RESEARCH ARTICLE

WRF-SMOKE-CMAQ modeling system for air quality evaluation in SãoPaulo megacity with a 2008 experimental campaign data

Taciana Toledo de Almeida Albuquerque1,2& Maria de Fátima Andrade3

& Rita Yuri Ynoue3&

Davidson Martins Moreira2,4& Willian Lemker Andreão1

& Fábio Soares dos Santos1 &

Erick Giovani Sperandio Nascimento4

Received: 18 June 2018 /Accepted: 23 October 2018 /Published online: 29 October 2018# Springer-Verlag GmbH Germany, part of Springer Nature 2018

AbstractAtmospheric pollutants are strongly affected by transport processes and chemical transformations that alter their composition andthe level of contamination in a region. In the last decade, several studies have employed numerical modeling to analyzeatmospheric pollutants. The objective of this study is to evaluate the performance of theWRF-SMOKE-CMAQmodeling systemto represent meteorological and air quality conditions over São Paulo, Brazil, where vehicular emissions are the primarycontributors to air pollution. Meteorological fields were modeled using the Weather Research and Forecasting model (WRF),for a 12-day period during the winter of 2008 (Aug. 10th–Aug. 22nd), using three nested domains with 27-km, 9-km, and 3-kmgrid resolutions, which covered the most polluted cities in São Paulo state. The 3-km domain was aligned with the Sparse MatrixOperator Kernel Emissions (SMOKE), which processes the emission inventory for the Models-3 Community Multiscale AirQuality Modeling System (CMAQ). Data from an aerosol sampling campaign was used to evaluate the modeling. The PM10 andozone average concentration of the entire period was well represented, with correlation coefficients for PM10, varying from 0.09in Pinheiros to 0.69 in ICB/USP, while for ozone, the correlation coefficients varied from 0.56 in Pinheiros to 0.67 in IPEN.However, the model underestimated the concentrations of PM2.5 during the experiment, but with ammonium showing smalldifferences between predicted and observed concentrations. As the meteorological model WRF underestimated the rainfall andoverestimated the wind speed, the accuracy of the air quality model was expected to be below the desired value. However, ingeneral, the CMAQ model reproduced the behavior of atmospheric aerosol and ozone in the urban area of São Paulo.

Keywords Air quality modeling . CMAQ . SMOKE .WRF .Measurement campaign . Aerosol . Ozone

Introduction

Urban air quality has been receiving widespread attention, asover half of the world’s population now lives in urban centers,54.3% as of 2016 (The Word Bank 2018). If populationgrowth trends continue, air quality will deteriorate further un-less serious control measures are put in place. The effects are

expected to be especially serious in developing country megacities (cities populations of over 10 million), such as SãoPaulo city, Brazil. The potential for these quick changes com-bined with a developing interest in cleaner air has impressedthe need to enhance their capacity to control air pollution uponpolicymakers (Johnson et al. 2011).

Particulate matter, especially fine particles (FPM; diameterless than 2.5 μm), is one of most concerning factors that con-tribute to adverse health effects in urban areas. Several studiesindicate that mortality and morbidity are correlated to particleconcentration (Pope et al. 2002, 2004; Laden et al. 2006;Krewski et al. 2009; Crouse et al. 2012; Cesaroni et al.2013; Gouveia and Junger 2018). Cohen et al. (2017) reported4.2 million deaths globally due to fine particle matter.

Atmospheric particles vary in size from a few nanometersto tens of micrometers, depending on their chemical compo-sition, which may reflect their source (Seinfeld and Pandis2006). They can be directly emitted from a source (primary

Responsible editor: Marcus Schulz

* Taciana Toledo de Almeida Albuquerquetaciana@desa.ufmg.br

1 Federal University of Minas Gerais, Belo Horizonte, Brazil2 Federal University of Espírito Santo, Vitória, Brazil3 University of São Paulo, São Paulo, Brazil4 SENAI CIMATEC, Salvador, Brazil

Environmental Science and Pollution Research (2018) 25:36555–36569https://doi.org/10.1007/s11356-018-3583-9

particles) or be formed in the atmosphere by chemical reac-tions (secondary particles) and remain in the atmosphere forseveral days, during which they undergo further changes byphysicochemical processes such as dilution, dispersion, coag-ulation, deposition, and chemical reaction.

Vehicular emissions are the main source of particulateemissions in urban Brazil (Alonso et al. 2010; Andradeet al. 2012; Miranda et al. 2012; Pacheco et al. 2017;Andrade et al. 2017; Alvim et al. 2017). Although eachcity has specific characteristics, the main fuels used for theBrazilian light vehicle fleet are ethanol (95% ethanol; 5%water), gasohol (75% gasoline; 25% ethanol), or com-pressed natural gas, while the heavy fleet (trucks and bus-ses) runs on diesel rather than biodiesel (Andrade et al.2017). Diesel is also used in light-duty commercial vehi-cles. Fine particle emissions are primarily associated withthe diesel fleet (Alves et al. 2015). Tunnel studies in SãoPaulo city have shown that diesel combustion is a signif-icant contributor to the emission of black carbon (BC) andgaseous pollutants (Andrade et al. 2012; Hetem andAndrade 2016).

In recent years, significant advances have been made intechniques used to estimate ambient air pollution levelsand identify emission sources, and policy maker techni-cians in environmental agencies around the world are be-ginning to rely on air quality models (AQMs). TheModels-3 Community Multiscale Air Quality ModelingSystem (CMAQ) (Binkowski and Shankar 1995; Jun andStein 2004; Byun and Schere 2006), CAMx (Pepe et al.2016) and the Weather Research and Forecast modelcoupled with Chemistry (WRF-Chem) (Grell et al. 2005)are examples of photochemical models that have been de-veloped considering emissions, gas-to-particle conversion,and chemical reactions in the atmosphere, as well as otherprocesses of dispersion and transport (Albuquerque et al.2012; Vara Vela et al. 2016). In relation to the CMAQmodel, the system is designed for environmental researchand has applications for multiscale air pollution problems(urban and regional) and multi-pollutants (oxidants, aciddeposition and particulates). Several recent applicationsin different parts of the world are observed in the literature(Wang et al. 2015a, b; Li et al. 2018; Syrakov et al. 2016;Chen et al. 2017; Jiang and Yoo 2018). The CMAQ modelwas chosen because it has an aerosol module, is computa-tionally economical, operates off-line, and is efficient inthe dynamic representation of particles in the atmosphere(Jun and Stein 2004). Nevertheless, in the scenario of im-portant urban areas in Brazil, such as the MetropolitanArea of São Paulo (MASP), the studies with air qualitymodels, as WRF-Chem, are limited (Silva Junior andAndrade 2013; Andrade et al. 2015; Hoshyaripour et al.2016; Vara Vela et al. 2016). In the MASP, this is the firststudy that applies CMAQ model.

