the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 03 Jan 2025
Modeling Total Phosphorus Transport in the European Riverine System: Parameterization and Projections under Climate and Socioeconomic Scenarios
Abstract. Eutrophication is recognized as a critical ecological challenge that detrimentally affects aquatic ecosystems in both riverine and marine environments. Understanding how future human actions may influence nutrient pollution is crucial for mitigating these effects. While studies have focused on phosphorus trends related to fertilizer use in cropland areas only, this study also considers land-use changes and human development as defined by the Shared Socioeconomic Pathways (SSPs). Phosphorus transport trends are estimated using a new parameterization in a hydrological model, taking into account the evolution of agricultural, urban, and natural land use types, in line with the SSP narratives, as well as aquaculture activities, atmospheric deposition, and weathering process. Additionally, the effects of global warming are integrated by incorporating simulated hydrological data following three Representative Concentration Pathways scenarios. Total phosphorus load budgets are estimated for the four semi-enclosed European seas. The findings indicate that phosphorus losses are primarily driven by human development and land-use expansion, outweighing the response from pollution control policies and technological advances and, to some extent, hydrological changes due to climate change. The scenario data generated, and the new parameterization implemented within an Earth system model framework, can serve as a valuable resource for ecosystem modeling efforts.
- Preprint
(4902 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2024-3645', Anonymous Referee #1, 08 Feb 2025
-
AC1: 'Reply on RC1', Alberto Elizalde, 19 Feb 2025
We appreciate your critical evaluation and the provided feedback. While we acknowledge your concerns, we believe after carefully reading RC1 that there might be a misunderstanding on the goal and methods from our study. Below, we provide a detailed response to each of the R1 comments, clarifying misunderstandings and incorporating the necessary improvements.
1) A deep understanding of the nutrient dynamics within the IMAGE-GNM model (or surface water fluxes in HydroPy) would be an added value for a well-versed reader, but it is not necessary to understand our results. Given the complexity and numerous processes involved in the models providing forcing data for our simulations, any attempt to describe them would result in an unnecessary lengthy and cumbersome text. Therefore, descriptions of processes within the models providing the forcing datasets are out of the scope of our paper (i.e. precipitation and temperature from GSWP3, evaporation, transpiration and surface water fluxes in the soil scheme from HydroPy, as well as phosphorus cycling from agricultural inputs, soil vegetation dynamics, nutrient transformation, etc. simulated by the IMAGE-GNM model). We note that HydroPy (incl. its predecessor MPI-HM) and IMAGE-GNM are well established models in their respective domain of the Earth System modelling. HydroPy is a state of the art global hydrology model and its predecessor MPI-HM has contributed to the WATCH Water Model Intercomparison Project (WaterMIP; Haddeland et al., 2011) and the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP; Warszawski et al., 2014). IMAGE-GNM is a widely utilized model that has made substantial contributions to policy development, environmental assessment, and climate studies. (https://models.pbl.nl/image/Applications). These models are well documented and properly cited in our manuscript. Our research is focused solely on the transport of surplus water and phosphorus, which is assumed to be available for mobility through hydrological processes, but not on the precedent mechanisms to produce such forcings.
2) We believe that the reviewer R1 misunderstood the intention of our study and the associated model development. Our approach does not focus on a process-based modeling, which is typically used at the plot or field scales. Instead, our objective was to introduce a model setup for large-scale applications on global or continental scales. In the respective large-scale models, sub-grid scale processes cannot be explicitly resolved, so that we developed a suitable parameterization within the more conceptual-based HD model. In such a model, conceptual or statistical approaches are appropriate, e.g. in form of a scheme or parameterization. Given this objective, phosphorus transport (Section 2.3.4) and the distinction between total phosphorus and particulate/dissolved forms dynamics in rivers (Section 3.1) are addressed and discussed in our manuscript. These discussions were critical in shaping the newly developed parameterisation.Our lateral phosphorus transport approach with the hydrological HD model is similar to that of Beusen et al. (2016). In their study, nutrient transport was estimated using nutrient concentration data from the IMAGE model and discharge data from the ERA40 reanalysis dataset to force the hydrological model PCR-GLOBWB with a yearly timestep. According to Beusen et al. (2016), there is no feedback from PCR-GLOBWB to IMAGE, meaning the models operate in one-way coupling mode. In our manuscript, Section 2.3, line 148, clearly states a brief explanation on how P loads were calculated in Beusen et al. (2016). To enhance clarity to this matter, we will apply the following change to the introduction section:
Line 53: Following similar approach to Beusen et al. (2016) for modelling TP transport in river systems, we employ the Hydrological Discharge (HD) model ...
