Synergizing grassland and soil system model expertise by coupling GRASSMIND (v2.0) and BODIUM (v1.2)

Kantzenbach, Matthes; König, Sara; Reitz, Thomas; Schädler, Martin; Taubert, Franziska

doi:10.5194/egusphere-2026-1713

Preprints

https://doi.org/10.5194/egusphere-2026-1713

Preprints

04 May 2026

| 04 May 2026

Synergizing grassland and soil system model expertise by coupling GRASSMIND (v2.0) and BODIUM (v1.2)

Matthes Kantzenbach, Sara König, Thomas Reitz, Martin Schädler, and Franziska Taubert

Abstract. Ecological models often have a specific focus, simplifying other system components. In the context of landscapes under climate change, it is increasingly important to include all relevant components and their interactions in detail in the models. Grassland models, advising management strategies for this important vegetation type of European landscapes, often lack detailed and reliable hydrological and soil resource dynamics that influence plant growth in grasslands. This study investigates the potential to overcome this issue by coupling an existing grassland with a soil system model, making use of their expertise in a specialized area. Here, the individual- and process-based grassland model GRASSMIND is coupled to the systemic soil model BODIUM using the coupling framework FINAM. The influence of soil water on grassland dynamics is shown to be more reliable with the coupled models than with GRASSMIND alone. In addition, the coupling offers the potential to tackle shortcomings in the representation of other plant processes such as root growth. However, the most urgent challenge is to overcome the ambiguity in the parametrization of GRASSMIND itself. Our experience suggests that maintaining the native models as independent components provides flexibility for future improvements but also complicates updating parametrizations in the combined system as the individual models evolve.

Received: 27 Mar 2026 – Discussion started: 04 May 2026

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

Matthes Kantzenbach, Sara König, Thomas Reitz, Martin Schädler, and Franziska Taubert

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-1713', Anonymous Referee #1, 19 May 2026

The integration of GRASSMIND and BODIUM is technically robust, and the improvements in soil moisture modeling are acknowledged as valuable. Nevertheless, the reviewer identifies three principal areas requiring attention.
First, a major concern is the restricted accessibility of GRASSMIND v2.0. As noted in the manuscript (p. 28), the code is currently "only available on request," hindering transparency and reproducibility.
Second, the calibration process reveals parameter ambiguity. For example, the "overlap factor" (p. 10) consistently reaches its maximum limit of 1.0 across all simulations (Table 2), and the rooting parameters (r1–r2) exhibit considerable variability between calibration attempts (Tables 2–3).
Third, the post-2020 validation is inadequately analysed. Specific issues include a ~7–9 percentage point overestimation of deep-layer soil moisture (p. 17) and an unsatisfactory representation of early reseeding yields, particularly in the absence of sensitivity analyses (p. 16).
Beyond these primary concerns, this reviewer identifies several interpretative ambiguities. These include circular reasoning in water-limitation claims, as the coupled model's water limitation is judged "more reliable" primarily because it inherits BODIUM's moisture calculations (p. 22). There is also no detailed breakdown of drought-year yield losses, meaning the individual contributions of water, nitrogen and temperature to the actual yield reduction (not just limitation factor) are never quantified (Figure 7). Moreover, the significance of the self-coupling test is overstated: while technically valid, its scope is limited to verifying data-transfer mechanisms rather than confirming ecological equivalence (Appendix D).
Several minor points also warrant attention. These include figure captions lacking full clarity (e.g. specifying that GRASSMIND executes first in each timestep), unweighted objective functions despite variations in measurement uncertainty (e.g. early-year biomass and sensor-based soil moisture with ±3% accuracy are given equal weight; p. 11), an inconsistent treatment of the 2020 plowing/tillage event (GRASSMIND is restarted while BODIUM continues running and retains soil memory; p. 9, p. 24), the ambiguous presentation of model‑only nitrogen outputs (Figure 6 shows D2–D3 without measurements, though the caption states only D1 was measured) and inadequate discussion of plot‑to‑plot variability (the measured variance across the five GCEF parcels is never compared to the model's narrow ensemble envelopes; p. 26). This reviewer also suggests that Table 1 could be split into two tables (differentiating variables passed from GRASSMIND to BODIUM and vice versa) and recommends clarifying why "rain seeping into soil" is marked as "additional", implying BODIUM typically processes precipitation directly.
In addition, this reviewer notes that: (i) modifications to the shading scheme (Appendix B2, p. 31–33) might influence seedling competition dynamics, as switching from own leaf area index to overall leaf area index for shading calculations could exacerbate the poor early growth after reseeding; (ii) a spin-up phase with diverse initial soil profiles could better encompass observed plot-to-plot heterogeneity (p. 26), as the authors acknowledge that initial soil conditions vary across parcels but are not modelled; and (iii) the significance attributed to FINAM appears disproportionate to its actual contribution, as the authors state that FINAM "offered no additional value" and "complicated the workflow" (p. 27), yet it appears prominently in the abstract.

