Science of the Total Environment 454–455 (2013) 61–72
Contents lists available at SciVerse ScienceDirect
Science of the Total Environment
j ourna l homepage: www.e lsev ie r .com/ locate /sc i totenv
A computational fluid dynamic modelling approach to assess the representativenessof urban monitoring stations
Jose Luis Santiago, Fernando Martín ⁎, Alberto MartilliAtmospheric Pollution Division, Environmental Department, CIEMAT, Avda. Complutense 40, Ed. 3, 28040, Madrid, Spain
H I G H L I G H T S
• Methodology to reconstruct NOx maps around urban air quality stations using CFD simulations• Application of the methodology to two urban air quality stations in Spanish cities• Evaluation of the spatial representativeness (SR) of urban air quality stations
Article history:Received 3 December 2012Received in revised form 7 February 2013Accepted 22 February 2013Available online 26 March 2013
Keywords:Urban air qualityStation representativenessRANS-CFD modelStreet-canyon modelling
Air quality measurements of urban monitoring stations have a limited spatial representativeness due to thecomplexity of urban meteorology and emissions distribution. In this work, a methodology based on a set ofcomputational fluid dynamics simulations based on Reynolds-Averaged Navier–Stokes equations (RANS-CFD)for different meteorological conditions covering several months is developed in order to analyse the spatialrepresentativeness of urban monitoring stations and to complement their measured concentrations. Themethodology has been applied to two urban areas nearby air quality traffic-oriented stations in Pamplonaand Madrid (Spain) to analyse nitrogen oxides concentrations. The computed maps of pollutant concentra-tions around each station show strong spatial variability being very difficult to comply with the European leg-islation concerning the spatial representativeness of traffic-oriented air quality stations.
In urban areas, air quality assessment is usually based onmeasuredpollutant concentrations from networks of urban monitoring stations.It is based on the assumption that the pollutant concentration in aregion around the station does not differ significantly than the con-centrationmeasured at the station. The European Air Quality Directive(EC/2008/50), for example, specifies that “…a sampling point must besited in such a way that the air sampled is representative of air qualityfor a street segment no less than 100 m in length at traffic-orientedsites…”. However, the complex air flow patterns caused by theurban morphology (streets and buildings), and the irregular spatialdistribution of traffic emissions, generate strong spatial gradients inthe concentration fields inside the urban canopy layer. As a conse-quence, the scientific questions that motivate this paper are:
Can a point measurement be representative of the air quality in acertain urban area (streets, squares or district)? Is there amethodol-ogy to estimate the representativeness of an urban monitoring sta-tion? More broadly, how can we link the concentration measured
ín).
rights reserved.
in a certain point with the 3D field of concentrations or with the2 m concentrations (air breathed)?
These questions are very relevant if we want to know the qualityof the air breathed by citizens. Moreover, they will allow clarifyingwhether an urban air quality station can be really representative asstated by the Directive or not. This study is addressed to traffic sta-tions where the highest values of concentration are found.
European projects have devoted efforts to tackle the station repre-sentativeness question, for example, the European project AIR4EU de-veloped between 2004 and 2006 or presently FAIRMODE (Forum forair quality modelling in Europe), in which there is a specific workinggroup dealing with the combination of measurements and modelsand spatial representativeness of air quality stations.
An answer to the station representativeness questions can beobtained by organizing specific measurement campaigns with a largeamount of passive samplers deployed around a monitoring station dur-ing weeks or months. The advantage is that these samplers are cheaperand smaller than the standard monitoring station itself, and can beinstalled easily. The disadvantage is that they can provide only longterm concentration averages (over weeks or months) (Krochmal andKalina, 1997; Liu et al., 1997; Galan Madruga et al., 2001; Parra et al.,2009). In addition, wind tunnel experiments can be carried out to
62 J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
investigate the representativeness of monitoring stations, as done byRepschies et al. (2007) for wind measurements in urban areas.
Alternatively, there are methodologies based on the use of modelsand/or some surrogate indicators related to emission sources distribu-tion. For example, Janssen et al. (2008, 2012) have used land use datato take into account the local influence of the air pollutant concentra-tionsmeasured at specific stations, in theirmethodology for air qualityassessment in Belgium. Other methodologies are based on climatic-topographic criteria, which can be recommended specially for ruralbackground stations (European Commission, 2011). Spangl et al.(2007) made a very complete review of the criteria and methods forair quality classification and representativeness estimate. Other au-thors have used models for estimating spatial representativeness ofair quality stations ranging from rural to urban stations, includingfew studies for stations located in streets, as those of Lohmeyer et al.(2005) or Schatzmann et al. (2006). Vardoulakis et al. (2005) usedparametric street-canyonmodels to check how they simulate the spa-tial distribution of pollutants near an urban air quality station whereexperimental campaigns were carried out. Other authors have usedmore complex models as computational fluid dynamic (CFD) tostudy the distribution of pollutants in streets, taking advantage ofthe high resolution of those models (Schlünzen et al., 2003; Parra etal., 2010; Santiago et al., 2010; Buccolieri et al., 2011, among others).
There is a consensus in the scientific community that CFD modelsare needed to simulate the complex flow and dispersion influencedby the presence of buildings (Schatzmann et al., 2010). The CFDmodels resolve explicitly the flow and pollutant dispersion aroundurban obstacles (building, trees…) on spatial domains with a size ofseveral hundreds of meters. They need a very dense computationalgrid with high resolution (order of meters or finer). There are severaltypes of CFD depending on the phenomena parameterized, that havedifferent computational requirements. The cheapest one in terms ofcomputational burden, solves the Reynolds Averaged Navier–Stokes(RANS) equations, and parameterizes all the turbulent motions. Itprovides steady-state simulations for fixed inlet wind and boundaryconditions, and in general it is considered accurate for the estimateof the mean concentration. More refined approaches are Large EddySimulation (LES) that parameterise small eddies (in general smallerthan the grid size) and resolves explicitly the largest turbulent eddiesor Direct Numerical Simulation (DNS) that resolves all the turbulentmotions. These last two techniques can provide not only the mean,but also the higher order statistics. However, the disadvantage isthat the computational time is significantly larger than for RANS (atleast a factor 100 for LES and even more for DNS). For many applica-tions, including the one proposed in this article, the mean is the mostrelevant information, and the RANS approach is a good compromisebetween accuracy and CPU time (Santiago et al., 2007, 2008, 2010;Parra et al., 2010).