Additionally, we will make clear that our model setup and development aims at large-scale applications on global or continental scales. Consequently, we will point this out more thoroughly in the abstract and introduction:
Line 6: Phosphorus transport trends are estimated using a novel parameterization designed for large-scale applications at global and continental scales within a hydrological model.
Line 54: Due to the lack of explicit P concentration data in future scenarios, a new scheme was developed to parameterize P losses from agricultural sources for large-scale applications at global and continental scales, based on N data from fertilizer application information and land-use cover, both from the Land-Use Harmonization 2 (LUH2) database (LUH, 2018; Hurtt et al., 2020).
3) Unfortunately, we do not understand how the reviewer has concluded this comment. We dedicated Section 2.3.4 to provide a full description for the calculation of the P loads from croplands for the scenario simulations. That section provides detailed explanations of each decision made in the implementation. The methodology for calculating P loads from croplands (and P aggregates in general) in the hindcast simulation is described in Section 2.3.
4) Wastewater is not treated differently to the other components of total phosphorus. All TP aggregates are covered in Section 2.2.5. Both, P concentrations and loads (from all aggregates) are addressed and relevant, as they are essential for understanding the model output. Analysing only P loads would not allow for distinguishing between the signal from phosphorus inputs due to changing societal dynamics and the hydrological signal in the water cycle by global warming effects. Throughout the manuscript (in text and figures), whenever concentrations or loads are analysed, it is clearly indicated which quantity is being referred to.
5) We used a conservative remapping for the forcings when transitioning between grids with different projections and resolutions. This ensures the preservation of P concentration and runoff budgets. This detail will be included in the next version of the manuscript by modifying:
Line 132: Forcing fields, surface runoff and drainage, were interpolated from the original grid resolution of 0.5° to 5 arcmin using a conservative remapping method to ensure the budgets remain unchanged.
Line 144: Same as in for the hydrological forcing fields, P concentrations fields were interpolated from a 0.5° grid to 5 arcmin resolution using a conservative remapping approach, ensuring that budget integrity is maintained.
6) We disagree with the suggestion that ‘transport’ implies internal river ‘processes’ as this appears to be a personal interpretation by the reviewer R1. Modeling biogeochemical processes related to P transformation in rivers is beyond the scope of our approach. In our study, P is treated as an agent transported by river discharge, as explained in our response to point 2). To rule out any impact of biogeochemical processes on the general signal of TP, we investigated seasonality in observational station data (Sections 2.3.2 and 3.1) and found no significant effect. We do not see how the title could lead to misinterpretation.
7) The concept and implementation of the fPloss parameter is carefully described and discussed in section 2.3.1. fPloss conceptually represents the fraction of P lost from croplands due to fertilizers application. Given its high uncertainty, this factor is used as an adjustable tuning parameter to optimise the model’s output to better match IMAGE-GNM data, with its purpose being to reduce model uncertainties. The optimised value of 3.5%, as noted in the mentioned section, was determined using a trial-and-error method, comparing averaged catchment values from our scheme results with original IMAGE-GNM data. The trial-and-error method is a valid and commonly used approach for tuning large-scale models, given the complexity of the process involved. Direct analytical optimisation is infeasible in this case due to the multiple pathways through which phosphorus travels from land to rivers, as highlighted in the referenced literature of our manuscript (Hart et al, 2004, Lun et al., 2018; Hua and Zhu, 2020.). Any other smaller or larger values would result in a degradation of model performance. We see no reason why this optimisation process should raise concerns about the validity and robustness of our results, especially since Section 3.2 presents a detailed model validation, direclty comparing the model output to independent observational station data. However, we will add a brief description of how the optimisation was done in the next version of the manuscript.
Line 196: Therefore, in this parametrization, this value is treated as a tuning factor. We determine its value using a trial-and-error method, comparing averaged catchment values from our scheme results with original IMAGE-GNM data. The trial-and-error method is a valid and commonly used approach for tuning large-scale models, given the complexity of the process involved. Direct analytical optimisation is infeasible in this case due to the multiple pathways through which phosphorus travels from land to rivers, as highlighted in Hart et al, 2004, Lun et al., 2018 and Hua and Zhu, 2020. Any other smaller or larger values would result in a degradation of model performance.
8) This comment from R1 is incorrect. P emissions from industry and manure are accounted for in the P concentrations within the IMAGE-GNM model, which we use here as forcings. As described in the IMAGE-GNM model documentation, industry emissions are included under the concept of wastewater, while manure is under agricultural land emissions. All phosphorus inputs provided by IMAGE-GNM model are listed in Section 2.2.5.9) The new parametrization related to phosphorus is not process-based. We did not intend to develop a process-based model. Please refer to our answer to point 1). The suggested literature will be considered for the next version of the manuscript.