Citation: https://doi.org/10.5194/egusphere-2026-1713-RC1
- AC1: 'Reply on RC1', Matthes Kantzenbach, 22 Jun 2026
  
  First, a major concern is the restricted accessibility of GRASSMIND v2.0. As noted in the manuscript (p. 28), the code is currently "only available on request," hindering transparency and reproducibility.
  → Response: We fully agree with your concern. Please note that the GRASSMIND model has been previously developed within the code structure of the FORMIND model which comes with its own publication policy and license. Therein, the model code of GRASSMIND v2.0 is freely available without restrictions, but requires a user registration before access (see also https://formind.org/downloads/download-formind-model/). We are aware that this adds an extra step before accessing the code and may potentially compromise transparency. Therefore, we are currently working with the FORMIND developers to find a solution for the revision of this manuscript that will make the GRASSMIND-specific parts of the code directly accessible without violating FORMIND’s publication guidelines.
  Second, the calibration process reveals parameter ambiguity. For example, the "overlap factor" (p. 10) consistently reaches its maximum limit of 1.0 across all simulations (Table 2), and the rooting parameters (r1–r2) exhibit considerable variability between calibration attempts (Tables 2–3).
  → Response: Thank you very much for your constructive comment.
  The main focus of our article is on the evaluation of the bidirectional dynamic coupling of two complex process-based models rather than on their calibration. While we could have also use virtually created data for the comparison of the standalone and coupled model versions, we decided here to use measured data from an experimental study site for a more applicable illustration. The two calibration attempts thereby were intended to just illustrate calibration ambiguity and to emphasize complexity in the calibration process of process-based models itself, which will be the focus of a manuscript currently in progress (as 2^nd chapter of a PhD thesis). Given your comment, we are now aware that this might have shifted the focus from our originally intended scope of the manuscript.
  Although we have already mentioned calibration ambiguity at a few sections in the manuscript (ll. 9, 527, 666), we aim now to revise the abstract and introduction to more clearly highlight the intended scope of our study and will elaborate on the topic more in the discussion section.
  Third, the post-2020 validation is inadequately analysed. Specific issues include a ~7–9 percentage point overestimation of deep-layer soil moisture (p. 17) and an unsatisfactory representation of early reseeding yields, particularly in the absence of sensitivity analyses (p. 16).
  → Response: Thank you for pointing this out. The overestimation of the deep-layer soil moisture is already discussed on pages 21-22 in lines 440-470 (section Discussion). The unsatisfactory representation of early yield after seeding (both for the initial seeding and for reseeding) is also discussed already on pages 24-25 in lines 551-579 (section Discussion). However, we agree that these aspects were not sufficiently highlighted in the context of model calibration and its post-2020 evaluation. We will therefore restructure these sections for more clarity and improve the discussion accordingly.
  Beyond these primary concerns, this reviewer identifies several interpretative ambiguities. These include circular reasoning in water-limitation claims, as the coupled model's water limitation is judged "more reliable" primarily because it inherits BODIUM's moisture calculations (p. 22).
  → Response: Thank you for pointing out that this claim was too strong. Our intention was to highlight the improved consistency of GRASSMIND-BODIUM. We will therefore revise the text accordingly.
  There is also no detailed breakdown of drought-year yield losses, meaning the individual contributions of water, nitrogen and temperature to the actual yield reduction (not just limitation factor) are never quantified (Figure 7).
  → Response: Thank you for this comment. We agree that a quantitative attribution of drought-year yield losses to water, nitrogen, and temperature limitations would provide valuable ecological insights; however, the scope of this study is the evaluation of the coupling of two complex process-based models instead of drought impact analysis. For this purpose, we consider the comparison of the limitation factors to be appropriate, as it allows differences in the underlying model processes to be identified. A detailed analysis of the individual contributions of water, nitrogen, and temperature limitation to yield reductions would be an interesting topic for future work but is beyond the scope of the present study. We understand that some formulation put too strong focus on drought, which was not our intention. We will revise the respective sentences in the introduction.
  Moreover, the significance of the self-coupling test is overstated: while technically valid, its scope is limited to verifying data-transfer mechanisms rather than confirming ecological equivalence (Appendix D).
  → Response: Thank you for this comment. It was not our intention to claim that the self-coupling test can confirm ecological equivalence. We will revise the text to clarify its purpose for solely verifying that the data transfer between the deactivated model component and the remaining component is correctly handled in the models.
  Several minor points also warrant attention. These include figure captions lacking full clarity (e.g. specifying that GRASSMIND executes first in each timestep),
  → Response: Thank you for pointing this out. We agree that the captions could be improved and will revise them accordingly.
  unweighted objective functions despite variations in measurement uncertainty (e.g. early-year biomass and sensor-based soil moisture with ±3% accuracy are given equal weight; p. 11),
  → Response: Thank you for this comment. Indeed, the objective function does not account for measurement uncertainty and was designed as a rather general metric for comparing the modeling approaches. We agree that the choice of objective function deserves further discussion and will add an additional paragraph to the discussion section (new subsection on calibration). Therein, we aim to explain the rationale behind the selected formulation and its limitations, including the lack of reliable uncertainty estimates for the calibration data (the ±3% accuracy refers to the measurements after 2020 that were not used for the calibration).
  an inconsistent treatment of the 2020 plowing/tillage event (GRASSMIND is restarted while BODIUM continues running and retains soil memory; p. 9, p. 24),
