Applied Energy 182 (2016) 625–633

Contents lists available at ScienceDirect

Applied Energy

journal homepage: www.elsevier .com/locate /apenergy

Defining the Influence Region in neighborhood-scale CFD simulations fornatural ventilation design

http://dx.doi.org/10.1016/j.apenergy.2016.08.0980306-2619/� 2016 Elsevier Ltd. All rights reserved.

⇑ Corresponding author at: Center for Green Buildings and Cities, Graduate Schoolof Design, Harvard University, Cambridge, MA 02138, USA.

E-mail address: ztong@gsd.harvard.edu (Z. Tong).

Zheming Tong ⇑, Yujiao Chen, Ali MalkawiCenter for Green Buildings and Cities, Harvard University, Cambridge, MA 02138, USAGraduate School of Design, Harvard University, Cambridge, MA 02138, USA

h i g h l i g h t s

� The accuracy of natural ventilation analysis relies largely on how the Influence Region is chosen.� Only including the adjacent layer of surrounding buildings is not sufficient.� Three layers of surrounding buildings are typically required for modeling low-rise neighborhoods.� Fewer surrounding buildings are required for wide canyons and high-rise landscapes.� Downstream buildings can be moderately excluded in the Influence Region.

a r t i c l e i n f o

Article history:Received 31 March 2016Received in revised form 27 June 2016Accepted 17 August 2016

Keywords:Natural ventilationInfluence RegionStreet canyonAir change rate (ACH)CFD

a b s t r a c t

Natural ventilation is one of the most important design options for green buildings, which reduces energyuse and improves thermal comfort. Computational Fluid Dynamics (CFD) simulations have been usedincreasingly for natural ventilation design in urban neighborhoods. The accuracy of such simulationsrelies largely on how the CFD domain is chosen. In the domain, we define the Influence Region as the areawhere the surrounding buildings must be modeled explicitly to predict the ventilation flow rate accu-rately. This study presents the early efforts to determine the adequate size of the Influence Region inthe CFD domain using a coupled indoor-outdoor CFD simulation, in which the air change rate (ACH)no longer varies noticeably with increasing number of surrounding obstacles. Convergence charts ofACH as a function of an increasing number of surrounding building layers are generated using variousurban parameters (e.g., wind condition, aspect ratio, building height relative to surroundings, down-stream obstacles, and non-idealized surroundings). Our analysis demonstrated that only including theadjacent layer of surrounding obstacles is not sufficient for predicting correctly the ACH because of theartificial channeling effect between buildings. For both normal and oblique wind directions, three layersof surroundings are required for regular street canyons with an aspect ratio H/W = 1. In the case of widecanyons (H/W = 1/3), two layers of surroundings are needed because there is less flow interferencebetween upstream and downstream obstacles. For the urban configuration, where the target buildingis significantly taller than nearby structures, the ACH on higher floors does not vary much with increasingamount of surroundings, which significantly reduces the required number of buildings in the InfluenceRegion. In addition, buildings at the side and downstream of the target building can be moderatelyexcluded in the Influence Region as long as the most adjacent downstream layer of obstacles is modeled.A real urban configuration with non-uniform spacing among buildings is evaluated. We showed that therequired size of the Influence Region that is derived from uniform building arrays still generally applies tonon-idealized landscapes. This study demonstrates the importance of assessing the sensitivity of theselected Influence Region in CFD simulations to reduce unintended modeling errors and computingexpense.

� 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Natural ventilation has become an increasingly attractivedesign option in the building industry because of the recent focus

626 Z. Tong et al. / Applied Energy 182 (2016) 625–633

on sustainable design [1–3]. It is the process that utilizes naturaloutside air movement to both cool and ventilate a building pas-sively without the use of any mechanical system. Studies havedemonstrated that the cooling energy consumption of naturallyventilated buildings can be reduced by as much as 40–50% in com-parison with an air-conditioned building [4–15].

With the aid of the recent advance in computing technology,CFD is widely used to understand the urban physics in the loweratmospheric boundary layer, due to the relatively low cost, highdata resolution, and adequate validation with experimental data[16–26]. Neighborhood-scale simulation is critical to study variousphysical mechanisms such as local energy demand, pollutanttransport, and thermal comfort [23,27–34]. Successful designs ofnatural ventilation system rely largely on the configuration of sur-rounding neighborhood and the airflow over and through naturallyventilated buildings [35]. Therefore, CFD has been shown as a valu-able tool for natural ventilation designs [3,36–46]. Although CFD isa promising tool, appropriate implementation is required to obtainreliable simulation results. The existing literature has providedbest practice guidelines to determine the size of the CFD domain,boundary conditions, and proper mesh resolution [17,47,48].

In Fig. 1, the size of the CFD domain is determined based on theblockage ratio that avoids artificial acceleration by non-physicalboundaries. In particular, the inlet, lateral, and top boundariesshould be at least 5H away from the Influence Region where build-ings are explicitly modeled (H is the height of the target building).The outflow boundary is recommended to be at least 15H awayfrom the Influence Region. The obstacles in the upstream, down-stream, and lateral areas of the domain are not included explicitlybut their effect on the flow can be parameterized in terms of sur-face roughness in the wall function. In the highlighted InfluenceRegion (Fig. 1), obstacles such as individual buildings and streetsthat are in close vicinity to the target building must be modeledexplicitly with their geometrical shapes. Researchers have demon-strated that the pressure distribution on a building is greatly influ-enced by surrounding structures, i.e., the sheltering effect, throughboth CFD simulation and wind tunnel experiments [35,49–54].

