CFD Modeling of Reactive Pollutants Dispersion in Simplified Urban Configurations with Different Chemical Mechanisms

An accurate understanding of urban air quality requires considering a coupled behavior between dispersion of reactive pollutants and atmospheric dynamics. Currently, urban air pollution is mostly dominated by traffic emissions, being nitrogen oxides (NOx) and Volatile Organic Compounds (VOCs) the primary emitted pollutants. However, modeling reactive pollutants with a large set of chemical reactions using a computational fluid dynamics (CFD) model requires a large amount of computational time. In this sense, the selection of the chemical reactions needed in different atmospheric conditions becomes 5 essential to give the best compromise between CPU time and accuracy. The purpose of this work is to assess the differences in NO and NO2 concentration considering three chemical approaches: a) passive tracers (non-reactive), b) the NOx−O3 photostationary state, and c) a reduced complex chemical mechanism based on 23 species and 25 reactions. The appraisal of the effects of chemical reactions focuses on studying the NO and NO2 dispersion in comparison with the tracer behavior within the street. In turn, the effect of including VOC reactions is also analysed taking into account several VOC/NOx ratios of 10 traffic emission. Given that the NO and NO2 dispersion can also be affected by the atmospheric conditions such as wind flow or the background concentration from season-dependent pollutants, in this work, the influence of wind speed and background O3 concentration are studied. Results show that the presence of ozone in the street takes an important role in NO and NO2 concentration. Therefore, greater differences linked to the chemical approach used are found with higher O3 concentration and faster wind speed. This bears relation to the vertical flux as a function of ambient wind speed, since it increases the pollutants 15 exchange between the street and the overlying air. This detailed study allows to ascertain under which atmospheric conditions the inclusion of chemical reactions are necessary for the study of NO and NO2 dispersion. The conclusions can be applied to future studies in order to establish the chemical reactions needed in terms of an accurate modeling of NO and NO2 dispersion and the CPU time required in a real urban area.

only the fastest chemical reactions have an impact on pollutants concentration. So some slow reactive compounds like CO or hydrocarbons can be considered as practically inert species at microscale. However, the NO 2 dissociation in presence of light and the NO titration by reduction of O 3 take place rather fast (Vardoulakis et al., 2003). And besides, Volatile Organic Compounds (VOCs) are also involved in this complex chain of reactions.
Currently, modeling urban air quality using a computational fluid dynamics (CFD) model is a big challenge. Besides pollu-5 tants dispersion, the simulation of chemical conversions of reactive pollutants in complex urban areas increases the computational time considerably. To better understand the problem, it is firstly necessary to thoroughly study the basic dynamical effects controlling non-reactive pollutants dispersion in simplified geometries. Many previous studies have investigated the main factors that affect the passive tracers distribution within the canopy such as the inflow conditions (Tominaga and Stathopoulos, 2010;Kim and Baik, 2004), the street-canyon aspect ratio (Chang and Meroney, 2003), thermal effects of a heating surface 10  or the vegetation location in the street (Buccolieri et al., 2011).
However, the main pollutants in an urban area are largely reactive compounds and for that reason, the photostationary scheme was included in an attempt to examine the dispersion of pollutants in street canyons. Baker et al. (2004) and Baik et al. (2007) studied the dispersion of reactive pollutants in a street canyon considering the NO x -O 3 photostationary steady state. Baik et al. (2007) also incorporated a heating surface at the bottom of the street and found that the magnitude of the chemical term for 15 O 3 was comparable to the advection or turbulent diffusion terms, but that was not the case either for NO or for NO 2 . Both researches showed that the O 3 concentration was faster depleted within the canopy because of the high NO emission at ground level. These studies underlined the importance of including the O 3 photochemical reaction in order to analyze the NO and NO 2 dispersion.