Citation: https://doi.org/10.5194/egusphere-2024-3645-AC1
-
AC1: 'Reply on RC1', Alberto Elizalde, 19 Feb 2025
-
RC2: 'Comment on egusphere-2024-3645', Anonymous Referee #2, 23 May 2025
Modeling Total Phosphorus Transport in the European Riverine System: Parameterization and Projections under Climate and Socioeconomic Scenarios by Elizalde et al.
First of all, this topic has my interest! The authors claim that the phosphorus (P) models have received less attention than nitrogen (N). I agree with them. Also, because modeling phosphorus is much harder than nitrogen (in my view).
However, using the same approach for the transport of P from land to sea as N, is not going to work. I made this claim based on the introduction, but I am not sure what the authors really did. The main reason is that there is no description of for example TP is transported in the river. Is there any loss process activated here? No description how they convert yearly data from Beusen et al. (2016) into daily/monthly timesteps. No description how to correct the TP input on the land use maps of LUH2. It is possible that for example Beusen et al. claims a grid cell has agricultural land, but LUH2 has only natural land. It seems like they focus on the hydrology, but not on the TP. It also does not help that “load” and “concentration” are mixed up and wrongly used. In the abstract, they claim to calculate the future trends based on all sources (more than fertilizer), but I only see one description of the use of N fertilizer with a conversion factor. I don’t think this is a valid assumption (see e.g. papers from Mogollon et al.). To my surprise, they don’t compare their results with Beusen et al. (2016), who also calculated loads to the coast. Why not?
To conclude, there is so much wrong with this article, that I don’t want to give specific comments. My recommendation is that this article should be rejected, and my advice is not to publish this somewhere else in this poor state!
Citation: https://doi.org/10.5194/egusphere-2024-3645-RC2 -
AC2: 'Reply on RC2', Alberto Elizalde, 17 Jun 2025
Comments from RC2:
First of all, this topic has my interest! The authors claim that the phosphorus (P) models have received less attention than nitrogen (N). I agree with them. Also, because modeling phosphorus is much harder than nitrogen (in my view).However, using the same approach for the transport of P from land to sea as N, is not going to work. I made this claim based on the introduction, but I am not sure what the authors really did. The main reason is that there is no description of for example TP is transported in the river. Is there any loss process activated here? No description how they convert yearly data from Beusen et al. (2016) into daily/monthly timesteps. No description how to correct the TP input on the land use maps of LUH2. It is possible that for example Beusen et al. claims a grid cell has agricultural land, but LUH2 has only natural land. It seems like they focus on the hydrology, but not on the TP. It also does not help that “load” and “concentration” are mixed up and wrongly used. In the abstract, they claim to calculate the future trends based on all sources (more than fertilizer), but I only see one description of the use of N fertilizer with a conversion factor. I don’t think this is a valid assumption (see e.g. papers from Mogollon et al.). To my surprise, they don’t compare their results with Beusen et al. (2016), who also calculated loads to the coast. Why not?
To conclude, there is so much wrong with this article, that I don’t want to give specific comments. My recommendation is that this article should be rejected, and my advice is not to publish this somewhere else in this poor state!
Reply to RC2:
“... there is no description of for example TP is transported in the river. Is there any loss process activated here?”
This point is discussed in Section 2.1, paragraph 3. The hydrological model HD includes a subroutine for transporting substances within river flow, using the same water flow calculations described in paragraph 1 of the same section. This is the same approach was used by Beusen et al. (2016) for modeling nitrogen (N) and total phosphorus (TP) transport in rivers, although they employed the PCR-GLOBWB hydrological model, whereas we use the HD model.Beusen et al. (2016) do not report any feedback from PCR-GLOBWB to the IMAGE model, indicating a one-way coupling between the models. They also do not account for TP outgassing from water bodies. Our approach follows the same assumptions.
To improve clarity on this matter, we will implement the following modifications in the manuscript:
In the introduction section:
Line 53: Following similar approach to Beusen et al. (2016) for modelling TP transport in river systems, we employ the Hydrological Discharge (HD) model ...In Section 2.1 paragraph 3:
Line 73: This method has been applied in previous studies, such as Gehlot et al. (2024), to simulate the transport of dissolved organic carbon. A similar approach was used by Beusen et al. (2016) for modeling N and TP transport in rivers, although they employed the PCR-GLOBWB hydrological model, whereas we use the HD model. They also do not account for TP outgassing from water bodies. Our approach follows the same assumptions.
“No description how they convert yearly data from Beusen et al. (2016) into daily/monthly timesteps.”
To perform this conversion, we divided the yearly values by the number of days in the corresponding year, following a Gregorian calendar, to obtain daily values. Based on our analysis of TP observation data, we found no evidence of a seasonal signal in TP (see Section 3.1). This supports our decision to use a straightforward conversion from yearly to daily data. For clarification we will add:
In Section 2.3, paragraph 1:
Line 152: To perform the conversion from yearly to daily data, we divided the yearly values by the number of days in the corresponding year, following a Gregorian calendar, to obtain daily values. Our analysis of TP observation data revealed no discernible seasonal signal in TP concentrations (see Section 3.1), thereby supporting the use of this simplified annual-to-daily conversion.