  → Response: Thank you for this valuable comment. The unequal treatment of the 2020 plowing event resulted from a technical limitation of GRASSMIND, which does not include a mechanism to remove all vegetation through a plowing event. Therefore, in the original simulations, GRASSMIND was restarted after plowing. In the coupled GRASSMIND-BODIUM setup, this procedure has little effect because only the plant component of GRASSMIND is active, while soil state variables and their memory are retained within BODIUM. However, for the standalone GRASSMIND simulations, we agree that restarting the model introduced an inconsistency that deserved closer attention.
  To address this issue, we reran the standalone GRASSMIND simulations using a continuously running model instance. Instead of restarting the model, we implemented the plowing event by killing all existing plants, removing germinating seedlings and subsequently applying the second seeding event (hard-coded specifically for these model runs). The revised simulations show noticeable quantitative differences in biomass dynamics following the second seeding (Supplement Fig. 1) while the simulated soil moisture dynamics and the agreement with observations changed only marginally (Supplement Fig. 2). Further, the overall behavior of the model and the qualitative comparison between the model setups remain unchanged. Consequently, the main conclusions of the study and the evaluation of model coupling performance are not affected by this revision. According to the changes in biomass dynamics we will update the corresponding results, figures and text in the manuscript.
  the ambiguous presentation of model‑only nitrogen outputs (Figure 6 shows D2–D3 without measurements, though the caption states only D1 was measured)
  → Response: Thank you for pointing this out. The intention of presenting the D2 and D3 layers, despite the lack of measurements, was to compare the simulated nitrogen dynamics among the different model setups throughout the soil profile. However, we agree that the original presentation could lead to ambiguity regarding which model outputs are supported by observations.
  To address this concern, we will revise Figure 6 and its figure caption to clearly indicate that mineral nitrogen measurements were only available for layer D1, whereas the results shown for layers D2 and D3 represent model outputs only.
  and inadequate discussion of plot‑to‑plot variability (the measured variance across the five GCEF parcels is never compared to the model's narrow ensemble envelopes; p. 26).
  → Response: Thank you for this comment. We already touched upon this issue in the paragraph on variability on page 26 in lines 601–633. However, we agree that it was not discussed in sufficient detail. We will therefore further improve the discussion and include quantitative information on the variance of both the measurements and the model results.
  This reviewer also suggests that Table 1 could be split into two tables (differentiating variables passed from GRASSMIND to BODIUM and vice versa)
  → Response: Thank you for this constructive suggestion. We will adopt it and split Table 1 into two separate tables to better distinguish the variables passed from GRASSMIND to BODIUM and vice versa.
  and recommends clarifying why "rain seeping into soil" is marked as "additional", implying BODIUM typically processes precipitation directly.
  → Response: Thank you for pointing that out. We will add a clarification that the interface replaces BODIUMs direct input of precipitation data, as BODIUM does not account for interception by plants but GRASSMIND does.
  In addition, this reviewer notes that: (i) modifications to the shading scheme (Appendix B2, p. 31–33) might influence seedling competition dynamics, as switching from own leaf area index to overall leaf area index for shading calculations could exacerbate the poor early growth after reseeding;
  → Response: Thank you for this insightful suggestion. To evaluate whether the modification of the shading scheme influenced the simulated establishment dynamics after reseeding, we performed additional simulations with the parametrizations of the different calibration setups using the previous shading formulation. Figure 3 in the supplement show exemplary the comparison between the biomass results from the revised shading formulation and the old shading formulation for GRASSMIND (setup 1a) and GRASSMIND-BODIUM (setup 2a).
  The comparison indicates that the overall dynamics and the poor early growth following reseeding are very similar for both shading formulations (as well for GRASSMIND as for GRASSMIND-BODIUM). While minor quantitative differences occur during some periods, the general establishment patterns and biomass trajectories remain largely unchanged. This suggests that the limited early growth after reseeding is not primarily caused by the revised shading approach, but rather arises from other processes represented in the model.
  We will add this comparison to the appendix to clarify the effect of the shading formulation on model behavior.
  (ii) a spin-up phase with diverse initial soil profiles could better encompass observed plot-to-plotheterogeneity (p. 26), as the authors acknowledge that initial soil conditions vary across parcels but are not modelled;
  → Response: Thank you, this sounds like an interesting approach. The measurements revealing heterogeneous initial soil conditions encompass heterogeneities in horizon depth and bulk density (Schädler et al., 2019). However, the measurements do not include hydraulic soil properties like permanent wilting point, field capacity (or other parameters on soil water retention), which are required for the model parameterization.
  While it would be possible to incorporate some aspects of the observed heterogeneity based on the available information, this would raise further methodological questions (how to combine plot specific and site specific measurements) and would require additional discussion (e.g. treatment of within-plot heterogeneity). Addressing these issues would constitute a substantial extension of the present study and is therefore beyond its scope.
  However, we agree that this was not sufficiently explained in the manuscript. We will therefore add a short explanation describing why soil heterogeneities were not included in the simulations.
  and (iii) the significance attributed to FINAM appears disproportionate to its actual contribution, as the authors state that FINAM "offered no additional value" and "complicated the workflow" (p. 27), yet it appears prominently in the abstract.
  → Response: Thank you for pointing this out. Our original idea was to set up the complete workflow using FINAM being beneficial for simulating larger areas with heterogeneous environmental conditions. However, in this study we focused only on one small parcel with homogeneous environmental conditions – an application where FINAMs benefits do not come to light. We therefore will set the expectations for the readers correctly by not mentioning FINAM in the abstract anymore, shortening the corresponding paragraphs in the manuscript and by shifting FINAM more to the discussion as an outlook for coupled model system applications in regional- to large-scale analysis.
  
