 Research
 Open access
 Published:
Simulation of a groundmounted prism in ABL flow using LES: on overview of error metrics and distribution
Advances in Aerodynamics volume 5, Article number: 9 (2023)
Abstract
The accuracy of wind loading predictions using Large Eddy Simulation (LES) is usually influenced by numerous model parameters, which can influence the obtained results. The validation of numerical simulations with traditional Wind Tunnel Test (WTT) is still an important task, necessary to increase our a priori knowledge of possible inaccuracies and set up mitigation strategies. In this study, LES is used to simulate the wind fields around an isolated model highrise building, under seven wind attack angles and validated with WTT results. The influence of various settings and parameters on the model performance is studied. For the angle of attack showing higher inaccuracy, different mesh refinement strategies and turbulence models are tested. Results indicate that LES can accurately predict the mean and local maxima of the pressure coefficients for both perpendicular and skew wind attack angles, as well as reproduce global forces and their envelopes with very good accuracy. Conversely, pronounced errors are found in the prediction of the pressure coefficient standard deviation and the local minima. The highest deviations between LES and WTT are found close to the leading edge in correspondence of flow separations which are observed in WTT and not in LES for skew flows. The addition of boundary layer cells and the use of different subgrid models have very limited effectiveness in modifying the obtained results for the analysed case.
1 Introduction
Large Eddy Simulation, LES, is known to be a powerful tool for the investigation of wind effects on structures such as highrise buildings [1,2,3,4]. However, comparison with traditional Wind Tunnel Test (WTT) sometimes shows discrepancies, so that their validation and the individuation of critical aspects which might affect results is still an important research activity. Numerous studies can be found in the literature validating LES results in terms of wake flow characteristics [5, 6], global aerodynamic forces [7, 8] and pressure distribution [9,10,11]. The usefulness of such validations mainly consists in the individuation of critical aspects which can be sometimes moderated by appropriate modelling choices. When moderation strategies are not found or lead to excessive computational burden, their individuation anyway allows to build up knowledge, which can help in correctly interpreting results obtained by numerical models when WTT results are not available.
From the studies above and analogous works [12, 13], it can be found that the model performance of LES is usually good but, in some cases, sensitive to various model parameters. Among them, the accurate reproduction of flow characteristics in the Atmospheric Boundary Layer (ABL) is known to be crucial for the evaluation of wind effects on structures [14, 15]. Nevertheless, numerous other factors can influence the simulation [16,17,18], which include, as pointed out in [19], ground roughness, subgrid scale model, domain size, nearwall treatment and sensitivity to mesh and timestep.
Numerous contributions have been devoted to investigating the influence of some of the aforementioned model parameters on the model accuracy. Unfortunately, due to the variability of the adopted computational setup and analysed flows, contradictory results are sometimes found.
In particular, Wang et al. [20] assessed the effects of the mesh type and boundary mesh on the timeaveraged and fluctuating wind characteristics around an isolated highrise building standing in ABL. They found that the case with hexahedral cells has the best agreement with experiments. They also found that the boundary layer mesh does not improve the numerical accuracy in any circumstance, despite the fact that the nondimensional wall distance (\(y^{+}\)) is often considered to be a very important quantity to be controlled. Wijesooriya et al. [21] showed that the choice of the SubGrid Scale (SGS) model does impact the flow field. The wall adapting local eddy viscosity (WALE) model was found to be the most suitable among the tested models including SmagorinskyLilly model, Dynamic model, kinetic energy sub grid scale model and wall modeled LES. Liu et al. [22] examined the effect of mesh resolution and different SGS models on the prediction of the pedestrian level wind fields around building arrays. They observed that the dynamic SmagorinskyLilly method gave the best model performance and recommended appropriate values for time step and sampling period.
The aim of the present paper is to assess the numerical accuracy of LES in predicting wind loads on a highrise building and to study the influence of some important model parameters on the accuracy of the obtained results. The highrise building model is an isolated flattopped box with an aspect ratio of 1:3:5, standing in an ABL. It is one of the highrise building models from the public wind tunnel database constructed by Tokyo Polytechnic University (TPU), Japan, which have been considered also in some of the already available contributions [19, 23,24,25].