“No description how to correct the TP input on the land use maps of LUH2. It is possible that for example Beusen et al. claims a grid cell has agricultural land, but LUH2 has only natural land. It seems like they focus on the hydrology, but not on the TP.”
This is explained in Section 2.3, paragraph 2. To transform data fields between grids, we applied horizontal interpolation using a conservative remapping method. This approach preserves the integrated quantity and accounts for partial overlaps between grid cells. As described, the transformation is applied to each LUH2 land-use type individually.Conflicted grid boxes are a common problem in grid transformations. However, because the conservative method ensures the preservation of budgets, and the HD model is designed for large-scale applications at global and continental levels, occasional local inconsistencies have a negligible impact on our overall results.
“It also does not help that “load” and “concentration” are mixed up and wrongly used.”
We carefully reviewed the entire manuscript and could not identify any instance where "load" and "concentration" were incorrectly used or confused. Throughout the text and figures, we have explicitly stated whether the analysis pertains to concentrations or loads, and the corresponding units are clearly indicated in each case.
“In the abstract, they claim to calculate the future trends based on all sources (more than fertilizer), but I only see one description of the use of N fertilizer with a conversion factor. I don’t think this is a valid assumption (see e.g. papers from Mogollon et al.).”
As explained in Section 2.3 paragraph 3: We use indeed all sources of TP in our scenario simulations: agricultural and natural land, weathering, allochthonous organic matter input to rivers, aquaculture, wastewater and atmospheric deposition.
This paragraph also includes how these aggregates were processed to be used in the scenario simulations: we use inferred concentrations from aquaculture and allochthonous organic matter from IMAGE-GNM. Concentrations from wastewater, natural land, and pasture areas from IMAGE-GNM were remapped into urban, forest and pasture areas using the corresponding land use types from LUH2 dataset. This transformation assigns concentrations to each of the LUH2 land types, ensuring that values remain conservative at the catchment scale for the years with the common period. Concentrations for weathering, atmospheric deposition and cropland were calculated are described in Sections 2.3.1, 2.3.2 and 2.3.3, respectively.
“To my surprise, they don’t compare their results with Beusen et al. (2016), who also calculated loads to the coast. Why not?”
We compared our results with those of Beusen et al. (2016). We used the P concentrations from Beusen et al. (2016) as the driving dataset for our simulations. As stated in Section 2.2.5, our hindcast simulation is forced with only the Beusen et al. (2016) data. Therefore, this simulation produces identical results to those of Beusen et al. (2016). In fact, comparing our hindcast simulation with the Beusen et al. (2016) dataset was part of the validation process. As stated in the Methods section (Section 2.3, paragraph 1, line 150):“To verify that no bias was introduced by the use of different hydrological datasets, we compared the inferred concentration from the weathering aggregate with the concentration calculated directly using the weathering parameterization by Hartmann et al. (2014b). Both magnitudes were consistent.”
Moreover, the hindcast simulation was also evaluated using statistical analysis using observed data, as described in the Model Performance section (Section 3.2, paragraph 2, line 258):
“Results for TP loads from the hindcast simulation (driven with forcings from the reanalysis dataset and TP concentrations from IMAGE-GNM) generally exceed this -0.4 threshold (left panel in Fig. 1).”
And compared with our historical simulation on the next paragraph (Section 3.2, paragraph 3, line 1):
“The comparison of the hindcast and the historical simulation is done using two of the KGE components, CV and mean ratios, for TP concentrations and discharge (Fig. 2).”Citation: https://doi.org/10.5194/egusphere-2024-3645-AC2
-
AC2: 'Reply on RC2', Alberto Elizalde, 17 Jun 2025
Status: closed
-
RC1: 'Comment on egusphere-2024-3645', Anonymous Referee #1, 08 Feb 2025
The manuscript of Elizalde and co-authors is entitled “Modeling Total Phosphorus Transport in the European Riverine System: Parameterization and Projections under Climate and Socioeconomic Scenarios”. After a thorough evaluation, I regret to inform the authors that I cannot recommend this manuscript for publication in its current form. Below are the detailed reasons for my decision:
- Dependence on the IMAGE-GNM Model: The manuscript requires a deep understanding of the IMAGE-GNM model (Beusen et al., 2015, 2016) for comprehension. This reliance on a specific model limits the accessibility of the manuscript to a broader audience who may not be familiar with this model.