  Citation: https://doi.org/10.5194/egusphere-2026-1713-AC1
RC2:
'Comment on egusphere-2026-1713', Anonymous Referee #2, 28 May 2026

The authors try to replace GRASSMIND's simplified hydrology with BODIUM's detailed soil-water physics while keeping both models independently maintainable by coupling them using the FINAM framework. They introduce a "self-coupling test" to identify and verify interface variables, and they evaluate the coupled system against multi-year biomass, yield, soil-moisture and mineral-nitrogen data. The coupled model reproduces soil moisture far better than GRASSMIND alone and gives a more credible balance of water- and nitrogen-limitation on growth. The coupling is carefully executed, the self-coupling test is a genuinely transferable idea, and the topic is core GMD territory. That said, several issues need to be resolved before the quantitative conclusions can be trusted and my substantive concerns are below, followed by minor points:

Major:

The Code and Data Availability section states that GRASSMIND v2.0 is "only available on request," and the v3.0 fork is described as not yet tested or documented. GMD's policy does not permit this as the model code described in the paper must be deposited in a persistent public archive under an open-source-compatible license at the time of submission, and "on request" plus a registration-gated, non-redistributable license does not meet that bar. The input data are archived exemplarily, and BODIUM and the coupling scripts are available, so the gap is concentrated on the one model that matters most.
The weather preparation uses a Thornthwaite PET, which is well documented to underestimate evapotranspiration in temperate lowlands relative to Penman-Monteith. The discussion's recurring puzzle is that the deeper soil layers stay too wet and that soil evaporation appears too low, which is exactly the symptom a low-biased PET would produce. Before attributing this to BODIUM's evaporation barrier being "too strong," the authors should rule out the forcing. Radiation and temperature are already available, so a Penman-Monteith comparison run is feasible and would clarify whether the residual is a model problem or an input problem.
The 2013 to 2020 measurements come from a Kern DBS60-3, which reports gravimetric water content (the figure axis confirms "% per fresh weight"), while the 2020 to 2025 SMT100 data and the model output are volumetric ("% per fresh volume"). No bulk-density conversion is described, yet soil moisture enters the calibration objective and underpins the main claim that the coupled model matches observations. Need further explanation.
In Tables 2 and 3 the overlap factor pins to its upper bound in essentially every run, the rooting parameters and the height-width ratio vary substantially between the two repetitions of the same setup, and several other parameters sit on their bounds. The authors note the rooting parameters are "ambiguous," but running each setup only twice samples the response surface too sparsely to characterize that. A global sensitivity screen to identify the influential parameters, followed by a Bayesian or ensemble calibration returning posterior distributions, would either resolve the ambiguity or document it properly. It would also give the ensemble spread in Figures 3 to 9 a real basis rather than one driven solely by mortality stochasticity.
The objective function is under-specified, and equal weighting is questionable for me. The cost is "the sum of the absolute differences divided by the mean of the measurement values," but with biomass, yield and soil moisture combined, and very different observation counts and measurement uncertainties per type (the sensors carry a stated ±3% accuracy), the implicit weighting can dominate the optimum. Please further explain.
Minor:
The sentence "it is increasingly important to include all relevant components and their interactions in detail in the models" overstates the case and contradicts the paper's own sensible simplifications (a single grass functional type, deactivated components, FINAM judged unnecessary). No model includes everything, as "All models are wrong, but some are useful". I would recommend rewording to emphasize selective process detail where it matters for the question.
The text calls setup 1a "calibrated with soil moisture" and 1b "calibrated without," which reverses the definitions in Section 3.4.
All hydraulic and texture values derive from a single 2005 survey, while the run extends to 2025. This sits awkwardly against the discussion of drought-driven changes in porosity and conductivity.

Citation: https://doi.org/10.5194/egusphere-2026-1713-RC2
- AC2: 'Reply on RC2', Matthes Kantzenbach, 09 Jul 2026
  
  Major:
  