Located in southeastern Brazil, the city of São Paulo is thecapital of the State of São Paulo and acts as the central nucleusof the MASP. The MASP is certainly one of the largest mega-cities in the world, with the largest urban motor vehicle fleet,with about 21 million inhabitants and 7 million automobiles(CETESB 2017). Pacheco et al. (2017) reviewed the emis-sions and concentrations of particulate matter in MASP anddiscussed the significance of vehicle fleet and fuel type used inroad vehicles on the emissions. Unfavorable conditions forpollutant dispersion are observed mainly during winter (dryseason), with frequent subsidence and thermal inversionlayers, which may lead to higher concentrations of pollutants(CETESB 2009, 2017).

Atmospheric models of dispersion, transport, and chemis-try utilize information from emission inventories to predictconcentrations of air pollutants in the atmosphere. Therefore,it is necessary to resolve emission inventories in space andtime and to allocate them to specific locations (i.e., gridsquare, smoke stack). When developing an emission invento-ry to be used as input for air quality numerical simulations,precursors of secondary pollutants such as SO2, NOx, andVOCs need to be accounted for because they can generatesulfates, nitrates, and secondary organic aerosols, respec-tively (Seinfeld and Pandis 2006; Borge et al. 2008; Borgeet al. 2014). Model results should also be evaluatedthrough comparisons with measured ambient levels to testthe model implementation and the accuracy of the devel-oped emission inventory.

In this context, the objective of this study was to evaluatethe performance of the WRF-SMOKE-CMAQ modeling sys-tem to represent the meteorological and air quality conditionsover São Paulo city, as well as to investigate the spatial andtemporal variability of the variables simulated. Additionally,an aerosol sampling campaign was performed over 10 days inwinter 2008 (Aug. 12th–Aug. 22nd) in the MASP to obtainlocal data to evaluate the numerical models.

Methodology

Experimental field campaign set up

A field campaign was conducted during August 12th–22nd,2008 (winter) by Laboratório de Análise dos ProcessosAtmosféricos (LAPAt–IAG/USP). PM10 and PM2.5 were sam-pled and measured at the Institute of Biomedical Sciences ofUniversity of São Paulo (ICB/USP). A gravimetry methodwas employed to determine mass concentrations, reflectancemethod to quantify BC concentrations, X-ray fluorescence tocharacterize elemental composition, and ion chromatographywas used to determine the composition and concentrations ofanions and cations. HNO3 and NH3 were also measured forthe first time in São Paulo.

36556 Environ Sci Pollut Res (2018) 25:36555–36569

A 13-stage cascade-type Microorifice Uniform DepositImpactor (Nano-MOUDI) was used to collect size-resolvedaerosol samples twice a day at two specific periods during thecampaign (daytime—07 to 17 h and nighttime—17 to 07 h),

using polycarbonate filters of diameter 47 mm and thickness of8 μm (Nuclepore ®) and a Teflon after-filter of 37 mm indiameter, which allows for the collection of particles smallerthan 0.1 μm. The MOUDI sampling flow rate was 20 L min−1.

The PM mass concentrations from Nano-MOUDI sampleswere obtained gravimetrically using an electronic high-precision microbalance (Mettler), with a sensitivity of 1 μg.Assessments were performed in a controlled-atmosphere en-vironment (25 °C and 60% relative humidity). Filters wereequilibrated for 24 h prior to weighing. Electrostatic chargeswere controlled using radioactive 210 Po sources. The con-centration of BCwas determined using a reflectance technique(Reid et al. 1998), using a Smoke Stain Reflectometer-Model43 (Longo et al. 1999).

The filters were analyzed to establish their elemental com-positions (Al, As, Ca, Cd, Cl, Cr, Cu, Fe, K, Mn, Ni, Pb, S, Sb,Se, Si, Sr, Ti, V, and Zn) by X-ray fluorescence with EDX(energy dispersive x-ray detector), considering they were inthe oxidized form. Then, the same samples were subjected toion chromatographic (IC) analysis (761 Compact IC, Metrohn)

for anion (F−, Cl−, Br−, NO−2 NO

−3 , SO

2−4 , and PO3−

4 ) and cation(Caþ2 , K

+, Mg2+, Na+, and NHþ4 ) concentrations.

Mass balance of the insoluble material from the soil and theelements associated with combustion processes were estimat-ed considering that the main inorganic species would be pres-ent in the oxidized form, and then the following compoundswere added: Al2O3, SiO2, CaCO3, K2O, TiO, VO, MnO2,Fe2O3, NiO, Cu2O, ZnO, Se, Br, Sr, Zr, and Pb, calculatedby EDX, based on soluble compounds from chromatographyanalysis (ammonium, sulfate, nitrate, and NaCl) and the

Fig. 1 Study area and domains. The marks represent the measurement stations

Table 1 Temporal, spatial, physical, and dynamic parameters for WRF

Temporal parameters

Initial date Aug. 11, 2008 (00 UTC)

Final date Aug. 23, 2008 (18 UTC)

Simulated period 306 h

Spatial parameters

Grid resolution 27 km 9 km 3 km

Column numbers 34 52 109

Row numbers 34 52 76

Vertical layers 21 (p_top: 5000 m)

Grid center − 23, 55°S; − 46, 49°W

Physical and dynamic options—WRFv 3.0.1

Microphysical Thompson

Longwave radiation Rrtm

Short wave radiation Dudhia

Surface layer Pleim–Xu

Surface–earth Pleim–Xu

Boundary layer ACM2 (Pleim)

Cumulus Kain–Fritsh (new ETA)

Soil layers Surface model Pleim–Xu

Diffusion, dissipation, advection Third-order Runge-Kutta

Turbulence and mixing Second-order diffusion term

Eddy coefficient First-order Smagorinsk scheme

Environ Sci Pollut Res (2018) 25:36555–36569 36557

following expressions were obtained: Al*1.88, Si*2.14,K*1.2, Ca*2.9, Ti*1.33, V*1.31, Mn*1.58, Fe*1.43,Ni*1.27, Cu*1.13, and Zn*1.25. The water aerosol was esti-mated using the ISORROPIA thermodynamic equilibriummodel (Nenes et al. 1998; Nenes et al. 1999). The systemmodeled by ISORROPIA calculates the composition of inor-ganic species of the atmospheric aerosol and divides the spe-cies into gas, liquid, and solid phases.

