CFD Simulation of Near-Field Pollutant Dispersion in the Urban Environment: A Review of Current Modeling Techniques

TL;DR: In this paper, a review of current modeling techniques in CFD simulation of near-field pollutant dispersion in urban environments and discusses the findings to give insight into future applications is presented.

About: This article is published in Atmospheric Environment.The article was published on 2013-11-01 and is currently open access. It has received 372 citations till now. The article focuses on the topics: CFD in buildings.

Summary

Introduction

• Air pollution near and around buildings is an important environmental problem.
• This model is specialized for use in building design, but has limited applicability and less accuracy concerning building configuration details (Hajra et al. 2010; 2011) .
• Furthermore, for evaluating the quality of CFD simulations, it is necessary to analyze its sensitivity and uncertainty appropriately.
• Section 3 identifies key features of near-field pollutant dispersion around buildings from previous studies and discusses their relevance in CFD modeling.

2.1 Dispersion around an isolated building

• Among the studies on dispersion around a generic building model, it is important to investigate not only the concentration field but also the relationship between dispersion and flow structure, particularly because most studies focused only on concentration distributions.
• Therefore, in order to validate and evaluate CFD performance on such phenomena, model results should be compared with experimental data on both the concentration and velocity fields.
• Robins and Castro (1977b) measured concentration field downstream in the vicinity of a surface mounted cube with a source in a simulated atmospheric boundary layer.
• Their results were discussed by considering the influence of the flow field in the vicinity of the cube measured in a series of experiments (Robins and Castro, 1977a) .
• Recently, Yoshie et al. (2011) examined flow and dispersion fields with gas emission from the wake region of a single building by detailed experiment, in which instantaneous wind velocity, temperature, and concentration, enabling measurement of turbulent heat flux and turbulent concentration flux, measured simultaneously.

2.2 Dispersion in and around a single street canyon

• Street canyons are typical architectural structures in urban environments, and they represent highly polluted zones around buildings.
• In order to minimize the effect of these pollutants in built-up environments, it is necessary to model and accurately predict contaminant dispersion properties in street canyons.
• Vardoulakis et al. (2003) and Ahmad et al. (2005) reviewed the measurements and modeling techniques for wind and pollutant transport within street canyons.
• A 3-D street canyon, which consists of two or more building blocks, has three determining factors of flow regime, namely the relative height (H), width (W) and length (L) of the canyon, in contrast to a 2-D canyon with two factors, H and W.
• As summarized in Li et al. (2006) , several studies for 3-D street canyons investigated 3-D lateral and secondary flows, which are absent in 2-D simulations, although strongly influence the vertical mixing of pollutant concentration (Hunter et al., 1992; Leitl and Meroney, 1997) .

2.3 Dispersion in and around building arrays (continuous street canyons)

• Boppana et al. (2010) demonstrated the influence of the roughness morphology on the dispersion processes and the power of LES for obtaining physically important scalar turbulent flux information.
• The results of DNS by Branford et al. (2011) largely helped to further understand the processes affecting the plume structure, e.g. channeling, lateral dispersion, detrainment, secondary-source dispersion and plume skewing.

2.4 Dispersion around building complexes

• Wind tunnel modeling has also been conducted at the University of Hamburg (Kastner-Klein et al., 2004) to measure high resolution flow and dispersion data sets that supplement JU2003 field data.
• Several CFD studies have been applied to this subject (Lien et al., 2008) .
• Hanna et al., (2006) compiled computational results of tracer gas dispersion using the urban atmospheric boundary layer scenario in New York City.
• They found that the five CFD models they compared produce similar wind flow patterns, as well as good agreement with winds observed during a field experiment.
• Recently, a relevant tracer dispersion experiment conducted in London has been referred to in several CFD simulations (Xie and Castro, 2009; Xie, 2011; Xie et al., 2013) .

3 Key features of near-field pollutant dispersion around buildings

• Based on the studies discussed in the preceding section, some features of near-field pollutant dispersion around buildings can be identified by considering bluff body aerodynamics and atmospheric dispersion.
• The relationship between these features and reminders of the applicability of CFD modeling are explained in this section.

3.1 Three-dimensionality of mean flow

• Fig. 3 shows main features of flow around cuboids at 0° and 45° to the approach flow in a thick boundary layer (Robins and Macdonald, 2001) .
• Since the production of turbulence statistics, such as Reynolds stress, are closely related to the strain-rate tensors, the turbulence characteristics around a bluff body become complex due to the complicated distribution of the strain-rate tensors.
• Therefore, in the near-field, the pollutant diffusion phenomenon also has a fully three-dimensional nature due to the flow around a building.
• It can be even more prevalent in a group of buildings (e.g., where contaminants can travel upstream in the cavity of one building to the sidewall eddy induced by another building).
• Furthermore, as pointed out by Meroney (1990) , the aerodynamic characteristics of a dispersion plume depend upon the shape and intensity of motion within separated flow regions around the obstacle.

