Introduction
The Sao Paulo Metropolitan Area (SPMA), in the southeast region of Brazil,
is considered a megalopolis comprised of Sao Paulo city and more than 38 municipalities. One of the main concern in the SPMA is the occurrence of
violations of air quality standards for ozone and fine particles at
different air quality stations from the Sao Paulo Environmental Agency
(CETESB). The air pollutant emissions in the SPMA are related to the burning
of the following fuels: ethanol, gasohol (gasoline with 25 % ethanol), and diesel.
Recent work of Carvalho et al. (2015) reported a substantial increase in
number of road vehicles from 1 million in 2000 to almost 7 million in 2014,
together with an overview of the pollutants concentration, fuel use in the
SPMA and the relationship between the emissions and the improvement in the
air quality in past years.
They constitute the main cause of impairment to air quality in the SPMA, but
the number of air quality standard violations has decreased for almost all
pollutants with the exception of PM2.5 and O3. Both these
pollutants are impacted by the vehicular emissions and have experienced an
increase in the number of violations of local air quality standards as
discussed in detail by Carvalho et al. (2015). Pérez-Martínez et
al. (2015) have analysed the monthly mean values for the regulated
pollutants from 2000 to 2013 for the air quality stations in the SPMA. They
found a decrease in the average concentration of NOx, CO, and PM10 by
0.65, 0.37, and 0.71 % month-1, respectively, although the sales of
the fuels (gasoline, ethanol, and diesel) had increased by 0.26, 1.96, and
0.38 % month-1, respectively.
A recent report from CETESB (CETESB, 2013) highlighted that, in 2012, the
vehicles contributed with about 40 % of the total PM10 mass
concentrations through direct emissions. If we consider the secondary
aerosols, which were about 25 % of PM10 as estimated by CETESB (2013), these were mainly found to be formed by chemical reactions between
gases released from exhaust of vehicles.
The implementation of the Program for the Control of Vehicular Emission
(PROCONVE) established by the Brazilian Government in the 80's, enforcing
measures such as use of catalytic converters and ethanol as additive to
gasoline in substitution of tetraethyllead, led to a decrease in emissions of
CO and VOCs and hence their ambient concentration. Although the emissions
have been controlled by regulations, the number of vehicles has increased
substantially and faster than the replacement of the old vehicles by the new
ones (Pérez-Martínez et al., 2014). According to CETESB (2013), the
road vehicles contributed up to about 97, 87, and 80 % of CO, VOCs, and NOx
emissions in 2012, respectively, being most of NOx associated to diesel
combustion and most of CO and VOCs from gasohol and ethanol combustion.
Receptor modelling studies applied to six capital cities in Brazil (Andrade
et al., 2012) showed that only 13 % of PM2.5 in the SPMA is
associated to the emission by the industrial processes (oil burning and
secondary aerosols).
To date, many studies assessing the impact of biofuels on the air quality
have been performed in Brazil. For example, Anderson (2009) conducted a
review concerning the use of ethanol fuel in Brazil. His work highlighted
that the atmospheric concentrations of acetaldehyde and ethanol are much
higher in Brazil in comparison with the other areas of the world. Costa and
Sodré (2010) showed that exhaust emissions of hydrous ethanol reduced CO
and Hydrocarbons (HC), but increased CO2 and NOx levels.
A number of past studies has shown the significant participation of the
carbonaceous compounds in the concentration of fine particles in the SPMA
(Albuquerque et al., 2011; Miranda and Andrade, 2005; Ynoue and Andrade,
2004; Castanho and Artaxo, 2001). Studies conducted on ambient air pollution
in the SPMA have also shown that BC explains 21 % of mass concentrations
of fine particles (PM2.5; ≤ 2.5 µm in diameter) compared with
40 % of organic carbon (OC), 20 % of sulfates, and 12 % of soil dust
(Andrade et al., 2012). Most of the observed ambient PM2.5 mass
concentration usually originates from precursors gases such as sulphur
dioxide (SO2), ammonia (NH3), nitrogen oxides (NOx), and volatile
organic compounds (VOCs) as well as through the physico-chemical processes
such as the oxidation of low-volatile hydrocarbons noted above transferring
to the condensed phase (McMurry et al., 2004; Heal et al., 2012). Since
these processes are often photo-chemically driven, the resultant aerosol
usually falls into the category of secondary photochemical pollutant (Jenkin
and Clemitshaw, 2000). Oxidation of VOCs can produce species of sufficiently
low vapour pressure to be condensable, leading to the formation of secondary
organic aerosol (SOA) (Kroll and Seinfeld, 2008). Fine particles in SPMA
have a great participation on its composition of SOA, formed from the
emissions of VOCs, which have the same origin of the primary compounds
involved in the formation of ozone, from the burning of fuels. The
participation of the biogenic emission is considered to be small in the
formation of particles in the metropolitan area of the city according to
previous studies of Martins et al. (2006).
The impact of the fine particles has been discussed in previous works, with
evaluation of the scattering and absorbing effects of the aerosol (e.g. Li
et al., 2005; Real and Sartelet, 2011). Vehicular emissions of particulate matter
(PM) in the SPMA have a high percentage of BC (Brito et al., 2013), which
after emitted to the atmosphere can enhance the absorption coefficient and
thus the attenuation rates.
Description of aerosol sampling campaigns performed at IAG-USP.