In this paper, we propose amethodology based on CFD simulationsfor different inlet wind directions and several simple assumptions(non-reactive pollutants, negligible thermal effects, etc.) to estimatethe spatial representativeness of urban air quality stations. The advan-tage is that it can provide 3D fields of concentration with virtually anytime resolution (from hours to years). Concerning the type of CFDmodel used, two aspects must to be taken into account: CPU timeand accuracy. As commented previously, LES is able to reproducemore accurately the atmospheric wind flow, however, CPU time re-quired is much higher than for RANS (about two order of magnitude).In this work, we are interested in the mean concentration fields and alarge number of CFD simulations (one for each wind direction, 16 intotal) are needed. Dejoan et al. (2010) simulated the concentrationfield measurements within an array of shipping containers (MUST ex-periment, Biltoft, 2001) using LES and RANS. The differences betweenLES and RANS and experimental concentration were partly explainedby small fluctuation of inlet wind direction. Taking into account thatfor the methodology proposed in this study we use only a discrete
number of CFD simulations (16 inlet wind directions) and the evalua-tion of spatial variability carried out of the whole methodology usingRANS as CFD model by Parra et al. (2010) for a large period of time(explained further on in Section 3), a RANS model is selected to runthe set of simulations with different wind directions. This methodolo-gy (described in Section 2) is an extension of the methodology pro-posed in Parra et al. (2010), and it has been applied to two locationswith very different characteristics: a monitoring station located in acentral part of a small Spanish city (Pamplona, North of Spain, (Loca-tion 1)) and a station located close to a big park in Madrid (Location2). Results are presented and discussed in Section 3. Conclusions arein Section 4.
2. Methodology
For the purpose of this study (representativeness of urban monitor-ing stations),maps of time averaged pollutant concentration nearby theurbanmonitoring station are needed. Frequently, air quality limit valuesare associated to large periods of time and therefore pollutant concen-trations should be averaged for large period of time. In addition, thesemaps should have spatial resolution high enough (~m) to catch thestrong spatial heterogeneities in the distribution of pollutants insidethe streets. Taking into account that it is not possible (within a suitableCPU time) to run an unsteady CFDmodel for these large periods of time,the solution proposed is to run with a steady CFD model only a set ofinlet wind directions and use a numerical combination of these to com-pute the final results. This methodology was developed by Parra et al.(2010). A brief description and the modifications made in this paperare explained in the next sections, together with a description of thetwo locations investigated. This study is focused to nitrogen oxides con-centration around two traffic-oriented air quality stations in Pamplonaand Madrid.
2.1. Modelling approach
Numerical simulations were based on the steady state Reynolds-averaged Navier–Stokes equations (RANS) and k–ε turbulence modelsusing STARCCM+ (from CD-Adapco) code. The pollutant concentra-tion was simulated using additional transport equations for passivescalars. More information about the models and a discussion aboutthe selection of RANS turbulence scheme can be found in Santiagoet al. (2007) and Parra et al. (2010). The particular modelling setupsfor the two locations are explained in the next section.
The wind flow and the dispersion of several passive tracers insidethe urban zones studied is simulated for 16 different inlet wind direc-tions, i.e. from 0° to 360° with an increment of 22.5° (N is for 0°, NNEfor 22.5° and so on). The notation used is sector 1 for N, sector 2 forNNE and so on, clockwise. The passive tracers represent traffic emis-sions and they are all released at the same rate from line sources indifferent streets (one passive tracer for each street). The total concen-tration Ctotal(t) at hour t in a certain point, is the sum of the concentra-tion of each tracer (i), corrected to account for the emissions andwind speed of this hour, and it can be written as:
Ctotal tð Þ ¼ M1
vin tð Þ∑iCi Sector tð Þð Þ⋅ Li
VsourceiNi tð Þ ð1Þ
where,
- Sector(t) is the wind direction sector at hour t.- i indicates the tracer emitted in street i.- Ci(Sector(t)) is the concentration computed for Sector(t) for a givenemission from street i and for a reference inlet wind speed.
- Li is the length of the street i.- Vsourcei is the volumeof the rowof computational cellswhere emis-sion of the street i is located.
63J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
- Ni(t) is the number of cars per unit time running in street i at hour t.- vin(t) is the wind speed measured at hour t in a nearby station.- Μ is a normalizing factor, calculated to ensure that modelled andmeasured average concentrations over the whole period are thesame.
The measured meteorological conditions and traffic data at eachhour are used to select one of the 16 simulations (based on the winddirection) previously run and to compute the values of the followingfactors: Ni ⋅ Li/Vsourcei and 1/vin (Eq. (1)). Note that as in these casesthe emissions were unknown, the tracer emissions were consideredthe same in the CFD simulations inside each street and for this reasonCi(Sector(t)) needs to be multiplied by Ni ⋅ Li/Vsourcei. Therefore, inthese cases, M is a normalizing factor computed as the ratio betweenmeasured concentration averaged over a large period of time (severalmonths) and computed concentration without the M factor averagedover the same period of time. More details are explained in Parraet al. (2010).
This approach is valid under the following assumptions made:
1) Pollutant reactivity should be small during the time period studied.2) Thermal effects are negligible in comparison with dynamical ef-
fects. This is because the simulations are performed for a neutralatmosphere, e.g. no heat exchanges are considered between build-ings or soil and the atmosphere. The inclusion of the thermaleffects may result in a significant increase of the simulations need-ed to characterize the dispersion.
3) Emissions inside each street at a selected hour are proportional totraffic intensity (number of vehicles) at that hour.
4) Concentration at a selected hour only depends on the emissionswithin the modelling domain and meteorological conditions atthat hour.
5) Concentration is proportional to 1/wind speed (Parra et al., 2010).Using this approach only one simulation for each wind direction isneeded.