Firstly, the numerical results are compared with WTT data in the form of several model performance metrics for 7 angles of attack ranging from \(0^{\circ }\) to \(90^{\circ }\). Then, for the worst case, corresponding to a skew flow impinging at \(45^{\circ }\), the spatial distribution of the pressure coefficient statistics (i.e. timeaverage, standard deviation and peak values) and the corresponding Prediction Error (PE), defined as the difference between numerical and experimental results, are analysed. For such worst case, different numerical setups are evaluated varying mesh configuration and testing three SGS models, namely, WALE, \(k\omega\) SSTSAS and Smagorinsky.
The paper is organized as follows. The descriptions of WTT and LES are provided in the next section, along with the procedure used to generate inflow conditions. Then, the numerical results are validated in Section 3. Subsequently, the sensitivity studies of the PE to the mesh configuration and the SGS model are provided in Section 4. Finally, conclusions are drawn in Section 5.
2 WTT and LES description
2.1 WTT description
The WTT were performed in the Boundary Layer Wind Tunnel (BLWT) of TPU, Japan. It is an open circuit tunnel with 2.2 m width and 1.8 m height. The ABL in the experiments was generated through spires and square blocks acting as roughness elements. It corresponds to a terrain category IV in the Architectural Institute of Japan (AIJ) standard [26]. The wind speed at building height is around 11.11 m/s, and the streamwise turbulence intensity (\(I_{u\_WTT}\)) is approximately 15%.
The tested model is a flattopped prismatic box with 0.5 m height (H) and 0.3 m (L) ∗ 0.1 m (W) rectangular crosssection. The length scale reduction factor of the experiments is 1:400, leading to a building height equal to 200 m in real scale. 480 wind pressure taps are positioned on the four side faces, sampling synchronously with a frequency of 1000 Hz for a duration of 32.8 s. The origin of the coordinate system is the center of the building base. The geometry of the building is presented in Fig. 1 (a).
Incidence angles ranging from \(0^{\circ }\) to \(100^{\circ }\) with an increment of \(5^{\circ }\) are considered in WTT. The \(0^{\circ }\) is defined as the condition in which the flow is moving along the \(+x\) direction and it is perpendicularly hitting the long building side, as indicated in Fig. 1 (b). The complete database and detailed description of the WTT are publicly accessible [27].
2.2 LES description
LES are designed to reproduce the WTT using the open source software OpenFoam v6.0. A 3D view of the computational domain is presented in Fig. 2 (a). The domain has the dimensions of 3 m, 2.2 m and 1.8 m in streamwise, lateral and vertical directions, respectively. The distance between the inlet patch and the building location is 1 m, which equals the double of the building height H. The blockage ratio of the simulations is approximately 3.0% for \(0^{\circ }\) and is 1.0% for \(90^{\circ }\), which are both lower than recommendations [28].
The turbulent flow at the inflow patch is generated using a synthetic turbulence generation technique called PRFG\(^3\). It is an extended version of the Prescribedwavevector Random Flow Generator (PRFG) [29], aiming at giving control over threedimensional spectral densities, in other words, explicitly targeting all integral length and time scales of turbulence. Interested readers are invited to the references [29, 30] for details regarding the method. The top and two sides of the domain are treated as symmetry conditions, mimicking the effect of wind tunnel walls. The domain bottom and the building surfaces are treated as walls. In particular, the ground of the domain is modeled as a rough wall with a roughness height of 0.015 m (model scale) and a roughness constant of 0.5. The building surfaces are modeled as smooth walls. An inletoutlet condition is used for the outlet patch.
An overview of the mesh is shown in Fig. 2 (b). The snappyHexMesh is used for meshing. A rotorstator method is adopted. Specifically, the rotor is a cylinder of radius 0.25 m (reduced scale) with an axis oriented along the zdirection passing through the building center. It is rigidly rotated when studying the wind effects at different attack angles without remeshing the geometry. The stator is the rest of the domain, as indicated in Fig. 2 (b) and (c). As can be seen, a hexahedral mesh is used for the whole domain and cell expansion gradients are used, leading to the fact that the mesh size at the domain top and outlet is around two times bigger than at the domain center. The mesh size is around 0.016 m at the inflow patch and it is around 0.004 m up to a distance of 0.1 m from the building surfaces. This yields the number of cells for the building H, L and W to be 125, 75 and 25, respectively, which fulfills the requirements in [18]. The total number of cells is around 1.8 million.