In an urban environment, buildings are grouped closelytogether. To predict the ventilation flow rate accurately, it is essen-tial to explicitly model a sufficient number of surrounding build-ings in the Influence Region. A key question that has not beenaddressed thoroughly in the literature is how many surrounding

Fig. 1. Graphic illustration of the CFD domain and Influence Regi

obstacles should be included in the Influence Region of the CFDdomain. In general, if the modeled Influence Region is larger, theCFD domain needs to be larger, and, therefore, more computationaltime is required. Conversely, underrepresenting the surroundingsleads to incorrect prediction of the ventilation rate at the targetbuilding. To fully address this question regarding the required sizeof the Influence Region, we chose ACH as a primary indicator toderive the proper size of the Influence Region that would capturethe effect of surroundings sufficiently, yet maintain a reasonablecomputational cost.

In practice, ACH is often considered as a target criterion toassess the adequacy of natural ventilation for acceptable indoorair quality and thermal comfort. It is a straightforward measurefor building engineers and it is often used as a ‘‘rule of thumb” inventilation design. There are two common approaches to deter-mine ACH in the design stage. The most common strategy is basedon the orifice equation as described in Eq. (1).

ACH ¼ CD � A � uref �ffiffiffiffiffiffiffiffiffiDCp

p

ffiffiffiffiffi2�

pV

ð1Þ

where CD is the discharge coefficient. A is the area of the opening.uref is the reference wind speed. DCp is the difference in surface-averaged wind pressure coefficient between windward and lee-ward walls, and V is the volume of the cross-ventilated room. Theliterature suggests that CD is in the range of 0.60–0.65 for sharp-edged openings. The value of Cp can be obtained from either directmeasurement (i.e., field experiment and wind tunnel), or indirectmeasurement that includes sources such as existing Cp database(AIVC, ASHRAE) or regression models that are based on a largeamount of empirical data from wind tunnel studies (e.g., TNO Cp

generator) [55–58]. Although the sheltering effect by buildingscan be partially considered in these databases using variousapproximations, de Wit and Augenbroe [59] pointed out that thesemethods, which are based on interpolation or extrapolation ofexisting wind pressure coefficients, can introduce considerableuncertainties. Overall, the orifice equation is derived based on Ber-noulli’s assumption. However, in reality, flow through windowopenings is never laminar. In addition, past studies have shown thatthe use of CD involves many uncertainties and CD also varies withseveral flow variables [60–62]. In wind tunnel studies, the volumet-ric flow rate through cross-ventilated buildings cannot be

on. H is the height of target building in the Influence Region.

Table 1Urban parameters selected to determine the size of Influence Region.

Variables Value

Number of

Z. Tong et al. / Applied Energy 182 (2016) 625–633 627

measured directly. Instead, Cp is measured on solid buildings sur-faces, which assumes the flow field around the building withoutopenings is equivalent to that with openings. It also implies thatthe approaching flow stagnates at the window and the turbulentkinetic energy is dissipated at the window opening. For cross-ventilated buildings with large openings, this approach can intro-duce significant errors [44,63]. The second approach is to computeACH through a coupled indoor-outdoor CFD simulation, whichdirectly calculates the volumetric flow rate through openings. Thecoupled indoor-outdoor CFD simulation resolves the outdoor andindoor wind flow within the same computational domain and isable to capture the dynamic interaction between the outdoor andindoor environment at the window openings [44,46]. This method,in contrast to the first one, eliminates assumptions and uncertain-ties in the orifice equation such as the estimation of CD and Cp.

In this study, we present the early efforts to determine therequired size of the Influence Region in CFD simulation in whichACH no longer varies noticeably with increasing number of sur-rounding obstacles using a coupled indoor-outdoor CFD simula-tion. Convergence charts of ACH as a function of number ofsurrounding building layers are generated using various urbanparameters (e.g., wind condition, aspect ratio, relative buildingheight to surroundings, downstream obstacles, and non-idealizedbuilding configuration). Our goal is to address an important, butoften overlooked, issue in the setup of neighborhood-scale CFDsimulations, and complement existing best practice guidelines forthe design of natural ventilation in dense urban landscapes. Thepaper is organized as follows. We first describe the setup of theCFD simulation. The CFD model is then evaluated against measure-ments from a wind tunnel experiment. Secondly, we derive therequired size of Influence Region by generating convergence chartsof ACH using various urban parameters. A summary of the findingsis provided in the conclusion.

layers: n = 1,2, 3, 4

Normal,Oblique (45�)

H/W = 1(regularcanyon); H/W = 1/3 (widecanyon)

1H, 3H

Number ofdownstreamlayers = 1, 2, 3

2. Model description

2.1. Turbulence modeling

Large Eddy Simulation (LES) with a dynamic subgrid model isemployed here. It is the appropriate turbulent model for simulat-ing highly unsteady and three dimensional flows in urban streetcanyons. LES resolves large-scale unsteady motions and requiresmodeling only the small-scale, unresolvable turbulent motion thatis less influenced by physical boundary conditions [26]. AlthoughLES is more computationally demanding, it shows superior perfor-mance for modeling flows that are highly three dimensional inurban street canyons, and general modeling guidelines were devel-oped [23,64–66].