3.2 Unsteadiness of large-scale flow structure

• A large-scale organized motion in the flow field around a bluff body can be explained easily for the flow past a 2D cylinder.
• When an isolated bluff body has a simple shape, the contribution of the large scale unsteadiness, like vortex shedding, is relatively large.
• Because RANS is derived by ensemble average, it is basically applicable to non-stationary flows such as periodic or quasi-periodic flows involving deterministic structures.
• Most of the turbulence models used to close the equations are valid only as long as the time over which these changes in the mean occur is large compared to the time scales of the turbulent motion containing most of the energy.
• Therefore, special attention to turbulence models used is necessary to use URANS successfully.

3.3 Anisotropy of turbulent scalar fluxes

• For transport and dispersion around buildings, large eddies act to physically transport contaminants from one point in space to another.
• A pollutant released in a street canyon is intermittently ejected out of the canyon due to the instantaneous flow field, and it results in periods of low and high concentration in the canyon.
• Several algebraic formulations for the turbulent scalar flux have been proposed and some of them applied successfully to turbulent mass transfer in complex flows (Rogers et al., 1989; Abe and Suga, 2001; Younis et al., 2005) .
• Recently, Rossi and Iaccarino (2009) showed that the SGDH failed to predict the streamwise component of the scalar flux through a numerical study of a line source downstream of a square obstacle.
• Gousseau et al. (2011b) indicated that the counter-gradient mechanism occurs for not only cubic buildings, but also when the source is higher and less affected by the building-generated turbulence.

4 Prediction accuracy of various turbulence models

• In terms of the accuracy of CFD simulations, the main sources of errors and uncertainties should be identified.
• The ERCOFTAC guidelines categorize errors and uncertainties in CFD simulations as model uncertainty, discretization or numerical error, iteration or convergence error, round-off errors, application uncertainties, user errors and code errors (Casey & Wintergerste, 2000) .
• The model uncertainty, which is due to the difference between the real flow and the exact solution of the model equations, is mainly discussed in this paper.
• As ensured in the ERCOFTAC guidelines, it should be noted that the relevance of turbulence modeling only becomes significant in CFD simulations when other sources of error, in particular the numerical and convergence errors, have been removed or properly controlled.

4.1 Comparison of various RANS models

• It is well known that the standard k-ε model (SKE; Launder and Spalding, 1972) poorly represents separation flow due to the overestimation of turbulent kinetic energy (TKE) near the upwind corner of a building (Murakami, 1993) .
• Therefore, when a source exit exists in the recirculation regions on the roof and walls, poor concentration prediction is observed.
• This is why the overestimation of TKE near the upwind corner in SKE causes large turbulent diffusion at the side and leeward walls of the building.
• The Reynolds-stress models (RSM; Launder et al., 1975) has often provided the worst results in comparative studies of various turbulence models, although it can occasionally capture the near-wall flow phenomena (Murakami et al., 1996; Wang and McNamara, 2006; Koutsourakis et al., 2012) .
• This is mainly because RSM requires optimization of many numerical parameters due to the comparatively high number of differential equations to be solved, and the simulations generally have a higher dependency on the chosen mesh and more difficulty converging when compared to the k-ε models.

4.3 Comparison of RANS and LES

• LES computation can yield important information on instantaneous fluctuations of concentration that cannot be directly obtained by RANS computations.
• In compensation, a model solving the transport equation of the concentration variance has been known as the prediction method of concentration fluctuation in RANS model (Launder, 1978) .
• With the availability of measurements of concentration fluctuation in a plume dispersing in an urban area, efforts to model the concentration variance for urban plumes have been undertaken by several researchers (Andronopoulos et al., 2002; Hsieh et al., 2007; Mavroidis et al., 2007; Wang et al., 2009; Milliez and Carissimo, 2008; Yee et al., 2009) .
• Generally, it has been demonstrated that these models can be used to reproduce concentration variance in an idealized urban area under laboratory conditions.
• The models for closure of the transport equations for concentration variance need to be further validated using other available urban flow and pollutant dispersion data from both laboratory and full-scale experiments.

5 Issues with modeling of boundary conditions

• The accuracy of CFD is also affected by uncertainties in the precise geometry, uncertain data or models that need to be specified as boundary conditions such as turbulence properties at an inlet (Casey & Wintergerste, 2000) .
• This section describes two important issues with modeling of boundary conditions, i.e., inflow boundary condition and expression of unresolved obstacles, affecting near-field pollutant dispersion in the urban environment.