As for the numerical schemes utilized for the simulations, the pressure velocity coupling is imposed using the PISO algorithm. The time discretization is performed using the Crank Nicolson scheme with a blending coefficient of 0.85. The face fluxes for all the quantities are calculated using the bounded Gauss linear scheme except for the velocity which is calculated using the secondorder LUST scheme. The kEquation is used as the LES subgrid scale turbulence model. The nondimensional time step (\(\Delta t*(U_r/H)\)) for the calculation is 0.01, yielding the Courant number of the calculation to be around 3.0 on average and only attain 6.0 in some small size cells.
2.3 Inflow
As mentioned before, synthetic turbulence is applied at the inflow. In particular, PRFG\(^3\) is used for generating the ABL flow. It targets a Category IV profile in the AIJ standard code, which is in agreement with the one adopted in WTT. The mean velocity and the alongwind turbulence intensity, \(I_u\), distribution along height are set following the power law in the code prescription, being 0.27 the exponential and 550 m the reference height of ABL. The turbulence intensities in the y and z directions are set to be 0.75 and 0.5 times of the turbulence intensity in the x direction, respectively. We mention that in order to obtain the target wind field characteristics at the building location, the turbulence intensity of each velocity component has been increased of about 20% at the inflow to compensate for the energy dissipation between the inflow and the building location. The reference velocity, \(U_r\), is chosen as the timeaveraged wind speed at the reference height, \(H_r\), which is the height of the building.
In order to check the performance of the turbulence generator, an empty domain with the synthesized turbulent inflow condition is tested first. The wind profiles at the location where the building will be placed are shown in Fig. 3. As can be seen, the mean wind speed profiles from the AIJ code, WTT and LES agree well with each other. In the same figure the turbulence intensity profiles are reported, being \(I_{u\_LES}\), \(I_{v\_LES}\) and \(I_{w\_LES}\) at \(H_r\) equal to 12%, 10% and 8%, respectively. Furthermore, the simulated turbulence integral length scales at \(H_r\), evaluated from velocity timehistories and assuming frozen turbulence, \(L_{ux}\), \(L_{vx}\) and \(L_{wx}\), nondimensionalized to the WTT length scale, are 0.375, 0.2 and 0.125, respectively.
The power spectral density of the velocity components u, v and w at the building height are presented in Fig. 4 (a), (b) and (c), respectively, showing a good agreement with the von Karman (indicated as "Tar. Karman" in the figures) wind spectrum up to the cutoff frequency, equal to approximately 65 Hz. Overall, the simulated turbulent inflow shows a satisfactory agreement with WTT and with the targeted profiles.
3 Results and discussion
3.1 Mesh
As is well known, it is ideally necessary to ensure that results obtained using numerical simulations are independent from the adopted mesh size [28]. When using LES, it is actually more appropriate to state that results of interest shall be independent or vary within acceptable ranges with the mesh size, as a complete independency is actually not expected. We start by considering the case at \(0^\circ\). For such case, we use three meshes, namely, coarse mesh (CM), medium mesh (MM) and fine mesh (FM), to check the grid dependency of the numerical solution. The number of cells of the three meshes is 0.1, 1.8 and 6.6 million, respectively. In other words, the ratio between the cell size in each direction between two consecutive meshes is approximately 1.5, following the suggestion in [31]. A summary of the characteristics of the three meshes is provided in Table 1. The computational resources needed to perform each simulation are also reported.
In order to characterize the obtained results, the timeaverage, rms, maximum and minimum values of \(C_p\), respectively denoted as \(\bar{C}_p\), \(C^{\prime }_p\), \(\hat{C}_p\) and \(\check{C}_p\), are considered. Peak values are calculated by fitting to the twominute extreme using a Gumbel cumulative probability distribution function and then, adopting the wellknown shifting property of the Gumbel distribution [32], to extract the 10minute extremes associated with a nonexceedance probability equal to 80%, in agreement with the Cook and Mayne approach [33].