In LES, a low-pass filtering operation is performed so that theresulting velocity field ~ui can be resolved on a relatively coarsegrid. A dynamic subgrid model is chosen, which allows theSmagorinsky constant to vary in space and time [67]. The constantis dynamically computed and based on the information providedby the resolved scales of motion. The dynamic model removessome issues associated with the constant coefficient Smagorinskymodel by eliminating the need to prescribe a length scale andnear-wall correction [68].

In LES, filtered continuity and momentum equations are shownin Eqs. (2) and (3).

@q@t

þ @q~uj

@xj¼ 0 ð2Þ

@q~ui

@tþ @q~ui~uj

@xj¼ � @~p

@xiþ @sij

@xjþ @rij

@xjð3Þ

sij is the filtered stress tensor, and rij is the subgrid-scale Reynoldsstress, which is modeled using the Boussinesq hypothesis in Eq. (4).

rij � 13dijrkk ¼ �2ltSij ð4Þ

Sij is the rate of strain tensor of the resolved scale. The subgrid vis-cosity lt is modeled according to Eq. (5) [67].

lt ¼ qðLSÞ2jSj ð5ÞLs is the mixing length scale that depends on the size of the compu-tational cell and the dynamically computed Smagorinsky constant.

2.2. Model setup

The coupled indoor-outdoor CFD simulation resolves the out-door and indoor wind flow within the same computational domainand is able to capture the dynamic interaction between the out-door and indoor environment at the window openings [44,46].Depending on the simulated scenarios in Table 1, the dimensionsof the computational domain are 210–1100 m in length,

628 Z. Tong et al. / Applied Energy 182 (2016) 625–633

210–1100 m in width, and 60–180 m in height, which satisfy thedirectional blockage ratio suggested in the guidelines. The compu-tational domain contains 2–11 million unstructured cells depend-ing on the modeling scenario. Although a grid-independentsolution is not available for LES, grid sensitivity studies are con-ducted to ensure the time-averaged results of the selected griddo not vary noticeably with finer grid resolution. The grids aremore refined near the solid cube. Prism layer mesh is applied nearthe wall in order to reduce numerical diffusion. The horizontalhomogeneity is evaluated by creating a flow simulation in anempty domain of the same size as suggested by Blocken et al.[18], which ensures that there is no unintended difference betweenthe inlet and incident profiles.

The numerical setup of the boundary conditions in this studyfollows the CFD guidelines for simulating flows in the urban neigh-borhood [17,47,48]. For the inlet, a logarithmic inlet velocity pro-file was employed according to Eq. (6).

uðzÞ ¼ u�

jln

zþ z0z0

ð6Þ

where j is von Karman constant (0.42), and z0 is aerodynamicroughness height (0.025 mm). u� is the friction velocity. Turbulentkinetic energy k is estimated according to Eq. (7) [48]. The turbu-lence dissipation rate is derived from Eq. (8).

kðzÞ ¼ ðIðzÞuðzÞÞ2 ð7Þ

�ðzÞ ¼ u�3

jðzþ z0Þ ð8Þ

The time-dependent feature of the inlet turbulence profile issimulated with the vortex method, where random vortices in theinlet flow plane for the wall-normal components are generated,providing a spatial correlation [23,69,70]. Symmetry boundaryconditions are applied for the two sides as slip walls with zeroshear. Outflow boundary condition is specified at the end of thedomain. The standard wall function with a roughness modificationis employed [71,72]. The sand-grain roughness height ks androughness constant Cs are estimated from the aerodynamic rough-ness height z0 based on the relationship derived by [18]. Second-order discretization scheme is used in order to reduce numericaldiffusion.

Fig. 2. Schematics of the numerical setup for the CFD evaluation. (a

2.3. CFD validation

Validation is an essential part of model evaluation that providesquantitative views of the model performance. The accuracy of thesimulations was examined by comparing the numerical resultswith the results of wind tunnel experiments conducted by Karavaet al. [73]. The flow field through and over a scaled cross-ventilatedbuilding model in a boundary layer wind tunnel was measuredwith Particle image velocimetry (PIV). Building models with differ-ent openings were built from 2 mm cast transparent polymethyl-methacrylate sheet at a scale of 1:200. The modeled buildingshave a dimension of L �W � H = 100 � 100 � 80 mm3 at wind-tunnel scale and 20 � 20 � 16 m3 at real scale. The thickness ofthe wall is 2 mm. Configurations with both openings at the centerof the two opposite walls and with Window-to-Wall ratio (WWR)of 5% and 10% are shown in Fig. 2a and b. The height of the openingis 18 mm for both configurations while the width is varied to pro-vide the corresponding porosity. The PIV measurements were con-ducted in the vertical plane of symmetry through the openings.

The numerical setup of the model validation case is similar tothat from Section 2.2. The inlet boundary condition is obtainedbased on measured wind speed and turbulent intensity. Symmetryboundary conditions are applied for the two sides as slip walls withzero shear. Outflow boundary condition is specified at the end ofthe domain. Additional details regarding the model setup are pre-sented in our previous study [46]. Fig. 3 shows a comparison ofpredicted and measured time-averaged streamwise flow velocityprofiles. Overall, the performance of the LES model is satisfactory.For the 5%WWR case, the model was able to accurately capture theflow acceleration near both openings, as well as deceleration insidethe building. The 10% WWR case presents a less accurate agree-ment. This is likely due to the size and position of the upstreamvortex in the PIV experiment not being able to be reproduced inthe LES simulation. Discrepancies are observed at the inlet and out-let openings (x/D � 0) and floor (x/D � 1.1). As discussed by Karavaet al. [73], the experimental data is not reliable in the immediatevicinity of the openings due to the effects of shadows or reflectionsthat might have produced uncertainties in the PIV measurementsat these positions.