For a big city like Madrid, and out of the winter months, assump-tions 2) and 4) are not completely fulfilled. For this reason, somemod-ifications to this methodology (explained in the next sections) wereintroduced to take into account weak winds and background concen-tration in big cities.
2.2. Cases studied: description and modelling
Two urban locations have been considered.
2.2.1. Location 1. A square in Pamplona (Spain)Inside this square, there is a traffic-oriented air quality station
(Fig. 1). This is the location of the study of Parra et al. (2010) wherethemethodologywas successfully evaluated first. Pamplona is amedi-um size city of Spain (200,000 inhabitants) and the zone of the study isa rectangular square (120 mwide) surrounded by 15 m-height build-ings. The streets around the square are 10 m wide (Fig. 1). Measure-ments of wind speed and direction were provided by the RegionalGovernment of Navarra from a meteorological station located in anopen area nearby the simulated area (Parra et al., 2010).
The domain is shown in Fig. 1. The geometry of the buildings wasmodelled using an irregular mesh of 3.5 · 106 computational cells.All buildings have the same size and each one is resolved with atleast 10 grid points per direction. A test of the grid-independence ofthe results has beenmade and it showed that this resolution is accept-able. Boundary conditions at buildings and ground surfaces aremodelled by means of standard wall functions. Inlet profiles of windspeed are logarithmic (Richards and Hoxey, 1993). Traffic inside 12streets around the square is modelled bymeans of emission of passivescalars in rows of numerical cells close to the ground along the streets.Hourly traffic data inside each street without discriminated by vehicle
type was provided by local authorities. As explained in the previoussection, the emissionswere unknown and the comparisonwith exper-imental data was made by means of a normalization (M factor inEq. (1)). A detailed description of numerical domain, mesh andboundary conditions can be found in Parra et al. (2010). The time pe-riod used in this study is January–February 2007.
The methodology was successfully applied to this location byParra et al. (2010) for January and February 2007 as the time evolu-tion of hourly mean NOx and PM10 concentration estimated by thismethodology fit well the observations from an urban monitoring sta-tion. The highest differences were found for low wind speed hours. Inaddition, experimental two-weeks averaged concentrations (usingpassive samplers) of NO2 and BTEX pollutants in three locations (in-side high, medium and very low traffic streets) from four samplingcampaigns (Parra et al., 2009) were also compared with model results(details in Section 3).
2.2.2. Location 2. A street intersection close to an urban park in Madrid(Spain)
In this zone there are several streets and avenueswith intense roadtraffic and also there is a large green urban area (El Retiro). Buildingswith different heights are located in this area; most of them have aheight between 18 m and 24 m, although the tallest building is 90 mhigh approximately. The traffic-oriented air quality station (EA) is lo-cated in a garden very close to the sidewalk in the intersection of thetwo more important avenues (Fig. 2). Meteorological data were pro-vided by AEMET (Spanish Meteorological Agency) from the meteoro-logical station located in El Retiro (the urban park of location 2 atSouth of EA station) but outside of the numerical domain (Fig. 2).The wind rose for the time period studied is shown in Fig. 3.
The size of the numerical domain was 700 m × 800 m, approxi-mately (Fig. 2). Similar boundary conditions to the previous locationwere used. Also, the traffic was modelled in a similar way. However,hourly traffic data inside each street was not available, but this informa-tion was estimated from the daily average traffic intensity (TI) insideeach street (provided by the city council of Madrid) multiplied by thetime evolution of traffic for different types of days (TF) (obtained fromPalacios, 2001). In this location, 29 passive tracers were released (onefor each street). An irregular mesh of 3 · 106 grid points with a resolu-tion of about 1 m–3 m close to the buildings is used. A test about gridindependence was made, and it showed that this resolution is accept-able. The time period used in this study is January–May 2011.
In this zone, there is an urban park that produces not negligible dy-namical effects on pollutant dispersion. These dynamical effects of thevegetation are modelled considering trees as porous medium. In thisway, sources/sinks terms in the momentum, turbulent kinetic energyand turbulent dissipation rate equations were added in the numericalcells with vegetation. Inside the park, vegetation is present in the nu-merical cells located from the ground to 18 m-height (trees of 18 m)but for the trees within the street only the cells located at the crownlevel have vegetation. Note that the vegetation modelling proposedwas tested against data from Raupach et al. (1987) experiments overforest-clearing–forest configurations. A suitable agreement wasobtained. In Appendix A, the procedure on how dynamical effects ofthe vegetation were taken into account is described.
The air quality station studied in Madrid (location 2) is different insome aspects to the Pamplona case (location 1) and some general as-sumptions made in the methodology are not fulfilled for this location.Firstly, Madrid is amuch bigger city (more than 3.5 million inhabitants)than Pamplona (200.000 inhabitants), and it is expected to have a sig-nificant urban background concentration. In addition, the domain cho-sen for the simulation is surrounded (except on the South side), bydensely populated neighbourhoods with elevated emissions levels,that are expected to influence the inflow boundary conditions. More-over, the period studied is not onlywinter, as for Pamplona, but extendsuntil spring (May), when the solar radiation, and consequently the
Fig. 1. a) Aerial view of the location 1 (circle (1) is the location of the urban monitoring station and one passive sampler and triangles (2–3) are the locations of the other two passivesamplers). The limits of the simulated domain are represented by the white line. b) Modelled geometry of simulated domain (4 lines represent the street categories corresponding to4 different traffic emissions). Solid and dashed lines are high traffic density streets, dotted line is medium traffic density street and dashed-dotted line is very low traffic density street.Note that, in this case, hourly traffic intensities in these streets are known and used.
64 J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
thermal effects, start to be significant in Madrid. Therefore, all these as-pects have to be taken into account in the Madrid case.