Meteorological model

The Advanced Research WRF (WRF-ARW v3.0.1) meso-scale model developed by the National Center forAtmospheric Research (NCAR), USA, was used to simu-late the meteorological conditions at the same period of

the experimental campaign, applying three nested domainswith 27-km grid resolution (34 × 34 cells), 9-km grid res-olution (52 × 52 cells), and a high-resolution domain of 3-km grid resolution (109 × 76 cells), as shown in Fig. 1.Domain and configurations are given in Table 1.Simulations were conducted for August 11–23, 2008 (awinter period). The model was initialized at 00 UTC onAugust 11, 2008 and integrated for 306 h. The three-dimensional National Centers for EnvironmentalPrediction (NCEP) Global Forecasting System (GFS) me-teorological analysis data available at a resolution of 1degree was used for the initial conditions. The boundaryconditions to the outer domain were updated from 3 hour-ly GFS forecasts. The US Geological Survey (USGS) el-evation data, Food and Agriculture Organization (FAO)

Fig. 2 Spatial distribution patternof vehicular emissions for thestudy region. White spots denotethe locations of emission. SãoPaulo city is the inner white spot

Fig. 3 Maximum emissionscenario of CO (moles/s) for thestudy area

36558 Environ Sci Pollut Res (2018) 25:36555–36569

soil data, and USGS land use data available at 10 arc min,5 arc min, and 30 arc sec resolutions were used to definethe surface fields in the model. For verification of the accuracy

of the model, the data measured at the Pinheiros meteorolog-ical station were used because it was located close to theexperiment site and data for other stations were not available.

Fig. 4 Hourly emission profiles for NH3 (moles/s), CO (moles/s), NOx (moles/s), SO2 (moles/s), and PM10 (g/s)

Table 2 Emission inventory usedin SMOKE Emission inventory [ton/year]

Fuel MP NOx SO2 CO VOC NH3

Gas 68,498.8 69,541.9 1738.6 3,911,732.0 297,291.7 15,907.8

Ethanol – 14,081.3 – 792,073.7 113,530.6 –

Flex–gas or ethanol – 623.8 – 23,391.7 6081.9 –

Diesel c 63,667.9 231,902.8 14,757.5 1,328,170. 6 139,774.2 –

Diesel EII 2798.1 10,191.7 64.9 5837.0 4780.9 –

Diesel EIII 2424.9 8832.8 56.2 5058.8 529.9 –

Vehicular natural gas – 3265.6 – 17,416.4 9578.9 –

Motorcycle 49,077.3 49,824.7 1245.7 2,802,638.5 319,500.8 –

Environ Sci Pollut Res (2018) 25:36555–36569 36559

Emission inventory

The SparseMatrix Operator Kernel Emission (SMOKE) (DOIhttps://doi.org/10.5281/zenodo.1321279) is an operationalsystem that processes emission inventories. The SMOKEemissions model was used to create a 3D emission inventoryfrom vehicular sources, varying temporally and spatially tothe MASP and its surroundings. For georeferencing ofspatial distribution of vehicles, it was used map of lightsspots provided by the Operational Linescan System sensor(OLS) satellites of the Defense Meteorological Satellite

Program (DMSP). This satellite map was used by Martinset al. (2008, 2010) to estimate the carrier density. In Martinset al. (2008, 2010), satellite images were combined to con-struct urbanization maps of prominent Brazilian cities. EachBcity light intensity value^ was equivalent to 24.8 vehicleskm−2 (Martins et al. 2010). The resulting urbanization mapshowed good correlation with number of vehicles and popu-lation and was essential in running an emission inventory withthe SMOKE model in MASP.

SMOKE system was used in the 3-km domain of WRFsimulation. Figure 2 illustrates the pattern of spatial distribu-tion of emissions used in the simulation based onMartins et al.(2008, 2010), while Fig. 3 is an example of the maximum COemission rate scenario using the lights mapping methodology,which was inserted into the SMOKEmodel. It is observed thatthe urban area of theMASPwas well represented by the trafficarea of the region.

The emission factors were derived from field experi-ments in city tunnels described by Martins et al. (2006),which were used to estimate the emission of NOx, SO2,CO, VOC, and particulate matter. For ammonia gas,emission factors estimated by Fraser and Cass (1998) wereused. Two temporal emission profiles were used, one forheavy vehicles (CETESB 2009) and another for light ve-hicles (Lents et al. 2004). Speciation of organic com-pounds was established based on results reported inMartins et al. (2006), and the partitioning of the particu-late material was based on Sánchez-Ccoyllo et al. (2009).The Fleet distribution accounted for 5.5 million vehicles

Table 4 Description of temporal,spatial, physical and dynamicoptions parameters for CMAQmodel

Temporal parameters

Initial date 08/11/2008 (00 UTC)

Final date 08/23/2008 (18 UTC)

Simulated period 306 h

Spatial parameters

Grid resolution 3 km

Column numbers 102

Row numbers 69

Vertical layers 21

Grid center −23,55°S; −46,49°WPhysical and dynamic options–CMAQV4.6

Numerical scheme CTM—Yamartino (Advection with mass conservation)

Photochemical mechanism JPROC (Profile provided by the model)

Chemical model for the gas phase Ebi_cb05 (Euler scheme adapted to the chemicalmechanism Carbon Bond V)

Aerosol model Aero4 (Aerosol model including sea salt emissionsand their thermodynamic properties)

Cloud model Cloud_ACM (Based on the cloud RADM processor,which uses the asymmetric convection model tocalculate the convective mixture)

Chemical mechanism Cb05_ae4_aq (CB-05 gas phase mechanism. Aero4considering sea salt, chemistry of gas phase/cloud)

Table 3 Initial andboundary concentration(ICON/BCON) for thesetup of CMAQ

ICON-BCON Concentration(ppm)

SO2 0.001

NO2 0.005

NO 0.001

O3 0.03

H2O2 0.0001

NH3 0.00001

ETH 0.0003

ETOH 0.0006

CO 0.2

OLE 0.0001

IOLE 0.0001

FORM 0.0001

PM* (μg m−3) 40

36560 Environ Sci Pollut Res (2018) 25:36555–36569

(CETESB 2009). Figure 4 represents the temporal profilefor the SO2, VOCs, NOx, and PM10 inventoried, where theexistence of two profiles can be observed: one related to theemission of CO and NH3, with two peaks, related to the light-duty vehicles and another with only one peak for the pollut-ants SO2, NOx, and PM10, related to the heavy-duty vehicles.Table 2 presents the final emission inventory obtained.

Air quality model

The Community Multiscale Air Quality (CMAQ) modelingsystem is a Eulerian photochemical model developed by theUS Environmental Protection Agency (USEPA) in whichcomplex interactions between atmospheric pollutants on ur-ban, regional, and hemispheric scales are well treated

Fig. 6 Aerosol mass balance for adaytime and b night time

Fig. 5 Temporal evolution ofconcentration of coarse particlesand fine particulate matter duringthe field campaign collected byNano-MOUDI

Environ Sci Pollut Res (2018) 25:36555–36569 36561

considering a consistent framework. It is used in the impactassessment of multiple pollutants; for example, troposphericozone and other oxidants, and aerosols and acid deposition(Byun and Schere 2006).