3. Determining the Influence Region

The Influence Region is defined here as the area where thestructures near the location of interest must be explicitly modeled

) Window-to-Wall ratio = 5%; (b) Window-to-Wall ratio = 10%.

Z. Tong et al. / Applied Energy 182 (2016) 625–633 629

in order to accurately predict the ventilation flow rate. To deter-mine the required size of Influence Region, we generated conver-gence charts of time-averaged ACH as a function of increasingnumber of surrounding layers under various parameters. Table 1presents key urban parameters including wind condition, aspectratio, building height relative to surroundings, and number ofdownstream obstacles. The selection of these parameters are basedon existing Cp generators derived from a large amount of windmeasurement [56,57]. As shown in Table 1, the target cross-ventilated building has a dimension of 10 � 10 � 10 m3 with aWWR of 10%. To reach certain generalizability, matrices of cubicalbuildings are employed to represent idealized urban geometries.We selected a square matrix of buildings with a centered targetbuilding (height H = 10 m and aspect ratio H/W = 1) under normalwind direction as the baseline. A non-idealized building configura-tion with non-uniform spacing based on a real urban neighborhoodin Massachusetts, USA was also examined at the end (Section 4.5).

4. Results and discussion

Fig. 4 presents the ACH computed directly from CFD as a func-tion of the number of surrounding layers (n) for the baseline con-dition. The ACH declines substantially with increasing size of theInfluence Region due to the sheltering effect. In the case of a singlebuilding without surroundings, the wind strikes the building and

Fig. 4. Convergence chart of ACH as a function of number of surrounding layers (n) fornumber of surrounding layers.

Fig. 3. Comparison between experimental and numerical results for the normalized streafrom the study by Karava et al. [73].

induces a positive pressure on the windward face and negativepressure on the leeward face, which then creates a large air flowthrough the openings. The maximum pressure occurs at the localstagnation point where the oncoming flow impinges the buildingsurface. Transitioning from an isolated building configuration tothe configuration surrounded by layers of buildings, the ACH decli-nes significantly due to the addition of upwind buildings. This isbecause the target building at the center is no longer struckdirectly by the approaching flow. As visualized in Fig. 5a, a low-pressure wake region with counter-rotating vortices is formedbehind each building. The approaching flow is channeled into thestreet canyon between buildings. The flow is accelerated in parallelto the oncoming wind direction in the canyon, and deceleratesgradually with increasing distance from the leading edge of thefirst building row. This observed flow pattern is consistent withthe literature on flow passing a matrix of cubical obstacles[26,74–77]. The magnitude of flow acceleration or the channelingeffect is strongest in the canyons at the first building row thatare directly attacked by the approaching flow. As a result, the pres-sure at the wake zone behind the first row is lower than thatbehind the next building row, which creates a reverse flow fromdownwind to upwind direction. Our analysis shows that the influ-ence of each additional surrounding layer on ACH decreases withincreasing distance from the target building at the center. Thechanneling effect around the target building attenuates as the

a regualar street canyon under normal and oblique incident wind of 45�. n is the

mwise velocity component. a) WWR = 5%; b) WWR = 10%. The experimental data is

Fig. 5. Velocity contour plot of the flow field with arrows indicating the flow pattern of a 5 by 5 matrix under (a) normal and (b) oblique (45�) wind directions.

630 Z. Tong et al. / Applied Energy 182 (2016) 625–633

upwind distance increases. The effect becomes negligible near thetarget building, and ACH reaches a convergence when the numberof surrounding building layers increases to three (n = 3).

4.1. Wind direction and speed

The effect of wind direction (or attack angle) is very importantto the flow pattern through and around the target building. In con-trast to normal wind direction from the baseline condition, theflow field around the building is more ‘‘streamlined” under theoblique wind direction (45�). Instead of striking the windward faceand creating a stagnation zone, the flow initially diverges at thewindward corner and travels along the surface until a separationoccurs at the leeward sharp corner (visualized in Fig. 5b). Onestream of the flow feeds into two symmetric counter-rotating vor-tices formed with respect to the 45� diagonal line behind the build-ing. The other stream channeled into the far side of the street re-converges with the one from the symmetric side before hittingthe downstream building. As a result, the ACH for the oblique winddirection is noticeably greater than that of the normal directionexcept for the isolated configuration where the sheltering effectof surroundings is absent. Similar to the normal wind direction,ACH reaches a plateau at n = 3 (three surrounding layers).

Fig. 6a illustrates the effect of wind speed on ACH under twowind speeds (u and 2u) and wind directions (normal and oblique).

Fig. 6. (a) Convergence chart of ACH as a function of increasing number of surrounding laconvergence chart of ACH under two street canyon aspect ratios. n is the number of sur

In general, increasing the wind speed increases the cross ventila-tion and ACH at the target building. However, the increase inACH attenuates with a growing number of surrounding layersdue to the sheltering effect. For the single building configurationwithout any surroundings, doubling the wind speed doubles theACH under both wind directions. However, in the case of multi-layer configurations, the percentage increase in ACH falls withincreasing number of surrounding layers under normal wind direc-tion. In comparison, the percentage increase in ACH remains nearlyunchanged with increasing number of layers under oblique winddirection. This is because the high windward pressure created bythe stagnation effect is greatly reduced when the attack wind angleis oblique (Fig. 5b). With regard to the convergence of ACH, ouranalysis suggests that three layers of surrounding buildings(n = 3) are still necessary regardless of attack wind angle.