In addition, the VOCs related to traffic emission are also involved in urban air pollution. This becomes necessary to model 20 a vast number of chemical interactions NO, NO 2 , O 3 and VOCs. In more recent studies, complex chemical mechanisms have been implemented in CFD models in order to reproduce the NO and NO 2 dispersion in a street canyon. Kwak and Baik (2012) developed a CFD model coupled with the carbon bond mechanism IV (CBM-IV) and they evaluated the O 3 sensitivity with respect to different VOC and NO x emission levels in a street canyon. They found that the O 3 removal from the NO reaction was more important than the NO 2 photolysis given by the continuous and high NO emission from vehicles, which causes that the 25 O 3 behavior had a negative correlation with NO x emission level. Likewise, the O 3 concentration was weakly correlated with the amount of VOC emission. Kwak et al. (2013) examined the dispersion and photochemical evolution of reactive pollutants in a street canyon with different canyon aspect ratios. It confirmed the relation O 3 with NO x and VOC emission levels. And they found a slight influence of wind speed whereby increasing the wind speed enhances the exchange with overlying air which strengthens the downward O 3 transport. These recent studies highlight the importance of the O 3 oxidation process as 30 well as the OH oxidation process in describing the photochemical reactions in the street. Park et al. (2015) analyzed different aspect ratios and several VOC/NO x emission scenarios in three-dimensional street canyons. The main effects of NO x and VOC emission levels in O 3 concentration are similar to the results obtained in two-dimensional studies. In contrast, small variations are locally found owing to different flow patterns generated by the geometry of the domain.
To further analyze the sensitivity with the chemical mechanism used in a CFD model, Kim et al. (2012) compared the hourly concentration of NO, NO 2 and O 3 with the photostationary steady state and a full chemical mechanism (110 species and 343 reactions) with identical conditions. The results revealed similar values of NO and NO 2 concentration, whereas the O 3 concentration with the simple mechanism was lower than that obtained with the full scheme. Bright et al. (2013) compared the O 3 -NO x chemical system with a full chemical mechanism (51 chemical species and 136 reactions) using a LES model. 5 They studied in depth the effects of VOC chemical processes and found that by means of a more detailed chemistry, the values of NO, NO 2 and O 3 differ from the photostationary steady state, which reflects additional transformation of NO into NO 2 resulting from the intrinsic VOC degradation processes.
The coupled behavior between chemistry and dynamic processes is also a research goal. The relative importance of chemistry versus turbulent transport have been also investigated for homogeneous surface layers in the past (Galmarini et al., 1997;10 Molemaker and Vilà-Guerau de Arellano, 1998). The relative differences between turbulent and chemical time scales were evaluated using the Damkholer number. However, we estimate that this approach is not appropriate in the urban canopy layer (UCL). Basically because the transport is not only turbulent (as it is in the homogeneous surface layer), but also driven by the mean motions induced by the presence of the obstacles. Instead of adding a new time scale -linked to the mean transportand trying to deal with the strong heterogeneity of the flow and associated time scales typical of the UCL, we found it more 15 straightforward to perform simulations with and without chemical reactions, and to assess the relative importance of chemistry based on the difference between the simulations. In turn, regarding to the chemical scheme used, the best compromise between accuracy and CPU time required is also part of the main conclusions, since a complex chemical scheme implies more than twice the CPU time needed for the photostationary steady state. Therefore, the aim of this work is to determine in which conditions result essential to incorporate more chemical reactions 20 in order to model the dispersion of NO and NO 2 in the urban canopy. To achieve this goal, the NO and NO 2 dispersion is studied in two simplified geometries (2D and 3D geometries) under different chemical approaches: (a) non-reactive, (b) with photostationary steady state and (c) with a reduced complex chemical scheme (25 reactions and 23 species). In addition, knowing the impact of atmospheric conditions on reactive pollutants dispersion, the effects of the background O 3 concentration, the wind speed and the VOC/NO x emission ratio are jointly assessed. 25

Chemical Schemes Used
The evaluation of NO and NO 2 dispersion regarded as reactive pollutants is here studied using two chemical schemes: the photostationary steady state and a reduced complex chemical scheme (CCM-CFD). The photostationary steady state is the most simple chemical mechanism implemented in the CFD and it is defined by a three-reaction system of NO, NO 2 and O 3 .