Parameter
Sampling frequency
Period of sampling
Sampling device
Aerosol mass size
24 h
July–September
MOUDI
distribution
impactor
PM2.5 and PM10
12 h
July–September
Dichotomous
concentration
sampler
OC and EC
12 h
August–September
Sunset OC-EC
concentration
analyser
One of the most important aspects of this work is the quantitative analysis
of the formation of PM2.5 and ozone (O3) in the SPMA. Photolysis of
O3 by ultraviolet light in the presence of water vapour is the main
source of hydroxyl radical (OH), the most important radical in the atmosphere
in terms of reactivity (Monks, 2005). At the same time, OH levels in the
atmosphere directly determine the oxidation rate of the precursors of
secondary aerosols. Oxidation products of VOCs and semi-VOCs by OH are the
most important precursors of SOA (Li et al., 2011a). Although VOCs and
NOx are precursors of both O3 and a fraction of atmospheric PM
(NO3- and secondary organics) while they influence indirectly the
formation of the rest of the secondary PM components like SO4=, their
control strategies that are optimal for O3 controls may even increase
PM2.5 concentrations (McMurry et al., 2004). Such an analysis is
important to evaluate the contribution of the vehicular fleet using different
kind of fuels to the concentration of fine particles. In this sense, a
numerical study with an adequate physical approach, representing particles in
the modelling system, is important to understand the formation of secondary
aerosols from primary emission of gases in a metropolitan area where the
composition of fuel in vehicular fleet has changed significantly over the
past years. Therefore, the goal of the present study is to evaluate the
impact of vehicular emissions on the formation of fine particles in the SPMA,
focusing especially on the potential formation of secondary particles from
the primary emission of gases coming from on-road vehicles. The impact of
aerosol particles on the ozone photochemistry is also examined by means of
numerical simulations. Measurements were performed to provide input data to
evaluate the modelling performance and estimate the vehicular emission
factors. Aerosol measurements were taken from field campaigns that were
carried out as part of the Narrowing the Uncertainties on Aerosol and Climate
Changes in Sao Paulo State (NUANCE-SPS) project
(http://nuance-lapat.iag.usp.br/). These campaigns took place
between July and September 2012. An online-coupled meteorology and chemistry
model, i.e. the Weather Research and Forecasting with Chemistry (WRF-Chem)
model, has been used to characterize and describe the physico-chemical
processes involved in both the formation and growth of new particles over the
SPMA in southern Brazil. The details of the experimental campaigns, WRF-Chem
model and emissions are described in Sect. 2. Results from modelling
experiments and comparison with measurements are presented in Sect. 3.
Finally, the summary and conclusions are given in Sect. 4.
Downtown area of the 3 km modelling domain (d03) showing the
locations of measurement sites and WRF topography in the vicinity of the
SPMA. Red dots represent sites with information on O3 and PM. Yellow
dots represent only sites with information on PM. Blue dot represents the
location of the IAG-USP's climatological station.
Description of measurement sites.
Site
Initials
Latitude
Longitude
Classification
Measured species
Nossa S. do O
NSO
-23.4796
-46.6916
Urban
PM10, O3
Santana
SAN
-23.5055
-46.6285
Urban
PM10
Parque D. Pedro
PDP
-23.5448
-46.6276
Urban
PM10, O3
Mooca
MOO
-23.5497
-46.5984
Urban
PM10, O3
Cerqueira Cesar
CCE
-23.5531
-46.6723
Urban
PM10
IAG-USP
IAG-USP
-23.5590
-46.7330
Suburban
PM10, PM2.5, OC, EC,
aerosol mass size distrib.a
IPEN-USP
IPEN-USP
-23.5662
-46.7374
Suburban
PM2.5, O3, NOx, CO
Ibirapuera
IBI
-23.5914
-46.6602
Suburban
PM10, O3, NOx, CO
Congonhas
CON
-23.6159
-46.6630
Urban
PM10, PM2.5
AF-IAG
AF-IAG
-23.6500
-46.6167
Suburban
T, RH, WS, WDb
Santo Amaro
SAM
-23.6545
-46.7095
Urban
PM10
Interlagos
INT
-23.6805
-46.6750
Urban
PM10, O3, T, RH, WS, WD
a Includes SO42-, NO3-, NH4+, Na+,
Cl- and PM10.
b T, RH, WS, and WD denote temperature, relative humidity, wind speed
and wind direction, respectively.
Methodology
Observational data sets
The study period starting from 7 August until 6 September 2012 was selected
for comparison with the modelled results (Sect. 2.2) due to the
availability of experimental data from the NUANCE-SPS project. The aim of
NUANCE-SPS was to evaluate the impact of emissions in the SPMA on the air
quality and changing climatic conditions, and feedback mechanisms between
climatic perturbations produced by both primary and secondary emissions and
urban atmospheric processes. Aerosol observation data sets used in this work
were collected using a Dichotomous sampler (Wedding et al., 1980) and a
Micro-Orifice Uniform Deposit Impactor (MOUDI, model 100; MSP Corporation;
Marple et al., 1986). The MOUDI impactor collected particles in 10 size
classes with nominal 50 % cut-off diameters: 10, 5.6, 3.2, 1.8, 1.0, 0.56,
0.32, 0.18, 0.1, and 0.06 µm. Particles smaller than 0.06 µm
were collected in a subsequent stage or after-filter. The samples collected
with the MOUDI impactor were deposited on a polycarbonate membrane filter
with 0.4 µm porous and for the Dichotomous sampler the substrate was
a teflon membrane filter with 2 µm porous. The after-filter in the
MOUDI impactor is a 33 mm teflon membrane filter, which was not submitted to
the reflectance analysis. The collected membrane filters sampled with the
Dichotomous and MOUDI samplers were analysed to the identification of trace
elements of mass through X-ray diffraction analysis, mass concentration
through gravimetric analysis, and black and organic carbon through
reflectance and thermo analysis using a thermal-optical transmittance (TOT)
(Sunset Laboratory Inc.; Birch and Cary, 1996). Ion concentrations were
evaluated through the ion chromatography analysis of the soluble material
collected on the membrane filters (sulphate, nitrate, ammonium, sodium, and
chloride). All these samplings were performed on the roof of the main
building of the Institute of Astronomy, Geophysics and Atmospheric Sciences
of the University of Sao Paulo (IAG-USP) (hereafter also referred as IAG-USP
measurement site or simply IAG-USP), which is inside a small green-park
(approximately 7.4 km2), with local traffic during the day and
surrounded by major roads with intense traffic by light and heavy-duty
vehicles (Nogueira et al., 2014). Table 1 lists the aerosol instrumentation
deployed roughly at the IAG-USP measurement site. In addition, ambient data
from the CETESB's air quality monitoring network and the IAG-USP's
climatological station (hereafter also referred as AF-IAG) were also
considered for evaluation of numerical simulations. The locations of
measurement sites are depicted in Fig. 1 whereas geographic coordinates,
urban-suburban classification, and the list of pollutants and meteorological
parameters monitored at each site is available in Table 2.