Figure 5 presents the scatter plots of the \(C_p\) statistics from simulations using the aforementioned three meshes. As shown in Fig. 5 (a)  (c), no remarkable differences in terms of \(\bar{C}_p\), \(C^{\prime }_p\) and \(\hat{C}_p\) can be observed, except from a very slight underestimation of \(\bar{C}_p\) in CM (the difference is approximately 0.1). Regarding \(\check{C}_p\), results obtained from CM only arrive at 2.5. However, they attain approximately 4.0 in the simulations with MM and FM, showing no apparent variation between the two meshes. Some localized points have more severe extreme suctions in MM and FM than in CM, while the results from CM are closer to WTT, which seem nonintuitive. However, as indicated in an analogous work [34], a finer grid model might underestimate the turbulence dissipation rate and give a higher turbulent kinetic energy prediction, yielding overall better performance with coarser meshes. In all, results highlight that CM is probably excessively coarse, as it substantially differs from the other two. Results appear to be quite similar with respect to all analyses quantities for MM and FM, so that MM is chosen for the next analyses.
3.2 Pressure coefficients statistics
After choosing the mesh size to be used, we proceed at simulating seven attack angles, equally spaced between \(0^\circ\) and \(90^\circ\). The LES results are compared with WTT results for all the simulated wind attack angles in the form of three error metrics, namely, the mean error (ME), the standard deviation of error (SE) and the coefficient of determination (\(R^2\)) [12], as reported below.
where the subscript i represents the pressure tap number and the total number of monitors, N, is equal to 480. The \(\overline{WTT}\) (or \(\overline{LES}\)) represents the timeaverage and \(\sigma _{WTT}\) (or \(\sigma _{LES}\)) represents the standard variation of the WTT (or LES) dataset. The error metrics regarding different \(C_p\) statistics for all the considered cases are reported in Table 2. Notice that ME for \(\bar{C}_p\) is not reported, as it might be simply related to the choice of the reference pressure, so being irrelevant.
Looking at Table 2, an overall correspondence in terms of \(\bar{C}_p\) is obtained between WTT and LES, with the average \(R^2\) around 0.99. Similarly, good agreement could also be found regarding the \(\hat{C}_p\), whose average \(R^2\) is 0.93.
Considering all the wind attack angles, we observe that no clear trend emerges for the error metrics when the angle is varied. For instance, \(R^2\) of \(C^{\prime }_p\) for \(0^{\circ }\) is 0.72, lower than some skew angles such as 0.86 of \(30^{\circ }\) and 0.76 of \(75^{\circ }\). However, at \(90^{\circ }\) we obtain the highest value of \(R^2\) among all the cases, equal to 0.89. Substantially, a good matching is obtained when the flow is orthogonal to the short side, leading to a reattached flow. In other conditions, results do not follow simple trends. The prediction of fully detached flow expected at \(0^{\circ }\) is accurate, but it is not more accurate than other skew angles.
Figures 6 and 7 show the local peak values of \(C_p\) considering at the same time wind incident angles from \(0^{\circ }\) to \(90^{\circ }\), i.e. the extremes from all attack angles are enveloped together, as for design purposes. In particular, we show scatter plots reporting in abscissa WTT results and in ordinates LES results. In order to allow to individuate taps belonging to different building faces, pressure taps are grouped as \(face\ I\), \(face\ II\), \(face\ III\) and \(face\ IV\), following the indication in Fig. 1 (a). The results in terms of the peak \(\hat{C}_p\) appear to be characterized by much higher accuracy with respect to suctions. The extreme negative pressures near the building edges on \(face\ II\), \(face\ III\) and \(face\ IV\) reach values of 4.0 in WTT but only attain 3.5 in LES. Conversely, LES overestimates the negative extremes near the center bottom of \(face\ IV\), where the \(\check{C}_p\) arrives at 5.0 in LES but reaches only 3.5 in WTT.
3.3 Global forces and moments
The global forces for all simulated wind attack angles are reported in dimensionless form \(C_{Fx}\), \(C_{Fy}\), \(C_{Mx}\) and \(C_{My}\), representing the force and moment components in alongwind (x) and crosswind (y) directions, as shown in Figs. 8 and 9. Specifically, the time series of \(C_p\) is integrated over the building surfaces to obtain time histories of the nondimensional force coefficients. Then, for each incidence angle, the peak values (including the maximum and minimum) of forces and moments are calculated following the same procedure used for the pressure peaks explained in Section 3.1. Again, for design purposes, the peaks from all the angles are enveloped and indicated as \(LES\ env.\) and \(WTT\ env.\) in the figures.
