the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 04 Dec 2024
Technical Note: Influence of building representation in flood hydrodynamic modelling: the case of the 2021 Ahr valley flood
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. 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 can affect both 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. At coarser resolutions, the Building Hole method produces the most accurate flood extents, while the increased resistance method performs better at finer scales. 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 impact calculations. These variations are vital for flood impact assessments, especially in dense urban areas. Our results emphasize the importance of selecting appropriate building representation methods based on model resolution to enhance urban flood modeling and impact assessment accuracy.
- Preprint
(34234 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 26 Feb 2025)
-
RC1: 'Comment on hess-2024-314', Anonymous Referee #1, 31 Jan 2025
reply
This technical note shows how different strategies to represent buildings in 2D flood modeling, performed with the model RIM2D, influence the numerical results. The 2021 flood in the Ahr valley is used as case study.
The paper is generally well written and structured.
While the topic of building representation in numerical flood modeling applications is certainly of interest to the HESS readers, I’m concerned by i) the way the paper presents the problem, ii) the lack of critical information and iii) I miss the key message.
First of all, no information is provided in the paper about the study area (extend, topography, river characteristics, normal flow conditions…), the percentage of the study area concerned by buildings or the flood characteristics. This means we have no idea of the global flood characteristics compared to river capacity, flood features (discharge? hydrogram?) and on the possible magnitude of building effects on the flood.
Then, key information is missing about numerical model features, which solves an approximation of shallow water equations: is it based on a structured or unstructured grid? How is the buildings footprint linked to the model meshing?
Finally, it is emphasized several times in the paper that “the study revealed a strong interaction between model resolution and building representation accuracy”. The Authors conclude that at coarser resolutions (e.g. dx = 10m), excluding buildings footprint from the computation domain (BH method) produces the “best” results, while at finer resolutions (e.g. dx = 1 or 2 m), replacing buildings by increased roughness area produces “better” results in terms of flood extend prediction but with a wrong representation of water depths. What is the meaning of such conclusions? In fact, it is simply shown that model calibration is required (calibration of the friction factors, such as explained on lines 103-105) and that this calibration depends on model resolution and buildings representation method (and also probably of the flow solver features). This is particularly visible on figure 1 where the “best” results are gained with the calibration modeling scenario (dx=5m, BH approach) but not with a finer mesh (dx=1 or 2m) while it is clear that a finer mesh enable a better representation of the topography and also probably of the building footprint closer to the reality. This simple but very important message is not clear over the paper.
As a basic rule in numerical modeling, the results cannot depend on grid resolution (grid convergence). If this is not the case, grid related approximations are hidden in the calibration parameters, i.e. usually friction coefficients in 2D flow models. In this paper, building representation technique is another source of approximation, whose influence combines with the grid resolution one, making general conclusions impossible.
In addition to these general remarks, the way the accuracy in water depths prediction is evaluated with equation on line 177 is wrong. The norm of simulated to observed values must be computed to avoid compensation errors between over and under estimations. This would certainly affect a large part of the subsequent analysis.
In conclusion, I cannot recommend this paper for publication.
Citation: https://doi.org/10.5194/hess-2024-314-RC1 -
AC1: 'Reply on RC1', Shahin Khosh Bin Ghomash, 13 Feb 2025
reply
Dear RC1,
We appreciate your detailed review and the opportunity to clarify and strengthen our manuscript. While we acknowledge the points you raised, below we systematically respond to each concern and outline the revisions we have made.
This technical note shows how different strategies to represent buildings in 2D flood modeling, performed with the model RIM2D, influence the numerical results. The 2021 flood in the Ahr valley is used as case study.
The paper is generally well written and structured.
While the topic of building representation in numerical flood modeling applications is certainly of interest to the HESS readers, I’m concerned by i) the way the paper presents the problem, ii) the lack of critical information and iii) I miss the key message.
First of all, no information is provided in the paper about the study area (extend, topography, river characteristics, normal flow conditions…), the percentage of the study area concerned by buildings or the flood characteristics. This means we have no idea of the global flood characteristics compared to river capacity, flood features (discharge? hydrogram?) and on the possible magnitude of building effects on the flood.
REPLY: We have already included details about the study area in lines 45 -51, about the basin size, stream characteristics, and vulnerability to floods. We have also described the 2021 flood event and its impact. In Lines 107 – 113, we have provided information related to the source of the discharge data used and how we have fed it as an upstream boundary to the model. As this is a technical note, we aimed to keep it concise and focused on the core methodological aspects. However, we recognize the value of additional context, so we have made the following improvements:
-Added a supplementary figure illustrating the study area and simulation extent.