WRF-Chem model
The WRF-Chem model is a fully coupled online meteorological and chemical
transport model (Grell et al., 2005), supported by National Center for
Atmospheric Research (NCAR) of the USA and several other research
institutions around the world. This model is a system of two key components.
The WRF-Chem meteorological component, the Weather Research and Forecasting
(WRF), is a system configured for both research and operational
applications. The dynamical core used in this study is the Advanced Research
WRF (ARW). Model's equations into ARW are solved to non-hydrostatic
conditions on a fully compressible atmosphere. Further details on the
modelling system can be found on the WRF model website
(http://www.wrf-model.org). On the other hand, the WRF-Chem chemical
component treats chemical processes such as dry deposition, gas-phase
chemistry, photolysis rates, and aerosols chemistry. A detailed description
of the WRF-Chem model can be found on its website
(http://ruc.noaa.gov/wrf/WG11). Since both meteorological and chemical
components are fully coupled, the transport of all chemical species is
on-line. The gas-phase chemistry and aerosol modules employed in this study
are the Regional Acid Deposition Model, version 2 (RADM2) (Chang et al.,
1989) and the Modal Aerosol Dynamics Model for Europe – Secondary Organic
Aerosol Model (MADE–SORGAM) (Ackermann et al., 1998; Schell et al., 2001),
respectively. The inorganic species included in the RADM2 mechanism are 14 stable species, 4 reactive intermediates, and 3 abundant stable species
(oxygen, nitrogen, and water). Atmospheric organic chemistry is represented
by 26 stable species and 16 peroxy radicals. The RADM2 mechanism represents
organic chemistry through a reactivity aggregated molecular approach
(Middleton et al., 1990). Similar organic compounds are grouped together in
a limited number of model groups through the use of reactivity weighting.
The aggregation factors for the most emitted VOCs are given in Middleton et
al. (1990).
On the other hand, the most important process for the formation of secondary
aerosol particles is the homogeneous nucleation in the sulfuric acid-water
system. It is parameterized in MADE, following the method of Kulmala et al. (1998). Aerosol growth by condensation occurs in two steps: the production
of condensable material (vapour) by the reaction of chemical precursors, and
the condensation and evaporation of ambient volatile species on aerosols.
The inorganic chemistry system, based on the Model for an Aerosol Reacting
System (MARS) (Saxena et al., 1986) and its modifications by Binkowski and
Shankar (1995), calculates the chemical composition of a
sulphate-nitrate-ammonium-water aerosol according to equilibrium
thermodynamics. The organic aerosol chemistry is based on the SORGAM, which
assumes that SOA compounds interact and form a quasi-ideal solution (Grell
et al., 2005). The SOA formation in SORGAM follows the two-product approach
(Odum et al., 1996) where the oxidation of hydrocarbons produces two types
of modelled semivolatile compounds that are partitioned between the gas and
particle phases after considering the absorptive partitioning theory
(Pankow, 1994a, b). The primary organic aerosol (POA) in MADE is calculated
from the primary anthropogenic emission of OC. Then, one may calculate the
predicted OC concentration from the sum of both SOA and POA. The concurrent
organic matter (OM) can be obtained from the OC concentration by application
of a conversion factor. Brown et al. (2013) showed that the average OM : OC
ratio was 1.54 (with a standard deviation of 0.2) for sites with low amount
of secondary aerosol formation. It is important to note that this ratio can
change from one place to another. In areas impacted by biomass burning the
ratio can be higher. Gorin et al. (2006) assumed a ratio of 1.6 for the
conversion from OC to OM over an area that experiences a significant wood
smoke influence.
Selected WRF-Chem configuration options.
Atmospheric process
WRF-Chem option
Longwave radiation
RRTM
Shortwave radiation
Goddard
Surface layer
Monin–Obukhov
Land surface
Noah
Boundary layer
YSU
Cumulus clouds∗
Grell 3D
Cloud microphysics
Lin
Gas-phase chemistry
RADM2
Aerosol chemistry
MADE/SORGAM
Photolysis
Fast-J
∗ Outer domains only.
Model configuration
WRF-Chem version 3.6 was configured with three nested grid cells: coarse
(75 km), intermediate (15 km), and fine (3 km). The coarse grid cell
covered a big region of Brazil and also of the Atlantic Ocean. The
intermediate grid covered southeast Brazil while the fine grid cell covered
barely the SPMA and metropolitan areas nearest to it. Figure 1 shows the
arrangement of measurement sites and topography in the downtown area of the
3 km modelling domain. The initial and boundary meteorological conditions
are from the National Center for Environmental Prediction's Final Operational
Global Analysis with 1∘ of grid spacing, 26 vertical levels and are
available every 6 hours: 00:00, 06:00, 12:00, and 18:00 UTC
(http://rda.ucar.edu/datasets/ds083.2/). The initial and boundary
chemical conditions for representing gases and aerosols background
concentration were obtained from the Model for Ozone and Related chemical
Tracers, version 4 (MOZART-4; Emmons et al., 2010). This model was driven by
meteorological inputs from the Goddard Earth Observing System Model,
version 5 at a horizontal resolution of 1.9∘ × 2.5∘,
56 vertical levels that are also available every 6 hours. Table 3 lists the
WRF-Chem configuration options employed by this study.
WRF-Chem simulation with coupled primary aerosol (dust, sea salt and
anthropogenic) and gas (biogenic and anthropogenic) emission modules,
together with the direct effect of aerosol particles turned on, is performed
as the control simulation in order to evaluate the model performance
(hereafter referred to as Case_0). For secondary aerosols, a
simulation scenario (Case_1) with biogenic and anthropogenic
gases emission is performed to evaluate its formation potential. An
additional simulation (Case_2) is also performed to evaluate
the impact of aerosols on ozone photochemistry. Notation and description of
simulations are listed in Table 4. The first 7 days of each simulation
were not analysed and used for model spin-up.
Emissions
Anthropogenic emissions
Because on-road vehicles are the most important sources of air pollution in
southeast Brazil's metropolitan areas, particularly in SPMA where, according
to CETESB, more than 80 % of pollutant emissions are generated by
vehicular emissions; the anthropogenic emissions of trace gases and
particles in both 3 and 15 km modelling domains were considered to include
emissions only coming from on-road vehicles through the use of a vehicular
emission model developed by the IAG-USP's Laboratory of Atmospheric
Processes (LAPAt). Basically, this model considers the number of vehicles,
vehicular emission factors, and average driving kilometres for vehicle per
day as basic parameters for the calculation of exhaust emissions considering
different vehicle types (light-duty vehicles, heavy-duty vehicles, and
motorcycles) and different fuel types (ethanol, gasohol, combination of any
proportion of gasohol and ethanol, and diesel) according to CETESB (2012).