CMAQ version 4.6 model simulations have been performedin the 3-km domain of WRF simulation, which covers the mostpolluted cities in the MASP (São Paulo, Campinas, Sorocaba,São José dos Campos and Cubatão). Aerosol processes andaqueous chemistry in CMAQ were used (AERO4—thefourth-generationmodal CMAQ aerosol model with extensionsfor sea salt emissions and thermodynamics), with Carbon Bond

Vas a gas phase mechanism (Yarwood et al. 2005). The simu-lation period was from Aug. 11, 2008 to Aug. 23, 2008, with1 day accounted for spin-up.

The initial and boundary conditions available as defaultmodels were modified to make them more realistic for model-ing scenarios. The initial and boundary chemical concentra-tions (ICON and BCON modules), presented in Table 3, arederived from averages of measurements from local monitoringstations or previous numerical tests with CMAQ that adjustedthe values. The organic parcel was derived from Martins et al.(2006). Table 4 shows the settings used in CMAQ.

Fig. 7 Temporal evolution for a air temperature (°C), b relative humidity, and c wind velocity, measured at Pinheiros station and modeled using WRF

Fig. 8 Wind rose at Pinheiros station obtained from WRF simulations (left side) and from measurements (right side)

36562 Environ Sci Pollut Res (2018) 25:36555–36569

PM concentrations obtained during the experimental cam-paign at USP and O3 monitoring data from three monitoringstations of the Companhia Ambiental do Estado de São Paulo

(CETESB, São Paulo State Environmental Protection Agency):IPEN, Pinheiros and Ibirapuera were used to evaluate the nu-merical results.

Results

Experimental field campaign

The mass concentrations obtained from gravimetric analysesfrom Nano-MOUDI samples showed large differences be-tween day and night time values. Figure 5 shows the temporalevolution of the coarse particles (CPM; aerodynamic diameterbetween 2.5 and 10 μm) and fine particulate matter (FPM;aerodynamic diameter less than 2.5 μm) during the period. Itis worth noting that the concentrations of FPM were largerthan the CPM for all the days, which confirms the significantpresence of vehicular sources at the study area. August 18 hadthe maximum concentration peak for FPM (85 μg m−3) andCPM (29 μg m−3), during the night.

Figure 6 presents the mass balance calculated for theexperimental campaign, considering size distribution fordaytime—07 h to 17 h (a) and night time—17 h to 07 h(b) periods. During the day, a significant fraction of FPMremained unexplained, particularly for the lower diame-ters. However, the difference between the mass concentra-tion and the sum of the component concentrations mightbe attributable to organic carbon (OC) being in fine mode.Castanho and Artaxo (2001) found that for São Paulo, BC,and OC constitute more than 70% of the FPM. Ynoue andAndrade (2004) found that OC explained 38% of the dif-ference and almost all of the missing diurnal and nocturnalmass. During the night period, the participation of ions,metals, and water content increases with respect to day-time values.

Atmospheric aerosol collected was present more in theaccumulation fine mode, with a cutting diameter of0.32 μm in nano-MOUDI samples. However, BC particleswere found, on average, with the cutting diameter of0.05 μm. Considering the unexplained PM as OC, duringthe day, it appears that OC makes up the largest compo-nent of the mass of fine particulate matter collected in theMASP, reaching about 80% in the lower stages of nano-MOUDI. During the night period, the participation ofions, metals, and water content increased with respect tothe daytime.

Among the mass fraction of the ions extracted from the gaschromatograph, the largest mass participation is characterizedby fine sulfate aerosols, followed by ammonium and nitrate.Results from ion chromatographic experiments agree with themeasured concentration of nitrates, where the largest concen-tration occurs during the night time.

Fig. 10 Temporal evolution of PM10 concentrations measured andsimulated using CMAQ at CETESB stations: Ibirapuera (a), CerqueiraCésar (b), and Pinheiros (c)

Fig. 9 Temporal evolution of day and nighttime average concentration ofPM10 measured at ICB/USP and modeled using WRF

Environ Sci Pollut Res (2018) 25:36555–36569 36563

The temporal evolution of fine particulate matter is ex-plained by the measured concentration of ammonia gas andnitric acid. The concentration of FPM followed these gaseousconcentrations. The reaction betweenNO andNO3 is extremelyfast, and therefore, they cannot coexist in high concentration.During the night, the concentration of NO decreased due toreactions with O3 (Seinfeld and Pandis, 2006). The con-centration of nitrate increases and, consequently, the con-centration of aerosol also increases.

WRF model

Temperature, relative humidity (RH), and wind speed fieldsimulated by WRF (August 12th–August 23th) are presentedrespectively in Fig. 7a–c, while wind direction is presented inFigs. 7d and 8. During this week, the synoptic conditions wereassociated with two cold fronts passages (Aug. 13–14 andAug. 20–21), which influenced the increase of the RH anddecrease of temperature. Usually, cold fronts are associatedwith higher wind speeds, humid air sometimes accompaniedby rain, and cleaner air (Albuquerque et al. 2012). Higher

pollutant concentrations can be observed in the presence of ahigh-pressure system, with stagnation of air circulation andmore stable atmospheric conditions. The WRF modelunderestimated the rainfall and overestimated the wind speedduring the study period. The temperature field had a goodrepresentation showing the variation between days and nights,but the model did not represent the behavior of these variableswell during the cold front episodes (Aug 13th and 21st).

In general, WRFwas capable of reproducing the atmospher-ic conditions of the study area regarding air temperature andRH. During the experiment, calm winds (less than 0.5 m s−1)accounted for 51.5%, which the model did not capture well.Numerical models generally are not sensitive enough to simulatevery low velocity speeds (Shimada et al. 2011; Zang et al. 2013).

Air quality model

The performance of the CMAQ model has been evaluatedboth for PM and ozone. For PM data from the experimentalfield, campaign and monitoring stations were used.Figure 9 shows the temporal evolution of the averagePM10 measured in ICB/USP and modeled by CMAQ, sep-arated into daytime and nighttime. It is observed that dur-ing this period, there were no exceedances of the Brazilianair quality standard (24 h average is 150 μg m−3). Themaximum average value measured was 113 μg m−3 onthe evening of August 18, and the model underestimatedthis value by approximately 80 μg m−3, simulating33 μg m−3. Meanwhile, the maximum simulated averageoccurred during August 19 (64 μg m−3).