It can be seen that the LES reproduces the global forces and moments very well with only a slight underestimation in terms of the maximum values of positive \(C_{Fx}\) and \(C_{Fy}\). The strongest deviation is on \(15^{\circ }\), where the extreme values of both forces and moments in LES are about 25% lower than in WTT. Again, discrepancies can be found for \(45^{\circ }\), for which the peak values of forces (and moments) have smaller values in LES than in WTT. Similar results are also found in another work [35], which declared that the wind attack angle of \(15^{\circ }\) approaches the critical angle or glancing angle and has the minimum drag force coefficient and maximum mean lift magnitude, and the Strouhal number (St) is maximized.
3.4 Prediction errors for \(45^{\circ }\) case
As the \(45^{\circ }\) case appears to be the one presenting higher discrepancies between WTT and LES, we further investigate it in this section in more detail. Hereafter the Prediction Error (PE) is used as an error metric and is defined as \(PE_S = LES_S  WTT_S\), where the subscript S represents the \(C_p\) statistics, i.e., \(\bar{C}_p\), \(C^{\prime }_p\), \(\hat{C}_p\) and \(\check{C}_p\).
Figure 10 presents the scatter plots of \(C^{\prime }_p\) and \(\check{C}_p\) for the \(45^{\circ }\) case. The scatter plots of \(\bar{C}_p\) and \(\hat{C}_p\) are not reported here, since they are satisfactorily accurate. It can be seen that \(C^{\prime }_p\) on \(face\ II\) shows systemic underestimations from the numerical model. In fact, monitors with \(C^{\prime }_p\) approaching 0.5 in WTT are lower than 0.3 in LES. As expected, underestimations also regarding \(\check{C}_p\) can be found on these probes, with the negative extremes in LES being 60% less pronounced than in WTT. Besides, some localized differences regarding \(\check{C}_p\) can be found on \(face\ III\). The values of \(\check{C_p}\) of these points are around 3.0 in WTT but are lower than 2.0 in LES.
Figures 11 and 12 present the surface distributions of \(C^{\prime }_p\) and \(\check{C}_p\) as well as their prediction errors, i.e., \(PE_{C^{\prime }_p}\) and \(PE_{\check{C}_p}\). More specifically, it can be clearly seen that the underestimations mentioned above are located near the leading edge of \(face\ II\). Looking at Fig. 11 (a) and (b), the higher values of \(C^{\prime }_p\) are also located on this region.
As for the independent spots near the top edge of \(face\ III\) where the highest \(PE_{\check{C}_p}\) appears, one possible explanation can be found in a relative study [23]: the extreme value is recorded downstream of the leading edge, from which strong vortices are expected to be shed.
Figure 13 presents the frequency distributions of \(PE_{C'_p}\) and \(PE_{\check{C}_p}\). It can be seen that the \(PE_{C'_p}\) concentrates around 0.05. It can also be found that deviations between LES and WTT in terms of extreme suctions (\(PE_{\check{C}_p}\)) are mainly inside the range 1.0 and 0.0, which indicates a tendency of the simulation to underestimate such quantity and provides a measure of the expected underestimations.
4 Sensitivity study
In this section, the influence of parameters including mesh setting and LES subgrid scale (SGS) model on numerical results for the \(45^{\circ }\) case is investigated, attempting to ameliorate the previously obtained prediction errors.
4.1 Mesh
We firstly consider the possibility to ameliorate results by refining the mesh. In Section 3, we observed that the selected meshing parameters were able to guarantee a low sensitivity of the obtained results to the cell size. Nevertheless, the study was performed at \(0^\circ\), so that it cannot be automatically extended to \(45^\circ\).
We here consider two mesh refinement strategies: in the first one we refine all the fluid volume around the building, while in the second one we add boundary layer cells. Then, the two strategies are combined.
