the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 May 2026
FluidUrban v1.0: Enhancing Urban Ventilation and Pollutant Dispersion Modelling with Three-dimensional Dynamic Adaptive Meshes Optimisation
Abstract. Simulating urban airflow and pollutant dispersion requires resolving multiscale physical processes, from large-scale meteorological forcing to highly localized building-induced turbulence. To accurately capture these multiscale urban flow fields, this study introduces FluidUrban v1.0, an advanced modelling system built upon the Fluidity solver and centred on a three-dimensional Dynamic Adaptive Mesh Optimization (DAMO) framework. By dynamically adapting mesh resolution in response to the evolution of flow physics and scalar gradients, DAMO concentrates computational resources on critical high-gradient regions such as building wakes, shear layers and scalar sharp plume. The model's performance is systematically evaluated against high-fidelity “WOTAN” wind-tunnel experimental data under varying surface roughness conditions and inflow directions. The results demonstrate that the FluidUrban with DAMO framework consistently outperforms traditional non-uniform fixed meshes (FIXM) by accurately capturing complex urban wind fields and pollutant concentration. For normalized wind speed, FluidUrban with DAMO achieved a Mean Absolute Error (MAE) of 0.187, representing a notable reduction from the 0.214 simulated by FIXM. In terms of wind direction, the model reduced the MAE by up to 38.4 % in medium roughness and 36.1 % in high roughness conditions, respectively, during realistic oblique inflow scenarios. Furthermore, for pollutant dispersion, the model effectively suppresses numerical diffusion and maintained sharply plume gradients, achieving an 89 % compliance rate with established atmospheric model evaluation standards (FB, NMSE, and MG), compared to only 50 % for FIXM. While DAMO introduces runtime cost for mesh regeneration, this cost is strategically offset by the optimization of the accuracy-efficiency balance. Following the systematic evaluation, FluidUrban v1.0 was applied to a realistic urban scenario, demonstrating its robust capability to resolve the complex flow fields and spatial heterogeneity within real urban morphologies. Thus, FluidUrban v1.0 demonstrates to be a robust aerodynamic tool for resolving the transient, small-scale flow structures critical to pollutant transport, establishing a solid foundation for the future integration of comprehensive urban physical components, including radiation, vegetation, and full energy-balance physics.
- Preprint
(10650 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-1685', Anonymous Referee #1, 19 May 2026
-
RC2: 'Comment on egusphere-2026-1685', Anonymous Referee #2, 13 Jul 2026
General comments
The authors validated their automated dynamic mesh system by comparing its results with wind tunnel experiments. In particular, they investigated the applicability of this mesh system to physical phenomena involving strong spatial gradients such as building wakes, shear layers and scalar sharp plume.
However, in their comparison with wind tunnel data, they limit their evaluation to assessing only average wind speed and average concentration. Is evaluating average wind speed sufficient to determine whether the system accurately captures complex turbulent behaviors? Shouldn't they also need to assess standard deviation and Reynolds stresses? Furthermore, shouldn't they also examine the continuity of concentration spatial distribution? I believe it is necessary to reconsider the approach of systematical evaluattion.
Specific comments
The authors should show the turbulence characteristics of LES inflow driven by a synthetic eddy method.
They should show spatial difference schemes for flow and scalar fields and type of subgrid models, and Turbulent Schmidt number, etc.
Citation: https://doi.org/10.5194/egusphere-2026-1685-RC2 -
RC3: 'Comment on egusphere-2026-1685', Anonymous Referee #3, 01 Aug 2026
The proposed article introduces FluidUrban a very interesting modelling framework for urban air flows, that differs from other such models by the use of dynamic mesh adaptation, and presents an analysis of its results compared to wind tunnel experimental data. The topic is of significant interest to the community. Although adaptive mesh methods are widely used in aeronautics, their application to environmental engineering remains comparatively limited, making contributions in this direction particularly welcome.
The resulting model appears promising, and the comparison with experimental data is carried out thoroughly, showing a noticeable improvement due to mesh adaptation for the test case considered. However, my understanding is that the paper aims both to present the new model FluidUrban and to validate it through a "systematic and quantitative evaluation". In my opinion, the manuscript does not yet fully achieve either objective. The model itself is not described in sufficient detail, and validation on a single benchmark case, however thorough, is difficult to regard as fully conclusive.
I find the main focus of the paper somewhat unclear : is the paper primarily intended to introduce the complete FluidUrban modelling framework, as suggested by the title, or is its main contribution the incorporation of DAMO within this framework? In the latter case, has the underlying model already been described elsewhere?
I understand the model is based on Fluidity, and that the main additions are related to boundary conditions and meshes. I would nevertheless appreciate a clearer description of the numerical methods employed. Could the authors please specify the spatial and temporal discretisation schemes that are used, together with the main numerical parameters employed in the simulations? In addition, some evidence that the underlying solver behaves satisfactorily on fixed meshes before introducing adaptivity would strengthen the paper.