- Lack of familiarity with P transport processes: The authors a ppear to lack sufficient knowledge of phosphorus (P) transport processes in rivers and soils. This gap in understanding is evident throughout the manuscript and affects the credibility of the research.
- Non-traceable P load calculations: The methodology for calculating P loads from croplands is not traceable. The manuscript does not provide a clear and transparent description of the methods used, which is essential for reproducibility and validation.
- Ambiguity in wastewater mapping: There is a lack of clarity regarding the mapping of wastewater loading or concentration. It is unclear whether the manuscript focuses on loadings or concentrations, leading to confusion and undermining the reliability of the findings.
- Inadequate description of concentration mapping: The process of mapping concentrations from 0.5 degree to 5 arcmin is not described. Detailed methodological descriptions are crucial for understanding and replicating the research.
- Misleading title: The title of the manuscript is misleading as it suggests that total phosphorus (TP) transport would consider riverine processes, which are not adequately addressed in the manuscript.
- Unexplained optimization of factor fPloss: The manuscript does not describe how fPloss was optimized. This omission raises questions about the validity and robustness of the results.
- Exclusion of key P emission sources: The manuscript does not consider P emissions from industry and manure, which are significant sources and should be included in a comprehensive study of P transport.
- Non-process-based model and literature gaps: The model used is not process-based, at least processes are not or not well described. The authors should consider additional literature describing TP modeling, such as papers on the Marina and Global News models, Grizetti et al. (2021), and Beusen et al. (2022).
Given these significant issues, I recommend rejecting the manuscript.
Citation: https://doi.org/10.5194/egusphere-2024-3645-RC1 -
AC1: 'Reply on RC1', Alberto Elizalde, 19 Feb 2025
We appreciate your critical evaluation and the provided feedback. While we acknowledge your concerns, we believe after carefully reading RC1 that there might be a misunderstanding on the goal and methods from our study. Below, we provide a detailed response to each of the R1 comments, clarifying misunderstandings and incorporating the necessary improvements.
1) A deep understanding of the nutrient dynamics within the IMAGE-GNM model (or surface water fluxes in HydroPy) would be an added value for a well-versed reader, but it is not necessary to understand our results. Given the complexity and numerous processes involved in the models providing forcing data for our simulations, any attempt to describe them would result in an unnecessary lengthy and cumbersome text. Therefore, descriptions of processes within the models providing the forcing datasets are out of the scope of our paper (i.e. precipitation and temperature from GSWP3, evaporation, transpiration and surface water fluxes in the soil scheme from HydroPy, as well as phosphorus cycling from agricultural inputs, soil vegetation dynamics, nutrient transformation, etc. simulated by the IMAGE-GNM model). We note that HydroPy (incl. its predecessor MPI-HM) and IMAGE-GNM are well established models in their respective domain of the Earth System modelling. HydroPy is a state of the art global hydrology model and its predecessor MPI-HM has contributed to the WATCH Water Model Intercomparison Project (WaterMIP; Haddeland et al., 2011) and the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP; Warszawski et al., 2014). IMAGE-GNM is a widely utilized model that has made substantial contributions to policy development, environmental assessment, and climate studies. (https://models.pbl.nl/image/Applications). These models are well documented and properly cited in our manuscript. Our research is focused solely on the transport of surplus water and phosphorus, which is assumed to be available for mobility through hydrological processes, but not on the precedent mechanisms to produce such forcings.
2) We believe that the reviewer R1 misunderstood the intention of our study and the associated model development. Our approach does not focus on a process-based modeling, which is typically used at the plot or field scales. Instead, our objective was to introduce a model setup for large-scale applications on global or continental scales. In the respective large-scale models, sub-grid scale processes cannot be explicitly resolved, so that we developed a suitable parameterization within the more conceptual-based HD model. In such a model, conceptual or statistical approaches are appropriate, e.g. in form of a scheme or parameterization. Given this objective, phosphorus transport (Section 2.3.4) and the distinction between total phosphorus and particulate/dissolved forms dynamics in rivers (Section 3.1) are addressed and discussed in our manuscript. These discussions were critical in shaping the newly developed parameterisation.Our lateral phosphorus transport approach with the hydrological HD model is similar to that of Beusen et al. (2016). In their study, nutrient transport was estimated using nutrient concentration data from the IMAGE model and discharge data from the ERA40 reanalysis dataset to force the hydrological model PCR-GLOBWB with a yearly timestep. According to Beusen et al. (2016), there is no feedback from PCR-GLOBWB to IMAGE, meaning the models operate in one-way coupling mode. In our manuscript, Section 2.3, line 148, clearly states a brief explanation on how P loads were calculated in Beusen et al. (2016). To enhance clarity to this matter, we will apply the following change to the introduction section:
Line 53: Following similar approach to Beusen et al. (2016) for modelling TP transport in river systems, we employ the Hydrological Discharge (HD) model ...