2.3. Modification of methodology to address urban background concen-tration and weak winds in big cities
Concentration at an urban monitoring station is the sum of the con-tribution of the emissions within the domain (simulated by CFD-RANS)and the contribution of the emissions outside the domain (that we willcall ‘background contribution’). This contribution is not negligible and itis difficult to estimate without a very dense network of measurementsin big cities. The background contribution depends on wind direction(transport of outer emissions), e.g. for the location 2, where the windcan come from a big urban park (with low emissions) or from a zonewith streets with dense traffic (high emissions). In an ideal situation,it could be estimated with stations just outside the numerical domainin every direction. However, it is not occurring here, except for the
Fig. 2. a) Aerial picture of the zone including the location of the urbanmonitoring station (dot)b)Modelled geometry (29 lines represent the street categories corresponding to 29 different tris indicated by shades (dark lines show high intensities).
South side. To illustrate the importance of background concentration,the urbanmonitoring station of location 1 is studied using the availabledata. The measurements used correspond to the hourly concentrationsof NOx (considered as passive) provided by two nearbymonitoring sta-tions: one is the monitoring station studied at location 2 (EA station)and the other is a station located south of the domain inside the park(Retiro station) (Fig. 2). Therefore, for cases of Southwindwith high ve-locity (V > 2 m s−1 at 10 m-height, this choice is justified below)mea-surements from the Retiro station can be considered as backgroundconcentration (e.g. representative of the inflow condition). In Fig. 4,the concentration measured at EA station and the concentration at EAminus the concentration at Retiro are fitted to amodelled concentrationproportional to,
∑iCi Sector tð Þð Þ⋅TIi tð Þ⋅ Li
Vsourcei⋅TF tð Þ⋅ 1
vin tð Þ: ð2Þ
and background station (triangle), the simulated domain is represented by the white line.affic emissions). Relative daily average of traffic intensity in themain streets of the domain
Fig. 4. Scatter plots (measurements vsmodel) of a) HourlyNOx EA and b)Hourly (NOx_EA–NOx_Retiro) for South wind cases with V > 2 m s−1.
Fig. 3. The wind rose for location 2 in the study period.
65J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
The number of vehicles inside each street at hour t is estimated asTIi(t) · TF(t), where TIi is the daily average traffic intensity inside eachstreet (provided by the city council of Madrid) and TF is the time evo-lution of traffic during the day for different types of days (obtainedfrom Palacios, 2001).
In order to simplify the notation, all variables depending on streetare grouped and hereafter named CFD(Sector(t)),
Fig. 4 shows that removing the background concentration mea-sured at Retiro improves the fitting.
Unfortunately, for the other directions (different than South), thereis no air quality station that can give the background concentration andusing Retiro station for all directions does not work. Moreover, asshown in Fig. 5, even for South direction, when we consider also situa-tionswith lowwind speed, the correlation betweenmodelled andmea-sured valuesworsens. It is necessary, then, to develop amethodology toestimate the background concentration for different wind directions,and for low and high wind speeds.
In order to model concentration for the whole period, it is necessaryto estimate urban background concentration for every wind direction.Depending on the number of monitoring stations and available data,the urban background could be estimatedmore accurately, but for Loca-tion 2 only one background monitoring station can be used. Then, theassumption adopted is to compute the background depending on trafficevolution along the day (TF), 1/vint and h0/h, (h is themixing height andh0 is the mixing height at night) as follows,
Cbackg ¼ α⋅ TFv int
h0h
þ β ð4Þ
α and β are the fitted parameters. h is computed usingmeasurements ofradiation and temperature. A simple method has been used. Measure-ments of temperature and solar radiation are used to compute longwaveradiation (Idso and Jackson, 1969) and the sensible heat fluxes are
calculated following the hysteresismodel of Grimmond et al. (1991). Fi-nally, the computation of mixing height is made as (Arya, 2001),
h tð Þ ¼ h20 þ2 1þ Cð Þ
γ∫tt0
wθ� �
0dt� �1=2
ð5Þ
where h0 is the minimum (nocturnal) planetary boundary layer height(a value of 100 m is used, which is rather typical in urban areas),wθ� �
0 is the surface heat flux, C is a constant to parameterize the en-trainment heat flux as a fraction of surface heat flux (=0.2) and γ istemperature lapse rate (its value is unknown with the measurementsavailable, and a value of 5.5 · 10−3 K m−1 is used).
For Southwind cases (sector 9)with V > 2 m s−1, we can evaluatethis approach usingmeasurements from the Retiro monitoring station(assuming that there are no large emissions close to the station).
CRetiro≈Cbackg ¼ α⋅ TFvint
h0h
þ β: ð6Þ
Fig. 6 shows that this approach is suitable, e.g. the concentration atthis station can be represented as the sum of a constant term (β), plusa term directly proportional to traffic intensity and inversely to windspeed and mixing height.
Fig. 5. Scatter plots (measurements vs model) of hourly (NOx_EA–NOx_Retiro) for South wind cases with different wind speeds. Note that Fig. 4b shows the South wind cases forV > 2 m s−1.
Fig. 6. Hourly NOx concentration at Retiro station versus background concentrationmodelled for sector 9 (S).
66 J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
Therefore, concentration for everywind directionwithV > 2 m s−1
We assume that A is the same for every wind direction and takethe value obtained from sector South using the concentration at theRetiro station as background. For every wind sector (except for sectorSouth) α and β are computed by fitting concentration measured at EAstation versus model concentration given by Eq. (8) as a line with aslope close to 1 and an independent term close to 0. In other words,a set of values of α and β are used to compute the modelled concen-tration and a linear regression is performed between modelled andmeasured concentration. The values of α and β are selected takinginto account the optimum value of the correlation of the different re-gressions obtained. For sector South, α and β are taken from the linearfitting of Fig. 6 and the concentration modelled has a very good corre-lation in comparison with concentration measured at EA station(Fig. 7a). For the other wind sectors due to the approaches assumedand the uncertainties in some variables the correlation is lower butit can be considered suitable (Table 1). For example, for other sectorslike sector NE (predominant wind direction) the modelled concentra-tion (obtain as described above, Eq. (8)) also has a good correlationwith measurements (Fig. 7b).