5.1 Inflow boundary condition

• Additionally, some researchers also indicated that the influence of solar radiation has significant influence on pollutant diffusion inside street canyons (Sini et al.
• These studies reveal that the differential heating of street surfaces due to solar radiation can largely influence the flow's capability to transport and exchange pollutants.

5.2 Unresolved obstacles

• Many obstacles affect near-field dispersion in urban environments.
• Canopy models have been developed for representing surface roughness effect of obstacles on wind profiles for a long time.
• The ensemble-averaged properties in experiments can only be compared with RANS CFD results.
• A 3-D Eulerian-Lagrangian approach to moving vehicles that takes into account the traffic-induced flow and turbulence was proposed by Jicha et al. (2000) for pollutant dispersion in a street canyon.
• On the other hand, an explicit treatment of moving vehicles can be undertaken by computations using the moving finite element/control volume method (e.g. Fluidity, 2012) .

6.1 Evaluation of CFD models

• Recently, based on accumulated knowledge from the meteorological field, the COST Action 732 (2005) (2006) (2007) (2008) (2009) proposed and tested a new protocol for evaluation and quality assurance of numerical models for micro-scale urban meteorology (Britter and Schatzmann, 2007) .
• The proposed evaluation protocol has several distinct elements: a scientific evaluation process, a verification process, the provision of appropriate and quality assured validation data sets, a model validation process, and an operational evaluation process that reflects the needs and responsibilities of the model user.
• The maximum concentration is a key metric for most air quality applications.
• Furthermore, accurate prediction of plume location and coverage is very important when developing and evaluating urban dispersion models, but it poses a difficult challenge (Brown, 2004) .
• In such cases, LES, which can directly evaluate a specific percentile of concentration, has a great advantage when compared to the RANS approach.

6.2 Measurements data for validation

• In particular, it is important that the required boundary conditions, such as inflow conditions, are carefully provided for computations.
• As pointed out by Schatzmann and Leitl (2011) , atmospheric variability may be critical when CFD results are compared with field measurement data.
• The constraint is felt particularly strongly in short-range urban dispersion because of the high degree of inherent natural variability, coupled with confounding factors such as traffic conditions (Belcher et al., 2012) .
• The database will be extended step by step by more complex configurations.
• Detailed wind tunnel experiments were also performed in order to study flow and pollutant dispersion in the real urban environment (Carpentieri et al., 2009) .

7 Conclusions

• In the past decade, CFD has become a more accessible tool due to the continued progress of modeling studies and the rapid increase of computational resources.
• In particular, for the prediction of the pedestrian wind environment, CFD has already been a viable method using several practical guidelines.
• For near-field pollutant dispersion around buildings, it is still needed to discern the applicability of CFD further with paying careful attention to the above quote.
• As demonstrated in this paper, inevitably, high quality CFD is often time consuming and costly.
• The validity of the level of expertise required and the time (cost) involved should be carefully evaluated on the basis of its purposes by comparing them with those of other assessment methods.

Figures

Fig. 6. Periodic and stochastic fluctuations in the turbulent vortex shedding flow. f ~ : periodic fluctuation; "f : stochastic turbulent fluctuation; 'f : total fluctuation (periodic + stochastic); <f>p: phase-averaged value; <f> t: time-averaged value (Murakami and Mochida, 1995).

Fig. 3. Main features of flow around cuboids in a thick boundary layer. (a) Mean flow patter around a cuboid normal to the approach flow (lines marked ‘sa’ denote mean separation/attachment lines). (b) Mean flow pattern around a cuboid at 45 o incidence to the approach flow (Robins and Macdonald, 2001).

Fig. 7. Contours of normalized concentrations (K) on building surfaces and surrounding streets for different turbulence models; (a) SKE and (b) LES (Gousseau et al., 2011a).

Fig. 4. Distribution of velocity gradients around a simple bluff body. xi : three components of spatial coordinates (i=1,2,3: streamwise, lateral, vertical); ui : three components of wind velocity; < >: ensemble average. (Murakami, 1993).

Fig. 1. Schematic diagram of flow and contamination patterns around a rectangular building (ASHRAE 2011).

Fig. 2. Iso-surfaces of instantaneous scalar concentration C/Cm = 10 -3 from the DNS run with Re = 5,000: upper left panel, t* = 8, upper right panel, t* = 11, lower left panel, t* = 18, lower right panel, t* = 24 (Rossi et al., 2010).

Fig. 5. Schematic diagram of plume dispersion through an array of obstacles (Davidson et al., 1995).