Consequently, three adjusted meshes based on MM are obtained. The details of the tested meshes are listed below: (I) boundary layer cells, which have 8 layers and the minimum thickness equaling 0.0004 m and scratching ratio of 1.15, are added to the building surfaces, leading to the mesh named BL; (II) the mesh size is refined to 0.004 m up to a distance of 0.2 m from the building and to 0.002 m to a distance of 0.025 m from the building, resulting in the mesh named BR; (III) the two aforementioned strategies are applied at the same time, yielding the mesh named BL_BR. The mesh zones far away from the building are kept unchanged. An overview of the newly considered meshes in the proximity of the building is provided in Fig. 14 and their characteristics are provided in Table 3. It is worth noting that the addition of boundary layer cells dramatically increases (double) the computational costs, though a relatively lower (half) value of \(y^{+}\) can be obtained at the building surfaces.
The error metrics of the three cases are presented in Table 4. Unexpectedly, the mesh with boundary layer cells, i.e., BL, has worse performance in terms of \(C'_p\) prediction compared to the original mesh, MM. Specifically, the \(R^2\) of \(C'_p\) for mesh MM is 0.60 while for the mesh BL it decreases to 0.54. Differently, the BR mesh shows better performance in terms of \(C'_p\), reaching \(R^2\) equal to 0.65. As it can be seen, despite such variations, the overall error does not change substantially, especially if we consider the high increase in computational resources needed to make the refinements. We also show that the combination of the two refinement strategies BL_BR does not show a more accurate prediction of \(C'_p\).
Now we see the simulated results in terms of \(\check{C}_p\). Again, slightly better performance can be found from BR than from BL, while the results from BR and BL_BR are close to each other.
Figures 15 and 16 show the surface distributions of the prediction errors \(PE_{C^{\prime }_p}\) and \(PE_{\check{C}_p}\) for the three meshes. The range of \(PE_{C^{\prime }_p}\) for BL is close to the original simulation, MM, still showing high error values on \(face\ II\), which seems to indicate a detachment of the flow at the edge for WTT not predicted by LES. This can be deduced also by Fig. 11 (a) and (b) in which \(C^{\prime }_p\) is shown. The distribution of \(PE_{\check{C}_p}\) for all the analysed meshes is shown in Fig. 16, which does not show any major difference between the three.
4.2 Subgrid scale model
It has been seen that refining the mesh did not yield strong improvements of the results, despite its high increase of computational costs. We thus here investigate the effect of changing the turbulence model. Table 5 reports error metrics of \(C_p\) statistics for simulations on mesh MM with different turbulence models (namely, WALE, \(k\omega\) SSTSAS and Smagorinsky). Overall, none of them shows better model performance than the original turbulence model, i.e., kEquation, whose ME for \(C^{\prime }_p\) and \(\check{C}_p\) are 0.03 and 0.16, respectively. Only marginal improvements can be seen, for instance, regarding \(R^2\) for \(\check{C}_p\), with the value increasing from 0.68 in the original simulation to 0.71 in the simulation with the Smagorinsky model.
From the distributions of \(PE_{C^{\prime }_p}\) (Fig. 17) and \(PE_{\check{C}_p}\) (Fig. 18), the differences between the three cases are also insignificant. In a summary, the prediction errors of \(C^{\prime }_p\) and \(\check{C}_p\) for \(45^{\circ }\) are not sensitive to changes in the adopted turbulence model.
5 Conclusion
Wind fields around an isolated highrise building model with an aspect ratio of 1:3:5 are simulated using LES to evaluate the model performance and investigate the influences of several model parameters on the prediction errors. Simulations are performed for seven wind attack angles ranging from \(0^{\circ }\) to \(90^{\circ }\) with an increment of \(15^{\circ }\). The main conclusions of the study are hereby drawn as below:

1.
Globally, LES is capable of capturing the mean and maximum wind pressure on the building for both perpendicular and skew wind attack angles, with the correlation (\(R^2\)) between the experimental and predicted results being 0.99 for \(\bar{C}_p\) and 0.93 for \(\hat{C}_p\) on average. The model performance on the prediction of global forces (moments) is also extremely satisfactory, showing very good agreement with WTT results. However, the numerical accuracy in terms of \(C^{\prime }_p\) and \(\check{C}_p\) is less satisfactory, especially in a few cases. The \(R^2\) for \(C^{\prime }_p\) and \(\check{C}_p\) are 0.74 and 0.78 on average, respectively;

2.