  The Code and Data Availability section states that GRASSMIND v2.0 is "only available on request," and the v3.0 fork is described as not yet tested or documented. GMD's policy does not permit this as the model code described in the paper must be deposited in a persistent public archive under an open-source-compatible license at the time of submission, and "on request" plus a registration-gated, non-redistributable license does not meet that bar. The input data are archived exemplarily, and BODIUM and the coupling scripts are available, so the gap is concentrated on the one model that matters most.
  We fully agree with your concern. Please note that the GRASSMIND model has been previously developed within the code structure of the FORMIND model which comes with its own publication policy and license. Therein, the model code of GRASSMIND v2.0 is freely available without restrictions, but requires a user registration before access (see also ). We are aware that this adds an extra step before accessing the code and may potentially compromise transparency. Therefore, we are currently working with the FORMIND developers to find a solution for the revision of this manuscript that will make the GRASSMIND-specific parts of the code directly accessible without violating FORMIND’s publication guidelines.
  The weather preparation uses a Thornthwaite PET, which is well documented to underestimate evapotranspiration in temperate lowlands relative to Penman-Monteith. The discussion's recurring puzzle is that the deeper soil layers stay too wet and that soil evaporation appears too low, which is exactly the symptom a low-biased PET would produce. Before attributing this to BODIUM's evaporation barrier being "too strong," the authors should rule out the forcing. Radiation and temperature are already available, so a Penman-Monteith comparison run is feasible and would clarify whether the residual is a model problem or an input problem.
  This is a valid point and good suggestion to explore in detail. We agree with the well-documented underestimation of Thornthwaite’s approach; however, we would like to highlight that we here used an adapted approach for daily estimates (Pereira and Pruitt 2004), which addresses this exact issue. This approach follows a photoperiodic effective daily temperature and is equivalent to PET estimates using the often-recommended Penman–Monteith approach (i.e. see Fig. 1 and 2 in Pereira and Pruitt 2004). Given that both approaches estimate equally well PET, our final choice for the adapted Thornthwaite’s approach traces back to the minimum required number of variables (i.e. air temperature and day length) compared to Penman-Monteith.
  To clarify this choice and processing step, we will include more information and references on the chosen PET estimation approach to the methods section and to the weather generation scripts already uploaded on Zenodo (), including the comparison plots of both PET approaches for the respective study site below (Figure 1, Figure 2).
  Figure 3 shows some differences in the simulated growth dynamics between simulations using Thornthwaite’s PET estimates and Penman–Monteith PET estimates (e.g. stronger growth at the beginnig of the year with Penman–Monteith PET). However, the overall pattern is similar. There are also some differences in soil moisture between the approaches (Figure 4); however, these differences are negligible when evaluating the simulation results against the GCEF sensor data. With this background information, we will extend the discussion to inform readers on the sensitivity of the chosen PET approach to simulated soil moisture and add the above figures to the appendix.
  Reference: Pereira, A. R., & Pruitt, W. O. (2004). Adaptation of the Thornthwaite scheme for estimating daily reference evapotranspiration. Agricultural water management, 66(3), 251-257.
  The 2013 to 2020 measurements come from a Kern DBS60-3, which reports gravimetric water content (the figure axis confirms "% per fresh weight"), while the 2020 to 2025 SMT100 data and the model output are volumetric ("% per fresh volume"). No bulk-density conversion is described, yet soil moisture enters the calibration objective and underpins the main claim that the coupled model matches observations. Need further explanation.
  → Response: Thank you for pointing this out. We agree that the description of the soil-moisture measurements and the corresponding unit conversions was not sufficiently detailed and should be clarified in the manuscript.
  The measurements from 2013 to 2020 are reported as gravimetric soil water content, whereas the SMT100 measurements (2020–2025) are expressed as volumetric soil water content. In the GRASSMIND-BODIUM simulations, BODIUM explicitly tracks both soil volume and soil mass, allowing both gravimetric and volumetric soil moisture to be derived as model output. In the standalone GRASSMIND simulations, the soil is static and soil mass is not represented; hence, only volumetric soil water content is provided. For comparison with the measurements from 2013 to 2020, a conversion to volumetric soil water content was performed using bulk density values reported by Altermann et al. (2005). We will add a more detailed description of these procedures to the manuscript (i.e. methods section and figure caption).
  Reference: Altermann, M., Rinklebe, J., Merbach, I., Körschens, M., Langer, U. and Hofmann, B. (2005), Chernozem—Soil of the Year 2005. Z. Pflanzenernähr. Bodenk., 168: 725-740.
  In Tables 2 and 3 the overlap factor pins to its upper bound in essentially every run, the rooting parameters and the height-width ratio vary substantially between the two repetitions of the same setup, and several other parameters sit on their bounds. The authors note the rooting parameters are "ambiguous," but running each setup only twice samples the response surface too sparsely to characterize that. A global sensitivity screen to identify the influential parameters, followed by a Bayesian or ensemble calibration returning posterior distributions, would either resolve the ambiguity or document it properly. It would also give the ensemble spread in Figures 3 to 9 a real basis rather than one driven solely by mortality stochasticity.
  → Response: Thank you for this valuable suggestion. We agree that a comprehensive assessment of parameter identifiability through global sensitivity analysis and Bayesian or ensemble-based calibration would provide a more rigorous characterization of parameter uncertainty.
  However, the primary objective of the present study is to present and evaluate the opportunities, limitations and challenges of the coupling approach while your suggestion would constitute a valuable but also substantial extension beyond this scope (further, a comprehensive, ind-depth uncertainty analysis of (coupled) model system parametrization will be the focus of another manuscript currently in progress, i.e. 2^nd chapter of a PhD thesis).
  The prominent presentation of the calibration results in Tables 2 and 3 was intended to transparently document model limitations and to avoid overstating the certainty of the calibrated parameter values. Given your comment (and a similar one of the first reviewer), we are now aware that this might have shifted the focus from our originally intended scope of the manuscript.
  We will add a paragraph on the parameter ambiguity to the discussion, and one sentence on the scope of the study to the abstract. We will also revise the wording in the Methods section to clarify the purpose of the repeated calibrations and move the table containing the calibration parameters from the repeated calibration run (former Table 3) to the Appendix. This change emphasizes that these parameter values are provided as supporting information and sets the primary focus of the study again to the evaluation of the coupling approach.
  The objective function is under-specified, and equal weighting is questionable for me. The cost is "the sum of the absolute differences divided by the mean of the measurement values," but with biomass, yield and soil moisture combined, and very different observation counts and measurement uncertainties per type (the sensors carry a stated ±3% accuracy), the implicit weighting can dominate the optimum. Please further explain.
  → Response: Thank you for pointing this out. Indeed, the formulation is wrong: the used cost is actually "the sum of the absolute differences divided by the sum of the mean measurement values". So, the observation count does not influence the calibration. We will correct the sentence.
  However, we agree that the objective function does not account for measurement uncertainty and the choice of objective function deserves further discussion. Given also the first reviewer’s comment on our calibration procedure, we have decided to add an explicit paragraph on model calibration (bundling now calibration-related aspects that were previously distributed across different paragraphs). We will include a paragraph explaining the rationale behind the selected formulation and its limitations, including the lack of reliable uncertainty estimates for the calibration data (the ±3% accuracy refers to the measurements after 2020 that were not used for the calibration).
  Minor:
  The sentence "it is increasingly important to include all relevant components and their interactions in detail in the models" overstates the case and contradicts the paper's own sensible simplifications (a single grass functional type, deactivated components, FINAM judged unnecessary). No model includes everything, as "All models are wrong, but some are useful". I would recommend rewording to emphasize selective process detail where it matters for the question.
  → Response: Thank you for this comment. We agree that the original formulation was too general and could be interpreted as advocating an unrealistic level of model complexity. This was not our intention. Our aim was rather to emphasize that relevant model components and their interactions should be assessed to avoid missing key feedbacks that may be important for the questions addressed. We will therefore change the sentence in the abstract as well as similar formulations in the introduction.
  The text calls setup 1a "calibrated with soil moisture" and 1b "calibrated without," which reverses the definitions in Section 3.4.
  → Response: Thank you for pointing this out. We accidentally reversed “a” and “b” in that sentence and will correct it, ensuring that the definitions in Section 3.4 are used consistently throughout the text.
  All hydraulic and texture values derive from a single 2005 survey, while the run extends to 2025. This sits awkwardly against the discussion of drought-driven changes in porosity and conductivity.
  → Response: Thank you for your comment. Currently, the hydraulic characteristics derived from the 2005 survey are used to initialize the soil profile and parameterize the model. However, the simulations themselves begin in 2013 and the soil is not treated as static thereafter.
  Within BODIUM, soil structure evolves dynamically in response to management and soil processes. In particular, the model simulates changes in the volume fractions of different pore-size classes, as well as the consolidation of soil structure following tillage operations.
  We recognized that the discussion of drought-related changes in porosity and conductivity should refer more to processes represented by the model than to the model parametrization. We will therefore change the corresponding paragraph in the discussion.
  