Where M represents a molecule that absorbs excess energy and thereby stabilizes O 3 molecule formed (Seinfeld and Pandis, 1998). In the photochemical reaction, the photolysis rate constant is dependent on zenith angle (θ) and it is defined as J N O2 = A exp(B/cos(θ)), where A and B are chemical constants.
The complex chemical scheme used, which involves the VOC interactions, needs to be as reduced as possible due to the large load of computational time required to simulate a vast number of chemical reactions using CFD modelling and the number of 5 simulations needed to carry out this detailed study. The aim was to reduce the Regional Atmospheric Chemistry Mechanism (RACM) -with 77 species and 237 reactions (Stockwell et al., 1997)-as far as possible without losing much accuracy during the first 30 minutes of simulation for urban conditions in a boxmodel. The CCM-CFD was created applying the software CHEMATA (CHEmical Mechanism Adaptation to Tropospheric Applications) (Junier et al., 2005;Kirchner, 2005). A wide set of different urban pollution conditions was investigated and special emphasis was given to cases where the results for the 10 RACM mechanism differ considerable from the results of the photostationary steady state. Three main manners of mechanism reduction are applied: -Modification of the organic peroxy radical parametrisation: instead of using the 20 RACM RO2 species, we applied the approach described by Kirchner (2005) with the RO2 species CH3O2 (called MO2) and RC(O)O2 (called ACO3) as well as the two operator species XO2 and XO2N which work in the following way. The real reactions: are parameterised by; and all its reactions. Also, low reactive species as CH 4 and H 2 can be eliminated under urban conditions.

35
The resulting CCM-CFD mechanism consists on 23 species and 25 chemical reactions (Table 1). Compared to the photostationary steady state, the CCM-CFD includes the peroxy radical production from VOCs, CO and SO 2 as well as NO x loss processes by the formation of PAN, HNO 3 and organic nitrates. Whereas the peroxy radical production increases the NO 2 /NO ratio as therefore the NO 2 concentrations, the NO x loss processes decrease NO x and therefore NO 2 . As well, O 3 loss by the reaction with olefins is considered. Therefore, the effect on NO and NO 2 concentration after including the O 3 −NO x reactions 5 can be quantified with the deviation from the non-reactive pollutant concentration. And besides, the influence of including reactions involving VOCs can be obtained by the differences on NO and NO 2 concentration between using the CCM-CFD and the photostationary steady state.

10
The CFD model used is based on the Reynolds-averaged Navier-Stokes equations (RANS) with a k-ε turbulence model. This model allows to evaluate the effect of several parameters using a wide set of simulations within a reasonable CPU time. The turbulence can be solved more accurately by other models such as Large Eddy Simulation or Direct Numerical Simulation, however the CPU load increases considerably and it would limit the number of simulations. In this work, the CFD-RANS model is STARCCM+ (from CD-ADAPCO). It has been used in previous researches in order to study dynamical processes, 15 thermal forcing and passive tracers dispersion in simplified and real urban geometries Martilli et al., 2013;Santiago et al., 2014).
The NO x -O 3 photostationary state and the complex chemical mechanism described in previous section are implemented in the CFD model, in a way that the chemical term is added to the transport equation for each pollutant. The chemical term is defined as formation and depletion of a compound in the chemical reactions and modeled as a production rate in the transport 20 equation for each chemical specie ([∆C] Chem ). The traffic emission is added like a source term (S Ci ). Hence the transport equations are given by, Where C i is the concentration of the ith specie and D and K c are molecular diffusivity and eddy diffusivity of pollutants, respectively.