For the considered cases, the worst predictions are obtained with an attack angle of \(45^\circ\). The high values of \(PE_{C^{\prime }_p}\) and \(PE_{\check{C}_p}\) for this case concentrate on the leading edge in correspondence of the flow separations. It appears that a flow separation occurs in WTT which is not predicted by the adopted numerical models;

3.
The prediction errors of \(C^{\prime }_p\) and \(\check{C}_p\) for \(45^{\circ }\) are not sensitive to the addition of boundary layer cells for the nearwall treatment. Some marginal improvements are obtained by refining the mesh in the surroundings of the building, but this will dramatically increase the computational time;

4.
No major improvement is obtained by changing the adopted turbulence model.
Overall, it is difficult to individuate with certainty the cause of the observed discrepancies. Actually, many possible causes can be individuated. Such causes range from small deviations of the geometry with respect to WTT to differences in the incoming turbulence, despite the fact that a good matching was obtained in terms of turbulence intensity, and length scales were prescribed according to usual practice. Surely, we observe that the considered case is not particularly sensitive to the adopted turbulence model and mesh size, once an appropriate mesh has been initially selected. It must be remarked, that the addition of boundary layers proved, beside not ameliorating results, had only a very limited effect on the obtained results.
Availability of data and materials
Data used and/or analysed during the current study is included in this published article and are available from the corresponding author on reasonable request. Part of the data can be openly accessed at TPU.
Change history
17 March 2023
The heading of Acknowledgements has been updated.
References
Jiang G, Yoshie R (2020) Side ratio effects on flow and pollutant dispersion around an isolated highrise building in a turbulent boundary layer. Build Environ 180:107078
Kataoka H, Ono Y, Enoki K (2020) Applications and prospects of CFD for wind engineering fields. J Wind Eng Ind Aerodyn 205:104310
Luo Y, Liu H, Xue H et al (2019) Largeeddy simulation evaluation of wind loads on a highrise building based on the multiscale synthetic eddy method. Adv Struct Eng 22(4):997–1006
Vita G, Salvadori S, Misul DA et al (2020) Effects of inflow condition on RANS and LES predictions of the flow around a highrise building. Fluids 5(4):233. Accessed 1 March 2022
Tominaga Y, Mochida A, Murakami S et al (2008) Comparison of various revised k–ε models and LES applied to flow around a highrise building model with 1: 1: 2 shape placed within the surface boundary layer. J Wind Eng Ind Aerodyn 96(4):389–411
Zu G, Lam KM (2018) LES and wind tunnel test of flow around two tall buildings in staggered arrangement. Computation 6(2):28
Alminhana GW, Braun AL, LoredoSouza AM (2018) A numericalexperimental investigation on the aerodynamic performance of CAARC building models with geometric modifications. J Wind Eng Ind Aerodyn 180:34–48
Liu J, Hui Y, Wang J et al (2021) LES study of windwardfacemountedribs’ effects on flow fields and aerodynamic forces on a square cylinder. Build Environ 200:107950
Buffa E, Jacob J, Sagaut P (2021) LatticeBoltzmannbased largeeddy simulation of highrise building aerodynamics with inlet turbulence reconstruction. J Wind Eng Ind Aerodyn 212:104560
Ke J (2019) RANS and hybrid LES/RANS simulations of flow over a square cylinder. Adv Aerodyn 1(1):10
Yu Y, Yang Y, Xie Z (2018) A new inflow turbulence generator for large eddy simulation evaluation of wind effects on a standard highrise building. Build Environ 138:300–313
Papp B, Kristóf G, Gromke C (2021) Application and assessment of a GPUbased LES method for predicting dynamic wind loads on buildings. J Wind Eng Ind Aerodyn 217:104739
Thordal MS, Bennetsen JC, Capra S et al (2020) Towards a standard CFD setup for wind load assessment of highrise buildings: Part 1 – Benchmark of the CAARC building. J Wind Eng Ind Aerodyn 205:104283
Lamberti G, Gorlé C (2020) Sensitivity of LES predictions of wind loading on a highrise building to the inflow boundary condition. J Wind Eng Ind Aerodyn 206:104370