Concerning meshing and remeshing aspects, I have several questions. More generally, since this is presented as the "core advancement" of the framework, I believe these aspects deserve a more detailed and self-contained description.
* Section 2.1.1 details how the initial mesh is generated. Interesting techniques are mentioned, but with little detail :
(i) "a custom utility designed to extract building polygons directly from high-resolution satellite imagery utilizing a supervised Support Vector Machine (SVM) classification algorithm" seems to be quite a piece of software, could the authors please tell us more about it, or provide a reference?
(ii) I presume that the geometry recovered from the satellite images is simplified for scientific computing purposes. How is this simplification done? How much detail is preserved, how was the choice made, and how does the tool ensure that?
(iii) the initial mesh is apparently not uniform. What determines the local mesh size in the initial mesh? Is it entirely user-defined?
* Section 2.4.2 details the remeshing procedure, and should be the heart of the article. However, I found it difficult to understand the proposed procedure from the current description, and had to infer parts of it from previous publications on adaptivity in Fluidity.
(1) Does this paragraph mean that there is a priori (before the metric-based refinement) refinement based on first and second derivatives ? If so, how does it work ? Otherwise, when are gradients considered in the following procedure ?
(2) (i) could the authors please define a "metric tensor" ? How is it used ?
(ii) could the authors please explicit how this metric tensor is obtained from the Hessian matrix ?
(iii) could the author define the $\cap$ operator ? And explain the procedure for combining such operators ?
(iv) is there any evidence that metric M actually results in an error of $\epsilon$ ? If so, could the author provide such evidence, or reference ?
(3) (i) could the authors please specify which "composite metric tensor" they are referring to ?
(ii) If a global scaling is applied to control the number of nodes, am I wrong to think that \epsilon is not really reached ? f this is correct, what role should be attributed to this error bound?
(iii) How is the adapted mesh generated ? Is the meshing tool the same as in (Pain 2005) ?
* How was the spatial accuracy of the fixed mesh (FIXM) chosen ?Some more questions:
* section 3: I don't think the mesh sizes (number of elements, min size, max size...) are specified for both adaptive and fixed meshes. Could the author give those figures ?
* Figure 6: I would have expected more turbulent variability behind the building. Could the adaptation process selected introduce some dissipation that would filter part of the turbulence structures ?
* Table 3 and others : have the authors looked at statistical dispersion indicators as well ?
* Figure 9 : is DAMO better than FIXM here ?
* Figure 12 : the flow for a fixed mesh (c) seems fairly different from measurements. Why does the model not perform better on this case ? Why is mesh adaptation guaranteed to perform better ?
* Section 3 in general : improvements like MAE changing from 0.214 to 0.187 ar reported. As LES is inherently stochastic, how are the authors sure that these differences are not within expected sampling variability
* Section 3.4 performs a performance assessment on fixed and adaptive meshes with a similar mesh size. What would the comparison results be if, instead, the meshes were chosen to as to provide a similar accuracy ? I believe the "superior balance between accuracy and computation cost" (conclusion) would be more convincingly demonstrated by fixing alternatively the two measures
* The paper argues that DAMO improves the accuracy-efficiency balance. Yet, the only example where performance is evaluated exhibits at 17% overhead for DAMO, which is only moderately convincing. Could the authors provide evidence that the accuracy-efficiency trade-off becomes favourable for larger or more complex problems?Finally, the behaviour of mesh adaptation is often subtle and typically requires validation over a range of configurations before general conclusions can be drawn. This would be all the more true when turbulence modelling is involved, as accurately capturing the relevant flow structures is known to be particularly complex to model and a challenge for usual mesh adaptation strategies. I am curious to know whether FluidUrban has been assessed on additional academic or realistic benchmark cases, and whether a sensitivity analysis of the adaptive parameters has been carried out. Could the author please comment and provide more validation cases ?
Overall, I find the topic timely and the proposed modelling framework promising. The comparison with wind-tunnel experiments is carefully conducted and suggests that the adaptive mesh strategy can provide tangible benefits. However, in its current form, I believe the manuscript would benefit from substantial revision before publication. In particular, the description of the numerical methods and the mesh adaptation strategy should be made more self-contained, the validation should be strengthened or its scope more clearly acknowledged, and several claims regarding performance and efficiency should be supported by additional evidence or discussion. Addressing the points above would, in my opinion, considerably improve the manuscript.
Citation: https://doi.org/10.5194/egusphere-2026-1685-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 264 | 98 | 24 | 386 | 24 | 23 |
- HTML: 264
- PDF: 98
- XML: 24
- Total: 386
- BibTeX: 24
- EndNote: 23
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
To accurately capture multiscale urban flow fields, this work introduces FluidUrban v1.0, an advanced modelling system built upon the Fluidity solver and centred on a three-dimensional Dynamic Adaptive Mesh Optimization (DAMO) framework. Dynamic Adaptive Mesh Optimization is the key to model urban flows. It paves a solid foundation for the future integration of comprehensive urban physical components. It is a good model and presents a nice work.