  Citation: https://doi.org/10.5194/egusphere-2026-1713-AC2
RC3:
'Comment on egusphere-2026-1713', Anonymous Referee #3, 31 May 2026
This manuscript presents the coupling of the grassland model GRASSMIND and the soil system model BODIUM through the FINAM framework, aiming to combine the strengths of both modelling approaches while maintaining their independent development. The study addresses a relevant challenge in grassland modelling, namely the representation of interactions between vegetation dynamics and soil water processes under changing environmental conditions. The coupling strategy is well motivated, and the concept of integrating the species- and process-oriented representation of GRASSMIND with the more detailed soil hydrology of BODIUM is scientifically sound. The manuscript demonstrates the potential of this approach to improve the representation of soil moisture dynamics and to provide additional insights into plant-soil interactions. My main comments are provided below, followed by a number of minor remarks.
Minor comments:
The notation of nitrogen species should be checked throughout the manuscript and Table 1 to ensure correct chemical formatting (e.g., NH₄⁺ and NO₃⁻ subscripts/superscripts are not consistently displayed).

The study objective is introduced relatively late in the introduction (line 58), which weakens the narrative flow. As aspects of the model coupling are already introduced earlier (lines 29-30), the later statement appears somewhat repetitive. We recommend presenting the research gap and study objective earlier in the introduction and then introducing the modelling approach. A clearer progression from gap to aim to methodology would improve readability and reduce redundancy.

In Figure 6, the caption states that mineral nitrogen was only measured in layer D1; however, model results are also shown for layers D2-D3. This may lead to ambiguity regarding which outputs are supported by observations and which are model-only results. We recommend clarifying this distinction in either the figure caption or the figure itself.

For Figure 7, the different model setups are difficult to distinguish visually. Consider using a more contrasting colour palette and/or different line styles to improve readability.

The statement "The case of BODIUM and GRASSMIND is somehow unusual..." (line 65) would benefit from further clarification. As many process-based ecosystem and agroecosystem models explicitly represent plant and soil processes while using weather and management as external inputs, it is not immediately clear what aspect of the BODIUM-GRASSMIND combination is considered unusual. We recommend clarifying the specific novelty or distinguishing characteristic being referred to.

Major:
1. The coupling strategy introduces a temporal lag in the exchange of soil water and nitrogen information between BODIUM and GRASSMIND. As described in lines 197-202 and illustrated in Figure 2, GRASSMIND uses soil water potential and nitrogen concentrations originating from the previous coupling step, while updated soil states become available only after BODIUM has completed its calculations. This effectively alters the original process sequence of GRASSMIND. While such sequential coupling approaches are common, additional discussion would be helpful regarding the implications of this one-step delay. In particular, could the authors elaborate on the expected effects on plant water and nitrogen uptake during periods of rapid environmental change (e.g., drought onset, rewetting events, or fertilization)? Was the potential influence of the lag assessed or considered negligible relative to the daily time step?
2. The weather input data originate from multiple sources and processing steps, including site measurements, gap-filled observations, Copernicus-derived radiation data, and derived estimates of potential evapotranspiration. Additional information on the implications of these preprocessing choices would be helpful. In particular, temperature gaps exceeding 30 days were filled using linear interpolation. Given that the study focuses on drought-related processes, it would be useful to discuss the potential impact of this approach on the representation of climatic variability and extreme events. Furthermore, a brief justification of the selected radiation data source and the overall consistency of the assembled weather dataset would strengthen the methodological transparency of the study.

3. Tables 2 and 3 suggest potential parameter identifiability issues. Several parameters are estimated at or near their calibration bounds (e.g., overlap factor), while others, particularly the rooting parameters and height-width ratio, vary substantially between repeated calibrations of the same setup. We recommend discussing the implications of this parameter ambiguity and its potential impact on the robustness and interpretation of the model results.
4. While the study demonstrates improved simulation of soil moisture through the coupling with BODIUM, the analysis of plant water availability appears to rely primarily on soil moisture dynamics within the simulated profile. Since rooting depth and drought response are central themes of the manuscript, additional information on groundwater conditions and potential groundwater contributions to plant water uptake would be valuable. It remains unclear whether groundwater interactions, capillary rise, or fluctuating water tables are represented in the modelling framework or are assumed to be negligible. This aspect should be clarified, as it may influence the interpretation of both rooting behaviour and drought-related water limitation.
Citation: https://doi.org/10.5194/egusphere-2026-1713-RC3
- AC3: 'Reply on RC3', Matthes Kantzenbach, 09 Jul 2026
  
