are annual datasets, as indicated in Table 1) are determined would enhance clarity. This is particularly important as the timing can be set by the user or is automatically applied by SWAT based on a specified nitrogen stress threshold (according to Neitsch et al., 2005). Additionally, it would be helpful to include a short description of how the 5-year point source input data from Beusen et al. (2022) were downscaled, as well as how the 0.5-degree resolution input data were spatially refined to 2 km resolutions (Table 1).
- Section 2.1 refers to the N and P cycles and associated nutrient pools. However, it remains unclear how the model ensures mass balance and closure of these nutrient pools across space (e.g. line 117 refers to basins whereas the model runs on HRU scale) and time (e.g. negative balances).
- In contrast to the calibrated SWAT+ model, the new CoSWAT-WQ model is uncalibrated. Yet, it remains unclear which model parameters used to be calibrated and how the uncalibrated alternative works. I suggest the authors to include a table which specifies which model parameters are calibrated in SWAT+ and how the uncalibrated approach works.
- The Results section provides a comparison of the model results with other models and monitoring data. However, it lacks a presentation of novel insights from the newly developed model. For example, what new understanding do the high-spatial-temporal-resolution results offer regarding global and local water quality assessments? From my point of view, highlighting such novelties would strengthen the value of the study and fit the aim of the study.
- The Discussion section could be strengthened by a more thorough examination of certain modeling choices and their implications on the study’s findings.
- Line 162, “The ratios are conservative and will be updated in future versions…”. I suggest reflecting on the implications of the conservative ratios on the model outputs.
- The study used and downscaled several global input datasets. In Section 4.1, the uncertainties in the input data are acknowledged in the Discussion section of the manuscript. Yet, the authors could elaborate more on the implications of these uncertainties on the results.
- To my understanding, the SWAT+ model accounts for five crops that represent different croplands (Table 2 and Figure 2, Nkwasa et al., (2022)). However, globally, many different crops exist, with each having a characteristic cropping pattern. Hence, this raises the question of whether the representative crops as used in SWAT+ are also representative for a global application of the model (e.g. rice does not seem to be included). I suggest the authors justify their choice and reflect on uncertainties associated with the crop selection.
- Lines 282-284, “…but also to improve processes such as lakes and reservoir implementations as they have a big influence on nutrient enrichment and residence times”. Could you provide some more insights on this? For example, include a reference or define ‘a big influence’.
- I suggest the authors to check the citations used.
- In Section 2.3, several global water quality models are mentioned, including DynQual, MARINA, and IMAGE-GNM. However, I recommend the authors to check the citations. For IMAGE-GNM, the model version of Beusen et al., (2015) is cited, while a more updated version exist as described in Beusen et al., (2022). For the MARINA model, the calibrated model version of Strokal et al., (2016) is cited, while the authors refer to the uncalibrated version (e.g. Micella et al., (2024).
- In Section 3.1, the authors compare the global TN export estimates of CoSWAT-WQ with other global models. Here, I also suggest the co-authors to check the values and the citations. For example, according to the manuscript estimates for total N export include “39.1 Tg/yr from IMAGE-GNM”, whereas Table 1 in Beusen et al. (2022) reports 41 Tg/yr. Perhaps the authors have excluded aquaculture for comparison purposes. If this is the case, I suggest the authors specify this. Next to the IMAGE-GNM model, estimates of the MARINA model are provided “34.8 Tg/yr from the MARINA model (Strokal et al., 2021)” whereas Strokal et al., (2021) report only inputs to rivers from urban sources.
- Lastly, I would like to suggest some textual changes:
- The title refers to a “community global SWAT+ water quality model”. Yet, the word “community” does not appear in the rest of the manuscript. In the Conclusion, the authors refer to the free accessibility and highly customizable aspects of the model. I suggest the authors to integrate the word ‘community model’ somewhere to strengthen this message.
- The abbreviations for N and P are introduced multiple times in the Introduction section; the same applies to HRU. I suggest only introducing them once.
- Line 68, “This makes physically based model approaches…”, could be rephrased to make the connection between process-based models and physically based models more clear.
- Line 82, “Despite being data-intensive, SWAT(+) models can now benefit from…”, I would start the sentence with “SWAT(+) models can now benefit from….”, as the first part may cause confusion due to the aforementioned lack of data and the computational intensity already becomes clear from the rest of the sentence.
- Line 80, “application” to “applications”
- Lines 90-91, “to identify hotspots and trends”, does this refer to seasonal trends or future trends? Or both?