The details of this model are available in Andrade et al. (2015). In the
case of VOCs, there are another two relevant emissions (fuel transfer and
evaporative processes) associated with the vehicles, besides the exhaust
Description of WRF-Chem simulations.
Label
Description
Case_0
Emission of gases
(Baseline simulation)
Emission of aerosols
Aerosol-radiation feedback turned on
Case_1
Emission of gases
No emission of aerosols
Aerosol-radiation feedback turned on
Case_2
Emission of gases
Emission of aerosols
Aerosol-radiation feedback turned off
emissions. Because of the complexities in the spatial representation due to numerous factors such as emissions at service stations, such emission
sources are assumed to be emitted by exhaust of vehicles for the sake of
simplicity. The vehicular fleet and intensity of use data sets are provided
by the National Department of Traffic (DENATRAN) and the Sao Paulo
Transporte (SPTrans), respectively. Emission factors for road vehicles for
most pollutants were considered from previous studies performed inside road
tunnels (i.e. Janio Quadros, referred as JQ tunnel, and the tunnel 3 of the
Rodoanel Mario Covas that is referred hereafter as RA tunnel) located within
the SPMA (Pérez-Martínez et al., 2014; Nogueira et al., 2014).
However, emission factors for VOCs are considered from dynamometer protocols
(CETESB, 2010). VOCs and PM speciation profiles used by gas-phase and
aerosol chemical modules were also obtained from NUANCE-SPS experimental
campaigns performed in 2011 (tunnel measurements) and 2012 (ambient data).
It is important to note that due to the lack of information on vehicular
emission factors and intensity of use for most of the other metropolitan
areas inside both modelling domains (e.g. the Campinas Metropolitan Area,
which is shown by the second largest grey stain in Fig. 2), the calculation
of vehicular emissions for these urban areas was carried out on the basis of
the parameters found for the SPMA. The number of vehicles in any modelling
domain is calculated from the sum of the number of vehicles in each one of
the main urban areas inside the modelling domain in question.
Emission rates for Aromatic VOCs at 19:00 UTC in the 3 km modelling
domain.
Spatial distribution of emissions for the 3 km modelling domain resolution
was based on road density products compiled by the OpenStreetMap project and
extracted from the Geofabrik's free download server
(http://download.geofabrik.de). Urban areas were assumed to allocate high
emissions since these concentrate a road density greater than other areas.
In the case of the 15 km modelling domain, emissions are based on night-time
lights data from the Defense Meteorological Satellite Program
(http://ngdc.noaa.gov/eog/dmsp/downloadV4composites.html). These images are
30 arc second grids, spanning from -180∘ to +180∘ longitude and
-65∘ to +75∘ latitude and contain the lights from cities, towns
and other sites with persistent lighting, including gas flares. Cleaned up
night-time light points with no ephemeral events such as forest fires are
used to allocate emissions. To estimate the number of vehicles in each grid
point of both domains, the sum of individual intensities at each point (i.e.
total road length for the 3 km modelling domain and night-time light for the
15 km modelling domain) is firstly normalized by the total fleet, and then
distributed uniformly using the total fleet distribution so that emissions
in urban areas are mainly represented by emissions coming from their
vehicles. Furthermore, due to the complexity involved in describing the
temporal variation of emissions at each grid point, median values for
vehicular traffic obtained from measurements inside the JQ and RA tunnels
(Pérez-Martínez et al., 2014) were used for distributing the
emissions during the day in both domains. This approximation followed the
approach used by Fast et al. (2006) where emission profiles were calculated
from median diurnal variations on weekdays and weekends. We have applied the
same constant diurnal cycle at all grid points where emissions have values
greater than zero. VOC and PM emission profiles were assumed to be the same
as for CO and NOx emission profiles since these pollutants are also
characteristic tracers of emissions of light-duty and heavy-duty vehicles,
respectively. Figure 2 shows the maximum hourly emission rates for aromatic
VOCs in the 3 km modelling domain. Anthropogenic emissions were not
considered in the 75 km modelling domain.
The Another Assimilation System for WRF-Chem (AAS4WRF) chemical emissions
pre-processor developed by the Latin American Observatory (OLE2; Muñoz
et al., 2010, 2012) was used to scale emission rates on WRF curvilinear
coordinates. AAS4WRF is appropriate to write chemical emission rates from
both surface and elevated sources in the proper WRF data file format,
providing an alternative tailored way to assimilate emissions to WRF-Chem.
The method is explained in the OLE2 Wiki pages in detail
(http://www.cmc.org.ve/mediawiki/index.php?title=Calidad_de_Aire).
Other emissions
Biogenic emissions are calculated online based on the Guenther scheme
(Guenther et al., 1993, 1994). The Guenther biogenic emissions model
calculates the emission rates using temperature, photo-synthetically active
radiation flux and land-use data as the U.S. Geological Survey (USGS)
land-use cover system classification if coupled with the WRF model. However,
as indicated in the WRF-Chem emissions guide
(http://ruc.noaa.gov/wrf/WG11/Emission_guide.pdf), several
key chemical species would have been representing relatively low emission
rates because of the limited vegetation types in the simulation, and thus
their impacts are anticipated to be much lower than those from vehicular
emissions.
Dust and sea salt emissions are calculated online following the works of
Ginoux et al. (2001) and Gong (2003), respectively. The calculation of
Ginoux et al. (2001) for the uplifting of dust particles is based on the
surface wind speed, wetness, and information on soil characteristics. The
model then solves the continuity equation including the emission, chemistry,
advection, convection, diffusion, dry deposition, and wet deposition of each
species. The parameterization of sea salt aerosol source function of Gong (2003) is an extended parameterization of Monahan et al. (1986), which
scales the generation of marine aerosols from mechanical disruption of wave
crests by the wind and sea surface covered by whitecaps.