Simulation Setup
The distribution and the residence time of pollutants in the streets are mainly affected by the flow defined as a result of buildings configuration. The pollutant dispersion is analyzed in two types of geometries with the purpose of understanding the behavior of reactive pollutants in the street and the dependence on geometry. Figure 1 shows the computational domains used: a single street-canyon (2D-geometry) and a staggered array of cubes (3D-geometry). The domain sizes of street-canyon and staggered 30 array of cubes are 24x40x64 m and 64x64x64 m in x, y and z directions respectively. The aspect ratio defined as the ratio of building height (H) to the street width (W) is H/W = 1 in both geometries. The top of the domains is fixed at 4H based on the results obtained in Coceal et al. (2006). The three-dimensional array of cubes is represented by staggered cubes with buildings width (L), resulting a packing density λ = 0.25. A grid independence test was carried out and the mesh resolution was selected to 1 m in all directions.
Symmetry boundary conditions are assumed in the spanwise direction (y-direction) and cyclic boundary conditions are 5 imposed in the streamwise direction in order to simulate an infinite number of streets. The flow is driven in x-direction by a pressure gradient equal to ρu 2 τ /4H, where u τ is a reference velocity. Two velocities are studied in this work, u τ =0.45 m s −1 and u τ =0.23 m s −1 . For u τ =0.23 m s −1 , the corresponding wind speeds at 1.5H are 1.9 m s −1 and 1.5 m s −1 in the 2D and 3D geometries respectively, and the double for the u τ =0.45 case. For flow and turbulence equations, symmetry conditions are established at the top of the domain that enforce parallel flow and zero normal derivatives for the dynamic variables (Santiago 10 and . The traffic emission sources are located at ground level in both domains (Fig.1). Knowing that the diurnal variation of traffic flow in an urban area registers two peaks of traffic around 08:00 LST and 20:00 LST and a valley around 13:00 LST, we focus on analyzing the NO and NO 2 dispersion in maximum emissions conditions at 08:00 LST. The traffic emissions are simulated considering a NO x emission of 0.5 g km −1 per vehicle (Baker et al., 2004;Baik et al., 2007) and the volumetric ratio of 15 NO and NO 2 emission is 10:1 (Buckingham, 1997). The NO and NO 2 emissions are fixed with rates of 112 µg m −1 s −1 and 17 µg m −1 s −1 respectively, which is equivalent to 930 vehicles per hour representative of medium traffic (Baker et al., 2004). In addition, the VOC traffic emissions are included on using the CCM-CFD. The emitted VOC/NO x ratio by vehicles is dependent on several factors such as kind of vehicle or traffic flow speed. For that, two VOC/NO x emission ratio are simulated: where k is the rate reaction constant and it is set assuming isothermal condition (T=298 K). With the purpose of evaluating

CFD Model Evaluation
Due to a lack of experimental data that can be used to evaluate the chemical mechanisms, the results from the CFD model are evaluated in two steps. Firstly, the dispersion in the canyon using the non-reactive case is compared with experimental 10 data (Meroney et al., 1996). Secondly, the implementation in the CFD model of the chemical terms as a part of the complex chemical scheme (CCM-CFD) is evaluated with the results from a chemical box model well-established (Junier et al., 2005;Kirchner, 2005).
The non-reactive pollutant dispersion is evaluated with wind tunnel experimental data (Meroney et al., 1996). The experimental system consists on a single 2D street canyon with H/W = 1. The wind blows perpendicular to the canyon in x-direction 15 and the emission source is located at the bottom of the domain. This scenario is similar to our 2D-CFD simulations with the exception of the cyclic boundary conditions imposed in streamwise direction. In the comparison with experimental data, the inlet concentration is dismissed from the concentration inside the canyon. We focus on comparing pollutant concentration from simulation results with the measurements of urban roughness experimental case at different points close to walls within the canyon (Fig. 2). The concentration at each measurement point is normalized with respect to the concentration at the point 7 20 within the street (C/C 7 ). The same normalization was used in Santiago and Martín (2008) to evaluate the CFD model results with same experiment. Comparison between experimental and computed normalized concentration is depicted in Fig. 3.
