How can building representation influence flood hazard and impact modelling: Insights from the 2021 Ahr Valley Flood

Khosh Bin Ghomash, Shahin; Devi Nallasamy, Nithila; Apel, Heiko

doi:10.5194/egusphere-2025-2304

Preprints

https://doi.org/10.5194/egusphere-2025-2304

Preprints

17 Jun 2025

| 17 Jun 2025

How can building representation influence flood hazard and impact modelling: Insights from the 2021 Ahr Valley Flood

Shahin Khosh Bin Ghomash, Nithila Devi Nallasamy, and Heiko Apel

Abstract. The increasing flood risk in urban areas, driven by rising urbanization and climate change, underscores the need for accurate representation of buildings and urban features in flood hydrodynamic models. This study investigates the impact of different building representation techniques on flood hydrodynamic and impact modeling, using the 2021 flood event in the Ahr Valley, Germany, as example. Three methods — Building Block (BB), Building Hole (BH), and Building Resistance (BR) —are applied across varying model resolutions to assess their influence on flood extent, water depths, and flow velocities.

Our findings reveal that building representation affects both simulated flood extent and flow dynamics. The Building Block and Building Hole approaches generally lead to larger flooded areas with deeper water and higher velocities, while increased resistance or omitting buildings results in smaller flood extents, shallower water, and slower flow. Additionally, we show a strong link between building representation and model resolution. Our findings show that at coarser resolutions, the choice of building representation is more critical, with larger differences in flood extent across setups. We show that while all methods produce acceptable flood extents, variations in water depths and velocities highlight the importance of choosing the right building representation for accurate flood simulations—particularly in dense urban areas where accurate flood impact assessments rely on realistic flow dynamics. Our results emphasize the importance of selecting appropriate building representation methods based on model resolution to enhance urban flood modeling and impact assessment accuracy, with a general recommendation to include buildings as physical obstacles (BH, BB) in hydraulic models.

Received: 16 May 2025 – Discussion started: 17 Jun 2025

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

Shahin Khosh Bin Ghomash, Nithila Devi Nallasamy, and Heiko Apel

Status: final response (author comments only)

RC1:
'Comment on egusphere-2025-2304', Anonymous Referee #1, 18 Jul 2025
The manuscript provides a comprehensive evaluation of different building representation strategies (Building Block – BB, Building Hole – BH, and Building Resistance – BR) in urban flood modeling, using the 2021 Ahr Valley flood as a case study. The use of multiple spatial resolutions and performance metrics is commendable and offers a rich dataset for comparative analysis. However, several points merit further attention or clarification:
Evaluation of Calibration Strategy: The paper emphasizes that no calibration was performed for the hydrodynamic model. While this allows for a cleaner intercomparison between building representation methods, the lack of calibration likely contributes significantly to RMSE and bias values, especially for water depths. It would be beneficial for the authors to elaborate more explicitly on how calibration might alter the relative performance of BR, BB, and BH approaches. The brief mention in Section 4.3 (lines 305–310) could be expanded or perhaps discussed earlier in the Methods section to better contextualize the results. Including hypothetical scenarios or past studies where calibration impacted outcomes could provide a more nuanced understanding.

Human Instability Indicator Use: The adoption of the Jonkman and Penning-Rowsell (2008) human instability threshold is justified but would benefit from additional detail. How sensitive are the results to this 1 m²/s threshold? Have alternative thresholds been tested in the literature or in this study to assess robustness? This is important since conclusions on public safety and hazard reduction hinge on this criterion. If possible, a brief sensitivity analysis or reference to similar studies using different thresholds could strengthen this section.

Computational Cost vs. Resolution Trade-offs: The manuscript rightly points out the steep increase in computational time for higher-resolution simulations. However, it would be helpful to contextualize this cost-benefit tradeoff more clearly. For example, do the improvements in velocity and depth prediction at 1–2 m justify the 10–50× increase in runtime compared to 5 m? A clearer cost-benefit comparison, possibly in tabular form, would enhance practical guidance for modelers. Including specific examples or case studies where high resolution was critical could also provide practical insights.