As commented above, forweakwinds the situation ismore complex.Buoyancy effects are not negligible and some assumptions are not ful-filled. Even for south wind, the measurements of concentration at theRetiro station does not correspond to background concentration at the
EA station because the transport velocity is too small. In order tomake an evaluation of these low velocity cases, a simple parameteriza-tion is proposed introducing a factor γ(Sector(t)) (e.g. a different valuefor each sector, and not the same value as before with the factor A in(Eq. (8))), adapting the values of α(Sector(t)) and β(Sector(t)) for thebackground and introducing a lower limit of 1 m s−1 for the windspeed (see Eq. (9)). This is a very crude approximation, but it was suffi-cient to improve the results (without these changes therewas no corre-lation between modelled and measured concentration as shown inFig. 5), however future investigations should study in depth this topic.
Fig. 7. Modelled concentration using Eq. (8) versus hourly NOx concentration at the EAmonitoring station a) for wind sector 9 (S); b) for wind sector 3 (NE).
67J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
This modification is applied for cases with V b 2 m s−1.
C tð Þ ¼ CFD Sector tð Þð Þ⋅TF tð Þmax v int tð Þ;1ð Þ γ Sector tð Þð Þ þ α Sector tð Þð Þ⋅ TF tð Þ
max v int tð Þ;1ð Þh0
h tð Þ þ β Sector tð Þð Þ:
ð9Þ
As for V > 2 m s−1, for everywind sector, concentrationmeasuredat the EA station versus model concentration given by Eq. (9) is fittedto a line with a slope close to1 and an independent term close to 0 inorder to obtain a set of constants (γ, α, β) for each wind sector. Thefitting procedure is similar to that described above. The correlationvalues are shown in Table 2.
Table 1Fitted parameters and correlation values for every wind direction for V > 2 m s−1 cases. V
a Values from the fitting of Fig. 5.b Few data (only 1 value above 2 m s−1).
This methodology can be applied to other similar cases to take intoaccount the contribution of the emissions outside the domain.
3. Results
3.1. Evaluation
Before to analyse the pollutant distribution in location 2, an evalu-ation with the observed data at the air quality station was carried outfor hourly data of NOx for the simulation period (January–May 2011).Fig. 8 shows the scatter plot of NOx modelled and measured and thetemporal series of NOx modelled and measured for the time periodstudied. The large majority of data (93%) are within a factor 2 (dashlines). In addition, hourly concentration averaged for each month iscomputed and shown in Fig. 9. Highest differences are found in Febru-ary, this can be due to uncertainties on traffic data (the same trafficpattern is used for every month). However, in general, the results(Figs. 8 and 9) are quite good showing that simulated data fit wellthe observations in spite of the uncertainties of some input data likethe traffic or urban background concentration.
The results of the 16 CFD-RANS simulations (one for each winddirection)made for each location are combined following themethodol-ogy described above to obtain time series of hourly concentration. Aver-ages for the simulated periods were computed from the estimatedhourly concentrations. Note that the same EA station data is used forthe fitting and for the evaluation. This can be a limitation but it is neces-sary taking into account the available data. In addition, for the spatialrepresentativeness study shown here, this approach can be consideredappropriate because the model simulations are done to provide infor-mation about the gradient of pollutant concentrations in the urbandomain, which complements the measured concentration data in thetraffic-oriented station to give rise to a map of the pollution aroundthe station. Related to this, an experimental campaign using passivesamplers in the frame of the SERCA project (Lumbreras, 2012) was car-ried out close to Location 2within thenumerical domain analysed in thispaper. The concentrations gradients measured in this campaign were ofthe same order (a factor 2 between monitoring station and some pointsin other nearby streets) than those given in our study, despite the factthat the campaign took place in a different period.
In location 1, the temporal series of NOx modelled andmeasured forthe timeperiod studiedwas compared by Parra et al. (2010). In addition,the spatial variability of the mean concentration (over large periods oftime) provided by this methodology (including the combination of theCFD simulated meteorological situations), was evaluated by means ofa comparison against experimental data of two-weeks averaged NO2
concentrations from passive samplers in three points for Location 1(Fig. 1) performed by Parra et al. (2010). Fig. 10 shows the normalisedconcentration of NO2 in the 3 points (Fig. 1a, data from Parra et al.,2010). The agreement of modelling results with the experimental datais good and the spatial variability of the mean concentration in these 3points is captured. As commented previously, themethodology assumesthat modelled pollutants are non-reactive. NOx fulfils this criterion bet-ter than NO2. However, Parra et al. (2010) shown that, in this case, NO2
concentrations (NOxwas not available in the database of the campaign)are well reproduced probably because in the studied period (January–February) the reactivity of nitrogen oxides is not very high.
alues of fitted parameters are normalised by the values corresponding to sector 9.
Table 2Fitted parameters and correlation values for every wind direction for V b 2 m s−1. Values of fitted parameters are normalised by the values corresponding to sector 9.
68 J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
3.2. Representativeness analysis
To determine the representativeness of a monitoring station it isnecessary to define the concept in a quantitative way. We define therepresentativeness area (RA hereafter) of the monitoring station, asthe area where concentrations are within an interval of ±20% of theconcentration at the monitoring station. This criterion is based onthe legal error allowed for the measurements by the European AirQuality Directive (2008/50) and was used for the two studied loca-tions. The idea is that, if the pollutant concentration in any place dif-fers less than 20% of the concentrations at the air quality station,both places basically have similar pollutant concentrations.
3.2.1. Location 1Here we analyse only the representativeness of the traffic monitor-
ing stations, in Fig. 11a, the computed mean concentration (normalisedby concentration at station location) is shown in the square, where theurban air quality station is located. Strong spatial variability in the pol-lutant concentration is modelled in this area. There can be differences
Fig. 8. a) Scatter plot of hourly NOxmeasured andmodelled for the whole period studied.Straight line is fitted line and dash lines indicate a factor 2 of the fitted line. b) Temporalseries of hourly NOx measured and modelled for the whole period studied.
larger than a factor 3, between concentrations in some streets and con-centrations in the square. Areaswith line patterns representwhere con-centrations are within an interval of ±20% of the concentration at themonitoring station which is the RA. The RA has a donut shape coveringa portion of the square and small areas in few streets. It means that thisair quality station does not meet the requirement of representativenessstated the European Air Quality Directive, because it is not representa-tive of a segment of street of 100 m. This shape of RA does not corre-spond to a specific microscale condition but it is due to an averagesituation taken for the different wind directions taking into accounttheir frequency and the corresponding emissions. Note that, as emis-sions are unknown, the values of concentrations of NO2 or NOx comput-ed are obtained by means of the normalization using the valuesmeasured. In this case, the normalised maps are similar.