Results and discussion
Characterization of meteorological conditions
In order to study and understand the spatial and temporal variability of
atmospheric aerosols, O3, and other pollutants (i.e. CO, NOx) during
the study period, it was first necessary to analyse the behaviour of main
meteorological systems acting on the atmospheric environment of the SPMA and
surrounding areas.
Hourly accumulated precipitation and relative humidity observed at
the IAG-USP's climatological station during the study period.
According to the monthly climate reports from the IAG-USP's Climate Research
Group (GrEC), the observed precipitation rates were lower than the
climatological value in SPMA (anomaly of -38.6 mm) and larger part of the
Sao Paulo State during August 2012. Negative anomalies on the precipitation
were caused by the intensification of the South Atlantic Subtropical High
(SASH). These conditions established an easterly wind anomaly pattern at the
850 hPa level. Conditions were unfavourable for relative humidity coming from
the Amazon due to the Low Level Jet (LLJ) and less intense Alisian winds in
the Tropical Atlantic (GrEC, 2012a). However, the action of frontal systems
favoured the rain accumulation in September 2012, mainly in western Sao Paulo
State where the greater positive amount of anomalies were observed.
Precipitation events were predominantly observed during the second half of
the month. In this case, the wind pattern showed an opposite configuration
to that observed in August 2012 as a result of the weakening of the SASH
(GrEC, 2012b). The IAG-USP's climatological station recorded an accumulated
precipitation of about 1.3 mm on 3 days of occurrence (28 and 30 August and 4 September 2012) and an easterly wind pattern with a median
intensity of 2 m s-1 during the period between 7 August and 6 September 2012. Figure 3 shows the hourly accumulated precipitation and
relative humidity observed at the IAG-USP's climatological station.
HYSPLIT 3-day backward trajectories and locations of fires in
Sao Paulo State and part of central-west region of Brazil. Pink markers
represent the observed fire locations during the study period considering
different satellite products (GOES, AQUA, TERRA, NOAA). The backward
trajectories starting at IAG-USP were calculated for the days 9 and 31 August
and 5 September 2012 at three different altitudes: 500 m (red lines),
1000 m (blue lines), and 2000 m (green lines).
Analysis of aerosol species
Aerosol analysis included species such as organic carbon (OC), elemental
carbon (EC), sulphate, nitrate, ammonium, sodium and chloride in addition to
other elemental constituent of PM. All the samplings for these species were
performed at IAG-USP. Results showed that the major contributors to the
concentration of fine particles are OM (55.7 %; OM : OC ratio of 1.5 found
by Brito et al., 2013) and EC (15 %), followed by sulphate (2.9 %),
ammonium (2.1 %), sodium (1.9 %), nitrate (0.5 %), and chloride
(0.3 %). The remaining mass (21.6 %) is calculated by determining the
difference between the total mass of PM2.5 (from the gravimetric
analysis) and the sum of the masses of seven individual compounds, as noted
above. Part of this remaining mass is related to the water content of
aerosols (Andrade et al., 2012).
On the other hand, PM2.5, PM10 and size distribution of particles
measured at IAG-USP show that the study period was characterized by a
reduction in the concentrations up to the end of August 2012 when their
minimum values were achieved. This reduction was related to the action of a
semi-stationary front between the coasts of Sao Paulo and Parana States.
After the passage of this system, aerosol concentrations have significantly
increased what could be related to an increase in relative humidity once the
SASH system is moved away from the continent, as well as the transport of
aerosol particles produced by forest fires in the central-west region of
Brazil and the Sao Paulo State. Several studies have shown the contribution
of forest fires on the atmospheric aerosol concentrations in SPMA
(Vieira-Filho et al., 2013; Vasconcellos et al., 2010). One way to
qualitatively evaluate the contribution of forest fires on aerosol
concentrations is by using the air mass trajectories. The Hybrid
Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Draxler
and Hess, 1998) was used to calculate backward trajectories of air masses in
order to identify atmospheric transport of air mass from forest fire areas.
Figure 4 shows the 3-day backward trajectories of air masses starting at
IAG-USP for the days 9 and 31 August and 5 September, when increases in the
OC and EC concentrations were observed at IAG-USP. The pink markers on the
map represent the observed fire locations during the study period
considering different satellite products (GOES, AQUA, TERRA, NOAA).
Daily (top), diurnal (bottom), and nocturnal (middle) mean
concentrations for EC, OC, PM10, PM2.5–10, PM2.5 (left
panels), and Na, Fe2SO3, SiO2, K2O, and S (right
panels). The PM2.5–10 aerosol variable is defined as particulate matter
with aerodynamic diameter between 2.5 and 10 µm. The grey line
indicates the WHO air quality standard for PM2.5 (25 µg m-3).
Figure 5 shows the concentration of OC, EC and some species of PM2.5 during the study period at IAG-USP. We can observe 11 exceedances of
PM2.5 concentration with respect to the air quality standard of 25 µg m-3 (see grey line in Fig. 5a) established by the World
Health Organization (WHO). These exceedances have mainly occurred at the
beginning and at the end of the study period when an increase in the
concentrations of OC and EC were observed. The increasing organic matter
could be associated to traffic incidents which may raise the emissions,
which in case of less favourable meteorological conditions (e.g. lower height
of lower planetary boundary layer, PBL, or slow transport of air pollutants)
may have led to a more efficient formation of secondary particles. Castanho
and Artaxo (2001) analysed the behaviour of the aerosol composition in the SPMA
and showed the increase in the concentration of inorganic and organic
material in the winter season compared to the summer season, explaining this
behaviour with the meteorological characteristics: dry conditions with low
height inversion layer in the wintertime and a rainy summer.
Size distributions of aerosol mass indicate that the majority of sulphate,
ammonium and PM10 mass concentration is distributed in the size range
with diameters between 0.1 and 1 µm, commonly known as accumulation
mode particles (Kumar et al., 2010). In the cases of nitrate, sodium, and
chloride, most part of mass was concentrated in particles with diameters
greater than 1 µm.
The predicted average of wind vectors at 10 m and temperature at
2 m from the baseline simulation (Case_0) for the whole study
period in the 3 km modelling domain. Blue dots represent the locations of
the measurement sites, whereas cyan numbers represent the observed average
temperature in those sites: 17.7 ∘C in AF-IAG and 17.8 ∘C in INT.