Comparison to Observations – Uncertainty in Depths: While the limitations of the water mark dataset are acknowledged (lines 285–290), these uncertainties are substantial and should be more fully integrated into the discussion of validation results. Could the authors provide uncertainty bounds or sensitivity ranges for RMSE/Bias considering the observational imprecision? Including a graphical representation of these uncertainties might also help in conveying this point more effectively.

Observed Flood Extent: The use of LfU flood extent data as a validation baseline is valuable. Still, it would be helpful to elaborate on whether the observed extent includes buildings or open surfaces only, and how this influences the comparison across scenarios (e.g., BR might underperform if buildings were ignored in the observed extent). Clarifying this aspect could help in understanding potential discrepancies between observed and simulated extents.

Generalizability: The conclusions appropriately highlight that the findings are specific to the Ahr Valley’s geomorphology and event type. However, since generalization is mentioned in the final paragraph (lines 350–355), it would be useful to elaborate briefly on where the authors believe the results might not generalize—e.g., in urban deltas or larger floodplains with more heterogeneous flow paths. Including a discussion on potential adaptations or adjustments needed for different settings could make this section more robust.
Citation: https://doi.org/10.5194/egusphere-2025-2304-RC1
- AC2:
  'Reply on RC1', Shahin Khosh Bin Ghomash, 07 Nov 2025
  Dear Reviewer, thank you for your thorough and constructive review of our manuscript. Below, we address each point in detail, indicating planned revisions to the manuscript where applicable.
  We agree that the lack of calibration, while intentional for isolating the effects of building representation methods under standardized conditions, warrants further discussion to contextualize its potential influence on relative performance. As noted in the manuscript (now expanded in the Methods section, Section 3.2.1), our approach aligns with similar comparative studies (e.g., Schubert and Sanders, 2012; Jiang et al., 2022; Iliadis et al., 2024), where consistent parameters allow for a "clean" intercomparison without confounding factors from scenario-specific tuning. To address your suggestion, we will elaborate earlier in the Methods section by adding a paragraph discussing how calibration could alter outcomes. For instance, calibration (e.g., via Manning's n adjustments) might reduce overall RMSE and bias across all methods but could disproportionately benefit BR approaches in urban areas by compensating for underrepresented flow resistance, potentially narrowing differences with BB and BH. We will reference studies like Caviedes-Voullième et al. (2020), where calibration improved depth predictions in urban models by 10–20%, and Alipour et al. (2022), which showed that parameter tuning can shift relative method performance in high-resolution setups. Hypothetically, in our case, calibrating for observed water marks might enhance BR by mimicking obstruction effects, but less so for BB/BH, which already capture physical barriers. This will be cross-referenced in Section 4.3 for consistency.
  
  Thank you for highlighting the need for more detail on the human instability threshold. We selected the 1 m²/s criterion from Jonkman and Penning-Rowsell (2008) due to its widespread use in flood hazard assessments for adults in flowing water, as it balances empirical data from lab experiments and field observations. To enhance robustness, we will add a brief sensitivity analysis in Section 4.2 (Flood Impact Assessment).
  
  We appreciate your point. This specific trade-off has already been addressed in detail in our recent publications on the same case study and modeling framework (Khosh Bin Ghomash et al., 2024; Khosh Bin Ghomash et al., 2025). In Khosh Bin Ghomash et al. (2024), we performed a comparative assessment of two 2D shallow-water solvers (RIM2D and SERGHEI) at resolutions from 1 to 10 m, showing a rapid escalation in computational cost at finer resolutions. Improvements in velocity and depth predictions were notable below 5 m (e.g., better-resolved urban topography reducing differences by 10–20%), justifying higher costs for detailed hazard mapping in complex valleys like the Ahr, but coarser resolutions sufficed for broader extent simulations. Similarly, Khosh Bin Ghomash et al. (2025) used Monte Carlo sensitivity analysis to evaluate resolutions (e.g., 5 m and 10 m), finding coarser DEMs adequate for flood extent and impact assessments (balancing efficiency with accuracy), while finer resolutions were critical for channel dynamics despite higher costs. To integrate this into the manuscript, we will add a brief subsection in Section 4.4 referencing these studies and summarizing key insights,
  