In order to check whether other locations are more representativeand comply the directive in terms of representativeness, hypotheticallocations of the new monitoring stations were also assessed based onthese results. For example, locations in another zone of the squareand in a nearby street were studied (Fig. 11b and c). It can be ob-served that the spatial extension and shape of the RA is strongly de-pendent on the position of the station. In this case, the location inthe street seems to be more representative of the air pollution inthe area than the locations in the square. In this case, the Directive re-quirement on representativeness is almost complied. Concentrationin the narrow streets is relatively homogeneous, whilst largest gradi-ents are present in the square.
3.2.2. Location 2The computed mean concentration in Location 2 (normalised by
concentration at station position) is plotted in Fig. 12. The averagednormalised concentration map shows a RA (white zone in Fig. 12)extending along the sidewalk close to the northern border of the park,a relatively small area around the station and some streets of less trafficintensity. There is a very remarkable variability in the pollutant concen-trations in the main streets (up to a factor of 2) in spite of the impor-tance of background concentration.
The RAs of this traffic station have some interesting features. TheRA covers the gardenwhere the station is located and the north borderof the park (south sidewalk of the avenue crossing west to east) alongwith some narrow streets. However, the north sidewalk of the W–Eavenue have much higher concentrations because the street scale cir-culations tend to transport and accumulate pollutant there. The con-centrations in most of the other avenues are also higher in spite thetraffic intensity is similar or lower than in the W–E avenue. This canbe explained by the fact that the urban park (El Retiro) covers alarge area without emissions from the W–E avenue to the south bor-der of the domain. In addition, this area with trees and without build-ings allows better ventilation in the W–E avenue and gives rise to amore effective pollutant dispersion. The concentrations in the narrowstreets are similar to those at the station due to the low emissions cor-responding to low traffic intensity. In this case, maps of modelled NOx
are analysed. Air Quality Directive holds limit values for NO2, but ourpurpose goes beyond the air quality directive and is to show the strongspatial variability of concentration of a generic pollutant around a traf-fic monitoring station, even for large time period averages.
Fig. 9. Hourly concentration averaged for each month. Bars represent the standard deviations.
69J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
3.2.3. An analysis towards the EC/2008/50 representativeness requirementsIn locations 1 and 2 the time period study is 2 and 5 months
respectively whilst the EU Air Quality Directive use a 1-year timeframe for some air quality limit values. However, the time periodsstudied are large enough and the maps do not represent only a partic-ular situation but an average situation of the representative conditionsof each zone, that is, the resulted maps would have the main featuresof spatial distribution of pollutants around the station for a longer av-eraging times. In the two cases studied, strong gradients of pollutantconcentrations were computed in the main streets and squares. Thisindicates that the concentration at the monitoring station, even ifthe point complies with the representativeness requirements of Euro-pean Air Quality Directive (EC/2008/50) (something that can be diffi-cult to obtain), is not enough to characterize the air quality in the area.
In the case of Location 1, model results indicate that there are largeconcentration gradients inside the squaremaking that the actual posi-tion of the traffic monitoring station does not comply with the repre-sentativeness requirements of European Directive (there is not asegment of 100 m with concentrations within +/−20% of the valueat the monitoring station). More importantly, none of the points inthe square has this characteristic. On the other hand, the largest repre-sentative area is obtained when the traffic-oriented station is locatedin one of the narrow streets close to the square. This homogeneity inthe pollutant concentration is due to the fact that all the narrowstreets share the same conditions (emissions, size, etc.). In this case,the representativeness requirements of the Directive are almost met.However, it is likely that people will spend much more time outdoorin the square, where there are recreational areas and shops, than in
Fig. 10. Comparison between experimental andmodelled two-week averaged normalisedconcentrations of NO2 (location 1, data from Parra et al., 2010).
Fig. 12. Mean concentration map normalised by concentration at station location 2(represented by a star) for NOx. White zone shows the area with concentrations into±20% around the station concentration.
70 J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
the streets nearby, that have small sidewalks. Model results indicatethat concentrations in the square are a factor 2–3 lower than in thestreets. In this case, then, moving the station to a street will increasethe extension of the RA, but it is likely that the concentration mea-sured will be less representative of what the majority of people reallybreathe.
In the case of Location 2, the highest concentrations are in the ave-nues with intense vehicle traffic near the station site. Strong differences(more than a factor 2) in concentration between sidewalks of the sameavenue are estimated. In this case, the RA of the station ismostly limitedto the sidewalk close to the park and some other low traffic streets. In
Fig. 11. Mean concentration map normalised by concentration at station location 1(represented by a circle). a) Present location of the urban quality station. b) Hypothet-ical location of urban AQ station in other square corner. c) Hypothetical location ofurban AQ station in a street close to the square. The axis units arem. Areas with line pat-terns represent where concentrations are within an interval of ±20% of the concentra-tion at the monitoring station (RA).
this case, the very large concentration gradients inside the avenuesmake it very difficult to meet the Directive requirements becausethere are important differences between sidewalks of the same avenue.However, some low traffic narrow streets have 100 m segments insidethe ±20% interval, but those are not connected with traffic stationlocation.
However, within the small simulation domain (less than 1 km2),there are other points with an RA with extension of 100 m or more(e.g. they comply with the representativeness requirements of Direc-tive), and concentrations significantly different (up to a factor 2) thanthe concentration at the actual monitoring point.
4. Conclusions
In this paper, a methodology to estimate the representativenessarea of traffic-oriented air quality stations has been proposed and ap-plied to two cases in Pamplona and Madrid (Spain). The methodologyis based on high resolution, steady state CFD-RANS simulations oftraffic-emitted pollutant dispersion taking into account the effect ofthe buildings on air flow and pollutant dispersion in order to providemaps of pollutant distributions. In this study, NOx were the simulatedpollutants.