Figure 10 represents the hourly evolution of the PM10 at the(a) Ibirapuera, (b) Cerqueira César, and (c) Pinheiros stations.Although the three stations are located in neighboring cells inthe CMAQ model, each CETESB station has unique charac-teristics influenced by the type of road near where it is located

Table 5 Statistical indexes forPM10, PM2.5, BC, NH4, NO3,SO4, and O3

Pollutant Location Average CMAQ(μg m−3)

Average observed(μg m−3)

Mean bias(μg m−3)

Ratio r

PM10 ICB/USP 41.19 64.48 − 23.29 0.67 0.69

Ibirapuera 39.79 45.86 − 4.03 1.11 0.24

Pinheiros 41.62 72.16 − 28.50 0.92 0.09

Cerq. César 40.22 53.60 − 12.73 0.95 0.24

PM2.5 ICB/USP 27.51 42.41 − 14.90 0.77 –

BC ICB/USP 4.90 5.42 − 0.52 1.65 –

NH4 ICB/USP 1.07 1.08 − 0.01 1.45 –

NO3 ICB/USP 1.21 0.92 0.29 1.51 –

SO4 ICB/USP 1.95 3.27 − 1.31 0.75 –

O3 IPEN 0.024a 0.016a 0.008a 4.64 0.67

Ibirapuera 0.023a 0.015a 0.008a 2.46 0.64

Pinheiros 0.023a 0.006a 0.02a 10.50 0.56

a ppm

Fig. 11 Temporal evolution of day and night time average concentrationof PM2.5 measured at ICB/USP and modeled using CMAQ

36564 Environ Sci Pollut Res (2018) 25:36555–36569

next to, which complicates the representation of the PM con-centration values in CMAQ. Thus, it is observed that thePinheiros station, which is located on the Pinheiros RiverRoad, has one with the highest PM10 concentrations, thuspresenting the largest discrepancies between the modeledand measured values. Additionally, the abrupt variations inthe observed concentrations in a short time are representedwith difficulty by air quality models. The CMAQ model, de-spite reproducing an average pattern during this period, couldnot represent the peaks recorded in these three CETESB sta-tions. The peaks observed occurred on August 16, with max-imum concentrations of 119 μg m−3 for the station located atIbirapuera, 124 μg m−3 at Cerqueira César street station, and198 μg m−3 at Pinheiros station. The CMAQ presented max-imum concentrations of PM10 during the night of August 18,of 112.5 μg m−3 at the Ibirapuera station grid cell, 128 μg m−3

at the Cerqueira César station, and 116.5 μg m−3 at thePinheiros station. Although the peaks are not coincident, theaverage concentrations during the whole period were wellreproduced by the model, with Cerqueira César presenting amean ratio of 0.95 between the modeling and the measure-ment. The best correlation coefficient (r = 0.69) obtained be-tween the model and the measured was for the data collectedduring the ICB/USP experiment. Table 5 presents some sta-tistical indexes for PM10.

For fines particles, the model underestimated the concen-trations during the experiment, as shown in Fig. 11. As themeteorological modelWRF underestimated the rainfall duringthe study period and overestimated the wind velocity, the

accuracy of the air quality model was expected to be belowthan the desired value. Another limiting factor for the accura-cy of the CMAQ model was the lack of detailed informationon the emission inventory, considering not only vehicularsources but also industrial and biogenic sources as well asother relevant sources. In general, the model could representthe daytime concentrations better. An increase in measuredconcentration during nighttime is also observed, which theCMAQmodel was unable to represent. Table 5 presents somestatistical indexes for PM2.5.

The BC variation was underestimated by the model, espe-cially during the most polluted period (August 18–19), shownin Fig. 12a. On the other hand, the highest concentration ofinorganic aerosols was better represented. Figure 12b–d pre-sents the model and measured concentration of fine aerosolsthat originated from sulfate and ammonium and nitrate,respectively. The sulfate achieved a maximum measuredconcentration of 9.25 μg m−3 on August 20, and a maxi-mum simulated concentration of 3.27 μg m−3 duringAugust 19 (daytime).

The fine mode of ammonium and nitrate aerosol was abouttwice as small as the fraction of the fine mode of sulfate aero-sol. The maximum concentration of ammonium aerosol ob-served during the experiment was 2.7 μg m−3 on August 13and also during the night of August 20. The maximum con-centration of nitrate was simulated on August 21(4.39 μg m−3), and the maximum measured concentrationoccurred during the night of August 18 (1.5 μg m−3).During the same period, the maximum concentrations of

Fig. 12 Temporal evolution of day and nighttime average concentration of a black carbon, b sulfate, c ammonium, and d nitrate, measured at ICB/USPand modeled using CMAQ

Environ Sci Pollut Res (2018) 25:36555–36569 36565

PM2.5, PM10, and BC pollutants were also obtained. It isfound that on average, the aerosol ammonium was better rep-resented by the model, showing the smallest differences be-tween predicted and observed values. Table 5 presents somestatistical indexes for BC, NH4, NO3, and SO4.

To evaluate the CMAQ performance in reproducing theozone formation, model results have been compared with datafrom IPEN, Ibirapuera, and Pinheiros CETESB stations(Fig. 13). The model represented the evolution of the ozoneformation well, but in general overestimated the concentrationvalues in the whole simulation period, in almost all the mea-surement stations. The maximum that occurred on August 19was overestimated by the model at every station. The maxi-mum measured concentrations achieved by the monitoringstations were 0.092 ppm at IPEN, 0.072 ppm at Ibirapuera,and 0.065 ppm in Pinheiros, while the maximum modeledconcentrations were 0.115 ppm at IPEN grid cell, 0.115 ppmin the Pinheiros station grid cell, and 0.113 in Ibirapuera

station grid cell. Table 5 presents some statistical indexes forO3. The station that was best represented by the CMAQmodelwas IPEN, which presented lower mean bias (0.008 ppm) andthe highest correlation coefficient (r = 0.67). The differencesobserved between the stations can mainly be attributed to theirlocation: the IPEN station is located in a high ground of thecity and distant from high traffic routes and it is one the sta-tions strategically implemented to monitor ozone; theIbirapuera station is located in a green area surrounded byurbanized areas where land use is predominantly residentialand is far from vehicular and industrial sources of direct influ-ence; and the Pinheiros station is manly influenced by trafficemissions of one of the most important expressways in themunicipality.

The chemical species established as initial and boundaryconditions influences the modeling of secondary pollutantsformation. Borge et al. (2010) showed that dynamic boundaryconditions improve the results in CMAQ as compared to staticconcentrations prescribed, while Hogrefe et al. (2017) bringsthe importance of the initial conditions, which depends on thegeographic domain and the species of interest, and its influ-ence decreases with the simulation time. Problems in definingthese conditions usually cause inconsistencies in the resultsand are sometimes not sufficient to fit the modeling of thestudy area (Samaali et al. 2009). Additionally, Appel et al.(2017) highlight the importance of the clouds and photolysiscalculations between theWRF and CMAQmodels in the gen-eration of the O3 mixing ratio.