  Thanks for the comment. We would like to point out that the water mark dataset used in our study has been used in various previous studies of the 2021 Ahr Valley flood (e.g., Apel et al., 2022; Mohr et al., 2022), providing a suitable foundation for validation. While some uncertainty in these observations may exist, though the exact magnitude is unknown, this uncertainty is consistent across all building representation setups (BB, BH, BR) due to the uniform application of the same dataset. Thus, it does not disproportionately affect the relative performance comparison between methods. We will expand the discussion in Section 4.3 to reflect this, emphasizing the dataset's reliability as observed data and its equal impact across all setups, suggesting that any uncertainty does not undermine our conclusions on method differences.
  
  The LfU flood extent data, derived from a combination of satellite images, aerial photography, and ground surveys of the edge of the flood extent, does not include building footprints. In the flood extent comparisons, we have excluded the cells flagged as buildings from the analysis to ensure a consistent comparison across all setups. This exclusion means that the representation of buildings (whether as BB, BH, BR, or omitted) does not influence the comparison with observed extents. We will add details in Section 4.3 to clarify this.
  
  We agree that while our findings highlight the importance of building representation in steep, narrow valleys like the Ahr, generalizability should be discussed with caveats. In the conclusions, we will expand to note limitations: Results may not directly apply to flat urban deltas or large floodplains, where diffuse flow dominates and BR could suffice due to less channeling. In such settings, BB/BH might overemphasize obstacles, leading to computational inefficiency without proportional gains (e.g., as in Kim et al., 2015). Adaptations could include hybrid methods (e.g., porosity for coarse scales) or resolution-dependent choices. We recommend testing in diverse geomorphologies, referencing Dewals et al. (2021) for porosity in flat terrains.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2304-AC2
RC2:
'Comment on egusphere-2025-2304', Anonymous Referee #2, 26 Jul 2025

This study investigated the impacts of different building representation techniques and DEM resolutions on flood model outputs based on a case study of 2021 Ahr Valley flood in Germany. The performances of different model configurations were compared in terms of flood inundation extents, water depths, and flow velocities. Overall, the study is interesting, and the findings seem to be meaningful. However, I still have several comments and suggestions as follows.
1) More details regarding the hydrodynamic model, RIM2D, should be presented: What is the size of the structured grid, and would it affect the simulation accuracy and efficiency? What is the downstream boundary condition of the model? What is the time step used in the numerical simulation, and is it sensitive to the model stability?
2) Eqns. (1)-(3): Please explain more about the terms q_x and q_y. It seems that the dimensions of some terms in Eqn. (2) are not the same on both sides of the equation.
3) Figures 2, 7, and 8: To avoid confusion, please change the legend from “lines” to “points”.
4) Figures 3 and 4: It may be unfair to only employ “BR10x” instead of “BR2x” and other configurations within the category of “BR” for comparison purposes.
5) Figure 6: Why was “BB” not included in the instability comparison?
6) Line 290: The Bias may not be a good metric, because over-estimations and under-estimations would be canceled out. Also, given the sampling uncertainty and measurement errors in both temporal and spatial data, it is suggested to present the values of metrics through a statistical distribution instead of a fixed number. The authors can refer to the article below for more information about the limitations of some commonly used evaluation metrics in flood modeling.
Reference: “Beyond a fixed number: Investigating uncertainty in popular evaluation metrics of ensemble flood modeling using bootstrapping analysis” (https://doi.org/10.1111/jfr3.12982)
7) Figures 7 and 8: Why did the models with finer resolutions not necessarily yield a better performance in terms of the evaluation metrics? The results also indicated that “BR2x” was more consistent and robust with the change in DEM resolutions.
8) Any guidance or suggestions on how to make a choice between BB and BH methods?
Minor Issues
9) Line 5: Change “extent” to the plural form “extents”.
10) Line 117: It is suggested to provide link access to RIM2D if it is open source.
11) Line 255: Did the different building representation methods significantly affect the computational costs? Can the model RIM2D run using CPU? If yes, how about the CPU computational cost for this case?