- “CoSWAT-WQ” or “COSWAT-WQ” both spellings are used
- “nonpoint” and “non-point” are both used, choose one to stay consistent
- Section 2.1 is titled “SWAT+ model description”, change to “CoSWAT-WQ model description”? As I would expect a description of the newly developed model.
- Line 156, “was extracted” to “were extracted”
- Lines 259-261, “Hydrology plays a particularly important role in model results, though this discussion focuses on nutrient-related processes. Particularly, there is a need to improve river flow modelling and to incorporate water management features such as lakes and reservoirs, as highlighted in Chawanda et al. (2025).” The use of the word ‘particularly’ feels a bit odd here, as the previous sentence highlights the focus on nutrient-related processes.
References included in this review [all references suggested here are already included in the manuscript or are linked to water quality models that have already been referred to in the manuscript]
- Nkwasa, A., Chawanda, C. J., Jägermeyr, J., & Van Griensven, A. (2022). Improved representation of agricultural land use and crop management for large-scale hydrological impact simulation in Africa using SWAT+. Hydrology and Earth System Sciences, 26(1), 71-89.
- Jones, E. R., Graham, D. J., van Griensven, A., Flörke, M., & van Vliet, M. T. (2024). Blind spots in global water quality monitoring. Environmental Research Letters, 19(9), 091001.
- Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., King, K.W., 2005. SWAT theoretical documentation. Soil Water Res. Lab. Grassl. 494, 234–235.
- Beusen, A. H., & Bouwman, A. F. (2022). Future projections of river nutrient export to the global coastal ocean show persisting nitrogen and phosphorus distortion. Frontiers in Water, 4, 893585.
- Micella, I., Kroeze, C., Bak, M. P., Tang, T., Wada, Y., & Strokal, M. (2024). Future scenarios for river exports of multiple pollutants by sources and sub‐basins worldwide: Rising pollution for the Indian Ocean. Earth's Future, 12(11), e2024EF004712.
Citation: https://doi.org/10.5194/egusphere-2025-703-RC1 -
AC1: 'Reply on RC1', Albert Nkwasa, 26 Jun 2025
We sincerely thank the reviewer for their thoughtful and encouraging comments, as well as their interest in our work. We appreciate the recognition of the potential and significance of the CoSWAT-WQ model for advancing global water quality research. We acknowledge the concerns raised and will incorporate all suggestions to improve the quality and clarity of the manuscript.
Please see attached a PDF of our detailed point-by-point response, where we also indicate changes/alterations that we intend to make in the manuscript (in blue).
- One of the main novelties of the new CoSWAT-WQ model is its high spatial and temporal resolution at the global scale. Yet, this also comes with some concerns:
-
RC2: 'Comment on egusphere-2025-703', Anonymous Referee #2, 11 May 2025
Overall, the paper represents an important advance in global water quality modeling using SWAT+, offering a solid foundation for future community-driven improvements. However, uncertainty, representation of in-stream processes, and data generalization remain key challenges.
My comments mainly concern the quality and scope of the validation.
I feel that the authors primarily consider spatial validation, comparing modeled and observed nutrient loads across catchments. At the same time, temporal validation (e.g. monthly or seasonal nutrient concentrations) is largely absent. In my opinion, uncertainty analysis is also sorely lacking. While some indicators are reported, uncertainty in input data (especially for fertilizers/manures and point sources) is insufficiently considered.
Regarding fertilizer and manure inputs, these are based on national statistics, which introduce large uncertainties in local management practices.
In the paper, the authors do not assess the sensitivity of nutrient results to stream process parameters - a significant gap in water quality modeling at this scale.
The authors assume that the SWAT+ structure and parameters are universally applicable - but in reality, hydrological and biogeochemical processes vary with climate, soil, and management. For example, the same set of parameters may perform poorly in tropical systems but well in temperate ones. This is not explained.
Citation: https://doi.org/10.5194/egusphere-2025-703-RC2 -
AC2: 'Reply on RC2', Albert Nkwasa, 26 Jun 2025
We sincerely thank the reviewer for their positive feedback and welcoming our work. We appreciate the recognition of the potential and significance of the CoSWAT-WQ model for advancing global water quality research. We also acknowledge the concerns regarding the discussion of uncertainty and validation, and we will address these points carefully in the revised manuscript.
Please see attached a PDF of our detailed point-by-point response, where we also indicate changes/alterations that we intend to make in the manuscript (in blue).
-
AC2: 'Reply on RC2', Albert Nkwasa, 26 Jun 2025
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 653 | 119 | 26 | 798 | 41 | 16 | 27 |
- HTML: 653
- PDF: 119
- XML: 26
- Total: 798
- Supplement: 41
- BibTeX: 16
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1