-Included a new section describing the characteristics of the Ahr Valley, including topography, river features, and urbanization.
-Added the more details of the boundary forcing of the model and the hydrograph as a supplementary figure to provide more clarity on flood characteristics.These additions should provide the necessary context while maintaining the brevity appropriate for a technical note.
Then, key information is missing about numerical model features, which solves an approximation of shallow water equations: is it based on a structured or unstructured grid? How is the buildings footprint linked to the model meshing?
REPLY: Regarding this point, the entire Section 2.1 of the manuscript already provides information and multiple references to RIM2D and its underlying numerical framework, which the reviewer appears to have overlooked. Additionally, again, as this is a technical note, our goal was to keep the paper concise. Nevertheless, we have now incorporated the requested details into Section 2.1 to address this concern.
Finally, it is emphasized several times in the paper that “the study revealed a strong interaction between model resolution and building representation accuracy”. The Authors conclude that at coarser resolutions (e.g. dx = 10m), excluding buildings footprint from the computation domain (BH method) produces the “best” results, while at finer resolutions (e.g. dx = 1 or 2 m), replacing buildings by increased roughness area produces “better” results in terms of flood extend prediction but with a wrong representation of water depths. What is the meaning of such conclusions? In fact, it is simply shown that model calibration is required (calibration of the friction factors, such as explained on lines 103-105) and that this calibration depends on model resolution and buildings representation method (and also probably of the flow solver features). This is particularly visible on figure 1 where the “best” results are gained with the calibration modeling scenario (dx=5m, BH approach) but not with a finer mesh (dx=1 or 2m) while it is clear that a finer mesh enable a better representation of the topography and also probably of the building footprint closer to the reality. This simple but very important message is not clear over the paper.As a basic rule in numerical modeling, the results cannot depend on grid resolution (grid convergence). If this is not the case, grid related approximations are hidden in the calibration parameters, i.e. usually friction coefficients in 2D flow models. In this paper, building representation technique is another source of approximation, whose influence combines with the grid resolution one, making general conclusions impossible.
REPLY: There are several points to address here. First, the study does indeed demonstrate a strong interaction between model resolution and building representation accuracy, as the most significant differences in simulated flood extents occur at coarser resolutions, while these differences diminish at finer resolutions. This clearly implies that the choice of building representation is resolution-dependent: the coarser the resolution, the more critical this choice becomes.
Second, your statement regarding Figure 1 is incorrect. In Figure 1a, the dx = 2m setup with increased resistance (BR10X and BR5X) yields the highest CSI values, and in many other cases, the finer dx = 1m and 2m setups achieve better scores than the 5m BH calibration model. In Figure 1b, it is evident that the finer dx = 1m and 2m setups result in clear higher scores with the HR indicator. Similarly, Figure 1c, which presents the Bias Percentage Indicator, supports the same conclusions as Figure 1a. So when you mention that 'the “best” results are gained with the calibration modeling scenario (dx=5m, BH approach)', this is cleary a false statment. The figure conveys what we have inferred in the manuscript.
Thirdly, in response to your question, "What is the meaning of such conclusions?", the meaning is clear: the paper provides practical guidance on selecting the appropriate building representation method based on the desired simulation outcome. We explicitly highlight which techniques yield more accurate flood extents at different resolutions and which methods produce better water depth predictions across the domain. With this information, users can make an informed decision on the most suitable building representation and resolution for their specific application, depending on whether flood extent or water depth accuracy is more critical to their analysis.
Fourth, regarding the calibration point, it is important to clarify that while calibration generally leads to better model results, it is not necessary for this analysis. The primary objective of this study is not to determine which building representation best simulates the Ahr 2021 flood but rather to evaluate the impact of different representation techniques on simulation outcomes, particularly flood properties such as extent, depth, and velocity, and to highlight the significance of building representation choices in flood modeling and impact analysis. Additionally, the past studies that analyzed either the impact of building representation or grid resolution (Shubert and Sanders, 2012; Jiang et al., 2022; Illiadis et al., 2023) have used only one of their scenarios as a benchmark and used those calibrated values throughout.
Our focus is on how variations in building representation affect simulation results while keeping all other model parameters constant. Our findings, such as “The Building Block and Building Hole approaches typically result in larger flooded areas with deeper water and higher velocities, whereas increasing resistance or omitting buildings leads to smaller flood extents, shallower water, and slower flow,” remain valid regardless of the assigned Manning roughness values. The only instance where roughness can influence the results is in the Building Block setups (except the obvious case of the increased building resistance approach), and only when water levels rise above the buildings — a rare occurrence. In such cases, the roughness value assigned to the urban land category may affect the simulation outcomes.