4.2. Aspect ratio H/W

The aspect ratio (H/W) of surrounding street canyons is shownto have a great impact on the flow pattern around buildings [78]. Inaddition to a regular canyon (H/W = 1, H = 10 m), a wide canyon(H/W = 1/3, H = 10 m) is investigated, which is observed with lessflow interference between upstream and downstream buildingsin the literature [79]. In contrast with the regular canyons, theACH for the wide street canyons is noticeably greater because there

yers (n) for baseline wind speed (u) and twice the baseline wind speed (2u); (b) therounding layers.

Fig. 7. Convergence chart of ACH as a function of increasing number of surrounding layers (n) under (a) normal wind direction and (b) oblique wind direction (45�). ACH iscomputed on each floor. n is the number of surrounding layers.

Z. Tong et al. / Applied Energy 182 (2016) 625–633 631

is a sufficient distance between obstacles that allows the airflow tore-converge before encountering the next obstacle. In Fig. 6b, a dipin ACH is observed at the n = 1 due to strong channeling effect andlow pressure wake zone behind the first row of buildings, whichdiminishes when the number of surrounding layers increases. Inthe case of a wide canyon, two layers of surrounding buildings(n = 2) are required to be modeled explicitly in the InfluenceRegion.

4.3. High-rise building

If the relative height of the target building is significantlygreater than the surroundings, the required size of the Influenceregion will be affected. As shown in Table 1 (4), we designedanother configuration where a cross-ventilated high-rise is sur-rounded by low-rise. Fig. 7 depicts the convergence charts ofACH at each floor with increasing number of layers. The ACH atthe ground level is mostly affected by the size of the surroundings.On the other hand, the ACH of the third floor does not vary muchwith increasing number of surrounding layers, because the portionof the building that is taller than the surroundings acts essentiallyas an isolated building. This suggests that cross-ventilated high-rise is less influenced by nearby obstacles and therefore requiresonly two layers of surroundings (n = 2).

Fig. 8. Convergence chart of ACH as a function of removed number of downwindlayers.

4.4. Downstream and lateral obstacles

The presence of downstream obstacles may play a less impor-tant role in ACH than the upstream obstacles. Here, we exploredthe impact of downstream buildings on ACH of the target buildingby removing one downstream layer at a time as displayed inTable 1 (5). This sensitivity study is based on a three-layer config-uration derived previously. The resulting convergence chart of ACHas a function of the number of layers removed is presented in Fig. 8.Instead of modeling the downstream area with the same amount ofupwind surrounding buildings, we demonstrate that only the mostadjacent downstream layer is required under both normal andoblique wind direction, which to a large extent, reduces the com-putational cost.

4.5. Non-idealized landscape

In addition to idealized building matrices with uniform spacingemployed previously, we chose a non-idealized landscape withnon-uniform spacing and unequal building dimensions based ona real urban neighborhood in Massachusetts, USA to examine thesize of the Influence Region. In Fig. 9, each surrounding layer iscolor coded. Similar to the uniform building arrangement, the con-vergence chart of ACH is shown in Fig. 10 where ACH falls withincreasing number of layers. The same channeling effect is

Fig. 9. Non-idealized building array based on a real neighborhood. Each buildinglayer is color coded as described in the legend.

Fig. 10. Convergence chart of ACH as a function of increasing number ofsurrounding layers (n) under normal and oblique wind directions. n is the numberof surrounding layers.

632 Z. Tong et al. / Applied Energy 182 (2016) 625–633

observed at the one-layer configuration where there is a drop inACH at n = 1 (Fig. 10). For both wind directions, the convergenceis found at n = 3, which is equivalent to that from uniform buildingarrays. Although the irregularity alters the flow pattern to certainextent, the substantial decrease in wind speed towards the centerof the building matrix reduces the ACH greatly regardless of thebuilding arrangements. Therefore, our findings indicate that therequired size of the Influence Region that is derived from uniformbuilding arrays can be generally applied to non-idealizedconfigurations.

5. Concluding remarks

The accuracy of natural ventilation analysis relies largely onhow the Influence Region is chosen in neighborhood-scale CFDsimulations. A sufficient amount of buildings near the point ofinterest (i.e., a cross-ventilated building) must be modeled explic-itly in order to accurately predict the ventilation flow rate and ACH.Although the sheltering effect by surrounding buildings is recog-nized in the literature, the derivation of the required size of theInfluence Region has not been analytically investigated. In thisstudy, we present the early efforts to derive the required size ofthe Influence Region in the CFD domain in which the ACH at thecross-ventilated building no longer varies with increasing numberof surrounding obstacles using a coupled indoor-outdoor CFD sim-ulation. The impact of various urban parameters (e.g., wind condi-tion, aspect ratio, building height relative to surroundings,downstream obstacles, and non-idealized surroundings) on ACHare evaluated against the number of surrounding layers. The con-clusions from our analysis are briefly outlined. Only including theadjacent layer of surrounding obstacles is not adequate to correctlypredict the ACH due to the artificial channeling effect betweenbuildings. We found that three layers of surroundings (n = 3) mustbe modeled explicitly for regular street canyons (H/W = 1) undernormal and oblique wind directions whereas the Influence Regioncan be reduced to two layers (n = 2) for wide canyons (H/W = 1/3)and high-rise configuration. Buildings at the side and downstreamof the target building can be moderately excluded in the InfluenceRegion as long as the most adjacent downstream obstacles aremodeled. A real urban configuration with non-uniform spacingand dimensions is also evaluated. Our analysis indicates that theirregularity does not noticeably change the required size of theInfluence Region.

