| 31 Aug 2026
Combining an individual-based dynamic vegetation model with a distributive hydrologic model to improve coupled water-carbon modelling
Abstract. Vegetation and hydrology are tightly coupled: forest structure regulates evapotranspiration and soil moisture, while water availability governs tree growth, mortality, and succession. Yet hydrological and dynamic vegetation models typically maintain a domain-specific focus, limiting their ability to represent the feedbacks between forest dynamics and catchment-scale water dynamics. We address this gap by coupling the distributed hydrological model mHM with the individual-based forest model FORMIND, and apply the coupled framework (FORMIND-mHM) to the Selke river catchment in Central Germany, a mixed deciduous catchment with approximately 40 % forest cover. We examine how explicit vegetation–water interactions affect hydrological flux partitioning, and whether catchment-scale discharge carries exploitable information about stand-scale gross primary production (GPP). The coupled framework achieved discharge performance comparable to stand-alone mHM while substantially altering ET partitioning, shifting the transpiration fraction (Et/ET) from 0.50 in stand-alone mHM to 0.73 in the coupled framework, closer to observed ranges for temperate forests. Furthermore, a seasonal GPP–discharge relationship emerged consistently across all gauges, demonstrating that river discharge carries information about forest productivity, an analytical pathway accessible only through coupled modelling. FORMIND-mHM is therefore particularly suited to applications where vegetation–water feedbacks are central, including long-term projections of forest dynamics under changing climatic conditions.
This manuscript presents a coupled vegetation–hydrology modelling framework that integrates the individual-based forest model FORMIND with the distributed hydrological model mHM through the FINAM coupling framework. The topic is relevant to Geoscientific Model Development, and the motivation for coupling stand-scale vegetation dynamics with catchment-scale hydrology is generally clear. However, in its current form, I have several concerns regarding the methodological justification, reproducibility, and robustness of the model evaluation.
In particular, the large spatial-scale mismatch between FORMIND and mHM, the representativeness of the 9 ha forest samples within 4 km × 4 km hydrological grid cells, and the parameter optimization/evaluation strategy require further clarification and, where feasible, additional sensitivity analysis. Overall, I recommend major revision. The Introduction would benefit from substantial reorganization and clearer positioning, while the Methods require additional details and stronger justification of several key modelling choices. Detailed comments are provided below.
Major comments:
Lines 40–50. The discussion of the limitations of conventional hydrological models is useful, but the current wording appears somewhat too general. Some relatively simple hydrological models may indeed rely on prescribed or static LAI and empirical PET formulations, and may not explicitly distinguish different components of evapotranspiration. However, these limitations do not apply to all hydrological models. Many more process-based or integrated hydrological models are able to represent dynamic vegetation properties, and some explicitly simulate evaporation and transpiration as separate processes. I therefore suggest avoiding an overly broad characterization of “conventional hydrological models.” Instead, the authors could distinguish between simpler conceptual or empirical hydrological models and more process-based or integrated models, and provide examples showing which models use static LAI or empirical PET formulations and which models explicitly represent dynamic vegetation and separate evaporation and transpiration processes. This additional review would help clarify the specific model limitations that motivate the proposed coupling framework and better position the methodological contribution of this study.
Lines 64–77. These two paragraphs contain a considerable amount of model-specific and setup-specific information that may be more appropriate for the Methods section. For example, details such as the 20 m × 20 m patch size of FORMIND, its previous applications, the typical grid resolution of mHM, and particularly the choice of a 4 km × 4 km resolution in this study are methodological details rather than part of the general motivation. In the Introduction, I suggest retaining only the information needed to explain why FORMIND and mHM are complementary and why coupling these two models is scientifically or methodologically useful. The detailed descriptions of the individual models and the specific model setup could then be moved to the Methods section. This would also provide more space in the Introduction for positioning the proposed framework relative to existing coupled vegetation–hydrology modelling approaches.
Lines 78–88. The listed questions/objectives are not fully parallel in either content or sentence structure. Question (1) concerns model coupling, while Questions (2)–(3) focus on model behavior and evaluation. Question (4) is then introduced separately as a further application of the coupled framework. The grammar is also inconsistent: “We therefore ask” is followed by “(1) how...”, but then switches to “(2) examine...” and “(3) analyse...”. I suggest revising this paragraph to distinguish the main research questions from the subsequent application and to use a more consistent parallel structure.