These results show a suitable performance of the CFD-RANS simu-lations when compared with the observations. The methodologyworks better when detailed information on traffic counting per streetare available, and when the large part of the emissions determiningthe concentration are within the modelling domain, like in small cit-ies. For big cities, where the transport of pollutant from urban areasoutside the modelling domain may be relevant, it is important tohave good information on the evolution of the background values. Inthe Madrid station studied, this type of information has been estimat-ed from measurements data of another station for one wind directionand from traffic and mixing height evolution for the other directions(where no observational data are available). Another option (not in-vestigated in this study) could be to use urban mesoscale models athigh resolution to have an estimation of the background concentra-tion in the studied zone for each wind direction. Keeping in mindthese points, results give confidence to use CFD-RANS to estimatethe distribution of pollutants in the streets near the air quality stationsand then, to estimate the spatial representativeness of the air qualitystations by analyzing how similar is the station concentration to thatof its surroundings. Even in the case where background concentrationis important, large pollutant concentration horizontal gradients arefound. This is because the complex air flow and emissions distributiongive rise to strong heterogeneities in the pollutant concentration
71J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
patterns in streets and squares of the urban areas. It means that thepollutant concentration measured at a traffic-oriented station couldnot provide enough information about the air quality in the areawhere it is located. However, CFD street-canyonmodels could provideinformation about the 3D distribution of pollutants complementingthe traffic-oriented station observations giving rise to a more actualpicture of the air quality around the station.
To reconstruct the 3D distribution of pollutants, information on theemission patterns, urbanmorphology, andmeteorology are also need-ed. In this study we propose a methodology to perform this recon-struction, based on a set of CFD-RANS simulations. In this context,the station measurement is used to normalise the modelled valuesand reduce the uncertainty. The best station location is not necessarilya location representative of the air quality in a segment of 100 m, but alocation that allows this reconstruction. In particular we think that themeasurement point must:
• Be far from sharp gradients of concentration. This is because such re-gions are usually much more difficult to simulate, and also to avoiderrors due to uncertainty in the position of the station.
• Be close to an area where people stay. This is because, by normalizingthe model results with the concentration at the station, we expect tohave the smallest errors close to this point. This is important becausethe aim should be to estimate the quality of the air that most peoplebreathe.
In our opinion, the air quality in an urban area cannot be assessedwith confidence using only measurements from a monitoring station.These measurements can be hardly considered representative of theair quality of a region sufficiently wide (as those mentioned in theEU Directive, for example) for the purpose. The use of street scale dis-persion models and maps of population density and residence timeshould help to resolve this problem getting a more complete and pre-cise view of the air pollution in an urban area by means of air qualitymaps (generated by validated models) associated to every air qualitystation. We think that this is one of the research lines that should beinvestigated in the next future.
Acknowledgements
Authorswould like to thank the local authorities of Pamplona andMa-dridCouncils and theRegionalGovernment ofNavarra andAEMET (Span-ish Meteorological Agency) for providing information and their support.This study has been also supported by the SpanishMinistry of Agriculture,Food and Environment and by the Projects Supercomputation andE-Science (SyeC) from the Spanish CONSOLIDER Programme andModelizacion de la Influencia de la Vegetacion Urbana en la Calidad delAire y Confort Climático a Micro y Mesoscala (CGL2011-26173).
Appendix A
In themomentum equation a sink term tomodel the formdrag dueto vegetation is added:
Sui ¼ −ρAf cd Uj jui ðA:1Þ
where Af is the leaf area density in the cell, cd is the drag coefficient ofvegetation, ∣U∣ is the wind speed and ui is wind velocity in direction i.Note that the values used for Af and cd were 0.11 and 0.2, respectively.These values are similar to those used in the modelling of Raupach etal. (1987) experiments. In the park, vegetation cells are consideredfrom the ground to 18 m-height (trees of 18 m) but for the trees with-in the street only the crown is considered as vegetation cells.
The source term in the turbulent kinetic energy (k) equation ismodelled as:
Sk ¼ ρAf cd βp Uj j3−βd Uj jk� �
ðA:2Þ
where βp is the fraction of mean kinetic energy converted into k bymeans of drag and takes a value between 0 and 1. βd is the dimen-sionless coefficient for the turbulence cascade short-circuiting, thathas no clear physical basis (Sanz, 2003). Concerning turbulent dissi-pation rate (ε) equation the source term used is:
Sε ¼ ρAf cd Cε4βpεkUj j3−Cε5βd Uj jε
� �: ðA:3Þ
Values of βd, Cε 4 and Cε 5 are based on analytical expressions com-puted by Sanz (2003) with βp = 1.
βd ¼ C0:5μ
2α
� 23
βp þ3σk
ðA:4Þ
Cε4 ¼ Cε5ð Þ ¼ σk2σε
−C0:5μ
62α
� 23
Cε2−Cε1ð Þ !
: ðA:5Þ
We assume Cε4 = Cε5 and use α = 0.05 (Dalpé andMasson, 2009)and (Cμ, σk, σε, Cε1, Cε2) = (0.09, 1, 1.3, 1.44, 1.92).
References
Arya SP. Introduction to micrometeorology. San Diego: Academic Press; 2001 [420 pp.].Biltoft CA. Customer report for Mock Urban Setting Test (MUST). DPG document
WDTC-TP-01-028, West Desert Test Center, U.S. Army Dugway Porving Ground,Dugway, Utah, 58 pp.; 2001.
Buccolieri R, Salim SM, Leo LS, Di Sabatino S, Chan A, Ielpo P, et al. Analysis of local scaletree atmosphere interaction on pollutant concentration in idealized street canyonsand application to a real urban junction. Atmos Environ 2011;45:1702–13.
Dalpé B, Masson C. Numerical simulation of wind flow near a forest edge. J Wind EngInd Aerodyn 2009;97:228–41.