Overall, a good agreement is obtained in spite of the slight difference at point 4 and 13 where the concentration is lower than F AC2 = fraction of data that satisfy 0 Where n is the number of points, the O i are the measurements at each point, and O is the measurement mean. And P i and P are the computed values at each point and its corresponding mean. The values of FAC2=1, NMSE=0.030 and FB= 0.068 reveal a good fit between computed and experimental data with a slight underestimation. Therefore, the pollutant dispersion 5 is accurately simulated in base to the model acceptance criteria established for NMSE<1.5 and -0.3<FB<0.3 in Chang and Hanna (2005) and as a threshold of the correlation coefficient R>0.8 in Goricsán et al. (2011) (Table 3).
Secondly, the accuracy of solving a vast number of chemical reactions using the CFD model is evaluated with the chemical box model used to develop the CCM-CFD (Section 2). For this comparison, the CFD model was run as a box and the chemical evolution for the reactive species of the CCM-CFD is simulated during 30 min. Similar initial conditions and the related-traffic s is quantified using the quadratic relative differences given by, where C k and C k is the horizontal average concentration at several vertical levels with t s =0.1 s or 2 s and t s =1 s, respectively.
The results of the QD for NO, NO 2 and O 3 from the simulations with the time step 0.1 s and 2 s are enclosed in Table 4. By comparing these results, we establish the time step at 1 s in order to simulate the dispersion of reactive pollutants.

Results and discussion
The CFD model is used to simulate turbulent flow and dispersion of reactive species in two different geometries: a 2D street canyon and a 3D staggered array of cubes. The interaction between the atmosphere and buildings configuration induces complex flow patterns within the urban canopy and this induces a heterogeneous pollutant distribution in the streets. Herein the canopy height is regarded as the height of buildings (H). The ventilation of the street in a 2D-geometry is more reduced than 5 in a 3D-geometry and accordingly, pollutants reside longer in the street (Vardoulakis et al., 2003). Thus, the residence time of pollutants in the street is determined by building configurations and wind flow. The time that these pollutants remain within the street affects to the chemical interactions, and as a consequence, to the amount of concentration at pedestrian level. For that reason, the effects of including chemical reactions in the CFD simulations are evaluated in 2D and 3D domains. To facilitate the analysis, the concentration of the study pollutants are normalized using the emission area (A Em ), the reference velocity 10 (u τ ) and the source emission rate (Q) given by, This work is focused on analyzing NO and NO 2 dispersion within the street and their normalized concentration are referred as [NO] N and [NO 2 ] N . In addition to quantify the effect of considering chemical reactions, the deviation in concentration (δC) from the non-reactive compound behavior is computed as, where C N and C N (T ) are the normalized concentration of the pollutants regarded as reactive and non-reactive, respectively.

Ozone Influence on Reactive Pollutants.
The O 3 concentration is generally dependent on the photochemical regime but in the street, it is also affected by chemical interactions with the NO x and VOC levels owing to the VOC-NO x traffic emission ratio. Kwak and Baik (2012) analyzed the 20 effect of several VOC-NO x emission ratios with a background [O 3 ] fixed. In addition, with NO x -VOC fixed emission rates, the diurnal variation of NO x and the exchange of O 3 at roof level was studied in Kwak and Baik (2014  % and more than 100 % inside the street respectively, and besides the difference of including VOC reactions is remarkable. As In addition, the difference among chemical mechanisms is also influenced by VOC levels. The concentration of a non-reactive pollutant is inversely proportional to wind speed (Parra et al., 2010), but the non-linearity of chemical reactions induces changes in this behavior. The effect due to wind speed on NO and NO 2 dispersion is obtained considering two wind speeds changing the reference velocity (u τ ): Case 1 and Case 2 ( Table 2). As discussed in the previous section, the higher O 3 case is used in this section given that chemical reactions have more effect on NO and NO 2 concentrations below the canopy top. Figure 6 shows the δNO and δNO 2 for different values of wind speed in the 2D-geometry. In the faster wind speed case, the differences on normalized concentration are larger in comparison to the tracer behavior. Overall, for all chemical scenarios, [NO] N is reduced from the tracer around 18 % and 10% within the canopy with u τ =0.45 m s −1 and u τ =0.23 m s −1 respectively. 5 However, the [NO 2 ] N increases a factor 2.7 and 2 from the normalized tracer concentration for u τ =0.45 m s −1 and u τ =0.23 m s −1 , respectively. Therefore, greater influence of chemical reactions on NO and NO 2 dispersion is obtained with the faster wind speed since it increases the O 3 inflow to the street acting as a trigger between pollutant interactions. Note that for a tracer, the normalized concentration is proportional to concentration and inversely proportional to wind speed. However, the behavior of the normalized concentration of a reactive pollutant is not lineal with the wind speed.