The derived conclusions are not intended to be exhaustive, andthey may vary to certain degrees depending on the actual land-

scape. Our aim here is to address an important, but often over-looked, issue in the setup of neighborhood-scale CFD simulations,and complement existing best practice guidelines for the designof natural ventilation in dense urban areas. Given the complexityof real urban landscapes, it is advisable to assess the sensitivityof any selected Influence Region following the steps presented hereto avoid unintended modeling errors and unnecessary computingexpense.

Acknowledgment

Z.T. and Y.C. are grateful for the postdoctoral fellowship fromthe Center for Green Buildings and Cities (CGBC) at HarvardUniversity.

References

[1] Allocca C, Chen Q, Glicksman LR. Design analysis of single-sided naturalventilation. Energy Build 2003;35:785–95.

[2] Brager GS, de Dear RJ. Thermal adaptation in the built environment: aliterature review. Energy Build 1998;27:83–96.

[3] Chen Q. Ventilation performance prediction for buildings: a method overviewand recent applications. Build Environ 2009;44:848–58.

[4] Cao B, Zhu Y, Li M, Ouyang Q. Individual and district heating: a comparison ofresidential heating modes with an analysis of adaptive thermal comfort.Energy Build 2014;78:17–24.

[5] Cardinale N, Micucci M, Ruggiero F. Analysis of energy saving using naturalventilation in a traditional Italian building. Energy Build 2003;35:153–9.

[6] Gratia E, De Herde A. Natural cooling strategies efficiency in an office buildingwith a double-skin façade. Energy Build 2004;36:1139–52.

[7] Kolokotroni M, Aronis A. Cooling-energy reduction in air-conditioned officesby using night ventilation. Appl Energy 1999;63:241–53.

[8] Li X, Wen J. Review of building energy modeling for control and operation.Renew Sustain Energy Rev 2014;37:517–37.

[9] Li X, Wen J, Malkawi A. An operation optimization and decision framework fora building cluster with distributed energy systems. Appl Energy2016;178:98–109.

[10] Oropeza-Perez I, Østergaard PA. Energy saving potential of utilizing naturalventilation under warm conditions – a case study of Mexico. Appl Energy2014;130:20–32.

[11] Ramponi R, Angelotti A, Blocken B. Energy saving potential of nightventilation: sensitivity to pressure coefficients for different Europeanclimates. Appl Energy 2014;123:185–95.

[12] Wang R, Wu Y, Ke W, Zhang S, Zhou B, Hao J. Can propulsion and fuel diversityfor the bus fleet achieve the win–win strategy of energy conservation andenvironmental protection? Appl Energy 2015;147:92–103.

[13] Wang Y, Zhao F-Y, Kuckelkorn J, Spliethoff H, Rank E. School building energyperformance and classroom air environment implemented with the heatrecovery heat pump and displacement ventilation system. Appl Energy2014;114:58–68.

[14] Tong Z, Chen Y, Malkawi A, Liu Z, Freeman RB. Energy saving potential ofnatural ventilation in China: The impact of ambient air pollution. Appl Energy2016;179:660–8.

[15] Li X, Wen J, Bai E-W. Developing a whole building cooling energy forecastingmodel for on-line operation optimization using proactive systemidentification. Appl Energy 2016;164:69–88.

[16] Baik J-J, Kim J-J, Fernando HJS. A CFD model for simulating urban flow anddispersion. J Appl Meteorol 2003;42:1636–48.

[17] Blocken B. Computational Fluid Dynamics for urban physics: importance,scales, possibilities, limitations and ten tips and tricks towards accurate andreliable simulations. Build Environ 2015;91:219–45.

[18] Blocken B, Stathopoulos T, Carmeliet J. CFD simulation of the atmosphericboundary layer: wall function problems. Atmos Environ 2007;41:238–52.

[19] Britter RE, Hanna SR. Flow and dispersion in urban areas. Annu Rev Fluid Mech2003;35:469–96.

[20] Di Sabatino S, Buccolieri R, Pulvirenti B, Britter R. Simulations of pollutantdispersion within idealised urban-type geometries with CFD and integralmodels. Atmos Environ 2007;41:8316–29.

[21] Gromke C, Buccolieri R, Di Sabatino S, Ruck B. Dispersion study in a streetcanyon with tree planting by means of wind tunnel and numericalinvestigations – evaluation of CFD data with experimental data. AtmosEnviron 2008;42:8640–50.

[22] Tong Z, Baldauf RW, Isakov V, Deshmukh P, Zhang K Max. Roadside vegetationbarrier designs to mitigate near-road air pollution impacts. Sci Total Environ2016;541:920–7.

[23] Tong Z, Zhang KM. The near-source impacts of diesel backup generators inurban environments. Atmos Environ 2015;109:262–71.