Conclusions

Aerosol and gases were measured at the MASP during winterand were used to highlight the significance of vehicle emis-sion of gases and the formation of secondary aerosols. Massconcentration analyses from the aerosol sampled during thefield campaign showed large differences between daytime andnighttime periods.

The strong emissions of trace gases and aerosol by vehi-cles, coupled with low rainfall, favor high soil dust resuspen-sion and, under the unfavorable natural conditions of disper-sion, contributed significantly to the high concentrations ofpollutants observed in the entire study area.

Air quality modeling simulations arise from modeling ofcomplex atmospheric processes. For the first time, the CMAQmodeling system was used in São Paulo. Sensitivity tests areneeded to improve the modeled results because air qualitymodel is credible for assessing emission control strategiesand can be used for these applications. The CMAQ modelbetter reproduced the behavior of atmospheric aerosol inICB/USP grid cell, with a correlation coefficient of 0.69, al-though the mean values were underestimated in the evaluatedmonitoring stations. The ozone was better represented at

Fig. 13 Temporal evolution of O3 concentration measured and modeledusing CMAQ at CETESB stations: IPEN (a), Ibirapuera (b), andPinheiros (c)

36566 Environ Sci Pollut Res (2018) 25:36555–36569

IPEN, with a correlation coefficient of 0.64, while Pinheirosstation grid cell presented the lowest correlation coefficient(0.56). Although the model underestimated the concentrationsof PM2.5, the BC, NH4, NO3, and SO4 concentrations showedsmall differences between predicted and observed concentra-tion. However, a better meteorological field and a more de-tailed emissions inventory may improve the results.

Funding information This research was partially funded by Fundação deAmparo à Pesquisa do Estado de São Paulo (FAPESP), Coordenação deAperfeiçoamento de Pessoal de Nível Superior (CAPES) and ConselhoNacional de Desenvolvimento Científico e Tecnológico (CNPq), Brazil.

References

Albuquerque TT, Andrade MF, Ynoue RY (2012) Characterization ofatmospheric aerosols in the city of São Paulo, Brazil: comparisonsbetween polluted and unpolluted periods. Environ Monit Assess184:969–984. https://doi.org/10.1007/s10661-011-2013-y

Alonso MF, Longo K, Freitas S, Fonseca R, Marécal V, Pirre M, KlennerL (2010) An urban emission inventory for South America and itsapplication in numerical modeling of atmospheric chemical compo-sition at local and regional scales. Atmos Environ 44:5072–5083.https://doi.org/10.1016/j.atmosenv.2010.09.013

Alves CA, Lopes DJ, Calvo AI, Evtyugina M, Rocha S, Nunes T (2015)Emissions from light-duty diesel and gasoline in-use vehicles mea-sured on chassis dynamometer test cycles. Aerosol and Air Qual Res15:99–116. https://doi.org/10.4209/aaqr.2014.01.0006

Alvim DS, Gatti LV, Corrêa SM, Chiquetto JB, Rossatti CS, Pretto A,Santos MH, Yamazaki A, Orlando JP, Santos GM (2017) Mainozone-forming VOCs in the city of Sao Paulo: observations, model-ling and impacts. Air Qual Atmos Health 10:421–435. https://doi.org/10.1007/s11869-016-0429-9

Andrade MF, Miranda RM, Fornaro A, Kerr A, Oyama B, Andre PA,Saldiva P (2012) Vehicle emissions and PM2.5 mass concentrationsin six Brazilian cities. Air Qual Atmos Health 5:79–88. https://doi.org/10.1007/s11869-010-0104-5

Andrade MF, Ynoue RY, Freitas ED, Todesco E, Vara Vela A, Ibarra S,Martins LD, Martins JA, Carvalho VSB (2015) Air quality forecast-ing system for southeastern Brazil. Front Environ Sci 23:1–14.https://doi.org/10.3389/fenvs.2015.00009

Andrade MF, Kumar P, Freitas ED, Ynoue RY, Martins J, Martins LD,Nogueira T, Perez-Martinez P, Miranda RM, Albuquerque T,Gonçalves FLT, Oyama B, Zhang Y (2017) Air quality in the mega-city of São Paulo: evolution over the last 30 years and future per-spectives. Atmos Environ 159:66–82. https://doi.org/10.1016/j.atmosenv.2017.03.051

Appel KW, Napelenok SL, Foley KM, Pye HOT, Hogrefe C, LueckenDJ, Bash JO, Roselle SR, Pleim JE, Foroutan H, Hutzell WT,Pouliot GA, Sarwar G, Fahey KM, Gantt B, Gilliam RC, HeathNK, Kang D, Mathur R, Schwede DB, Spero TL, Wong DC,Young JO (2017) Description and evaluation of the communitymultiscale air quality (CMAQ) modeling system version 5.1.Geosci Model Dev 10:1703–1732. https://doi.org/10.5194/gmd-10-1703-2017

Binkowski FS, Shankar U (1995) The regional particulate matter model:1. Model description and preliminary results. J Geophys Res-Atmos100:26191–26209. https://doi.org/10.1029/95JD02093

Borge R, López J, Lumbreras J, Narros A, Rodríguez E (2010) Influenceof boundary conditions on CMAQ simulations over the IberianPeninsula. Atmos Environ 44:2681–2695. https://doi.org/10.1016/j.atmosenv.2010.04.044

Borge R, Lumbreras J, Pérez J, de la Paz D, Vedrenne M, de Andrés JM,Rodríguez ME (2014) Emission inventories and modeling require-ments for the development of air quality plans. Application toMadrid (Spain). Sci Total Environ 466-467:809–819. https://doi.org/10.1016/j.scitotenv.2013.07.093

Borge R, Lumbreras J, Rodríguez E (2008) Development of a high-resolution emission inventory for Spain using the SMOKEModelling system: a case study for the years 2000 and 2010.Environ Model Softw 23:1026–1044. https://doi.org/10.1016/j.envsoft.2007.11.002

Byun D, Schere KL (2006) Review of the governing equations, compu-tational algorithms, and other components of the Models-3 commu-nity multiscale air quality (CMAQ) modeling system. Appl MechRev 59:51–77. https://doi.org/10.1115/1.2128636

Castanho ADA, Artaxo P (2001) Wintertime and summertime São Pauloaerosol source apportionment study. Atmos Environ 35:4889–4902.https://doi.org/10.1016/S1352-2310(01)00357-0