10
To analyze the [NO] and [NO 2 ] within the canopy, the vertical profiles of horizontal spatial average of the normalized concentration in the 2D and 3D geometries are depicted in Fig. 7. Different vertical distributions of the normalized concentration are obtained for each geometry. However, the same deviation of the normalized concentrations including chemical reactions from its corresponding tracer is obtained in the 2D and 3D geometries with different wind speed. In the faster wind speed case, the normalized concentration tends to be further from the non-reactive normalized concentration, however the variation among 15 chemical scenarios is small. The [NO] N and [NO 2 ] N with CCM1/2 just differ from PSS around 1% and 7% respectively. In contrast, with lower wind speed, the [NO 2 ] N with CCM1/2 and CCM1/5 increases 17% and 2% respectively from the PSS.
In this case, the accumulation of NO x emissions within the street is greater, which contributes to quickly consume the [O 3 ], reducing the NO 2 production and the NO removal. Likewise, the VOC levels given by the increase of emissions in the street intensify the NO 2 production and NO depletion, developing the opposite effect. This means that the concentration from re-20 active pollutants is higher with faster wind speeds since it produces a greater exchange of pollutants between the street and overlying air. Whereas when the wind speed is reduced, the difference from tracer behavior is reduced but, the importance of reactions involving VOC increases in comparison with just regarding the NO x − O 3 system.

Vertical transport
The vertical transport of NO, NO 2 and O 3 is analyzed in both geometries in order to determine the atmospheric conditions as   respectively. However, in the higher O 3 conditions, the differences of the simple scheme and by including VOC reactions are greater and the dependency with the wind speed is significant. As it has seen before, despite of the deviation in comparison with tracer is larger with the u τ =0.45 m s −1 , the higher differences between chemical scenarios are obtained with u τ =0.23 m s −1 . The maximum differences on [NO] N and [NO 2 ] N are obtained with more VOC emissions, and their variation switch from 10 % to 20 % when the inflow velocity changes of u τ =0.45 m s −1 to the half. Therefore, the chemical reactions are strongly determined by the available O 3 of background and subsequently, they have an slight dependency with the wind speed.
With the objective to deepen on the exchange of ozone inflow into the street, the vertical fluxes are analyzed. Given that, 5 there are only small differences between both computational domains with the same aspect ratio H/W=1. The pollutant removal from the street and pollutant entrainment into the street are analyzed by means of vertical fluxes of NO, NO 2 and O 3 in the 3Dgeometry. The total vertical flux (F T,i ) is defined as a sum of vertical mean flux (F m,i ) and turbulent flux (F t,i ). The horizontal spatial average (<>) of F T,i of ith species is computed at several vertical levels every 1 m using the following equations.
Here w and w are the vertical velocity and its fluctuation respect to the mean, w. And C i and C i are the corresponding variables to the concentration of ith species. To examine the atmospheric parameters effect on NO and NO 2 regarded as reactive pollutants through vertical fluxes, the NO and NO 2 fluxes are normalized following the method developed for a 15 passive tracer in Martilli et al. (2013). For a passive tracer, taking into account an homogeneous emission at surface, the steady state and neglecting vertical variations of air density with height, the following holds, within the canopy (Eq. 13) and above (Eq. 14).