[24] Wang YJ, Nguyen MT, Steffens JT, Tong Z, Wang Y, Hopke PK, et al. Modelingmulti-scale aerosol dynamics and micro-environmental air quality near a largehighway intersection using the CTAG model. Sci Total Environ2013;443:375–86.

Z. Tong et al. / Applied Energy 182 (2016) 625–633 633

[25] Wang YJ, Zhang KM. Modeling near-road air quality using a computationalfluid dynamics model, CFD-VIT-RIT. Environ Sci Technol 2009;43:7778–83.

[26] Xie Z, Castro I. LES and RANS for turbulent flow over arrays of wall-mountedobstacles. Flow Turbul Combust 2006;76:291–312.

[27] Chua KJ, Chou SK, Yang WM, Yan J. Achieving better energy-efficient airconditioning – a review of technologies and strategies. Appl Energy2013;104:87–104.

[28] Oh MS, Ahn JH, Kim DW, Jang DS, Kim Y. Thermal comfort and energy saving ina vehicle compartment using a localized air-conditioning system. Appl Energy2014;133:14–21.

[29] Tong Z, Wang YJ, Patel M, Kinney P, Chrillrud S, Zhang KM. Modeling spatialvariations of black carbon particles in an urban highway-buildingenvironment. Environ Sci Technol 2012;46:312–9.

[30] Zhang S, Wu Y, Hu J, Huang R, Zhou Y, Bao X, et al. Can Euro V heavy-dutydiesel engines, diesel hybrid and alternative fuel technologies mitigate NOXemissions? New evidence from on-road tests of buses in China. Appl Energy2014;132:118–26.

[31] Tong Z, Whitlow TH, Landers A, Flanner B. A case study of air quality above anurban roof top vegetable farm. Environmental Pollution 2016;208(PartA):256–60.

[32] Chen Y, Samuelson HW, Tong Z. Integrated design workflow and a new tool forurban rainwater management. Journal of Environmental Management2016;180:45–51.

[33] Tong Z, Whitlow TH, MacRae PF, Landers AJ, Harada Y. Quantifying the effect ofvegetation on near-road air quality using brief campaigns. EnvironmentalPollution 2015;201:141–9.

[34] Zhang S, Wu Y, Un P, Fu L, Hao J. Modeling real-world fuel consumption andcarbon dioxide emissions with high resolution for light-duty passengervehicles in a traffic populated city. Energy 2016;113:461–71.

[35] Hooff Tv, Blocken B. On the effect of wind direction and urban surroundings onnatural ventilation of a large semi-enclosed stadium. Comput Fluids2010;39:1146–55.

[36] Asfour OS, Gadi MB. Using CFD to investigate ventilation characteristics ofvaults as wind-inducing devices in buildings. Appl Energy 2008;85:1126–40.

[37] Cheung JOP, Liu C-H. CFD simulations of natural ventilation behaviour in high-rise buildings in regular and staggered arrangements at various spacings.Energy Build 2011;43:1149–58.

[38] Chow WK. Wind-induced indoor-air flow in a high-rise building adjacent to avertical wall. Appl Energy 2004;77:225–34.

[39] Ding W, Hasemi Y, Yamada T. Natural ventilation performance of a double-skin façade with a solar chimney. Energy Build 2005;37:411–8.

[40] Ji Y, Cook MJ, Hanby V. CFD modelling of natural displacement ventilation inan enclosure connected to an atrium. Build Environ 2007;42:1158–72.

[41] Linden PF. The fluid mechanics of natural ventilation. Annu Rev Fluid Mech1999;31:201–38.

[42] Malkawi A, Yan B, Chen Y, Tong Z. Predicting thermal and energy performanceof mixed-mode ventilation using an integrated simulation approach. BuildSimul 2016;9:335–46.

[43] Mistriotis A, Bot GPA, Picuno P, Scarascia-Mugnozza G. Analysis of theefficiency of greenhouse ventilation using computational fluid dynamics. AgricFor Meteorol 1997;85:217–28.

[44] Ramponi R, Blocken B. CFD simulation of cross-ventilation flow for differentisolated building configurations: validation with wind tunnel measurementsand analysis of physical and numerical diffusion effects. J Wind Eng IndAerodyn 2012;104–106:408–18.

[45] Tan G, Glicksman LR. Application of integrating multi-zone model with CFDsimulation to natural ventilation prediction. Energy Build 2005;37:1049–57.

[46] Tong Z, Chen Y, Malkawi A, Adamkiewicz G, Spengler JD. Quantifying theimpact of traffic-related air pollution on the indoor air quality of a naturallyventilated building. Environ Int 2016;89–90:138–46.

[47] Franke J, Hellsten A, Schlunzen KH, Carissimo B. The COST 732 Best PracticeGuideline for CFD simulation of flows in the urban environment: a summary.Int J Environ Pollut 2011;44:419–27.

[48] Tominaga Y, Mochida A, Yoshie R, Kataoka H, Nozu T, Yoshikawa M, et al. AIJguidelines for practical applications of CFD to pedestrian wind environmentaround buildings. J Wind Eng Ind Aerodyn 2008;96:1749–61.

[49] Bauman F, Ernest D, Arens EA. The effects of surrounding buildings on windpressure distributions and natural ventilation in long building rows. AshraeTrans 1988;94.

[50] Cóstola D, Blocken B, Hensen JLM. Overview of pressure coefficient data inbuilding energy simulation and airflow network programs. Build Environ2009;44:2027–36.