  Major:
  1. The coupling strategy introduces a temporal lag in the exchange of soil water and nitrogen information between BODIUM and GRASSMIND. As described in lines 197-202 and illustrated in Figure 2, GRASSMIND uses soil water potential and nitrogen concentrations originating from the previous coupling step, while updated soil states become available only after BODIUM has completed its calculations. This effectively alters the original process sequence of GRASSMIND. While such sequential coupling approaches are common, additional discussion would be helpful regarding the implications of this one-step delay. In particular, could the authors elaborate on the expected effects on plant water and nitrogen uptake during periods of rapid environmental change (e.g., drought onset, rewetting events, or fertilization)? Was the potential influence of the lag assessed or considered negligible relative to the daily time step?
  → Response: Thank you for this valuable comment. Indeed, the coupling approach effectively alters the original process sequence of GRASSMIND and we agree that the potential implications of this lag deserve further consideration.
  To assess its influence, we performed an additional simulation in which the delay was increased by one further coupling step (i.e., GRASSMIND receives the soil state from BODIUM at the end of two days ago instead of the last timestep/day). We then compared simulation results on aboveground biomass with those of the standard coupling approach. While there are some distinct differences when comparing the simulations of a single run due to stochasticity (Figure 1), these differences became negligible in 100 runs (each with different stochastic seed; Figure 2). The same holds true for the simulation results on yield, soil moisture, and soil nitrogen (Figures 3–5). This indicates that the effects of such short delays on the simulation results are not systematic and the influence of short lags (i.e. the altered process sequence in GRASSMIND) can be considered negligible.
  We will add this evaluation to the appendix.
  2. The weather input data originate from multiple sources and processing steps, including site measurements, gap-filled observations, Copernicus-derived radiation data, and derived estimates of potential evapotranspiration. Additional information on the implications of these preprocessing choices would be helpful. In particular, temperature gaps exceeding 30 days were filled using linear interpolation. Given that the study focuses on drought-related processes, it would be useful to discuss the potential impact of this approach on the representation of climatic variability and extreme events. Furthermore, a brief justification of the selected radiation data source and the overall consistency of the assembled weather dataset would strengthen the methodological transparency of the study. → Response: Thank you for this constructive comment. We agree that more clarification is required for our study on the chosen weather variables and data sources, its processing and implications for the results.
  The five replicate plots (or parcels) of the GCEF measurements used for calibration include daily weather measurements only for air temperature and precipitation; however, radiation and potential evapotranspiration were not available.
  Further, sensor measurements of temperature and precipitation started in 2015 while the experiment already runs since 2013. We therefore decided to complement the missing data in 2013 and 2014 with measurements of a nearby weather station from the German Meteorological Service (Deutscher Wetterdienst, DWD). The station was directly located at the same site just a few meters from the experiment and was maintained from 1997 until 2018. We used overlapping years between both data sources (i.e. 2015 – 2018) to identify a correction factor for adjusting air temperature data in 2013 and 2014 (to be consistent with data measured at the plots from 2015 onwards). As spatial autocorrelation in precipitation data is comparably low, such a correction was not done for precipitation.
  For radiation, we decided to use data from the ERA5-Land hourly data (Copernicus Climate Data Store) for the entire simulation period (2013 – 2025), because the local weather station at GCEF closed in 2018 and further meteorological stations of the DWD in about 50 km distance revealed long data gaps or complete missing data. However, for overlapping years (2013-2018) we found very good consistency of the ERA5-Land radiation data with the local measurements.
  For estimating potential evapotranspiration, we used an approach of Thornthwaite that has been adapted for daily time steps and overcomes the well-documented underestimation in arid regions (already pointed out by reviewer 2; see response to the comment of reviewer 2). We will add more information on the chosen approach as well as comparison figures of PET estimates for both approaches and their impact on the results to the manuscript appendix and the weather generation scripts already uploaded at Zenodo (https://zenodo.org/records/19129783).
  With respect to gap-filling (only required for plot-specific temperature and precipitation data in 2015-2025), we derived the window length of the moving median approach for each time-series of precipitation and air temperature of each of the five parcels separately by stepwise increasing window length until no further gaps have been detected. This was successful for precipitation data, but for air temperature data gaps still existed after our preset maximum of 30 days (remaining data gaps have been filled by linear interpolation). We appreciate your suggestion and tested the impact of various modifications of our gap-filling approaches on the average weather input file (mean across the five plots/parcels; used for driving the model simulations) and on the resulting simulation output (e.g. aboveground biomass, yield and soil moisture).
  When either using moving median only (with plot-specific derived window lengths) or interpolation only (of different methods like ‘linear’ or ‘spline’), the highest deviations to the weather data of the current gap-filling approach occur for the case of spline interpolation only (see Figure 6). Combinations of different interpolation methods and moving median (window lengths of 7 or 30 days) reveal much less deviations than presented in Figure 6. We therefore decided to test the impact of such changes in gap-filling on our simulation results on the example of spline interpolation only. The simulation results on aboveground biomass and soil moisture (Figures 7-9) show only minor differences between the current gap-filling approach and the spine interpolation. We therefore conclude that the choice of the gap-filling approach does not have any impact on the evaluations in our study.
  To improve transparency and clarity for readers, we will add a table to the appendix which summarizes the chosen weather variables, their data sources and processing steps, as well as their impact on simulation results (i.e. for temperature and PET). As we believe that the background information described above will be of interest for a wider readership, we decided to add those details to the weather generation scripts already uploaded at Zenodo (https://zenodo.org/records/19129783) and will refer to them in the newly added table in the appendix and in the methods section.
  3. Tables 2 and 3 suggest potential parameter identifiability issues. Several parameters are estimated at or near their calibration bounds (e.g., overlap factor), while others, particularly the rooting parameters and height-width ratio, vary substantially between repeated calibrations of the same setup. We recommend discussing the implications of this parameter ambiguity and its potential impact on the robustness and interpretation of the model results.
  → Response: Thank you for this comment. We agree that the calibration results indicate parameter identifiability issues.
  As also noted by the other reviewers, the implications of this issue were not sufficiently discussed in the original manuscript. We will therefore expand the discussion section accordingly to explicitly address the fact that no distinct global minimum has been identified during calibration and thus, the resulting parametrization should not be interpreted as unique or universally applicable. At the same time, we emphasize that the resulting simulations show very similar patterns in yield, biomass, and soil moisture despite the differences in parameter values. Consequently, the parameter ambiguity does not affect the main objective of this study, which is the evaluation of the coupled modeling framework, rather than the derivation of ecological conclusions from a specific parametrization.
  Given this comment, we are now aware to more emphasize our intended scope of the manuscript. The two calibration attempts were intended to just illustrate calibration ambiguity and to emphasize complexity in the calibration process of process-based models itself, which will be the focus of a manuscript currently in progress (as 2^nd chapter of a PhD thesis). We will add a statement to the abstract, change the formulations in the methods section and move table 3 into the appendix to more clearly highlight the primary objective of this study, i.e. the evaluation of coupling two complex process-based models.
  4. While the study demonstrates improved simulation of soil moisture through the coupling with BODIUM, the analysis of plant water availability appears to rely primarily on soil moisture dynamics within the simulated profile. Since rooting depth and drought response are central themes of the manuscript, additional information on groundwater conditions and potential groundwater contributions to plant water uptake would be valuable. It remains unclear whether groundwater interactions, capillary rise, or fluctuating water tables are represented in the modelling framework or are assumed to be negligible. This aspect should be clarified, as it may influence the interpretation of both rooting behaviour and drought-related water limitation.
  → Response: Thank you for this comment. We agree that groundwater conditions and their potential influence on plant water uptake can be important for the interpretation of rooting behaviour and drought responses. However, according to site information for the GCEF in Bad Lauchstädt, the groundwater table is located approximately 32 m below the surface (Wu et al., 2025). This is far below both, the rooting depths and the simulated soil profile considered in this study, making direct groundwater contributions to plant water uptake highly unlikely. We will clarify this assumption in the manuscript and add a short paragraph on this to the discussion.
  Reference: Mengqi Wu, Christiane Roscher, Martin Schädler et al. Legacy effects of climate extremes on deep soil water storage and water use efficiency across different land-use systems, 21 July 2025, PREPRINT (Version 1) available at Research Square [https://doi.org/10.21203/rs.3.rs-7062058/v1]
  