Dejoan A, Santiago JL, Martilli A, Martin F, Pinelli A. Comparison between large-eddysimulations and Reynolds-averaged Navier–Stokes computations for the MUSTfield experiment. Part II: effects of incident wind angle deviation on the meanflow and plume dispersion. Bound-Layer Meteorol 2010;135:133–50.
European Commission. Establishing guidelines for the agreements on setting up commonmeasuring stations for PM2.5 under Directive 2008/50/EC on ambient air quality andcleaner air for Europe. Commission Staff Working Paper. Brussels 14.01.2011, SEC(2011) 77 final; 2011.
Galan Madruga D, Fernández Patier R, Diaz Ramiro E, Herce Garraleta M. Study of thesuperficial ozone concentrations in the atmosphere of Comunidad de Madridusing passive samplers. Rev Salud Ambient 2001;1(1):20–9.
Grimmond CSB, Cleugh HA, Oke TR. An objective urban heat storage model and itscomparison with other schemes. Atmos Environ 1991;25B:311–26.
Idso SB, Jackson RD. Thermal radiation from the atmosphere. J Geophys Res 1969;74:5397–403.
Janssen S, Dumont D, Fierens F, Mensink C. Spatial interpolation of air pollution mea-surements using CORINE land cover data. Atmos Environ 2008;42:4884–903.
Janssen S, Dumont G, Fierens F, Deutsch F, Maiheu B, Celis D, et al. Land use to character-ize spatial representativeness of air quality monitoring stations and its relevance formodel validation. Atmos Environ 2012;59:492–500.
Krochmal D, Kalina A. Measurements of nitrogen dioxide and sulphur dioxide concentra-tions in urban and rural areas of Poland using a passive sampling method. EnvironPollut 1997;96(3):401–7.
Liu LJS, Delfino R, Koutrakis P. Ozone exposure assessment in a Southern Californiacommunity. Environ Health Perspect 1997;105(1):58–65.
Lohmeyer A, BächlinW, Frantz H, MüllerWJ. Modellierung von Emission und Ausbreitungzur Untertstützung der Messplanung und Interpretation der Messergebnisse –
Zusammenarbeit nützt! Vortrag beim VDI-Kolloquium: Neuere Entwicklung bei derMessung und Beurteilung der Luftqualität, 8.–9. Juni 2005, Schwäbisch Gmünd; 2005.
Lumbreras J. Sistema de Evaluación de Riesgos por Contaminación Atmosférica en la Pen-ínsula Ibérica, SERCA. Final Project Report for Ministry of Environment, Rural andMarine Affairs. Project 058/PC08/3-18.1. Contribution from ETSII-UPM, IDAEA-CSICand ISC-III; 2012.
PalaciosM. Influencia del tráfico rodado en la generación de la contaminación atmosférica.Aplicación de un modelo de dispersión al área de influencia de la Comunidad deMadrid (Document written in Spanish). Doctoral Thesis; 2001.
Parra MA, Elustondo D, Bermejo R, Santamaría JM. Ambient air levels of volatile organiccompounds (VOC) and nitrogen dioxide (NO2) in a medium size city in NorthernSpain. Sci Total Environ 2009;407:999-1009.
72 J.L. Santiago et al. / Science of the Total Environment 454–455 (2013) 61–72
Parra MA, Santiago JL, Martín F, Martilli A, Santamaría JM. A methodology to urban airquality assessment during large time periods of winter using computational fluiddynamic models. Atmos Environ 2010;44:2089–97.
Raupach MR, Bradley EF, Ghadiri H. Wind tunnel investigation into the aerodynamic ef-fect of forest clearing on the nesting of Abbott's Booby on Christmas Island. Internalreport, CSIRO Centre for Environmental Mechanics, Canberra; 1987.
Repschies D, Schatzmann M, Leitl B. How dense is dense enough? — systematic evalua-tion of the spatial representativeness of flow measurements in urban areas. Interna-tional Workshop on Physical Modelling of Flow and Dispersion Phenomena, Orleans,Aug. 28–31, 2007; 2007.
Richards PJ, Hoxey RP. Appropriate boundary conditions for computationalo wind en-gineering models using k–ε turbulence model. J Wind Eng Ind Aerodyn 1993;46 &47:145–53.
Santiago JL, Martilli A, Martín F. CFD simulation of airflow over a regular array of cubes.Part I: three-dimensional simulation of the flow and validation with wind tunnelmeasurements. Bound-Layer Meteorol 2007;122:609–34.
Santiago JL, Parra MA, Martín F, Santamaría JM. Numerical and experimental study of airquality in the Pamplona downtown (Navarra, Spain). Proceedings of HARMO12.Cavtat (Croatia); 2008.
Santiago JL, ParraMA,Martilli A, Martín F, Santamaría JM. Analysis of spatial representative-ness of urban monitoring stations using steady CFD-RANS simulations. Proceedings of
9th Symposium on the Urban Environment of American Meteorological Society. Key-stone. Colorado. USA; 2010.
Sanz C. A note on k–ε modelling of vegetation canopy air-flows. Bound-Layer Meteorol2003;10:423–53.
SchatzmannM, BächlinW, Emeis S, Kühlwein J, Leitl B, MüllerWJ, et al. Development andvalidation of tools for the implementation of European Air Quality Policy in Germany(Project VALIUM). Atmos Chem Phys 2006;6:3077–83.
Schatzmann M, Olesen H, Franke J, editors. COST732 model evaluation case studies: ap-proach and results. COST Office Brussels3-00-018312-4; 2010.
Schlünzen KH, Hinneburg D, Knoth O, Lambrecht M, Leitl B, López S, et al. Flow and trans-port in the obstacle layer: first results of the micro-scale modelMITRAS. J Atmos Chem2003;44(2):113–30.
Spangl W, Schneider J, Moosmann L, Nagl C. Representativeness and classification ofair quality monitoring stations. FINAL REPORT. REPORT REP-0121. Vienna:Umweltbundesamt; 2007.
Vardoulakis S, Gonzalez-Flesca N, Fisher BEA, Pericleous K. Spatial variability of air pollu-tion in the vicinity of a permanent monitoring station in central Paris. Atmos Environ2005;39:2725–36.