[51] Hu C-H, Wang F. Using a CFD approach for the study of street-level winds in abuilt-up area. Build Environ 2005;40:617–31.

[52] Jiang Y, Alexander D, Jenkins H, Arthur R, Chen Q. Natural ventilation inbuildings: measurement in a wind tunnel and numerical simulation withlarge-eddy simulation. J Wind Eng Ind Aerodyn 2003;91:331–53.

[53] Stathopoulos T. Wind loads on low-rise buildings: a review of the state of theart. Eng Struct 1984;6:119–35.

[54] Wirén BG. Effects of surrounding buildings on wind pressure distributions andventilative heat losses for a single-family house. J Wind Eng Ind Aerodyn1983;15:15–26.

[55] ASHRAE. ASHRAE handbook – fundamentals; 2009.[56] Grosso M. Wind pressure distribution around buildings: a parametrical model.

Energy Build 1992;18:101–31.[57] Knoll B, Phaff J, De Gids W. Pressure simulation program. Proceedings of the

conference on implementing the results of ventilation research. AIVC; 1995.pp. 233–233.

[58] Swami M, Chandra S. Correlations for pressure distribution on buildings andcalculation of natural-ventilation airflow. ASHRAE Trans 1988;94:243–66.

[59] de Wit S, Augenbroe G. Analysis of uncertainty in building design evaluationsand its implications. Energy Build 2002;34:951–8.

[60] Chu CR, Chiu YH, Chen Y-J, Wang Y-W, Chou CP. Turbulence effects on thedischarge coefficient and mean flow rate of wind-driven cross-ventilation.Build Environ 2009;44:2064–72.

[61] Heiselberg P, Sandberg M. Evaluation of discharge coefficients for windowopenings in wind driven natural ventilation. Int J Vent 2006;5:43–52.

[62] Sawachi T, Narita K-i, Kiyota N, Seto H, Nishizawa S, Ishikawa Y. Wind pressureand air flow in a full-scale building model under cross ventilation. Int J Vent2004;2:343–57.

[63] Kato S, Murakami S, Mochida A, Akabayashi S-i, Tominaga Y. Special Issue 8thInternational Conference on Wind Engineering 1991 Velocity-pressure field ofcross ventilation with open windows analyzed by wind tunnel and numericalsimulation. J Wind Eng Ind Aerodyn 1992;44:2575–86.

[64] Gousseau P, Blocken B, Stathopoulos T, van Heijst GJF. CFD simulation of near-field pollutant dispersion on a high-resolution grid: a case study by LES andRANS for a building group in downtown Montreal. Atmos Environ2011;45:428–38.

[65] Tseng Y-H, Meneveau C, Parlange MB. Modeling flow around bluff bodies andpredicting urban dispersion using large eddy simulation. Environ Sci Technol2006;40:2653–62.

[66] Walton A, Cheng AYS, Yeung WC. Large-eddy simulation of pollutiondispersion in an urban street canyon—Part I: Comparison with field data.Atmos Environ 2002;36:3601–13.

[67] Germano M, Piomelli U, Moin P, Cabot WH. A dynamic subgrid-scale eddyviscosity model. Phys Fluids A 1991;3:1760–5.

[68] Pope SB. Turbulent Flows. Cambridge University Press; 2000.[69] Kraichnan RH. Diffusion by a random velocity field. Phys Fluids

1970;13:22–31.[70] Mathey F, Cokljat D, Bertoglio JP, Sergent E. Assessment of the vortex method

for large eddy simulation inlet conditions. Prog Comput Fluid Dynam Int J2006;6:58–67.

[71] Cebeci T, Bradshaw P. Momentum transfer in boundary layers. Washington,DC/New York: Hemisphere Publishing Corp./McGraw-Hill Book Co.; 1977. 407p; 1.

[72] Launder BE, Spalding DB. The numerical computation of turbulent flows.Comput Methods Appl Mech Eng 1974;3:269–89.

[73] Karava P, Stathopoulos T, Athienitis AK. Airflow assessment in cross-ventilatedbuildings with operable façade elements. Build Environ 2011;46:266–79.

[74] Cheng Y, Lien FS, Yee E, Sinclair R. A comparison of large Eddy simulations witha standard k–e Reynolds-averaged Navier-Stokes model for the prediction of afully developed turbulent flow over a matrix of cubes. J Wind Eng Ind Aerodyn2003;91:1301–28.

[75] Coceal O, Thomas T, Castro I, Belcher S. Mean flow and turbulence statisticsover groups of urban-like cubical obstacles. Bound-Layer Meteorol2006;121:491–519.

[76] Davidson MJ, Mylne KR, Jones CD, Phillips JC, Perkins RJ, Fung JCH, et al. Plumedispersion through large groups of obstacles—a field investigation. AtmosEnviron 1995;29:3245–56.

[77] Hanna SR, Tehranian S, Carissimo B, Macdonald RW, Lohner R. Comparisons ofmodel simulations with observations of mean flow and turbulence withinsimple obstacle arrays. Atmos Environ 2002;36:5067–79.

[78] Oke TR. Street design and urban canopy layer climate. Energy Build1988;11:103–13.

[79] Vardoulakis S, Fisher BEA, Pericleous K, Gonzalez-Flesca N. Modelling airquality in street canyons: a review. Atmos Environ 2003;37:155–82.