Citation: https://doi.org/10.5194/egusphere-2025-2304-RC2
- AC1:
  'Reply on RC2', Shahin Khosh Bin Ghomash, 07 Nov 2025
  Thank you for reviewing our manuscript. We appreciate your comments and suggestions, which will help improve the clarity and depth of the paper. Below, we address each point, indicating planned revisions where applicable.
  We have now added more information on the domain sizes under different resolutions, the outflow boundary, and the timestep used for the simulations in Section 3.1.
  
  q_x and q_y represent the unit discharges (specific discharges) in the x and y directions, respectively, defined as q_x = h u and q_y = h v, where h is water depth and u, v are depth-averaged velocities. These are standard in the local inertia approximation (Bates et al., 2010; de Almeida and Bates, 2013). Regarding dimensions in Eqn. (2), all terms are consistent with units of acceleration times depth (m²/s²), as the friction term g n² q |q| / h^{7/3} balances the temporal derivative ∂q/∂t (m²/s²) when derived from the Manning formula. This is a common formulation in simplified shallow-water models, and no dimensional inconsistency exists. We will clarify this in Section 3.1 by adding definitions and a note on dimensional consistency.
  
  We will update the legends to "points" in the revised figures.
  
  We selected BR10x for these figures as it represents an extreme case of increased resistance (10 times baseline Manning's n), highlighting the maximum differences in flow dynamics and extents compared to BB, BH, and NoB. However, we agree that including other BR variants (e.g., BR2x, BR5x) would provide a more balanced view. In supplementary analyses, BR2x shows intermediate extents (~5–10% less than BH/BB), while BR5x is closer to BR10x. We will revise Figures 3 and 4 to include BR2x and BR5x overlays,
  
  BB was omitted from Figure 6 because its human instability patterns were nearly identical to BH, with very mild differences in unstable areas due to similar obstruction effects (infinite vs. finite height, but flood depths exceeded typical building heights in the event). Including BB would add redundancy without new insights. We will add a note in the Figure 6 caption clarifying this and provide BB results in the supplementary material for completeness.
  
  We acknowledge the limitations of Bias, as over- and under-estimations can cancel out, potentially masking errors (as discussed in Huang et al., 2024). To provide different perspectives, we report both RMSE and Bias. We will expand Section 4.3 (around lines 290–300) to discuss these limitations, reference Huang et al. (2024), and include a table with summary statistics of RMSE and Bias distributions.
  
  Finer resolutions do not always improve metrics because, without calibration, they amplify topographic details (e.g., micro-relief, building edges) that may introduce localized errors if not perfectly represented, leading to higher RMSE in uncalibrated setups (consistent with Caviedes-Voullième et al., 2020). BR2x appears more robust as mild resistance smooths flow, reducing sensitivity to resolution changes compared to BB/BH, which emphasize obstacles and vary more with grid refinement. We will add more explanation on this in Section 4.3, with cross-references to related studies.
  
  As shown in our results, BB and BH perform similarly for flood extents and depths in events like the 2021 Ahr flood, but BH is simpler and computationally cheaper (no need for building height data). We would recommend choosing BB when detailed height data is available and vertical effects matter (e.g., multi-story flooding); otherwise, BH suffices as a reflective boundary equivalent to infinite height. Both are recommended over BR for realistic dynamics in urban areas. We will add more on this in the conclusions.
  