Additionally, we will make clear that our model setup and development aims at large-scale applications on global or continental scales. Consequently, we will point this out more thoroughly in the abstract and introduction:
Line 6: Phosphorus transport trends are estimated using a novel parameterization designed for large-scale applications at global and continental scales within a hydrological model.
Line 54: Due to the lack of explicit P concentration data in future scenarios, a new scheme was developed to parameterize P losses from agricultural sources for large-scale applications at global and continental scales, based on N data from fertilizer application information and land-use cover, both from the Land-Use Harmonization 2 (LUH2) database (LUH, 2018; Hurtt et al., 2020).
3) Unfortunately, we do not understand how the reviewer has concluded this comment. We dedicated Section 2.3.4 to provide a full description for the calculation of the P loads from croplands for the scenario simulations. That section provides detailed explanations of each decision made in the implementation. The methodology for calculating P loads from croplands (and P aggregates in general) in the hindcast simulation is described in Section 2.3.
4) Wastewater is not treated differently to the other components of total phosphorus. All TP aggregates are covered in Section 2.2.5. Both, P concentrations and loads (from all aggregates) are addressed and relevant, as they are essential for understanding the model output. Analysing only P loads would not allow for distinguishing between the signal from phosphorus inputs due to changing societal dynamics and the hydrological signal in the water cycle by global warming effects. Throughout the manuscript (in text and figures), whenever concentrations or loads are analysed, it is clearly indicated which quantity is being referred to.
5) We used a conservative remapping for the forcings when transitioning between grids with different projections and resolutions. This ensures the preservation of P concentration and runoff budgets. This detail will be included in the next version of the manuscript by modifying:
Line 132: Forcing fields, surface runoff and drainage, were interpolated from the original grid resolution of 0.5° to 5 arcmin using a conservative remapping method to ensure the budgets remain unchanged.
Line 144: Same as in for the hydrological forcing fields, P concentrations fields were interpolated from a 0.5° grid to 5 arcmin resolution using a conservative remapping approach, ensuring that budget integrity is maintained.
6) We disagree with the suggestion that ‘transport’ implies internal river ‘processes’ as this appears to be a personal interpretation by the reviewer R1. Modeling biogeochemical processes related to P transformation in rivers is beyond the scope of our approach. In our study, P is treated as an agent transported by river discharge, as explained in our response to point 2). To rule out any impact of biogeochemical processes on the general signal of TP, we investigated seasonality in observational station data (Sections 2.3.2 and 3.1) and found no significant effect. We do not see how the title could lead to misinterpretation.
7) The concept and implementation of the fPloss parameter is carefully described and discussed in section 2.3.1. fPloss conceptually represents the fraction of P lost from croplands due to fertilizers application. Given its high uncertainty, this factor is used as an adjustable tuning parameter to optimise the model’s output to better match IMAGE-GNM data, with its purpose being to reduce model uncertainties. The optimised value of 3.5%, as noted in the mentioned section, was determined using a trial-and-error method, comparing averaged catchment values from our scheme results with original IMAGE-GNM data. The trial-and-error method is a valid and commonly used approach for tuning large-scale models, given the complexity of the process involved. Direct analytical optimisation is infeasible in this case due to the multiple pathways through which phosphorus travels from land to rivers, as highlighted in the referenced literature of our manuscript (Hart et al, 2004, Lun et al., 2018; Hua and Zhu, 2020.). Any other smaller or larger values would result in a degradation of model performance. We see no reason why this optimisation process should raise concerns about the validity and robustness of our results, especially since Section 3.2 presents a detailed model validation, direclty comparing the model output to independent observational station data. However, we will add a brief description of how the optimisation was done in the next version of the manuscript.
Line 196: Therefore, in this parametrization, this value is treated as a tuning factor. We determine its value using a trial-and-error method, comparing averaged catchment values from our scheme results with original IMAGE-GNM data. The trial-and-error method is a valid and commonly used approach for tuning large-scale models, given the complexity of the process involved. Direct analytical optimisation is infeasible in this case due to the multiple pathways through which phosphorus travels from land to rivers, as highlighted in Hart et al, 2004, Lun et al., 2018 and Hua and Zhu, 2020. Any other smaller or larger values would result in a degradation of model performance.
8) This comment from R1 is incorrect. P emissions from industry and manure are accounted for in the P concentrations within the IMAGE-GNM model, which we use here as forcings. As described in the IMAGE-GNM model documentation, industry emissions are included under the concept of wastewater, while manure is under agricultural land emissions. All phosphorus inputs provided by IMAGE-GNM model are listed in Section 2.2.5.9) The new parametrization related to phosphorus is not process-based. We did not intend to develop a process-based model. Please refer to our answer to point 1). The suggested literature will be considered for the next version of the manuscript.