< w C > + < w C > = S (13) Where S is the constant emission flux at the surface and A street and A roof are the areas of the streets and roofs of buildings respectively. Hereafter the spatial horizontal average of total fluxes of NO and NO 2 are normalized with S (within the canopy) and S A street /(A street + A roof ) above, referred as < w NO > N and < w NO 2 > N . Figure 9 illustrates the vertical profiles of the deviation of < w NO > N , < w NO 2 > N from the normalized total flux of tracer 25 and the vertical profile of < w O 3 > spatially averaged for the CCM1/2 and PSS scenarios in the 3D-geometry. The CCM1/5 results are not shown since their results are quite similar to PSS results. The < w NO > N and < w NO 2 > N taking into account chemical reactions are compared with the normalized flux of tracer. The vertical transport of a tracer above the canopy top is only turbulent, given that the mean vertical velocity is zero, but within the canopy in the middle part, it is mainly dominated by mean flux. This behavior is obtained for NO and NO 2 regarded as reactive pollutants as well. Unlike the tracer fluxes, the 30 turbulent fluxes of NO and NO 2 are affected by chemical interactions, mainly with O 3 . In Fig. 9, we can observe that the effect of the chemical reactions involving VOC (with the VOC/NO x =1/2) on vertical transports of NO and NO 2 are significant in the higher [O 3 ] and faster wind speed case. In this case, the < w NO > N and < w NO 2 > N are slightly smaller and larger respectively, corresponding to a greater vertical transport of O 3 in all heights and more photochemical activity. In contrast, in the other cases, the influence of introducing chemical reactions on < w NO > N and < w NO 2 > N just can be observed above the canopy top whereas inside the street is quite similar to the tracer flux. For a better understanding of the relative importance of including more chemical reactions and so to characterize the NO and In Fig. 10 In summary, the impact of the chemical reactions below the canopy top is affected by the O 3 entrainment within the canopy. The purpose of this work is to optimize the modeling of NO and NO 2 dispersion with respect to chemical reactions over a real urban area. Accordingly, an analysis of the relative importance of chemical reactions respect to mean and turbulent transport has been carried out by means of a CFD model. This is based on three different model set-ups: a simulation where chemical reactions are not considered (passive tracer), one where only the NO x − O 3 cycle is simulated (Photostationary steady state), 5 and the last one where the reactions involving VOC are also implemented (a Complex Chemical Mechanism reduced for the CFD model). With normalized concentration the source emission dependency is dismissed, and the deviation with respect to tracer is only due to the chemical reactions of pollutants with each other. As well, this helps to extract the conclusion in the 2D and 3D configurations used without dependence on geometry.
As it can be seen in this study, the general conclusions linked to the use of chemical reactions are obtained from both 10 geometries. With high available ozone in the street, the difference between chemical schemes is intensified by increasing the amount of VOC emission. The lower wind speed conditions encourages the increment of VOC within the canopy and this bears relation to the NO reduction and NO 2 production, the canopy and the chemical interactions increase the NO conversion into NO 2 .
Likewise, the formation of O 3 related to the photolysis of NO 2 along with the O 3 concentration of background increase the difference on the [NO 2 ] N considering the VOC reactions with respect to the simple chemical mechanism. 25 Nowadays, the big concern about air quality in big cities is carrying out many studies about dispersion of the primary pollutants related to traffic emissions. It focuses on NO 2 since it is the most detrimental for citizens health. To simulate its dispersion in a real urban area is a big challenge from the perspective of modeling at microscale with a CFD model. To obtain the best compromise between accuracy and the CPU time required in order to simulate the NO 2 dispersion within the streets, this work seeks the chemical reactions needed as a function of several atmospheric conditions. In winter conditions when the 30 O 3 concentration within the canopy is usually below 5 ppb and the solar radiation is low, the implementation of NO       Table 3. Statistic metrics obtained for the validation of CFD results with experimental data (Meroney et al., 1996) and the acceptance criteria proposed by (Chang and Hanna, 2005) for urban configuration.