  We will make this correction.
  
  The link is already provided in the Code Availability section at the end of the manuscript.
  
  We will clarify in Section 4.4 the differences in computational cost between building representation methods. RIM2D is GPU-optimized and cannot run on CPU.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2304-AC1
RC3:
'Comment on egusphere-2025-2304', Anonymous Referee #3, 30 Jul 2025

Overview
This manuscript presents a detailed investigation into how different building representation approaches (Building Block, Building Hole, Building Resistance) influence flood hazard and impact modelling, using the catastrophic 2021 Ahr Valley flood as a case study and the RIM2D model as a hydrodynamic simulator.

General Comment
The work is relevant and addresses a question of high practical importance for flood modellers and planners. The use of multiple model resolutions and a consistent comparison framework is commendable. The study is methodologically sound and features a well-structured comparison of seven building representation scenarios across four spatial resolutions. The use of consistent model setups allows for a clear isolation of the effects of different building representation techniques.
However, while I appreciate the effort made by the authors, I cannot recommend the paper for publication in its current form due to several major concerns outlined below:
1- Although the paper is methodologically rigorous and well-executed, its contribution to the field appears somewhat incremental. Building representation approaches and their effects on flood modelling have already been extensively explored in previous studies, many of which are either overlooked or only partially addressed in the current manuscript. The literature review needs to be significantly strengthened to adequately represent the state of the art in this specific area, clearly identify the knowledge gaps, and justify the novelty and motivation of the present work. As it stands, the manuscript does not convey which specific gap in the literature the study aims to address.
2- The contribution also appears incremental concerning the authors' prior works, which are cited in the manuscript. It seems that this paper merely expands on a specific aspect of their earlier studies. However, it is unclear whether this extension is sufficient to constitute a significant and novel contribution. Greater emphasis should be placed on clarifying the unique aspects and added value of this study compared to the authors' previous publications.
3- The comparison with observation is, in part, misleading. The results of the simulations are not only dependent on the different techniques used for modelling the buildings, but also on several sources of uncertainty that you have not quantified at all. Among these, the roughness values assumed in the domain (which in turn is also influenced by the grid resolution). This kind of comparison confused me a lot.
4- I appreciated the attempt to interpret the results in terms of flood impact, which is a key aspect from an emergency management perspective. The authors introduced a threshold (1 m²/s) to define the impact. While I acknowledge the practical motivation behind this choice, I believe it is an overly simplified approach that may obscure important differences among the simulation results. A possible direction to enhance the novelty of the work would be to evaluate all simulations using multiple impact criteria. I suggest referring to the study DOI: 10.1007/s11269-024-03988-5 as a practical example to support a discussion of the variability in your results, to the corresponding variability in flood impact assessment when applying different criteria. This would allow the authors to offer new insights, not only to their previous work, but also in comparison to similar studies in the literature. Please consider revisiting your work with the following guiding question in mind: What is the variability introduced by the building representation approaches compared to the uncertainty inherent in flood impact indicators?
5- The manuscript currently lacks a dedicated discussion section, as the discussion is merged with the presentation of results. I strongly recommend separating these two sections. A standalone discussion is essential to emphasise the novelty of the work in the context of existing literature. This section should clearly articulate what new insights the reader gains from this study that are not already available in previous works. Which aspects of the findings confirm or contradict earlier studies? How do the results expand the current understanding of building representation in urban flood modelling?
6- The conclusions are difficult to generalise, as they are drawn from a single case study with highly specific geomorphological and hydrodynamic characteristics that may not be representative of other urban flood contexts. Could you please discuss more on this?
7- The conclusions are tightly linked to the use of RIM2D; more sophisticated models based on fully dynamic shallow water equations could lead to different outcomes, as the velocity field around buildings may differ significantly, especially when using the BH approach. There are several papers in the literature which show comparisons among different complexity models in urban areas with different treatments of buildings. I recommend discussing the soundness of your conclusions in light of these works.
Overall, the paper has practical merit, but in its current form, it is better suited as a technical note rather than a full research article. I hope these comments will help the authors enhance the research novelty of the manuscript by incorporating additional analyses and broader considerations.