Citation: https://doi.org/10.5194/egusphere-2024-3645-AC1
-
RC2: 'Comment on egusphere-2024-3645', Anonymous Referee #2, 23 May 2025
Modeling Total Phosphorus Transport in the European Riverine System: Parameterization and Projections under Climate and Socioeconomic Scenarios by Elizalde et al.
First of all, this topic has my interest! The authors claim that the phosphorus (P) models have received less attention than nitrogen (N). I agree with them. Also, because modeling phosphorus is much harder than nitrogen (in my view).
However, using the same approach for the transport of P from land to sea as N, is not going to work. I made this claim based on the introduction, but I am not sure what the authors really did. The main reason is that there is no description of for example TP is transported in the river. Is there any loss process activated here? No description how they convert yearly data from Beusen et al. (2016) into daily/monthly timesteps. No description how to correct the TP input on the land use maps of LUH2. It is possible that for example Beusen et al. claims a grid cell has agricultural land, but LUH2 has only natural land. It seems like they focus on the hydrology, but not on the TP. It also does not help that “load” and “concentration” are mixed up and wrongly used. In the abstract, they claim to calculate the future trends based on all sources (more than fertilizer), but I only see one description of the use of N fertilizer with a conversion factor. I don’t think this is a valid assumption (see e.g. papers from Mogollon et al.). To my surprise, they don’t compare their results with Beusen et al. (2016), who also calculated loads to the coast. Why not?
To conclude, there is so much wrong with this article, that I don’t want to give specific comments. My recommendation is that this article should be rejected, and my advice is not to publish this somewhere else in this poor state!
Citation: https://doi.org/10.5194/egusphere-2024-3645-RC2 -
AC2: 'Reply on RC2', Alberto Elizalde, 17 Jun 2025
Comments from RC2:
First of all, this topic has my interest! The authors claim that the phosphorus (P) models have received less attention than nitrogen (N). I agree with them. Also, because modeling phosphorus is much harder than nitrogen (in my view).However, using the same approach for the transport of P from land to sea as N, is not going to work. I made this claim based on the introduction, but I am not sure what the authors really did. The main reason is that there is no description of for example TP is transported in the river. Is there any loss process activated here? No description how they convert yearly data from Beusen et al. (2016) into daily/monthly timesteps. No description how to correct the TP input on the land use maps of LUH2. It is possible that for example Beusen et al. claims a grid cell has agricultural land, but LUH2 has only natural land. It seems like they focus on the hydrology, but not on the TP. It also does not help that “load” and “concentration” are mixed up and wrongly used. In the abstract, they claim to calculate the future trends based on all sources (more than fertilizer), but I only see one description of the use of N fertilizer with a conversion factor. I don’t think this is a valid assumption (see e.g. papers from Mogollon et al.). To my surprise, they don’t compare their results with Beusen et al. (2016), who also calculated loads to the coast. Why not?
To conclude, there is so much wrong with this article, that I don’t want to give specific comments. My recommendation is that this article should be rejected, and my advice is not to publish this somewhere else in this poor state!
Reply to RC2:
“... there is no description of for example TP is transported in the river. Is there any loss process activated here?”
This point is discussed in Section 2.1, paragraph 3. The hydrological model HD includes a subroutine for transporting substances within river flow, using the same water flow calculations described in paragraph 1 of the same section. This is the same approach was used by Beusen et al. (2016) for modeling nitrogen (N) and total phosphorus (TP) transport in rivers, although they employed the PCR-GLOBWB hydrological model, whereas we use the HD model.Beusen et al. (2016) do not report any feedback from PCR-GLOBWB to the IMAGE model, indicating a one-way coupling between the models. They also do not account for TP outgassing from water bodies. Our approach follows the same assumptions.
To improve clarity on this matter, we will implement the following modifications in the manuscript:
In the introduction section:
Line 53: Following similar approach to Beusen et al. (2016) for modelling TP transport in river systems, we employ the Hydrological Discharge (HD) model ...In Section 2.1 paragraph 3:
Line 73: This method has been applied in previous studies, such as Gehlot et al. (2024), to simulate the transport of dissolved organic carbon. A similar approach was used by Beusen et al. (2016) for modeling N and TP transport in rivers, although they employed the PCR-GLOBWB hydrological model, whereas we use the HD model. They also do not account for TP outgassing from water bodies. Our approach follows the same assumptions.
“No description how they convert yearly data from Beusen et al. (2016) into daily/monthly timesteps.”