  Minor comments:
  1. The notation of nitrogen species should be checked throughout the manuscript and Table 1 to ensure correct chemical formatting (e.g., NH₄⁺ and NO₃⁻ subscripts/superscripts are not consistently displayed).
  → Response: Thank you for pointing this out. We will carefully check the notation of nitrogen species throughout the manuscript and ensure that chemical formulas and charges are formatted correctly and consistently.
  2. The study objective is introduced relatively late in the introduction (line 58), which weakens the narrative flow. As aspects of the model coupling are already introduced earlier (lines 29-30), the later statement appears somewhat repetitive. We recommend presenting the research gap and study objective earlier in the introduction and then introducing the modelling approach. A clearer progression from gap to aim to methodology would improve readability and reduce redundancy.
  → Response: Thank you for this suggestion. We agree that the narrative flow of the introduction was not optimal and that the study objective was stated somewhat repetitively.
  We will revise the introduction and combine the statement of the study objective in line 58 with that in lines 29-30, such that the progression from research gap to the aim of the study and to the methodology becomes clearer.
  3. In Figure 6, the caption states that mineral nitrogen was only measured in layer D1; however, model results are also shown for layers D2-D3. This may lead to ambiguity regarding which outputs are supported by observations and which are model-only results. We recommend clarifying this distinction in either the figure caption or the figure itself.
  → Response: Thank you for this comment. We agree that the original presentation could lead to ambiguity regarding which mineral nitrogen results are supported by observations and which represent model outputs only (this issue was also raised by reviewer 1). We will update Figure 6 by clearly indicating that mineral nitrogen measurements were available only for layer D1 (0–15 cm), while the results shown for layers D2 (15–30 cm) and D3 (30–50 cm) represent model outputs. In addition, the figure caption will be revised accordingly to explicitly clarify this distinction. (Figure 10)
  4. For Figure 7, the different model setups are difficult to distinguish visually. Consider using a more contrasting colour palette and/or different line styles to improve readability.
  → Response: Thank you for this suggestion. We agree that the original version of Figure 7 did not clearly distinguish all model setups. To maintain consistency with the other figures in the manuscript and to preserve comparability across panels, we retained the original color scheme, but improved visual distinction by introducing different line styles for the model setups. In addition, we applied a 14-day moving average to the plotted time series as a visual aid to reduce short-term variability and improve interpretability of differences between setups (this is now stated in the figure caption). We believe these changes substantially improve the readability and separation of the individual model configurations. (Figure 11)
  5. The statement "The case of BODIUM and GRASSMIND is somehow unusual..." (line 65) would benefit from further clarification. As many process-based ecosystem and agroecosystem models explicitly represent plant and soil processes while using weather and management as external inputs, it is not immediately clear what aspect of the BODIUM-GRASSMIND combination is considered unusual. We recommend clarifying the specific novelty or distinguishing characteristic being referred to.
  → Response: Thank you for this comment. We agree that the original wording was not sufficiently clear.
  We will revise the sentence to better explain what is meant. In particular, both models already represent the same general system, i.e. an explicitly resolved plant and soil component driven by external meteorological and management inputs. The coupling is therefore not a typical case of linking two structurally different models, but rather an exchange of an existing internal model system component by an external model system component.
  
  Citation: https://doi.org/10.5194/egusphere-2026-1713-AC3

Matthes Kantzenbach, Sara König, Thomas Reitz, Martin Schädler, and Franziska Taubert

Viewed

Total article views: 323 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
259	41	23	323	34	31

HTML: 259
PDF: 41
XML: 23
Total: 323
BibTeX: 34
EndNote: 31

Views and downloads (calculated since 04 May 2026)

Month	HTML	PDF	XML	Total
May 2026	212	28	15	255
Jun 2026	21	7	4	32
Jul 2026	26	6	4	36

Cumulative views and downloads (calculated since 04 May 2026)

Month	HTML	PDF	XML	Total
May 2026	212	28	15	255
Jun 2026	21	7	4	32
Jul 2026	26	6	4	36

Viewed (geographical distribution)

Total article views: 307 (including HTML, PDF, and XML) Thereof 307 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 28 Jul 2026

Short summary

To address uncertainties in grassland modeling due to simplified soil processes, we couple the grassland model GRASSMIND with the soil model BODIUM and evaluate the coupled system using multi-year field data. The results show a more reliable representation of plant–soil interactions and reveal new opportunities to improve model processes. This demonstrates that linking specialized models can strengthen model development and improve their usefulness for studying environmental systems.


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