Cesaroni G, Badaloni C, Gariazzo C, Stafoggia M, Sozzi R, Davoli M,Forastiere F (2013) Long-term exposure to urban air pollution andmortality in a cohort of more than a million adults in Rome. EnvironHealth Persp 121:324–331. https://doi.org/10.1289/ehp.1205862

CETESB (2009) Qualidade do ar no estado de São Paulo, 2008.Companhia Ambiental do Estado de São Paulo (São Paulo StateEnvironmental Protection Agency). http://ar.cetesb.sp.gov.br/publicacoes-relatorios/ (accessed April 20, 2018)

CETESB (2017) Qualidade do ar no estado de São Paulo, 2016.Companhia Ambiental do Estado de São Paulo (São Paulo StateEnvironmental Protection Agency). http://ar.cetesb.sp.gov.br/publicacoes-relatorios/ (accessed April 20,2018)

Chen D, Liu X, Lang J, Zhou Y, Wei L, Wang X, Guo X (2017)Estimating the contribution of regional transport to PM2.5 airpollution in a rural area on the North China plain. Sci TotalEnviron 583:280–291. https://doi.org/10.1016/j.scitotenv.2017.01.066

Cohen AJ, Brauer M, Burnett R, Anderson HR, Frostad J, Estep K,Balakrishnan K, Brunekreef B, Dandona L, Dandona R, Feigin V,Freedman G, Hubbell B, Jobling A, Kan H, Knibbs L, Liu Y, MartinR,Morawska L, Pope CA III, Shin H, Straif K, Shaddick G, ThomasM, Van Dingenen R, Van Donkelaar A, Vos T, Murray CJL,Forouzanfar MH (2017) Estimates and 25-year trends of the globalburden of disease attributable to ambient air pollution: an analysis ofdata from the global burden of diseases study 2015. Lancet 389:1907–1918. https://doi.org/10.1016/S0140-6736(17)30505-6

Crouse DL, Peters PA, Van Donkelaar A, Goldberg MS, Villeneuve PJ,Brion O, Khan S, Atari DO, Jerrett M, Pope CA III, Brauer M,Brook JR, Martin RV, Stieb D, Burnett RT (2012) Risk ofnonaccidental and cardiovascular mortality in relation to long-termexposure to low concentrations of fine particulatematter: a Canadiannational-level cohort study. Environ Health Persp 120:708–714.https://doi.org/10.1289/ehp.1104049

Fraser MP, Cass GR (1998) Detection of excess ammonia emissions fromin-use vehicles and the implications for fine particle control. EnvironSci Technol 32:1053–1057. https://doi.org/10.1021/es970382h

Gouveia N, Junger WL (2018) Effects of air pollution on infant andchildren respiratory mortality in four large Latin-American cities.Environ Pollut 232:385–391. https://doi.org/10.1016/j.envpol.2017.08.125

Grell GA, Peckham SE, Schmitz R, Mckeen SA, Frost G, SkamarockWC, Eder B (2005) Fully coupled Bonline^ chemistry within theWRF model. Atmos Environ 39:6957–6975. https://doi.org/10.1016/j.atmosenv.2005.04.027

Hetem IG, AndradeMF (2016) Characterization of fine particulate matteremitted from the resuspension of road and pavement dust in themetropolitan area of São Paulo, Brazil. Atmosphere 7(31):1–10.https://doi.org/10.3390/atmos7030031

Environ Sci Pollut Res (2018) 25:36555–36569 36567

Hogrefe C, Roselle SJ, Bash JO (2017) Persistence of initial conditions incontinental scale air quality simulations. Atmos Environ 160:36–45.https://doi.org/10.1016/j.atmosenv.2017.04.009

Hoshyaripour G, Brasseur G, Andrade MF, Gavidia-Calderón M,Bouarar I, Ynoue RY (2016) Prediction of ground-level ozone con-centration in São Paulo, Brazil: deterministic versus statistic models.Atmos Environ 145:365–375. https://doi.org/10.1016/j.atmosenv.2016.09.061

Jiang X, Yoo E (2018) The importance of spatial resolutions of commu-nity multiscale air quality (CMAQ) models on health impact assess-ment. Sci Total Environ 627:1528–1543. https://doi.org/10.1016/j.scitotenv.2018.01.228

Johnson TM, Guttikunda S, Wells GJ, Artaxo P, Bond TC, Russell AG,Watson JG, West J (2011) Tools for improving air quality manage-ment: a review of top-down source apportionment techniques andtheir application in developing countries. World Bank EnergySector Management Assistance Program, Formal Report 339/11,The World Bank Group, Washington, DC

Jun M, Stein ML (2004) Statistical comparison of observed and CMAQmodeled daily sulfate levels. Atmos Environ 38(27):4427–4436.https://doi.org/10.1016/j.atmosenv.2004.05.019

Krewski D, Jerrett M, Burnett RT, Ma R, Hughes E, Shi Y, Turner MC,Pope III CA, Thurston G, Calle EE, Thun MJ (2009) Extendedfollow-up and spatial analysis of the American Cancer Society studylinking particulate air pollution and mortality. HEI Research Report140. Boston: Health Effects Institute

Laden F, Schwartz J, Speizer FE, Dockery DW (2006) Reduction in fineparticulate air pollution and mortality: extended follow-up of theHarvard six cities study. Am J Resp Crit Care 173:667–672.https://doi.org/10.1164/rccm.200503-443OC

Lents J, Nikkila N, Osses M, Davis N (2004) São Paulo vehicle activitystudy. International Sustainable Systems Research Center (ISSRC),California, USA, 86 p. Center (ISSRC), California, EUA, 86p

Li J, ZhangM, Tang G,Wu F, Alvarado LMA, Vrekoussis M, Richter A,Burrows JP (2018) Investigating missing sources of glyoxal overChina using a regional air quality model (RAMS-CMAQ). JEnviron Sci 71:108–118. https://doi.org/10.1016/j.jes.2018.04.021

Longo KM, Thompson AM, Kirchhoff VWJH, Remer LA, Freitas SR,Dias MAFS, Artaxo P, Hart W, Spinhirne JD, Yamasoe MA (1999)Correlation between smoke and tropospheric ozone concentration inCuiabá during smoke, clouds, and radiation-Brazil (SCAR-B). JGeophys Res-Atmos 104:12113–12129. https://doi.org/10.1029/1999JD900044

Martins LD, Andrade MF, Freitas ED, Pretto A, Gatti LV, AlbuquerqueEL, Tomaz E, Guardani ML, Martins MHRB, Junior OMA (2006)Emission factors for gas-powered vehicles traveling through roadtunnels in São Paulo, Brazil. Environ Sci Techno 40:6722–6729.https://doi.org/10.1021/es052441u