Section 2.2/2.3. The use of a 4 km × 4 km mHM grid appears relatively coarse compared with the much finer FORMIND representation and may smooth important spatial heterogeneity in topography, soils, and land cover. The authors should further justify this resolution choice and discuss the limitations associated with using such a coarse hydrological grid. If computationally feasible, a sensitivity test using one or more finer mHM resolutions would be valuable to assess whether the main conclusions are robust to spatial resolution.
Sections 2.3–2.4.1. The spatial upscaling of the FORMIND simulations would benefit from further justification. For each 4 km × 4 km mHM grid cell, the authors simulate a representative 300 m × 300 m forest area, and the initialization is taken from the nearby inventory with the highest forest fraction; within that inventory, the 300 m × 300 m tile with the highest tree density is selected. Please clarify why this sampling strategy is considered representative of the forested fraction of the full mHM grid cell, and whether selecting the highest-density tile could systematically influence simulated forest structure or transpiration.
Section 2.4.1. The 5% forest-fraction threshold used to determine whether a 4 km × 4 km cell is treated as a forest cell should be justified. Since this threshold controls where the FORMIND coupling is activated, it may influence the spatial extent of the coupled simulation and the resulting catchment-scale fluxes. The authors should explain the basis for selecting 5%, and, if feasible, provide a sensitivity analysis using alternative threshold values to assess the robustness of the results.
Section 2.4.4. The parameter optimization procedure is not described in sufficient detail for reproducibility. Since the optimized configurations are central to the comparisons in Section 3.3, please specify which mHM parameters were optimized, their parameter ranges, the optimization algorithm, and the objective function used. The calibration period is reported as 2005–2015, but the corresponding evaluation/validation setup should also be clarified.
Sections 3.1–3.2. The use of eddy covariance observations to evaluate ET and GPP at the stand scale is a strength of the study. However, at the catchment scale, AET and soil saturation are mainly compared between FORMIND-mHM and stand-alone mHM, without an independent observational benchmark. The authors may consider including remotely sensed ET and, if available, soil moisture/saturation products to provide an additional catchment-scale evaluation of these variables, rather than relying only on differences between the two model configurations.
Section 3.3. Since the manuscript explicitly includes a calibration step for the optimized configurations, I suggest also reporting model performance over an independent evaluation period. This would help assess the out-of-sample performance and transferability of both mHM and FORMIND-mHM, rather than relying mainly on performance over the calibration or full simulation period.
Section 4.1. The discussion of “cross-domain parameter inference” may be somewhat stronger than what is directly demonstrated in the present analysis. The results show sensitivity of discharge to WUE and a relationship between GPP and discharge, which clearly indicates potential for cross-domain parameter constraint. However, a formal parameter-inference or identifiability analysis is not performed. I suggest slightly moderating the terminology or clarifying this distinction.
Minor comments:
Lines 90–97. This paragraph mainly describes the study area, observational datasets, and model evaluation strategy, and may be more appropriate for the Methods section.
Section 2.4.2. The statement that soil parameters were “adjusted according to the local soil type” is too vague for reproducibility. I would suggest specifying which soil parameters were adjusted and how the corresponding values were determined in Supplementary Information.
Figure 3. The day-of-year color coding is somewhat difficult to interpret. The colorbars are narrow and the seasonal gradient is not very distinct in the scatter points. Please consider using a clearer seasonal colormap and/or a more prominent colorbar with more intuitive month labels.
Section 3.4. It would be helpful to briefly remind readers what the two tested parameters represent physically and how they enter the respective models. In particular, please clarify which processes and model responses are directly controlled by WUE in FORMIND and by the canopy interception constant in mHM. This would make the sensitivity results easier to interpret.
Section 4.2. Since computational cost is highlighted as an important limitation of the coupled framework, please consider providing quantitative runtime information for the main model configurations, together with the computing platform and resources used (e.g., HPC system and number of cores), preferably in the Supplementary Information.