Performance statistics for WRF-Chem predictions at all
sites∗.
Index
PM2.5
PM10
O3
NOx
CO
T
RH
WS
WD
MB
-8.84
-14.10
-0.85
-8.75
-0.27
0.65
-5.74
0.54
31.12
MFB (%)
-47.62
-38.19
22.63
12.68
-32.53
1.94
-7.95
41.21
31.66
MFE (%)
47.90
39.90
72.85
82.82
80.93
14.16
23.84
71.12
54.40
RMSEUB
6.83
10.59
27.45
30.35
0.57
3.21
20.06
1.08
79.38
R
0.73
0.72
0.63
0.42
0.54
0.71
0.62
0.41
0.43
∗ Values are averaged from all the individual indexes found at the
measurement sites. Individual indexes are calculated from both hourly
observed and predicted values.
Comparison of baseline simulation with observations
All the numerical results presented in this section, for the purpose of
comparison with the measurements, were obtained from the baseline simulation
(Case_0). The predicted temperature, humidity, and 10 m wind
speed and direction have been compared to measurements from the AF-IAG and
INT measurement sites. Overall, the model captured the diurnal variation of
temperature, relative humidity, and wind directions reasonably well.
However, the predicted wind speeds were higher than the observed values. To
evaluate the model performance in solving the meteorology and chemical
species, we computed the statistics correlation coefficient (R), mean bias
(MB), mean fractional bias (MFB), mean fractional error (MFE), and root mean
square error UB (RMSEUB). The definitions of these statistics are given
in the Appendix. Figure 6 shows the predicted average of 10 m wind vectors and
2 m temperature for the whole study period in the 3 km modelling domain.
Blue dots represent the locations of AF-IAG and INT sites, while the numbers
in cyan indicate the observed average temperatures (i.e. 17.7 ∘C
at AF-IAG and 17.8 ∘C at INT). On an average, the predicted wind
direction was easterly in SPMA, which has somewhat affected the spatial
distribution of aerosol particles as examined later in this section.
The observed and predicted daily variations of PM2.5
concentrations at three sites in SPMA for the 3 km modelling domain.
Likewise, the statistics used to quantify the model performance in the
representation of PM concentration show that, in general, most of
prediction-observation pairs present good correlation coefficients, mainly
those for PM10, but with negative biases and standard deviations lower
than those for observations (see Fig. 10). Table 5 summarizes the
performance statistics used in this study showing comparisons between
WRF-Chem predictions and observations. The evaluation of WRF-Chem
predictions for meteorology and chemical species on a site-by-site basis is
presented in the Sects. 1 and 2, respectively, of the Supplement. Figures 7, 8, and 9 show the observed and predicted temporal
variations of PM2.5, PM10, and O3 concentrations at 3, 10 and
6 sites in the SPMA, respectively, with some measurement sites sharing the
same grid point for comparisons due to the geographical proximity (e.g. the
sites IAG-USP and IPEN-USP both separated around 900 m from each other).
The observed and predicted daily variations of PM10
concentrations at 10 sites in SPMA for the 3 km modelling domain.
These figures suggest that predicted concentrations did not present any
significant spatial variation in the downtown SPMA and were generally
underestimated when compared to measurements. This under prediction could be
associated with an underestimation on the vehicular emissions as well as
other emission sources (e.g. emissions coming from industry) that are
disregarded in this study, in addition to predicted surface winds more
intense than those observed, leading to a dilution of aerosol particles in
the SPMA. The high concentrations of PM2.5 and PM10 observed at
the beginning and at the end of the study period, whose variability and
trends were reasonably well captured by the model, could be related with the
emission of high aerosol loadings due to traffic incidents as well as the
establishment of lower PBL heights, commonly observed under post-frontal
situations. The results for this simulation (Case_0) show
that overall the predicted PBL heights (not shown here) have a regular
diurnal variation in the downtown SPMA with averaged daily values around
500 m at both the beginning and the end, and of up to 700 m in the middle of the
study period, when lower concentrations of aerosols were observed.
The observed and predicted hourly variations of O3
concentrations at six sites in SPMA for the 3 km modelling domain.
Taylor diagram (Taylor, 2001) showing the individual correlation
coefficients, mean biases, and normalized standard deviations for the
PM10, PM2.5, OC, and EC concentrations.
Figures 11–13 show the predicted average surface distribution of PM2.5,
PM10, and PM2.5 : PM10 ratio for the 3 km modelling domain,
respectively. Red dots and cyan numbers represent the locations and the
observed mean PM concentrations (or mean PM concentration ratios) at the
measurement sites, respectively. Major contributions of PM2.5 on the
total PM10 concentration were observed mainly over offshore continental
areas (see Fig. 13). High PM2.5 : PM10 concentration ratios would be
firstly associated with the transportation of fine particles and gases from
upwind regions (see Fig. 6), followed by a production of fine particles from
biogenic emissions. Additional comparisons between the observed and
predicted concentrations of OC and EC at IAG-USP (the only site with
measurements of OC and EC) are shown in Fig. 14. As it has been pointed out
in Sect. 2 of the Supplement, under predicted OC
concentrations could be associated, among others (e.g., underestimation of
POA emissions, inaccurate meteorological predictions), with an
underestimation of SOA, probably due to the absence of oxidation of
monoterpenes and a limited treatment of anthropogenic VOCs oxidation in the
RADM2 mechanism, as discussed by Tuccella et al. (2012). The SORGAM aerosol
module considers the formation of anthropogenic SOAs from the oxidation of
alkane, alkene, and aromatic VOCs as well as the biogenic SOA formation from
the oxidation of alpha-pinene, limonene and isoprene VOCs. Recent studies
coupling non-traditional SOA models (volatility basis set approaches) in
WRF-Chem show improvements in the predicted SOA concentrations, although
these are still lower than those observed (e.g. Li et al., 2011b; Ahmadov et
al., 2012; Shrivastava et al., 2013).
The predicted average surface distribution of PM2.5
concentrations for the whole study period in the 3 km modelling domain. Red
dots represent the locations of the measurement sites with information on
PM2.5, whereas cyan numbers represent the observed average PM2.5
concentration in those sites: 23.4 µg m-3 in IPEN-USP,
21.3 µg m-3 in IAG-USP, and 22.2 µg m-3 in CON.