Citation: https://doi.org/10.5194/egusphere-2025-2304-RC3
- AC3:
  'Reply on RC3', Shahin Khosh Bin Ghomash, 08 Nov 2025
  Thank you for your review. Your comments have helped us identify areas for improvement, particularly in emphasizing novelty, expanding the literature review, and enhancing discussions on uncertainties and impacts. We will revise the manuscript accordingly to address these concerns. Below, we respond to each point:
  Although we have already mentioned works that have previously studied building representation in the manuscript, we will further expand Section 1 to include additional references and articulate the gap more clearly. The gap we address is the limited integration of building methods with multi-resolution analysis in extreme flash flood contexts like the Ahr event, where high velocities and complex topography amplify differences, unlike milder urban pluvial floods in most prior studies. Moreover, our uncalibrated, comparative approach with a physics-based model isolates building effects, providing practical guidance for operational models where data scarcity limits tuning. Our results also extends beyond general method comparisons by linking to impact assessments in a real catastrophic event.
  
  This study builds on but distinctly advances our previous publications. Khosh Bin Ghomash et al. (2025) focused on Monte Carlo sensitivity of roughness and resolution in RIM2D, without varying building representations. Khosh Bin Ghomash et al. (2024) compared RIM2D (local inertia) and SERGHEI (full SWE) solvers with a fixed building setup (BH), emphasizing computational feasibility for early warning. Here, we specifically dissect building methods (BB, BH, BR, NoB) across resolutions, quantifying their isolated effects on extents, depths, velocities, and impacts, novel aspects not covered before. The added value includes recommendations for method selection in steep valleys and links to human instability, absent in our prior sensitivity or solver-focused works. We will clarify this more in Section 1 and the conclusions.
  
  We agree that uncertainties can affect absolute performance, and our uncalibrated approach prioritizes relative comparisons. To address this, we will expand Section 4.3 to explicitly discuss these uncertainties, referencing our Monte Carlo study (Khosh Bin Ghomash et al., 2025), and to what extent roughness variations can alter simulated depths. Grid resolution interacts with roughness (finer grids reduce effective roughness via better topography), but our fixed Manning's n highlights building-driven variances. We'll note that while absolute metrics (e.g. RMSE) reflect uncalibrated biases, relative rankings (e.g. BB/BH) hold, as uncertainties apply uniformly.
  
  Thanks for the suggestion. We will enhance Section 4.2 using Bellos et al. (2024) and testing additional thresholds for our impact indicator.
  
  We will create a new Discussion (Section 5), moving interpretations and comparisons to other studies from the other parts of the manuscript.
  
  We will expand the conclusions to discuss the transferability of the results more, mentioning that the findings best apply to similar confined valleys (e.g., Alpine/ Appalachian flashes). And that in flat deltas, diffuse flows may favor BR or porosity methods.
  
  While RIM2D's local inertia suits subcritical flows, full SWE solvers may better capture supercritical dynamics around buildings, potentially altering BH velocities (e.g., more reflections). Our conclusions hold for similar simplified models but may differ in full SWE, as shown in our comparison (Khosh Bin Ghomash et al., 2024), where SERGHEI (full SWE) showed increased velocities by 10–20% in urban zones. We willl add more discussions on this matter in the manuscript.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2304-AC3
RC4:
'Comment on egusphere-2025-2304', Anonymous Referee #4, 30 Jul 2025

Thank you for submitting this article. The research is well described and clear. The methods are valid. A few points of enhancement can be performed, especially in the introduction and in the description of results/discussion. Please see my comments below:

More specifically, into 95 to 115, the local-inertial approximation neglecting convective acceleration substantially improves diffusive-wave or kinematic-wave models by considering local acceleration. However, it presumes sub-critical flows. Since the event simulated is unprecedented and the channels are steep, I wonder if the approximation is valid. Supporting information for this would be the computed Froude numbers from a full momentum model. Local-inertial models when faced super-critical flows, tend to produce mass balance errors. I encourage the authors to provide both: (1) mass balance computational errors and (2) a map of computed maximum Froude numbers in the domain. If possible, it will certainly enhance the confidence in the results. As shown by de Almeida (2012), the local-inertial model has a good performance under sub-critical flows but its performance for super-critical flows is still a topic of further research.

Paragraph 130: What specific modifications were made in the DEM for NoB?

In Fig A1, around 2000 min, a sudden fall of the stage-hydrograph is shown. Any reason for that? In addition, the authors mentioned that the reconstruction of inflow boundary condition was made upon the observed stage-hydrograph. I wonder how much volume it yielded and how much that would represent in mm from the upstream catchment, so a comparison can be made to validate the approach since rainfall most likely is available for the upstream catchment of this gauge. The authors already mentioned that calibrating roughness for all scenarios is not the purpose of the study. But simulating an event without checking the appropriateness of it nor estimating proper manning roughness coefficients for the scenarios might look like an extensive theoretical exercise rather than an application focused on explaining the observed reality.

Increasing resistance scenarios resulting in lower maximum depths seems a little counteractive to me, since larger roughness coefficients typically lead to higher flood depths, for a same flow discharge. I’d like the authors to include more physical explanations on why this opposing effect occurred for scenarios like BR10x. Also, please cite relevant literature to compare your results with previous results when discussing (220-225)

In line 230, please justify the reason or at least contextualize on why not to use a human instability risk metric based on human body physics flood momentum calculations. Recent examples of these approaches can be seemed in:

Gomes, Marcus N., Vijay Jalihal, Maria Castro, and Eduardo M. Mendiondo. "Exploring the impact of rainfall temporal distribution and critical durations on flood hazard modeling." Natural Hazards 121, no. 9 (2025): 10989-11012.

Postacchini, Matteo, Gabriele Bernardini, Marco D’Orazio, and Enrico Quagliarini. "Human stability during floods: Experimental tests on a physical model simulating human body." Safety science 137 (2021): 105153.

In Fig. 6 the authors mentioned a critical velocity of “1m2/s”?, which I believe it is 1 m/s. The authors should clearly provide a more extensive literature review on human instability models on their introduction.

Fig 1: please fix the name Atenahr to fit in the box, not crossing it. Also, add country boundaries in the left inset chart, not just state the coordinates. For the DEM image, is this a DEM or a DTM? I presume the white parts are human developments, but this is not clear from the figure. Please make the figure more descriptive. Finally, no need to use scientific notation for 5e1 to 5e2.

Citation: https://doi.org/10.5194/egusphere-2025-2304-RC4
- AC4:
  'Reply on RC4', Shahin Khosh Bin Ghomash, 09 Nov 2025
  Thank you for the review and your positive feedback on our manuscript. We appreciate your suggestions for enhancements. Below, we address each comment in detail.
  Thanks for the comment, and yes, the explicit numerical scheme from Bates et al. (2010) can become unstable in near- or super-critical flow regimes or with fine grid resolutions. However, RIM2D incorporates the stabilization method introduced by de Almeida et al. (2012), based on numerical diffusion. Moreover, an additional automatic stability control has been introduced to keep the model stable even if the numerical diffusion is not able to capture model instabilities, which can occur in case of supercritical flow conditions. For the Ahr event, we evaluated this in our prior comparison of RIM2D (local-inertial) and SERGHEI (full SWE) (Khosh Bin Ghomash et al., 2024), where maximum Froude numbers reached ~1.5–2 in narrow reaches, but mass balance errors in RIM2D remained low. We will add mass balance errors and some more discussion on this topic to the manuscript.
  