Thordal MS, Bennetsen JC, Koss HHH (2019) Review for practical application of CFD for the determination of wind load on highrise buildings. J Wind Eng Ind Aerodyn 186:155–168
Bruno L, Salvetti MV, Ricciardelli F (2014) Benchmark on the aerodynamics of a rectangular 5:1 cylinder: An overview after the first four years of activity. J Wind Eng Ind Aerodyn 126:87–106
Schatzman M (2010) COST 732 model evaluation case studies: approach and results. Meteorological Inst
Tominaga Y, Mochida A, Yoshie R et al (2008) AIJ guidelines for practical applications of CFD to pedestrian wind environment around buildings. J Wind Eng Ind Aerodyn 96(1011):1749–1761
Guichard R (2019) Assessment of an improved random flow generation method to predict unsteady wind pressures on an isolated building using largeeddy simulation. J Wind Eng Ind Aerodyn 189:304–313
Wang W, Cao Y, Okaze T (2021) Comparison of hexahedral, tetrahedral and polyhedral cells for reproducing the wind field around an isolated building by LES. Build Environ 195:107717
Wijesooriya K, Mohotti D, Chauhan K et al (2019) Numerical investigation of scale resolved turbulence models (LES, ELES and DDES) in the assessment of wind effects on supertall structures. J Build Eng 25:100842
Liu J, Niu J, Du Y et al (2019) LES for pedestrian level wind around an idealized building array—assessment of sensitivity to influencing parameters. Sustain Cities Soc 44:406–415
Cao Y, Liu X, Zhou D et al (2022) Investigation of local severe suction on the side walls of a highrise building by standard, spectral and conditional POD. Build Environ 217:109047
Ricci M, Patruno L, Kalkman I et al (2018) Towards LES as a design tool: wind loads assessment on a highrise building. J Wind Eng Ind Aerodyn 180:1–18
Wang Y, Chen X (2020) Simulation of approaching boundary layer flow and wind loads on highrise buildings by wallmodeled LES. J Wind Eng Ind Aerodyn 207:104410
Tamura Y, Ohkuma T, Kawai H, Uematsu Y, Kondo K (2004) Revision of AIJ recommendations for wind loads on buildings. Structures 2004: Building on the Past, Securing the Future 2004:1–10
Tokyo Polytechnic University (2007) TPU aerodynamic database. http://wind.arch.tkougei.ac.jp/system/eng/contents/code/tpu. Accessed 1 March 2022
Franke J, Hellsten A, Schlünzen H et al (2007) Best practice guideline for the CFD simulation of flows in the urban environmenta summary. COST Office, Belgium
Patruno L, Ricci M (2017) On the generation of synthetic divergencefree homogeneous anisotropic turbulence. Comput Methods Appl Mech Eng 315:396–417
Bervida M, Patruno L, Stanič S et al (2020) Synthetic generation of the atmospheric boundary layer for wind loading assessment using spectral methods. J Wind Eng Ind Aerodyn 196:104040
Ferziger JH, Perić M (2002) Computational methods for fluid dynamics, 3rd edn. Springer, Berlin, Heidelberg
Dyrbye C, Hansen SO (1997) Wind loads on structures. Wiley, Chichester
Cook NJ, Mayne JR (1980) A refined working approach to the assessment of wind loads for equivalent static design. J Wind Eng Ind Aerodyn 6(1–2):125–137
Gousseau P, Blocken B, van Heijst GJF (2013) Quality assessment of largeeddy simulation of wind flow around a highrise building: validation and solution verification. Comput Fluids 79:120–133
Cao Y, Tamura T, Kawai H (2019) Investigation of wall pressures and surface flow patterns on a wallmounted square cylinder using very highresolution Cartesian mesh. J Wind Eng Ind Aerodyn 188:1–18
Acknowledgements
The author is thankful to CINECA for providing the HPC facilities which allowed the completion of the present study. The author is also thankful to Tokyo Polytechnic University for providing the wind tunnel data. The support provided by the China Scholarship Council (CSC) is acknowledged.
Funding
Not applicable.
Author information
Authors and Affiliations
Contributions
Jin Xing: Methodology, investigation, conceptualization, writing, and visualization. The author read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The author declares that she has no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Xing, J. Simulation of a groundmounted prism in ABL flow using LES: on overview of error metrics and distribution. Adv. Aerodyn. 5, 9 (2023). https://doi.org/10.1186/s4277402300139z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s4277402300139z