In order to make all these raised points more clear for the readers we have now revised some text of the manuscript.
References:
-Jiang, Weiwei, et al. "Understanding the effects of digital elevation model resolution and building treatment for urban flood modelling." Journal of Hydrology: Regional Studies 42 (2022): 101122.
-Iliadis, Christos, Vassilis Glenis, and Chris Kilsby. "Representing buildings and urban features in hydrodynamic flood models." Journal of Flood Risk Management 17.1 (2024): e12950.
-Schubert, Jochen E., and Brett F. Sanders. "Building treatments for urban flood inundation models and implications for predictive skill and modeling efficiency." Advances in Water Resources 41 (2012): 49-64.
In addition to these general remarks, the way the accuracy in water depths prediction is evaluated with equation on line 177 is wrong. The norm of simulated to observed values must be computed to avoid compensation errors between over and under estimations. This would certainly affect a large part of the subsequent analysis.REPLY: Thank you for your comment. While it is true that bias alone does not account for the compensation of over- and under-predictions, one could argue whether such compensation is necessarily a drawback. Bias provides valuable information by indicating differences in the means, which is something that metrics like RMSE or NSE do not capture when normalizing or using absolute values. To provide a more comprehensive assessment, we have now included RMSE values alongside bias as a companion metric, ensuring that both overall trends and individual deviations in water depth predictions are properly evaluated.
The claim that the calculated Bias values "affect a large part of the subsequent analysis" is incorrect, as the subsequent analysis was not solely based on the bias, but also the other indicators.In conclusion, I cannot recommend this paper for publication.
Reply/Comment:
We would like to highlight that to the best of our knowledge previous studies have not considered the possible implications of using both different building representation approaches and DEM resolution (Shubert and Sanders, 2012; Jiang et al., 2022; Illiadis et al., 2023). Particularly, RIM2D is a multi-GPU-supported hydraulic model that has an well balanced trade-off between computational costs and accuracy (Apel et al., 2022, 2024). Therefore, it is well-suited for being employed in forecasting chains that can save lives and properties from flood damage (Najafi et al., 2024) and scenario simulations that require multiple simulations (Vorogushyn et al., 2024). In that regard, a technical note highlighting the suitability of different building approaches and recommendations will be, from our point of view, indeed helpful to the modellers in developing an operational or scenario simulation framework.References,
https://www.sciencedirect.com/science/article/pii/S2214581822001355
https://onlinelibrary.wiley.com/doi/full/10.1111/jfr3.12950
https://www.sciencedirect.com/science/article/pii/S0309170812000425
https://www.nature.com/articles/s41467-024-48065-y
https://nhess.copernicus.org/preprints/nhess-2024-97/
https://www.frontiersin.org/journals/water/articles/10.3389/frwa.2024.1310182/full
https://nhess.copernicus.org/articles/22/3005/2022/nhess-22-3005-2022.htmlCitation: https://doi.org/10.5194/hess-2024-314-AC1
-
AC1: 'Reply on RC1', Shahin Khosh Bin Ghomash, 13 Feb 2025
reply
-
RC2: 'Comment on hess-2024-314', Anonymous Referee #2, 18 Feb 2025
reply
The study utilizes the 2021 Ahr Valley flood as a case study to evaluate the impact of different building representation methods—Building Block (BB), Building Hole (BH), and Building Resistance (BR)—on the accuracy of flood modeling. This is a highly relevant topic that sparked scientific debate nearly two decades ago when technological constraints limited the use of high-resolution grids due to computational time restrictions. More recently, significantly more complex techniques have been introduced, incorporating the concept of urban porosity.
However, it is not entirely clear from the article how this study advances the field beyond the existing literature. Additionally, relying solely on comparisons of flood extents provides a weak basis for selecting one method over another. Moreover, the comparison of maximum water depths and velocities, based only on visual inspection of the resulting maps from different methodologies, appears insufficient to definitively determine the superior approach. I suggest that the authors, if feasible, include a point-by-point comparison with observed data to strengthen their analysis.
Citation: https://doi.org/10.5194/hess-2024-314-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 196 | 58 | 11 | 265 | 8 | 13 |
- HTML: 196
- PDF: 58
- XML: 11
- Total: 265
- BibTeX: 8
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1