To perform this conversion, we divided the yearly values by the number of days in the corresponding year, following a Gregorian calendar, to obtain daily values. Based on our analysis of TP observation data, we found no evidence of a seasonal signal in TP (see Section 3.1). This supports our decision to use a straightforward conversion from yearly to daily data. For clarification we will add:
In Section 2.3, paragraph 1:
Line 152: To perform the conversion from yearly to daily data, we divided the yearly values by the number of days in the corresponding year, following a Gregorian calendar, to obtain daily values. Our analysis of TP observation data revealed no discernible seasonal signal in TP concentrations (see Section 3.1), thereby supporting the use of this simplified annual-to-daily conversion.
“No description how to correct the TP input on the land use maps of LUH2. It is possible that for example Beusen et al. claims a grid cell has agricultural land, but LUH2 has only natural land. It seems like they focus on the hydrology, but not on the TP.”
This is explained in Section 2.3, paragraph 2. To transform data fields between grids, we applied horizontal interpolation using a conservative remapping method. This approach preserves the integrated quantity and accounts for partial overlaps between grid cells. As described, the transformation is applied to each LUH2 land-use type individually.Conflicted grid boxes are a common problem in grid transformations. However, because the conservative method ensures the preservation of budgets, and the HD model is designed for large-scale applications at global and continental levels, occasional local inconsistencies have a negligible impact on our overall results.
“It also does not help that “load” and “concentration” are mixed up and wrongly used.”
We carefully reviewed the entire manuscript and could not identify any instance where "load" and "concentration" were incorrectly used or confused. Throughout the text and figures, we have explicitly stated whether the analysis pertains to concentrations or loads, and the corresponding units are clearly indicated in each case.
“In the abstract, they claim to calculate the future trends based on all sources (more than fertilizer), but I only see one description of the use of N fertilizer with a conversion factor. I don’t think this is a valid assumption (see e.g. papers from Mogollon et al.).”
As explained in Section 2.3 paragraph 3: We use indeed all sources of TP in our scenario simulations: agricultural and natural land, weathering, allochthonous organic matter input to rivers, aquaculture, wastewater and atmospheric deposition.
This paragraph also includes how these aggregates were processed to be used in the scenario simulations: we use inferred concentrations from aquaculture and allochthonous organic matter from IMAGE-GNM. Concentrations from wastewater, natural land, and pasture areas from IMAGE-GNM were remapped into urban, forest and pasture areas using the corresponding land use types from LUH2 dataset. This transformation assigns concentrations to each of the LUH2 land types, ensuring that values remain conservative at the catchment scale for the years with the common period. Concentrations for weathering, atmospheric deposition and cropland were calculated are described in Sections 2.3.1, 2.3.2 and 2.3.3, respectively.
“To my surprise, they don’t compare their results with Beusen et al. (2016), who also calculated loads to the coast. Why not?”
We compared our results with those of Beusen et al. (2016). We used the P concentrations from Beusen et al. (2016) as the driving dataset for our simulations. As stated in Section 2.2.5, our hindcast simulation is forced with only the Beusen et al. (2016) data. Therefore, this simulation produces identical results to those of Beusen et al. (2016). In fact, comparing our hindcast simulation with the Beusen et al. (2016) dataset was part of the validation process. As stated in the Methods section (Section 2.3, paragraph 1, line 150):“To verify that no bias was introduced by the use of different hydrological datasets, we compared the inferred concentration from the weathering aggregate with the concentration calculated directly using the weathering parameterization by Hartmann et al. (2014b). Both magnitudes were consistent.”
Moreover, the hindcast simulation was also evaluated using statistical analysis using observed data, as described in the Model Performance section (Section 3.2, paragraph 2, line 258):
“Results for TP loads from the hindcast simulation (driven with forcings from the reanalysis dataset and TP concentrations from IMAGE-GNM) generally exceed this -0.4 threshold (left panel in Fig. 1).”
And compared with our historical simulation on the next paragraph (Section 3.2, paragraph 3, line 1):
“The comparison of the hindcast and the historical simulation is done using two of the KGE components, CV and mean ratios, for TP concentrations and discharge (Fig. 2).”Citation: https://doi.org/10.5194/egusphere-2024-3645-AC2
-
AC2: 'Reply on RC2', Alberto Elizalde, 17 Jun 2025
Data sets
Total phosphorus transport and river runoff over Europe Alberto Elizalde and Stefan Hagemann https://doi.org/10.26050/WDCC/cD_Priver_Eur
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 413 | 58 | 18 | 489 | 16 | 38 |
- HTML: 413
- PDF: 58
- XML: 18
- Total: 489
- BibTeX: 16
- EndNote: 38
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript of Elizalde and co-authors is entitled “Modeling Total Phosphorus Transport in the European Riverine System: Parameterization and Projections under Climate and Socioeconomic Scenarios”. After a thorough evaluation, I regret to inform the authors that I cannot recommend this manuscript for publication in its current form. Below are the detailed reasons for my decision:
Given these significant issues, I recommend rejecting the manuscript.