  For the NoB (No Buildings) scenario, the DEM was unmodified from the base 1 m resolution LiDAR data, meaning buildings were not explicitly removed or altered, flow simply propagates over the raw topography without building treatments.
  
  The sharp drop around 2000 minutes originates from the reconstructed water-level time series provided by the Flood Warning Center of Rhineland-Palatinate and no we don’t have a concrete reason why that is. We added a clarification in the manuscript to explicitly note that this feature is part of the official reconstruction and not an artifact of our modeling workflow. We will add some information on the inflow volume from the reconstructed hydrograph and compare it to catchment-scale rainfall. And regarding calibration, we agree that event-specific calibration is essential when the goal is to reproduce observed flood dynamics as accurately as possible. However, the focus here is on the relative influence of different building representation methods under controlled and identical model settings. Calibrating each scenario separately would confound the comparison by introducing scenario-specific parameter compensation effects. Moreover, in our previous study on this flood (Khosh Bin Ghomash et al., 2025, Monte Carlo-based sensitivity…), we have already carried out extensive calibration and sensitivity analyses for this case, demonstrating both the model’s capabilities and the dominant controls on performance. The main goal here is therefore not to re-calibrate the event, but to isolate structural differences between building-representation approaches. This approach using consistent roughness and uncalibrated setups to evaluate structural contrasts is standard in previous building-representation studies as mentioned in the manuscript.
  
  In our case, treating buildings as resistance rather than obstacles allows water to spread over a larger effective area (including cells occupied by buildings) leading to lower water levels and reduced local ponding. We will expand the physical explanation in Section 4.1, also citing further references.
  
  We chose the method by Jonkman and Penning-Rowsell (2008) for its empirical basis in lab/field data on adult stability in flowing water, widely used in flash flood assessments. While physics-based metrics may offer refined momentum calculations, they are beyond our scope, as we focus on building-driven flow differences rather than advanced biomechanics. We will justify this in Section 4.2, also referencing your suggested studies as alternatives for future work
  
  The threshold is correctly 1 m²/s (depth-velocity product, not velocity alone), as per Jonkman and Penning-Rowsell (2008). We will add a more extensive review of human instability models, including empirical to the manuscript.
  
  We will revise Fig. 1: adjust the "Altenahr" label to fit within the box, add country boundaries, clarify the white areas (representing building footprints that are cut out), and use standard notation instead of scientific. The figure caption will be updated for clarity and descriptiveness.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2304-AC4

Shahin Khosh Bin Ghomash, Nithila Devi Nallasamy, and Heiko Apel

Viewed

Total article views: 899 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
773	106	20	899	13	27

HTML: 773
PDF: 106
XML: 20
Total: 899
BibTeX: 13
EndNote: 27

Views and downloads (calculated since 17 Jun 2025)

Month	HTML	PDF	XML	Total
Jun 2025	108	24	5	137
Jul 2025	99	18	10	127
Aug 2025	96	22	1	119
Sep 2025	397	12	2	411
Oct 2025	39	19	1	59
Nov 2025	34	11	1	46

Cumulative views and downloads (calculated since 17 Jun 2025)

Month	HTML	PDF	XML	Total
Jun 2025	108	24	5	137
Jul 2025	99	18	10	127
Aug 2025	96	22	1	119
Sep 2025	397	12	2	411
Oct 2025	39	19	1	59
Nov 2025	34	11	1	46

Viewed (geographical distribution)

Total article views: 888 (including HTML, PDF, and XML) Thereof 888 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 15 Nov 2025

Short summary

This study explores how the way buildings are represented in flood models influences predictions of flood extent, water depth, flow speed, and overall impact. Using a major flood event in Germany as a case study, we evaluate different representation methods across various model resolutions. The results support more accurate flood modeling and impact assessments, helping cities better prepare for and respond to future floods.


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