The predicted average surface distribution of PM10
concentrations for the whole study period in the 3 km modelling domain. Red
dots represent the locations of the measurement sites with information on
PM10, whereas cyan numbers represent the observed average PM10
concentration in those sites: 49.5 µg m-3 in IAG-USP and
38.7 µg m-3 in CON.
The predicted average surface distribution of the
PM2.5 : PM10 ratio for the whole study period in the 3 km modelling
domain. Red dots represent the locations of the measurement sites with
information on both PM2.5 and PM10, whereas cyan numbers represent
the observed average PM2.5 : PM10 ratio in those sites: 0.43 in
IAG-USP and 0.57 in CON.
The observed and predicted daily variations of OC and EC
concentrations at IAG-USP.
The observed and predicted average aerosol mass size distribution
for SO4, NO3, NH4, Na, Cl, and other PM10 constituents
at IAG-USP. The observed aerosol distributions were collected in 10 size
classes using a rotated impactor (MOUDI) and joined adequately according to
the three modes used by the MADE aerosol scheme: Aitken (< 0.1 µm),
accumulation (0.1–1 µm), and coarse (> 1 µm).
The five inorganic ions carried in MADE are only calculated for
the Aitken and accumulation modes. The WRF's PM10 aerosol variable
does not include neither OC nor EC for this comparison.
On the other hand, measurements of mass size distribution were also made with
a MOUDI impactor at IAG-USP, following the protocol described in Miranda and
Andrade (2005). Constituents of aerosol were subsequently determined by X-Ray
fluorescence analysis and ion chromatography analysis. As previously
indicated in this section, the main identified species are SO4,
NO3, NH4, Na, and Cl. The observed average aerosol composition is
derived using measurements from both MOUDI impactor and SUNSET analyzer. To
perform the comparisons of mass size distribution, we adequately joined the
MOUDI bin sizes according to the three modes used by the MADE aerosol module:
Aitken (< 0.1 µm), accumulation (0.1–1 µm) and
coarse (> 1 µm). The observed and predicted aerosol
mass size distributions averaged over the same sampling time period (16 days
along the study period) are shown in Fig. 15. Over the downtown SPMA, both
the observed and predicted fine particles from accumulation mode account for
the majority of the total PM2.5 mass. Since the formation-growth
processes of aerosols in question are explicitly treated in the Aitken and
accumulation modes, the predicted concentrations for particles larger than
1 µm are assumed to be zero. In this case, the mass of particles
larger than 1 µm is allocated to the PM10 aerosol variable
(see Fig. 15). The comparison between the observed and predicted average
contributions for the main identified aerosol constituents at IAG-USP is
shown in Fig. 16. Both the observed and predicted OC and EC make up the
largest fraction of PM2.5 mass with contributions of 55 and 40 %,
respectively. In addition, it was found that the predicted SOA concentrations
contribute 17 % of the predicted total OC concentration at this
measurement site. Various global and regional scale SOA simulations have been
conducted using mass-based yield and partitioning coefficients, but they have
underestimated the SOA concentrations by roughly an order of magnitude,
especially over urban regions (Matsui et al., 2014). Using the same SOA
formation approach employed by this study and a conversion factor of 1.6 to
convert the emissions of OC to OM, Tuccella et al. (2012) found simulated
SOA : OM ratios in the 5–40 % range against the observed range of
50–80 %. Although the predicted average PM2.5 concentration
(14.48 µg m-3) was lower than observed
(22.32 µg m-3), the mean aerosol chemical composition was
reasonably well represented by the model (see Fig. 16).
The observed and predicted average contributions for the main
identified constituents of PM2.5 at IAG-USP.
Contribution of dust-sea salt and coarse anthropogenic aerosols to
PM concentration
The evaluation of the contribution of dust and sea salt aerosols on
PM10 concentration is performed from the sum of their concentrations
divided by the PM10 concentration. The simulated average ratio between
dust–sea salt aerosols and the total PM10 mass concentration is shown
in Fig. 17b. High concentration ratios have been observed over the ocean
where sea salt emissions are by far the most important aerosols source.
Unlike high concentration ratios over the ocean, lower concentration ratios
are observed over the continent far away from the coast. In this region, the
main sources of atmospheric aerosols would be the emission of primary
biological aerosol, SOA formed from the emission of biogenic volatile
organic compounds (BVOCs), and forest fires. However, particles could also
be transported from remote areas. In addition, we can also observe that dust
and sea salt aerosols have a contribution between 40 and 50 % of the total
PM10 concentration in the downtown SPMA. Furthermore, it is possible to
estimate the contribution of all the other PM10 (i.e., the coarse
anthropogenic aerosol) to the total PM10 mass concentration. It may be
directly calculated from the model or estimated from Figs. 13 and 17b
once the sum of concentrations of PM2.5, dust and sea salt, and coarse
anthropogenic aerosol represents 100 % of the total PM10 mass
concentration. For example, we found that the coarse anthropogenic aerosol
represents around 10 % of PM10 in the downtown SPMA.
Evaluation of secondary aerosol formation
As described in Sect. 2.1, aerosol module employed by this study
(MADE/SORGAM) includes the homogeneous nucleation in the sulphuric
acid-water system. The sulphuric acid is the most significant condensable
molecule formed in the atmosphere, which has also been long recognised as
the most important molecule from the point of view of the nucleation of new
particles (Jenkin and Clemitshaw, 2000; Seinfeld and Pandis, 2006). However,
for the SPMA, the importance of SOA formed from the anthropogenic emission
of fuel used by the transport sector was noted (Salvo and Geiger, 2014).