Martins JA,Martins LD, Freitas ED,Mazzoli CR, Hallak R, AndradeMF(2008) Aplicação de imagens de satélite no desenvolvimento deinventários de Emissão de alta resolução. Paper presented at theXV Congresso Brasileiro De Meteorologia. São Paulo, Brazil

Martins JA,Mazzoli CR, OliveiraMGL, Ynoue RY, AndradeMF, FreitasED,Martins LD (2010) Desenvolvimento de inventários de emissãode alta resolução: Intensidade de luzes noturnas e distribuiçãoespacial de veículos. Paper presented at the XVI CongressoBrasileiro de Meteorologia. Belém, Brazil

Miranda RM, Andrade MF, Fornaro A, Astolfo R, de Andre PA, SaldivaP (2012) Urban air pollution: a representative survey of PM2.5 massconcentrations in six Brazilian cities. Air Qual Atmos Health 5:63–77, 2012. https://doi.org/10.1007/s11869-010-0124-1

Nenes A, Pandis SN, Pilinis C (1998) ISORROPIA: a new thermody-namic equilibrium model for multiphase multicomponent inorganic

aerosols. Aquat Geoch 4:123–152. https://doi.org/10.1023/A:1009604003981

Nenes A, Pilinis C, Pandis SN (1999) Continued development and testingof a new thermodynamic aerosol module for urban and regional airquality models. Atmos Env 33:1553–1560. https://doi.org/10.1016/S1352-2310(98)00352-5

Pacheco MT, Parmigiani MMM, Andrade MF, Morawska L, Kumar P(2017) A review of emissions and concentrations of particulate mat-ter in the three metropolitan areas of Brazil. J Transp Health 4:53–72. https://doi.org/10.1016/j.jth.2017.01.008

Pepe N, Pirovano G, Lonati G, Balzarini A, Toppetti A, Riva GM,Bedogni M (2016) Development and application of a high resolu-tion hybrid modelling system for the evaluation of urban air quality.Atmos Environ 141:297–311. https://doi.org/10.1016/j.atmosenv.2016.06.071

Pope CA III, Burnett RT, ThunMJ, Calle EE, Krewski D, Ito K, ThurstonGD (2002) Lung cancer, cardiopulmonary mortality, and long-termexposure to fine particulate air pollution. J Am Med Assoc 287:1132–1141. https://doi.org/10.1001/jama.287.9.1132

Pope CA III, Burnett RT, Thurston GD, Thun MJ, Calle EE, Krewski D,Godleski JJ (2004) Cardiovascular mortality and long-term expo-sure to particulate air pollution: epidemiological evidence of generalpathophysiological pathways of disease. Circulation 109:71–77.https://doi.org/10.1161/01.CIR.0000108927.80044.7F

Reid JS, Hobbs PV, Liousse C, Martins JV, Weiss RE, Eck TF (1998)Comparisons of techniques for measuring shortwave absorption andblack carbon content of aerosols from biomass burning in Brazil. JGeophys Res-Atmos 103:32031–32040. https://doi.org/10.1029/98JD00773

Samaali M, Moran MD, Bouchet VS, Pavlovic R, Cousineau S, Sassi M(2009) On the influence of chemical initial and boundary conditionson annual regional air quality model simulations for North America.Atmos Environ 43:4873–4885. https://doi.org/10.1016/j.atmosenv.2009.07.019

Sánchez-Ccoyllo OR, Ynoue RY, Martins LD, Astolfo R, Miranda RM,Freitas ED, Borges AS, Fornaro A, Freitas H, Moreira A, AndradeMF (2009) Vehicular particulate matter emissions in road tunnels inSao Paulo, Brazil. Environ Monit Assess 149:241–249. https://doi.org/10.1007/s10661-008-0198-5

Seinfeld JH, Pandis NS (2006) Atmospheric chemistry and physics: fromair pollution to climate change, 2nd edn. Wiley – IntersciencePublication, USA

Silva Junior RS, Andrade MF (2013) Validação de poluentesfotoquímicos e inclusão do inventário de emissões no modelo dequalidade do ar WRF/CHEM, para a região metropolitana de SãoPaulo. Rev Brasil Meteorol 28:105–121. https://doi.org/10.1590/S0102-77862013000100010

Shimada S, Ohsawa T, Chikaoka S, Kozai K (2011) Accuracy of the windspeed profile in the lower PBL as simulated by the WRF model.SOLA 7:109–112. https://doi.org/10.2151/sola.2011-028

Syrakov D, Prodanova M, Georgieva E, Etropolska I, Slavov K (2016)Simulation of European air quality by WRF–CMAQ models usingAQMEII-2 infrastructure. J Comput Appl Math 293:232–245.https://doi.org/10.1016/j.cam.2015.01.032

The World Bank (2018) Urban population (% of total). https://data.worldbank.org/indicator/SP.URB.TOTL.IN.ZS (accessed April20, 2018)

Vara Vela A, Andrade MF, Kumar P, Ynoue RY, Muñoz AG (2016)Impact of vehicular emissions on the formation of fine particles inthe Sao Paulo metropolitan area: a numerical study with the WRF-Chem model. Atmos Chem Phys 16:777–797. https://doi.org/10.5194/acp-16-777-2016

36568 Environ Sci Pollut Res (2018) 25:36555–36569

Yarwood G, Rao S, Yocke M, Whitten G (2005) Updates to the carbonbond chemical mechanism: CB05. Final Report to the US EPA, RT-0400675. Available at http://www.camx.com/publ/pdfs/CB05_Final_Report_120805.pdf

Ynoue RY, Andrade MF (2004) Size-resolved mass balance of aerosolparticles over the São Paulo metropolitan area of Brazil. Aerosol SciTechnol 38(S2):52–62. https://doi.org/10.1080/02786820490466756

Wang L, Wei Z, Wei W, Fu JS, Meng C, Ma S (2015a) Source apportion-ment of PM2.5 in top polluted cities in Hebei, China using theCMAQ model. Atmos Environ 122:723–736. https://doi.org/10.1016/j.atmosenv.2015.10.041

Wang Z, Pan L, Li Y, Zhang D, Ma J, Sun F, Xu W, Wang X (2015b)Assessment of air quality benefits from the National PollutionControl Policy of thermal power plants in China: a numerical sim-ulation. Atmos Environ 106:288–304. https://doi.org/10.1016/j.atmosenv.2015.01.022

Zang H, Pu Z, Zang X (2013) Examination of errors in near-surfacetemperature and Wind from WRF numerical simulations in regionsof complex terrain. Weather Forecast 28:893–914. https://doi.org/10.1175/WAF-D-12-00109.1

Environ Sci Pollut Res (2018) 25:36555–36569 36569