According to the official emission inventory developed by the Sao Paulo
Environmental Protection Agency (CETESB, 2013), the SOA explains 51 % of
the fine particle mass concentration, with the vehicular emission being its
main source. The subsequent growth processes involve aerosol growth by
condensation of condensable material onto existing particles, and by
coagulation of particles to form larger particles (Kumar et al., 2011,
2014). For example, particles in the accumulation mode emerge through
coagulation of particles from the Aitken mode (Kumar et al., 2011). It is
important to emphasize that the boundaries were updated with gas and aerosol
background concentrations coming from the 15 km modelling domain during the
whole simulation period. Thereafter, the impact of vehicular emissions on
the formation of fine particles was calculated from the predicted PM2.5
concentration considering an emission scenario (Case_1) in
which only emission of gases from vehicles and vegetation are taken into
account to be emitted to the atmosphere from the surface. The predicted
average PM2.5 (Case_1) : PM2.5 (Case_0) ratio is shown in Fig. 17a. A contribution between
20 and 30 % in the predicted baseline PM2.5 concentration in downtown
SPMA is found to correspond to the fine particles formation and
transportation processes. Higher concentration ratios over the SPMA
surroundings (30–50 %) could be associated with more efficient biogenic
emissions. Overall, it is observed that the formation efficiency increases
towards the northwest from the ocean. Deep red areas in Fig. 17a could also
be associated with the transportation of fine particles and gases from other
regions, in addition to having a more efficient production of fine particles
from biogenic emissions. For example, given the distribution of winds in
Fig. 6, the northern boundary could represent the main source of particles
and gases over this part of the simulation domain. Additionally, the
comparison between the predicted and observed OC and EC concentrations at
IAG-USP shown in Fig. 14 includes the Case_1 simulation in
which only emission of primary gases is taken into account in the assessment
of fine particles formation. The concentration peaks observed at the
beginning and at the end of the study period may be associated with the
transport of aerosol particles from both biomass and fossil fuel burning
areas (see Fig. 4). Considering the Case_1 simulation, we can
observe very low concentrations for EC (mean concentration of 0.01 µg m-3), as expected. This is because these particles are not produced by
photochemical processes in the atmosphere, but associated mainly with the
diesel exhaust.
The impact of (a) emissions of primary gases on the fine
particles formation, (b) emissions of dust-sea salt aerosols on the
PM10 concentration, and (c) aerosol direct effect on the ground level
O3 concentrations at 16:00 (local time).
Aerosol impact on O3 photochemistry
Ozone photochemistry production mainly depends on the two key photolysis
rates, as shown in Reactions (R1) and (R2), i.e. shortwave radiation able to
reach the surface to break molecules of O3 and NO2.
O3+hv→O2+O(1D)(λ< 320nm)NO2+hv→NO+O(3P)(λ< 420nm)
Therefore, the impact of aerosols on O3 photochemistry has been
evaluated from the impact of aerosols on downward shortwave radiation.
Attenuation (scattering and absorption) of downward shortwave radiation by
aerosols may substantially modify the photolysis rates, and thereby
affecting the ozone photochemistry production.
The average percentage change in surface O3 concentrations at 16:00
(local time) with and without the aerosol-radiation feedback module turned
on are shown Fig. 17c. Overall O3 is destroyed or formed (incoming
transport from other regions) in small quantities between -1 and +1 % in
relation to its total concentration. In addition, it was observed that the
surface O3 concentration decreased by around 2 % in the downtown
SPMA. Li et al. (2011a) found that the impact of aerosols on O3
formation in Mexico City was most pronounced in the morning with the O3
reduction of 5–20 %, but the reduction is less than 5 % in the
afternoon. Low reductions in the O3 concentration in the downtown SPMA
compared to results from other studies may be explained by the lower
predicted PM10 concentrations, which can lead to a minor attenuation of
the incoming solar radiation. Simulated mean downward shortwave fluxes at
ground surface (not shown) were up to 5 % higher for the
Case_2 than for the Case_0 during the
afternoon. The inclusion of the direct effect of aerosol particles was found
to have small reductions in the surface temperature (changes by around
2 %), presumably due to an increase in the number of atmospheric processes
involving downward longwave fluxes over this area. Forkel et al. (2012)
found an underestimation of predicted downward longwave radiation over the
southern Baltic Sea when the direct effect of aerosol particles was
neglected. Despite the highly non-linear behaviour of tropospheric O3,
the reduction in the predicted O3 concentrations indicates a high
efficiency of aerosols to attenuate the downward shortwave radiation, what
is plausible once it was found that low PM10 concentrations have a
capability to reduce ground-level O3 concentrations in a few ppb.
Summary and conclusions
The WRF-Chem community model has been used to evaluate the impact of
vehicular emissions on the fine particles formation in the SPMA. Three
31-day simulations, covering a period from 7 August to 6 September
2012, have been performed. The aims were to evaluate the impact of fine
particles formation (both inorganic and SOA) from gases emitted by road
vehicles as well as the aerosol impacts on the ozone formation
photochemistry. The results were compared with the measurements available
from the NUANCE-SPS project.
The predicted temporal variations of meteorology, PM2.5, PM10, and
O3 were found to agree well with the measurements at most of the sites
during the entire simulation period. However, the predicted concentrations
of PM2.5, PM10, and O3 (but in minor intensity) were lower
than the observed values. This difference could be associated with an
underestimation of the vehicular emissions and other emission sources such
as industry, heating, and cooking, which are not considered in this study.
Wind speed and direction played an important role in the distribution of
fine particles over the simulation domain. Backward trajectories analysis
suggested that aerosol particles from biomass burning were transported to
SPMA, impacting on the PM concentration over this region.
The baseline simulation (Case_0) showed that dust and sea
salt aerosols made a contribution between 40 and 50 % of the total
PM10 concentration in the downtown SPMA. On the other hand, the
Case_1, which represents simulations with gaseous emissions
only, indicates that the emissions of primary gases coming mainly from
vehicles have a potential to form new particles between 20 and 30 % in
relation to the baseline PM2.5 concentration found in the downtown
SPMA. Finally, the Case_2, which represents simulations with
aerosol-radiation feedback turned on, reveals a reduction in the surface
O3 concentration by around 2 % in the afternoon (16:00; local time)
when the aerosol-radiation feedback is taken into account.
This study provides a first step to understand the impact of vehicular
emissions on the secondary particles formation in the SPMA. Nevertheless,
more experimental campaigns are recommended for future work in order to
characterize aerosols in ambient air and to improve their emission estimates
so that a better understanding of physical and chemical properties and their
formation can be established. This study also evaluates the importance of
the VOCs in the formation of not only O3 but also of fine particles.
These compounds play an important role concerning health impacts and climate
change, and the control of their concentrations requires the description of
their formation mechanisms.